All-Optical Signal Processing

Springer Series in Optical Sciences 194
StefanWabnitz
BenjaminJ.Eggleton Editors
All-Optical
Signal
Processing
Data Communication and Storage
Applications
Springer Series in Optical Sciences
Volume 194
Founded by
H.K.V. Lotsch
Editor-in-Chief
William T. Rhodes, Georgia Institute of Technology, Atlanta, USA
Editorial Board
Ali Adibi, Georgia Institute of Technology, Atlanta, USA
Toshimitsu Asakura, Hokkai-Gakuen University, Sapporo, Japan
Theodor W. Hnsch, Max-Planck-Institut fr Quantenoptik, Garching, Germany
Ferenc Krausz, Ludwig-Maximilians-Universitt Mnchen, Garching, Germany
Bo A.J. Monemar, Linkping University, Linkping, Sweden
Herbert Venghaus, Fraunhofer Institut fr Nachrichtentechnik, Berlin, Germany
Horst Weber, Technische Universitt Berlin, Berlin, Germany
Harald Weinfurter, Ludwig-Maximilians-Universitt Mnchen, Mnchen,
Germany
The Springer Series in Optical Sciences, under the leadership of Editor-in-Chief William T. Rhodes,
Georgia Institute of Technology, USA, provides an expanding selection of research monographs in all
major areas of optics: lasers and quantum optics, ultrafast phenomena, optical spectroscopy techniques,
optoelectronics, quantum information, information optics, applied laser technology, industrial applica-
tions, and other topics of contemporary interest.
With this broad coverage of topics, the series is of use to all research scientists and engineers who need
up-to-date reference books.
The editors encourage prospective authors to correspond with them in advance of submitting a manu-
script. Submission of manuscripts should be made to the Editor-in-Chief or one of the Editors. See also
www.springer.com/series/624
Editor-in-Chief
William T. Rhodes
School of Electrical and Computer Engineering
Georgia Institute of Technology
Atlanta, GA 30332-0250
USA
e-mail: bill.rhodes@ece.gatech.edu
Editorial Board
Ali Adibi Bo A.J. Monemar
School of Electrical and Computer Engineering Department of Physics and Measurement Technology
Georgia Institute of Technology Materials Science Division
Atlanta, GA 30332-0250 Linkping University
USA 58183 Linkping, Sweden
e-mail: adibi@ee.gatech.edu e-mail: bom@ifm.liu.se
Toshimitsu Asakura Herbert Venghaus

Faculty of Engineering Fraunhofer Institut fr Nachrichtentechnik
Hokkai-Gakuen University Heinrich-Hertz-Institut
1-1, Minami-26, Nishi 11, Chuo-ku Einsteinufer 37
Sapporo, Hokkaido 064-0926, Japan 10587 Berlin, Germany
e-mail: asakura@eli.hokkai-s-u.ac.jp e-mail: venghaus@hhi.de
Theodor W. Hnsch Horst Weber

Max-Planck-Institut fr Quantenoptik Optisches Institut
Hans-Kopfermann-Strae 1 Technische Universitt Berlin
85748 Garching, Germany Strae des 17. Juni 135
e-mail: t.w.haensch@physik.uni-muenchen.de 10623 Berlin, Germany
e-mail: weber@physik.tu-berlin.de
Ferenc Krausz
Ludwig-Maximilians-Universitt Mnchen Harald Weinfurter
Lehrstuhl fr Experimentelle Physik Sektion Physik
Am Coulombwall 1 Ludwig-Maximilians-Universitt Mnchen
85748 Garching, Germany and Schellingstrae 4/III
Max-Planck-Institut fr Quantenoptik 80799 Mnchen, Germany
Hans-Kopfermann-Strae 1 e-mail: harald.weinfurter@physik.uni-muenchen.de
85748 Garching, Germany
e-mail: ferenc.krausz@mpq.mpg.de
More information about this series at http://www.springer.com/series/624

Stefan Wabnitz Benjamin J. Eggleton

Editors
All-Optical Signal
Processing
Data Communication and Storage
Applications
123
Editors
Stefan Wabnitz Benjamin J. Eggleton
Dipartimento di Ingegneria CUDOS, School of Physics
dellInformazione University of Sydney
Universit degli Studi di Brescia Sydney, NSW
Brescia Australia
Italy
ISSN 0342-4111 ISSN 1556-1534 (electronic)

ISBN 978-3-319-14991-2 ISBN 978-3-319-14992-9 (eBook)
DOI 10.1007/978-3-319-14992-9
Library of Congress Control Number: 2015933154
Springer Cham Heidelberg New York Dordrecht London

Springer International Publishing Switzerland 2015
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or
dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt
from the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained
herein or for any errors or omissions that may have been made.
Printed on acid-free paper
Springer International Publishing AG Switzerland is part of Springer Science+Business Media

(www.springer.com)
Preface
Although it holds the promise for substantial processing speed improvements, in

todays communication infrastructure optics remains largely conned to the signal
transport layer, as it lags behind electronics as far as signal processing is concerned.
This situation is bound to change in the near future as the tremendous growth of
data trafc requires the development of new, energy efcient, and fully transparent
all-optical networks for telecom and datacom applications. This book provides a
comprehensive review of the state of the art of all-optical devices based on non-
linear optical materials for applications to optical signal processing. Contributors to
this book present breakthrough solutions for enabling a pervasive use of optics in
data communication and signal storage applications. The book content ranges from
the development of innovative materials and devices, such as graphene and pho-
tonic crystal structures, to the use of nonlinear optical signal processing for secure
quantum information processing, for increasing the transmission channel capacity,
and for enhancing the performance of broadband radio frequency signal processing.
The book is expected to benet all researchers in the elds of optical communi-
cations, photonic devices for optical signal processing, nonlinear guided wave
optics, quantum information processing, and microwave photonics, including
senior undergraduate and postgraduate students and industry researchers.
Chapter 1 summarizes the recent progress in materials and structures for all-
optical signal processing that employ either second- or third-order optical nonlin-
earities. The dominant choice for quadratic materials is provided by periodically
poled lithium niobate waveguides. For cubic nonlinearities, the range of materials
ranges from glasses to both active and passive semiconductor devices: a brief
summary of the advantages and disadvantages of each class of materials and device
structure is provided. In Chap. 2, recent advances in new nonsilicon CMOS-
compatible platforms for nonlinear integrated optics are revised, focusing on Hydex
glass and silicon nitride. The promising new platform of amorphous silicon is also
briefly discussed. These material systems have opened up new functionalities such
as on-chip optical frequency comb generation, ultrafast optical pulse generation,
and measurement. Chapter 3 overviews the principles of optical switching devices,
based on either optical or electrical control signals, which permit to avoid the
v
vi Preface
necessity of electro-optic conversion. Discussed devices include nonlinear mode

couplers and interferometers based on optical bers, and integrated waveguides
based on photonic crystal structures or surface wave interactions in novel materials
such as graphene. Chapter 4 reviews the recent progress on using nonlinear optical
bers for optical pulse shaping in the temporal and spectral domains. Signicant
examples that are most relevant for applications include the synthesis of specialized
temporal waveforms, the generation of ultrashort pulses, and optical
supercontinuum.
Given the exponential growth rate of the total volume of data transported across
the communication network, energy consumption, alongside increased information
capacity, has become a critical driver in deploying new technologies. In addition to
transponders at the end terminals of an optical network, certain signal processing
functions, such as regeneration, format conversion, wavelength conversion, and
arbitrary waveform generation, are often proposed. The following chapters of the
book discuss how many of these intermediate functions may be performed all-
optically, with the primary advantage of increased bandwidth and consequent
resource sharing. In Chap. 5 the need, the general principles, and the approaches
used for the all-optical regeneration of mainly phase encoded signals of differing
levels of coding complexity are discussed. The key underpinning technology and
the current state of the art of optical regeneration, including a historic perspective,
are presented.
Chapter 6 presents the theory and experiments of photonic signal processing,
logic operations, and computing. These functionalities take advantage of nonlinear
processes with ultrafast response time to perform high speed operations either on
analog or digital optical signals directly in the optical domain. Practical all-optical
frequency generation and conversion requires highly efcient parametric interac-
tions across a wide spectral band. Chapter 7 presents a new class of traveling wave
parametric mixers for efcient, cavity-free frequency generation. Driven by
continuous-wave seeds, these devices combine inherently more stable lasers with
distributed noise inhibition in dispersion-managed parametric processes. The
operating principles, the design methodology, and the performance limits of
parametric mixers are discussed, together with applications to signal multicasting
and ultrafast channel processing.
Optical pulse shaping techniques are an active area of research for increasing the
spectral efciency of optical modulation formats in dense wavelength division
multiplexed (DWDM) transmissions, by avoiding interchannel and intersymbol
interference. In Chap. 8 the main pulse shapes of interest are introduced, the dif-
ferent available techniques for their generation are presented, along with the
associated signal multiplexing schemes, namely orthogonal frequency division
multiplexing (OFDM) and Nyquist pulse modulation. The relative performances of
electronic and optical signal processors for implementing Fourier transforms and
Nyquist pulse generation are discussed.
As previously mentioned, energy saving will be a main driver for the transition
from electronic to optical signal processing solutions. Chapter 9 describes advanced
functionalities for optical signal processing with reduced energy consumption using
Preface vii
optical time lenses. This approach permits broadband optical processing, which is
also capable of handling many bits in a single operation. In this way, the processing
energy is shared by the many bits, and the energy per bit is reduced. The basic
functionality is serial-to-parallel conversion in a single time lens. Combining time
lenses into telescopic arrangements allows for more advanced signal processing
solutions, such as conversion of OFDM signals into DWDM-like signals, which
can be separated passively, i.e., without additional energy consumption. The pre-
viously discussed signal processing functions may also be performed by using the
very optoelectronic devices used in the transponders themselves, either including
decision circuitry and/or forward error correction or as linear media converters. In
Chap. 10, the performance and energy consumption of digital coherent tran-
sponders, linear coherent repeaters, and modulator-based pulse shaping/frequency
conversion is analyzed, thus setting an important benchmark for the proposed all-
optical implementations.
The exponential growth in demand for information transmission capacity
requires a rethinking of the maximum or Shannon capacity of the ber-optic
communication channel, in the presence of ber nonlinearity. Chapter 11 addresses
the problem of estimating the Shannon capacity for nonlinear communication
channels, and discusses the potential of different optical signal coding, transmis-
sion, and processing techniques to improve the information capacity and increase
the system reach of ber-optic links.
Information and communication technologies based on quantum optics princi-
ples may lead to greatly improved functionalities such as enhanced sensing,
exponentially faster computing, and the secure transfer of information. Chapter 12
overviews techniques for the nonlinear optics-based encoding and fully secure
transfer of information in a quantum communication optical network. Chapter 13
discusses how classical optical signal processing techniques can be extended to
nonclassical entangled photon states, thus permitting unprecedented control of the
time frequency correlations shared by these light quanta. Moreover, in Chap. 13 it is
shown that quantum properties produce interesting effects that are not observable
with classical elds. Examples include Fourier transform pulse shaping, which
relies on programmable spectral ltering and electro-optic modulation, where the
temporal phase or amplitude of the entagled photon state is manipulated by means
of an electrical signal.
Chapter 14 discusses how nonlinear optical effects in photonic chip scale devices
may be exploited for enhancing the performance of radio frequency (RF) signal
processing in microwave photonics applications. Specic examples presented in
Chap. 14 include frequency agile and high suppression microwave bandstop lters,
general purpose programmable analog signal processors, and high performance
active microwave lters. Finally, Chap. 15 presents recent advances in optical
signal processing techniques for wireless RF signals. Specically, Chap. 15 dis-
cusses photonic architectures for wideband analog signal processing, including RF
beamforming, co-channel interference cancelation, and physical layer security.
viii Preface
Besides broadband operation, photonics offers reduced size, weight, and power, in
addition to low transmission loss, rapid recongurability, and immunity to elec-
tromagnetic interference.
We would like to thank Dr. Claus E. Ascheron, Springer Executive Editor for
Physics, for inviting us, during the 2013 Conference on lasers and electro-optics in
Munich, Germany, to bring a volume on all-optical signal processing to a wider
audience. We also acknowledge helpful comments and suggestions by Dr. Herbert
Venghaus of the Fraunhofer Institute for Telecommunications, and Editor of the
Springer Series in Optical Sciences. Last but not the least, we are most grateful to
all colleagues who contributed to this book for their brilliant work and continued
effort in bringing this project to reality.
Brescia, Italy Stefan Wabnitz

Sydney, Australia Benjamin J. Eggleton
Contents
1 Materials and Structures for Nonlinear Photonics . . . . . . . . . . .. 1

Xin Gai, Duk-Yong Choi, Steve Madden and Barry Luther-Davies
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 All-Optical Processing Using (2) Nonlinearities . . . . . . . . . . . 4
1.3 All-Optical Processing Using (3) Nonlinearities . . . . . . . . . . . 8
1.3.1 Properties of (3) Materials and Devices . . . . . . . . . . . 10
1.4 All-Optical Processing in Semiconductor Optical Amplifiers. . . 18
1.4.1 Types and Origins of Nonlinear Effects
in Semiconductor Optical Amplifiers . . . . . . . . . . . .. 18
1.4.2 Impairments in SOA Devices . . . . . . . . . . . . . . . . .. 19
1.4.3 Current State of the Art and Future Prospects
for SOA Nonlinear Processing . . . . . . . . . . . . . . . .. 23
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 26
2 CMOS Compatible Platforms for Integrated Nonlinear Optics ... 35

David J. Moss and Roberto Morandotti
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.2 Platforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.3 Low Power Nonlinear Optics in Ring Resonators . . . . . . . . . . 46
2.4 Microresonator-Based Frequency Combs . . . . . . . . . . . . . . . . 48
2.5 Advanced Frequency Comb Generation . . . . . . . . . . . . . . . . . 53
2.6 Supercontinuum Generation . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.7 Comb Coherence and Dynamic Properties . . . . . . . . . . . . . . . 56
2.8 Ultrashort Pulsed Modelocked Lasers . . . . . . . . . . . . . . . . . . 60
2.9 Ultrafast Phase Sensitive Pulse Measurement . . . . . . . . . . . . . 60
2.10 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
ix
x Contents
3 Optical Guided Wave Switching . . . . . . . . . . . . . . . . . . . . . .... 71

Costantino De Angelis, Daniele Modotto, Andrea Locatelli
and Stefan Wabnitz
3.1 Introduction: Optical Switching Using Guided-Waves . . . . . . . 71
3.2 All-Optical Pulse Switching in Optical Fibers . . . . . . . . . . . . . 72
3.2.1 Nonlinear Mode Coupling . . . . . . . . . . . . . . . . . . . . 72
3.2.2 Nonlinear Fiber Couplers . . . . . . . . . . . . . . . . . . . . . 74
3.2.3 Nonlinear Mach-Zehnder Interferometers . . . . . . . . . . 77
3.2.4 Nonlinear Loop Mirrors. . . . . . . . . . . . . . . . . . . . . . 78
3.2.5 Nonlinear Passive Loop Resonators. . . . . . . . . . . . . . 81
3.2.6 Optical Soliton Switching . . . . . . . . . . . . . . . . . . . . 82
3.3 Optical Switching in Integrated Optical
Waveguide Structures . . . . . . . . . . . . . . . . . . . . . . . . . .... 83
3.3.1 All-Optical Switching in Photonic
Crystal Couplers . . . . . . . . . . . . . . . . . . . . . . . .... 83
3.3.2 Graphene-Assisted Control of Coupling Between
Surface Plasmon Polaritons . . . . . . . . . . . . . . . .... 90
Optical Waveguides . . . . . . . . . . . . . . . . . . . . .... 93
3.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 99
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 100
4 Temporal and Spectral Nonlinear Pulse Shaping Methods

in Optical Fibers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 105
Sonia Boscolo, Julien Fatome, Sergei K. Turitsyn,
Guy Millot and Christophe Finot
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 105
4.2 Pulse Propagation in Optical Fibers. . . . . . . . . . . . . . . . . . .. 106
4.3 Pulse Shaping in Normally Dispersive Fibers:
From the Generation of Specialized Temporal
Waveforms to Spectral Sculpturing . . . . . . . . . . . . . . . . . . .. 108
4.3.1 Generation of Specialized Temporal Waveforms . . . .. 108
4.3.2 Spectral Sculpturing . . . . . . . . . . . . . . . . . . . . . . .. 112
4.4 Pulse Shaping in the Anomalous Dispersion Regime:
From Ultrashort Temporal Structures to Ultra-Broad Spectra .. 115
4.4.1 Generation of High-Repetition-Rate
Ultrashort Pulses. . . . . . . . . . . . . . . . . . . . . . . . . .. 115
4.4.2 Generation of Frequency-Tunable Pulses . . . . . . . . .. 117
4.4.3 Supercontinuum Generation and Optical
Rogue Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 119
4.5 Conclusions and Perspectives . . . . . . . . . . . . . . . . . . . . . . .. 122
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 123
Contents xi
5 Optical Regeneration . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..... 129

Francesca Parmigiani, Radan Slavk, Joseph Kakande,
Periklis Petropoulos and David Richardson
5.1 Introduction to Optical Regeneration . . . . . . . . . . . . . . ..... 129
5.2 Optical Regenerators for Simple Amplitude
Encoded Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
5.3 Regeneration of Phase-Only Encoded Signals . . . . . . . . . . . . . 134
5.3.1 Modulation Format Specific PSA Regenerators. . . . . . 138
5.3.2 Modulation Format Transparent PSA. . . . . . . . . . . . . 144
5.4 Regeneration of Amplitude and Phase Encoded Signals . . . . . . 145
5.4.1 Modulation Format Specific PSA Regenerators. . . . . . 146
5.4.2 Modulation Format Transparent PSA Schemes . . . . . . 150
5.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
6 Photonic Signal Processing for Logic and Computation . . . . . . . . 157

Antonella Bogoni and Alan Willner
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
6.2 Photonic Logic and Computation Functions for Multi-Format
Data Communication and Storage . . . . . . . . . . . . . . . . . . . . . 158
6.3 Overview of Nonlinear Processes . . . . . . . . . . . . . . . . . . . . . 158
6.3.1 Wave Mixing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
6.3.2 Phase Modulation . . . . . . . . . . . . . . . . . . . . . . . . . . 163
6.4 Enabling Technologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
6.4.1 Optical Fiber . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
6.4.2 Semiconductor Devices . . . . . . . . . . . . . . . . . . . . . . 164
6.4.3 Photonic Crystals . . . . . . . . . . . . . . . . . . . . . . . . . . 165
6.4.4 Periodically Poled Lithium Niobate Waveguides . . . . . 165
6.4.5 Silicon Devices. . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
6.5 State of the Art for Logic . . . . . . . . . . . . . . . . . . . . . . . . . . 166
6.5.1 Overview of Logic Functions
and Achieved Results . . . . . . . . . . . . . . . . . . . . . . . 166
6.5.2 OOK 640 Gbit/s Logic Operations . . . . . . . . . . . . . . 166
6.5.3 PSK 160 Gbit/s Logic Functions. . . . . . . . . . . . . . . . 168
6.5.4 Hexadecimal 16PSK Addition . . . . . . . . . . . . . . . . . 170
6.6 State of the Art for Computation. . . . . . . . . . . . . . . . . . . . . . 171
6.6.1 Overview of Tapped Delay Lines . . . . . . . . . . . . . . . 171
6.6.2 Fundamental Tools to Enable Photonic TDL . . . . . . . 171
6.6.3 Optical 1D Correlation Results Using Nonlinearities
and On-Chip MZIs . . . . . . . . . . . . . . . . . . . . . . . . . 173
6.6.4 Optical WDM Correlator and 2D Correlation . . . . . . . 176
6.6.5 Discrete Fourier Transforms Using Nonlinearities
and On-Chip MZIs . . . . . . . . . . . . . . . . . . . . . . . . . 178
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
xii Contents
7 Wide-Band and Noise-Inhibited Signal Manipulation

in Dispersion-Engineered Parametric Mixers . . . . . . . . . . . . . . .. 185
Bill P.-P. Kuo and Stojan Radic
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
7.2 Fundamentals of Parametric Mixers. . . . . . . . . . . . . . . . . . . . 187
7.2.1 One-Pump Parametric Mixing. . . . . . . . . . . . . . . . . . 187
7.2.2 Two-Pump Parametric Mixing . . . . . . . . . . . . . . . . . 189
7.2.3 Signal Processing Functions of Parametric Mixers. . . . 190
7.2.4 Effect of Chromatic Dispersion. . . . . . . . . . . . . . . . . 191
7.3 Dispersion-Stable Waveguide Engineering for Wide-Band
Parametric Mixer Synthesis . . . . . . . . . . . . . . . . . . . . . . . .. 192
7.3.1 Wide-Band Parametric MixingThe
Atomic-Scale Challenge. . . . . . . . . . . . . . . . . . . . .. 193
7.3.2 Post-Fabrication Dispersion Fluctuations
Rectification. . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 193
7.3.3 Waveguide Design Methods for Achieving
Intrinsic Dispersion Stability . . . . . . . . . . . . . . . . .. 197
7.4 Inhomogeneous Dispersion Engineering for Noise-Inhibited
Parametric Mixing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 201
7.4.1 Self-seeded Two-Pump Parametric
MixingHomogeneous Limit . . . . . . . . . . . . . . . . . 201
7.4.2 Inhomogeneous Dispersion Engineering. . . . . . . . . . . 202
7.4.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
7.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
8 All-Optical Pulse Shaping for Highest Spectral Efciency . . . . . .. 217

Juerg Leuthold and Camille-Sophie Brs
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
8.2 Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
8.3 Orthogonal Frequency Division Multiplexing (OFDM) . . . . . . 223
8.3.1 OFDM Tx and Rx Implementations . . . . . . . . . . . . . 223
8.3.2 Optical Fourier Transform Processors
and Optical OFDM . . . . . . . . . . . . . . . . . . . . . . . . . 227
8.3.3 OFDM Tx and RxExperimental Implementations. . . 234
8.4 Nyquist Pulse Shaping. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
8.4.1 Electronic Nyquist Processing. . . . . . . . . . . . . . . . . . 241
8.4.2 Digital Signal Processing Based Generation . . . . . . . . 243
8.4.3 Optical Processors. . . . . . . . . . . . . . . . . . . . . . . . . . 244
8.4.4 Implementations . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
Contents xiii
9 Energy-Efcient Optical Signal Processing Using

Optical Time Lenses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 261
Leif Katsuo Oxenlwe, Michael Galili,
Hans Christian Hansen Mulvad, Hao Hu, Pengyu Guan,
Evarist Palushani, Mads Lillieholm and Anders Clausen
9.1 Introduction: Energy-Efficient Solutions Using Nonlinear
Optical Signal Processing. . . . . . . . . . . . . . . . . . . . . . . . . .. 261
9.1.1 Energy-Efficient Optical Signal Processing:
Many Bits per Operation . . . . . . . . . . . . . . . . . . . .. 262
9.2 Time-Domain Optical Fourier Transformation/Time
Lens Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
9.2.1 Time Lens Principle and Time-Domain OFT . . . . . . . 266
9.2.2 Important OFT/Time Lens References . . . . . . . . . . . . 269
9.3 Serial-to-Parallel Conversion . . . . . . . . . . . . . . . . . . . . . . . . 270
9.3.1 640 Gbit/s OOK OTDM-to-DWDM Conversion . . . . . 270
9.3.2 Serial-to-Parallel Conversion of Data
with Advanced Modulation Formats . . . . . . . . . . . .. 273
9.3.3 Wavelength-Preserving Serial-to-Parallel
Conversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 273
9.3.4 Nyquist-OTDM to OFDM-Like
Signal Conversion. . . . . . . . . . . . . . . . . . . . . . . . .. 274
9.4 Spectral Telescopes and Applications . . . . . . . . . . . . . . . . .. 277
9.4.1 WDM Nonlinear Optical Signal ProcessingWDM
Grid Manipulation. . . . . . . . . . . . . . . . . . . . . . . . .. 277
9.4.2 OFDM-to-DWDM Conversion by Spectral
Magnification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280
9.5 Time-Domain Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
9.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
10 Signal Processing Using Opto-Electronic Devices . . . . . . . . . . ... 291

Mary McCarthy, Simon Fabbri and Andrew Ellis
10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
10.2 Coherent Transponders . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
10.2.1 Power Consumption of Opto-Electronic Devices . . . . . 293
10.2.2 Energy Consumption of DSP . . . . . . . . . . . . . . . . . . 295
10.3 Link Power Consumption . . . . . . . . . . . . . . . . . . . . . . . . . . 302
10.4 Optical Signal Modulation Using Electro-Optic Modulators . . . 306
10.4.1 Pulse Generation Using Electro-Absorption
Modulators. . . . . . . . . . . . . . . . . . . . . . . . . . . . ... 306
10.4.2 Pulse Generation and Optical Sampling Using
Mach Zehnder Modulators . . . . . . . . . . . . . . . . . ... 308
10.4.3 Optical Comb Generation. . . . . . . . . . . . . . . . . . ... 310
xiv Contents
10.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316

References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
11 Optical Information Capacity Processing . . . . . . . . . . . . . . . . ... 325

Mariia Sorokina, Andrew Ellis and Sergei K. Turitsyn
11.1 Introduction: Information Capacity in Optical
Communications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
11.2 Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
11.2.1 Shannon Channel Capacity. . . . . . . . . . . . . . . . . . . . 329
11.2.2 Numerical Computation of Shannon Capacity. . . . . . . 330
11.3 Linear Additive White Gaussian Noise Channel . . . . . . . . . . . 331
11.3.1 Capacity of Linear Additive White Gaussian
Noise Channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
11.4 Nonlinear Fiber Channel . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
11.4.1 Basic Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
11.4.2 Simplified Nonlinear Channel Models . . . . . . . . . . . . 335
11.5 Optical Signal Processing to Improve Signal Transmission . . . . 343
11.5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
11.5.2 Quasilossless Transmission . . . . . . . . . . . . . . . . . . . 343
11.5.3 Optical Phase Conjugation . . . . . . . . . . . . . . . . . . . . 344
11.5.4 Phase-Conjugated Twin Waves. . . . . . . . . . . . . . . . . 345
11.5.5 Optical Regeneration . . . . . . . . . . . . . . . . . . . . . . . . 346
11.6 Conclusion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
12 Nonlinear Optics for Photonic Quantum Networks . . . . . . . . . ... 355

Alex S. Clark, Lukas G. Helt, Matthew J. Collins, Chunle Xiong,
Kartik Srinivasan, Benjamin J. Eggleton and Michael J. Steel
12.1 Introduction: Photonic Quantum Networks . . . . . . . . . . . . ... 356
12.1.1 Photonic Qubits, Gates and Algorithms . . . . . . . . ... 357
12.2 Nonlinear Optics for Single Photon Generation. . . . . . . . . ... 362
12.2.1 Single Photon Sources and the Heralding
of Single Photons . . . . . . . . . . . . . . . . . . . . . . . ... 362
12.2.2 Photon Pair Sources and Photon Statistics . . . . . . ... 365
12.2.3 Pair Generation Processes in Second-Order
Nonlinear Media. . . . . . . . . . . . . . . . . . . . . . . . ... 370
12.2.4 Pair Generation Processes in Third-Order
Nonlinear Media. . . . . . . . . . . . . . . . . . . . . . . . . . . 376
12.2.5 Quantum Description of Photon Pair States . . . . . . . . 384
12.2.6 Multiplexing: A Route to On-demand Photons . . . . . . 394
12.3 Nonlinear Optics for Quantum Frequency Conversion . . . . . . . 398
12.3.1 Second-Order Nonlinear Media. . . . . . . . . . . . . . . . . 399
12.3.2 Third-Order Nonlinear Media . . . . . . . . . . . . . . . . . . 402
Contents xv
12.3.3 Future Directions with Quantum

Frequency Conversion . . . . . . . . . . . . . . . . . . . . . . . 404
12.4 Nonlinear Optics for Quantum Communication. . . . . . . . . . . . 406
12.4.1 Single Photon Schemes . . . . . . . . . . . . . . . . . . . . . . 407
12.4.2 Entanglement-Based Schemes. . . . . . . . . . . . . . . . . . 408
12.4.3 Long Distance Quantum Communication . . . . . . . . . . 409
12.5 Conclusion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 412
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 412
13 Biphoton Pulse Shaping . . . . . . . . . . . . . . . . . . . . . . . ........ 423

Joseph M. Lukens and Andrew M. Weiner
13.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423
13.2 Classical Pulse Shaping . . . . . . . . . . . . . . . . . . . . . . . . . . . . 425
13.3 Biphoton Pulse Shaping: Theory. . . . . . . . . . . . . . . . . . . . . . 428
13.4 Biphoton Pulse Shaping: Important Experiments . . . . . . . . . . . 432
13.4.1 Ultrafast Coincidence Detection . . . . . . . . . . . . . . . . 432
13.4.2 Additional Biphoton Shaping Experiments . . . . . . . . . 434
13.5 Detailed Example I: Cancellation of Dispersion
or Modulation . . . . . . . . . . . . . . . . . . . . . . . . . . ........ 435
13.6 Detailed Example II: Encoding and Decoding
of Biphoton Wavepackets . . . . . . . . . . . . . . . . . . ........ 439
13.7 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ........ 444
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ........ 446
14 Harnessing Nonlinear Optics for Microwave Signal Processing . .. 449

David Marpaung, Ravi Pant and Benjamin J. Eggleton
14.1 Microwave Photonics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 449
14.1.1 MWP System Performance . . . . . . . . . . . . . . . . . . . 451
14.1.2 Integrated Microwave Photonics . . . . . . . . . . . . . . . . 452
14.1.3 Nonlinear Integrated Microwave Photonics. . . . . . . . . 452
14.2 Stimulated Brillouin Scattering . . . . . . . . . . . . . . . . . . . . . . . 453
14.2.1 On-Chip SBS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453
14.3 Reconfigurable Microwave Filters Using SBS . . . . . . . . . . . . 454
14.4 SBS Tunable Delay Line and Phase Shifter . . . . . . . . . . . . . . 458
14.4.1 On-Chip SBS Tunable Delay Lines . . . . . . . . . . . . . . 459
14.4.2 On-Chip SBS Phase Shifter . . . . . . . . . . . . . . . . . . . 459
14.5 FWM and XPM for MWP Signal Processing . . . . . . . . . . . . . 460
14.6 Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 462
14.6.1 General Purpose Analog Processor . . . . . . . . . . . . . . 462
14.6.2 Highly Integrated Tunable RF Filter . . . . . . . . . . . . . 464
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 464
xvi Contents
15 Ultrafast Optical Techniques for Communication

Networks and Signal Processing . . . . . . . . . . . . . . . . . . . . . . . . . 469
Bhavin J. Shastri, John Chang, Alexander N. Tait, Matthew P. Chang,
Ben Wu, Mitchell A. Nahmias and Paul R. Prucnal
15.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 469
15.1.1 Scenario of Interest . . . . . . . . . . . . . . . . . . . . . . . . . 473
15.2 Primer on Antenna Arrays and Beamforming . . . . . . . . . . . . . 474
15.2.1 Narrowband Beamforming . . . . . . . . . . . . . . . . . . . . 476
15.2.2 Wideband Beamforming . . . . . . . . . . . . . . . . . . . . . 478
15.3 Microwave Photonic Filters . . . . . . . . . . . . . . . . . . . . . . . . . 480
15.3.1 Requirements for MPFs . . . . . . . . . . . . . . . . . . . . . . 481
15.3.2 Popular MPF Architectures . . . . . . . . . . . . . . . . . . . 482
15.3.3 Optical Technologies for MPFs . . . . . . . . . . . . . . . . 484
15.4 Photonic Adaptive Beamformers . . . . . . . . . . . . . . . . . . . . . . 485
15.4.1 State of the Art Photonic Beamformers . . . . . . . . . . . 486
15.4.2 Highly Scalable Adaptive Photonic Beamformer . . . . . 490
15.4.3 Photonic Beamforming for Physical Layer Security . . . 494
15.5 Summary and Concluding Remarks. . . . . . . . . . . . . . . . . . . . 498
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 499
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 505
Contributors
Antonella Bogoni CNIT, National Laboratory of Photonic Networks, Pisa, Italy

Sonia Boscolo School of Engineering and Applied Science, Aston Institute of
Photonic Technologies, Aston University, Birmingham, UK
Camille-Sophie Brs Institute of Electrical Engineering, Ecole Polytechnique
Fdrale de Lausanne (EPFL), Lausanne, Switzerland
John Chang Princeton University, Princeton, NJ, USA
Matthew P. Chang Princeton University, Princeton, NJ, USA
Duk-Yong Choi Centre for Ultrahigh-bandwidth Devices of Optical Systems,
Laser Physics Centre, RSPE, The Australian National University, Canberra, ACT,
Australia
Alex S. Clark School of Physics, Centre for Ultrahigh bandwidth Devices for
Optical Systems (CUDOS), Institute of Photonics and Optical Science (IPOS),
University of Sydney, Camperdown, NSW, Australia
Anders Clausen DTU Fotonik, Department of Photonics Engineering, Technical
University of Denmark, Kongens Lyngby, Denmark
Matthew J. Collins School of Physics, Centre for Ultrahigh bandwidth Devices
for Optical Systems (CUDOS), Institute of Photonics and Optical Science (IPOS),
Costantino De Angelis Dipartimento di Ingegneria dellInformazione, Universit
degli Studi di Brescia, Brescia, Italy
Benjamin J. Eggleton School of Physics, Centre for Ultrahigh bandwidth Devices
for Optical Systems (CUDOS), Institute of Photonics and Optical Science (IPOS),
Andrew Ellis School of Engineering and Applied Science, Aston Institute of
xvii
xviii Contributors
Simon Fabbri School of Engineering and Applied Science, Aston Institute of

Julien Fatome Laboratoire Interdisciplinaire CARNOT de Bourgogne, UMR
6303 CNRS-Universit de Bourgogne, Dijon, France
Christophe Finot Laboratoire Interdisciplinaire CARNOT de Bourgogne, UMR
6303 CNRS-Universit de Bourgogne, Dijon, France
Xin Gai Centre for Ultrahigh-bandwidth Devices of Optical Systems, Laser
Physics Centre, RSPE, The Australian National University, Canberra, ACT,
Australia
Michael Galili DTU Fotonik, Department of Photonics Engineering, Technical
Pengyu Guan DTU Fotonik, Department of Photonics Engineering, Technical
Lukas G. Helt MQ Photonics Research Centre and CUDOS, Department of
Physics and Astronomy, Macquarie University, North Ryde, NSW, Australia
Hao Hu DTU Fotonik, Department of Photonics Engineering, Technical Univer-
sity of Denmark, Kongens Lyngby, Denmark
Joseph Kakande Bell Laboratories, Alcatel-Lucent, Holmdel, NJ, USA
Bill P.-P. Kuo Department of Computer and Electrical Engineering, University of
California San Diego, La Jolla, CA, USA
Juerg Leuthold Institute of Electromagnetic Fields, ETH-Zurich, Zurich,
Switzerland
Mads Lillieholm DTU Fotonik, Department of Photonics Engineering, Technical
Andrea Locatelli Dipartimento di Ingegneria dellInformazione, Universit degli
Studi di Brescia, Brescia, Italy
Joseph M. Lukens Purdue University, West Lafayette, IN, USA
Barry Luther-Davies Centre for Ultrahigh-bandwidth Devices of Optical Sys-
tems, Laser Physics Centre, RSPE, The Australian National University, Canberra,
ACT, Australia
Steve Madden Centre for Ultrahigh-bandwidth Devices of Optical Systems, Laser
Physics Centre, RSPE, The Australian National University, Canberra, ACT,
Australia
David Marpaung School of Physics, Centre for Ultrahigh Bandwidth Devices for
Contributors xix
Mary McCarthy School of Engineering and Applied Science, Aston Institute of

Guy Millot Laboratoire Interdisciplinaire CARNOT de Bourgogne, UMR 6303
CNRS-Universit de Bourgogne, Dijon, France
Daniele Modotto Dipartimento di Ingegneria dellInformazione, Universit degli
Roberto Morandotti INRS-EMT, Varennes, QC, Canada
David J. Moss School of Electrical and Computer Engineering, RMIT University,
Melbourne, VIC, Australia
Hans Christian Hansen Mulvad DTU Fotonik, Department of Photonics Engi-
neering, Technical University of Denmark, Kongens Lyngby, Denmark
Mitchell A. Nahmias Princeton University, Princeton, NJ, USA
Leif Katsuo Oxenlwe DTU Fotonik, Department of Photonics Engineering,
Technical University of Denmark, Kongens Lyngby, Denmark
Evarist Palushani DTU Fotonik, Department of Photonics Engineering, Techni-
cal University of Denmark, Kongens Lyngby, Denmark
Ravi Pant School of Physics, Centre for Ultrahigh Bandwidth Devices for Optical
Systems (CUDOS), Institute of Photonics and Optical Science (IPOS), University
of Sydney, Camperdown, NSW, Australia
Francesca Parmigiani Optoelectronics Research Centre, University of South-
ampton, Southampton, Hampshire, UK
Periklis Petropoulos Optoelectronics Research Centre, University of Southamp-
ton, Southampton, Hampshire, UK
Paul R. Prucnal Princeton University, Princeton, NJ, USA
Stojan Radic Department of Computer and Electrical Engineering, University of
California San Diego, La Jolla, CA, USA
David Richardson Optoelectronics Research Centre, University of Southampton,
Southampton, Hampshire, UK
Bhavin J. Shastri Princeton University, Princeton, NJ, USA
Radan Slavk Optoelectronics Research Centre, University of Southampton,
Southampton, Hampshire, UK
Mariia Sorokina School of Engineering and Applied Science, Aston Institute of
Kartik Srinivasan Center for Nanoscale Science and Technology, National
Institute of Standards and Technology (NIST), Gaithersburg, MD, USA
xx Contributors
Michael J. Steel MQ Photonics Research Centre and CUDOS, Department of

Physics and Astronomy, Macquarie University, North Ryde, NSW, Australia
Alexander N. Tait Princeton University, Princeton, NJ, USA
Sergei K. Turitsyn School of Engineering and Applied Science, Aston Institute of
Stefan Wabnitz Dipartimento di Ingegneria dellInformazione, Universit degli
Andrew M. Weiner Purdue University, West Lafayette, IN, USA
Alan Willner Department of Electrical and Computer Engineering, University of
Southern California, Los Angeles, CA, USA
Ben Wu Princeton University, Princeton, NJ, USA
Chunle Xiong School of Physics, Centre for Ultrahigh bandwidth Devices for
Chapter 1
Materials and Structures for Nonlinear
Photonics
Xin Gai, Duk-Yong Choi, Steve Madden and Barry Luther-Davies
Abstract In this chapter we summarize progress in materials and structures for

all-optical signal processing that employ either the second or third order optical
nonlinearity. Three-wave mixing and cascading in periodically-poled lithium nio-
bate waveguides dominates signal processing in second order materials. In the case
of third-order nonlinearities, four wave mixing is the dominant physical process but
here the range of materials that have been employed is wider and ranges from
glasses to both active and passive semiconductor devices. This chapter provides a
brief summary of the advantages and disadvantages of each class of materials and
device structure as well as the state-of-the-art for each case.
1.1 Introduction
Research into all-optical processing dates back to the 1980s when the so-called
electronic bottleneck, the limited speed of electronics, was seen as the major
barrier that would limit the information capacity of ber-optic networks. This was
the era pre-dating the development of wavelength division multiplexing (WDM)
and, hence, research was exclusively focused on time division multiplexed (TDM)
signals on a single wavelength channel [1]. In this rst era of optical signal
processing the main idea was to use the ultra-fast third order nonlinearity of Kerr-
like materials to create fast switches so that an interleaved signal could be de-
multiplexed efciently to different output ports of a device. Although there was a lot
of work on self-switched devices, where routing was controlled by varying the
signal power itself, the most practical implementation involved switching the state
of the device using separate control pulses.
X. Gai D.-Y. Choi S. Madden B. Luther-Davies (&)

Centre for Ultrahigh-bandwidth Devices of Optical Systems,
Laser Physics Centre, RSPE, The Australian National University,
Canberra, ACT 2600, Australia
e-mail: Barry.Luther-Davies@anu.edu.au
Springer International Publishing Switzerland 2015 1

S. Wabnitz and B.J. Eggleton (eds.), All-Optical Signal Processing,
Springer Series in Optical Sciences 194, DOI 10.1007/978-3-319-14992-9_1
2 X. Gai et al.
A canonical optical switch consisted of a nonlinear Michelson interferometer in

which a third order nonlinear element existed in only one of the arms. At the input
port the beam was split in two and one half propagated through the linear arm and
the other through the nonlinear arm to the output. At the output the signals inter-
fered at a second beam splitter and the power was divided between the two output
ports depending on the relative phase of the beams. At low power if the phase shift
in the two arms was identical, the two beams interfered constructively to create an
output at the through port. However, if the power was increased (or a control
pulse was present) such that the nonlinear phase change that occurred in the non-
linear arm was nl = kon2IL = (L is the length of the nonlinear element; n2 its
nonlinear refractive index, I the light intensity and ko the free space wavenumber),
the power would be switched to the cross port. This type of nonlinear optical
switch, therefore, relied on a nonlinear phase shift caused by the real part of the
third order nonlinearity.
Of course, in this description, losses are ignored and no account is taken of the
time dependence of the pulse that can lead to partial switching and pulse distortion.
This can only be eliminated for the special case where the propagating pulses are
temporal solitons [2] or when the control pulse is flat-topped and long compared
with the signal pulse. Losses are, however, invariably present and can have a
number of important consequences. Firstly, linear (and nonlinear) losses lead to
power dissipation and this causes heating of the device. Via the thermo-optic effect
this creates a slow phase change that can imbalance the interferometer. Thermal
effects can be particularly serious in high average power applications, such as
processing of telecommunications signals, or when micro-resonators with very
small thermal mass are used as a switch [3]. Secondly, losses linear or nonlinear,
reduce the effective path length in the nonlinear medium and this can limit the
available nonlinear phase change because the power decays in the propagation
direction. Finally, in a case of a simple Michelson interferometer, loss can imbal-
ance the arms reducing the contrast of the switch. These matters were considered
extensively in the early 1990s and a number of gures of merit were proposed to
compare different switching devices and materials [1]. Whilst there is now little
research into these true optical switches, except in the case of photonic crystal
cavities, these remain relevant today.
One of the most widely quoted gures of merit (FOM) is associated with two-
photon absorption [4]. The nonlinear refractive index, n2, and the two-photon
absorption (2PA) coefcient, 2PA, reflect the real and imaginary components of the
third order susceptibility. They are related via the Kramers-Kronig relation just as is
the case of the linear refractive index and the linear absorption coefcient. As a
consequence, two-photon absorption can be present in any third order nonlinear
material operated at photon energies close to half the bandgap. Figures of merit
(FOM) dene the nonlinear phase change that can be achieved over an absorption
length. For two-photon absorption, the nonlinear phase change and the nonlinear
absorption coefcient are both linearly dependent on intensity and hence the
resulting FOM2PA depends only on material parameters and is written
FOM2PA = n2/(2PA). For efcient all-optical processing we ideally require
1 Materials and Structures for Nonlinear Photonics 3
FOM2PA 1. A similar FOM can be written for linear absorption as reads

FOM1PA = n2Imax/() [1] where Imax is the maximum intensity that the material can
sustain (e.g. limited by optical damage, or by the maximum power available for the
device) and is the linear absorption coefcient. Taking this further, in materials
where three photon absorption may exist (3PA) we obtain FOM3PA = n2/(3PAI)
where I is the beam intensity. Clearly in the case of 3PA the FOM3PA is no longer a
constant but gets smaller with as the intensity is increased.
There have been a number of attempts to predict the dispersion of the real and
imaginary components of the third order nonlinearity. One of the more useful
results was published by Sheik-Bahae et al. [57], and was based on a simple two-
band model for direct-gap semiconductors. Their analysis resulted in expressions
for the dependence of n2 and 2PAon the ratio of the photon tobandgap
energies,
2 4 2 3
and had the form n2 A= n0 Eg G2 h-=Eg and b2PA B= no Eg F2 h-=Eg
where A and B are constants; no is the linear refractive index; and Eg the bandgap
energy. The functions F 2 and G 2 are reproduced in Fig. 1.1a. These expressions
have been shown to provide a reasonable t to experimental measurements for a
wide range of materials [5, 6]. For the case of indirect-gap semiconductors the
situation is somewhat more complex but was analyzed by Dinu [8] whose results
are shown in Fig. 1.1b. In both these cases the dispersion of the nonlinear coef-
cients contain similar features. Firstly, as the frequency increases, the third order
nonlinearity becomes resonantly enhanced above its value at DC frequency as the
two-photon resonance close to half the band gap of the material is approached.
Once this resonance is passed, two photon absorption (F 2 , ()) grows and
maintains a high value all the way up to the band edge of the material. The
influence of the single photon resonance has a marked effect on the real part of the
nonlinearity and, in general, results in a reversal in the sign of the nonlinearity from
positive to negative. In the case of the direct gap semiconductors this is predicted to
occur at 0.7Eg whilst for indirect materials it is located within the bandgap of the
material at around 1.25Eg. In both these models 2PA = 0 when h < Eg/2, thus in
Fig. 1.1 a Dispersion functions for the nonlinear refraction (G 2 ) and two photon absorption (F
2 )
for direct gap materials (from [7]). b Similar relationships for n2() (=n2) and () (=2PA) for
indirect gap materials (from [8])
4 X. Gai et al.
this regime the FOM2PA becomes very large. Whilst this is true, one must remember
that the nonlinear polarization of the material contains higher order terms. For
example, the imaginary part of (5) gives rise to 3PA and this is not necessarily
small as, for example, has been demonstrated recently for silicon [9]. Thus, beyond
half its band-gap silicon has the intensity dependent FOM3PA mentioned above. In
fact as the frequency is progressively reduced we expect that higher and higher
orders of nonlinear absorption can appear. For the case of the direct gap materials,
the dispersion of the high order multi-photon absorption was predicted by Wherrett
[10] and consists of a series of similar curves to that shown for 2PA in Fig. 1.1a, but
displaced progressively lower frequencies with a cut-off frequencies corresponding
to h < Eg/3, <Eg/4. etc. To our knowledge no similar analysis exists for indirect
gap materials. Of course it is generally assumed that such higher order absorption
can be neglected but this may not always be the case as has been illustrated by
recent results in silicon, which shows that strong nonlinear absorption can exist at
wavelengths well into the mid infrared [9].
So far we have concentrated on the ultrafast Kerr nonlinearity, but intensity-
dependent changes in the phase of the light can also be due to other mechanisms.
Apart from the thermal effects, mentioned above, the most common of these is due
to the creation of free carriers that lead to a change of index as described by the
Drude model. Free carrier effects in semiconductors can be strong and have proved
very useful for all-optical switching. However, in passive materials such as silicon,
free carriers created by single or multi-photon absorption lead to strong free carrier
absorption (FCA) and this is generally detrimental. However, in active devices
based on IIIV semiconductors, the losses can be compensated using laser action
and, hence, the semiconductor optical amplier (SOA) has proved to be a very
effective structure for nonlinear optics. As outlined in more detail below, SOAs
main limitation arises because of the nite lifetime of the free carriers which leads
to a speed limitation.
Finally in the late 1990s, the so-called cascaded second order nonlinearity was
(re-)discovered and showed that a nonlinear phase shift could be generated using
second order nonlinear materials [11]. Implemented in PPLN this has led to very
efcient all-optical processors that are effective even at the single photon level.
Their general characteristics are described below.
In what follows we summarize the properties of materials and devices using
Kerr, cascaded and free carrier nonlinearities outlining the material and device
properties and some of their capabilities and limitations for all-optical signal
processing.
1.2 All-Optical Processing Using (2) Nonlinearities
In the late 1990s it was realized that an intensity-dependent phase shift that was
used for all-optical switching could also be obtained by cascading two second-order
((2)) frequency conversion processes, creating an effective third order nonlinearity
without nonlinear loss [11]. In this early work, cascading involved imperfectly
phase matched second harmonic generation and degenerate parametric down con-
version operating simultaneously. Because of the phase mismatch, power flowed
from the low frequency pump to its second harmonic and back again. In this
process the down-converted signal originating from the second harmonic accu-
mulated a small phase shift (positive or negative depending on the sign of the phase
mismatch) relative to the unconverted pump so when these combined coherently,
the phase of the pump was slightly modied. If the coherence length of the SHG/
DFG processes was small compared with the device length (the case of large phase
mismatch) the phase shift increased linearly with device length, and also linearly
with pump intensity mimicking the behavior of a (3) nonlinear material. Various
schemes for all-optical switching using cascading were reported [12].
The advent of WDM signicantly changed the emphasis of the research into all-
optical signal processing away from switching devices. For example, the challenge
in a WDM network was not, at least until recently, to de-multiplex signals at speeds
beyond the limits of electronics, but to transfer data from one to another wavelength
or to multi-cast data on many channels simultaneously. Conventionally such pro-
cesses used direct detection and electronic modulation of a wavelength shifted laser
or multiple lasers but this can become power-hungry and complex as channel counts
rise. An alternative and potentially better solution is to use nonlinear optics to
directly convert light at one wavelength to another using nonlinear mixing processes.
Thus, optical signal processing using (3) nonlinear optics became mostly based on
four-wave mixing (4WM) whilst in the case of (2) devices three-wave mixing
(3WM) was employed. Many of the wave mixing devices are agnostic to the data
format and can readily preserve the amplitude and phase of the incoming signal.
4WM and 3WM are shown schematically in Fig. 1.2. In the case of 4WM two
pump photons at frequency p mix with a signal photon at s to create an idler
photon at frequency i for which energy conservation requires 2p = s + i. In
this case p = s,i where s,i which means that all three frequencies lie
in the telecommunications band where there are readily available sources. Since
two photons at the pump are destroyed and new photons created at s, i, the signal
and idlers experience gain.
Fig. 1.2 a Schematic of 4WM in which two pump photons (p) mix with a signal photon (s) to
create an idler (i) such that 2p = s + i; b schematic of 3WM in which difference frequency
mixing between a single high frequency pump photon (p) and a signal (s) creates an idler (i)
symmetrically positioned around p/2 such that p = s + i; c cascaded 3WM where second
harmonic generation converts a pump at p to 2p which then undergoes 3WM with the signal to
create an idler
6 X. Gai et al.
The case of 3WM is quite similar, however, here a single pump photon decays
into signal and idler photons so that the energy conservation condition now reads
p = s + i. Again p = s,i but now s,i and hence a high frequency
pump is required that lies well outside the telecommunication band where sources
are generally less compatible with telecommunications technology. This difference
between the pump and signal/idler frequencies in 3WM would be a large disad-
vantage were it not possible to eliminate it using cascading. To achieve this a strong
pump at frequency p s,i in the telecommunications band is frequency doubled
in the (2) material to create a new second harmonic pump at 2p. Difference
frequency generation then allows this second harmonic pump to mix with the signal
to create a new idler spaced symmetrically around the original low frequency
pump. For this case the pump, signal and idler all exist in the telecommunications
band although it must be remembered that the process inevitably generates a new
high frequency pump at 2p which needs to optimally couple with the signal and
idler waves to achieve high overall conversion efciency. Just like in the 4WM
process, two pump photons are thus destroyed and their energy transferred to the
signal and idler exactly mimicking 4WM in a 3 material.
To achieve low operating powers, compatible with telecommunications systems,
3WM must take place in a non-centrosymmetric crystalline waveguide exhibiting
very strong (2) nonlinearity with perfect phase matching. The most common
material of choice has been periodically-poled lithium niobate (PPLN) for which
there are several well-established routes for waveguide fabrication. In addition, the
technology for electric eld poling of lithium niobate is well developed and
complex grating patterns can be engineered, for example, to increase the bandwidth
for phase matching. Of the approaches to waveguide fabrication that have been
developed, the most widely used is the so-called reverse proton exchange process
(RPE) [13, 14]. This involves, rstly, the preparation of a substrate using electric
eld poling to periodically invert the ferroelectric domains. This is then followed by
proton exchange and annealing steps that dene the waveguide channel creating as
annealed proton exchange (APE) waveguides [15, 16]. The losses, nonlinearity and
eld overlap between second harmonic pump and signal and idler elds are not
optimum in APE structures and so a second exchange process with a lithium-rich
melt is used to replace proton by lithium near the channel surface thus creating a
RPE waveguide. These waveguides have lower losses because the interacting elds
are pushed further from the top surface, and better overlap between the interacting
modes which also enhances the nonlinearity. RPE waveguides typically have losses
are 0.10.2 dB/cm and modes well-matched to standard SMF-28 optical ber which
reduces coupling loss [17]. Whilst RPE waveguides have been used in many
demonstrations of all-optical signal processing, they do suffer from some issues
with photorefractive damage. This can be alleviated by operation at elevated
temperatures or by using alternative materials and fabrication methods such as Ti
in-diffusion. A fuller description of the various fabrication approaches can be found
in the references [17].
An important parameter for any (2) waveguide is its normalized efciency cor.
For the cascaded process the output power at the idler can be written in the non-
depleted pump approximation as Pout = (corL2Pp/2)2Ps where Pp and Ps are the
pump and signal powers respectively. Typically cor 1 W1 cm2 and this implies
that if the losses are negligible, 100 % conversion can be obtained for a pump
power of only 80 mW in a 5 cm long device at low signal levels. In fact one of the
remarkable features of the devices based on cascaded nonlinearities is their high
dynamic range and ability to up-convert single signal photons to access high
detector sensitivity [18].
The second important parameter is the phase matching bandwidth for the
interaction. For DFG the conversion scales with sinc2(kL/2), where is the device
length and k = kp ks ki Kg is the phase mismatch with kp,s,i are the k-vectors
of the pump, signal and idler at frequencies p, s and i, respectively, and
Kg = 2/g with g being the grating period. Similarly for the SHG process
k = kp 2kp/2 Kg. In either case the output falls to half it maximum when
k 0.89/L. The DFG bandwidth substantially exceeds that for SHG and is
typically 70 nm FWHM for a 50 mm long device, however, for SHG the bandwidth
is only 0.2 nm for a similar device length. This rather narrow pump bandwidth can
be overcome either by engineering multiple gratings into the crystal to allow several
different pumps to be used simultaneously [19] or by using two tunable pump
beams [17].
PPLN based all-optical processors have proven be highly capable in applications
such as wavelength conversion [20]; dispersion compensation via phase conjuga-
tion [21]; digital signal processing including header recognition, time-slot inter-
change [22], etc. For an extensive review of their capabilities the reader is referred
to recent reviews by Langrock et al. [17] and Willner et al. [23]. In terms of
limitation there are three of consequence. The rst arises from group velocity
dispersion (GVD) because of the need to convert the pump in the telecom band near
1550 nm to its harmonic around 775 nm. This large frequency difference results in
GVD of 0.3 ps/mm which means that 50 mm long devices are restricted to operate
with pulses >15 ps in duration. Whilst techniques have been proposed to circum-
vent this issue, such as the introduction of delays lines to retime the fundamental
and second harmonic waves [24], the large GVD makes it more difcult to process
signals at very high rates. Nevertheless optical processing at 640 Gbit/s has been
reported [25]. A second limitation arises from the single polarization nature of the
interaction, and generally requires the use of polarization diversity techniques.
Finally, in a photonic world increasingly dominated by silicon waveguide devices,
PPLN remains a material that is difcult to integrate to create highly functional
circuits and because of the interactions are phase matched require temperature
stabilization. Countering these are the ability of PPLN devices to operate with low
noise; minimal cross talk, high bandwidth and high efciency and no chirp making
them a powerful platform for all-optical signal processing.
8 X. Gai et al.
1.3 All-Optical Processing Using (3) Nonlinearities
3WM can achieve efcient all-optical processing but the use of the (2) nonlinearity
requiring a non-centro-symmetric material limits its applicability to only a few
materials and these are generally incompatible with current CMOS processing
technology. As a result 4WM using the (3) nonlinearity that exists in all materials
has started to dominate all-optical processing. In this case, the pump signal and idler
all lie in the telecommunication band, group velocity dispersion can be small, and
phase matching can be achieved by engineering the structure of the waveguide so
that the dispersion is anomalous. A large number of materials can be used for FWM
and combined with the simply device geometry and, in many cases, compatibility
with CMOS processing, allow all-optical devices to be integrated at low cost. In this
section we describe some of the materials dependent factors that influence the
efciency and bandwidth for FWM.
For waveguides that utilize the (3) nonlinearity, the nonlinear phase change is
generally written in the form nl = PL, where P is the laser power; L is the
propagation length; and is the waveguide nonlinear parameter = kon2/Aeff, with
ko = 2/ is the free space wavenumber; and Aeff is the effective area of the
waveguide mode. In a lossless waveguide, the 4WM conversion efciency from
signal to idler in the non-depleted pump approximation, can be written
4WM = (nl)2 = (PL)2, and is directly determined by nonlinear phase change. It is
worth noting here that compared with switching devices where nl needs to be
large (>), FWM can be quite efcient for small values of nl 0.3. The nonlinear
parameter is important for any (3) waveguide because a high value minimizes
P and L, and small, low power devices are essential in most applications. In practice
P is often limited by energy constraints in the system, and L has a maximum value
determined by the loss of the waveguide. In a lossy waveguide, L is replaced by an
effective length written Leff = (1 exp(L))/ where is the linear loss coefcient.
Because of these constraints, as high a value of as possible is always preferable
and this motivates the use of materials with high n2 and waveguides with small Aeff.
The phase matching condition for 4WM is generally written 4P < k 22 < 0,
where k = 2kp ks ki is the phase mismatch with kp,s,i are the k-vectors of the pump,
signal and idler at frequencies p,s,i respectively, = |p s,i| is the frequency
difference between pump and signal (or idler), and 2 is the second order dispersion of
waveguide. The dispersion of waveguide () = k() = 2neff/ can be expanded as a
Taylor series as (0 + ) = 0 + 1 + 22/2 + 33/6 + , where m = dm/
dm (m=1, 2,) at 0; neff is the waveguide effective index. The rst order of the
dispersion 1 = 1/vg = ng/c0 where vg is the group velocity, ng is the group index and c0 is
the light speed in free space. The second orders of the dispersion 2 = d1/d = 2D/
2c0, where D is the group velocity dispersion (GVD) of waveguide. 1 and 2 are the
most critical parameters for all-optical processing because the pulse envelope prop-
agates at vg and pulse broadening is relating to GVD.
In order to achieve phase matching for FWM, the waveguide dispersion must be
anomalous, that is 2 < 0 and when this is achieved, power in the signal and idler
waves grows exponentially at the expense of the pump. However, most materials
exhibit normal dispersion, 2 > 0, at telecommunication frequencies and it is,
therefore, necessary to engineer the waveguide structure to produce sufcient
anomalous waveguide dispersion so that the normal material dispersion is com-
pensated. This means that for any particular material, only specic waveguide
designs can be used and these determine the minimum mode area, Aeff, and hence
the of the waveguide. Figure 1.3 is an example of such dispersion engineering for
a 630 nm wide chalcogenide nanowire and shows the effect of changing the
waveguide height on the TM mode dispersion [26]. In this example, the waveguide
has to be >450 nm thick in order to achieve anomalous dispersion at 1550 nm. In
other materials, like crystalline silicon, anomalous dispersion occurs in smaller
structures potentially leading to higher . Thus material dispersion becomes the
dominant parameter that ultimately determines the achievable value of at a par-
ticular wavelength.
In the phase matching relation, represents the bandwidth of 4WM and it is
determined by 2 and P. 2 close to zero and a large P will support 4WM over a
very wide bandwidth. This is important for many devices required for telecom-
munications and allows 4WM to span the whole S- C- and L-bands. As a result, the
zero-dispersion is a target for dispersion engineering as well as a large value.
The phase-matching condition of 4WM, in fact, demonstrates the relation
between the nonlinear phase change and the dispersive phase change. However, this
conventional phase matching relation ignores the influence of other nonlinearities,
notably Raman scattering. In fact Raman scattering can cause an additional phase
change that originates from the real part of the Fourier transform of the Raman
response function Re[hR()] [27, 29]. Thus, in materials where the bandwidth for
FWM extends as far as the Raman bands, the phase matching relation must be
modied to include this additional phase term and thus reads 4P4PfR(Re[hR()
1]) < k 22 < 0, where fR is the fractional Raman factor which normally
range from 0.1 to 0.2 depending on material. When the Re[hR()] 1, the
Fig. 1.3 The group velocity

dispersion (GVD) of
Ge11.5As24Se64.5 nanowire as
a function of lm thickness
and wavelength. The
waveguide width is xed at
630 nm [26]
10 X. Gai et al.
Fig. 1.4 a The imaginary part of Fourier transform of the Raman response function of
Ge11.5As24Se64.5, As2S3 and SiO2 [27]. b The real part of Fourier transform of the Raman response
function of Ge11.5As24Se64.5, As2S3 and SiO2 [27]. c The experimental result of Raman modulated
4WM in a Ge11.5As24Se64.5 chalcogenide nanowire [28]
inequality becomes 4P < k 22 < 0, and Raman scattering does not

contribute to the phase change. When Re[hR()] > 1, it causes a positive phase
change which broadens the phase-matching condition increases and the gain for
4WM. On the other hand, if Re[hR()] < 1, then the Raman phase change is
negative, the phase-matching condition narrows and the 4WM gain is reduced.
Figure 1.4a, b shows the imaginary and real parts of hR() for SiO2, As2S3 and
Ge11.5As24Se64.5 glasses [27]. Typically Re[hR()] remains 1 for small detuning,
and then exhibits a resonant behavior to increase up several times right after the
Raman peak of Im[hR()] before dropping sharply. At large detuning, Re
[hR()] 0, but never returns to its original value of unity. As a result, the
bandwidth of 4WM is restricted by the spectrum of Raman scattering even if 2 is
near zero and is large. Figure 1.4c shows the effect of Re[hR()] on 4WM for a
Ge11.5As24Se64.5 chalcogenide nanowire [28]. When the Re[hR()] recovers to zero
at large detuning, the gain is less than 70 % of that Re[hR()] 1. The maximum
bandwidth for high gain 4WM can be predicted using the Raman detuning fre-
quency for the material, and is 130 nm for Ge11.5As24Se64.5; 180 nm for As2S3;
190 nm for SiO2; 220 nm for amorphous Si; and over 250 nm for crystalline Si
assuming a pump at 1550 nm.
1.3.1 Properties of (3) Materials and Devices
Very many materials and device structures have been considered for (3) nonlinear
optics although only a few of these have actually been used in demonstrations of
optical signal processing. One of the most successful has been silica-based highly
nonlinear optical ber (HLNF) that has been used for parametric signal processing
for more than two decades [30]. Despite its inherent compatibility with ber optic
systems, the small material nonlinearity implies large device lengths and additionally
the ber geometry makes integration of multiple structures for signal processing
impractical. Several alternative media with much larger material nonlinearity have
emerged including bismuth oxide glass; crystalline and amorphous silicon; silicon
nitride; AlGaAs; chalcogenide glasses; and high-index doped silica-based glasses. In
this section we summarize the characteristics of a few of these materials and discuss
device properties relevant to a few on-chip platforms that use the nonlinear Kerr
effect and 4WM based all-optical processing. Because of the large amount of work
based on the following materials we focus on crystalline silicon (cSi); hydroge-
nated amorphous silicon (aSi:H); and silicon nitride that are all compatible to
complementary metal oxide semiconductor (CMOS) processes and the somewhat
less CMOS-compatible chalcogenide glasses.
1.3.1.1 Crystalline Silicon
In the past decade a new application for silicon has emerged as a material for
photonics. The main driver for this has been the interconnect bottleneck now facing
electronic processors. The major advantage of silicon photonic devices is they can
leverage the precision of advanced CMOS processing technology. However, in
order to create a small-scale communication network, various functional photonic
components must be integrated onto the silicon platform. This has resulted in some
remarkable developments including germanium light emitters [31] and germanium
photo-detector [32] integrated onto silicon chips as well as high-speed optical
modulators [33]. There has also been substantial interest in all-optical signal pro-
cessing utilizing the (3) nonlinearity of Si.
Si is transparent from 1100 to 7000 nm and linear absorption can, therefore, be
low across the whole optical communications bands between 1260 and 1675 nm,
but two-photon absorption is large. This results in a relatively low FOM2PA of *0.4
at 1550 nm corresponding to a Kerr nonlinear coefcient n2 = 4.4 1014 cm2/W
and 2PA = 8.4 1010 cm/W [34]. This means that Si is, perhaps, not the best
material for nonlinear photonics but since it can be used to produce complex
circuits at low-cost in high-volume there has been a large incentive to explore its
use in all-optical signal processing. Si also has excellent thermal conductivity and
high damage resistance and because of its high refractive index can conne light
into a very small mode and this can lead to exceptionally large values of . In
addition, waveguide designs that lead to near optimal connement also correspond
to those which achieve anomalous dispersion.
The silicon-on-insulator (SOI) platform has become the foundation of silicon
photonics and comprises of 220 nm thick top silicon layer sitting on thermally
oxidized (12 m SiO2 layer) silicon substrate. Figure 1.1a shows a schematic of a
typical silicon photonic nanowire, and the optical eld distribution in the funda-
mental TE mode. The large refractive index contrast between Si (n = 3.45) and SiO2
(n = 1.45) or air (n = 1) leads to a strong light connement which makes it possible
to scale down the size of the waveguides mode to approximately 0.1 m2. Com-
bining high n2 and small Aeff yields an extremely large nonlinear parameter of
300 W1 m1 [35]. On the other hand this rectangular geometry makes much lower
effective index in TM so that the mode area is a few times larger than that of TE.
12 X. Gai et al.
Silicon photonic wire waveguides are typically fabricated by plasma etching

with electron-beam or deep-ultraviolet (DUV) lithography used to dene the sub-
micron pattern. Nowadays, propagation losses of 12 dB/cm [36] are routinely
achieved mostly limited by light scattering from the sidewalls but this is low
enough since the devices are short (a few cm or less). A more difcult issue has
been light coupling into the photonic wire due to a large mode mismatch between
nanoscale silicon waveguides and bers. Two coupling schemes are commonly
employed to overcome this issue are to use surface grating couplers [36] and Si
inverse tapers [37] for which the coupling efciency reaches around 4070 %.
In spite of its many advantages over other media, silicon-based all-optical signal
processing inevitably suffers from nonlinear loss and patterning effects resulting
from large TPA and free carrier absorption (FCA). The lifetime of free carriers
generated by two-photon absorption is in the range of several hundred ps to several
hundred ns [38] and slow carrier dynamics can limit the speed of signal processing.
Several approaches have been proposed to mitigate the issue as shown in Fig. 1.5b
d. The main idea is that direct light-silicon interaction can be avoided by engi-
neering the Si waveguide structure. The rst is to employ a thin strip waveguide and
cover it with a material possessing both a large nonlinear Kerr coefcient (n2) and a
small nonlinear loss (2PA) [39]. In this case the structure makes TM mode extend
far into the cladding material so that most of the nonlinear effect can be due to the
upper cladding. Figure 1.5c shows a so-called slotted waveguide where trenches
several tens to hundreds nanometer wide are formed along the middle of Si
waveguide, and this slot is lled with another nonlinear medium. With an appro-
priate design the light intensity in the slot is enhanced a few times relative to the
unstructured waveguide [39]. Photonic crystal waveguides have also been dem-
onstrated to further increase nonlinearity by slowing-down the light speed [40].
This slow-light effect can boost the light-matter interaction proportional to the
group index squared (c/g)2.
Fig. 1.5 Silicon waveguidesilicon waveguide structures and their respective electric eld
distributions. a Strip waveguide using silicon nonlinearity in core. b Strip waveguide using cover
nonlinearities. c Slot waveguide using nonlinearities in the slot. d Slot slow-light waveguide [35]
To conclude, it is worthwhile noting that in spite of the limitations of silicon, two

recent reports clearly show that these can be overcome with careful attention to
device design. Forster et al. [41] achieved peak conversion efciencies >10 % and
conversion bandwidths >150 nm in carefully dispersion-engineered Si nanowire
waveguides. For this they used rather fat and thick Si core (300 nm tall and 500 nm
wide) and operated with the TM mode. The work of Denmark Technical University
group is also remarkable in that ultra-high bit rate signals were processed in a
nominal Si nanowire without incurring TPA and associated FCA problems [42].
They demonstrated the wavelength conversion of 640 Gbit/s signal by low-power
4WM in a 3.6 mm long silicon waveguide with a switching energy of *110 fJ/bit,
which is low enough to reduce nonlinear absorption.
1.3.1.2 Hydrogenated Amorphous Silicon (aSi:H)
Over the years aSi:H has been extensively deployed in applications from photo-
voltaics to flat-panel displays but its applications in lightwave technology is just
evolving. As a nonlinear medium for all-optical signal processing, amorphous
silicon outperforms crystalline silicon in some respects. First of all, this material
retains CMOS-compatibility so that low-cost mass manufacturing is available.
Another attractive feature of aSi:H comes from the fact that it can be deposited at
low temperature (below 400 C) on almost any substrate including glass, metal and
even plastic, whilst crystalline silicon (SOI) does not offer this flexibility. More-
over, the low temperature deposition allows back-end integration of aSi:H pho-
tonic components on pre-processed CMOS electronic chips without any damage to
the underlying metal routing wires [43]. In addition it also proves to be a highly
promising nonlinear material for power efcient photonic devices due to its ultra-
high optical nonlinearity and low nonlinear loss relative to crystalline Si.
Plasma-enhanced chemical vapour deposition (PECVD) is a well-established
technique for depositing high quality amorphous silicon-hydrogen alloy (*10 at.%
H) at low temperature. Silane (SiH4) is the dominant precursor gas, mixed with
hydrogen, helium, or argon as a diluent. If amorphous silicon is deposited by
sputtering it has mid-bandgap absorption in the near infrared due to dangling bonds.
These Si dangling bonds, however, can be saturated by incorporating hydrogen
during PECVD, resulting in the low optical attenuation below 1 dB/cm [44]. The
fabrication of aSi:H photonic nanowaveguides is identical to that of SOI; i.e.
optical or electron-beam lithography followed by plasma etching and SiO2 PECVD
for cladding. aSi:H nanowires have similar or slightly smaller dimension com-
pared with crystalline Si because of its higher index. The linear propagation losses
achieved in aSi:H waveguides lies in the same range as cSi; that is a few dB/cm,
and even *1 dB/cm has been achieved by adopting a delicate thickness control
technique [45].
The nonlinear characteristics of the material are favourable with a Kerr non-
linearity (n2) reported to be several or up to 10 times [46] higher than that of cSi. It
is debatable as to whether TPA in aSi:H is a still the dominant nonlinear loss
14 X. Gai et al.
mechanism in the telecommunication band, since TPA should vanish since the
photon energy is less than the half the bandgap (Eg/2 * 0.85 eV) of the medium.
Nevertheless, it is generally considered to have similar level of nonlinear absorption
to cSi. Because of the high nonlinearity, can reach 3000 W1 m1 in a 500 nm
wide, 200 nm high nanowire [47]. Utilizing this extremely high nonlinear parameter
in aSi:H, Wang et al. [47] demonstrated wavelength conversion through nonlinear
parametric processes with maximum conversion efciency of 13 dB using mere
15 mW of pump peak power. In addition on-chip parametric amplication with
26.5 dB gain was attained in aSi:H photonic wire at telecommunication wave-
lengths [48].
A key concern of hydrogenated amorphous silicon devices has been its lack of
stability. Kuyken et al. [49] observed a decrease of parametric amplication with
the time of exposure to pump light. They asserted that this results from material
degradation and has the same origin to the Staebler-Wronski effect [49], which is
well known to the aSi solar cell community. Most recent results, however, dem-
onstrated that deterioration in amorphous silicon may not be an issue. The large
scatter in n2, 2PA (or 3PA), FOM2PA, , and free carrier lifetime in the literature is
another nagging problem. For instance, the measured 2PA spans from 41 1010 to
2.5 1010 cm/W. The origin of this large variation remains an open question, but
it is a general thought to be due to differences in the lm deposition process. Even
so, aSi:H is quite a promising platform for ultra-compact all-optical signal pro-
cessing chips operating at low power.
Silicon nitride (SiN): This is another CMOS-compatible material that has
recently emerged as a candidate for nonlinear photonics. Since SiN can be
deposited from gas precursors in a standard silicon processing environment, it
benets from flexibility in deposition parameters which produce different lm
characteristics. Despite n2 of SiN (n2 = 2.5 1015 cm2/W at 1.55 m) being more
than an order of magnitude lower than cSi, the most signicant advantage of SiN
over crystalline and amorphous Si comes from the absence of 2PA or multi-photon
absorptions and associated FCA due to its wide bandgap (Eg * 5 eV) [50].
Low-pressure chemical vapour deposition (LPCVD) at high temperature
(*800 C) is the preferred method to grow low loss SiN layer and leads to minimal
hydrogen contamination in the lm. Using this method, propagation losses of
0.5 dB/cm loss was obtained in a SiN waveguide with 1 m2 cross section [51]. In
order to facilitate back-end integration with pre-existing electronic circuits, how-
ever, PECVD below 400 C was also demonstrated to produce SiN waveguides
with only *1 dB/cm propagation losses. Here, the absorption in the near IR was
controlled by reducing the number of hydrogen bonds (SiH and NH) in the
grown SiN by employing techniques such as adding helium into the plasma [52]
and replacing NH3 with a N2 precursor [53].
The group at Cornell University [51] demonstrated on-chip multiple wavelength
source exploiting 4FM optical parametric oscillation in a SiN micro-ring (Fig. 1.6).
They used a single pump laser tuned to the resonance of the ring, then *1000 times
higher light power circulating in the ring which leads to cascaded 4WM, and
allowing narrow multi-wavelength lines.
Fig. 1.6 On-chip optical parametric oscillator in SiN waveguide. a A single pump laser tuned to
the resonance of an integrated SiN microring allows the generation of numerous narrow linewidth
sources at precisely dened wavelengths. b A scanning electron micrograph of a SiN microring
resonator coupled to a bus waveguide [51]
Fig. 1.7 Minimum pump

power required to achieve a
given CE for the different
materials [54]
Diaz et al. [54] reported a systematic comparison of eight nonlinear media for
parametric all-optical signal processing using degenerate 4WM wavelength con-
version. Figure 1.7 represents the minimum pump power required to attain a given
conversion efciency (CE). Here CE is the ratio of the idler power at the waveguide
output to the signal power at its input. Crystalline and amorphous silicon can
operate at low pump power because of their very high , but their CE is limited to
around 10 % due to TPA and FCA. Whilst the performance of SiN is inferior to
16 X. Gai et al.
cSi and aSi because of its smaller nonlinear parameter, it can handle much higher
pump power without suffering nonlinear losses. An additional difculty, however,
is that the large mode area needed for dispersion engineering which is around 1 m2
results in the footprint of SiN being two orders higher than the equivalent devices
made in cSi and aSi:H and this makes it difcult to achieve compact devices.
Chalcogenide glass: Chalcogenide glasses have attracted a lot of interest for all-
optical signal processing. These are amorphous, highly nonlinear materials con-
taining the chalcogen elements S, Se and Te covalently bonded with glass forming
materials such as Ge, As, P, Sb or Si. Chalcogenide glasses provide low optical loss
in the telecommunication bands (1310 and 1550 nm) and transmission out to
beyond 8 m in the infrared. These glasses have high linear refractive indices (2.0
3.0 at 1550 nm) which leads to small mode volumes in dispersion-engineered
nanowires and a Kerr nonlinear index similar to cSi (a few hundred silica) [55,
56]. Most importantly they generally have negligible 2PA which leads to
FOM2PA > 100 [5759]. In most chalcogenide glasses, there are no free carrier
effects and this provides a distinct advantage compared with the semiconductors.
As2S3 is one of the most well known chalcogenide glasses with a linear index of
2.43 at 1550 nm, a nonlinear index over 120 times silica and FOM2PA over 320 and
a glass transition temperature over 170 C. The rst reports demonstrating the
efcacy of As2S3 ber for all-optical processing were produced by Asobe [60].
More recently a large number of ultra-fast nonlinear devices have been demon-
strated using rib waveguides made from this glass including a signal regenerator
[61]; RF spectrum analyser [62, 63]; dispersion compensator using mid-span
spectral inversion [64]; TDM demultiplexers [65]; wavelength convertor [66];
parametric amplier [29, 67]; as well as a supercontimuum source [68].
Although, As2S3 glass has, therefore, exhibited some good properties for all-
optical processing, even better chalcogenide materials would be of advantage
particularly if higher nonlinearity and an absence of photosensitivity can be
obtained [69]. High nonlinear index and low TPA has been demonstrated in Ge
AsSe, GeAsSSe and AsSSe compositions [55, 56, 58]. Germanium rich
compositions have reported nonlinearities between 200 and 900 times that of fused
silica. It was found that the optical properties of the glasses can be improved by ne
tuning the composition in the GeAsSe system.
The Ge11.5As24Se64.5 glass is one of a small family of GeAsSe glass com-
positions that show high nonlinearity, particularly high thermal and optical stability
when exposed to near band edge light and good lm forming properties. It is known
that the physical properties of ternary chalcogenides vary signicantly as a function
of their chemical composition and mean coordination number (MCN = the sum of
the products of the valency times the atomic abundance of the constituent atoms)
and that MCN can be used to categorize the basic properties of the glass network. In
the case of GeAsSe glasses, it has been found that their linear and nonlinear
refractive indices; optical losses; elastic properties, etc., all vary strongly with
MCN. Ge11.5As24Se64.5, for which MCN = 2.47, lies in the so-called intermediate
phase (IP) which lies between the floppy and stressed-rigid glass networks.
Of particular signicance is that lms produced by thermal evaporation in this
region have similar properties to those of bulk glassesan unusual behaviour for
lms made from ternary chalcogenides.
Compared with As2S3 glass, which has been the workhorse for all-optical
devices to date, Ge11.5As24Se64.5 has a higher linear refractive index (2.65 com-
pared with 2.43) which leads to better mode connement and, as would be expected
from Millers rule, a higher nonlinear index, n2 (8.6 1014 cm2/W c.f. 3 1014
cm2/W for As2S3 at 1500 nm) [70]. No TPA could be detected during z-scan
measurements made on bulk Ge11.5As24Se64.5 samples, however, the FOM has been
found to be 60 from measurements of the power-dependence of the transmission
for Ge11.5As24Se64.5 rib waveguides, as shown in Fig. 1.8a [71]. Ge11.5As24Se64.5
nanowires with 135 W1 m1 have been reported with losses around 1.5
2.5 dB/cm [26, 28]. 4WM and SC were demonstrated in such nanowires as well as
polarization independent properties using the structure shown in Fig. 1.8b [28].
These nanowires were also used to demonstrate correlated quantum photon pair
generation [27, 72]. High Q (>700,000) photonic crystal cavities are also made
from this material for optical switching application, as shown in Fig. 1.9 [73].
Fig. 1.8 a The As2S3 rib waveguide. b The polarization independent Ge11.5As24Se64.5 nanowires
Fig. 1.9 SEM images of an end coupled Ge11.5 photonic crystal. a, b The prole from top surface.
c Cross section of holes cut by FIB and lled with Pt for imaging the side walls
18 X. Gai et al.
1.4 All-Optical Processing in Semiconductor Optical

Ampliers
Semiconductor optical ampliers (SOAs) have a long history in all-optical pro-

cessing with their popularity stemming from the range and strength of nonlinear
effects available; from the early commercial availability of bre pigtailed devices;
from the capacity for on chip integration of multiple functions/devices; and from the
potential for low power operation with net conversion gain. Such is the diversity of
architectures that it is beyond the scope of this summary to cover these in their
entirety and the reader is instead referred to some of the excellent recent review
articles and books on the topic for the wider view, e.g. [7481]. Here the types of
nonlinear effects will be briefly reviewed, the limitations facing SOAs examined
along with some of the proposed remedies and the potential limits, and the best and
latest achievements briefly summarized.
1.4.1 Types and Origins of Nonlinear Effects

in Semiconductor Optical Ampliers
As electrically pumped devices, SOAs are dominated by carrier density driven

nonlinearities. These result in effects operating on the amplitude, phase and
polarization of optical signals. The rst and most obvious effect is due to the signal
power dependence of the SOA gain which saturates as a result of carrier depletion
as the input power is raised. Thus, cross gain modulation (XGM) occurs where an
intense pulse modulates a co-propagating CW beam at a different wavelength. This
process is commonly used in wavelength conversion. Gain saturation is also used in
optical limiting where the SOA output power becomes essentially independent of
the input power, thereby reducing amplitude noise on high signal levels. This is
particularly useful for non-return to zero (NRZ) modulation.
The Kramers-Kronig relation means gain or absorption changes are always
accompanied by changes in refractive index and also therefore phase change in the
propagating signals. Refractive index power dependence of course manifests as a 3
effect. Ascribing a precise value to the nonlinearity is troublesome due to the
differing contributing parts [82], but to give some comparative indication of the
effective size of n2, values from 2 1012 up to *1 109 cm2/W have been
reported, e.g. [8385]. However the response time of the electron plasma is not as
fast as the response of bound electron states in glasses, for example, leading to
speed limitations as will be discussed shortly. Cross phase modulation (XPM)
between a strong pulsed pump and a weaker CW probe can also be exploited in
SOAs for wavelength conversion by placing the SOA inside a Mach-Zehnder or
other interferometer.
Nonlinear polarisation rotation (NPR) in SOAs has also been increasingly har-
nessed for high speed nonlinear processing. This results from a combination of the
different gain saturation behaviours of the TE and TM modes, and also from the
intrinsic waveguide birefringence in the SOA (with both geometrical and stain
components) which can also be modulated by changes in the carrier density. Thus a
signal launched into both polarization states sees a waveplate with an effective
input power dependent retardation, and this can be used for all optical processing.
1.4.2 Impairments in SOA Devices
1.4.2.1 Carrier Induced Switching Speed Limitations
The strengths of SOAs for nonlinear processing stem from the strong carrier-based
nonlinearity in the devices, but this also leads to their main shortcoming. A major
driver for all-optical signal processing is speed, but the relaxation times of carrier-
based nonlinearities are long compared with those due to bound electron (Kerr)
nonlinearities in dielectrics. This is the most signicant impairment for SOA based
devices and so will be explored in some detail.
The dynamics of the gain compression and its associated phase shift has been
studied by quite a number of groups, e.g. [83, 86101] with good summaries of
the varying contributing processes and their distinct timescales being available e.g.
[75, 76, 102]. It should also be noted that the SOA gain region structure (bulk vs.
Quantum well (QW) vs. Quantum dot (QD) [103]) also strongly influences the
relative contributions of these components. The recovery time is also affected by
the input power, the polarization, the level of bias current applied to the SOA, the
dimensions and design of the active region, and the pump wavelength, e.g. [92, 94,
97101]. This makes it hard to compare recovery times in any absolute sense, but
nonetheless some broad generalizations of structural impacts, and mitigating
approaches is in order. It is also important to note that the phase recovery has
different characteristics to the gain recovery.
Hall et al. [83], undertook one of the most comprehensive earlier studies at
1550 nm in bulk and strained multi-quantum well devices, showing that carrier
injection and carrier heating effects dominated the dynamic recovery of the gain and
phase. Under the conditions studied in their pump-probe experiment, the QW
device exhibited faster gain recovery than the bulk albeit with a small but long lived
tail, but broadly similar phase recovery to the bulk device, however at about one
third the amplitude. Responses from the devices studied are reproduced from their
paper below in Fig. 1.10.
QD SOAs exhibit a different response again with very rapid gain recovery but
rather slower and smaller phase response as shown in Fig. 1.11. The reason for the
more rapid gain recovery in QD devices is considered to be the action of the wetting
layer used to grow the dots as a charge reservoir that feeds the dots locally and
therefore quickly. The wetting layer itself can also deplete but this has to rell via
normal interband mechanisms and so displays the characteristic long time constant
(visible as a very small and long exponential recovery tail in Fig. 1.11).
20 X. Gai et al.
Fig. 1.10 Bulk gain left and phase underneath, MQW to the right from [83]
Fig. 1.11 Gain and phase

recovery of typical QD SOA
(from [104])
Taking the published results as a whole, the obvious question is how short can
the recovery times go? As noted earlier operating parameters influence the recovery
times as was recently graphically illustrated by Cleary et al. [75, 105]. By opti-
mizing the pump and probe wavelengths relative to the SOA gain peak, they
achieved 1090 % gain recovery times from 35 ps in an MQW SOA, up to 7
improvement over some of the unoptimized conditions. The phase response was
improved by factors up to 2. QD SOAs have also demonstrated similar gain
recovery times [104, 106]. These results suggest bit rates of 3500 Gbit/s should be
possible at least using XGM effects.
There are other effects that can also be harnessed in the quest for ultimate speed.
To improve pure carrier recovery these include doping to reduce the carrier lifetime
in the bulk/feeding zones, exploiting tunnelling effects in QW and QD structures
[91], the use of pulsed or CW optical holding/reset beams [86, 107, 108], and
increasing the energy gap of QDs to the wetting layer [97] amongst others. Exactly
how far the intrinsic recovery time can be pushed is not yet clear, but modelling has
shown that speeds of 1 Tbit/s appear possible in some devices [93].
Sub-picosecond phase response times without long lived tails have also been
reported both experimentally and theoretically in MQW SOAs and have been used
in NPR based wavelength conversion experiments under carefully tailored low
current injection regimes where two photon absorption dominates [109111].
However the major issue with this mode of operation is the large pulse energy
required for switching (*10 pJ meaning an average power at the SOA input of 5 W
for a 1 Tbit/s PRBS RZ stream!) and the observation that the output pulses were
broadened considerably (nonetheless at 500 fs these are still the shortest pulses ever
wavelength translated in an SOA).
In addition to directly attacking the carrier recovery, architectural design can also
aid in improving the overall system response time. To cite a single example from
many, in XGM based wavelength conversion the nature of the phase response can
also be exploited as it causes the long recovery tail to be red shifted on the probe
and so a blue shifted bandpass lter can remove it, e.g. [112]. This effect was used
to demonstrate XGM based wavelength conversion at 640 Gbit/s SOAs with a
carrier recovery time of *60 ps [113]. A wide variety of other architectures have
also been developed to enhance overall switching speed as are discussed elsewhere,
e.g. [7481].
The discussion above relates to nonlinear devices utilising the gain or XPM
effects, but the Kerr effect in SOAs is large enough to enable Four Wave Mixing
(FWM) based processing. There are two distinct scenarios for FWM operation, one
where intense CW pumping is used and the pulsed input signal is relatively weak
(often for optical phase conjugation), and the other where the input pulsed signal is
strong and the CW probe used to generate the wavelength shifted idler is weak. In
the latter case all the recovery processes and timescales discussed above clearly
apply and this scenario has also been analysed by several authors for interactions
between pulses, e.g. [88, 114, 115]. In the former case, the recovery processes are
strongly suppressed due to the strong CW pump leading to the question of how fast
a signal can be accurately phase conjugated or wavelength translated by this
process?
Surprisingly there seem to be no analyses of the ultimate speeds achievable,
though Kikcuhi et al. [116] Summereld et al. [117] and Kim et al. [118] recog-
nized that the ratio of CW pump power to signal power is critical and that a trade off
exists with conversion efciency (due to carrier lifetime effects) and SNR for bulk
SOAs. In QD SOAs different and less stringent restrictions apply to the conversion
efciency e.g. [88, 103] but the trade off still exists. Experimentally, Kikuchi et al.
[119] demonstrated phase conjugation of a 1.6 ps pulse train in an SOA cascade and
successful transmission through 40 km of standard single mode bre limited only
by third-order bre dispersion differences resulting from the wavelength translation.
Taken together this suggests that the potential for Tbit/s CW pumped FWM
wavelength translation and OPC in SOAs are good.
22 X. Gai et al.
1.4.2.2 Other Limitations and Impairments of SOA Nonlinear Elements
Speed aside, there are other potential limitations in using SOAs as nonlinear ele-
ments. First and foremost all the nonlinear effects in an SOA except the Kerr effect
require amplitude modulation. Phase, subcarrier, and polarization encoding are
increasingly important in todays communication systems and amplitude modulation
is no longer implicit. Where modulation formats such as NRZ DQPSK are employed,
the signal has no amplitude modulation and XGM, XPM, and NPR cannot be har-
nessed for nonlinear signal processing. FWM however is a format transparent effect
and remains highly useful for wavelength translation and phase conjugation.
Another obvious issue in using a device with gain is the linear addition of
amplied spontaneous emission noise and its effect on optical signal to noise ratio,
but also the less well studied nonlinear interactions between noise and the signal.
OSNR is well understood and can be engineered and optimized with well proven
approaches. Nonlinear noise based interactions in SOAs have been studied by
numerous groups but most recently and with the least approximation by Connelly
et al. [120] for phase encoded signals in the context of line ampliers operating in
the saturated regime (i.e. as nonlinear elements are). However further study is
needed to clarify whether the effects of the various nonlinear noise mixing pro-
cesses in operation present a limiting factor to high speed data processing.
A further limitation often considered to limit the applicability of SOAs is
multichannel crosstalk. This relates to the imposition of the modulation of other
channels onto the channel under consideration by XGM, XPM, and NPR as a
channel ensemble passes through an SOA. In general this is an issue for devices
relying on these effects as the modulated signal is usually the pump and so it is
rather difcult to suppress the crosstalk as the nonlinear effect being exploited relies
upon it. However it can sometimes be managed enabling multichannel operation
with acceptable performance, e.g. [121].
There is however one potential exception to this which exploits the properties of
QD SOAs. QD SOAs have a range of dot sizes, each size amplifying a reasonably
narrow portion of the optical spectrum through the homogenous broadening of the
theoretically atom like DOS. Thus if the signals are sufciently separated in
wavelength that they each address different QD cohorts, then they will not impact
on each others gain to rst order [103] (at a more detailed level they are all fed by
the same wetting layer which may deplete based on the total instantaneous power,
this however as seen previously is a lower level effect in QD SOAs with appropriate
design and operating conditions).
Multichannel Crosstalk is also an issue in devices employing FWM. Here all the
same XGM mediated effects can apply if the powers are high enough, but in addi-
tion, there are also multiple interchannel FWM products that can fall inside the
desired signal bandwidth [117]. Whilst certain optimisations can minimise this, the
effects cannot be eliminated completely. Multichannel devices usually operate with a
strong CW pump that suppresses the gain compression based effects, and early work
demonstrated that even for purely intensity modulated data that the effects were
acceptable for four channel conversion [122]. With NRZ phase modulated data the
gain compression effects are absent leaving only the intermixing effects, and four
channel NRZ DQPSK modulated signals were wavelength converted by FWM in a
QD SOA with <4 dB penalties from intermixing [123]. Much further work however
remains to optimise the systems architecture to attain higher channel counts.
1.4.3 Current State of the Art and Future Prospects for SOA
Nonlinear Processing
Whilst the range of applications and therefore relevant results is too broad to
summarise meaningfully here, there are a number of key achievements for SOA
based nonlinear devices related to speed, conversion range, and conversion ef-
ciency which will now be briefly discussed. Taking speed rst, Table 1.1 gives an
overview of some of the fastest demonstrations using both pulses and full data
streams.
There are a number of pertinent points to consider from this summary. High bit
rates, up to 640 Gbit/s, have been achieved with error free transmission and under
special circumstances that even shorter pulses have been converted. However this
has not been achieved to date will all nonlinear effects, nor with very low detection
penalties. The detection penalties at best were 4 dB, this is not a small number
compared to the essentially penalty free results achieved in for example passive
waveguide demultiplexers using FWM at even 1.2 Tbit/s [65], although such
schemes are much less power efcient than SOAs. The origins of the penalties seem
to be connected in these two cases of 640 Gbit/s performance to the ltering
required to speed up the SOA response, particularly for the XGM based method
when used for demultiplexing.
Conversion efciencies for the system experiments were also relatively low
though the input powers were much lower than used in passive nonlinear wave-
guide devices. In particular, [126] shows performance that appears close to usable
in a real system. However there was only a single demonstration at high speed of a
device that could be format independent using FWM [128]. This would of course
allow for higher bit rates through multilevel coding schemes, and the use of phase
only modulation would also reduce XGM effects considered to be partly respon-
sible for the detection penalty (the output OSNR of *23 dB was though to make
up the remnant part).
With further research and perhaps utilizing some of the techniques discussed
earlier the detection penalties might be decreased in the future, conversion ef-
ciencies increased and possibly operation at 1.2 Tbit/s demonstrated as some
modelling has predicted [93].
Where speed is a lesser concern (e.g. for 40 Gbit/s RZ-OOK or 100 Gbit/s
coherent schemes), then some impressive results have been achieved in terms of
conversion efciency, range, and regenerative performance. Table 1.2 summarizes a
few of the more recent and impressive examples.
24
Table 1.1 Details of fastest and most recent nonlinear switching demonstrations in SOAs
References Nonlinear effect Data rate Data Pulse width Operation Conversion Pump Detection
(Gbit/s) format (ps) efciency power penalty (dB
(dB) @109 BER)
[113] XGM, blue shifted 640 RZ- 0.8 TDM demux, 640 to *22 8 dBm 13.5
bandpass lter OOK 40 Gbit/s Ave.
[124] XPM in symmetric 640 RZ- 0.38 TDM demux, 640 to ? 17 dBm 4
Mach-Zehnder DPSK 40 Gbit/s Ave.
[125] XGM, blue shifted 320 RZ- 1.2 TDM demux, 320 to *19 8 dBm 8
bandpass lter OOK 40 Gbit/s Ave.
[126] XGM, blue shifted 320 RZ- 1.2 Full C band 13 7 dBm 5
bandpass lter OOK wavelength shift Ave.
[127] XGM, blue shifted 320 RZ- 1.0 Wavelength shift *17 5.4 dBm *10
bandpass lter OOK Ave.
[128] FWM 320 RZ- 1.2 Wavelength shift and 18 9 dBm 4
OOK phase conjugation CW
[119] FWM Pulses 1.6 Wavelength shift and 5 7.3 dBm
phase conjugation CW
[110] NPR pulses 0.12 Wavelength shift 12 10 pJ per Output pulse
pulse 0.5 ps
X. Gai et al.
Table 1.2 High conversion efciency and multiple format NLO processing demonstrations in SOAs
References Nonlinear Operation Conversion range (nm @ X Pump power Detection penalty Bit Data No of
effect dB eff.) rate format Channels
[130] Single pump Wavelength shift/ 11.2 @ >0 6 dBm CW 1 20 QPSK 1
FWM OPC 28 @ >10 30 8 PSK
40 16 QAM
[129] Dual pump Wavelength shift/ 100 @ >0 3 dBm CW per ? error free max 20 QPSK 1
FWM OPC 140 @ >10 pump error vector mag. 30 8 PSK
14 % 40 16 QAM
40 OFDM
[134] Single pump Wavelength shift/ 32 @ >10 5 dBm CW per *1.5 256 16 QAM 1
1 Materials and Structures for Nonlinear Photonics
FWM OPC pump *2 320 64 QAM

[123] Single pump Multichannel OPC 10 dBm CW 4 50 NRZ- 4
FWM DQPSK
[135] Single pump Regeneration 10 @ >20 13.5 dBm CW Q raised from 10 NRZ- 1
FWM via saturation of 6 to >12.5 DPSK
FWM eff.
[121] XPM plus lter Regeneration N/A 1 dBm Ave/ 2 20 PM-RZ- 8
channel PSK
25
26 X. Gai et al.
The work of Contestabile et al. [129, 130] has conclusively demonstrated that
ultra-broadband wavelength conversion with gain is possible in a single device with
minimal signal degradation, and that it can be format and bitrate independent at
least to symbol rates <100 GBaud/s. Additionally the phase conjugation function
can also be obtained if desired for distortion compensation. Perhaps the only
shortcoming of their experiment was that the device was not polarization insensi-
tive, but this can be addressed with dual pump schemes [131, 132] and an
appropriately optimized QD SOA, or by using a polarization diversity approach to
obtain the best performance [133].
Filion et al. [134] recently demonstrated wideband conversion with >10 %
efciency of even high order multiplexed signals for the rst time. They also
demonstrated data rates up to 320 Gbit/s with low penalties including the phase
conjugation function.
Given the de facto nature of DWDM technology, the conversion of multiple
channels in a single device is clearly also a priority. Matsuura et al. [123] dem-
onstrated that acceptable penalties can be achieved for wavelength translation/OPC
in a single SOA even with coherent transmission. The penalties resulted from intra-
channel mixing products, meaning that to process a complete DWDM comb four or
less channel banding would be required, as partly demonstrated by Andriolli et al.
[136].
Taking this body of work, it is clear that the prognosis for system deployment of
format transparent highly efcient wideband wavelength translation devices using
in SOAs is very good. All optical regeneration of coherent signals in an SOA is
becoming competitive. Unlike On-Off Keying where an electronic regenerator can
be comparatively simple, coherent versions are relatively complex and expensive.
Saturation effects in SOAs can also be used to limit amplitude noise without
introducing signicant phase noise (or even reducing it in some cases) to improve
the signal Q factor or reduce the detection penalty. Porzi et al. [135] and Wu et al.
[121] demonstrated signicant improvements in the signal quality of coherently
coded data this way, and in the case of Wu et al. did this with 8 channels of
polarisation and wavelength division multiplexed data. In both cases these opera-
tions were performed in a simple integratable single SOA based device.
Finally there have also been developments to enable effects based on FWM to be
mimicked in SOAs. Optical phase conjugation has recently been accomplished for
the rst time based on non-parametric processes [137], and it remains to be seen
whether this enables new types of processing or superior performance to conven-
tional approaches.
References
1. G.I. Stegeman, E.M. Wright, All-optical wave-guide switching. Opt. Quant. Electron. 22,
95122 (1990)
2. N.J. Doran, D. Wood, Nonlinear-optical loop mirror. Opt. Lett. 13, 5658 (1988)
3. L.D. Haret, T. Tanabe, E. Kuramochi, M. Notomi, Extremely low power optical bistability in
silicon demonstrated using 1D photonic crystal nanocavity. Opt. Express 17, 2110821117
(2009)
4. V. Mizrahi, K.W. Delong, G.I. Stegeman, M.A. Sai, M.J. Andrejco, 2-Photon absorption as
a limitation to all-optical switching. Opt. Lett. 14, 11401142 (1989)
5. M. Sheikbahae, D.J. Hagan, E.W. Vanstryland, Dispersion and band-gap scaling of the
electronic kerr effect in solids associated with 2-photon absorption. Phys. Rev. Lett. 65, 96
99 (1990)
6. M. Sheik-Bahae, D.C. Hutchings, D.J. Hagan, E.W. Van Stryland, Dispersion of bound
electronic nonlinear refraction in solids. IEEE J. Quantum Electron. 27, 12961309 (1991)
7. M.H.M. Sheik-Bahae, Third order optical nonlinearities, in Handbook of Optics (OSA, 2000)
8. M. Dinu, Dispersion of phonon-assisted nonresonant third-order nonlinearities. IEEE
J. Quantum Electron. 39, 14981503 (2003)
9. X. Gai, Y. Yu, B. Kuyken, P. Ma, S.J. Madden, J. Van Campenhout, P. Verheyen, G.
Roelkens, R. Baets, B. Luther-Davies, Nonlinear absorption and refraction in crystalline
silicon in the mid-infrared. Laser Photonics Rev. 7, 10541064 (2013)
10. B.S. Wherrett, Scaling rules for multiphoton interband absorption in semiconductors. J. Opt.
Soc. Am. B: 1, 6772 (1984)
11. G.I. Stegeman, D.J. Hagan, L. Torner, Chi((2)) cascading phenomena and their applications
to all-optical signal processing, mode-locking, pulse compression and solitons. Opt.
Quantum Electron. 28, 16911740 (1996)
12. H. Kanbara, H. Itoh, M. Asobe, K. Noguchi, H. Miyazawa, T. Yanagawa, I. Yokohama, All-
optical switching based on cascading of second-order nonlinearities in a periodically poled
titanium-diffused lithium niobate waveguide. IEEE Photonic Tech. L. 11, 328330 (1999)
13. J.L. Jackel, J.J. Johnson, Reverse exchange method for burying proton exchanged wave-
guides. Electron. Lett. 27, 13601361 (1991)
14. Y.N. Korkishko, V.A. Fedorov, T.M. Morozova, F. Caccavale, F. Gonella, F. Segato,
Reverse proton exchange for buried waveguides in LiNbO3. J. Opt. Soc. Am. A: 15, 1838
1842 (1998)
15. M.L. Bortz, M.M. Fejer, Measurement of the 2nd-order nonlinear susceptibility of proton-
exchanged Linbo3. Opt. Lett. 17, 704706 (1992)
16. Y.N. Korkishko, V.A. Fedorov, F. Laurell, The SHG-response of different phases in proton
exchanged lithium niobate waveguides. IEEE J. Sel. Top. Quantum 6, 132142 (2000)
17. C. Langrock, S. Kumar, J.E. McGeehan, A.E. Willner, M.M. Fejer, All-optical signal
processing using chi((2)) nonlinearities in guided-wave devices. J. Lightwave Technol. 24,
25792592 (2006)
18. R.V. Roussev, C. Langrock, J.R. Kurz, M.M. Fejer, Periodically poled lithium niobate
waveguide sum-frequency generator for efcient single-photon detection at communication
wavelengths. Opt. Lett. 29, 15181520 (2004)
19. M.H. Chou, K.R. Parameswaran, M.M. Fejer, I. Brener, Multiple-channel wavelength
conversion by use of engineered quasi-phase-matching structures in LiNbO3 waveguides.
Opt. Lett. 24, 11571159 (1999)
20. G.W. Lu, S. Shinada, H. Furukawa, N. Wada, T. Miyazaki, H. Ito, 160 Gb/s all-optical
phase-transparent wavelength conversion through cascaded SFG-DFG in a broadband linear-
chirped PPLN waveguide. Opt. Express 18, 60646070 (2010)
21. M.H. Chou, I. Brener, G. Lenz, R. Scotti, E.E. Chaban, J. Shmulovich, D. Philen, S.
Kosinski, K.R. Parameswaran, M.M. Fejer, Efcient wide-band and tunable midspan spectral
inverter using cascaded nonlinearities in LiNbO3 waveguides. IEEE Photonics Technol. Lett.
12, 8284 (2000)
22. M.C. Cardakli, D. Gurkan, S.A. Havstad, A.E. Willner, K.R. Parameswaran, M.M. Fejer, I.
Brener, Tunable all-optical time-slot-interchange and wavelength conversion using
difference-frequency-generation and optical buffers. IEEE Photonics Technol. Lett. 14,
200202 (2002)
28 X. Gai et al.
23. A.E. Willner, O.F. Yilmaz, J.A. Wang, X.X. Wu, A. Bogoni, L. Zhang, S.R. Nuccio,
Optically efcient nonlinear signal processing. IEEE J. Sel. Top. Quantum 17, 320332
(2011)
24. J. Huang, J.R. Kurz, C. Langrock, A.M. Schober, M.M. Fejer, Quasi-group-velocity
matching using integrated-optic structures. Opt. Lett. 29, 24822484 (2004)
25. A. Bogoni, X.X. Wu, S.R. Nuccio, A.E. Willner, 640 Gb/s all-optical regenerator based on a
periodically poled lithium niobate waveguide. J. Lightwave Technol. 30, 18291834 (2012)
26. X. Gai, S. Madden, D.Y. Choi, D. Bulla, B. Luther-Davies, Dispersion engineered Ge(11.5)
As(24)Se(64.5) nanowires with a nonlinear parameter of 136 W(1)m(1) at 1550 nm. Opt.
Express 18, 1886618874 (2010)
27. X. Gai, R.P. Wang, C. Xiong, M.J. Steel, B.J. Eggleton, B. Luther-Davies, Near-zero
anomalous dispersion Ge11.5As24Se64.5 glass nanowires for correlated photon pair
generation: design and analysis. Opt. Express 20, 776786 (2012)
28. X. Gai, D.-Y. Choi, S. Madden, B. Luther-Davies, Polarization-independent chalcogenide
glass nanowires with anomalous dispersion for all-optical processing. Opt. Express 20,
1351313521 (2012)
29. X. Gai, D.Y. Choi, S. Madden, B. Luther-Davies, Interplay between Raman scattering and
four-wave mixing in As(2)S(3) chalcogenide glass waveguides. J. Opt. Soc. Am. B: 28,
27772784 (2011)
30. M.H.T. Okuno, T. Nakanishi, M. Onishi, Highly-nonlinear optical bers and their
applications. SEI Tech. Rev. 62, 3440 (2006)
31. R.E. Camacho-Aguilera, Y. Cai, J.T. Bessette, L.C. Kimerling, J. Michel, High active carrier
concentration in n-type, thin lm Ge using delta-doping. Opt. Mater. Express 2, 14621469
(2012)
32. L. Vivien, J. Osmond, J.M. Fedeli, D. Marris-Morini, P. Crozat, J.F. Damlencourt, E. Cassan,
Y. Lecunff, S. Laval, 42 GHz p.i.n Germanium photodetector integrated in a silicon-on-
insulator waveguide. Opt. Express 17, 62526257 (2009)
33. A.S. Liu, R. Jones, L. Liao, D. Samara-Rubio, D. Rubin, O. Cohen, R. Nicolaescu, M.
Paniccia, A high-speed silicon optical modulator based on a metal-oxide-semiconductor
capacitor. Nature 427, 615618 (2004)
34. V.G. Taeed, N.J. Baker, L.B. Fu, K. Finsterbusch, M.R.E. Lamont, D.J. Moss, H.C.
Nguyen, B.J. Eggleton, D.Y. Choi, S. Madden, B. Luther-Davies, Ultrafast all-optical
chalcogenide glass photonic circuits. Opt. Express 15, 92059221 (2007)
35. J. Leuthold, C. Koos, W. Freude, Nonlinear silicon photonics. Nat. Photonics 4, 535544
(2010)
36. X. Chen, C. Li, H.K. Tsang, Device engineering for silicon photonics. Npg Asia Mater. 3,
3440 (2011)
37. V.R. Almeida, R.R. Panepucci, M. Lipson, Nanotaper for compact mode conversion. Opt.
Lett. 28, 13021304 (2003)
38. A.R. Motamedi, A.H. Nejadmalayeri, A. Khilo, F.X. Kartner, E.P. Ippen, Ultrafast nonlinear
optical studies of silicon nanowaveguides. Opt. Express 20, 40854101 (2012)
39. C. Koos, L. Jacome, C. Poulton, J. Leuthold, W. Freude, Nonlinear silicon-on-insulator
waveguides for all-optical signal processing. Opt. Express 15, 59765990 (2007)
40. J.M. Brosi, C. Koos, L.C. Andreani, M. Waldow, J. Leuthold, W. Freude, High-speed low-
voltage electro-optic modulator with a polymer-inltrated silicon photonic crystal
waveguide. Opt. Express 16, 41774191 (2008)
41. M.A. Foster, A.C. Turner, R. Salem, M. Lipson, A.L. Gaeta, Broad-band continuous-wave
parametric wavelength conversion in silicon nanowaveguides. Opt. Express 15, 12949
12958 (2007)
42. H. Hu, J.D. Andersen, A. Rasmussen, B.M. Sorensen, K. Dalgaard, M. Galili, M. Pu, K.
Yvind, K.J. Larsen, S. Forchhammer, L.K. Oxenlowe, Forward error correction supported
150 Gbit/s error-free wavelength conversion based on cross phase modulation in silicon. Opt.
Express 21, 31523160 (2013)
43. F.G. Della Corte, S. Rao, Use of amorphous silicon for active photonic devices. IEEE Trans.
Electron Devices 60, 14951505 (2013)
44. G. Cocorullo, F.G. Della Corte, R. De Rosa, I. Rendina, A. Rubino, E. Terzi, Amorphous
silicon-based guided-wave passive and active devices for silicon integrated optoelectronics.
IEEE J. Sel. Top. Quantum 4, 9971002 (1998)
45. K. Furuya, K. Nakanishi, R. Takei, E. Omoda, M. Suzuki, M. Okano, T. Kamei, M. Mori, Y.
Sakakibara, Nanometer-scale thickness control of amorphous silicon using isotropic wet-
etching and low loss wire waveguide fabrication with the etched material. Appl. Phys. Lett.
100 (2012)
46. C. Lacava, P. Minzioni, E. Baldini, L. Tartara, J.M. Fedeli, I. Cristiani, Nonlinear
characterization of hydrogenated amorphous silicon waveguides and analysis of carrier
dynamics. Appl. Phys. Lett. 103 (2013)
47. K.Y. Wang, A.C. Foster, Ultralow power continuous-wave frequency conversion in
hydrogenated amorphous silicon waveguides. Opt. Lett. 37, 13311333 (2012)
48. B. Kuyken, S. Clemmen, S.K. Selvaraja, W. Bogaerts, D. Van Thourhout, P. Emplit, S.
Massar, G. Roelkens, R. Baets, On-chip parametric amplication with 26.5 dB gain at
telecommunication wavelengths using CMOS-compatible hydrogenated amorphous silicon
waveguides. Opt. Lett. 36, 552554 (2011)
49. B. Kuyken, H. Ji, S. Clemmen, S.K. Selvaraja, H. Hu, M. Pu, M. Galili, P. Jeppesen, G.
Morthier, S. Massar, L.K. Oxenlowe, G. Roelkens, R. Baets, Nonlinear properties of and
nonlinear processing in hydrogenated amorphous silicon waveguides. Opt. Express 19, 146
153 (2011)
50. D.J. Moss, R. Morandotti, A.L. Gaeta, M. Lipson, New CMOS-compatible platforms based
on silicon nitride and hydex for nonlinear optics. Nat. Photonics 7, 597607 (2013)
51. J.S. Levy, A. Gondarenko, M.A. Foster, A.C. Turner-Foster, A.L. Gaeta, M. Lipson, CMOS-
compatible multiple-wavelength oscillator for on-chip optical interconnects. Nat. Photonics
4, 3740 (2010)
52. G.N. Parsons, J.H. Souk, J. Batey, Low hydrogen content stoichiometric silicon-nitride lms
deposited by plasma-enhanced chemical vapor-deposition. J. Appl. Phys. 70, 15531560
(1991)
53. S.C. Mao, S.H. Tao, Y.L. Xu, X.W. Sun, M.B. Yu, G.Q. Lo, D.L. Kwong, Low propagation
loss SiN optical waveguide prepared by optimal low-hydrogen module. Opt. Express 16,
2080920816 (2008)
54. J.B. Nadal, J.M. Daz, D. Vukovic, F.Da Ros, E. Palushani, C. Peucheret, A comparison of
nonlinear media for parametric all-optical signal processing. IEEE Photonic Conf. ThG1.5
(2013)
55. F. Smektala, C. Quemard, L. Leneindre, J. Lucas, A. Barthelemy, C. De Angelis,
Chalcogenide glasses with large non-linear refractive indices. J. Non-Cryst. Solids 239, 139
142 (1998)
56. J.T. Gopinath, M. Soljacic, E.P. Ippen, V.N. Fuflyigin, W.A. King, M. Shurgalin, Third order
nonlinearities in GeAsSe-based glasses for telecommunications applications. J. Appl.
Phys. 96, 69316933 (2004)
57. J.S. Sanghera, L.B. Shaw, P. Pureza, V.Q. Nguyen, D. Gibson, L. Busse, I.D. Aggarwal, C.
M. Florea, F.H. Kung, Nonlinear properties of chalcogenide glass bers. Int. J. Appl. Glass
Sci. 1, 296308 (2010)
58. C. Quemard, F. Smektala, V. Couderc, A. Barthelemy, J. Lucas, Chalcogenide glasses with
high non linear optical properties for telecommunications. J. Phys. Chem. Solids 62, 1435
1440 (2001)
59. R.A. Ganeev, A.I. Ryasnyansky, M.K. Kodirov, T. Usmanov, Two-photon absorption and
nonlinear refraction of amorphous chalcogenide lms. J. Opt. Pure Appl. Opt. 4, 446451
(2002)
60. M. Asobe, Nonlinear optical properties of chalcogenide glass bers and their application to
all-optical switching. Opt. Fiber Technol. 3, 142148 (1997)
30 X. Gai et al.
61. S.J. Madden, D.Y. Choi, D.A. Bulla, A.V. Rode, B. Luther-Davies, V.G. Taeed, M.D.
Pelusi, B.J. Eggleton, Long, low loss etched As2S3 chalcogenide waveguides for all-optical
signal regeneration. Opt. Express 15, 1441414421 (2007)
62. M. Pelusi, F. Luan, T.D. Vo, M.R.E. Lamont, S.J. Madden, D.A. Bulla, D.Y. Choi, B.
Luther-Davies, B.J. Eggleton, Photonic-chip-based radio-frequency spectrum analyser with
terahertz bandwidth. Nat. Photonics 3, 139143 (2009)
63. T.D. Vo, M.D. Pelusi, J. Schroder, F. Luan, S.J. Madden, D.Y. Choi, D.A.P. Bulla, B.
Luther-Davies, B.J. Eggleton, Simultaneous multi-impairment monitoring of 640 Gb/s
signals using photonic chip based RF spectrum analyzer. Opt. Express 18, 39383945 (2010)
64. M.D. Pelusi, F. Luan, D.Y. Choi, S.J. Madden, D.A.P. Bulla, B. Luther-Davies, B.
J. Eggleton, Optical phase conjugation by an As(2)S(3) glass planar waveguide for
dispersion-free transmission of WDM-DPSK signals over ber. Opt. Express 18, 26686
26694 (2010)
65. T.D. Vo, H. Hu, M. Galili, E. Palushani, J. Xu, L.K. Oxenlowe, S.J. Madden, D.Y. Choi, D.
A.P. Bulla, M.D. Pelusi, J. Schroder, B. Luther-Davies, B.J. Eggleton, Photonic chip based
transmitter optimization and receiver demultiplexing of a 1.28 Tbit/s OTDM signal. Opt.
Express 18, 1725217261 (2010)
66. M. Galili, J. Xu, H.C.H. Mulvad, L.K. Oxenlowe, A.T. Clausen, P. Jeppesen, B. Luther-
Davies, S. Madden, A. Rode, D.Y. Choi, M. Pelusi, F. Luan, B.J. Eggleton, Breakthrough
switching speed with an all-optical chalcogenide glass chip: 640 Gbit/s demultiplexing. Opt.
Express 17, 21822187 (2009)
67. M.R.E. Lamont, B. Luther-Davies, D.Y. Choi, S. Madden, X. Gai, B.J. Eggleton, Net-gain
from a parametric amplier on a chalcogenide optical chip. Opt. Express 16, 2037420381
(2008)
68. M.R.E. Lamont, B. Luther-Davies, D.Y. Choi, S. Madden, B.J. Eggleton, Supercontinuum
generation in dispersion engineered highly nonlinear (gamma = 10/W/m) As(2)S(3)
chalcogenide planar waveguide. Opt. Express 16, 1493814944 (2008)
69. A. Zakery, P.J.S. Ewen, A.E. Owen, Photodarkening in As-S lms and its application in
grating fabrication. J. Non-Cryst. Solids 198, 769773 (1996)
70. A. Prasad, C.J. Zha, R.P. Wang, A. Smith, S. Madden, B. Luther-Davies, Properties of
GexAsySe1-x-y glasses for all-optical signal processing. Opt. Express 16, 28042815 (2008)
71. A. Prasad, GeAsSe chalcogenide glasses for all-optical signal processing, in Laser Physics
Center (Australian National University, Australia, 2010)
72. J. He, C. Xiong, A.S. Clark, M.J. Collins, X. Gai, D.Y. Choi, S.J. Madden, B. Luther-Davies,
B.J. Eggleton, Effect of low-Raman window position on correlated photon-pair generation in
a chalcogenide Ge11.5As24Se64.5 nanowire, J Appl. Phys. 112 (2012)
73. X. Gai, B. Luther-Davies, T.P. White, Photonic crystal nanocavities fabricated from
chalcogenide glass fully embedded in an index-matched cladding with a high Q-factor
(>750,000). Opt. Express 20, 1550315515 (2012)
74. E. Ciaramella, Wavelength conversion and all-optical regeneration: achievements and open
issues. J. Lightwave Technol. 30, 572582 (2012)
75. C.S. Cleary, High speed nonlinear optical components for next-generation optical
communications, PhD. Thesis, Photonic Systems Group, Tyndall National Institute,
Department of Physics, University College Cork, 2013
76. R. Bonk, Linear and Nonlinear Semiconductor Optical Ampliers for Next-Generation
Optical Networks (KIT Scientic Publishing, 2013)
77. J. Leuthold, W. Freude, G. Boettger, J. Wang, A. Marculescu, P. Vorreau, R. Bonk, All-
optical regeneration, in ICTON 2006: 8th International Conference on Transparent Optical
Networks, Vol 1, Proceedings: ICTON, MPM, INDUSTRIAL, PICAW, GOWN, ed. by
Marciniak, (2006), pp. 2831
78. J. Leuthold, J. Wang, T. Vallaitis, C. Koos, R. Bonk, A. Marculescu, P. Vorreau, S. Sygletos,
W. Freude, New approaches to perform all-optical signal regeneration, in ICTON 2007:
Proceedings of the 9th International Conference on Transparent Optical Networks, Vol 2,
ed. by Marciniak, (2007), pp. 222225
79. R.J. Manning, R. Giller, X. Yang, R.P. Webb, D. Cotter, Faster switching with
semiconductor optical ampliers, in 2007 Photonics in Switching (2007), pp. 145146
80. S. Diez, C. Schmidt, R. Ludwig, H.G. Weber, K. Obermann, S. Kindt, I. Koltchanov, K.
Petermann, Four-wave mixing in semiconductor optical ampliers for frequency conversion
and fast optical switching. IEEE J. Sel. Top. Quantum Electron. 3, 11311145 (1997)
81. H. Ishikawa, Ultrafast All-Optical Signal Processing Devices (Wiley, Chichester, 2008)
82. N. Christodoulides, I. Khoo, G. Salamo, G. Stegeman, E. Van Stryland, Nonlinear refraction
and absorption: mechanisms and magnitudes. Adv. Opt. Photonics 2, 60200 (2010)
83. K. Hall, G. Lenz, A. Darwish, E. Ippen, Subpicosecond gain and index nonlinearities in
InGaAsP diode lasers. Opt. Commun. 111, 589612 (1994)
84. R.S. Grant, W. Sibbett, Observatiobns of ultrafast nonlinear refraction in an InGaAsP optical
amplier. Appl. Phys. Lett. 58, 11191121 (1991)
85. G.P. Agrawal, N.A. Olsson, Self phase modulation and spectral braodening of optical pulses
in semiconductor laser ampliers. IEEE J. Quantum Electron. 25, 22972306 (1989)
86. K. Abedi, H. Taleb, Phase recovery acceleration in quantum-dot semiconductor optical
ampliers. J. Lightwave Technol. 30, 19241930 (2012)
87. L.G. Chen, R.G. Lu, S.J. Zhang, J.F. Li, Y. Liu, Time-resolved chirp properties of
semiconductor optical ampliers in high-speed all-optical switches, Opt. Eng. 52 (2013)
88. A.H. Flayyih, A.H. Al-Khursan, Theory of four-wave mixing in quantum dot semiconductor
optical ampliers, J. Phy. D-Appl. Phys. 46 (2013)
89. W. Freude, R. Bonk, T. Vallaitis, A. Marculescu, A. Kapoor, E.K. Sharma, C. Meuer, D.
Bimberg, R. Brenot, F. Lelarge, G.H. Duan, C. Koos, J. Leuthold, IEEE linear and nonlinear
semiconductor optical ampliers, in 2010 12th International Conference on Transparent
Optical Networks (2011)
90. S.R. Hosseini, M. Razaghi, N.K. Das, Analysis of ultrafast nonlinear phenomenas influences
on output optical pulses and four-wave mixing characteristics in semiconductor optical
ampliers. Opt. Quantum Electron. 42, 729737 (2011)
91. X. Huang, C. Qin, Y. Yu, X.L. Zhang, Acceleration of carrier recovery in a quantum well
semiconductor optical amplier due to the tunneling effect. J. Opt. Soc. Am. B-Opt. Phy. 29,
29902994 (2012)
92. J.H. Kim, Influence of optical pumping wavelength on the ultrafast gain and phase recovery
acceleration of quantum-dot semiconductor optical ampliers, J. Opt. 15 (2013)
93. A. Kotb, 1 Tb/s high quality factor NOR gate based on quantum-dot semiconductor optical
amplier. Opt. Quantum Electron. 45, 12591268 (2013)
94. T. Liu, K. Obermann, K. Petermann, F. Girardin, G. Guekos, Effect of saturation caused by
amplied spontaneous emission on semiconductor optical amplier performance. Electron.
Lett. 33, 20422043 (1997)
95. Y. Liu, L.G. Chen, X. Zheng, S.J. Zhang, R.G. Lu, Ultrafast nonlinear dynamics in
semiconductor optical ampliers for optical signal processing, in Semiconductor Lasers and
Applications V, eds. by Zhu, Li, Peters, Yu (Spie-Int Soc Optical Engineering, Bellingham,
(2012)
96. F.D. Mahad, A.S.M. Supaat, S.M. Idrus, D. Forsyth, Analyses of semiconductor optical
amplier (SOA) four-wave mixing (FWM) for future all-optical wavelength conversion.
Optik 124, 13 (2013)
97. K. Solis-Trapala, H.J.S. Dorren, Dynamic and static gain characteristics of quantum-dot
semiconductor optical ampliers operating at 1.55 mu m. Opt. Commun. 298, 106113
(2013)
98. R. Giller, R.J. Manning, G. Talli, R.P. Webb, M.J. Adams, Analysis of the dimensional
dependence of semiconductor optical amplier recovery speeds. Opt. Express 15, 17731782
(2007)
99. F. Ginovart, J. Simon, I. Valiente, Gain recovery dynamics in semiconductor optical
amplier. Opt. Commun. 199, 111115 (2001)
100. F. Girardin, G. Guekos, A. Houbavlis, Gain recovery of bulk semiconductor optical
ampliers. IEEE Photonics Technol. Lett. 10, 784786 (1998)
32 X. Gai et al.
101. R. Manning, A. Ellis, A. Poustie, K. Blow, Semiconductor laser ampliers for ultrafast all-
optical signal processing. JOSA B 14, 32043216 (1997)
102. J. Mrk, M. Nielsen, T. Berg, The dynamics of semiconductor optical amplier: modeling
and applications. Opt. Photonics News 14, 4248 (2003)
103. M. Sugawara, T. Akiyama, N. Hatori, Y. Nakata, K. Otsubo, H. Ebe, Quantum-dot
semiconductor optical ampliers, in APOC 2002: Asia-Pacic Optical and Wireless
Communications; Materials and Devices for Optical and Wireless Communications, eds. by
Chang Hasnain, Xia, Iga (2002), pp. 259275
104. T. Vallaitis, C. Koos, R. Bonk, W. Freude, M. Laemmlin, C. Meuer, D. Bimberg,
J. Leuthold, Slow and fast dynamics of gain and phase in a quantum dot semiconductor
optical amplier. Opt. Express 16, 170178 (2008)
105. C. Cleary, M. Power, S. Schneider, R. Webb, R. Manning, Fast gain recovery rates with
strong wavelength dependence in a non-linear SOA. Opt. Express 18, 2572625737 (2010)
106. T. Akiyama, H. Kuwatsuka, T. Simoyama, Y. Nakata, K. Mukai, M. Sugawara, O. Wada, H.
Ishikawa, Ultrafast nonlinear processes in quantum-dot optical ampliers. Opt. Quantum
Electron. 33, 927938 (2001)
107. A. Rostami, H. Nejad, R. Qartavol, H. Saghai, Tb/s optical logic gates based on quantum-dot
semiconductor optical ampliers. IEEE J. Quantum Electron. 46, 354360 (2010)
108. S. Alavizadeh, H. Baghban, A. Rostami, Quantum-dot semiconductor optical amplier
performance management under optical injection. J. Mod. Opt. 60, 509514 (2013)
109. H.J.S. Dorren, A.K. Mishra, Z.G. Li, H.K. Ju, H. de Waardt, G.D. Khoe, T. Simoyama, H.
Ishikawa, H. Kawashima, T. Hasama, All-optical logic based on ultrafast gain and index
dynamics in a semiconductor optical amplier. IEEE J. Sel. Top. Quantum Electron. 10,
10791092 (2004)
110. A.K. Mishra, X. Yang, D. Lenstra, G.D. Khoe, H.J.S. Dorren, Wavelength conversion
employing 120-fs optical pulses in an SOA-based nonlinear polarization switch. IEEE J. Sel.
Top. Quantum Electron. 10, 11801186 (2004)
111. X. Yang, A.K. Mishra, D. Lenstra, F.M. Huijskens, H. de Waardt, G.D. Khoe, H.J.S. Dorren,
Sub-picosecond all-optical switch using a multi-quantum-well semiconductor optical
amplier. Opt. Commun. 236, 329334 (2004)
112. J. Leuthold, R. Bonk, T. Vallaitis, A. Marculescu, W. Freude, C. Meuer, D. Bimberg, R.
Brenot, F. Lelarge, G.H. Duan, Linear and nonlinear semiconductor optical ampliers, paper
OThI3, in 2010 Conference on Optical Fiber Communication (OFC) Collocated with
National Fiber Optic Engineers Conference (NFOEC) (IEEE, 2010)
113. E. Tangdiongga, Y. Liu, H. de Waardt, G.D. Khoe, A.M.J. Koonen, H.J.S. Dorren, X. Shu, I.
Bennion, All-optical demultiplexing of 64040 Gbits/s using ltered chirp of a
semiconductor optical amplier. Opt. Lett. 32, 835837 (2007)
114. A.H. Flayyih, A.H. Al-Khursan, Four-wave mixing in quantum dot semiconductor optical
ampliers. Appl. Optics 52, 31563165 (2013)
115. J. Inoue, H. Kawaguchi, Highly nondegenerate four-wave mixing among subpicosecond
optical pulses in a semiconductor optical amplier. IEEE Photonics Technol. Lett. 10, 349
351 (1998)
116. K. Kikuchi, C. Lorattanasane, K. Saito, Telenor, Phase-conjugation characteristics of
semiconductor optical ampliers, in 22nd European Conference on Optical
Communications, Proceedings, Vols 16: Co-Located With: 2nd European Exhibition on
Optical CommunicationEeoc 96(1996), pp. C269-C272
117. M.A. Summereld, R.S. Tucker, Frequency-domain model of multiwave mixing in bulk
semiconductor optical ampliers. IEEE J. Sel. Top. Quantum Electron. 5, 839850 (1999)
118. Y. Kim, H. Lee, J. Jeong, Analysis of four-wave mixing in semiconductor optical ampliers
for optical frequency conversion and optical phase conjugation. J. Korean Phys. Soc. 34,
S582S586 (1999)
119. K. Kikuchi, K. Matsuura, Transmission of 2-ps optical pulses at 1550 nm over 40 km
standard ber using midspan optical phase conjugation in semiconductor optical ampliers.
IEEE Photonics Technol. Lett. 10, 14101412 (1998)
120. M.J. Connelly, C.L. Janer, Modeling of semiconductor optical amplier RIN and phase noise
for optical PSK systems. Opt. Quantum Electron. 44, 219225 (2012)
121. W.H. Wu, Y. Yu, B.R. Zou, W.L. Yang, X.L. Zhang, Single SOA based simultaneous
amplitude regeneration for WDM-PDM RZ-PSK signals. Opt. Express 21, 67186723
(2013)
122. J.P.R. Lacey, S.J. Madden, M.A. Summereld, R.S. Tucker, A.I. Faris, 4-Channel WDM
optical-phase conjugator using 4-wave-mixing in a single semiconductor optical amplier.
Electron. Lett. 31, 743744 (1995)
123. M. Matsuura, N. Calabretta, O. Raz, H.J.S. Dorren, Multichannel wavelength conversion of
50 Gbit/s NRZ-DQPSK signals using a quantum-dot semiconductor optical amplier. Opt.
Express 19, 560566 (2011)
124. T. Hirooka, M. Okazaki, T. Hirano, P.Y. Guan, M. Nakazawa, S. Nakamura, All-optical
demultiplexing of 640 Gb/s OTDM-DPSK signal using a semiconductor SMZ switch. IEEE
Photonics Technol. Lett. 21, 15741576 (2009)
125. M. Matsuura, O. Raz, F. Gomez-Agis, N. Calabretta, H.J.S. Dorren, Error-free 32040 Gbit/s
optical demultiplexing based on blueshift ltering in a quantum-dot semiconductor optical
amplier. Opt. Lett. 38, 238240 (2013)
126. M. Matsuura, O. Raz, F. Gomez-Agis, N. Calabretta, H.J.S. Dorren, Ultrahigh-speed and
widely tunable wavelength conversion based on cross-gain modulation in a quantum-dot
semiconductor optical amplier. Opt. Express 19, 551559 (2011)
127. Y. Liu, E. Tangdiongga, Z. Li, H. de Waardt, A.M.J. Koonen, G.D. Khoe, X.W. Shu, I.
Bennion, H.J.S. Dorren, Errior-free 320 Gb/s all-omtical wavelength conversion using a
single semiconductor optical amplier. J. Lightwave Technol. 25, 103108 (2007)
128. M. Matsuura, O. Raz, F. Gomez-Agis, N. Calabretta, H.J.S. Dorren, 320 Gbit/s wavelength
conversion using four-wave mixing in quantum-dot semiconductor optical ampliers. Opt.
Lett. 36, 29102912 (2011)
129. G. Contestabile, Y. Yoshida, A. Maruta, K. Kitayama, Ultra-broad band, low power, highly
efcient coherent wavelength conversion in quantum dot SOA. Opt. Express 20, 27902
27907 (2012)
130. G. Contestabile, Y. Yoshida, A. Maruta, K.I. Kitayama, Coherent wavelength conversion in a
quantum dot SOA. IEEE Photonics Technol. Lett. 25, 791794 (2013)
131. J. Lu, J.J. Yu, H. Zhou, Y. Li, L. Chen, Polarization insensitive wavelength conversion based
on dual-pump four-wave mixing for polarization multiplexing signal in SOA. Opt. Commun.
284, 53645371 (2011)
132. R.M. Jopson, R.E. Tench, Polarization independent phase conjugation of lightwave signals.
Electron. Lett. 29, 22162217 (1993)
133. J.P.R. Lacey, M.A. Summereld, S.J. Madden, Tunability of polarization-insensitive
wavelength converters based on four-wave mixing in semiconductor optical ampliers.
J. Lightwave Technol. 16, 24192427 (1998)
134. B. Filion, W.C. Ng, A.T. Nguyen, L.A. Rusch, S. LaRochelle, Wideband wavelength
conversion of 16 Gbaud 16-QAM and 5 Gbaud 64-QAM signals in a semiconductor optical
amplier. Opt. Express 21, 1982519833 (2013)
135. C. Porzi, A. Bogoni, G. Contestabile, Regenerative wavelength conversion of DPSK signals
through FWM in an SOA. IEEE Photonics Technol. Lett. 25, 175178 (2013)
136. N. Andriolli, S. Faralli, X.J.M. Leijtens, J. Bolk, G. Contestabile, Monolithically integrated
all-optical regenerator for constant envelope WDM signals. J. Lightwave Technol. 31, 322
327 (2013)
137. R.X. Yu, N.K. Fontaine, R. Proietti, B.B. Guan, S.J.B. Yoo, All-optical phase conjugation
using a 90 optical hybrid and nested SOA-MZIs, in 2013 Optical Fiber Communication
Conference and Exposition and the National Fiber Optic Engineers Conference (2013)
Chapter 2
CMOS Compatible Platforms
for Integrated Nonlinear Optics
David J. Moss and Roberto Morandotti
Abstract Nonlinear photonic chips are capable of generating and processing sig-
nals all-optically with performance far superior to that possible electronically
particularly with respect to speed. Although silicon-on-insulator has been the
leading platform for nonlinear optics, its high two-photon absorption at telecom-
munications wavelengths poses a fundamental limitation that is an intrinsic property
of silicons bandstructure. We review recent progress in new non-silicon CMOS-
compatible platforms for nonlinear optics, focusing on Hydex glass and silicon
nitride, and briefly discuss the promising new platform of amorphous silicon. These
material systems have opened up many new capabilities such as on-chip optical
frequency comb generation, ultrafast optical pulse generation and measurement.
We highlight their potential future impact as well as the challenges to achieving
practical solutions for many key applications.
2.1 Introduction
All-optical signal generation and processing [1, 2] have been highly successful at
enabling a vast array of capabilities, such as switching and de-multiplexing of
signals at unprecedented speeds [3, 4], achieving parametric gain [5] on a chip,
Raman lasing [6], wavelength conversion [7], optical logic [8], all-optical regen-
eration [9, 10], radio-frequency (RF) spectrometers operating at THz speeds
[11, 12], as well as entirely new functions such as ultra-short pulse measurement
[13, 14] and generation [15] on a chip, optical temporal cloaking [16], and many
others. Phase sensitive functions [14, 17], in particular, will likely be critical for
D.J. Moss (&)

School of Electrical and Computer Engineering,
RMIT University, Melbourne, VIC 3001, Australia
e-mail: david.moss@rmit.edu.au
R. Morandotti
INRS-EMT, 1650 Boulevard Lionel Boulet, Varennes, QC J3X 1S2, Canada

36 D.J. Moss and R. Morandotti
telecommunications systems that are already using phase encoding schemes

[18, 19]. The ability to produce these devices in integrated formphotonic inte-
grated circuits (PICs)will reap the greatest dividends in terms of cost, footprint,
energy consumption and performance, where demands for greater bandwidth,
network flexibility, low energy consumption and cost must all be met.
The quest for high performance platforms for integrated nonlinear optics has
naturally focused on materials with extremely high nonlinearity, and silicon-on-
insulator (SOI) has led this eld for several years [2]. Its high refractive index
allows for tight connement of light within nanowires that, combined with its high
Kerr nonlinearity (n2) has yielded extremely high nonlinear parameters of
= 200,000 W1 km1 ( = n2/c Aeff, where Aeff is the effective area of the
waveguide, c is the speed of light and is the pump frequency).
As a platform for linear photonics, SOI has already gained a foothold as a
foundation for the silicon photonic chip industry [20] aimed at replacing the large
network interface cards in optical communications networks, with the ability to
combine electronics and photonics on the same chip. These rst-generation silicon
photonic chips typically employ photonics for passive splitters, lters, and multi-
plexers and offer the substantial benet of potentially exploiting the enormous
global CMOS infrastructure, although not without challenges [21, 22]. What is
clear, however, is that any progress made in this direction for these rst generation
SOI chips will have a direct benet on future generation all-optical chips, whether
in SOI directly or in other CMOS-compatible platforms. Many of the issues that
make CMOS compatibility compelling for linear passive devices apply equally well
to all-optical-processing chips.
Indeed, silicon is so attractive as a platform for both linear and nonlinear pho-
tonics that were it not for one single issue, the quest for the ideal all-optical platform
might already be solved. In 2004 it was realized [23] that crystalline silicon (the
basis of SOI) suffers from high nonlinear absorption due to (indirect) two-photon
absorption (TPA) in telecommunications bands at wavelengths less than 2000 nm
(Fig. 2.1). Furthermore, the problem is compounded by the fact that TPA generates
free carriers which themselves produce signicant linear absorption. This was rst
realized as being a fundamental limitation, at that time in terms of achieving net
Raman gain on a chip [23]. Since then, many methods have been developed for
reducing the effect of TPA generated free carriers, such as sweeping carriers out by
the use of reverse-biased p-i-n junctions [24], reducing carrier lifetime by ion-
implantation, and other approaches.
However, this only solves the secondary effect of free carriersit has no effect
on silicons intrinsic nonlinear gure of merit (FOM = n2/ , where is the TPA
coefcient and the wavelength), which is only 0.3 near 1550 nm [2527]
(Fig. 2.1). This represents a fundamental limitationan intrinsic property of sili-
cons bandstructure. The fact that many impressive all-optical demonstrations have
been made in silicon despite its low FOM is a testament to how exceptional its
linear and nonlinear optical properties are. Nonetheless, the critical impact of sili-
cons large TPA was illustrated [28, 29] in 2010 by the demonstration of high
parametric gain at wavelengths beyond 2 m, where TPA vanishes. Indeed, it is
2 CMOS Compatible Platforms for Integrated Nonlinear Optics 37
Fig. 2.1 Source of two-photon absorption in silicon in the telecom band near 1550 nm and
nonlinear FOM of silicon from 1400 to 2400 nm obtained by taking an average of results reported
in the literature [2527], showing that the FOM is well below 1 in the telecom band due to the
indirect TPA, and exceeds 1 beyond 2000 nm, making it an attractive nonlinear platform in this
wavelength range
likely that in the mid-infrared wavelength range where silicon is transparent to both
one photon and two photon transitionsbetween 2 and 6 mit will undoubtedly
remain a highly attractive platform for nonlinear photonics, despite the influence of
higher order absorption [30].
For the telecom band, however, the search has continued for the ideal nonlinear
platform. Historically, two important platforms have been chalcogenide glasses [31]
and AlGaAs [32]. Chalcogenide glasses have achieved very high performance as
nonlinear devices, but pose some unique challenges. Realizing nanowires with
ultrahigh nonlinearity (>10,000 W1 km1) has proven elusive due to fabrication
challenges, as has been achieving a material reliability on par with semiconductors.
AlGaAs was the rst platform proposed for nonlinear optics in the telecom band
[32] and offers the powerful ability to tune the nonlinearity and FOM by varying the
alloy composition. A signicant issue for AlGaAs, however, is that nanowires
require very challenging fabrication [33] methods. Nonetheless, both platforms
offer signicant advantages and will undoubtedly play a role in future all-optical
photonic chips.
With the motivation of overcoming this fundamental shortcoming of silicon,
new platforms for nonlinear optics have recently been proposed [3438] that have
achieved considerable success and also offer CMOS compatibility. They are based
on silicon nitride SiN and high-index doped silica (trade-named Hydex reference
patents). Originally developed for linear optics [3941], these platforms are par-
ticularly promising due to their low linear loss, relatively large nonlinearities and,
most signicantly, negligible nonlinear loss at telecommunication wavelengths
[42]. In addition, their high quality CMOS-compatible fabrication processes, high
material stability, and the ability to engineer dispersion [36] make these platforms
highly attractive.
Indeed, within a short period of time signicant progress has been made with
respect to their nonlinear performanceparticularly in the context of optical
frequency comb (OFC) generation in microresonators [4350]. Since the demon-
stration of OFCs in SiN [36] and Hydex [37] in 2010, this eld has proliferated.
Extremely wide-band frequency combs [4244], sub 100-GHz combs [46], line-by-
line arbitrary optical waveform generation [47], ultrashort pulse generation [15, 49],
and dual frequency combs [50] have been reported. In addition to OFC generation,
optical harmonic generation [51] has been observed. These achievements have not
been possible in SOI at telecom wavelengths because of its low FOM.
Here, we review the substantial progress made towards nonlinear optical
applications of these new CMOS-compatible platforms as well as the newly
emerging promising platform of amorphous silicon. The high performance, reli-
ability, manufacturability of all of these platforms combined with their intrinsic
compatibility with electronic-chip manufacturing (CMOS) has raised the prospect
of practical platforms for future low-cost nonlinear all-optical PICs.
2.2 Platforms
Silicon nitride (Si3N4), a CMOS-compatible material well known in the computer

chip industry as a dielectric insulator, has been used as a platform for linear inte-
grated optics [39] for some time. However, only recently [34] has it been proposed
as a platform for nonlinear optics. Historically, the challenge for SiN optical devices
has been to grow low loss layers thicker than 250 nm, due to tensile lm stress.
Achieving such thick layers is critical for nonlinear optics since both high mode
connement as well as dispersion engineering for phase matching [36] are needed.
Thick (>500 nm) low loss SiN layers were recently grown (Fig. 2.2) by plasma-
enhanced chemical vapor deposition (PECVD) [34] as well as by low-pressure
chemical vapor deposition (LPCVD) [36] (Fig. 2.3). The latter approach employed
Fig. 2.2 Silicon nitride nanowires (a) SEM micrograph of the SiN/SiO2 waveguides a before and
b after SiO2 upper-cladding deposition, respectively [34]; Bottom calculated group index (a) and
group velocity dispersion (b) of SiN nanowires with different geometries from [35]
a thermal cycling process that resulted in very thick (700 nm) lms that yielded
nanowires with very low propagation loss (0.4 dB/cm).
The rst nonlinear optical studies of SiN waveguides were reported in 2008 [34]
showing nonlinear shifting of the resonances in 700 nm thick SiN ring resonators
with 200 mW CW optical pump power. Time resolved measurements enabled
thermal and Kerr contributions to be separated, resulting in an n2 * 10 that of
silica glass, which is consistent with Millers rule [52]. This value has been validated
in subsequent reports of nonlinear optics in SiN nanowires and resonators. Fig-
ure 2.2 also shows the group refractive index and linear dispersion 2 as a function
of wavelength for different nanowire dimensions. This represents the rst report of
dispersion engineering in SiN waveguides.
Parametric gain in SiN was rst demonstrated [36] in low loss nanowires by
centering the pump for the FWM process in the anomalous group-velocity
dispersion (GVD) regime near the zero-GVD point (Fig. 2.3). This allowed for
broad-bandwidth phase matching, and hence signal amplication, over a wide range
of wavelengths. Net gain was achieved in long (6 cm) SiN waveguides with
a nonlinear parameter of 1200 W1 km1 and a zero-GVD point near 1560 nm.
Fig. 2.3 Silicon nitride nanowires. a Theoretical dispersion [36] of a SiN nanowire grown by low
pressure CVD, showing a zero-GVD point at 1560 nm with anomalous dispersion in the C-band.
Inset shows a scanning electron micrograph (SEM) of the cross-section of the SiN nanowire
highlighting the trapezoidal shape of the core and the cladding materials; (b) parametric gain
achieved with a pump wavelength at 1550 nm, from [36]
An on/off signal gain as high as 3.6 dB was observed over a 150-nm bandwidth,
and since the total propagation loss through the waveguides was 3 dB, this
represented net parametric amplication.
Hydex glass was developed [41] as a low loss CMOS compatible optical platform
primarily for advanced linear lters. Its refractive index range of n = 1.51.9 is
slightly lower than SiN, being comparable to SiON, and so a buried waveguide
geometry is typically used rather than nanowires. Nonetheless, the core-cladding
contrast of 17 % still allows for relatively tight bend radii of 20 m. Its proprietary
composition is primarily aimed at reducing the need for annealing by reducing the
effect of NH bondsthe main source of absorption loss in the telecom band. This
platform has resulted in extraordinarily low linear propagation losses of 57 dB/m,
allowing for the use of extremely long waveguide spirals [14, 42, 53]. Figure 2.4
shows a schematic of a 45-cm-long spiral waveguide contained within a square area
of 2.5 mm 2.5 mm, pigtailed to single-mode ber via low loss on-chip beam
Fig. 2.4 Hydex waveguides. a SEM image of the cross-section of the Hydex waveguide [14] prior
to the nal deposition of the SiO2 upper cladding; b theoretical mode prole and c top down
schematic view of the 45 cm long spiral waveguide; d dispersion curves for the quasi-TE and
quasi-TM modes of the Hydex waveguides [54]. The shaded region indicates the experimentally
measured wavelength range (solid lines); the dispersion is extrapolated outside this region (dashed
lines). The two zero dispersion points are illustrated by the vertical arrows e parametric gain
curves from [53] when pumping with a pulsed source at peak powers up to 40 W with a
wavelength at 1540 nm
expanders and a SEM picture of its cross section (before cladding deposition). The
lms were fabricated with CMOS compatible processes that yielded exceptionally
low sidewall roughness in the core layer. Self-phase modulation experiments [42]
yielded a Kerr nonlinearity of n2 = 1.15 1019 m2/W, or *4.6 silica glass, and
roughly half as large as that for SiN, with a nonlinearity parameter 233 W1 km1
(*200 standard single-mode telecommunications ber). This enhancement in n2 is,
like SiN, in agreement with Millers rule [52], meaning that the proprietary chemistry
of Hydex is not relevant to its nonlinear optical performance. The waveguides were
engineered to yield anomalous dispersion [54] (Fig. 2.4), critical for efcient, wide
bandwidth FWM over most of the C-band with zero-GVD points being 1600 nm for
TE polarization and 1560 nm for TM. This resulted in a large FWM wavelength
tuning range with efcient parametric gain of +15 dB and a signal to idler conversion
efciency of +16.5 dB (Fig. 2.4) [53].
Table 2.1 Nonlinear parameters for CMOS compatible optical platforms

a-Si [65] c-Si [2, 27] SiN [3436] Hydex [42, 54]
n2 (fused silicaa) 700 175 10 5
[W1 m1] 1200 300 1.4 0.25
TPA [cm/GW] 0.25 0.9 Negligibleb Negligiblec
FOM 5 0.3 1 1
a
n2 for fused silica = 2.6 1020 m2 /W [1]
b
No nonlinear absorption has been observed in SiN nanowires
c
No nonlinear absorption has been observed in Hydex waveguides up to intensities of
25 GW/cm2 [42]
Table 2.1 summarizes the linear and nonlinear properties of these CMOS
compatible platforms. The empirical Millers rule [52] predicts that n2 increases as
the 4th power of n0. The price of increasing n0 in order to increase the Kerr
nonlinearity is that this also tends to increase propagation losses. The relatively low
refractive indices of Hydex (1.75) and Si3N4 (2.0) result in extremely low linear and
nonlinear lossestheir greatest strengthbut also results in fairly modest values of
n2. The low linear and nonlinear losses have largely accounted for the success of
these platforms for nonlinear optics in the telecommunications band, resulting in a
combination of many factors that include negligible TPA, a moderately high
nonlinearity, the ability to engineer dispersion, their high quality growth and fab-
rication processes, and their excellent material reliability. However, the ultimate
platform would be one with a nonlinearity ideally larger than silicon and with a
much larger FOM.
While amorphous silicon has been studied as a nonlinear material for some time
[55], and developed as a platform for linear photonics in the telecom band for a
number of years [56, 57], only recently has it been proposed [58] as an alternative
to SOI for nonlinear optics in the telecom band, with the hope that a-Si could
possibly offer lower TPA than silicon. Table 2.2 surveys the measured nonlinear
properties for a-Si, which show a near universal improvement in both the nonlin-
earity and FOM over c-Si. Although initial measurements yielded a FOM no better
than c-Si (*0.5) [58, 59], more recent results have yielded FOMs from 1 [60] to as
high as 2 [61, 62]. This has enabled very high parametric gain of more than +26 dB
over the C-band [63]. However, to date a key issue for this material has been a lack
of stability resulting in a signicant degradation in performance over relatively
short timescales [64].
Table 2.2 Nonlinear parameters from the literature, for amorphous silicon
References [65] [63, 64] [60] [59] [61] [62]
n2 (1017 m2/W) 2.1 1.3 4.2 0.05 0.3 7.43
1 1
(W m ) 1200 770 2000 35 N/A N/A
TPA (cm/GW) 0.25 0.392 4.1 0.08 0.2 4 from [58, 60]
FOM 5 0.3 2.2 0.4 0.66 0.3 0.4 0.97 1.1
Very recently, a-Si nanowires were demonstrated [65] that displayed a combi-
nation of high FOM, high nonlinearity, and good material stability at telecom
wavelengths. Figure 2.5 shows a cross section SEM picture of a hydrogenated
amorphous silicon nanowire (a-Si:H) waveguide fabricated in a 200 mm CMOS
pilot line at CEA-LETI, France. The a-Si:H lm was deposited by plasma enhanced
chemical vapor deposition (PECVD) at 350C on 1.7 m oxide deposited on a bulk
wafer. After deposition of a silica hard mask, two steps of 193 nm DUV lithography
and HBr silicon etching were used to dene grating couplers that were well aligned
with serpentine waveguides with varying lengths (1.2211 cm). The fabricated
waveguides are *220 nm in thickness and *500 nm in width. A 500 nm oxide
was deposited to provide an upper cladding. The group velocity dispersion for
the TE mode conned within a 500 nm 220 nm nanowire was calculated
with FEMSIM, yielding an anomalous second-order dispersion parameter
2 = 4.2 1025 s2/m at = 1550 nm.
Figure 2.5 shows a schematic of the experimental setup used for the measure-
ment of both the linear and nonlinear propagation characteristics of the a-Si:H
nanowires. A modelocked ber laser with near transform limited *1.8 ps long
pulses at a repetition rate of 20 MHz at 1550 nm was coupled into the TE mode of
the nanowires via in-plane gratings. The ber to waveguide coupling loss per
coupler was *10.6 and 12.4 dB per entry and exit, respectivelyhigher than
expected due to the grating couplers not being optimized. The propagation loss of
the TE mode was measured to be about 4.5 dB/cm, via a cut-back method on
serpentine waveguides with lengths varying from 1.22 to 11 cm. A comparison with
linear measurements performed on 1.3 mm long straight nanowires yielded a loss
contribution due to the bends (10 m radius) of about 4 dB, i.e. on the order of
0.04 dB/bend.
To determine the nonlinear parameters of the waveguides a series of self- phase
modulation (SPM) measurements in a 1.22 cm long nanowire were performed with
a coupled peak power up to *3 W. The output spectrum was then measured as a
function of input power.
Figure 2.5 shows the measured output spectra of the pulses for three coupled
peak powers as well as the experimental contour plots of the output spectra as a
function of coupled peak power ranging between 0.03 and 3 W. Strong spectral
broadening is observed, which is the signature of self-phase modulation of the pulse
propagating along the nonlinear nanowire. At higher coupled powers (>1.8 W), the
spectral broadening of the pulse was limited by the spectral transfer function of the
grating couplers, which had a 3 dB bandwidth of around *35 nm centered near
1550 nm. Split-step-Fourier method simulations were used to solve the nonlinear
Schrdinger equation governing the propagation of the picosecond optical pulse in
the nonlinear waveguide, in the presence of second-order dispersion and TPA. The
impact of dispersion was negligible, as expected for 2 = 4.2 1025 s2/m,
associated with a dispersion length exceeding 7 m for the 1.8 ps pulseswell over
the physical length of the waveguide. Figure 2.5 shows the output spectra from the
simulations, showing good agreement with measurement, when taking the TPA
contribution stated above and a nonlinear waveguide parameter = 1200/W/m,
(a) (b)
(c) (d)
3 3
Normalized power [Lin. scale]

Normalized power [lin. scale]
2.5 3.03W 2.5
2 2
1.5 1.5
1.21W
1 1
0.5 0.5
0.03W
0 0
1530 1535 1540 1545 1550 1555 1560 1565 1530 1535 1540 1545 1550 1555 1560 1565
Wavelength [nm] Wavelength [nm]
(e) x 10
-7 (f) x 10
-25
3 12 3
Coupled peak power [W]
Coupled Peak Power [W]
2.5
2.5 10 2.5
2
2 8 2
1.5
1.5 6 1.5
1
1 4 1
0.5 2 0.5 0.5
1535 1540 1545 1550 1555 1560 1535 1540 1545 1550 1555 1560
(g) (h)
-60
Power [dBm]
-70
-80
-90
1520 1540 1560 1580
Wavelength [nm]
Fig. 2.5 Amorphous silicon nanowires from [65]. Top a Experimental setup for measuring self
phase modulation and nonlinear transmission in order to extract the Kerr nonlinearity (n2) and the
two-photon absorption coefcient. b SEM image of nanowire cross-section. Output spectra for
0.03, 1.21 and 3.03 W coupled peak power c Experiment d Simulation. The curves are normalized
and shifted upwards with increasing powers for clarity. e Experimental and f theory 2D plots
showing the spectral broadening of the output pulse spectra versus coupled peak power. Note the
linear intensity scale at the right is relative. g self phase modulation broadened spectrum taken over
a period of an hour, showing negligible change. h Band diagram of silicon showing the principle
of simultaneous increasing the nonlinearity and nonlinear FOM
associated with a Ker nonlinearity of n2 = 2.1 1017 m2/W. The low TPA of a-Si:
H was also reflected in the absence of any blue shift in the output spectra. Figure 2.5
also shows the output spectra taken over a period of several hours, showing no
degradation in the nonlinear characteristics. Experimental measurements of self-
phase modulation and nonlinear transmission yielded a record high nonlinear FOM
of 5over 10 that of SOItogether with an n2 of 34 times c-Si.
The ability for a:Si to display both high nonlinearity and FOM has surprised
many researchers, since in most materials, higher nonlinearity is accompanied by
higher TPA. The key to the anomalous behavior of amorphous silicon may reside
in the difference between direct and indirect transitions. In direct bandgap semi-
conductors (GaAs/AlGaAs [66]), below the bandgap both n2 and TPA always
increase with decreasing bandgap energy. For these simple direct bandgap semi-
conductors this is largely a result of KramersKronig relations (adapted for nonlinear
response functions [67]) that predict that increasing the bandgap to decrease non-
linear absorption also decreases the nonlinear response. In Si, however, the situation
is more complex since TPA in the telecom band arises solely from indirect transi-
tions, while n2 (the real part of (3)) is dominated by direct transitions [67]. The direct
bandgap in Si is >3.2 eV while the indirect bandgap is only 1.1 eVwell below the
two-photon energy (1.6 eV) in the telecom band. For a:Si, it is likely that the
effective direct bandgap (or mobility gap, being amorphous) is decreased com-
pared to Si, thus increasing n2, while the contribution of indirect transitions is greatly
reduced, thus decreasing the TPA. This is entirely plausible since indirect transitions
involve phonons which depend on long range order which, even for optical phonons,
is clearly reduced in a:Si. Figure 2.5 shows the band structure of silicon and illus-
trates the principle of simultaneously increasing the nonlinearity and FOM.
A key goal for all-optical chips is to reduce device footprint and operating
power, and the dramatic improvement in the FOM of a-Si raises the possibility of
using slow-light structures [68, 69] to allow devices to operate at mW power levels
with sub-millimeter lengths.
Finally, we note that the broad use of the description CMOS compatible in this
context is intended to reflect a general compatibility in terms of growth tempera-
tures (<400 C) and materials that are familiar and used in the CMOS process
(silicon nitride, silicon oxynitride). It does not address the complexities and chal-
lenges of integrating optical and electronic devices with substantially different size
scales, nor does it address the challenges of adapting CMOS production lines to
optical device fabrication, both of which have been discussed at length [2022]. A
central issue in terms of integrating waveguides and nanowires with electronic
components is the rather thick nature of both the core and cladding lms. In this
regard, SiN, with a higher refractive index contrast of about 0.5, offers a signicant
advantage over Hydex where the index contrast is only about 0.3. Both glasses,
however, require noticeably thicker layers (both core and cladding) than SOI and
this is probably a key area where SOI out-performs these platforms. Nonetheless,
the concept of CMOS compatibility presented here, that these new platforms adhere
to, is a powerful one that will go a long way towards enabling the broad application
of CMOS techniques and manufacturing infrastructure to nonlinear photonic chips.
2.3 Low Power Nonlinear Optics in Ring Resonators
Optical microcavities, including ring resonators, can dramatically enhance nonlin-

ear processes [70] and offer a powerful approach to greatly reducing operating
powers for many applications such as frequency comb generation for spectroscopy
and metrology [70, 71]. Microcavities are particularly conducive to enhancing
FWM processes because FWM typically involves a pump, signal and idler beams
with frequencies (Pump, Idler, Signal) related by energy conservation:
xIdler 2xPump xSignal . This process can occur with (classical) or without
(spontaneous) a separate input signal at Signal, the classical effect being much
stronger and the basis for most all-optical signal processing. Very low CW oper-
ation was achieved in silica and single crystal micro-toroids and spheres with
Q-factors from 107 to 1010 [70, 71]. For microcavities, achieving phase-matching is
critical, which is equivalent to achieving equal resonance spacings (constant FSR)
(with due allowance for the Kerr induced resonance shifts [36]) that results in
pump, signal and idler all being in resonancea triple resonance that greatly
reduces the power requirement for the round-trip parametric gain to exceed the loss,
thus producing oscillation. A comprehensive review of the use of microcavities for
frequency comb generation is found in [48].
The rst demonstration of low power nonlinear optics in these CMOS platforms
was in Hydex [38] in 2008, where low power CW nonlinear optics was achieved
via FWM in integrated ring resonators. Figure 2.6 shows a four port micro-ring
resonator, radius 48 m, along with a scanning electron microscope (SEM) cross
section of the waveguide, having similar dimensions and composition to the spiral
structure. Figure 2.6 also shows the transmission spectrum of the resonator showing
a Q-factor of 65,000, a free-spectral range (FSR) of 575 GHz, and a full-width-
half-maximum (FWHM) of 3 GHz. The bus waveguides, buried in SiO2 beneath
the ring, couple light in and out of the resonator.
Through
Add
=Hydex
(a)
=SiO2
Ring
Resonator
Input Drop Bus waveguide
(b) (c)
SiO2
(d)
1.5m
Hydex 1.5m
Fig. 2.6 Hydex micro-ring resonator [38] with moderate Q factor = 65,000, resulting in a
bandwidth of 3 GHz, with an FSR of 575 GHz
Figure 2.7 shows the experimental setup as well as the basic physics behind
FWM in ring resonators. The experiments were conducted with the ring resonator
using only a few mW of CW pump power, with the signal tuned to an adjacent
resonance to the pump. FWM was achieved near 1558 nm with 5 mW of CW pump
power (in the waveguide) and a signal power of 570 W tuned to an adjacent
resonance. This yielded an idler that was almost exactly on resonance indicating
Pump Idler Signal
Dielectric medium
Pump
Signal
1st Idler
Pump
Signal
Pump
Idler = 2 Pump Signal

2 kPump = k Idler + kSignal
CW Pump Power = 5.6mW

Signal Power = 570W
Fig. 2.7 Low power FWM in Hydex ring resonators with moderate Q factor = 65,000, resulting in
a bandwidth of 3 GHz, with an FSR of 575 GHz [38]. Bottom External conversion efciency along
with estimated efciency of silicon RR reported in [72]
that the dispersion in the system was indeed negligible and conrmed that the
process was triply resonantly enhanced.
The experiments were performed with about 5 mW [38] of CW pump power
tuned to a ring resonance at 1553.38 nm (TM polarization) and directed to the
INPUT port, while a signal laser (TM polarization) with a power of 550 W was
tuned to an adjacent resonance at 1558.02 nm and directed to the ADD port.
Figure 2.7 shows the resulting output power spectra (TM polarization) as recorded
from both the THROUGH and the DROP ports after the pigtail; experiments carried
out with TE polarized modes led to similar results. Two detectable idlers were
generated by the FWM processthe rst-idler power in the bus waveguide
was 930 pW, whereas the second idler had a power of 100 pW. The ratio of the
powers for the two idlers agrees remarkably well with the ratio of the pump to the
signal power, Pidler(1)/Pidler(2) Ppump/Psignal, as expected for FWM. The rst idler
was almost exactly on resonance at 1548.74 nm, indicating that the dispersion in the
system is indeed negligible.
The theoretical conversion efciency takes into account the cavity enhance-
ment factor due to the ring geometry:
Pidler 4
g j2pRcj2 P2pump FEp FEs 2 FEi 2 2:1
Psignal
q

2 1 rl
p al pR
FEl rl 1 exp 2:2
2 1 rl al pR 2Finessel 2
where R is the ring radius, FE describes the eld enhancement of the ring and is
the mode linear loss coefcient for the mode . Finally, is the self-coupling
coefcient between ring and channels. The product (FEp)4(FEs)2(FEi)2 identies an
overall eld enhancement factor.
Figure 2.7 shows the external conversion efciency (including coupling losses),
along with the external conversion efciency of comparable experiments carried out
in silicon ring resonators [72], estimating the efciency from the coupling loss.
Although the conversion efciency of the Hydex device was predictably lower than
the SOI ring resonators due to the much lower , the negligible nonlinear absorption
potentially allows the scaling of pump powers to levels at which high bit rate
all-optical signal processing in silicon nanowires is typically performed [73, 74].
2.4 Microresonator-Based Frequency Combs
The area where these platforms have arguably had the greatest impact is in inte-
grated OPOs based on ring resonators. These devices have signicant potential for
many applications including spectroscopy and metrology as well as the ability to
provide an on-chip link between the RF and optical domains [75]. As discussed
above, micro-cavities enhance nonlinear optical processes, such as FWM involving

a continuous-wave (CW) pump, signal and idler beams with frequencies (Pump,
Idler, Signal) related by energy conservation: xIdler 2xPump xSignal . Achieving
phase-matching of the propagation constants for three interacting waves is essential
for efcient FWM, which for the microcavities is equivalent to having near-equal
resonance spacings, or a constant FSR, with due allowance for the Kerr-induced
resonance shifts [36]. This results in the pump, signal and idler waves all being in
resonancea triple resonance that greatly reduces the power requirement for the
round-trip parametric gain to exceed the loss, thus producing oscillation. Phase
matching, or low and anomalous waveguide dispersion, can be achieved with a
suitable design of the waveguide cross-section.
Reports in early 2010 of OPOs in both SiN [36] and Hydex [37] opened the door
to achieving these devices in practical CMOS-compatible integration platforms
with much lower Q factors than previous micro-cavity oscillators, and hence with
much less sensitivity to environmental perturbations and without the need for
delicate tapered bre coupling. This will allow these devices to benet not just the
scientic community but telecommunications, computing, and precision spectros-
copy and timekeeping.
Figure 2.8 shows a four port micro-ring resonator based in Hydex glass with a
Q factor of 1.2 million [37] along with a scanning electron microscope (SEM) of
the waveguide cross section and the corresponding optical transmission spectrum.
Figure 2.9 shows the experimental setup [37] for demonstrating an on-chip OPO
Fig. 2.8 Integrated ring resonators based in Hydex [37]. a Hydex four-port microring resonator
(bre pigtails not shown) with Q = 1.2 106. b SEM image of the Hydex ring resonator cross
section before depositing the upper cladding of SiO2; c bottom left linear transmission through ring
resonator from the INPUT to THROUGH port for TM and TE polarizations and d bottom right
high resolution transmission plot showing a resonance linewidth of 1.2 pm
Fig. 2.9 Experimental setup for OPO via external pumping with soft thermal locking [37]
using an externally pumped laser, highlighting the use of soft thermal locking to
achieve resonant coupling between the pump laser and the cavity resonances.
The optical frequency comb generated by the Hydex device (Fig. 2.10) exhibited
a very wide spacing of almost 60 nm when pumped at 1544.15 nm, in the
anomalous GVD regime (Fig. 2.3). The output power versus pump power shows a
very high single line differential slope efciency above threshold of 7.4 %, with a
maximum power of 9 mW achieved in all oscillating modes out of both ports,
representing a remarkable absolute total conversion efciency of 9 %. When
pumping at a slightly different wavelength closer to the zero-GVD wavelength (but
still anomalous), the device oscillated with signicantly a different spacing of
28.15 nm.
These observations are consistent with parametric gain based on a combination
of FWM and MI described above, where the spacing depends on the waveguide
dispersion characteristics and agrees well with calculations. Oscillation begins via
modulational instability (MI) gainessentially pure spontaneous (degenerate)
FWM (with only a pump present). Once oscillation is achieved, cascaded FWM
among different cavity modes takes over, resulting in the generation of a frequency
comb of precisely spaced modes in the frequency domain. However, the
enhancement of the elds in the cavity (all elds in the case of phase matching) that
is responsible for lowering the oscillation threshold, also enhances the nonlinear
losses, and it is primarily for this reason that oscillation in silicon ring resonators in
the telecom band where the FOM <1, has not been achieved.
Figure 2.11 shows the calculated MI gain curve when pumping at 1544 nm with
54 mW of power, using the experimentally measured dispersion curves. This shows
a gain peak at 1590 nmclose to the observed initial oscillation peaks. This
conrms that oscillation begins via MI gain near the peak. Once oscillation is
achieved, cascaded FWM can occur among many different cavity modes resulting
in the generation of a frequency comb of precisely spaced modes in frequency. This
illustrates the degree of freedom one can achieve in varying the frequency comb
spacingchiefly through dispersion engineering, and so one is not restricted by the
FSR of the resonator itself. The trade-off is that MI generated combs can themselves
further seed sub-combs that are poorly related in terms of coherence to the original
comb, limiting the degree to which modelocking, or ultrashort pulse generation, can
be achieved [47].
Fig. 2.10 (Top) Output spectra of Hydex hyper-parametric oscillator [37] with a pump power
varying from 50.8 mW (just above threshold) tuned to a resonance at 1544.15 nm (TM
polarization) to well above threshold at 101 mW. (Bottom) Output power vs pump power (at
1544.15 nm) in the drop port waveguide of the Hydex oscillator for a single oscillating mode at
1596.98 nm, showing a threshold of 54 mW with a differential slope efciency of 7.4 % (linear t
blue dashed line) above threshold
Fig. 2.11 Theoretical calculation of modulational instability (MI) gain [37] based on the
measured dispersion, at a pump power of 54 mW and pump wavelength near the zero dispersion
point at 1544 nm, showing a gain peak about 46 nm away from the pump wavelength
Fig. 2.12 Integrated OPO multiple wavelength sources in SiN [36] ring resonators [36]. a SEM
image of SiN microring resonator (58 m radius and Q = 500,000, FSR = 403 GHz) from [36]
coupled to a bus waveguide, with cross section height of 711 nm, base width of 1700 nm and a
sidewall angle of 20, giving anomalous GVD in the C-band and a zero-GVD point at 1610 nm;
b Output spectra of a 58-m-radius SiN ring resonator OPO with a single pump wavelength tuned
to a resonance at 1557.8 nm, numerous narrow linewidths at precisely dened wavelengths. The
87 generated wavelengths were equally spaced in frequency, with an FSR of 3.2 nm c spectra of a
20-m radius SiN ring resonator generating a comb with a different frequency spacing, pumped at
1561 nm and producing 21 wavelengths over a 200-nm span, with a spacing of 9.5 nm d output
power in the rst generated mode of SiN OPO versus pump power, showing a threshold of 50 mW
with a slope efciency of 2 %
Optical parametric oscillation was simultaneously reported in SiN [36] by res-

onantly pumping the rings with CW light near 1550 nm using a soft thermal lock
process in which the cavity heating is counteracted by diffusive cooling. Figure 2.12
shows a schematics of 2-port SiN micro-ring resonators: [36] a 58 m radius
resonator (Q factor = 500,000, FSR = 403 GHz) with dimensions designed to yield
anomalous GVD in the C-band with a zero GVD point at 1610 nm. Oscillation of
multiple lines over a very broad (>200 nm) wavelength range was achieved
(Fig. 2.12) at a pump threshold of 50 mW. Eighty-seven new frequencies were
generated between 1450 and 1750 nm, corresponding to wavelengths covering the
S, C, L and U communications bands. Several designs were employed with dif-
ferent ring radii, or FSR. A smaller ring with a Q factor of 100,000 generated
oscillation in 20 resonator modes when pumped with modest input powers
(150 mW) with THz mode spacing. These results represent a signicant step toward
a stabilized, robust integrated frequency comb source that can be scaled to other
wavelengths.
2.5 Advanced Frequency Comb Generation
Since these initial demonstrations of multiple wavelength oscillation, signicant

progress has been made in advanced comb generation, including both very wide
bandwidth octave spanning combs [45] and very low (sub 100 GHz) FSR spacing
combs [46].
The development of microresonator-based frequency combs with a free spectral
range (FSR) signicantly less than 100 GHz is critical to provide a direct link
between the optical and electrical domains in order to produce highly stable
microwave signals detectable with photodiodes. The challenge is that simple ring
geometries with sub-100 GHz FSR spacings do not t on typical single e-beam
elds and so novel ring geometries such as spirals need to be employed. Figure 2.13
shows spiral ring resonators with unique geometries for different FSRs below
100 GHz [46], all having a constant semicircular coupling region to enable critical
coupling between the bus and resonator, independent of path length. Bends in the
resonators had radii >100 m to ensure that bend induced dispersion was negligible,
a critical requirement for proper operation of the frequency comb. The experimental
spectra for 80, 40, and 20 GHz combs are shown in Fig. 2.13, typically requiring
about 2 W pump power to ll the entire comb spans.
Octave-spanning frequency combs are of great interest for spectroscopy, preci-
sion frequency metrology, and optical clocks and are highly desirable for comb self-
stabilization using f to 2f interferometry for precision measurement of absolute
optical frequencies [76]. Figure 2.14 shows an optical frequency comb in a SiN ring
resonator spanning more than an octave [45] from 1170 to 2350 nm, corresponding
to 128 THz, achieved by suitable dispersion engineering and employing higher
pump powers of up to 400 mW inside the waveguide, detuned slightly from a
Fig. 2.13 Advanced frequency combs in SiN ring resonators [46]. Sub-100GHz spacing SiN ring
resonators. Micrographs of the a 80 GHz, b 40 GHz, and c 20 GHz FSR resonators and the
corresponding linear transmission spectra (insets). A 2 nm section of each comb is inset in each
gure to illustrate the spacing of the comb lines. The nanowire cross section was 725 nm by 1650
nm, with a loaded Q of 100,000. Output spectra are d 300 nm wide for the d 80 GHz and e 40 GHz
FSR rings and f 200 nm for the 20 GHz FSR ring
cavity resonance. Figure 2.14 shows the simulated dispersion for nanowires with
varying widths (1200, 1650, and 2000 nm) indicating that large anomalous-GVD
bandwidths spanning nearly an octave are possible with appropriate design. These
results represent a signicant step toward a stabilized, robust integrated frequency
comb source that can be scaled to other wavelengths.
Fig. 2.14 Advanced frequency combs in SiN ring resonators [45]. a Top Optical spectrum of
octave-spanning parametric frequency comb generated in a SiN ring resonator and b Bottom
dispersion simulations for the fundamental TE mode of SiN waveguide with a height of 730 nm
and widths of 1200, 1650, and 2000 nm. The dashed curve shows the dispersion for bulk silicon
nitride
2.6 Supercontinuum Generation
By injecting ultrashort modelocked pulse trains into suitably designed waveguides

very broadband spectra can be generated. This super-continuum (SCG) spectra also
can result in octave-spanning frequency combs and has been of great interest for
spectroscopy, precision frequency metrology, and optical clocks and is highly
desirable for comb self-stabilization using f to 2f interferometry for precision mea-
surement of absolute optical frequencies [7680]. Wide bandwidth SCG has been
demonstrated in microstructured bers [8183], ChG waveguides [84], periodically
poled lithium niobate (PPLN) [85], and in Si [8688]. Hydex and SiN offer advantages
of much lower linear and nonlinear loss as well as transparency well into the visible.
Figure 2.15 compares the broadband frequency combs with SCG in Hydex
waveguides [89] where a spectral width >350 nm was achieved (limited by
experimental measurement capability), with that generated in 1100-nm wide SiN
nanowires [90] that yielded a spectrum spanning 1.6 octaves, from 665 to 2025 nm,
pumping at 1335 nm. The SiN results in particular represent the broadest recorded
SCG to date in a CMOS compatible chip. Both of these results were enabled by a
high effective nonlinearity, negligible TPA and most signicantly very flexible
dispersion engineering. SCG is signicantly enhanced if the pulse is launched near
a zero group-velocity dispersion (GVD) point or in the anomalous GVD regime.
The former minimizes temporal pulse broadening, thereby preserving high peak
Fig. 2.15 Supercontinuum generation in a Hydex waveguides [89] and b silicon nitride nanowires
[90] and c engineered dispersion in silicon nitride waveguides [90]
powers and thus maintaining a strong nonlinear interaction. The latter regime
enables soliton propagation, whose dynamics can contribute to spectral broadening.
The higher flexibility of dispersion engineering in SiN as compared to Hydex is
partly enabled by its higher available core/cladding index contrast and is probably
one of its most signicant advantages.
2.7 Comb Coherence and Dynamic Properties
As with conventional modelocked lasers, parametric frequency combs can also

potentially serve as sources of ultrashort laser pulses that, depending on the pump
laser and material system, can produce ultrashort pulses from the visible to the mid-
infrared at repetition rates in the GHz to THz regimes. The last few years have seen
signicant breakthroughs in understanding the dynamics and coherence behavior of
frequency comb formation [44, 47, 48]. These reports have revealed complex and
distinct paths to comb formation that can result in widely varying degrees of
coherence, wavelength spacing, and RF stability of these sources. This eld has
recently been highlighted by the achievement of an ultrastable, ultrashort optical

pulse source via modelocking [15] based on an integrated microcavity. Under-
standing and harnessing the coherence properties of these monolithic frequency
comb sources is crucial for exploiting their full potential in the temporal domain, as
well as for generally bringing this promising technology to practical fruition.
The rst investigation of the coherence properties of these on-chip oscillators
[47] focused on phase tuning of the comb via programmable optical pulse shaping
(Fig. 2.16). This line-by-line pulse shaping of microresonator frequency combs
was enabled by the relatively large mode spacings and represented a signicant
development in the eld of optical arbitrary waveform generation [7779]. In this
approach, transform limited can be realized for any spectral phase signature of
coherent combs by appropriately compensating the relative phase of the different
comb lines (Fig. 2.16). The ability to achieve successful pulse compression also
provides information on the passive stability of the frequency-dependent phase of
the Kerr combs. These time domain experiments revealed different pathways to
comb formation in terms of phase coherence properties, with the ability to effec-
tively modelock the combs varying signicantly depending on many factors such as
pump conditions and waveguide dispersion.
Recent studies of the dynamics of comb formation [44] (Fig. 2.17) have shown
that initial oscillation often begins at resonances near the parametric gain, or
modulational instability, peak with spacings varying widely from the FSR to values
as wide as 50100 nm. Cascaded FWM then replicates a comb with this initial
spacing. More complex dynamics arise when these comb lines themselves seed
their own mini-comb bands based on the local dispersion and pumping conditions,
often with a spacing at or near the cavity FSR in that wavelength region. While
these sub-combs maintain coherence within themselves, they are not coherently
related to the sub-combs generated by other lines of the initial, more widely spaced,
comb, and so this results in broadband comb with complex and generally limited
coherence. One approach to achieve high coherence is to design the device such
that the initial comb intrinsically oscillates at the FSR, producing a wideband comb
with high coherence properties. Figure 2.17 shows the output of a ring resonator
under different pumping conditions [44], dramatically showing the transition to
from an incoherent state where the frequency sub-combs are well separated and
hence incoherent, to a coherent state resulting in a much improved ability to lock
the modes to produce optical pulses, resulting in much lower RF noise in the output.
Further studies [49] of the temporal and optical and radio-frequency spectral
properties of a parametric frequency combs generated in SiN microresonators
demonstrate that the system undergoes a transition to a modelocked state and that
ultrashort pulse generation occurs. From a 25-nm ltered section of the 300-nm
comb spectrum, sub-200-fs pulses are observed and calculations indicate that the
pulse generation process is consistent with soliton modelocking, which is consistent
with very recent work involving comb generation in MgF2 microresonators [91]
and could be explained by the formation of temporal cavity solitons [91, 92], where
contributions from dispersion and loss are compensated by nonlinearity and a
coherent driving beam.
JFig. 2.16 a SEM image of SiN microring resonator from [47] (radius 40 m) along with b bre-
pigtailed coupling waveguide to resonator; c Optical transmission spectrum for TM polarized light
of ring resonator [46] including d high resolution spectrum for a mode at 1556.43 nm showing a
linewidth of 1.2 pm, corresponding to a loaded Q factor of 1.3 106 with a FSR of 4.8 nm. Bottom
Coherence and frequency comb formation dynamics. eh, Frequency comb coherence properties
from [47]. The output spectrum generated by a high-Q silicon nitride microring shows the ability
of type-I Kerr combs to perform pulse compression. Spectrum of the combs after processing with a
pulse shaper, together with the phase applied to the liquid-crystal modulator pixels required to
achieve optimum SHG signals e, f. Autocorrelation traces g, h corresponding to e, f [47]. The red
lines depict the compressed pulses after applying a phase correction, the blue lines are
uncompressed pulses and the black lines are calculated from the spectra shown in e and f by
assuming a flat spectral phase. The contrast ratios of the autocorrelations measured after phase
compensation are 14:1 (g) and 12:1 (h). Light grey traces show the range of the simulated
autocorrelation traces
Fig. 2.17 Studies of coherence evolution in SiN microring resonators [44] showing the transition
to a low phase noise Kerr comb. a Optical spectra of two microresonator comb states 1 (a) and 2
(b) (pump power 6W). State 2 evolves from state 1 when reducing the detuning of the pump laser.
A transition is observed from multiple sub-combs to a single (sub) comb over the bandwidth of the
Kerr comb reconstruction. In state 1, all sub-combs have the same mode spacing, but have different
offsets which differ by a constant of 66 MHz. c, d In the transmission from state 1 to state 2, the
amplitude noise peak resulting from the beating between overlapping offset sub-combs disappears
2.8 Ultrashort Pulsed Modelocked Lasers
The rst demonstration of stable modelocking of a frequency comb based on a

microresonator was recently achieved (Fig. 2.18) [15, 50, 93] by embedding the
resonator in an active ber loop. The unique aspect of this system is that the
microresonator is used as both a linear lter and as the nonlinear element. This
scheme, termed Filter-Driven Four-Wave-Mixing (FD-FWM), is an advancement
inspired by dissipative FWM (DFWM) [9496] where the nonlinear interaction
occurs in-bre and is then ltered separately by a linear FP lter. This new
approach is intrinsically more efcient and so allows for substantially reduced main
cavity lengths which in turn has enabled the achievement of highly stable operation
something that has so far eluded DFWM based lasers. A fundamental challenge
with DFWM lasers is that the main cavity mode spacing is typically much ner than
the microcavity, allowing many cavity modes to oscillate within each micro-reso-
nator mode, giving rise to so-called supermode instability. The FD-FWM
approach allows a substantial reduction in the main cavity length to the point where
only a very small number (<3) cavity modesand even a single modeexist
within each microresonator resonance, allowing for the possibility of only one main
cavity mode oscillating within each ring resonance. Hence, FD-FWM has achieved
highly stable operation at high repetition rates over a large range of operating
conditions, without a stabilization system, robust to external (i.e., thermal) pertur-
bations. In addition, it can potentially produce much narrower linewidths than
ultrashort cavity modelocked lasers because the long main cavity results in a much
smaller Schawlow-Towns phase noise limit [97]. Figure 2.19 compares the optical
spectra, time resolved optical waveforms (measured by autocorrelation) and radio-
frequency (RF) spectra of short and long cavity length lasers, clearly showing that
the RF spectra for the short cavity laser is highly stable whereas the long cavity
length laser not only shows large, long timescale, instabilities in the RF output, but
is also unstable on a short timescale where the optical autocorrelation traces show a
very limited contrast ratioa hallmark of unstable oscillation. Recently [50] stable
modelocked laser operation with two modes oscillating within each ring cavity was
demonstrated, which produced a highly pure RF tone over the modelocked train,
enabling high resolution RF spectral measurement of the optical linewidth, which
achieved the remarkably narrow width of about 10 kHz.
2.9 Ultrafast Phase Sensitive Pulse Measurement
Coherent optical communications [18, 19] have created a compelling need

for ultrafast phase-sensitive measurement techniques operating at milliwatt peak
power levels. Ultrafast optical signal measurements have been achieved using
time-lens temporal imaging on a silicon chip [13, 98] and waveguide-based
(a)
(b)
Unstable
Long Cavity = 33m
FSRc = 6MHz
(c) Oscillating modes

FSR Ring = 200GHz
Stable
Short Cavity = 3m
FSRc = 67MHz
Fig. 2.18 Ultrafast modelocked laser based on lter driven four-wave mixing in a modelocked
ber loop laser [15]. a Experimental conguration for ber loop modelocked laser based on lter
driven four wave mixing (FDFWM) based on a microring resonator, where the resonator performs
the dual function of linear ltering and nonlinear interaction. b, c oscillating modes of the laser
superimposed on the lter profle of the microresonator structure. b long (33 m) cavity length
showing 6 MHz spacing and c short (3 m) cavity showing much larger 67 MHz spacing
JFig. 2.19 Ultrafast modelocked laser based on lter driven four-wave mixing in a modelocked
ber loop laser [15]. Experimental optical output of unstable laser (a, b) and stable laser (c, d).
a optical spectrum and b autocorrelation trace of unstable (long length) laser, and c optical
spectrum and d autocorrelation trace of stable (short length) laser. Power output versus pump
power for unstable (e) and stable (f) lasers. The autocorrelation plots show theoretical calculations
(green) starting from the experimental optical spectra for a fully coherent and transform-limited
system calculated by considering each line of the experimental optical spectra as being perfectly
monochromatic and in-phase with the others, yielding an output pulse with a FWHM of 730 fs for
the highest excitation power condition. The measured autocorrelation for the long cavity laser
(b) shows a considerably higher background than the expected autocorrelation (50:1). Conversely,
the calculated autocorrelation trace for the short-cavity laser (d) perfectly matches the measured
trace, indicating stable modelocking, and corresponding to a transform-limited pulse width
(FWHM) of 2.3 ps with a peak to background ratio of 5:1. Stable oscillation was obtained by
adjusting the phase of the cavity modes for the short main cavity length laser relative to the ring
resonator modes via a free space delay line. Temporal waveforms measured via radio frequency
(RF) detectors (gi) and RF spectral (jl) output of the ber loop laser for different main cavity
lengths. (g, j) are for the long cavity (L = 33 m, FSR = 6.0 MHz) unstable laser, (h, k) are for the
short cavity laser (L = 3 m, FSR = 68.5 MHz) laser but with the main cavity modes detuned from
the microresonator mode, thus allowing more than one cavity mode to oscillate. (i, I) are for the
short cavity laser where the main cavity is tuned (via free space delay) such that the cavity modes
are aligned to the microcavity modethus restricting the laser to a single mode
Frequency-Resolved Optical Gating (FROG) [105], but these are either phase
insensitive or require waveguide-based tunable delay linesa difcult challenge.
Recently [14], a device capable of both amplitude and phase characterization of
ultrafast optical pulses was reported, operating via a novel variation of Spectral
Phase Interferometry for Direct Electric-Field Reconstruction (SPIDER) [100108]
based on FWM in a Hydex waveguides. Here, pulse reconstruction was obtained
with the aid of a synchronized incoherently related clock pulse.
SPIDER approaches are ultrafast, single-shot, and have a simple, direct and
robust phase retrieval procedure. However they traditionally have employed either
three-wave mixing (TWM) or linear electro-optic modulation, both of which require
a second-order nonlinearity that is absent in CMOS compatible platforms. Also,
typical SPIDER methods work best with optical pulses shorter than 100 fs (small
time-bandwidth products (TBP)) and peak powers >10 kW, and hence are not
ideally suited to telecommunications. The device reported in [14] measured pulses
with peak powers below 100 mW, with a frequency bandwidth greater than a
terahertz and on pulses as wide as 100 ps, yielding a record TBP >100 [105].
Figure 2.20 shows a schematic of the device. The pulse under test (PUT) is split
into two replicas and nonlinearly mixed with a highly chirped pump inside the
chip. The resulting output is then captured with a spectrometer and numerically
processed to extract the complete information (amplitude and phase) of the incident
pulse. Figure 2.21 compares the results from the X-SPIDER device using both a
standard algorithm and a new extended (Fresnel) phase-recovery extraction
Fig. 2.20 Experimental setup for the phase measurement of ultrafast pulses. Phase and amplitude
measurement of ultrafast optical pulses using spectral phase interferometry for direct electric-eld
reconstruction (X-SPIDER) device based on FWM in Hydex spiral waveguide [14]. The pulse
under test (PUT) is split in two replicas and nonlinearly mixed with a highly chirped pump inside
the chip (extracted from the same laser source in these experiments). The resulting output is
captured by a spectrometer and numerically processed by a suitable algorithm to extract the
complete information (amplitude and phase) of the incident pulse
algorithm [14, 108] designed for pulses with large TBP, with results from SHG-
based FROG measurements. As expected, for low TBP pulses (short-pulse regime)
the SPIDER device yielded identical results with both algorithms and agreed with
the FROG spectrogram. For large TBP pulses (highly chirped, long pulsewidths)
the X-SPIDER results obtained with the new algorithm agreed very well with the
FROG trace, whereas results using the standard phase-recovery algorithm were
unable to accurately reproduce the pulse.
SHORT PULSE LONG PULSE VERY LONG PULSE

Propagation ~0m SMF Propagation ~200m SMF Propagation ~700m SMF
Duration ~2ps Duration ~30ps Duration ~100ps
Time bandwidth product ~2 Time bandwidth product ~30 Time bandwidth product~100
(a) (b) (a) (b) (a) (b)
TIME SPECTRUM TIME SPECTRUM TIME SPECTRUM
(c) (d) (c) (d) (c) (d)
(e) (f) (g) (e) (f) (g) (e) (f) (g)
Fig. 2.21 Measurement of Ultrafast Pulses. X-SPIDER device based on FWM in Hydex spiral
waveguide [14]. Retrieved phase and amplitude proles for pulses with time bandwidth products
(TBP) of 2 (left), 30 (middle) and 100 (right). For each pulse, plots a are the temporal amplitude
output b the spectral amplitude c temporal phase and d spectral phase. The contour plots for each
pulse (short, long, ultra-long) are shown in eg. e was obtained by the X-SPIDER device using a
standard algorithm, f was obtained from the X-spider using a new algorithm [14] for large time-
bandwidth product (TBP) pulses, while g (plots at the right) were experimentally measured FROG
SHG measurements
2.10 Conclusions
We have reviewed the recent progress in CMOS-compatible, non c-Si integrated

optical platforms for nonlinear optics, focusing on SiN, Hydex glass, and amor-
phous Si. The combination negligible nonlinear (two-photon) absorption, low linear
loss, the ability to engineer dispersion, and moderately high nonlinearities has
enabled these new platforms to achieve new capabilities and novel devices not
possible in c-Si because of its low FOM. These include on-chip multiple wave-
length oscillators, optical frequency comb sources, ultrashort optical pulse gener-
ators, high-gain optical parametric ampliers, phase-sensitive optical pulse
measurement and many others. These platforms will likely have a signicant impact
on future all-optical devices, complementing the capabilities already offered by
silicon nanophotonics. The high performance, reliability, manufacturability of all of
these platforms combined with their intrinsic compatibility with electronic-chip
manufacturing (CMOS) raises the prospect of practical platforms for future low-
cost nonlinear all-optical PICs that offer a smaller footprint, lower energy con-
sumption and lower cost than present solutions.
References
1. B.J. Eggleton, D.J. Moss, S. Radic, in Optical Fiber Telecommunications V: Components

and Sub-systems, ed. by P.K. Ivan, T. Li, A.E. Willner. Nonlinear Optics in
Communications: From Crippling Impairment to Ultrafast Tools, Chapter 20 (Academic
Press, Oxford, 2008), pp. 759828
2. J. Leuthold, C. Koos, W. Freude, Nonlinear silicon photonics. Nat. Photonics 4, 535544
(2010)
3. H. Ji et al., 1.28-Tb/s Demultiplexing of an OTDM DPSK data signal using a silicon
waveguide. IEEE Photonics Technol. Lett. 22, 1762 (2010)
4. M. Galili et al., Breakthrough switching speed with all-optical chalcogenide glass chips:
640 Gb/s Demux. Opt. Express 17, 2182 (2009)
5. M. Foster et al., Broad-band optical parametric gain on a silicon chip. Nature 441, 04932
(2006)
6. H.S. Rong et al., A cascaded silicon Raman laser. Nat. Photonics 2, 170174 (2008)
7. W. Mathlouthi, H. Rong, M. Paniccia, Characterization of efcient wavelength conversion by
four-wave mixing in sub-micron silicon waveguides. Opt. Express 16, 1673516745 (2008)
8. F. Li, T.D. Vo, C. Husko, M. Pelusi, D.-X. Xu, A. Densmore, R. Ma, S. Janz, B.J. Eggleton,
D.J. Moss, All-optical XOR logic gate for 40 Gb/s DPSK signals via FWM in a silicon
nanowire. Opt. Express 19, 2036420371 (2011)
9. R. Salem et al., Signal regeneration using low-power FWM on a silicon chip. Nat. Photonics
2, 3538 (2008)
10. V.G. Taeed, M.Shokooh-Saremi, L.B.Fu, D.J. Moss, M.Rochette, I.C.M. Littler,
B.J. Eggleton, Y. Ruan, B.Luther-Davies. Integrated all-optical pulse regeneration in
chalcogenide waveguides. Opt. Lett. 30, 2900 (2005)
11. M. Pelusi, F. Luan, T. Vo, M. Lamont, S. Madden, D. Choi, B. Luther-Davies, B. Eggleton,
Photonic-chip-based radio-frequency spectrum analyser with terahertz bandwidth. Nat.
Photonics 3, 139 (2009)
12. B. Corcoran, T.D. Vo, M. Pelusi, C. Monat, D.-X. Xu, A. Densmore, R. Ma, S. Janz,
D.J. Moss, B.J. Eggleton, Silicon nanowire based radio-frequency spectrum analyser. Opt.
Express 18, 20190 (2010)
13. M. Foster et al., Silicon-chip-based ultrafast optical oscilloscope. Nature 456, 81 (2008)
14. A. Pasquazi, M. Peccianti, Y. Park, B.E. Little, S.T. Chu, R. Morandotti, J. Azaa,
D.J. Moss, Sub-picosecond phase-sensitive optical pulse characterization on a chip. Nat.
Photonics 5, 618 (2011)
15. M. Peccianti et al., Demonstration of an ultrafast nonlinear microcavity modelocked laser.
Nat. Commun. 3, 765 (2012)
16. M. Fridman, A. Farsi, Y. Okawachi et al., Demonstration of temporal cloaking. Nature 481,
6265 (2012)
17. R. Slavik et al., All-optical phase and amplitude regenerator for next-generation telecom
systems. Nat. Photonics 4, 690695 (2010)
18. P.J. Winzer, Beyond 100G ethernet. IEEE Commun. Mag. 48, 2630 (2010)
19. E. Rene-Jean et al., Capacity limits of optical ber networks. J. Lightwave Technol. 28, 662
(2010)
20. R. Won, M. Paniccia, Integrating silicon photonics. Nat. Photonics 4, 498499 (2010)
21. T. Baehr-Jones et al., Myths and rumours of silicon photonics. Nat. Photonics 6, 206208
(2012)
22. M. Hochberg, T. Baehr-Jones, Towards Fabless silicon photonics. Nat. Photonics 4, 492
(2010)
23. T.K. Liang, H.K. Tsang, Nonlinear absorption and Raman scattering in silicon-on-insulator
optical waveguides. IEEE J. Sel. Top. Quantum Electron. 10, 11491153 (2004)
24. A.S. Liu, H.S. Rong, M. Paniccia, Net optical gain in a low loss silicon-on-insulator
waveguide by stimulated Raman scattering. Opt. Express 12, 42614268 (2004)
25. F. Gholami et al., Third-order nonlinearity in silicon beyond 2350 nm. Appl. Phys. Lett. 99,
081102 (2011)
26. Q. Lin et al., Dispersion of silicon nonlinearities in the near infrared region. Appl. Phys. Lett.
91, 021111 (2007)
27. M. Dinu, F. Quochi, H. Garcia, Third-order nonlinearities in silicon at telecom wavelengths.
Appl. Phys. Lett. 82, 29542956 (2003)
28. S. Zlatanovic et al., Mid-infrared wavelength conversion in silicon waveguides using
ultracompact telecom-band-derived pump source. Nat. Photonics 4, 561564 (2010)
29. X. Liu et al., Mid-infrared optical parametric amplier using silicon nanophotonic
waveguides. Nat. Photonics 4, 557560 (2010)
30. X. Gai et.al., Nonlinear absorption and refraction in crystalline silicon in the midinfrared.
Laser Photon. Rev. 7, 10541064 (2013)
31. B.J. Eggleton, B. Luther-Davies, K. Richardson, Chalcogenide photonics. Nat. Photonics 5,
141148 (2011)
32. J.S. Aitchison et al., The nonlinear optical properties of AlGaAs at the half band gap. IEEE
33. K. Dolgaleva et al., Compact highly-nonlinear AlGaAs waveguides for efcient wavelength
conversion. Opt. Express 19, 1244012455 (2011)
34. K. Ikeda, R.E. Saperstein, N. Alic et al., Thermal and Kerr nonlinear properties of plasma-
deposited silicon nitride/silicon dioxide waveguides. Opt. Express 16, 1298712994 (2008)
35. D.T.H. Tan, K. Ikeda, P.C. Sun, Y. Fainman, Group velocity dispersion and self phase
modulation in silicon nitride waveguides. Appl. Phys. Lett. 96, 061101 (2010)
36. J.S. Levy et al., CMOS-compatible multiple-wavelength oscillator for on-chip optical
interconnects. Nat. Photonics 4, 3741 (2010)
37. L. Razzari et al., CMOS compatible integrated optical hyper-parametric oscillator. Nat.
Photonics 4, 4144 (2010)
38. M. Ferrera et al., Low-power continuous-wave nonlinear optics in doped silica glass
integrated waveguide structures. Nat. Photonics 2, 737740 (2008)
39. C.H. Henry, R.F. Kazarinov, H.J. Lee, K.J. Orlowsky, L.E. Katz, Low loss Si3N4-SiO2
optical waveguides on Si. Appl. Opt. 26, 26212624 (1987)
40. N. Daldosso et al., Comparison among various Si3N4 waveguide geometries grown within a
CMOS fabrication pilot line. IEEE J. Ligthtwave Technol. 22, 17341740 (2004)
41. B.E. Little et al., Very high-order microring resonator lters for WDM applications. IEEE
Photon. Technol. Lett. 16, 22632265 (2004)
42. D. Duchesne et al., Efcient self-phase modulation in low loss, high index doped silica glass
integrated waveguides. Opt. Express 17, 1865 (2009)
43. J.S. Levy et al., High-performance silicon nitride based multiple wavelength source. IEEE
Photonics Technol. Lett. 24, 13751377 (2012)
44. T. Herr et al., Universal formation dynamics and noise of Kerr-frequency combs in
microresonators. Nat. Photonics 6, 480487 (2012)
45. Y. Okawachi et al., Octave-spanning frequency comb generation in a silicon nitride
chip. Opt. Lett. 36, 33983400 (2011)
46. A. Johnson et al., Chip-based frequency combs with sub-100 GHz repetition rates. Opt. Lett.
37, 875877 (2012)
47. F. Ferdous et al., Spectral line-by-line pulse shaping of on-chip microresonator frequency
combs. Nat. Photonics 5, 770776 (2011)
48. T.J. Kippenberg, R. Holzwarth, S.A. Diddams, Microresonator-based optical frequency
combs. Science 332, 555559 (2011)
49. Y. Saha et al., Modelocking and femtosecond pulse generation in chip-based frequency
combs. arXiv:1211.1096 (2012)
50. M. Peccianti et al., Dual frequency comb mode-locked laser based on an integrated nonlinear
microring resonator. Opt. Express 20, 2735527362 (2012)
51. J.S. Levy et al., Harmonic generation in silicon nitride ring resonators. Opt. Express 19,
1141511421 (2011)
52. T.M. Monro et al., Progress in microstructured optical bers. Annu. Rev. Mater. Res. 36,
467495 (2006)
53. A. Pasquazi et al., Efcient wavelength conversion and net parametric gain via four wave
mixing in a high index doped silica waveguide. Opt. Express 18, 76347641 (2010)
54. M. Ferrera et al., Low power four-wave mixing in an integrated, microring resonator with
Q = 1.2 million. Opt. Express 17, 1409814103 (2009)
55. D.J. Moss, H. Driel, J. Sipe, Third harmonic generation as a structural diagnostic of ion
implanted amorphous and crystalline silicon. Appl. Phys. Lett. 48, 1150 (1986)
56. R. Orobtchouk, S. Jeannot, B. Han, T. Benyattou, J.M. Fedeli, P. Mur, in Proceedings of
SPIE 6183, conference on Integrated Optics, Silicon Photonics, and Photonic Integrated
Circuits. Ultra compact optical link made in amorphous silicon waveguide (Strasbourg,
France, 2006) paper 618304
57. J.M. Fedeli, M. Migette, L. Di Cioccio, L. El Melhaoui, R. Orobtchouk, C. Seassal,
P. Rojo-Romeo, F. Mandorlo, D. Marris-Morini, L. Vivien, in Proceedings of 3rd IEEE
International Conference on Group IV Photonics. Incorporation of a photonic layer at the
metallization levels of a CMOS circuit (2006) pp. 200202
58. K. Ikeda, Y.M. Shen, Y. Fainman, Enhanced optical nonlinearity in amorphous silicon and
its application to waveguide devices. Opt. Express 15, 1776117771 (2008)
59. Y. Shoji et al., Ultrafast nonlinear effects in hydrogenated amorphous silicon wire
60. K. Narayanan, S.F. Preble, Optical nonlinearities in hydrogenated amorphous silicon
waveguides. Opt. Express 18, 89989905 (2010)
61. S. Suda et al., Pattern-effect-free all-optical wavelength conversion using a hydrogenated
amorphous silicon waveguide with ultra-fast carrier decay. Opt. Lett. 37, 13821384 (2012)
62. K.-Y. Wang, A.C. Foster, Ultralow power continuous-wave frequency conversion in
hydrogenated amorphous silicon waveguides. Opt. Lett. 37, 13311333 (2012)
63. B. Kuyken et al., On-chip parametric amplication with 26.5 dB gain at telecommunication
wavelengths using CMOS-compatible hydrogenated amorphous silicon waveguides. Opt.
Lett. 36, 552554 (2011)
64. B. Kuyken et al., Nonlinear properties of and nonlinear processing in hydrogenated
amorphous silicon waveguides. Opt. Express 19, B146B153 (2011)
65. C. Grillet et al., Amorphous silicon nanowires combining high nonlinearity. FOM and optical
stability. Optics Express 20, 22609 (2012)
66. L. Caspani et al., Optical frequency conversion in integrated devices. JOSA B 28(12),
A66A81 (2011)
67. D. Moss, E. Ghahramani, J.E. Sipe, H.M. van Driel, Band structure calculation of dispersion
and anisotropy in (3) (3: ,,) for third harmonic generation in Si, Ge, and GaAs. Phys.
Rev. B 41, 1542 (1990)
68. B. Corcoran et al., Optical signal processing on a silicon chip at 640 Gb/s using slow-light.
Opt. Express 18, 77707781 (2010)
69. Corcoran B et al., Slow light enhanced nonlinear optics in dispersion engineered slow-light
silicon photonic crystal waveguides. IEEE J. Sel. Top. Quantum Electron. 16, 344356
(2010)
70. P. DelHaye et al., Optical frequency comb generation from a monolithic microresonator.
Nature 450, 12141217 (2007)
71. I.S. Grudinin, N. Yu, L. Maleki, Generation of optical frequency combs with a CaF2
resonator. Opt. Lett. 34, 878880 (2009)
72. A.C. Turner, M.A. Foster, A.L. Gaeta, M. Lipson, Ultra-low power parametric frequency
conversion in a silicon microring resonator. Opt. Express 16, 48814887 (2008)
73. H. Fukuta et al., Four-wave mixing in silicon wire waveguides. Opt. Express 13, 46294637
(2005)
74. H. Rong, Y. Kuo, A. Liu, M. Paniccia, High efciency wavelength conversion of 10 Gb/s
data in silicon waveguides. Opt. Express 14, 11821188 (2006)
75. A. Bartels et al., Femtosecond-laser-based synthesis of ultrastable microwave signals from

optical frequency references. Opt. Lett. 30, 667669 (2005)
76. Th. Udem, R. Holzwarth, T.W. Hnsch, Optical frequency metrology. Nature 416, 233237
(2002)
77. S.T. Cundiff, A.M. Weiner, Optical arbitrary waveform generation. Nat. Photonics 4,
760766 (2010)
78. Z. Jiang, C.B. Huang, D.E. Leaird, A.M. Weiner, Optical arbitrary waveform processing of
more than 100 spectral comb lines. Nat. Photonics 1, 463467 (2007)
79. M.H. Khan et al., Ultrabroad-bandwidth arbitrary radiofrequency waveform generation with
a silicon photonic chip-based spectral shaper. Nat. Photonics 4, 117122 (2010)
80. D. Jones et al., Carrier-envelope phase control of femtosecond mode-locked lasers and direct
optical frequency synthesis. Science 288, 635639 (2000)
81. J. K. Ranka, R. S. Windeler, A. J. Stentz, Visible continuum generation in air-silica
microstructure optical bers with anomalous dispersion at 800 nm. Opt. Lett. 25, 2527
(2000)
82. J.M. Dudley, G. Genty, S. Coen, Supercontinuum generation in photonic crystal ber Rev.
Mod. Phys. 78, 11351184 (2006)
83. J. Herrmann et al., Experimental evidence for supercontinuum generation by ssion of
higher-order solitons in photonic bers. Phys. Rev. Lett. 88, 173901 (2002)
84. M.R. Lamont, B. Luther-Davies, D.-Y. Choi, S. Madden, B.J. Eggleton, Supercontinuum
generation in dispersion engineered highly nonlinear ( = 10/W/m) As(2)S(3) chalcogenide
planar waveguide. Opt. Express 16, 14938 (2008)
85. C. Phillips, J. Jiang, C. Langrock, M.M. Fejer, M.E. Fermann, Supercontinuum generation in
quasi-phase-matched LiNbO3 waveguide pumped by a Tm-doped ber laser system. Opt.
Lett. 36, 39123914 (2011)
86. M.A. Foster, A.L. Gaeta, Ultra-low threshold supercontinuum generation in sub-wavelength
87. P. Koonath, D.R. Solli, B. Jalali, Limiting nature of continuum generation in silicon. Appl.
Phys. Lett. 93, 091114 (2008)
88. I.-W. Hsieh et al., Supercontinuum generation in silicon photonic wires. Opt. Express 15,
1524215249 (2007)
89. D. Duchesne et al., Supercontinuum generation in a high index doped silica glass spiral
90. R. Halir et al., Ultrabroadband supercontinuum generation in a CMOS-compatible platform.
Opt. Lett. 37, 1685 (2012)
91. T. Herr, V. Brasch, M.L. Gorodetsky, T.J. Kippenberg. Soliton Mode-locking in Optical
Microresonators, (2012). arXiv:1211.0733
92. F. Leo, S. Coen, P. Kockaert, S.-P. Gorza, P. Emplit, M. Haelterman, Temporal cavity
solitons in one-dimensional Kerr media as bits in an all-optical buffer. Nature Photon.
4, 471476 (2010)
93. A. Pasquazi, L. Caspani, M. Peccianti, M. Clerici, L. Razzari, M. Ferrera, D.Duchesne,
B. Little, S.T. Chu, D.J. Moss, R. Morandotti, Self-locked optical parametric oscillation in a
CMOS compatible microring resonator: a route to robust optical frequency comb generation
on a chip. Opt. Express 21(11), 1333313341 (2013)
94. M. Quiroga-Teixeiro et al., Passive mode locking by dissipative four-wave mixing. J. Opt.
Soc. Am. B 15, 13151321 (1998)
95. T. Sylvestre et al., Self-induced modulational instability laser revisited: normal dispersion
and dark-pulse train generation. Opt. Lett. 27, 482484 (2002)
96. M. Yoshida, A. Ono, M. Nakazawa, 10 GHz regeneratively modelocked semiconductor
optical amplier bre ring laser and its linewidth characteristics. Opt. Lett. 32, 35133515
(2007)
97. A.L. Schawlow, C.H. Townes, Infrared and optical masers. Phys. Rev. 112, 19401949
(1958)
98. A. Pasquazi A et al., Time-lens measurement of subpicosecond optical pulses in CMOS

compatible high-index glass waveguides. IEEE J. Sel. Top. Quantum Electron. 18, 629636
(2012)
99. R. Trebino, Frequency Resolved Optical Gating: The Measurement of Ultrashort Laser
Pulses (Kluwer Academic, Boston, 2002)
100. C. Iaconis, I.A. Walmsley, Spectral phase interferometry for direct electric eld
reconstruction of ultrashort optical pulses. Opt. Lett. 23, 792794 (1998)
101. C. Iaconis, I.A. Walmsley, Self-referencing spectral interferometry for measuring ultrashort
optical pulses. IEEE J. Quant. Electron. 35, 501509 (1999)
102. L. Gallmann et al., Characterization of sub-6-fs optical pulses with spectral phase
interferometry for direct electric-eld reconstruction. Opt. Lett. 24, 13141316 (1999)
103. C. Dorrer et al., Single-shot real-time characterization of chirped-pulse amplication systems
by spectral phase interferometry for direct electric-eld reconstruction. Opt. Lett. 24,
16441646 (1999)
104. J. Bromage, C. Dorrer, I.A. Begishev, N.G. Usechak, J.D. Zuegel, Highly sensitive, single-
shot characterization for pulse widths from 0.4 to 85 ps using electro-optic shearing
interferometry. Opt. Lett. 31, 35233525 (2006)
105. C. Dorrer, J. Bromage, High-sensitivity optical pulse characterization using Sagnac electro-
optic spectral shearing interferometry. Opt. Lett. 35, 13531355 (2010)
106. I.A. Walmsley, C. Dorrer, Characterization of ultrashort electromagnetic pulses. Adv. Opt.
Photon. 1, 308437 (2009)
107. M.E. Anderson, A. Monmayrant, S.P. Gorza, P. Wasylczyk, I.A. Walmsley, SPIDER: a
decade of measuring ultrashort pulses. Laser Phys. Lett. 5, 259266 (2008)
108. A. Pasquazi, M. Peccianti, J. Azana, D.J. Moss, R. Morandotti, FLEA: Fresnel-limited
extraction algorithm for SPIDER. Opt. Express 21(5), 57435758 (2013)
Chapter 3
Optical Guided Wave Switching
Costantino De Angelis, Daniele Modotto, Andrea Locatelli

and Stefan Wabnitz
Abstract Optical switching is a key functionality for enabling transparent

all-optical networks. We present an overview of optical switching devices, based on
either optical or electrical control signals, which permit to avoid the necessity of
optics-electronics-optics conversion. We describe the basic principles of various
guided wave optical switching devices, which exploit either relatively long inter-
action lengths in order to reduce the operating power requirements, or strong
transverse connement to reduce device dimensions. These devices include non-
linear mode couplers and interferometers based on optical bers, as well as inte-
grated waveguides based on photonic crystal structures or surface wave interactions
in novel materials such as graphene.
3.1 Introduction: Optical Switching Using

Guided-Waves
In recent years, telecommunication networks have witnessed a dramatic increase of

capacity, mostly driven by the exponential growth of IP trafc. Researchers had to
tackle and solve several problems, and certainly the challenge of realizing trans-
parent all optical switching was among the most important issues which have been
addressed. In this framework, the goal is to use optics not only at the transmission
level, but also at the switching level; this in turn requires to conceive devices which
can perform switching directly in the optical domain, thus overtaking the
unavoidable limitations of switching at the electronic level. Remarkably the goal
can be achieved using either an optical or an electrical control; the common
denominator is not the physical nature of the control signal, but the possibility of
C. De Angelis D. Modotto A. Locatelli S. Wabnitz (&)

Dipartimento di Ingegneria dellInformazione, Universit degli Studi di Brescia,
Via Branze 38, Brescia 25123, Italy
e-mail: stefan.wabnitz@unibs.it

72 C. De Angelis et al.
avoiding optics-electronics-optics (OEO) conversions which lead to inefcient and

nontransparent switching.
In particular, in this paper we review some basic ideas and devices for all-optical
switching in guided wave geometries, as this scenario offers important possibilities
for tuning the light connement, and thus exploring more efcient light-matter
interactions. In Sect. 3.2 we will discuss some relevant examples of all-optical
switching devices exploiting optical bers; in Sect. 3.3 we will move into the eld
of integrated optics and we will discuss optical switching in photonic crystal
waveguides, as well as using surface waves as a tool to enhance light matter-
interactions in novel nonlinear materials such as graphene.
3.2 All-Optical Pulse Switching in Optical Fibers
We review the different implementations of switching in optical bers, including

polarization effects in both high and low-birefringence bers, and nonlinear optical
loop mirrors. Both self-switching and two-beam, phase or cross-phase modulation
(XPM) controlled switching will be analyzed.
3.2.1 Nonlinear Mode Coupling
We start by considering the continuous wave (CW) case and focus our attention on
the linear and nonlinear coupling between the two modes of a structure composed
of two waveguides placed in close proximity. The basic model to describe the
evolution of the modal amplitudes A1;2 along the coupler [1] is given by the
following coupled equations:
dA1 h i
i CA2 c jA1 j2 qjA2 j2 A1 0
dz
dA2 h i
i CA1 c jA2 j2 qjA1 j2 A2 0 3:1
dz
where C is the linear coupling coefcient, c is the nonlinear coefcient and q is the
XPM factor. The previous equations were introduced for the nonlinear coherent
directional coupler (NLDC) [2, 3], and they work well for dual-core bers and for
integrated couplers [4].
In the case of a birefringent optical ber the XPM factor is 2=3 and nonlinear
four-wave mixing terms must be included, as well [57]:
3 Optical Guided Wave Switching 73

dax 1 2 1
i ax jax j jay j ax a2y ax 0
2 2
df 2 3 3

day 1 2 1
i ay jay j jax j ay a2x ay 0
2 2
3:2
df 2 3 3
where Ax;y are the complex envelopes of the fundamental modes polarized along the
x; y principal axes of the ber with propagation constants bx;y , respectively.
p
Moreover, ax;y c=Db expfibx by z=2gAx;y , f Dbz is a dimensionless
distance, Db bx by 2p=Lb is the linear ber birefringence and Lb the beat
length. Let us express (3.2) in a more compact form in terms of the circularly
p
polarized components a ax iay = 2 as
da 1 2h i
i a ja j2 2ja j2 a 0
df 2 3
da 1 2h i
i a ja j2 2ja j2 a 0 3:3
df 2 3
At relatively low powers, ber birefringence leads to the periodic exchange of

power among the two circular polarizations: the spatial period of such linear cou-
pling is equal to the linear beat length Lb 2Lc (Lc is the linear coupling length).
In order to better visualize the action of nonlinear coupling, it proves convenient
to rewrite (3.2) in terms of the real Stokes parameters si Si =S0 , where
S0 P jax j2 jay j2 , S1 jax j2 jay j2 , S2 ax ay c:c:, and S3 iax ay c:c::
ds1
2ps2 s3
df
ds2
s3 2ps1 s3
df
ds3
s2 3:4
df
where we dened the dimensionless input power p P=Pc , with the critical power
level Pc 3Db=2c. In vector notation, (3.4) can be written as
ds
XL XNL s s 3:5
df
which describes polarization evolution on the Poincar sphere as the motion of a

rigid body subject to the sum of the xed and the position-dependent angular
velocities XL 1; 0; 0 and XNL 0; 0; ps3 , respectively [6, 8]. Figure 3.1a
Fig. 3.1 a Poincar sphere description of switching in a nonlinear coupler; b evolution with
distance of power in the input mode for different input power levels (Figures adapted from [6, 8])
illustrates the radically different behavior of the polarization evolution trajectories

of the tip of the Stokes vector s as the input power is below or above the critical
power Pc . Below the critical power, all points on the Poincar sphere rotate around
the axis dened by XL : as a result, the two circular polarizations periodically
exchange their power along the ber length. On the other hand, for P [ Pc the
mode polarized along the fast axis of the ber (i.e., whose Stokes vector is
sF 1; 0; 0) loses its spatial stability and becomes an unstable saddle (see right
part of Fig. 3.1a) [6, 810]: as it is shown in Fig. 3.1b, an input right-handed
circular polarization (whose power is P jA j2 ) no longer periodically couples
into the orthogonal left-handed polarization whenever P [ 2Pc .
3.2.2 Nonlinear Fiber Couplers
The NLDC is a basic device which permits all-optical routing and switching
operations [2, 3]. Experiments have demonstrated nonlinear transmission and self-
switching in ber-optic NLDCs, such as: dual-core bers [1115], low birefrin-
gence bers [16], and periodically twisted birefringent lters [17, 18]. Even though
ber optic NLDCs exhibit ultrafast nonlinear response, whenever their length
L Lb , the associated switching power P Pc is relatively high, owing to the
weak nonlinearity of silica. Indeed, because of imperfections in ber fabrication (in
the case of dual core couplers) or of random ber birefringence (for polarization
couplers), beat lengths exceeding one or a few meters are not possible in practice.
Earlier experiments with ber NLDCs involved just one or two linear coupling
distances Lc . In this case, the associated nonlinear transmission exhibits a single
switching power, that is inversely proportional to Lc .
Nevertheless, full switching in a ber NLDC of any length L is still possible
when operating in the multi-beatlength regime, i.e., whenever L Lb [19, 20]. The
main advantage of the multi-beatlength NLDC is that the relevant switching power
is inversely proportional to the total length of the ber L. Therefore, the switching
performance is comparable to that of other nonlinear ber switches such as the
nonlinear Mach-Zehnder interferometer (MZI) and the nonlinear optical loop mirror
(NOLM). Using a circularly polarized beam of power P at the input of a bire-
fringent ber, the exact solution of (3.3) leads to the following expression for the
output power P in the same circular polarization
P
P 1 cnpL=Lc j~p 3:6
2
where ~
p p=2 and cn is a Jacobi elliptic function. From (3.6), one obtains that the
nonlinear period of polarization coupling between the two circular polarizations is
2Lc K~p
Lnc ~p 3:7
p
where K is the complete elliptic integral of the rst kind. Whenever ~p\1, one
obtains that

p ~p2
K~p 1 O~p2 3:8
2 4
Therefore for ~
p
1 (3.6) reduces to
P P cos2 pL=2Lnc ~p 3:9
The associated ONOFF switching power ps corresponds to the power-induced

increase of the coupling distance which leads to a p=2 shift in the argument of the
p
cosine in (3.9). One obtains ps 2 2Lb =L, i.e., in real units
12p
Ps p 3:10
c 2Lb L
The switching power of a ber-based NLDC operating in the multi-beatlength

regime may be further reduced by injecting an input elliptically (as opposed to
circularly) polarized beam [21, 22]. Let us set s1 f 0 e cos2/,
p
s1 f 0 e sin2/, and s3 f 0 1 e2 . From the exact solution of (3.3),
one obtains P P1 s3 L; p=2 for the right handed circular polarization
component at the output of a bre of length L, where
q
s3 1 e2 cos2/2 f pcos2prpL=Lb 3:11
~b p
and f p 1 pecos2/=2. In this case, the nonlinear beat length L
Lb =f p Lb 1 pe cos2/=2. By supposing jcos2/j 1, the switching
power ps is dened by 2prp 1L=Lb p, so that ps Lb =eL. In real units, the

switching power reads as
3p
Ps 3:12
ceL
In Fig. 3.2a we show an example of nonlinear CW transmissions computed from

(3.9) at the output of a L 20Lb long birefringent ber. Here we plot the trans-
mission dependence on input power for different input polarization angles /, and
we have set e 0:37, cL 1. As can be seen from Fig. 3.2a, the switching power
agrees well with the estimate Ps 25 W from (3.12). In Fig. 3.2 we also display the
change of the nonlinear transmission as the input polarization angle / is varied, for
L = 1020Lb and p 0:15.
An experimental conrmation of all-optical power and polarization controlled
switching in a multi-beatlength ber NLDC was performed using 205 m of York
ultralow-birefringence spun ber [19, 23]. The ber was wound on a drum with
radius of 15 cm, in order to introduce a bending-induced linear birefringence with
Lb 20 m, so that L 10Lb . By increasing the input power of a right-handed
circularly polarized beam from low values up to the switching power, an accu-
mulated nonlinear variation of the beat length equal to Lc or 10 m (or 5 % nonlinear
variation per beat length Lb ) was obtained.
Quasi-CW operating conditions were achieved using a mode-locked 100 ps
input pulse train from an Nd-YAG laser operating at 1:06 m. At the ber output,
the two circular polarization components of the emerging pulses were the two
output channels of the birefringent ber NLDC. These components were separated
by means of a calibrated Babinet-Soleil compensator, followed by a Wollaston
prism.
Figure 3.3a, b shows the streak camera pictures of the pulse proles emerging
from the bar (input) and the crossed circular polarization components at the ber
output, for an input average power of 320 mW [19]. As it can be seen, the center
(a) 1 (b) 1
=0 L/L =10
b
0.8 2=/4 0.8 L/L =20
b
p=0.15
Transmission
Transmission
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 20 40 60 80 100 0 20 40 60 80
Input Power (W) Input Polarization Angle (degrees)
Fig. 3.2 Theoretical CW transmissions at the output of L 20Lb long birefringent ber:
a transmission dependence on input power for different input polarization angles; b transmission
variation with input angle for two values of the total number of linear beat lengths Lb
(a) (c) 1
0.9
0.8
0.7
Transmission
0.6
Cross State
0.5 Bar State
(b) 0.4
0.3
0.2
0.1
0
0 100 200 300 400 500 600 700
Input Average Power (mW)
Fig. 3.3 Experimental pulse energy transmissions from orthogonal polarizations at the output of a
10 beat length long birefringent ber: a, b streak-camera traces of pulses emerging from the input
(bar state) and the orthogonal (cross state) polarizations; c average energy transmissions from
cross and bar states versus input average power (Figures adapted from [19, 22])
and the wings of the pulse have opposite handedness, leading to a symmetric pulse
breakup.
Figure 3.3c shows the measured circular polarization transmissions at the ber
output, as a function of the input average power into the ber [22]. Because of the
slow response time of the photodetectors, the transmission displays the average
power in each output polarization. At low powers, the input circular polarization
emerges from the ber almost unchanged. About 50 % switching of the trans-
mission is observed at about 250 mW: the corresponding peak pulse power is of
about 20 W, in relatively good agreement with the estimate of (3.12), since in the
experiment the condition cL 1 is veried.
3.2.3 Nonlinear Mach-Zehnder Interferometers
A relatively simple nonlinear switch may be implemented using an all-ber-based

Mach-Zehnder interferometer (MZI): here two 3-dB linear directional couplers are
used to split the input signal and then recombine the two arms of the interferometer
of lengths L1 and L2 , respectively (see Fig. 3.4). The components of the signal in
the two arms of the MZI thus experience different overall linear and SPM induced
phase shifts [2428]. The transmission through the bar port of the MZI reads as
TP sin2 D/L D/NL 3:13
where D/NL cPL1 L2 . Alternatively, a different nonlinear phase shift in the

two arms of the MZI may be induced whenever L1 L2 by using an unbalanced
Fig. 3.4 Structure of a ber based nonlinear MZ interferometer switch exploiting SPM or XPM
(in the presence of a control pump)
linear directional coupler, so that the two signal components have different powers.
A different implementation of a MZI type of switch consists of a highly birefringent
ber, where the two signal paths are represented by the two ber modes which are
orthogonally polarized along the principal axes of the ber. For an input beam of
power P that is initially linearly polarized with an orientation at the angle h with
respect to the x-axis of a birefringent ber of length L, one has Px P0 cos2 h,
Py P0 sin2 h, and D/NL cPL cos2h=3.
Besides SPM-activated switching, a ber MZI can be used to switch a signal
among its two output ports through XPM by injecting a control pump at a different
wavelength, that shifts the signal phase in the upper arm of the interferometer (see
Fig. 3.4). In the presence of a control pump composed by a pulse train, the input
CW signal may also be converted into a pulse train. The drawback of a ber MZI is
the presence of two separate paths for the signal, which make it sensitive to
environmental perturbations, so that active stabilization of the ber lengths may be
required [25].
3.2.4 Nonlinear Loop Mirrors
The environmental stability of an interferometric ber switch may be ensured by

using a Sagnac loop conguration [2833]: here the transmitted signal results from
the coherent superposition at the directional coupler output port of two signals that
have traveled in opposite directions around the same loop of ber (see Fig. 3.5a).
Clearly, whenever the directional coupler equally splits the input signal (i.e., a 3-dB
coupler is used), there is no differential nonlinear phase shift among the two signal
components that travel in opposite directions around the loop, and the Sagnac loop
acts as a perfect mirror by reflecting all of the incoming signal: the transmitted
signal vanishes. On the other hand, whenever an asymmetric coupler is used, there
is a differential nonlinear phase shift which leads to a signal to be transmitted at
high powers. Thus the Sagnac loop of Fig. 3.5a is also known as nonlinear optical
loop mirror (NOLM).
(a) (b) 1
K=0.4
K=0.45
K=0.48
0.8
Transmissivity
0.6
0.4
0.2
0
0 20 40 60 80 100
Input Power (W)
Fig. 3.5 a Structure of a ber based NOLM and b associated power dependence of transmission
for different power split coefcients K of the coupler
The nonlinear transmission of the NOLM can be simply obtained as follows.

The input amplitude of the waves traveling clockwise and counter-clockwise in the
p p
loop reads as AiCW K Ain and AiCCW i 1 K Ain , respectively, where K is the
power splitting ratio of the coupler and Ain is the input signal amplitude. After one
round-trip through the loop, the two elds traveling in opposite directions acquire
equal linear but different nonlinear (as determined by both SPM and XPM) phase
shifts, so that at the input of the coupler one has the two elds
n o
AoCW AiCW exp ibL ic jAiCW j2 2jAiCCW j2 L
n o
AoCCW AiCCW exp ibL ic jAiCCW j2 2jAiCW j2 L 3:14
where b is the linear propagation constant in the ber loop of length L. At the
p p
coupler output, one obtains At K AoCW i 1 K AoCCW , so that the transmis-
sivity of the NOLM reads as
TP jAt j2 =jAi j2 1 4K 1 K cos2 K 0:5cPin L 3:15
where Pin jAin j2 . The resulting input signal power dependence of the NOLM
transmissivity is illustrated in Fig. 3.5b for different values of the coupling ratio K,
and for cL 1. The associated switching power (corresponding to the rst unit
peak of the transmissivity as P is increased from zero) reads as
p
Ps 3:16
cL1 2K
which leads to Ps 16 W and Ps 31 W for K 0:4 and K 0:45,

respectively (see Fig. 3.5b).
As in the nonlinear MZI, signal switching in the NOLM may also be induced via
XPM by injecting in the loop, by means of a directional coupler, a pump pulse at a
different wavelength (see Fig. 3.6a) [3437]. The control pulse propagates in the
CCW direction, thus it only induces a nonlinear phase shift into the corresponding
component of the signal, so that propagation in the NOLM is un-balanced even
when K 0:5. In this case, the NOLM acts as a perfect reflecting mirror for a
signal in the absence of the control pump, whereas in the presence of the pump the
signal may be fully transmitted. For evaluating the nonlinear phase shift induced via
XPM on the signal by a short pump pulse, one should take into account the
temporal walk-off between pump and signal due to ber GVD. The total XPM
phase shift is
ZL
D/T 2c jAp T dxdx; 3:17
0
where T is time in the reference frame that moves with the group velocity at the
signal wavelength Vs , and d Vp1 Vs1 is the group velocity mismatch between

pump and signal. For a Gaussian pump pulse of the form Ap T P exp T 2 =T02
(so that the pump pulse full width at half maximum is Tfwhm 1:66T0 ), the
resulting nonlinear phase shift reads as
p

cLP pT0 T T Tw
D/T erf erf ; 3:18
Tw T0 T0
where erf x is the error function, and the total walk-off is Tw dL. The associated
nonlinear signal transmission is
T 1 4K 1 K cos2 D/=2; 3:19
Let us consider a balanced coupler with K 0:5, a peak XPM-induced phase

shift 2cLP p, and set T0 10 ps. Figure 3.6b shows examples of signal trans-
mission windows for different values of the walk-off time Tw . As it can be seen,
(a) (b) 1
0.8 T =1 ps
w
T =20 ps
w
Transmission
T =50 ps
0.6 w
0.4
0.2
0
20 0 20 40 60
Time (ps)
Fig. 3.6 a Schematic of XPM-controlled switching in a NOLM and b transmission function for
T0 10 ps and different values of the walk-off time Tw
whenever Tw
T0 the transmission window is a replica of the control pulse;
whereas as soon as Tw [ T0 the peak value of the transmission drops from unity
and the window broadens in time.
3.2.5 Nonlinear Passive Loop Resonators
A variant of the nonlinear Sagnac interferometer is provided by the nonlinear ber

ring resonator [3842]. As an example, consider the scheme of Fig. 3.7a: instead of
connecting the two output ports of the coupler, a ber loop now connects the rst
output port (port 3 in Fig. 3.7a) with the second input port (port 2) of the coupler.
The signal E1 at input port 1 enters the coupler and it continuously recirculates in
the loop; the cavity eld can be monitored at the second output port (port 4) of the
coupler. The complex amplitude transmission Y A4 =A1 of the ring resonator is
p p p p
obtained from the relations A3 1 s 1 K A1 i K A2 and A4 1 s
p p
i K A1 1 K A2 , where s is the fractional power loss of the coupler and
A2 A3 expaL ibL iD/NL with D/NL cjA3 j2 L. The result is provided by
the implicit equation
pp
1 s expid aL i 1 s K
Y pp ; 3:20
1 i K 1 s expid aL
where
pp
1s K K 1s
d bL cjA1 j2 L jYj2 i Y Y ; 3:21
1K 1K 1K
Clearly (3.20) is equivalent to two real equations, whose numerical solution yields
the power transmissivity T jA4 j2 =jA1 j2 of the nonlinear resonator (see Fig. 3.7b,
Fig. 3.7 a Schematic passive ber loop resonator and b its transmissivity for different values of
bL; here K 0:95 (Figures adapted from [40])
where the resonant coupling coefcient is M 1 s expaL 0:95). As it

can be seen, by varying the linear phase delay bL right below the resonance
condition bL 3p=2 2pm (where m is an arbitrary integer), one obtains quite
different nonlinear transmission behaviors. For small detunings from resonance
(i.e., bL 1:49p), it is only necessary to add a weak nonlinear phase shift (of the
order of p=100) in order to bring the resonator back into resonance. At higher input
powers, the resonator gets out-of-resonance and transmissivity rapidly increases.
On the other hand, for larger linear cavity detunings (e.g., for bL 1:45p), the
transmissivity becomes multi-valued, and optical multistability results: again, the
ONOFF switching power is relatively low, i.e., it is obtained for nonlinear phase
shifts cjA3 j2 L p=100, which is accessible at CW signal power levels.
3.2.6 Optical Soliton Switching
As we have seen in Sect. 3.2.1, the efciency of all-optical switching in nonlinear

couplers using optical pulses is severely limited unless square pulses (e.g., the non-
return-to-zero (NRZ) data modulation format) are used [15]. In fact, the CW
transmission curve is effectively averaged over the pulse prole, and pulse break-up
occurs at the device output, since different portions of the pulse prole are inde-
pendently switched according to their instantaneous power level (see Fig. 3.3a, b);
the same occurs with nonlinear inteferometers. The pulse break-up effect may be
avoided by operating in the soliton regime, that is whenever the signal pulses
represent optical ber solitons for the bers used in the switching device [25, 32,
33, 4345]. In the short pulse regime, the equations describing propagation in a
nonlinear coupler should be extended to include group velocity dispersion terms,
which leads to the coupled nonlinear Schrdinger equations
@u b @ 2 u
2 2
i jv juj qjvj u0
@n 2 @s2
dv b @ 2 v
2 2
i ju jvj qjuj v0 3:22
dn 2 @s2
where the dimensionless distance is n zjb2 j=ts2 , b2 is the GVD coefcient, b

1 for anomalous or normal dispersion, respectively, s z=ts where ts is a refer-
ence pulse width. Moreover, the dimensionless coupling coefcient j Cts2 =jb2 j
Pc =2Psol , where Pc 2C=c is the NLDC critical power, Psol jb2 j=cts2 is the
soliton power, and q is the XPM coefcient.
Figure 3.8a, b illustrates beam propagation solutions of (3.22) in the anomalous
GVD regime. Here q 0 (as in a linear dual-core ber coupler) and j 1=4. As it
can be seen, input soliton-like pulses of the form u0; s u0 sechs=s0 (here
v0; s 0 and s0 1) are entirely switched as a single entity from the cross to the
(a) (c)
(b)
Fig. 3.8 Evolution of power with distance in the bar and cross state of soliton NLDC at a low and
b high powers; c comparison of power-dependent transmission in the bar state for dispersive pulses in the
anomalous (solid curve) and normal (dashed curve) dispersion regime (Figures adapted from [43, 45])
bar output state as the input soliton peak power p ju0 j2 =Pc is increased from
p 2 to p 3. For p 2, the input pulse periodically couples back and forth
between the two channels, with a relatively small distortion of the pulse prole.
Whereas for p 3 the soliton transfer between the channels is inhibited. Therefore
the optical soliton exhibits a particle-like switching behavior, and the pulse break-
up which is observed in the absence of GVD (or for quasi-CW input signals) may
thus be avoided.
Figure 3.8c compares the fraction of energy transmitted in the bar state as a
function of the input peak power p, when using a NLDC in either the anomalous
(b [ 0) or normal (b\0) GVD regime, respectively. Here we have chosen
p
j p=2, and s0 1= 2p, so that twice the CW critical power is equal to the
fundamental soliton peak power. As it can be seen, with normal GVD the dispersive
pulse broadening combined with SPM and linear coupling nearly inhibits the self-
switching behavior. Conversely, in the anomalous GVD regime, the transmission is
similar to that obtained for CWs or ideal square pulses (besides an increase of the
effective switching power).
3.3 Optical Switching in Integrated Optical Waveguide

Structures
3.3.1 All-Optical Switching in Photonic Crystal Couplers
In recent years Photonic Crystals (PC) have received increasing attention from the
scientic community, especially for their ability to control the propagation of light
[46]. The basic building blocks for all-optical data processing such as waveguides
with sharp bends, high-Q resonant cavities, perfect mirrors and so on could indeed
be integrated on a single PC chip, in order to achieve complex functions with high
performance and small size [47]. PC structures seem to be the ideal choice to get
efcient nonlinear devices for optical switching, because of the strong connement
of the elds that permits to optimize the nonlinear interactions. In particular, the
development of nonlinear PCs exploiting the ultrafast Kerr nonlinearity has become
an important issue. The feasibility of bistable switching devices [48], optical diodes
and nonlinear bends [49], and optical isolators [50] has been reported. Directional
couplers are fundamental components for optical networks, and it has been dem-
onstrated that PC couplers exhibit smaller size and better performance than the
conventional ones [5153]. Here we describe the properties of an all optical switch
based on an ultrashort PC coupler. Switching is performed by exploiting the strong
Kerr nonlinearity of AlGaAs, by controlling the intensity of the input signal. The
reported two-dimensional nite-difference time-domain (2D FDTD, [54, 55])
analysis shows that the resulting ultra-compact device is characterized by a
switching power comparable with the one reported in the literature for centimeter-
long conventional nonlinear directional couplers [56].
The schematic view of the proposed structure is shown in Fig. 3.9. The PC is
formed by a square lattice of AlGaAs rods in air. AlGaAs seems to be a proper
material since it has a large nonlinear refractive index, with minimal linear and
nonlinear absorption in the 1550 nm telecommunications window [57]. The PC
lattice constant is a 400 nm, whereas the radius of the rods for the bulk crystal is
r 130 nm. The resulting structure has a wide bandgap for TM polarization in a
range of wavelengths between 1400 and 1750 nm. Two waveguides are formed
by introducing linear defects reducing the radius of the rods to rdif 70 nm [58].
Fig. 3.9 Schematic view of the PC structure. LB 140 m is the beat length of the coupler. It is
worth to note that the real length is scaled to t in the gure (Adapted from [56])
The reason for this choice is twofold. First, single-mode waveguides are essential to
get a directional coupler, with only one even and one odd supermode. It is well
known that reduced-index waveguides satisfy this requirement, whereas the
increased-index ones tend to be multimode [58]. Second, the need to optimize the
nonlinear interactions suggests to reduce only partially the radius of the defect rods,
in order to maximize the semiconductor ll factor. We have calculated that in the
previously described guiding structure over 50 % of the modal eld energy is
conned into the nonlinear dielectric defects [59]. The correct design of the coupler
section is the key issue for our problem. We have found that the critical parameter
to achieve switching with reasonable input intensity of the light is the distance
between the two waveguides. Therefore fully vectorial eigenmodes of Maxwells
equations for a set of couplers were computed using a freely available mode solver
[59], and 2D FDTD simulations [54, 55] were analyzed to study the device
behavior when the waveguides separation is varied.
Moreover, a very simple and powerful coupled-mode theory was developed to
model nonlinear propagation in PC couplers. As expected, there is a tradeoff
between length of the structure and switching intensity. If the waveguides are close
to each other the linear coupling is strong, thus it is very difcult to decouple them
exploiting the effects of the ultrafast Kerr nonlinearity. If the waveguides are far
away, the beat length of the coupler is very large, and the resulting device is not
ultra-compact. We have then chosen a coupler composed of two waveguides sep-
arated by ve lattice constants, as shown in Fig. 3.9. In Fig. 3.10 we report the
projected band structure of the coupler, evaluated through the mode solver [59]. It is
straightforward to calculate the beat length LB 2p=keven kodd of the device,
which is about 140 m at the wavelength k 1560 nm; keven and kodd are the
effective wavenumbers of the even and the odd supermode. Light can be coupled
from a dielectric slab waveguide (width W = 3 m) to the coupler through a tapered
input section [60], as shown in a schematic way in Fig. 3.9. More complex and
Fig. 3.10 Projected band structure of the coupler, with the normalized dispersion relations for the
odd (solid line) and the even (dash-dotted line) supermode. Note that in our case we are working
around a=k 0:256 (Adapted from [56])
efcient PC tapers have been proposed (e.g. in [61]), nevertheless a systematic

analysis of the input section of the PC chip in order to optimize the coupling
efciency is well beyond the aim of this section. The switch layout is completed
with the double sharp bend that decouples the two waveguides just in proximity of
the half beat length of the coupler (see Fig. 3.9). The nal structure, composed of
the tapered input section, the coupler and the double sharp bend is about 75 m
long.
In order to show the basic operation principle of the proposed structure we
exploit a coupled-mode theory [6264], developed to study coupled defects in
nonlinear PCs. In case of weak interactions between similar single-mode defects,
the modal eld of each defect can be considered unperturbed, so that only the eld
amplitudes vary in time. The evolution of the state of each PC waveguide, con-
sidered as a straight chain of resonators, is governed by a set of differential
equations
dan
i Cn an1 an1 cn jan j2 an 0 3:23
dt
where an is the eld amplitude in the n-th defect, Cn is the nearest-neighbor linear
coupling coefcient and cn is the self-phase modulation strength [6264]. This
theory was proposed for the analysis of propagation in coupled-resonator optical
waveguides (CROW). In our case, the individual defect rods composing the linear
defect are strongly coupled, therefore the accuracy of the model could be ques-
tionable. Nevertheless, we show that this rough and simple theory can help to
understand the behavior of the nonlinear PC coupler. We can introduce the coupling
between the two waveguides by just dening two sets of equations,
dan
i Cn an1 an1 Dn bn cn jan j2 an 0 3:24
dt
dbn
i Cn bn1 bn1 Dn an cn jbn j2 bn 0 3:25
dt
where an and bn are the eld amplitudes in the n-th defect of the rst and the second
waveguide (called WG1 and WG2 respectively), and Dn is the linear coupling
coefcient between the n-th defect in the rst and in the second waveguide. We
have solved numerically the system composed of (3.24) and (3.25) in the linear
regime (cn 0). The coupling coefcient Cn was xed to 2 s1 for every n, whereas
Dn 0:035 s1 for n\170 and Dn 0 elsewhere, in order to simulate the intro-
duction of the double sharp bend. Figure 3.11 shows that the coupled-defect theory
is able to reproduce the behavior of the PC coupler. We report the calculated values
of an and bn for a CW excitation when a steady-state is reached. The input eld is
injected into WG1 , and the eld flows toward WG2 because of the linear coupling
between the two waveguides. The sharp bend is in proximity of the half beat length
of the coupler, thus at the output all the optical energy is in WG2 . In Fig. 3.12 we
Fig. 3.11 Normalized

intensity of the eld in WG1
(at the top) and WG2 (on the
bottom) in linear regime
(cn 0) calculated through
the coupled defects model.
The sharp bend is placed near
the defect n 170 (see the
dash-dotted line) (Adapted
from [56])
report the numerical solution of the system composed of (3.24) and (3.25) in the
nonlinear regime (cn jain j2 0:75 s1 ). It is straightforward to note the effect of the
nonlinearity: now at the output all the optical energy is in WG2 , which shows that it
is possible to switch the output channel by varying the input eld intensity.
Now we show a rigorous 2D FDTD analysis of the feasibility of all-optical
switching in the previously described device. The problem of the implementation of
the structure on a real PC slab is a big issue that would require huge 3D FDTD
simulations, in order to estimate scattering losses in the third dimension. We
highlight that the 2D numerical modeling of the real 3D device is commonly used
in the literature for the study of phenomena due to Kerr effect in PC (see [4850]),
since this permits to focus the attention mainly on the nonlinear interactions. The
nonlinear PC coupler is simulated by injecting into the taper section a CW Gaussian
eld that approximates the fundamental mode of the dielectric slab waveguide. It is
worth to note that in this way we take into account the effects due to the coupling
efciency into the PC chip. By varying the maximum value of the Gaussian, and
then the related maximum intensity of the input light, we may characterize the
behavior of the power-controlled switch.
In Figs. 3.13, 3.14 and 3.15 we show the intensity of the electromagnetic eld in
the device for three different intensities of the input signal. Figure 3.13 shows the
simulation result for a maximum input intensity IIN max 1 GW/cm2. It is possible to
see that all the optical energy flows through the output port 2, thus the coupler is in
cross state. This behavior is in good agreement with the previously described design
procedure, in fact the double sharp bend decouples the two waveguides just near the
half beat length. In Fig. 3.14 we have increased the maximum input intensity to
IIN max 3:8 GW/cm2. Here it is clear the effect of the nonlinear phase shift: the
input intensity is equally divided between the two output waveguides, and the
structure behaves as a 50 % power splitter. Figure 3.15 shows the behavior of the
device increasing IIN max to 5 GW/cm2. In this case all the optical energy flows
Fig. 3.12 Normalized intensity of the eld in WG1 (at the top) and WG2 (on the bottom) in the
nonlinear regime (cn jain j2 0:75 s1 ) calculated through the coupled defects model. The sharp
bend is placed near the defect n 170 (see the dash-dotted line) (Adapted from [56])
Fig. 3.13 Intensity of the eld in the PC coupler with maximum input intensity 1 GW/cm2: the
coupler is in cross state, the power ratio is about 20 dB (Adapted from [56])
Fig. 3.14 Intensity of the eld in the PC coupler with maximum input intensity 3:8 GW/cm2: the
coupler behaves as a 50 % power splitter, the power ratio is about 0 dB (Adapted from [56])
through the output port 1, thus the coupler is in a bar state induced by the non-
linearity. It is worth pointing out that the result of the FDTD simulation shown in
Fig. 3.15 is in agreement with the solution of the coupled defects model (see
Fig. 3.12). The eld injected in WG1 initially couples to WG2 as in an asymmetric
coupler, but at the double sharp bend position all the optical energy is in WG1 , as
desired. Figure 3.16 summarizes the operation principle: a high-intensity signal
propagates along the input waveguide and flows through the output port 1, whereas
a low-intensity signal is switched toward output port 2.
In Fig. 3.17 we show the power ratio in decibel, i.e. the ratio between the optical
energy in the output ports 2 and 1 of the structure, versus the maximum value of the
input intensity at k 1555 nm, k 1560 nm and k 1565 nm, respectively. The
simulations demonstrate that the device can be considered an optically-controlled
Fig. 3.15 Intensity of the eld in the PC coupler with maximum input intensity 5 GW/cm2: the
coupler is in bar state, the power ratio is about 20 dB (Adapted from [56])
Fig. 3.16 The power-controlled switching function: a low-intensity signal is switched toward
output port 2 (linear coupler), whereas a high-intensity signal propagates along the input
waveguide because of the nonlinear phase shift (Adapted from [56])
Fig. 3.17 Power ratio versus

maximum input intensity of
the light at k 1560 nm
(solid line), k 1555 nm
(dash-dotted line) and k
1565 nm (dashed line)
(Adapted from [56])
switch, with a power ratio larger than 20 dB in a wide range of wavelengths. In

presence of pulsed excitations the performance of the power-controlled switch
decreases with respect to the CW case. In particular, as has already been described
in Sect. 3.2, with nonsquare input pulses the power ratio is reduced and output pulse
break-up is observed [65, 66]. Nevertheless these phenomena are strongly depen-
dent on the shape and the duration of the pulses, and they do not affect the validity
of the proof-of-principle nonlinear PC coupler. It is interesting to compare the
device behavior with what has already been reported in the literature for NLDCs
made by AlGaAs semiconductor waveguides, and operating in the third-telecom-
munications window. In this second case, both simulations and experiments show
switching powers from 50 to 90 W, for a few centimeters long devices [65, 66].
Assuming that the eld at the input of the PC chip is a Gaussian beam, with the
same spot-size w0 1 m we used in the 2D FDTD simulations, we can integrate
the input intensity nding an approximated power for the total switching of about
70 W, which is comparable with the previously given values. Nevertheless, it is
fundamental to note that the length of the nonlinear PC coupler is less than 80 m,
which is signicantly shorter than a standard waveguide NLDC. The one-pulse
scheme analyzed so far could be extended to a pump signal conguration, with a
strong pump acting through cross-phase modulation (XPM) on a weak signal at
different wavelength.

Surface Plasmon Polaritons
We discuss in this section the tuning of the coupling of surface plasmon polaritons
between two graphene layers with nanometer spacing. We demonstrate that, by
slightly changing the electrical doping and then shifting the chemical potential, a
graphene coupler can switch from the bar to the cross state. As a consequence, the
coupling coefcient in such structures can be easily controlled in an ultrafast
fashion either by means of an applied electrical signal [67] or by changing the
intensity of the signal at the device input. These ndings open the way to fully
exploit the huge nonlinearity of graphene for all optical signal processing: from one
side giving more degrees of freedom to already proposed devices [6872], from the
other side paving the way to new devices.
Graphene can sustain surface plasmon polariton (SPP) having unique properties
as compared to what we are used to with noble metals. In fact a single layer of
graphene can support either TE or TM polarized plasmons without suffering from
huge loss [7375]; moreover, as far as the TM polarization is concerned, the
extremely high connement factor is particularly favourable to explore the huge
v3 nonlinearity of graphene [68, 76, 77]. Experimental endeavors have demon-
strated the evidence of graphene plasmons by measuring the plasmon resonance of
graphene nanoribbon arrays [78], and by acquiring their near eld images [79, 80].
The coupling of SPP between separated graphene layers has been recently analyzed
in [81]; however the very interesting properties arising from the easily tunable
optical properties of graphene have not been exploited yet in this framework. Here
in particular we show that by slightly changing the chemical potential, a graphene
coupler can switch from the bar to the cross state.
In Fig. 3.18 we report the basic geometry that we are going to consider in this
section; two graphene layers are embedded in a dielectric structure: region 1 (of
width 2s) is the dielectric in between the two graphene layers. At the graphene
boundary we set the following conditions on the tangential components of the
electromagnetic eld:
Fig. 3.18 Schematic of the

graphene directional coupler:
the separation between the
layers is equal to 2s
(Adapted from [67])

E2;3 E1 ^x 0 3:26

H2;3 H1 ^x ixe0 erS12;3 Ejj x s
where Ejj is the electric eld tangent to the graphene layer and erS12 (erS13 ) is the
relative surface permittivity of the graphene layer between regions 1 and 2 (3). As
far as the electromagnetic constants of graphene are concerned, we write the linear
contribution to the relative complex permittivity as [82, 83]:
1 1
erS rR;I rR;R
erC 1 i erC;R ierC;I 3:27
dg dg xe0 dg xe0
1
where dg is the graphene thickness and the surface complex conductivity rR
1 1
rR;R irR;I (in Siemens) is obtained from theoretical models now well established
and experimentally validated [84, 85], which give the following dependence of the
real and imaginary parts of the conductivity on frequency (x), temperature (T) and
chemical potential ():

1 r0 hx 2l hx 2l
rR;R x tanh tanh
2 4kB T 4kB T

1 r0 4 2l3 hx 2l
rR;I x l 2 log 3:28
p hx 9t hx 2l
where t 2:7eV is the hopping parameter, h and kB are the reduced Plancks and
Boltzmanns constants, respectively, and r0 e2 =4h 6:0853 105 S, with e
the electron charge.
Note also that this model can be easily extended into the nonlinear regime by
1
adding a nonlinear correction to the surface conductivity as: rR rR
3
rR jEj2 Ey;z [74]. Moreover, thanks to the extremely small thickness of the graphene
layer, nonlinearity can be analyzed by introducing a parameter embedded into the
coefcients describing the continuity of the tangential components of the electro-
magnetic eld [68, 77].
To describe SPP propagation along z, we rst note that, at rst order, the y
dependence of the electromagnetic eld can be neglected; we then look for guided
modes with harmonic temporal dependence expixt and spatial variation
E1;2;3 x; z, H1;2;3 x; z expibz C1;2;3 x with C21;2;3 b2 er1;2;3 k02 .
Obviously the complex wavenumber b, through its real and imaginary parts,
describes the evolution of both the phase and the amplitude of the guided modes.
We can apply the above modeling to derive the dispersion relation of both TE and
TM modes. In the following we describe in details the TM polarization. We rst
consider a very general situation where the two graphene layers can be biased in a
different way to give rise to an asymmetric coupler. After straightforward algebra
we nd that coupled SPP in the system are determined by setting to zero the
determinant of the following matrix:
2 3
e C1 s eC1 s eC2 s 0
6 eC1 s eC 1 s 0 e C3 s 7
M6
4 ixe 1 C1 s ixe1 C1 s
C1 e g12 eC2 s
7
5 3:29
C1 e 0
1 C1 s ixe1 C1 s C3 s
ixe
C1 e C1 e 0 g13 e
where g12 and g13 take into account the contribution of the two graphene layers
in the continuity conditions:
ixe2 ixe3
g12 ixe0 erS;12 ; g13 ixe0 erS;13
C2 C3
where erS;12 and erS;13 refer to the relative dielectric constant of the two graphene
layers, which in general may have different values due to different carriers con-
centrations. This asymmetric coupler offers a wide variety of possible settings
which certainly deserve to be investigated both in the linear and in the nonlinear
regime. Here we describe a prototype example illustrating the possibilities offered
by the tunability of graphene parameters in this framework; we thus focus our
attention on a very particular situation corresponding to a linear and symmetric case
(e2 e3 and erS;12 erS;13 ); moreover we use T = 300 K and k 10 m. For the
sake of simplicity, we also set er1 er2 er3 2:25. In this regime the graphene
directional coupler has two different eigenstates: the even (odd) supermode corre-
sponding to the out of phase (in phase) hybridization of the SPP guided by the
single graphene layers. Note also that the even mode here has always the highest
value of the propagation constant.
Fig. 3.19 Effective index neff <eb=k0 of even and odd supermodes of the coupled graphene
layers as a function of the separation among the layers. Continuous (dashed) lines refer to a
chemical potential of l1 0:1 eV (l2 0:15 eV) (Adapted from [67])
In Fig. 3.19 we report the solution of the dispersion relation as a function of s for
two different situations: continuous lines here refer to the even and odd supermodes
corresponding to a chemical potential l1 0:1 eV in (3.28), while the dashed lines
refer to a choice of the chemical potential l2 0:15 eV in (3.28). It is straight-
forward to note that, for large enough s, the two supermodes of the coupler tend to
degeneracy, and their propagation constants approximate the propagation constant
of the SPP of a single graphene layer. The main message we can read from
Fig. 3.19 is that a very small change in the chemical potential can induce a very big
change in the behaviour of the coupler. In the following, we focus our attention to
the case with s = 5 nm. For this value of the separation between the layers, we
computed the beat length of the directional coupler as a function of the chemical
potential: the corresponding results are reported in Fig. 3.20. We can clearly see
there that a very small change of the chemical potential can be used to induce huge
changes of the beat length of the coupler. The two particular points (open square
and open circle) enlightened in Fig. 3.20 are the initial conditions in Fig. 3.21,
where we describe the propagation of the electromagnetic signal in the graphene
coupler. In both panels in Fig. 3.21 total propagation length is set to L 90 nm. On
the left panel in Fig. 3.21 the input condition corresponds to the square in Fig. 3.20
and the coupler is in the cross state; on the right panel the input condition corre-
sponds to the circle in Fig. 3.20 and the coupler is in the bar state.

Optical Waveguides
In this subsection we demonstrate that, thanks to the ultrafast tunability of losses

which are introduced by graphene layers deposited onto the structures, a careful
design of silicon on insulator ridge waveguides can be used to explore the so-called
Fig. 3.20 Beat length versus chemical potential for a graphene plasmon coupler. Here 2s = 10 nm
(Adapted from [67])
Fig. 3.21 Field evolution in a graphene plasmon directional coupler: left (right) refers to a
chemical potential of l1 0:0908 eV (l2 0:1367 eV). Here 2s = 10 nm (Adapted from [67])
passive parity-time (PT) symmetry breaking in directional couplers. A quantum

system characterized by a Hamiltonian H is PT-symmetric if H commutes with the
operator PT, where P is the parity operator and T is the temporal operator [84]. At
an exceptional point, two or more eigenvalues are degenerate. We prove that the
exceptional point of the coupler can be probed by varying the applied voltage,
which may lead to very compact photonic structures for the control of coupling
among waveguides, and for tailoring discrete diffraction in arrays [86].
In particular, in this section we present numerical results which demonstrate the
huge potential of graphene as a means to control coupling between optical dielectric
waveguides. The possibility of tuning losses in each waveguide by acting on a thin
loss element permits to break the symmetry of the coupled waveguides without
introducing a strong perturbation in each single waveguide. Remarkably, we
demonstrate that tunable losses induced by graphene and a careful design of ridge
waveguides allows to probe passive PT-symmetry breaking in directional couplers
[8789]. Moreover, the exceptional point of the coupler can be dynamically
controlled by varying the applied voltage. We will thus explore these properties to
mould energy exchange between waveguides and to nely tune discrete diffraction
in waveguide arrays. These results, together with the strong saturable absorption of
graphene [76, 90], suggest also the possible use of this conguration in nonlinear
devices with a strong pump beam used as a probe for all optical switching of weak
input signals from the bar to cross state.
In order to prove our statement we rst need to consider the behavior of silicon
waveguides on a silica substrate in a wavelength range between 1350 and 1600 nm.
The structure has been inspired by the modulator proposed in [91]. In particular, a
layer of silicon with thickness equal to 50 nm is deposited onto the substrate. The
400 nm wide ridge waveguide is composed of lower and higher layers made of
silicon (both with thickness 200 nm) which sandwich a central region including
three alternating layers of alumina (thickness 7 nm) and two absorption layers
composed of three graphene monolayers with thickness 0:34 nm. Graphene can be
electrically controlled in order to tune doping (and then conductivity), as suggested
in [9198]. The dielectric constants of silicon, silica and alumina were taken equal
to 12:1, 2:1 and 3. Figure 3.22a displays a schematic view of the structure.
The behavior of graphene in the optical regime has been numerically modeled by
following the approach suggested in [82, 83], as already described in the previous
subsection. Indeed, we assigned to each graphene monolayer with thickness D a
1 1
volume conductivity equal to rg;v rR =D, where rR is the conductivity of the
2D sheet (see (3.27) and (3.28)). It was demonstrated that, as a rst approximation,
few-layer graphene is characterized by the same band structure (and then by the
same excellent electronic properties) of the monolayer. Moreover, if N is small
enough the conductivity of N-layer graphene (N 3 in our design) can be eval-
uated as N times conductivity of the single layer [93].
We then performed a modal analysis of the waveguide in Fig. 3.22a by resorting
to nite-element simulations. We focused the attention on the TE-like mode which
is depicted in the inset of Fig. 3.22b, since the electric eld is tangential with
Fig. 3.22 Left Schematic view of the waveguide structure, with a detail of the central region with
graphene layers. Right Losses of the single waveguide (in dB/m) when graphene is in OFF state
(null voltage), and x-component of the electric eld of the TE-like mode (inset) (Adapted from [86])
respect to graphene layers. In the same gure we report the absorption of the TE-
like mode when the graphene layers are in OFF state (null voltage). Of course,
when graphene is in the ON state (a control voltage is applied), losses are close to
zero. First, it is important to note that the real part of the effective index of the mode
(not shown here) is barely affected by state of the graphene layers. Then, we
emphasize that when graphene is in OFF state losses are quite large (0.48 dB/m,
which corresponds to 1100 cm1 ) and almost constant over the entire bandwidth:
indeed, a 6 dB modulation contrast between ON and OFF states can be achieved
with a 12.5 m long waveguide. In the next paragraphs we will study the properties
of coupled waveguides wherein the described structure is the basic building block.
Full-wave simulations of photonic devices including graphene layers are char-
acterized by huge computational burden [99]. This effect is obviously emphasized
when structures composed of multiple waveguides must be analyzed. Therefore
conventional coupled-mode theory (CMT) [100, 101] has been reformulated to
study the present structures. Using full wave simulations, we rst numerically
proved that neither the prole of TE-like modes, nor the propagation constant b of
each isolated waveguide are affected by the status of graphene layers. Whereas
switching between ON and OFF states has the effect of turning off and on losses in
the single waveguide, which are modeled by the attenuation constant a. Notice that
modal evolution reads as expibz expaz.
Under these conditions, it is possible to verify that the system of governing
equations for A1;2 , which are the modal eld amplitudes in the rst and second
waveguide of a directional coupler composed of two identical graphene-based
waveguides, can be approximated as

d A1 b ia1 C A1
i ; 3:30
dz A2 C b ia2 A2
where a1;2 can be tuned between zero (ON state) and amax (OFF state) by con-
trolling the voltage applied to the graphene layers, and C is a complex coupling
coefcient [88].
When a1 0 and a2 a, the eigenvalues of the matrix in (3.29) read as
q
k1;2 b ia=2 jCj2 a=22 . Therefore the so-called exceptional point
(EP) for the onset of PT-symmetry breaking is located at the critical loss value
ac 2C. Indeed, when a\ac the two supermodes have different propagation
constants and the same attenuation constant a=2. Beyond the critical loss the su-
permodes coalesce: they are characterized by the same propagation constant b and
by different loss coefcients. In particular, one supermode experiences increasing
losses with increasing a, whereas the other supermode is characterized by the
opposite trend [87, 88]. Whenever a is much larger than C, one supermode is
characterized by losses which are close to the losses of a single waveguide, whereas
the other supermode is nearly loss-free.
At this point we may examine the behavior of the directional coupler which is
obtained by placing two identical graphene-based waveguides close to each other,
Fig. 3.23 Schematic view of the 300-nm-gap coupler (left), with the electric eld of the low-
(center) and high-loss mode (right) at 1530 nm. Graphene layers are in ONOFF states (Adapted
from [86])
separated by a 300 nm gap (see Fig. 3.23 on the left). We performed a modal
analysis at 1530 nm focusing the attention on the supermodes which originate from
the interaction between TE-like modes of the single waveguides, and we varied the
state of the graphene layers. Numerical results in Fig. 3.23 illustrate in a qualitative
way how the behavior of the structure can be controlled by exploiting the properties
of graphene. In particular, when graphene layers in both waveguides are in the same
state, the symmetry of the structure is preserved, and the two modes (not shown
here) are even and odd. Viceversa, Fig. 3.23 shows that when symmetry is broken
by switching to the ONOFF state, the effect of losses on the modal properties is
huge, and a trend toward decoupling between the waveguides is apparent.
The dispersive properties of the two modes have been characterized through
full-wave and CMT simulations, and these quantitative results conrm the intuitive
analysis we have reported above. Indeed, in Fig. 3.24a losses of the two super-
modes are depicted when the graphene layers are in the ONOFF state. In this case
symmetry is broken: as a consequence, one mode is characterized by absorption
which is close to zero, whereas the other mode experiences large losses, which are
very close to those of a single lossy waveguide. It is worth noting that this effect
tends to blur with increasing wavelength due to the dependence of coupling
coefcient on frequency (C gets larger with increasing wavelength). A thorough
treatment on phenomena arising from the wavelength dependence of the PT-sym-
metry condition is reported in [89]. The noticeable agreement in Fig. 3.24a between
simulations performed by using a full-wave mode solver and the results evaluated
by using CMT (in the latter case the imaginary part of k1;2 is reported) allows to
conrm the accuracy of CMT.
These phenomena stem from the breaking of passive PT-symmetry in complex
potentials. Indeed, in Fig. 3.24b we plot the attenuation constants of the two su-
permodes, evaluated by using CMT, as a function of the attenuation constant of the
single waveguide a. Data are normalized with respect to twice the coupling coef-
cient, so that we have the exceptional point when the abscissa is equal to 1. The
vertical dotted line indicates amax , i.e. the value of a when our structure is in the
OFF state: it is straightforward to see that one may operate beyond the exceptional
point, in agreement with the results of Fig. 3.24a. It is worth to emphasize that
graphene-based waveguides exhibit superior properties with respect to waveguides
wherein losses are introduced by depositing metal layers [88]. Losses induced by
sandwiching graphene layers inside silicon waveguides can be orders of magnitude
larger (thousands of cm1 with respect to tens of cm1 ), therefore it is possible to
Fig. 3.24 Left Losses of low- (red line) and high-loss mode (blue line) from mode solver (solid
line) and CMT (dashed-dotted line). Right Normalized attenuation constant of low- (red line with
squares) and high-loss mode (blue line with circles) versus normalized attenuation constant of the
single waveguide at k 1530 nm (Adapted from [86]). The vertical thin line indicates a amax
probe the exceptional point even in structures characterized by strong coupling.

Last, but not least, it is important to note that graphene is electrically tunable,
therefore losses in each single waveguide can be varied between zero (ON state)
and a maximum value amax , which is only determined by geometry (OFF state).
We envisage that switching of the state of one waveguide can be exploited to
nely tune coupling between waveguides. In order to verify the effectiveness of this
approach we applied CMT to our reference structure at the wavelength of 1530 nm,
and we show the results in Fig. 3.25. When the coupler is in the ONON state
losses are zero, and the predicted beat length LB p=beven bodd is around
80 m. Viceversa, when graphene layers are ON and OFF in the input and output
channels, the two waveguides tend to decouple and the eld intensity in the rst
waveguide is larger than in the second one. It is possible to justify this behavior by
recalling that when we inject light into the waveguide in ON state the low-loss
supermode is mainly excited.
These results have been validated by comparison with simulations of the 80 m
long coupler performed by using the commercial software CST Microwave Studio,
which allows to solve Maxwells equations in the time domain through the nite-
integration technique. Indeed, the ratio between output and injected power evalu-
ated by using CMT is 3 and 12 dB if the coupler is in the ONOFF state and we
consider as output port waveguides 1 and 2. CST simulations exhibit a good
agreement, in fact the corresponding calculated values are about 5 and 13 dB,
respectively.
The unique properties we have described in the previous paragraph open the way
to novel possibilities for controlling discrete diffraction in waveguide arrays
[100105]. Let us take for example an array composed of eleven identical wave-
guides, with the same geometrical and optical parameters that we have used
throughout the chapter, and the same spacing (300 nm) that we considered for the
coupler. The structure was simulated by using CMT in order to reduce the com-
putational burden. The input excitation covers only the central waveguide, and the
Fig. 3.25 Field amplitude in (left) rst and (right) second waveguide of the coupler (Adapted
from [86]). Graphene layers are in ONON (black line), and ONOFF (red line) states
Fig. 3.26 Discrete diffraction along the array. Left All the graphene layers are in ON state. Right
Only graphene layers inside the central waveguide are in ON state (Adapted from [86])
propagation length was taken equal to the beat length of the coupler (80 m).
Moreover, we assumed that the state of the graphene layers in each waveguide can
be controlled independently from each other.
In Fig. 3.26a we show the eld inside the structure when all the graphene layers
are in the ON state: the typical pattern of discrete diffraction is clearly visible, with
two pronounced outermost wings [100, 101]. In Fig. 3.26b all the waveguides
except for the central one are switched to the OFF state, and two phenomena can be
clearly noticed. First, beam broadening is reduced with respect to the previous case,
so that most of the optical energy remains concentrated into the central waveguide.
Second, losses are smaller with respect to the case of a single lossy waveguide.
3.4 Conclusions
Optical switching will be a key enabling functionality in future transparent

all-optical networks. In this chapter we have provided an overview of several
guided wave optical switching devices, where the inputoutput path of optical
signals is controlled by either optical or electrical signals, thus avoiding the need for
OEO conversion for optical signal processing. We have rst presented the basic
principles of ber optics switching devices, whose long interaction lengths permit to
signicantly reduce the operating power requirements. Next we have discussed
nonlinear couplers based on integrated waveguides with strong eld connement,
hence reduced device dimensions, thanks to photonic crystal structures or surface
wave interactions in graphene layers.
Acknowledgments This work was funded by Fondazione Cariplo (grants no. 2011-0395 and no.
2013-0736), the Italian Ministry of University and Research (grant no. 2012BFNWZ2), and the
US Army (grants no. W911NF-12-1-0590 and no. W911NF-13-1-0466).
References
1. A. Yariv, Coupled-mode theory for guided-wave optics. IEEE J. Quantum Electron. 9,

919933 (1973)
2. S.M. Jensen, The nonlinear coherent coupler. IEEE J. Quantum Electron. 18, 15801583
(1982)
3. A.A. Maier, Optical transistors and bistable devices utilizing nonlinear transmission of light
in systems with undirectional coupled waves. Sov. J. Quantum Electron. 12, 14901494
(1982)
4. G.I. Stegeman, E.M. Wright, All-optical waveguide switching. Opt. Quant. Electron. 22,
95122 (1990)
5. H.G. Winful, Self-induced polarization changes in birefringent optical bers. Appl. Phys.
Lett. 47, 213215 (1985)
6. B. Daino, G. Gregori, S. Wabnitz, New all-optical devices based on third-order nonlinearity
of birefringent bers. Opt. Lett. 11, 4244 (1986)
7. S. Trillo, S. Wabnitz, Nonlinear dynamics and instabilities of coupled waves and solitons in
optical bers, in Anisotropic and Nonlinear Optical Waveguides, ed. by C.S. Someda, G.I.
Stegeman (Elsevier, Amsterdam, 1992), pp. 185236
8. B. Daino, G. Gregori, S. Wabnitz, Stability analysis of nonlinear coherent coupling. J. Appl.
Phys. 58, 45124514 (1985)
9. H.G. Winful, Polarization instabilities in birefringent nonlinear media: application to ber-
optics devices. Opt. Lett. 11, 3335 (1986)
10. S. Wabnitz, E.M. Wright, C.T. Seaton, G.I. Stegeman, Instabilities and all-optical phase-
controlled switching in a nonlinear directional coherent coupler. Appl. Phys. Lett. 49, 838840
(1986)
11. D.D. Gusovskii, E.M. Dianov, A.A. Maier, V.B. Neustruev, E.I. Shklovskii, I.A.
Shcherbakov, Nonlinear light transfer in tunnel-coupled optical waveguides. Sov.
12. D.D. Gusovskii, E.M. Dianov, A.A. Maier, V.B. Neustruev, V.V. Osiko, A.M. Prokhorov,
KYu. Sitarskii, I.A. Shcherbakov, Experimental observation of the self-switching of radiation
in tunnel-coupled optical waveguides. Sov. J. Quantum Electron. 17, 724727 (1987)
13. S.R. Friberg, Y. Silberberg, M.K. Oliver, M.J. Andrejco, M.A. Sai, P.W. Smith, Ultrafast
all-optical switching in a dual-core ber nonlinear coupler. Appl. Phys. Lett. 51, 11351137
(1987)
14. A.M. Weiner, Y. Silberberg, S.R. Friberg, B.G. Sfez, P.W. Smith, Femtosecond all-optical
switching in a dual-core ber nonlinear coupler. Opt. Lett. 13, 904906 (1988)
15. A.M. Weiner, Y. Silberberg, H. Fouckhardt, D.E. Leaird, M.A. Sai, M.J. Andrejco, P.W.
Smith, Use of femtosecond square pulses to avoid pulse breakup in all-optical switching.
IEEE J. Quantum Electron. 25, 26482655 (1989)
16. S. Trillo, S. Wabnitz, R.H. Stolen, G. Assanto, C.T. Seaton, G.I. Stegeman, Experimental
observation of polarization instability in a birefringent optical ber. Appl. Phys. Lett. 49,
12241226 (1986)
17. S. Trillo, S. Wabnitz, W.C. Banyai, N. Finlayson, C.T. Seaton, G.I. Stegeman, R.H. Stolen,
Picosecond nonlinear polarization switching with a ber lter. Appl. Phys. Lett. 53, 837839
(1988)
18. S. Trillo, S. Wabnitz, W.C. Banyai, N. Finlayson, C.T. Seaton, G.I. Stegeman, R.H. Stolen,
Observation of ultrafast nonlinear polarization switching induced by polarization instability
in a birefringent ber rocking lter. IEEE J. Quantum Electron. 25, 104112 (1989)
19. P. Ferro, M. Haelterman, S. Trillo, S. Wabnitz, B. Daino, Polarization switching in spun
birefringent ber. Appl. Phys. Lett. 59, 20822084 (1991)
20. P. Ferro, M. Haelterman, S. Trillo, S. Wabnitz, B. Daino, All-optical polarization switch with
a long low-birefringence ber. Electron. Lett. 27, 14071408 (1991)
21. P. Ferro, S. Trillo, S. Wabnitz, All-optical polarization differential amplication with a
birefringent ber. Electron. Lett. 30, 16161617 (1994)
22. P. Ferro, S. Trillo, S. Wabnitz, Phase control of a nonlinear coherent coupler: the
multibeatlength twisted birefringent ber. Appl. Phys. Lett. 64, 28722874 (1994)
23. P. Ferro, S. Trillo, S. Wabnitz, Demonstration of nonlinear nonreciprocity and logic
operations with a twisted birefringent optical ber. Opt. Lett. 19, 263265 (1994)
24. N.J. Doran, D. Wood, Soliton processing element for all-optical switching and logic. J. Opt.
Soc. Am. B 4, 18431846 (1987)
25. N. Imoto, S. Watkins, Y. Sasaki, A nonlinear optical-ber interferometer for nondemolitional
measurement of photon number. Opt. Commun. 61, 159163 (1987)
26. B.K. Nayar, N. Finlayson, N.J. Doran, S.T. Davey, W.L. Williams, J.W. Arkwright, All-
optical switching in a 200-m twin-core ber nonlinear Mach-Zehnder interferometer. Opt.
Lett. 16, 408410 (1991)
27. M. Asobe, Effects of group-velocity dispersion in all-optical switching devices using highly
nonlinear optical waveguides. J. Opt. Soc. Am. B 12, 12871299 (1995)
28. N.J. Doran, D.S. Forrester, B.K. Nayar, Experimental investigation of all-optical switching in
a bre loop mirror device. Electron. Lett. 25, 267268 (1989)
29. K. Otsuka, Nonlinear antiresonant ring interferometer. Opt. Lett. 8, 471473 (1983)
30. D.B. Mortimore, Fiber loop reflectors. J. Lightwave Technol. 6, 12171224 (1988)
31. N.J. Doran, D. Wood, Nonlinear optical loop mirror. Opt. Lett. 13, 5658 (1988)
32. K.J. Blow, N.J. Doran, B.K. Nayar, Experimental demonstration of optical soliton switching
in an all-ber nonlinear Sagnac interferometer. Opt. Lett. 14, 754756 (1989)
33. M.N. Islam, E.R. Sunderman, R.H. Stolen, W. Pleibel, J.R. Simpson, Soliton switching in a
ber nonlinear loop mirror. Opt. Lett. 14, 811813 (1989)
34. M.C. Farries, D.N. Payne, Optical ber switch employing a Sagnac interferometer. Appl.
Phys. Lett. 55, 2526 (1989)
35. K.J. Blow, N.J. Doran, B.K. Nayar, B.P. Nelson, Pulse shaping, compression, and pedestal
suppression employing a nonlinear-optical loop mirror. Opt. Lett. 15, 248250 (1990)
36. M. Jinno, T. Matsumoto, Ultrafast, low power, and highly stable all-optical switching in
an all polarization maintaining ber Sagnac interferometer. IEEE Photon. Technol. Lett. 2,
349351 (1990)
37. M. Jinno, T. Matsumoto, Ultrafast all-optical logic operations in a nonlinear Sagnac
interferometer with two control beams Opt. Lett. 16, 220222 (1991)
38. L.F. Stokes, M. Chodorow, H.J. Shaw, All-single-mode ber resonator. Opt. Lett. 7,
288290 (1982)
39. H. Nakatsuka, S. Asaka, H. Itoh, K. Ikeda, M. Matsuoka, Observation of bifurcation to chaos
in an all-optical bistable system. Phys. Rev. Lett. 50, 109112 (1983)
40. B. Crosignani, B. Daino, P. Di Porto, S. Wabnitz, Optical multistability in a ber-optic

passive-loop resonator. Opt. Commun. 59, 309312 (1986)
41. M. Haelterman, S. Trillo, S. Wabnitz, Dissipative modulation instability in a nonlinear
dispersive ring cavity. Opt. Commun. 91, 401407 (1992)
42. S. Coen, M. Haelterman, P. Emplit, L. Delage, L.M. Simohamed, F. Reynaud, Experimental
investigation of the dynamics of a stabilized nonlinear ber ring resonator. J. Opt. Soc. Am.
B 15, 22832293 (1998)
43. S. Trillo, S. Wabnitz, E.M. Wright, G.I. Stegeman, Soliton switching in ber nonlinear
directional couplers. Opt. Lett. 13, 672674 (1988)
44. S. Trillo, S. Wabnitz, Weak-pulse-activated soliton switching in nonlinear couplers. Opt.
Lett. 16, 13 (1991)
45. M. Romagnoli, S. Trillo, S. Wabnitz, Soliton switching in nonlinear couplers. Opt. Quant.
Electron. 24, S1237S1267 (1992)
46. J.D. Joannopoulos, R.D. Meade, J.N. Winn, Photonic Crystals: Molding the Flow of Light
(Princeton University Press, Princeton, 1995)
47. J.D. Joannopoulos, P.R. Villeneuve, S. Fan, Photonic crystals: putting a new twist on light.
Nature 386, 143149 (1997)
48. M. Soljacic, M. Ibanescu, S.G. Johnson, Y. Fink, J.D. Joannopoulos, Optimal bistable
switching in nonlinear photonic crystals. Phys. Rev. E 66, 055601 (2002)
49. S.F. Mingaleev, Y.S. Kivshar, Nonlinear transmission and light localization in photonic-
crystal waveguides. J. Opt. Soc. Am. B 19, 22412249 (2002)
50. M. Soljacic, C. Luo, J.D. Joannopoulos, S. Fan, Nonlinear photonic crystal microdevices for
optical integration. Opt. Lett. 28, 637639 (2003)
51. S. Boscolo, M. Midrio, C.G. Someda, Coupling and decoupling of electromagnetic waves in
parallel 2-D photonic crystal waveguides. IEEE J. Quantum Electron. 38, 4753 (2002)
52. A. Martinez, F. Cuesta, J. Marti, Ultrashort 2-D photonic crystal directional couplers. IEEE
Photon. Technol. Lett. 15, 694696 (2003)
53. M. Thorhauge, L.H. Frandsen, P.I. Borel, Efcient photonic crystal directional couplers. Opt.
Lett. 28, 15251527 (2003)
54. A. Taflove, S.C. Hagness, Computational Electrodynamics: The Finite-Difference Time-
Domain Method (Artech House, Norwood, 2000)
55. R.M. Joseph, A. Taflove, FDTD Maxwells equations models for nonlinear electrodynamics
and optics. IEEE Trans. Antennas Propag. 45, 364374 (1997)
56. A. Locatelli, D. Modotto, D. Paloschi, C. De Angelis, All optical switching in ultrashort
photonic crystal couplers. Opt. Commun. 237, 97102 (2004)
57. J.S. Aitchison, D.C. Hutchings, J.U. Kang, G.I. Stegeman, A. Villeneuve, The nonlinear optical
properties of AlGaAs at the half band gap. IEEE J. Quantum Electron. 33, 341348 (1997)
58. S.G. Johnson, P.R. Villeneuve, S. Fan, J.D. Joannopoulos, Linear waveguides in photonic-
crystal slabs. Phys. Rev. B 62, 82128222 (2000)
59. S.G. Johnson, J.D. Joannopoulos, Block-iterative frequency-domain methods for Maxwells
equations in a planewave basis. Opt. Express 8, 173190 (2001)
60. T.D. Happ, M. Kamp, A. Forchel, Photonic crystal tapers for ultracompact mode conversion.
Opt. Lett. 26, 11021104 (2001)
61. P. Bienstman, S. Assefa, S.G. Johnson, J.D. Joannopoulos, G.S. Petrich, L.A. Kolodziejski,
Taper structures for coupling into photonic crystal slab waveguides. J. Opt. Soc. Am. B 20,
18171821 (2003)
62. A.L. Reynolds, U. Peschel, F. Lederer, P.J. Roberts, T.F. Krauss, P.J.I. de Maagt, Coupled
defects in photonic crystals. IEEE Trans. Microwave Theory Tech. 49, 18601867 (2001)
63. U. Peschel, A.L. Reynolds, B. Arredondo, F. Lederer, P.J. Roberts, T.F. Krauss, P.J.I. de
Maagt, Transmission and reflection analysis of functional coupled cavity components. IEEE
64. D.N. Christodoulides, N.K. Efremidis, Discrete temporal solitons along a chain of nonlinear
coupled microcavities embedded in photonic crystals. Opt. Lett. 27, 568570 (2002)
65. K. Al-hemyari, A. Villeneuve, J.U. Kang, J.S. Aitchison, C.N. Ironside, G.I. Stegeman,
Ultrafast all-optical switching in GaAlAs directional couplers at 1.55 m without
multiphoton absorption. Appl. Phys. Lett. 63, 35623564 (1993)
66. J.S. Aitchison, A. Villeneuve, G.I. Stegeman, All-optical switching in two cascaded
nonlinear directional couplers. Opt. Lett. 20, 698700 (1995)
67. A. Auditore, C. De Angelis, A. Locatelli, A.B. Aceves, Tuning of surface plasmon polaritons
beat length in graphene directional couplers. Opt. Lett. 38, 42284231 (2013)
68. A. Auditore, C. De Angelis, A. Locatelli, S. Boscolo, M. Midrio, M. Romagnoli, A.-D.
Capobianco, G. Nalesso, Graphene sustained nonlinear modes in dielectric waveguides. Opt.
Lett. 38, 631633 (2013)
69. D.A. Smirnova, A.V. Gorbach, I.V. Iorsh, I.V. Shadrivov, Y.S. Kivshar, Nonlinear switching
with a graphene coupler. Phys. Rev. B 88, 045433 (2013)
70. P.I. Buslaev, I.V. Iorsh, I.V. Shadrivov, P.A. Belov, Y.S. Kivshar, Plasmons in waveguide
structures formed by two graphene layers. JETP Lett. 97, 535539 (2013)
71. A. Locatelli, A.-D. Capobianco, G. Nalesso, S. Boscolo, M. Midrio, C. De Angelis,
Graphene based electro-optical control of the beat length of dielectric couplers. Opt.
Commun. 318, 175179 (2014)
72. Y.V. Bludov, D.A. Smirnova, YuS Kivshar, N.M.R. Peres, M.I. Vasilevskiy, Nonlinear TE-
polarized surface polaritons on graphene. Phys. Rev. B 89, 035406 (2014)
73. S.A. Mikhailov, K. Ziegler, New electromagnetic mode in graphene. Phys. Rev. Lett. 99,
016803 (2007)
74. S.A. Mikhailov, K. Ziegler, Nonlinear electromagnetic response of graphene: frequency
multiplication and the self-consistent-eld effects. J. Phys.: Condens. Matter 20, 110 (2008)
75. M. Jablan, H. Buljan, M. Soljacic, Plasmonics in graphene at infra-red frequencies. Phys.
Rev. B 80, 245435 (2009)
76. H. Zhang, S. Virally, Q. Bao, L.K. Ping, S. Massar, N. Godbout, P. Kockaert, Z-scan
measurement of the nonlinear refractive index of graphene. Opt. Lett. 37, 18561858 (2012)
77. A.V. Gorbach, Nonlinear graphene plasmonics: amplitude equation for surface plasmons.
Phys. Rev. A 87, 013830 (2013)
78. L. Ju, B. Geng, J. Horng, C. Girit, M. Martin, Z. Hao, H.A. Bechtel, X. Liang, A. Zettl, Y.R.
Shen, F. Wang, Nat. Nanotechnol. 6, 630 (2011)
79. J. Chen, M. Badioli, P. Alonso-Gonzalez, S. Thongrattanasiri, F. Huth, J. Osmond, M.
Spasenovic, A. Centeno, A. Pesquera, P. Godignon, A. Zurutuza Elorza, N. Camara, F. Javier
Garcia De Abajo, R. Hillenbrand, F.H.L. Koppens, Optical nano-imaging of gate-tunable
graphene plasmons. Nature 486, 7781 (2012)
80. Z. Fei, A.S. Rodin, G.O. Andreev, W. Bao, A.S. McLeod, M. Wagner, L.M. Zhang, Z. Zhao,
G. Dominguez, M. Thiemens, M.M. Fogler, A.H. Castro-Neto, C.N. Lau, F. Keilmann, D.N.
Basov, Gate-tuning of graphene plasmons revealed by infrared nano-imaging. Nature 487,
8285 (2012)
81. B. Wang, X. Zhang, X. Yuan, J. Teng, Optical coupling of surface plasmons between
graphene sheets. Appl. Phys. Lett. 100, 131111 (2012)
82. A. Vakil, N. Engheta, Transformation optics using graphene. Science 332, 12911294 (2011)
83. T. Stauber, N.M.R. Peres, A.K. Geim, Optical conductivity of graphene in the visible region
of the spectrum. Phys. Rev. B 78, 085432 (2008)
84. C.M. Bender, S. Boettcher, Real spectra in non-Hermitian Hamiltonians having PT
symmetry. Phys. Rev. Lett. 80, 52435246 (1998)
85. G.W. Hanson, Dyadic Greens functions for an anisotropic, non-local model of biased
graphene. IEEE Trans. Antennas Propag. 56, 747757 (2008)
86. A. Locatelli, A.-D. Capobianco, M. Midrio, S. Boscolo, C. De Angelis, Graphene-assisted
control of coupling between optical waveguides. Opt. Express 20, 28479 (2012)
87. S. Klaiman, U. Gunther, N. Moiseyev, Visualization of branch points in PT-symmetric
waveguides. Phys. Rev. Lett. 101, 080402 (2008)
88. A. Guo, G.J. Salamo, D. Duchesne, R. Morandotti, M. Volatier-Ravat, V. Aimez, G.A.

Siviloglou, D.N. Christodoulides, Observation of PT-symmetry breaking in complex optical
potentials. Phys. Rev. Lett. 103, 093902 (2009)
89. S. Yu, G.X. Piao, D.R. Mason, S. In, N. Park, Spatiospectral separation of exceptional points
in PT-symmetric optical potentials. Phys. Rev. A 86, 031802 (2012)
90. H. Yang, X. Feng, Q. Wang, H. Huang, W. Chen, A.T.S. Wee, W. Ji, Giant two-photon
absorption in bilayer graphene. Nano Lett. 11, 26222627 (2011)
91. K. Kim, J.Y. Choi, T. Kim, S.H. Cho, H.J. Chung, A role for graphene in silicon-based
semiconductor devices. Nature 479, 338344 (2011)
92. F. Bonaccorso, Z. Sun, T. Hasan, A.C. Ferrari, Graphene photonics and optoelectronics. Nat.
Photon. 4, 611622 (2010)
93. M. Liu, X. Yin, E. Ulin-Avila, B. Geng, T. Zentgraf, L. Ju, F. Wang, X. Zhang, A graphene-
based broadband optical modulator. Nature 474, 6467 (2011)
94. M. Liu, X. Yin, X. Zhang, Double-layer graphene optical modulator. Nano Lett. 12,
14821485 (2012)
95. M. Midrio, S. Boscolo, M. Moresco, M. Romagnoli, C. De Angelis, A. Locatelli, A.-D.
Capobianco, Graphene-assisted critically-coupled optical ring modulator. Opt. Express 20,
2314423155 (2012)
96. Q. Bao, H. Zhang, B. Wang, Z. Ni, C. Lim, Y. Wang, D.Y. Tang, K.P. Loh, Broadband
graphene polarizer. Nat. Photon. 5, 411415 (2011)
97. J.T. Kim, C.G. Choi, Graphene-based polymer waveguide polarizer. Opt. Express 20,
35563562 (2012)
98. Z.Q. Li, E.A. Henriksen, Z. Jiang, Z. Hao, M.C. Martin, P. Kim, H.L. Stormer, D.N. Basov,
Dirac charge dynamics in graphene by infrared spectroscopy. Nat. Phys. 4, 532535 (2008)
99. A.-D. Capobianco, A. Locatelli, C. De Angelis, M. Midrio, S. Boscolo, Finite-difference
beam propagation method for graphene-based devices. IEEE Photon. Technol. Lett. 26,
10071010 (2014)
100. D.N. Christodoulides, R.I. Joseph, Discrete self-focusing in nonlinear arrays of coupled
waveguides. Opt. Lett. 13, 794796 (1988)
101. T. Pertsch, T. Zentgraf, U. Peschel, A. Brauer, F. Lederer, Anomalous refraction and
diffraction in discrete optical systems. Phys. Rev. Lett. 88, 093901 (2002)
102. A. Locatelli, M. Conforti, D. Modotto, C. De Angelis, Diffraction engineering in arrays of
photonic crystal waveguides. Opt. Lett. 30, 28942896 (2005)
103. A. Locatelli, M. Conforti, D. Modotto, C. De Angelis, Discrete negative refraction in
photonic crystal waveguide arrays. Opt. Lett. 31, 13431345 (2006)
104. M. Guasoni, A. Locatelli, C. De Angelis, Peculiar properties of photonic crystal binary
waveguide arrays. J. Opt. Soc. Am. B 25, 15151522 (2008)
105. M. Conforti, M. Guasoni, C. De Angelis, Subwavelength diffraction management. Opt. Lett.
33, 26622664 (2008)
Chapter 4
Temporal and Spectral Nonlinear Pulse
Shaping Methods in Optical Fibers
Sonia Boscolo, Julien Fatome, Sergei K. Turitsyn, Guy Millot

and Christophe Finot
Abstract The combination of the third-order optical nonlinearity with chromatic

dispersion in optical bers offers an extremely rich variety of possibilities for
tailoring the temporal and spectral content of a light signal, depending on the
regime of dispersion that is used. Here, we review recent progress on the use of
third-order nonlinear processes in optical bers for pulse shaping in the temporal
and spectral domains. Various examples of practical signicance will be discussed,
spanning elds from the generation of specialized temporal waveforms to the
generation of ultrashort pulses, and to stable continuum generation.
4.1 Introduction
The third-order v3 optical nonlinearity in silica-based single-mode (SM) bers is

responsible for a wide range of phenomena such as third-harmonic generation,
nonlinear refraction (Kerr nonlinearity), and stimulated Raman and Brillouin
scattering [1]. In high-speed optical communications nonlinear effects generally
degrade the integrity of the transmitted signal, but the same effects, when properly
managed, can be used to realize a variety of optical functions that have practical
applications in the eld of lightwave technology. Nonlinear processes that have
been exploited in demonstrations and applications include stimulated Brillouin and
Raman scattering, as well as aspects of the Kerr effect variously called self-phase
S. Boscolo (&) S.K. Turitsyn

Aston Institute of Photonic Technologies, School of Engineering and Applied Science,
Aston University, Birmingham B4 7ET, UK
e-mail: s.a.boscolo@aston.ac.uk
J. Fatome G. Millot C. Finot
Laboratoire Interdisciplinaire CARNOT de Bourgogne, UMR 6303 CNRS-Universit de
Bourgogne, Dijon, France
e-mail: christophe.not@u-bourgogne.fr

106 S. Boscolo et al.
modulation (SPM), cross-phase modulation, four-photon (four-wave) mixing

(FWM) and parametric gain.
Techniques for generating, controlling, manipulating, and measuring ultrashort
optical pulses and specialized waveforms have become increasingly strategically
important in many scientic areas including, amongst others, ultrahigh-speed optical
communications, optical signal processing, and biophotonics. Picosecond and fem-
tosecond pulse shaping in the optical domain have been extensively implemented
using the Fourier-domain approach [2], which employs spectral manipulation of the
intensity and phase components of a pulse in order to create the desired eld dis-
tribution. Though powerful and flexible, this pulse shaping strategy has the drawback
that the bandwidth of the output spectrum is determined by the bandwidth of the input
spectrum. Indeed, a linear manipulation cannot increase the pulse bandwidth, and so
to create shorter pulses nonlinear effects must be used. In addition, a linear pulse
shaper can only subtract power from the frequency components of the signal while
manipulating its intensity, thereby potentially making the whole process extremely
power inefcient. The combination of third-order nonlinear processes and chromatic
dispersion in optical bers can provide efcient new solutions to overcome the
aforementioned limitations. As we will see in this chapter, nonlinear ber-based
pulse shaping has become a remarkable tool to tailor both the temporal and spectral
content of a light signal without the need for dedicated linear shapers. It is important
to note that though we will use a silica optical ber as a key example system to
demonstrate various effects, these general nonlinear science concepts can be applied
to a range of other nonlinear optical materials and media.
In this chapter, we review recent progress in the research on pulse shaping using
nonlinear effects in optical bers. We would like to note that the main attention in
this chapter will be focused on results obtained in our groups, and that it is not our
intention here to comprehensively cover all the possible examples of ber-based
pulse shaping. The chapter is organized as follows: Sect. 4.2 introduces the key
model governing the optical pulse propagation in a ber, and briefly discusses the
resultant qualitative features of the pulse evolution. Section 4.3 emphasizes how the
normal dispersion regime of a ber is particularly well-suited for the generation of
advanced temporal waveforms and stable continua with high power spectral den-
sity, as well as for spectral compression of the pulses. Section 4.4 highlights some
of the benets offered by the governing soliton dynamics in the anomalous dis-
persion region in terms of nonlinear shaping, namely, the possibility of generating
ultrashort temporal structures and ultrabroad spectra, and of tuning the central
frequency of the pulses. Finally, we conclude in Sect. 4.5.
4.2 Pulse Propagation in Optical Fibers
A variety of nonlinear phenomena associated with optical pulse propagation in SM

bers is well described by the nonlinear Schrdinger (NLS) equation [3], in spite of
the fact that this equation only includes two physical effects, namely, linear group
4 Temporal and Spectral Nonlinear Pulse Shaping 107
velocity dispersion (GVD) and nonlinear SPM. Here, a modied NLS equation for
the pulse envelope, wz; t, is taken, which includes the effect of linear gain,
namely,
ib2 1
wz w icjwj2 w gw; 4:1
2 tt 2
where z is the propagation distance, t is the reduced time, b2 and c are the respective
GVD and cubic nonlinearity coefcients of the ber, and g is the distributed gain
coefcient (which is nonzero only for an active ber). This equation neglects
higher-order linear effects such as third-order dispersion as well as higher-order
nonlinear effects such as self-steepening or intrapulse Raman scattering [3].
Although these effects can have noticeable impact on pulses shorter than 1 ps, here
we neglect them as the leading-order behavior is well approximated by (4.1).
Moreover, neglecting higher-order gain effects such as gain bandwidth and satu-
ration is well-suited to describe experiments that use high-gain, broadband rare-
earth ber ampliers [4].
It is useful to introduce here the dispersion length LD s20 =jb2 j and the nonlinear
length LNL 1=cP0 , which provide the length scales over which dispersive or
nonlinear effects become important for pulse evolution. The parameters s0 and P0 are
some characteristic temporal value and the peak power of the input pulse, respec-
tively. Linear propagation regime occurs when N 2 LD =LNL 1. In this case pulse
evolution can be simply determined by using the Fourier-transform method. As a
result of GVD, the phase of each spectral component of the pulse is changed by an
amount b2 x2 z=2, where x is the angular frequency of the pulse envelope. Even
though such phase changes do not affect the pulse spectrum, they lead to temporal
broadening of the pulse and a time dependence of the pulse phase, which can be
characterized by the instantaneous frequency shift or chirp dx /t . In the
opposite case when N 2 1, the nonlinearity-dominant regime is applicable. As a
result of the nonlinear term in (4.1), upon propagation in the ber the pulse acquires
Rz
an intensity-dependent nonlinear phase shift as /NL z; t cjw0; tj2 0 dz0
R z0
expGz0 with Gz0 0 df gf, namely the frequency chirp dx /NL t . In
the absence of GVD and for an initially transform-limited pulse, the presence of a
chirp causes a nonlinear broadening of the pulse spectrum.
When the effects of chromatic dispersion are considered in combination with the
Kerr nonlinearity, rich pulse dynamics arise from the interplay between dispersive
and nonlinear effects depending on the sign of the dispersion and the relative
magnitudes of the associated length scales. A well-known and fascinating example
is the formation of optical solitons [5, 6] in the anomalous dispersion b2 \0
regime of a passive ber as a result of a cooperation between GVD and SPM. In the
anomalous dispersion region, if N 1, the chirp induced by chromatic dispersion
can be compensated by the SPM-induced chirp, which has opposite sign. The pulse
shape adjusts itself during propagation to make such cancelation as complete as
possible, hence leading to soliton evolution. On the contrary, in the regime of
normal dispersion pulse dynamics are highly affected by the phenomenon of optical
wave breaking [710]. In the nonlinearity-dominant regime of propagation, the
combination of normal GVD and SPM makes a pulse broaden and change shape
toward an almost rectangular pulse form with a linear frequency chirp variation
over most of the pulse, as it is highlighted by Fig. 4.1 which shows the temporal
and spectral evolution of an initially unchirped hyperbolic secant pulse in the
normal and anomalous dispersion regimes of a ber using N 11:4 and g 0.
This evolution can be understood as arising from the frequency chirp, which makes
the red- (blue-) shifted light near the leading (trailing) edge travel faster (slower)
than the unshifted pulse parts, i.e., the pulse center and low-intensity wings. Ulti-
mately, when the shifted light overruns the pulse tails, the wave breaks; oscillations
appear in the wings of the pulse because of interference and, concomitantly, new
frequencies are generated and side lobes appear in the pulse spectrum (Fig. 4.4a).
In the next sections we will discuss some newly emerged qualitative features of
the optical pulse evolution in a ber under the combined action of dispersion and
nonlinearity.
4.3 Pulse Shaping in Normally Dispersive Fibers: From

the Generation of Specialized Temporal Waveforms
to Spectral Sculpturing
4.3.1 Generation of Specialized Temporal Waveforms
As a rst illustration of nonlinear pulse shaping in normally dispersive (ND) bers,

we discuss here the possibility of generating advanced temporal waveforms, such as
parabolic or triangular proles, which are of interest for various applications in
optical signal processing and manipulation [11].
4.3.1.1 Parabolic Pulse Generation
We have seen in the previous section that pulses that propagate in normal dispersion
media are susceptible to distortion and breakup owing to optical wave breaking,
which is a consequence of excessive nonlinear phase accumulation combined with
dispersion. In 1993 Anderson et al. showed that a sufcient condition to avoid wave
breaking is that a pulse acquires a monotonic frequency chirp as it propagates, and that
wave breaking-free solutions of the NLS equation (4.1) exist when the GVD is normal
[12]. These are asymptotic solutions of (4.1) in the quasi-classical or WKB limit
(i.e., the limit of high amplitude or small dispersion such that b2 jjwjtt j=2cjwj3 can
be neglected), and take the form
(a1) (a2)
(c)
(d)
Propagation distance (m)
(b1) (b2)
Intensity (arb. units)
Propagation distance (m)
Fig. 4.1 Evolution of an initially unchirped hyperbolic secant pulse with P0 1 W and s0
0:85 ps propagating in an highly nonlinear (HNL) ber with jb2 j 0:64 ps2 =km and
c 10 W km1 , as obtained by numerical integration of the NLS equation. Longitudinal
evolution of the a temporal and b spectral intensity proles in the normal (subplots 1) and
anomalous (subplots 2) dispersion regimes of the ber. Temporal c chirp and d intensity proles
after 350 m of propagation, i.e. 3.5 LNL , in the ber in the case of normal (black solid) and
anomalous (gray solid) GVD. Also shown is the initial temporal prole (dashed black)
q
jwz; tj az 1 t=sz2 hsz jtj; arg wz; t bzt2 /0 z; 4:2
and thus have parabolic intensity proles. Here, hx 1; x [ 0; hx 0; x\0

is the Heaviside function. Such a pulse maintains its shape, and is always a scaled
version of itself; i.e., it evolves self-similarly. In contrast to solitons, self-similar
pulses, which
R have been dubbed similaritons, can tolerate strong nonlinearity
(/NL;0 dzczP0 z p, where /NL;0 is the peak nonlinear phase shift) with-
out distortion or wave breaking (Fig. 4.4a). The normal GVD effectively linearizes
the accumulated phase of the pulse allowing for the spectral bandwidth to increase
without destabilizing the pulse. Similaritons were demonstrated theoretically and
10
Chirp (THz)
Chirp (THz)
.4
(a) 5 (b1) (c)
Power (W)
0
0
5 1 -.4

-5
(10 dB/div)
.8
x5
0 .6
Chirp (THz)
.2
(b2) .4
Intensity
0
.2
-.2 0
-4 0 4 -200 -100 0 100 200 -20 -10 0 10 20
Time (ps) Time (ps) Time (ps)
Fig. 4.2 Parabolic pulse generation at telecommunication wavelengths. a Experimental (solid)

and numerical (dashed) temporal intensity and chirp proles at the output of an erbium-doped ber
amplier. Adapted from [17]. b Results obtained in a Raman ber amplier: experimental
temporal intensity and chirp proles at the ber output (solid black) compared to a parabolic and a
linear t (circles), respectively. The initial intensity prole is also shown (solid gray). More details
on the experiment can be found in [20]. c Experimentally measured 40-GHz pulse train and chirp
prole at the input (dashed) and output (solid) of a passive segment of ND ber followed by an
HNL ber. Also shown are parabolic and linear ts to the output pulse shape and chirp (circles).
More details can be found in [73]
experimentally in ber ampliers in 2000 [4], and they continue to attract much
attention [13]. Similaritons in ber ampliers are along with solitons in passive
ber the most well-known classes of nonlinear attractors for pulse propagation in
optical bers [4, 14, 15], so they take on major fundamental importance. Various
experimental demonstrations relying on amplication from either rare-earth doping
[4, 1618] (Fig. 4.2a) or Raman scattering [19, 20] (Fig. 4.2b) have been achieved.
Experimental interest has been strongly driven by the possibility of fully canceling
the highly linear chirp developed during parabolic amplication and, thus, gener-
ating ultrashort high-power pulses with very low substructures [4, 1618]. Fur-
thermore, recent ber lasers that support similariton evolution in the passive [21] or
gain [2225] ber segment of the laser cavity have been demonstrated to achieve
pulse energy and peak power performances much higher than those of prior
approaches.
In addition to ber ampliers, similariton formation with nonlinear attraction can
be achieved in passive bers provided a suitable longitudinal variation of the dis-
persion is introduced [2628]. This approach is based on the observation that a
longitudinal decrease of the normal dispersion is formally equivalent to linear gain.
Recently, a simple approach to the generation of parabolic pulses that uses pro-
gressive nonlinear pulse reshaping in a ND ber with xed dispersion has been also
demonstrated [29] (Fig. 4.2c). However, in contrast with the asymptotic attracting
solutions obtained in ber ampliers, the generated parabolic waveforms represent
transient states of the nonlinear pulse evolution in the ber medium [30] and can be
associated with an intermediate asymptotic regime [31] of the pulse propagation. As
such, they have a nite life distance (characteristic length scale of the self-similar
pulse dynamics) that depends on the initial conditions. Nevertheless, stabilization of

the parabolic features is possible by use of a second propagation stage in a ber
with specially adjusted nonlinear and dispersive characteristics relative to the rst
ber [29, 30].
4.3.1.2 Triangular Pulse Generation
Parabolic shapes are not the only pulse waveforms that can be generated in a
passive ND ber. Indeed, it has been shown numerically that temporal triangular
intensity proles can result from the evolution of initially rectangular pulses in the
highly nonlinear (quasi-classical) regime of propagation [32]. Another approach
that has been recently demonstrated theoretically relies on the progressive reshaping
of initially parabolic pulses driven by the fourth-order dispersion (FOD) of the ber
[33]. The overall temporal effect of FOD on parabolic pulse propagation is to
stretch and enhance the power reduction in the pulse wings, leading to a triangular
prole. In [30], the combination of pulse prechirping and nonlinear propagation in a
section of ND ber has been introduced as a method for passive nonlinear pulse
shaping, which provides a simple way of generating various advanced eld dis-
tributions, including flat-top- and triangular-proled pulses with a linear frequency
chirp. In this scheme, Kerr nonlinearity and GVD lead to various reshaping pro-
cesses of an initial conventional eld distribution (e.g., a Gaussian pulse) according
to the chirping value and power level at the entrance of the ber. In particular,
triangular pulses can be generated for an anomalous (negative) initial chirp and
sufciently high energies. These theoretical results have been conrmed experi-
mentally by intensity (Fig. 4.3b) and phase measurements of the generated pulses
[34, 35]. In the experimental setup used (Fig. 4.3a), the control of the pulse pre-
chirping value was realized by propagation through different lengths of standard
SM ber with anomalous GVD, which imposed a negative chirp on the pulse. The
prechirped pulses were amplied to different power levels using an erbium-doped
1
(b1) (b2)
(a) SMF NDF .8
.6
.4
MLFL
.2
2.5 GHz
EDFA 0
-50 0 50 -50 0 50
Time (ps)
Fig. 4.3 a Experimental setup for triangular pulse generation. MLFL mode-locked ber laser.
b Measured (solid black) temporal intensity proles at the entrance (subplot b1) and the exit
(subplot b2) of the reshaping ber. Also shown is the corresponding triangular t (dashed gray).
More details can be found in [35]
ber amplier (EDFA) and then propagated through a ND ber to realize the pulse
reshaping.
Furthermore, the possibility of triangular pulse shaping in mode-locked ber
lasers has been rst reported in [36]. It has been numerically demonstrated that for
normal net dispersion, formation of two distinct steady-state solutions of stable
single pulses can be obtained in a laser cavity in different regions of the system
parameter space: the previously known similariton [21] and a triangular-proled
pulse with a linear chirp.
4.3.2 Spectral Sculpturing
4.3.2.1 Generation of Highly Coherent Continua
Nonlinear pulse shaping represents the ideal way of generating spectral bandwidth
that exceeds signicantly that of the input seed pulse. This is obviously not possible
with linear shaping techniques where one cannot shift energy outside the initial
pulse bandwidth. Optical ber-based supercontinuum (SC) sources have become a
signicant scientic and commercial success, with applications ranging from fre-
quency comb production to advanced medical imaging and telecommunication
applications [37, 38]. While typically the broadest spectra are generated in anom-
alously dispersive bers, where the broadening mechanism is strongly influenced
by soliton dynamics (Sect. 4.4), ND bers are undoubtedly advantageous as far as
the flatness and temporal coherence of the generated continuum is concerned.
Specically, nonlinear propagation in the normal dispersion regime inhibits mod-
ulation instability (MI) phenomena to occur, which reduces signicantly the
spontaneous amplication of noise and thereby shot-to-shot fluctuations [39].
Consequently, a high degree of coherence is maintained. However, the main limits
to spectral pulse quality in the normal dispersion regime are the spectral ripple
arising from SPM of conventional laser pulses, and the effects of optical wave
breaking which may lead to signicant changes in the temporal pulse shape and
severe energy transfer into the wings of the spectrum. These effects can in principle
be avoided by using preshaped input pulses with a parabolic temporal intensity
prole, which would preserve their shape whilst propagating within the ber and,
thus, result in spectrally flat, highly coherent pulses (Fig. 4.4a). Continuum with
low spectral ripple and high-energy density in the central part has been indeed
demonstrated by use of parabolic pulses generated in erbium-doped bers [40] or
resulting from linear shaping by ber Bragg gratings [41] (Fig. 4.4b) and recon-
gurable liquid-crystal-on-silicon devices [42]. More advanced techniques for
achieving even flatter spectra involve the use of a feedback loop relying on a
genetic algorithm to obtain a nearly rectangular output spectrum [43], as Fig. 4.5
shows. This demonstrates the potential of nonlinear pulse shaping assisted by a
linear shaping element as an additional sculpturing element for generating on-
demand tailored spectra without the requirement of expensive laser sources.
2.5
5
(a1) (a2) 10-3
2
(b)

4 -4
10
Power (W)
1.5
3
10-5
2 1
10-6
1 0.5
10-7
0 0 1530 1540 1550 1560
-5 0 5 -2 0 2
Wavelength (nm)
Time (ps) Frequency (THz)
Fig. 4.4 a Temporal (subplot a1) and spectral (subplot a2) intensity proles at the exit of a ND-
HNL ber with the same parameters as in Fig. 4.1 for initially unchirped hyperbolic secant (solid
black) and parabolic (solid gray) pulses with P0 5 W, as obtained by numerical integration of
the NLS equation. Also shown are the initial pulse temporal proles (dashed). b Experimental
spectrum of a parabolic-shaped pulse after propagation in a ND-HNL ber (black dashed) and
superposition of the measured spectra of 38 sliced channels. Adapted from [41]
The high degree of coherence of the spectra generated in the normal dispersion
regime is clearly attractive for all-optical pulse processing in HNL bres [11], or the
generation of SC covering the entire C-band of optical telecommunications. Indeed,
several studies have reported that such continua can be sliced into several tens of
high-quality, high-repetition-rate picosecond channels where isolated spectral lines
of the 10 or 40-GHz resulting comb provide several hundreds of continuous waves
[41, 44, 45] (Fig. 4.4b). Furthermore, high temporal coherence enables pulse
compression with very high compression ratios as the SPM-induced nonlinear chirp
can be compensated to a large extent [17, 46]. This technique has been efciently
exploited in photonic crystal (PC) bers where the dispersion characteristics can be
accurately controlled so as to obtain a ber with normal dispersion at all wave-
lengths [33], which enables the generation of highly broadband coherent spectra
spanning more than several hundreds of nanometers. Such spectra are particularly
attractive from the perspective of pulse compression down to durations as short as a
few optical cycles after compensation of the nonlinear phase [47, 48].
4.3.2.2 Spectral Compression
Spectral narrowing of optical pulses can be rather easily realized with standard
optical bandpass ltering, but at the expenses of signicant optical losses. Indeed,
as we mentioned in the introduction, linear ltering introduces a power penalty that
is at least proportional to the ratio of the target spectrum to the input one. The effect
of SPM on a pulse propagating in an optical ber is generally associated with
spectral broadening of the pulse. However, this is not always the case with chirped
(a) ND HNLF
0 (b)
Power (dBm)
linear pulse -2
shaper 30nm
ps laser OSA -4
35nm
HP-EDFA -6
computer
-8
1530 1540 1550 1560 1570
Wavelength (nm)
Fig. 4.5 a Experimental setup for the nonlinear generation of ultra-flat broadband frequency
combs assisted by adaptive pulse shaping. b SPM spectra generated from 2 ps Gaussian pulses
after the application of the adaptive system. More details can be found in [43]
pulses, where down-chirped pulses in the normal GVD region can experience
spectral narrowing [49]. Indeed, the SPM-induced nonlinear phase shift results in a
frequency downshift of the leading edge and an upshift of the trailing edge of the
pulse, i.e., a positive chirp. Consequently, for a negatively chirped input pulse the
linear and nonlinear phases cancel each other, leading to a redistribution of long and
short wavelengths toward the center wavelength and therefore to spectral com-
pression instead of spectral broadening. Spectral compression has been reported for
various parameters [5052]. However, in general an input pulse with a negative
linear chirp, jw0; tj expibt2 (with b [ 0), cannot be spectrally compressed to the
Fourier transform limit because the instantaneous frequency (in the absence of
dispersion) /t 2bt cjw0; tj2 t z cannot in general be made equal to zero for
all times, except for special cases [53] that require the balancing of the (higher-
order) dispersion and chirp. This results in residual side-lobes in the compressed
spectrum, as illustrated in Fig. 4.6b. On the other hand, if an input negatively
linearly chirped parabolic pulse is used, the instantaneous frequency /t 2b
ca2 z=s20 t can be chosen to yield zero for all t, for a suitable combination of c, a, z,
and b [54]. Because, at the same time, the parabolic temporal envelope with width
s0 remains unchanged (as long as dispersion in the ber can be ignored), this
expresses the fact that spectral compression to the transform limit takes place.
Experimental demonstrations at telecommunication wavelengths using a HNL ber
[55] or at the wavelengths around 1 m using a PC ber [56] have conrmed the
potential of this method for achieving high-quality compressed spectra with very
low substructures and enhanced brightness (Fig. 4.6c).
Furthermore, in-cavity nonlinear spectral compression in a mode-locked ber
laser has been recently demonstrated numerically [57]. A new concept of a ber
laser architecture has been presented, which supports self-similar pulse evolution in
the amplier ber segment of the laser cavity and nonlinear spectral compression in
the passive ber, thereby enabling the generation of highly chirped parabolic pulses
and nearly transform-limited spectrally compressed picosecond pulses from a single
device.
(a1) (b) (c1)

.2
Chirp (THz)
2
10
Input peak power (W)

0 1
8 60
-.2 0
6 40
(a2) (c2)
Power (W)
1 2
4 20
.5 2 1 0
0 0 0
-10 0 10 -.4 -.2 0 .2 .4 -100 -50 0 50 100
Time (ps) Frequency (THz) Frequency (GHz)
Fig. 4.6 a Temporal intensity and chirp proles, and b spectral intensity proles of an hyperbolic
secant pulse before (gray) and after (black) spectral compression in a ND-HNL ber (same
parameters as in Fig. 4.1), as obtained by numerical integration of the NLS equation.
c Experimentally measured dependence of the compressed spectral intensity prole on the input
peak power for initially negatively chirped hyperbolic secant (subplot c1) and parabolic (subplot
c2) pulses in an HNL ber. More details can be found in [55]
4.4 Pulse Shaping in the Anomalous Dispersion Regime:

From Ultrashort Temporal Structures to Ultra-Broad
Spectra
4.4.1 Generation of High-Repetition-Rate Ultrashort Pulses
In this section, we review the generation of high-repetition-rate ultrashort pulses

that can result from the nonlinear evolution of optical pulses in anomalously dis-
persive bers.
4.4.1.1 Solitonic Temporal Compression
As mentioned in Sect. 4.2, when b2 \0 the standard NLS equation ((4.1) with
g 0) has specic pulse-like solutions, known as optical solitons, that either do not
change along the ber length (fundamental solitons, resulting from an exact balance
between the GVD and SPM effects: N 1) or follow a periodic evolution pattern
(higher-order solitons, characterized by integers N [ 1). A technique for generating
high-quality ultrashort pulses relies on the adiabatic evolution of a fundamental
soliton that takes place in an anomalously dispersive ber with longitudinally
distributed gain (e.g., a Raman amplier) or decreasing dispersion (DD) [5860].
Indeed, in the presence of gain, the temporal width of a fundamental soliton must
decrease along the ber length to compensate for the power increase due to gain if
the pulse has to preserve its soliton character, i.e., if the requirement N 1 has to
be maintained. While physical distributed-amplication schemes require pumping along

the ber length, which entails an extra level of cost and complexity in the setup, soliton
compression in a ber can also be achieved by changing the dispersive properties of the
ber [61]. This passive approach is based on the fact that the NLS (4.1) with constant
gain coefcient g reduces to the standard NLS equation if the anomalous GVD of the
ber increases in magnitude exponentially along the ber length: b2 z=b2 0
expgz. Step-like dispersion-proled structures [62] using a set of short segments of
conventional bers with different dispersion values, or comb-like structures [63]
where the use of a large number of ber segments with alternate signs of dispersion
allows to approach an average local dispersion with the desired decrease, have been
successfully used to emulate continuous DD ber proles, whose realization requires
special, not commercially available ber-drawing technology.
A more straightforward approach to the generation of pulses with ultrashort
durations uses the stage of temporal compression that occurs in the periodic evo-
lution of higher-order solitons. Indeed, in the case of higher-order solitons, over a
period of the evolution pattern SPM dominates initially but GVD soon catches up
and leads to pulse contraction, as seen in Fig. 4.1. The higher the soliton number N
of the input pulse, the larger the compression achievable using this approach.
However, the compressed pulse exhibits a low-amplitude structure outside the
central lobe (Fig. 4.1). Moreover, given the scaling laws of the NLS equation,
increasing the input pulse peak power leads to a decrease of the distance in the ber
at which compression takes place, while higher-order nonlinear and dispersive
effects become important for pulse evolution. This may eventually make this
method very sensitive to power fluctuations of the input pulse. Feasibility of the
higher-order soliton compression scheme was rst shown in 1980 [64]. Various
experimental demonstrations in either HNL bers at telecommunication wave-
lengths [65] or in PC bers seeded by titanium-sapphire laser pulses [66] have been
achieved since then, enabling the generation of pulses as short as a few cycles.
4.4.1.2 High-Repetition-Rate Sources
High-repetition-rate ultrashort optical pulses are widely used in various areas such
as ultrafast physical processes, ultra-high-data-rate optical communications, optical
frequency metrology and high-accuracy measurements. While current electronics-
based and direct modulation technologies do not permit the generation of pulses at a
repetition rate higher than 40 GHz, the nonlinear transformation of a dual-frequency
signal in an optical ber can provide an attractive solution in this respect. A non-
linear technique based on MI of an intense pump wave induced by a weak signal
was rst suggested in [67] and realized experimentally in [68]. MI in the anomalous
dispersion regime of optical bers manifests itself as breakup of the initial pump
beam into a periodic pulse train whose period is inversely related to the frequency
difference between the pump and signal waves, X x0 x1 , with the generation
of two new spectral sidebands, and can be interpreted in terms of a FWM process
that is phase-matched by SPM. Physically, the energy of two photons from the
.2
(a1) (b)
Chirp (THz)

0
10
-5 (c)

-.2
(a2) 2
Power (W)
-10
1 10
0
-20 0 20 -.5 0 .5 0 25
Time (ps) Frequency (THz) Time (ps)
Fig. 4.7 Generation of high-repetition-rate pulse train using multiwave mixing compression of a
dual-frequency beat signal. a Temporal intensity and chirp proles, and b spectrum of the
compressed pulse train as obtained by numerical simulation of the NLS equation. The initial beat
signal is also shown (gray). c Intensity of the generated pulse train at 40 GHz recorded on an
optical sampling oscilloscope. More details on the experiment can be found in [72]
intense pump beam is used to create two different photons, one at the probe fre-
quency x1 and the other at the idler frequency 2x0 x1 . Since its rst demon-
stration, this technique has been used to create optical sources capable of producing
periodic trains of ultrashort pulses at high but controllable repetition rates.
When the rst-order sidebands become strong, higher-order sidebands located at
x0 mX, m 2; 3; . . . are created in a FWM process. An approach based on
multiple FWM in a single ber with anomalous constant dispersion has been
demonstrated experimentally for the generation of compressed pulse trains from the
propagation of a dual-frequency pump eld, and repetition rates ranging from a few
tens of gigahertz to a few terahertz have been achieved [6973]. The initial beat
signal in these experiments was obtained via temporal superposition of two con-
tinuous waves (CWs) with slightly different frequencies delivered by external
cavity lasers [70, 71], which permitted to achieve higher repetition rates than using
a single CW directly modulated by an intensity modulator. Exemplary experimental
results recorded on a high-speed oscilloscope show that high-quality compressed
pulses with a Gaussian intensity prole and nearly constant phase can be achieved
at a typical duty cycle of 1:5 [72] (Fig. 4.7c). Lower duty cycles can still be
obtained with a single ber, but at the expenses of degraded quality of the pulse
train with the appearance of low-intensity pedestal components and a nonuniform
phase across the pulses. Such issues can be partly avoided by using specially
designed arrangements of segments of bers [72, 73] or other comb-like structures.
4.4.2 Generation of Frequency-Tunable Pulses
One of the possibilities offered by the third-order nonlinearity of optical bers for
pulse shaping is to frequency shift the essential spectral content of ultrashort optical
(a) 1
(b) 1
(c) 1
Normalized gain coeff.

.8 .8 .8
.6 .6 .6
.4 .4 .4
.2 .2 .2
0 0 0
0 -5 -10 -1 5 -2 0 -25 1000 1100 1200 1300 -200 -100 0 100 200
Frequency detuning (THz) Wavelength (nm) Time (fs)
(d1) (d2)
160 0
Normalized intensity (dB)

140
Injected energy (pJ)
-5
120
-10
100
80 -15
60 -20
40
-25
20
-30
1000 1250 1500 1000 1250 1500
Wavelength (nm) Wavelength (nm)
Fig. 4.8 Raman-induced self-frequency shift of solitons. a Raman response of silica bers.
b Experimental spectra obtained at the output of a 2-m long PC ber for different input pulse
energies: 83 pJ (solid black), 125 pJ (dashed gray), and 173 pJ (dash-dot light gray).
c Autocorrelation traces of the frequency-shifted pulses compared with an hyperbolic secant t.
d Experimentally measured dependence of the output spectral intensity prole on the input energy
for 2-m long PC bers with a single (subplot d1) or two zero-dispersion wavelengths (subplot d2)
pulses whose width is close to or less than 1 ps. For pulses with a wide spectrum
(>1 THz), the Raman gain of the ber (Fig. 4.8a) can amplify the low-frequency
components of a pulse by transferring energy from the high-frequency components
of the same pulse. As a result of this phenomenon, called intrapulse Raman scat-
tering, the pulse spectrum shifts toward the low-frequency (red) side as the pulse
propagates inside the ber, a phenomenon referred to as the self-frequency shift
[74]. The physical origin of this effect is related to the delayed nature of the Raman
(vibrational) response [75]. The effects of intrapulse Raman scattering can be
included through the NLS equation by including the term icsR wjwj2 t on the left-
hand-side of (4.1), where the parameter sR is related to the slope of the Raman gain
spectrum [75]. For a soliton propagating inside a ber, the amplitude is not affected
by the Raman effect, but its frequency is shifted by dxR z 8jb2 jsR z=15s40 , as
illustrated experimentally in Fig. 4.8b. The temporal intensity prole remains close
to the hyperbolic secant soliton shape (Fig. 4.8c).
The soliton self-frequency shift (SSFS) has been exploited in the realization of
ber-based sources of spectrally tunable ultrashort optical pulses. Frequency shifts of
several hundreds of nanometers have been achieved in HNL bers. The advent of PC
bers has made it possible to observe SSFS at the operational wavelengths of tita-
nium sapphire lasers, and impressive shifts have been demonstrated, thereby paving
the way for efcient spectrally tunable laser sources for applications in, e.g., bio-
photonics or spectroscopy [76, 77]. In this context, the main factors that may limit
SSFS generation are the absorption due to hydroxil radicals contained in the core
glass of silica bers, the ssion of higher-order solitons into multiple sub-pulses, and
the existence of a second zero-dispersion wavelength for some bers, which induces
a stabilization of the soliton frequency accompanied by the generation of dispersive
waves [78] (Fig. 4.8d2). Furthermore, the combined effects of delayed nonlinear
response and higher-order dispersion can result in a gradual increase of the second-
order dispersion during the propagation. This leads to an adiabatic temporal broad-
ening of the soliton, thereby progressively limiting its Raman-induced frequency
shift [79]. To address this issue, optimization of the dispersion landscape seen by
the frequency-shifted pulse has been proposed as an efcient method to enhance the
performance of ber-based tunable frequency shifters [80].
4.4.3 Supercontinuum Generation and Optical Rogue Waves
Several books and review articles are now available to understand the physical
mechanisms driving the continuum formation and its modeling using the NLS equation
in its extended form (see, e.g., [3, 37, 38, 81]), so in this section we describe only the
basic temporal and spectral features of the continuum generated in the anomalous
dispersion regime of a nonlinear ber, which stem from the involved pulse dynamics.
We also discuss a possible solution to efciently harness the large temporal and spectral
fluctuations that are typically observed in this dispersion regime.
4.4.3.1 Supercontinuum Stability and Extreme Events
The nonlinear dynamics in the anomalous dispersion regime are dominated by an

initial stage of MI of the input picosecond pump pulse induced by noise, followed by
a stage of higher-order soliton temporal compression, and subsequent splitting of the
ultrashort coherent structure at the point of maximum compression due to third-order
dispersion or intra-pulse Raman scattering. Pulse splitting results in the emergence of
several ultrashort solitons with signicant shot-to-shot amplitude and timing jitters,
which prevents this type of SC to be sliced in multiple wavelength channels for
telecommunication applications. The spectral intensity prole is signicantly broader
than those typically obtained in ND bers but also suffers from large fluctuations.
Because the SC properties change signicantly from shot to shot both in the temporal
and spectral domains, the stability of the continuum obtained in the normal regime of
dispersion is generally not preserved in the anomalous dispersion region [39]. The
spectral phase fluctuations have been the subject of intense investigations as early as
2000 [82]. The rst experimental observation of signicant fluctuations in the tem-
poral domain has been reported by Solli et al. in 2007 [83]. Spectral ltering of the
most red-shifted part of the SC spectrum generated in a PC ber by anomalous
pumping resulted into a highly-skewed distribution of the statistical fluctuations of
the corresponding temporal intensity. By analogy with the typical L-shaped distri-
bution observed for extreme events, the intense and extremely rare outliers of the
distribution were named optical rogue waves. Since then, the subject has become
the focus of intense research in the international optical community. This enthusiasm
has indeed largely surpassed the connes of SC generation in PC bers, and optical
rogue waves have been reported in various nonlinear optical systems. A careful
analysis of various congurations has disclosed that the most red-shifted events
detected in the experiment of Solli et al. are fundamental solitons shifted in frequency
by the delayed Raman response of silica [84]. Because such giant solitons are fully
coherent and persist once they form, they are now generally referred to as rogue
solitons [85]. The emergence of such extreme solitons results from the process of
temporal collision between the substructures that are generated from the splitting of
the initial pump pulse. When two ultrashort substructures collide, a temporal peak
appears intermittently, thus reproducing one of the peculiar features of the infamous
oceanic freak waves [86]: their unexpected appearance and decay. Such collisions are
favored by convective mechanisms such as the third-order dispersion or the Raman
response of silica [87]. In the process of collision, an energy exchange also occurs,
leading to the progressive growth of a giant soliton, where most of the energy of the
system will eventually concentrate [88, 89].
The temporal and spectral analysis of the fluctuations associated with anomalous
pumping has led to reconsider the onset of SC generation where higher-order linear
and nonlinear effects do not play a signicant role, and pulse propagation can be
described in the framework of the standard NLS equation. In fact, the stage of MI
that governs the initial spectral broadening can be described analytically within the
framework of the Akhmediev-Kuznetsov breather formalism. Kuznetsov derived
non-stationary solutions of the NLS equation which are periodic in time in plasma
in 1977 [90]. The Akhmediev breathers are space-periodic solutions of the NLS
equation derived by Akhmediev and Korneev in 1986, which describe the single
growth-return cycle of an initially weak sinusoidal modulation [91]. Numerical
simulations and experiments have indeed conrmed that the properties of SC at the
early stages of propagation can be accurately described in terms of multiple
breathers corresponding to the excitation of the entire MI gain bandwidth [92].
4.4.3.2 Highly Coherent Supercontinuum Generation and Peregrine

Solitons
One way to limit the degradation of the SC coherence in the anomalous dispersion
regime is to use very short segments of ber or tapered bers [39]. An alternative
solution is to reduce the spontaneous instabilities that may emerge in the early
stages of propagation. By imposing an input deterministic seed located in the
5
2
(a) (b) 9 (c) ideal PS
experiment
4

1
6
3
Power (W)
Time (ps)
0
2
-1 3
1
-2
0 0
0 200 400 600 -2 0 2 -10 0 10
Distance (m) Time (ps) Time (ps)
Fig. 4.9 Peregrine soliton in ber optics. a Longitudinal evolution of the Peregrine soliton in an
optical ber as predicted by (4.3). b Temporal intensity prole of the Peregrine soliton at the point
of maximum compression (black) compared with the input condition (gray). The ber parameters
are the same as those used in Fig. 4.1, and P0 0:5 W. c Experimental demonstration of the
Peregrine soliton. The temporal intensity prole recorded on an optical sampling oscilloscope
(circles) is compared with the analytical waveform. More details on the experiment can be found
in [95]
frequency band of the instabilities on the (coherent) pump signal, one can expect
that the seed will prevail over the spontaneous effects so that enhanced stability and
coherence can be ensured. Such an approach has been used in various numerical
and experimental works to control the stability and spectral extent of the generated
continuum [93, 94]. An accurate spectral study has conrmed that a seeded gen-
eration can be well described by the mathematical formalism introduced by Ak-
hmediev et al. [95]. This kind of control can be used to generate Peregrine solitons,
which represent a limiting case of Akhmediev breathers for maximum temporal
localization or, equivalently, of Kuznetsov breathers for maximum spatial locali-
zation, and have the following analytical expression:
" #
p 1 41 2iz=LNL
wz; t P0 eiz=LNL : 4:3
1 4t=s0 2 4z=LNL 2
Consequently, these pulses over a continuous background of power P0 have the

particularity to appear from nowhere and to disappear without leaving a trace, thus
constituting an excellent prototype of oceanic rogue waves (Fig. 4.9). Peregrine
solitons have been rst predicted theoretically in 1983 in hydrodynamics [96], but
have never been demonstrated experimentally before 2010 [97], and using exper-
iments in optical bers. From an experimental point of view, a sinusoidal beating
created by the temporal overlap of two frequency-shifted CWs [97] or the direct
intensity modulation of a CW [95] can reshape into a Peregrine soliton [98].
Experimental results obtained by the frequency resolved optical gating technique
[97] or by an optical sampling oscilloscope [95] are well reproduced both
numerically and analytically (Fig. 4.9c). Longitudinal studies performed by a cut-
back technique conrm the strong spatial localization, though some replicas are
manifested for large propagation distances. These replicas can be theoretically
interpreted in terms of higher-order breather solutions through the Darboux trans-
formations, and linked to the process of higher-order MIs [99]. The Peregrine
soliton is only the rst-order solution of a wider class of rational solitons that have
been also found numerically in turbulent optical environments [84, 100] or in
hydrodynamics [101]. We would also like to note that among the few other ele-
mentary analytic solutions of the NLS equation, the rst soliton on a nite back-
ground solution to the NLS equation that was obtained [90] and is now known as
the Kuznetsov-Ma soliton [102, 103], is of intense current interest. Indeed, the
dynamics of the Kuznetsov-Ma soliton have been very recently conrmed by
quantitative experiments in optical bers [104].
4.5 Conclusions and Perspectives
In this chapter, we have provided an overview of several recent examples of the use
of nonlinear phenomena in optical bers for the shaping of optical pulses in the
temporal and spectral domains. The propagation of short pulses in optical bers is
connected with a large variety of interesting and practically important phenomena.
The unique dispersive and nonlinear properties of optical bers lead to various
scenarios of the pulse evolution which result in particular changes of the pulse
temporal shape, spectrum and phase prole. Because the nonlinear dynamics of
pulses propagating in bers with normal GVD are generally sensitive to the initial
pulse condition, it is possible to nonlinearly shape the propagating pulses through
control of the initial pulse temporal intensity and/or phase prole. In fact, this
approach can be flexibly applied to the generation of specialized temporal waveforms
for applications in optical signal processing and manipulation, or the generation of
highly coherent continua for optical telecommunications, or the generation of nar-
row-spectrum pulses for applications requiring high spectral resolution such as
nonlinear vibrational microscopy. Application of nonlinear pulse shaping at normal
dispersion has also provided signicant progress in the eld of high-power ber
ampliers and lasers. The temporal and spectral features of pulses propagating in
bers with anomalous GVD are typically governed by soliton dynamics. The unique
properties of optical solitons can be effectively used in the generation of ultrashort
pulses with very high repetition rates, which are widely used in various scientic and
technological areas, or the generation of frequency-tunable pulses, or the generation
of ultra-wide spectra with many disruptive applications and breakthroughs in elds
such as optical coherence tomography, metrology and spectroscopy, biophotonics,
and others. The quest for a better understanding and harnessing of the temporal and
spectral SC fluctuations stimulates the development of new experimental techniques
for a quantitative analysis [105]. Seeding the SC generation at anomalous dispersion
with a suitably chosen signal can signicantly improve the SC stability. Furthermore,
the same concept can be applied to stimulate the generation of Peregrine solitons,
which constitute a prototype of optical rogue waves. Additional combination with

linear preshaping can further widen the eld of exploration of these coherent
structures [106] as, for example, the recently demonstrated controlled collision of
optical breathers highlights [107].
From a purely scientic point of view, the results presented in this chapter are a
further illustration of how optical ber systems can provide useful experimental
testbeds for the study of a variety of nonlinear dynamical processes. Therefore, this
research area is interesting in its own right. Despite substantial research in this eld,
qualitatively new phenomena are still being discovered. With respect to application
requirements, integrated pulse shaping devices are desirable for applications such as
in optical communications where size and robustness are important. The develop-
ments of microstructured bers with extremely small effective core areas and
exhibiting enhanced nonlinear characteristics, and of bres using materials with
refractive indexes higher than that of the silica glass have enabled dramatic reduction
of the required ber lengths for nonlinear interactions compared to conventional
bers, thereby paving the way for integrated ber-based pulse shaping. With the
inclusion of higher-order propagation effects, nonlinear shaping can now be observed
at the integrated waveguide scale. The recent demonstration of soliton compression
in a silicon photonic crystal is an excellent example of this [108].
Acknowledgments We acknowledge important contributions of our colleagues into the original

papers discussed in this chapter: K. Hammani, B. Kibler, C.H. Hage, H. Rigneault, E. Andresen,
P. Harper, A.I. Latkin, B.G. Bale, F. Parmigiani, P. Petroupoulos, D.J. Richardson, J.M. Dudley,
G. Genty and N.N. Akhmediev. We would also like to acknowledge the nancial support of the
Leverhulme Trust (grant RPG-278), the Conseil Regional de Bourgogne (PARI Photcom) and the
Labex ACTION program (ANR-11-LABX-01-01). The experimental work has beneted from the
PICASSO Platform of the University of Burgundy.
References
1. Y.R. Shen, Principles of Nonlinear Optics (Wiley, New York, 1984)

2. A.M. Weiner, Femtosecond pulse shaping using spatial light modulators. Rev. Sci. Instrum.
71, 19291960 (2000)
3. G.P. Agrawal, Nonlinear Fiber Optics, 4th edn. (Academic Press, San Diego, 2006)
4. M.E. Fermann, V.I. Kruglov, B.C. Thomsen, J.M. Dudley, J.D. Harvey, Self-similar
propagation and amplication of parabolic pulses in optical bers. Phys. Rev. Lett. 84, 6010
6013 (2000)
5. H. Hasegawa, F. Tappert, Transmission of stationary nonlinear optical pulses in dispersive
dielectric bers I. Anomalous dispersion. Appl. Phys. Lett. 23, 142144 (1973)
6. L.F. Mollenauer, G.P. Gordon, Solitons in Optical Fibers: Fundamentals and Applications
(Academic Press, San Diego, CA, 2006)
7. D. Grischowsky, A.C. Balant, Optical pulse compression based on enhanced frequency
chirping. Appl. Phys. Lett. 41, 12 (1982)
8. W.J. Tomlinson, R.H. Stolen, A.M. Johnson, Optical wave-breaking of pulses in nonlinear
optical bers. Opt. Lett. 10, 457459 (1985)
9. D. Anderson, M. Desaix, M. Lisak, M.L. Quiroga-Teixeiro, Wave-breaking in nonlinear
optical bers. J. Opt. Soc. Am. B 9, 13581361 (1992)
10. C. Finot, B. Kibler, L. Provost, S. Wabnitz, Benecial impact of wave-breaking for coherent
continuum formation in normally dispersive nonlinear bers. J. Opt. Soc. Am. B 25, 1938
1948 (2008)
11. S. Boscolo, C. Finot, Nonlinear pulse shaping in bres for pulse generation and optical
processing. Int. J. Opt. 2012, 159057 (2012). doi:10.1155/2012/159057
12. D. Anderson, M. Desaix, M. Karlsson, M. Lisak, M.L. Quiroga-Teixeiro, Wave-breaking-
free pulses in nonlinear optical bers. J. Opt. Soc. Am. B 10, 11851190 (1993)
13. J.M. Dudley, C. Finot, D.J. Richardson, G. Millot, Self similarity in ultrafast nonlinear
optics. Nature Phys. 3, 597603 (2007)
14. S. Boscolo, S.K. Turitsyn, V.Y. Novokshenov, J.H.B. Nijhof, Self-similar parabolic optical
solitary waves. Theor. Math. Phys. 133, 16471656 (2002)
15. V.I. Kruglov, J.D. Harvey, Asymptotically exact parabolic solutions of the generalized
nonlinear Schrdinger equation with varying parameters. J. Opt. Soc. Am. B 23, 25412550
(2006)
16. J.P. Limpert, T. Schreiber, T. Clausnitzer, K. Zllner, H.-J. Fuchs, E.-B. Kley, H. Zellmer, A.
Tnnermann, High-power femtosecond Yb-doped ber amplier. Opt. Express 10, 628638
(2002)
17. C. Billet, J.M. Dudley, N. Joly, J.C. Knight, Intermediate asymptotic evolution and photonic
bandgap ber compression of optical similaritons around 1550 nm. Opt. Express 13, 3236
3241 (2005)
18. D.N. Papadopoulos, Y. Zaouter, M. Hanna, F. Druon, E. Mottay, E. Cormier, P. Georges,
Generation of 63 fs 4.1 MW peak power pulses from a parabolic ber amplier operated
beyond the gain bandwidth limit. Opt. Lett. 32, 25202522 (2007)
19. C. Finot, G. Millot, S. Pitois, C. Billet, J.M. Dudley, Numerical and experimental study of
parabolic pulses generated via Raman amplication in standard optical bers. IEEE J. Sel.
20. K. Hammani, C. Finot, S. Pitois, J. Fatome, G. Millot, Real-time measurement of long
parabolic optical similaritons. Electron. Lett. 44, 12391240 (2008)
21. F.. Ilday, J.R. Buckley, W.G. Clark, F.W Wise, Self-similar evolution of parabolic pulses
in a laser. Phys. Rev. Lett. 92, 213902(4) (2004)
22. W.H. Renninger, A. Chong, F.W. Wise, Self-similar pulse evolution in an all-normal-
dispersion laser. Phys. Rev. A 82, 021805(R) (2010)
23. B.G. Bale, S. Wabnitz, Strong spectral ltering for a mode-locked similariton ber laser. Opt.
Lett. 35, 24662468 (2010)
24. B. Oktem, C. lgdr, F.O. Ilday, Soliton-similariton bre laser. Nature Photon. 4, 307311
(2010)
25. C. Aguergaray, D. Mchin, V.I. Kruglov, J.D. Harvey, Experimental realization of a mode-
locked parabolic Raman ber oscillator. Opt. Express 18, 86808687 (2010)
26. T. Hirooka, M. Nakazawa, Parabolic pulse generation by use of a dispersion-decreasing ber
with normal group velocity dispersion. Opt. Lett. 29, 498500 (2004)
27. A. Plocky, A.A. Sysoliatin, A.I. Latkin, V.F. Khopin, P. Harper, J. Harrison, S.K. Turitsyn,
Experiments on the generation of parabolic pulses in waveguides with length varying normal
chromatic dispersion. JETP Lett. 85, 319322 (2007)
28. C. Finot, B. Barviau, G. Millot, A. Guryanov, A.A. Sysoliatin, S. Wabnitz, Parabolic pulse
generation with active or passive dispersion decreasing optical bers. Opt. Express 15,
1582415835 (2007)
29. C. Finot, L. Provost, P. Petropoulos, D.J. Richardson, Parabolic pulse generation through
passive nonlinear pulse reshaping in a normally dispersive two segment ber device. Opt.
Express 15, 852864 (2007)
30. S. Boscolo, A.I. Latkin, S.K. Turitsyn, Passive nonlinear pulse shaping in normally
dispersive ber systems. IEEE J. Quantum Electron. 44, 11961203 (2008)
31. G.I. Barenblatt, Scaling, Self-Similarity, and Intermediate Asymptotics (Cambridge
University Press, Cambridge, 1996)
32. Y. Kodama, S. Wabnitz, Analytical theory of guiding-center nonreturn-to-zero and return-to-

zero signal transmission in normally dispersive nonlinear optical bers. Opt. Lett. 20, 2291
2293 (1995)
33. B.G. Bale, S. Boscolo, K. Hammani, C. Finot, Effects of fourth-order ber dispersion on
ultrashort parabolic optical pulses in the normal dispersion regime. J. Opt. Soc. Am. B 28,
20592065 (2011)
34. H. Wang, A.I. Latkin, S. Boscolo, P. Harper, S.K. Turitsyn, Generation of triangular-shaped
optical pulses in normally dispersive bre. J. Opt. (2010). doi:10.1088/2040-8978/12/3/
035205
35. N. Verscheure, C. Finot, Pulse doubling and wavelength conversion through triangular
nonlinear pulse reshaping. Electron. Lett. 47, 11941196 (2011)
36. S. Boscolo, S.K. Turitsyn, Intermediate asymptotics in nonlinear optical systems. Phys. Rev.
A 85, 043811(5) (2012)
37. R.R. Alfano, The Supercontinuum Laser Source (Springer, New York, 2006)
38. J.M. Dudley, J.R. Taylor, Supercontinuum Generation in Optical Fibers (Cambridge
University Press, New York, 2010)
39. M. Nakazawa, K. Tamura, H. Kubota, E. Yoshida, Coherence degradation in the process of
supercontinuum generation in an optical ber. Opt. Fiber Technol. 4, 215223 (1998)
40. Y. Ozeki, Y. Takushima, K. Aiso, K. Taira, K. Kikuchi, Generation of 10 GHz similariton
pulse trains from 1,2 km-long erbium-doped bre amplier for application to multi-
wavelength pulse sources. Electron. Lett. 40, 11031104 (2004)
41. F. Parmigiani, C. Finot, K. Mukasa, M. Ibsen, M.A.F. Roelens, P. Petropoulos, D.
J. Richardson, Ultra-flat SPM-broadened spectra in a highly nonlinear ber using parabolic
pulses formed in a ber Bragg grating. Opt. Express 14, 76177622 (2006)
42. A.M. Clarke, D.G. Williams, M.A.F. Roelens, B.J. Eggleton, Recongurable optical pulse
generator employing a Fourier-domain programmable optical processor. J. Lightwave
Technol. 28, 97103 (2010)
43. Y. Xin, D.J. Richardson, P. Petropoulos, Broadband, flat frequency comb generated using
pulse shaping-assisted nonlinear spectral broadening. IEEE Photon. Technol. Lett. 25, 543
545 (2013)
44. L. Boivin, B.C. Collings, Spectrum slicing of coherent sources in optical communications.
Opt. Fiber Technol. 7, 120 (2001)
45. Z. Yusoff, P. Petropoulos, K. Furusawa, T.M. Monro, D.J. Richardson, A 36-channel10-
GHz spectrally sliced pulse source based on supercontinuum generation in normally
dispersive highly nonlinear holey ber. IEEE Photon. Technol. Lett. 15, 16891691 (2003)
46. W.J. Tomlinson, R.H. Stolen, C.V. Shank, Compression of optical pulses chirped by self-
phase modulation in bers. J. Opt. Soc. Am. B 1, 139149 (1984)
47. S. Demmler, J. Rothhardt, A.M. Heidt, A. Hartung, E.G. Rohwer, H. Bartelt, J. Limpert, A.
Tnnermann, Generation of high quality, 1.3 cycle pulses by active phase control of an
octave spanning supercontinuum. Opt. Express 19, 2015120158 (2011)
48. L.I. Hooper, P.J. Mosley, A.C. Muir, W.J. Wadsworth, J.C. Knight, Coherent
supercontinuum generation in photonic crystal ber with all-normal group velocity
dispersion. Opt. Express 19, 49024907 (2011)
49. S.A. Planas, N.L. Pires Mansur, C.H. Brito Cruz, H.L. Fragnito, Spectral narrowing in the
propagation of chirped pulses in single-mode bers. Opt. Lett. 18, 699701 (1993)
50. E.R. Andresen, J. Thogersen, S.R. Keiding, Spectral compression of femtosecond pulses in
photonic crystal bers. Opt. Lett. 30, 20252027 (2005)
51. J.P. Limpert, T. Gabler, A. Liem, H. Zellmer, A. Tnnermann, SPM-induced spectral
compression of picosecond pulses in a single-mode Yb-doped ber amplier. Appl. Phys. B:
Lasers Opt. 74, 191195 (2002)
52. J.P. Limpert, N. Deguil-Robin, I. Manek-Hnninger, F. Salin, T. Schreiber, A. Liem, F.
Rser, H. Zellmer, A. Tnnermann, A. Courjaud, C. Hnninger, E. Mottay, High-power
picosecond ber amplier based on nonlinear spectral compression. Opt. Lett. 30, 714716
(2005)
53. B. Washburn, J. Buck, S. Ralph, Transform-limited spectral compression due to self-phase

modulation in bers. Opt. Lett. 25, 445447 (2000)
54. C. Finot, A. Guenot, P. Dupriez, Spectral compression of optical parabolic similaritons. Ann.
Phys. France 32, 7174 (2007)
55. J. Fatome, B. Kibler, E.R. Andresen, H. Rigneault, C. Finot, All-ber spectral compression
of picosecond pulses at telecommunication wavelength enhanced by amplitude shaping.
Appl. Opt. 51, 45474553 (2012)
56. E.R. Andresen, J.M. Dudley, C. Finot, D. Oron, H. Rigneault, Transform-limited spectral
compression by self-phase modulation of amplitude shaped pulses with negative chirp. Opt.
Lett. 36, 707709 (2011)
57. S. Boscolo, S.K. Turitsyn, C. Finot, Amplier similariton ber laser with nonlinear spectral
compression. Opt. Lett. 37, 45314533 (2012)
58. S.V. Chernikov, P.V. Mamyshev, Femtosecond soliton propagation in bers with slowly
decreasing dispersion. J. Opt. Soc. Am. B 8, 16331641 (1991)
59. S.V. Chernikov, D.J. Richardson, E.M. Dianov, D.N. Payne, Picosecond soliton pulse
compressor based on dispersion decreasing bre. Electron. Lett. 28, 18421844 (1992)
60. M.L. Tse, P. Horak, J.H.V. Price, F. Poletti, F. He, D.J. Richardson, Pulse compression at
1.06 m in dispersion-decreasing holey bers. Opt. Lett. 31, 35043506 (2006)
61. V.A. Bogatyrev, M.M. Bubnov, E.M. Dianov, A.S. Kurkov, P.V. Mamyshev, A.M.
Prokhorov, S.D. Rumyantsev, V.A. Semenov, S.L. Semenov, A.A. Sysoliatin, S.V.
Chernikov, A.N. Guryanov, G.G. Devyatykh, S.I. Miroshnichenko, A single-mode ber
with chromatic dispersion varying along the length. J. Lightwave Technol. 9, 561565
(1991)
62. S.V. Chernikov, J.R. Taylor, R. Kashyap, Experimental demonstration of step-like dispersion
proling in optical bre for soliton pulse generation and compression. Electron. Lett. 30,
433435 (1994)
63. S.V. Chernikov, J.R. Taylor, R. Kashyap, Comblike dispersion-proled ber for soliton
pulse train generation. Opt. Lett. 19, 539541 (1994)
64. L.F. Mollenauer, R.H. Stolen, J.P. Gordon, Experimental observation of picosecond pulse
narrowing and solitons in optical bers. Phys. Rev. Lett. 45, 10951098 (1980)
65. B. Kibler, R. Fischer, R.A. Lacourt, E. Courvoisier, R. Ferriere, L. Larger, D.N. Neshev, J.M.
Dudley, Optimised one-step compression of femtosecond bre laser soliton pulses around
1550 nm to below 30 fs in highly nonlinear bre. Electron. Lett. 43, 915916 (2007)
66. A.A. Amorim, M.V. Tognetti, P. Oliveira, J.L. Silva, L.M. Bernardo, F.X. Krtner, H.M.
Crespo, Sub-two cycle pulses by soliton self-compression in highly nonlinear photonic
crystal bers. Opt. Lett. 34, 38513853 (2009)
67. A. Hasegawa, Generation of a train of soliton pulses by induced modulational instability in
68. K. Tai, A. Hasegawa, A. Tomita, Observation of modulation instability in optical bers.
Phys. Rev. Lett. 56, 135138 (1986)
69. S. Pitois, J. Fatome, G. Millot, Generation of a 160 GHz transform-limited pedestal-free
pulse train through multiwave mixing compression of a dual-frequency beat signal. Opt. Lett.
27, 17291731 (2002)
70. J. Fatome, S. Pitois, G. Millot, 20-GHz-to-1-THz repetition rate pulse sources based on
multiple four-wave mixing in optical bers. IEEE J. Quantum Electron. 42, 10381046
(2006)
71. J. Fatome, S. Pitois, C. Fortier, B. Kibler, C. Finot, G. Millot, C. Courde, M. Lintz, E. Samain,
Multiple four-wave mixing in optical bers: 1.53.4-THz femtosecond pulse sources and real-
time monitoring of a 20-GHz picosecond source. Opt. Commun. 283, 24252429 (2010)
72. I. El Mansouri, J. Fatome, C. Finot, M. Lintz, S. Pitois, All-bered high-quality stable 20-
and 40-GHz picosecond pulse generators for 160-Gb/s OTDM applications. IEEE Photon.
Technol. Lett. 23, 14871489 (2011)
73. C. Finot, J. Fatome, S. Pitois, G. Millot, All-bered high-quality low duty-cycle 20-GHz and
40-GHz picosecond pulse sources. IEEE Photon. Technol. Lett. 19, 17111713 (2007)
74. F.M. Mitschke, L.F. Mollenauer, Discovery of the soliton self-frequency shift. Opt. Lett. 11,
659661 (1986)
75. J.P. Gordon, Theory of the soliton self-frequency shift. Opt. Lett. 11, 662664 (1986)
76. E.R. Andresen, V. Birkedal, J. Thogersen, S.R. Keiding, Tunable light source for coherent
anti-Stockes Raman scattering microspectroscopy based on the soliton self-frequency shift.
Opt. Lett. 31, 13281330 (2006)
77. T. Johnson, S. Diddams, Mid-Infrared Upconversion Spectroscopy Based on a Yb: Fiber
Femtosecond Laser. Appl. Phys. B 107, 3139 (2012)
78. D.V. Skryabin, F. Luan, J.C. Knight, J.S. Russell, Soliton self-frequency shift cancellation in
photonic crystal bers. Science 201, 17051708 (2003)
79. P.V. Mamyshev, P.G.J. Wigley, J. Wilson, G.I. Stegeman, V.A. Semenov, E.M. Dianov, S.I.
Miroshnichenko, Adiabatic compression of Schrdinger solitons due to the combined
perturbations of higher-order dispersion and delayed nonlinear response. Phys. Rev. Lett. 71,
7376 (1993)
80. R. Pant, A.C. Judge, E.C. Magi, B.T. Kuhlmey, M. de Sterke, B.J. Eggleton,
Characterization and optimization of photonic crystal bers for enhanced soliton self-
frequency shift. J. Opt. Soc. Am. B 27, 18941901 (2010)
81. J.M. Dudley, G. Genty, S. Coen, Supercontinuum generation in photonic crystal ber. Rev.
Modern Phys. 78, 11351184 (2006)
82. K.L. Corwin, N.R. Newbury, J.M. Dudley, S. Coen, S.A. Diddams, K. Weber, R.S.
Windeler, Fundamental noise limitations to supercontinuum generation in microstructure
bers. Phys. Rev.Lett. 90, 113904(4) (2003)
83. D.R. Solli, C. Ropers, P. Koonath, B. Jalali, Optical rogue waves. Nature 450, 10541057
(2007)
84. B. Kibler, K. Hammani, C. Michel, C. Finot, A. Picozzi, Rogue waves, rational solitons and
wave turbulence theory. Phys. Lett. A 375, 31493155 (2011)
85. J.M. Dudley, C. Finot, G. Millot, J. Garnier, G. Genty, D. Agafontsev, F. Dias, Extreme
events in optics: challenges of the MANUREVA project. Eur. Phys. J. Special Topics 185,
125133 (2010)
86. C. Kharif, E. Pelinovsky, A. Slunyaev, Rogue Waves in the Ocean (Springer, Berlin, 2009)
87. M. Taki, A. Mussot, A. Kudlinski, E. Louvergneaux, M.I. Kolobov, M. Douay, Third-order
dispersion for generating optical rogue solitons. Phys. Lett. A 374, 691695 (2010)
88. G. Genty, C.M. de Sterke, O. Bang, F. Dias, N.N. Akhmediev, J.M. Dudley, Collisions and
turbulence in optical rogue wave formation. Phys. Lett. A 374, 989996 (2010)
89. K. Hammani, B. Kibler, C. Finot, A. Picozzi, Emergence of rogue waves from optical
turbulence. Phys. Lett. A 374, 35853589 (2010)
90. E.A. Kuznetsov, Solitons in a parametrically unstable plasma. Doklady Akamemii Nauk
SSSR 22, 507508 (1977)
91. N.N. Akhmediev, V.I. Korneev, Modulation instability and periodic-solutions of the
nonlinear Schrdinger equation. Theor. Math. Phys. 69, 10891093 (1986)
92. J.M. Dudley, G. Genty, F. Dias, B. Kibler, N.N. Akhmediev, Modulation instability,
Akhmediev breathers and continuous wave supercontinuum generation. Opt. Express 17,
2149721508 (2009)
93. G. Genty, J.M. Dudley, Route to coherent supercontinuum generation in the long pulse
regime. IEEE J. Quantum Electron. 45, 13311335 (2009)
94. D.R. Solli, C. Ropers, B. Jalali, Active control of rogue waves for stimulated supercontinuum
generation. Phys. Rev. Lett. 101, 233902(4) (2008)
95. K. Hammani, B. Kibler, C. Finot, P. Morin, J. Fatome, J.M. Dudley, G. Millot, Peregrine
soliton generation and breakup in standard telecommunications ber. Opt. Lett. 36, 112114
(2011)
96. D.H. Peregrine, Water waves, nonlinear Schrdinger equations and their solutions. J. Austral.
Math. Soc. Ser. B 25, 1643 (1983)
97. B. Kibler, J. Fatome, C. Finot, G. Millot, F. Dias, G. Genty, N. Akhmediev, J.M. Dudley,
The Peregrine soliton in nonlinear bre optics. Nature Phys. 6, 790795 (2010)
98. M. Erkintalo, G. Genty, B. Wetzel, J.M. Dudley, Akhmediev breather evolution in optical
ber for realistic initial conditions. Phys. Lett. A 375, 20292034 (2011)
99. M. Erkintalo, K. Hammani, B. Kibler, C. Finot, N. Akhmediev, J.M. Dudley, G. Genty,
Higher-order modulation instability in nonlinear ber optics. Phys. Rev. Lett. 107, 253901(4)
(2011)
100. N. Akhmediev, J.M. Soto-Crespo, A. Ankiewicz, Extreme waves that appear from nowhere:
on the nature of rogue waves. Phys. Lett. A 373, 21372145 (2009)
101. A. Chabchoub, N. Hoffmann, M. Onorato, N. Akhmediev, Super rogue waves: Observation
of a higher-order breather in water waves. Phys. Rev. X 2, 011015(6) (2012)
102. T. Kawata, H. Inoue, Inverse scattering method for the nonlinear evolution equations under
nonvanishing conditions. J. Phys. Soc. Japan 44, 17221729 (1978)
103. Y.C. Ma, The perturbed plane-wave solutions of the cubic Schrdinger equation. Stud. Appl.
Math. 60, 4358 (1979)
104. B. Kibler, J. Fatome, C. Finot, G. Millot, G. Genty, B. Wetzel, N. Akhmediev, F. Dias, J.M.
Dudley, Observation of Kuznetsov-Ma soliton dynamics in optical ber. Sci. Rep. 2, 463
(2012). doi:10.1038/srep00463
105. B. Wetzel, A. Stefani, L. Larger, P. Lacourt, J. Merolla, T. Sylvestre, A. Kudlinski, A.
Mussot, G. Genty, F. Dias, Real-time full bandwidth measurement of spectral noise in
supercontinuum generation. Sci. Rep. 2, 883 (2012). doi:10.1038/srep00882
106. B. Frisquet, A. Chabchoub, J. Fatome, C. Finot, B. Kibler, G. Millot, Two-stage linear-
nonlinear shaping of an optical frequency comb as rogue nonlinear-Schrdinger-equation-
solution generator. Phys. Rev. A 89, 023821 (2014)
107. B. Frisquet, B. Kibler, G. Millot, Collision of Akhmediev breathers in nonlinear ber optics.
Phys. Rev. X 3, 041032(9) (2013)
108. A. Blanco-Redondo, C. Husko, D. Eades, Y. Zhang, J. Li, T.F. Krauss, B.J. Eggleton,
Observation of soliton compression in silicon photonic crystals. Nature Commun. 5, 3160
(2014). doi:10.1038/ncomms4160
Chapter 5
Optical Regeneration
Francesca Parmigiani, Radan Slavk, Joseph Kakande,

Periklis Petropoulos and David Richardson
Abstract In this chapter we review the need, general principles and approaches
used to regenerate mainly phase encoded signals of differing levels of coding
complexity. We will describe the key underpinning technology and present the
current state-of-the-art, incorporating an appropriate historic perspective
throughout.
5.1 Introduction to Optical Regeneration
Data trafc on the worlds core optical networks is increasing at approximately

3040 % per annum driven largely by the emergence of new mobile and video-
centric applications, cloud computing and the ever increasing number of internet/
mobile users [1]. This rate of trafc growth is presenting huge challenges to net-
work hardware vendors and operators alike, particularly since in the laboratory the
ultimate capacity of conventional single mode ber technology is already being
approached as a result of the latest advances in the use of digital coherent trans-
mission and the use of advanced modulation format signals [2]. Use of the phase
(and often additionally the intensity) of an optical carrier rather than simple On-Off
Keying (OOK) is already a central feature of core network deployments today,
since as well as allowing for more spectrally efcient communications, it offers
advantages in terms of improved resilience to certain linear and nonlinear
transmission impairments [3]. However, the introduction of phase encoding means
that attention needs to be paid to nonlinear phase noise, which manifests itself as a
new dominant limitation to ultimate system performance [46]. In amplied
F. Parmigiani (&) R. Slavk P. Petropoulos D. Richardson

Optoelectronics Research Centre, University of Southampton,
Southampton, Hampshire SO17 1BJ, UK
e-mail: frp@orc.soton.ac.uk
J. Kakande
Bell Laboratories, Alcatel-Lucent, Holmdel, NJ 07733, USA

130 F. Parmigiani et al.
transmission systems, nonlinear phase noise originates from the combined effects of
noise introduced by the in-line optical ampliers and the nonlinear interaction
between the various co-propagating wavelength division multiplexing (WDM)
channels. For this reason it is benecial to periodically regenerate signals during
transmission over long distances in order to increase capacity and/or reach. In
currently installed transmission systems this is achieved by employing optical-to-
electronic-to-optical (O/E/O) conversion and fast electronic processors. However,
all-optical techniques that can perform regeneration by avoiding the need for O/E/O
conversion merit attention, since they are often associated with additional attractive
features, such as transparency to the symbol rate and a more favorable scaling of the
energy consumption. The ability to regenerate signals also ensures maximum
flexibility in network design and architecture, for example: allowing for a signi-
cant reduction in penalties due to accumulated cross-talk and lter concatenation in
all optical networks based on components such as Recongurable Optical Add
Drop Multiplexers (ROADMs), more flexible network design since performance is
less route dependent, and potentially providing greater tolerance to changes in
network load (adding and dropping channels).
As a consequence much research effort has been devoted over the years to the
implementation of all-optical solutions for signal regeneration using either speci-
ality optical bers or semiconductor optical amplier (SOA)-based devices as the
nonlinear optical processing medium. In the past the primary focus was on
regenerators for the then prevalent OOK modulation format signals, however in
more recent times attention has turned onto the regeneration of more advanced
modulation formats that incorporate phase encoding. In this chapter we rst briefly
review the early development of regenerators for OOK signals before turning our
attention to the even thornier problem of regenerating mainly phase-based modu-
lation format (M-ary phase shift keying, MPSK) signals with some initial efforts
towards amplitude and quadrature modulation (QAM) signals.
5.2 Optical Regenerators for Simple Amplitude Encoded

Signals
In accordance with the traditional optical transmission approach and technology, all
early types of optical regenerator were designed to operate on OOK signals. Such
regenerators were rst proposed in the 90s and were intensively studied throughout
the next decade or so [714].
In the most general form of regenerator, three basic signal-processing functions
are required: (i) Reamplication, (ii) Reshaping and (iii) Retiming. A regenerator
simultaneously providing each of these functions is referred to as a 3R-regenerator.
Applying this nomenclature, optical ampliers solely provide 1R-regeneration
whereas the extended class of devices that simultaneously provide reamplication
and reshaping are referred to as 2R-regenerators. A sketch of the generic layout
of a typical 3R regenerator is shown in Fig. 5.1, and ordinarily comprises:
5 Optical Regeneration 131
Fig. 5.1 A schematic of the generic layout of a typical 3R regenerator with a degraded OOK
signal present at the input. The input/output adaptation interfaces address issues such as input/
output power adjustment, initial (linear) pulse shaping, out-of-band noise ltering, format
conversion, control of the signal polarisation states, signal de-multiplexing/multiplexing (be it in
the time and/or frequency domain). The clock recovery unit retrieves an accurate clock from the
incoming data stream and provides appropriate synchronisation between the recovered clock and
the incident data within the nonlinear (optical) gate which provides for the 3R regeneration itself.
Adapted from [14, 15]
(i) input/output adaptation interfaces in order to convert the input data signal into a
form suitable for use within the regenerator, and/or to convert the internally
regenerated signal into the correct format for onward transmission, (ii) some form
of nonlinear gate, and (iii) a clock recovery/generation device [14, 15]. The non-
linear optical gate, which ultimately provides the reshaping and retiming function,
is generally the most critical element of any regenerator. Over the years many
nonlinear gates have been investigated. The greatest majority of work has used an
optical ber as the nonlinear medium [712] and has relied upon exploitation of
ultrafast nonlinear optical effects such as Self Phase Modulation (SPM), Cross
Phase Modulation (XPM), Four Wave Mixing (FWM) and Stimulated Raman
Scattering (SRS) to provide the required optical switching mechanism. However, a
wide variety of other nonlinear materials/devices have also been explored including
SOAs [1517], various saturable absorbers (SAs) [18], periodically poled lithium
niobate (PPLN) waveguides [19], silicon chips [20] and chalcogenide waveguides
[21], amongst others.
The majority of ber based gates are adaptations of devices based on either: (i)
spectral broadening and offset-ltering, the so-called Mamyshev regenerator [10,
12, 16, 21], whose schematic and working principle are reported in Fig. 5.2a; or (ii)
optical interferometers with a nonlinear medium in one of the arms, where one or
two signals (the noisy data and the clean clock signal) might be present at its input,
as shown in Fig. 5.2c [7, 8, 14, 15, 17]. While the nonlinear media and the
nonlinear processes exploited might differ, all amplitude regeneration schemes aim
to achieve a step-like power transfer function, dened in terms of the instantaneous
output versus input signal power. The slope of the power transfer function at the
logical 0 and 1 bit power levels denes the regenerative properties of the device
(see Fig. 5.2b, d).
Fig. 5.2 Examples of two forms of amplitude regenerators: ((a) the Mamyshev 2R-regenerator
and (c) a 3R-interferometric regenerator) and corresponding power transfer functions
(b and d) together with experimental eye diagrams before and after the regenerators (Adapted
from [12, 14])
Some OOK regenerators can still be used in the regeneration of modern

MPSK-encoded signals. To enable this, the adaptation stage converts an incoming
optical MPSK signal to one or more OOK signals, amplitude regeneration is then
performed on this signal before converting the signal back to an MPSK format at
the output adaptation stage [22]. Such regenerators are referred to as format con-
version based regenerators. Alternatively, those classes of amplitude regenerator
that preserve the phase of the optical signal, can be deployed on the basis that they
suppress any amplitude fluctuations on a MPSK signal without distorting the phase
information [2330]. Such regenerators are referred to as phase preserving ampli-
tude regenerators.
The extension to simultaneous multi-channel regeneration is a key issue in
todays high capacity systems. Such an extension is very challenging in practice as
this is generally limited by the occurrence of unwanted nonlinear interactions
amongst WDM channels, the so-called inter-channel crosstalk, which directly
competes with the nonlinear effects used for the regeneration process itself. In
optical bers, such cross-talk typically results from XPM and FWM interactions,
which manifest themselves as strong spectral distortions and/or spectrally depen-
dent power transfer functions due to the incomplete interaction of co-propagating
pulses during nonlinear propagation (i.e. the pulses involved have not fully walked
through each other during propagation).
Several techniques have been applied to simultaneously regenerate more than

one WDM channel. The main approaches are shown in Fig. 5.3.
The rst is based on wavelength-to-space demultiplexing where each wave-
length channel is split out into a separate path using a wavelength demultiplexer and
identical single channel regenerators are employed on a per-channel basis, see
Fig. 5.3a [14, 18, 30]. After regeneration, the channels are then recombined together
using a second wavelength multiplexer for further transmission. Such a per-channel
scheme clearly avoids nonlinear cross-talk-induced distortion; however, the com-
plexity (and cost) of the scheme increases linearly with the number of channels to
be simultaneously regenerated, which is far from desirable.
The second approach is based on time-interleaving of the input signals [3133].
Here the various WDM channels are aligned in time to avoid bit overlap before
entering the regenerator via some sort of WDM to optical time division multiplex
(OTDM) conversion process, see Fig. 5.3b. The individual WDM channel data
symbols are then sequentially processed in a single nonlinear device, guaranteeing a
reduction in component cost as the number of channels increases. However, this
approach requires both channel synchronisation and a low duty cycle (low spectral
efciency) signal format since the maximum time-slot that a channel can occupy is
now determined by the number of WDM channels simultaneously processed by the
regenerator.
The third approach is based on carefully engineered dispersive channel walk-
through maps in the regenerator to avoid inter-channel cross-talk [32, 3440]. In
the co-propagating geometry, a high local dispersion value ensures that nonlinear
inter-channel effects are reduced due to the fast walk-off time amongst the various
WDM channels (thereby reducing any bit-pattern dependence of the nonlinear
cross-talk), while a low overall average dispersion value (obtained via dispersion
Fig. 5.3 Schematic of multi-channel 2R regenerators based on a wavelength-to-space demulti-

plexing, b time interleaving and dispersive walk-through in c co- or d counter-propagating
geometries, exploiting SPM. (DE)MUX (de)multiplexer; PBS polarization beam splitter; (PM-)
HNLF (polarization maintaining-) highly nonlinear ber. Gray squares represent carrier-offset
lters
compensation) ensures that the nonlinear effect used for the regeneration process
itself (SPM for example) remains sufciently strong, Fig. 5.3c. The regenerator,
thus, consists of multiple modules incorporating appropriate combinations of highly
nonlinear bers (HNLFs) with high normal dispersion and associated periodic
group delay devices (PGDDs) or dispersion compensating bers to generate/emu-
late saw-tooth-like channelized dispersion proles. This approach enables in-line
processing of a (potentially quite large) number of WDM channels [32, 3437]
simultaneously in a single nonlinear device, guaranteeing a reduction in cost-per-
channel as the number of channels increases. However, while the previous two
approaches can potentially use any kind of nonlinear device and nonlinearity, this
third approach can only be implemented using SPM in bers and is thus only
applicable to amplitude-only signals, as SPM heavily distorts the signal phase.
In the counter-propagating geometry, Fig. 5.3d, an extremely high walk-off is
automatically guaranteed for counter propagating channels in the nonlinear gate. A
further two-fold channel scaling can be achieved using a polarization multiplexing
conguration which exploits the rapid interchannel pulse walk-off due to the large
birefringence-induced differential group delay achievable within the polarisation
maintaining (PM)-HNLF. A further three-fold reduction in XPM is also obtained
due to the tensorial nature of the Kerr effect between orthogonally-polarized signals
[3840]. So far it has been shown that up to four channels can be processed
simultaneously in a single nonlinear device [38, 39].
Having given a quick review of the status of the work on the regeneration of
OOK signals, largely for contextual setting, we turn our attention to more recent
work concerned with the regeneration of more modern and spectrally-efcient
MPSK and QAM signals.
5.3 Regeneration of Phase-Only Encoded Signals
As we briefly mentioned in Sect. 5.2, phase regeneration can be divided into two
broad categories: direct regeneration, which operates on the premise that a signal has
already been signicantly degraded in phase, and therefore signal quality can only be
improved by suppressing these phase perturbations; and indirect regeneration, which
acts to suppress amplitude fluctuations which could otherwise seed phase noise after
subsequent transmission (phase preserving amplitude regeneration) [2330]. How-
ever, to make it worthwhile, the latter process must not create signicant added
phase noise. Direct regeneration can be classied either as format conversion based
regeneration from MPSK to OOK and vice versa (without the requirement to be
phase-preserving) [22], or phase sensitive amplication (PSA) based regeneration,
which has been widely recognized as an effective way to regenerate phase-encoded
signals thanks to its phase squeezing capabilities [41, 42]. In the rest of the chapter
PSA-based phase regenerators only will be considered.
Phase sensitive ampliers exhibit gain characteristics that depend on the phase of
the input optical signal relative to a local optical reference; they operate by amplifying
one of the input signals quadratures whilst simultaneously de-amplifying the other
[43]. While the theory of PSAs is not recent, historically attention focused mainly on
exploiting second order nonlinearity in bulk crystals [44, 45], it is only relatively
recently that much experimental effort has been placed on this topic in the context of
optical ber communications. This growth in interest has largely been driven by
technological advances that have allowed a number of previously considered prac-
tical limitations to be overcome, as will be discussed below.
The implementation of PSAs using third order nonlinearity in optical bers
rather than second order nonlinearity in crystals was a rst step towards practical
network applications due to the increased robustness, improved power efciency
and ease of system integration that optical bers offer. Marhic et al. successfully
demonstrated the rst degenerate PSA by 1991 [43], but progress in the eld was
quite limited up until the demonstration of amplication with a 1.8 dB noise gure
by Imajuku et al. in 1999 [46]. Since then, PSAs have been demonstrated operating
over wide bandwidths in excess of 20 nm [4749] as well as with a record low
noise gure of 0.8 dB [50].
Phase sensitive amplication in bers can be accomplished either using inter-
ferometric schemes [41, 42], or schemes based on FWM [51]. The former approach
is advantageous in terms of the feasibility of phase locking (signal and pump
coincide) and broad bandwidth operation, however it is accompanied by severe
disadvantages such as strict single-wavelength operation, guided acoustic wave
Brillouin scattering (GAWBS) and low energy efciency compared to FWM based
counterparts, since the gain scales quadratically with pump power as compared to
exponentially in the latter case. For these reasons, non-interferometric PSA
implementations have been the preferred choice in most recent demonstrations of
phase regeneration.
A FWM-based PSA can be implemented using single- or dual- pump congura-
tions with signal, idler and pump(s) being along the same linear polarization (scalar
schemes) or different polarization axes (vector schemes)all possibilities cover
both non-degenerate and degenerate (signal and idler at the same wavelength) con-
gurations. In the literature, FWM-based PSAs that have been successfully used to
suppress noise on a data signal can be divided into two main categories. In the rst
one, the same data is carried on two signals at different wavelengthsthe rst signal
is the original data signal and the second its complex-conjugated copy (idler). The
PSA amplies equally both signal quadratures of the perfect phase-matched, power-
equalized signal-idler pair, which consequently experiences 6 dB higher gain than the
corresponding noise which is uncorrelated (and which consequently experiences
phase insensitive amplication, PIA, only) in the high gain regimea feature that is
inherent to a PSA [5254], see Fig. 5.4. This approach does not require any
knowledge of the modulation format used since the absolute input signal phase is
tracked by that of the conjugated idler, and, thus, we call it a Modulation Format
Transparent PSA. We intentionally do not use the term regenerator, in line with the
existing literature. The modulation format transparent PSA inherently requires a
Fig. 5.4 Basic principle of the modulation format transparent PSA. Adapted from [54]
non-degenerate scheme, as the signal and idler need to propagate through the network
at different wavelengths, accumulating uncorrelated noise, implying occupation of
twice the bandwidth as compared to a degenerate conguration, together with the
requirement of perfect phase synchronization amongst all the waves prior to the PSA.
Practically, the signal-idler pairs are usually generated in a parametric PIA before
the transmission link, which is generally referred to as a copier given its function
[53, 54]. Note that both copier and PSA should operate in a linear gain regime to avoid
gain-saturation-induced distortion. Finally it should be added that modulation format
transparent PSA schemes can also easily be extended to multi-channel amplication
for the same pump conguration.
The second approach requires only the original incident data signal to propagate
through the transmission ber link, guaranteeing optimum bandwidth occupation,
which is the standard approach used in todays telecommunications. To regenerate
such a signal, the modulation format must be known and, thus, we call such
schemes Modulation Format Specic PSA Regenerators. In such a conguration,
the idler involved in the PSA process is then generated locally (following a possible
transmission link) and it carries identical or an exact multiple of the original signal
phase information. Some examples of possible FWM congurations are shown in
Fig. 5.5a, d, respectively. As such, one signal quadrature will be amplied, whereas
the orthogonal one is de-amplied, differing in the operation principle from the
modulation format transparent PSA. In the simplest case of binary phase shift
keying (BPSK) modulation format regeneration, the dual-pump degenerate scalar
conguration is the most commonly used, see Fig. 5.5a. Here, the in-phase com-
ponent of the electric eld experiences gain, g, while the quadrature component
experiences de-amplication by 1/g in the linear regime [53]. Subsequently, the
output phase of the amplied signal is more closely aligned to the ampliers in-
phase axis (see Fig. 5.5b, c). This phase quantization effect, sometimes referred to
as phase squeezing, is inherently suitable for the regeneration of binary phase-
encoded signals, as the phase squeezed data bits are forced to adopt a phase of
either 0 or , thus restoring the delity of the signal prior to transmission [5559].
Note that squeezing in this context refers to a classical effect, rather than one of a
quantum nature. Figure 5.5b, c also show that a side-effect of the PSA action is the
introduction of unwanted amplitude variations added to the signal; the larger the
phase error the PSA corrects for, the larger the amplitude variation. However, this
Fig. 5.5 Sketches of two modulation format specic PSA congurations (a and d), corresponding
constellation diagrams (b and e) and transfer functions of the output amplitude/phase versus input
phase (c and f) derived from (5.1), where m1 = 1/(M1) and M = 2 (top line) and M = 4 (bottom
line), respectively
can be reduced by operating the PSA in saturation to further suppress any amplitude
noise present on the original phase-encoded signal, enabling simultaneous phase
and amplitude regeneration [5759].
As such schemes provide inherently single channel operation for the same pump
conguration there have been attempts to adapt this scheme to perform multi-
wavelength regeneration by employing more pumps [60, 61] (their number scales
linearly with the number of channels); however, a penalty is likely to be paid due to
strong and undesired pump-to-pump interactions [60, 61].
The Modulation Format Specic PSA regenerator scheme can be extended to
allow the phase regeneration of an MPSK signal with an arbitrary number M of
phase coding levels. This requires the realization of a staircase phase response that
includes the corresponding number of steps. This can be implemented through a
coherent addition of the (M 1)-th phase harmonic of the signal with the signal
itself, which mathematically can be expressed as [6264]:
Aeiht / ei/t m1 eiM1/t ; 5:1
where A and (t) are the output signal electric eld amplitude and phase, respec-
tively, (t) is the input phase and m1 is the phase harmonic weight. It is worth
emphasizing that also in this conguration, the (M 1)-th phase harmonic (idler) is
generated inside the regenerator, i.e. after transmission, so its phase noise is
strongly correlated to that of the original signal. Figure 5.5e, f show the corre-
sponding constellation diagrams and transfer functions derived from (5.1) for M = 4
levels of phase modulation and m1 = 1/(M 1). The periodicity of the phase transfer
function matches M, but the abruptness of the steps becomes less well dened, and
the depth of the phase-to-amplitude conversion is also seen to reduce. A sharper
step-like phase function could be achieved if more phase harmonics of the signal
were coherently added to the original signal.
Umeki et al. have compared optical-signal-to-noise-ratio (OSNR) and electrical
signal to noise ratio (SNR) at the input/output of the modulation format transparent
PSA and modulation specic PSA regenerator [65]. They showed experimentally
that, while it is possible to obtain 3 dB OSNR improvement with both schemes,
only the modulation transparent PSA allows a 3 dB SNR improvement. However, it
is worth emphasizing that in their study both PSAs were operating in the linear
(non-saturated) regime.
In general, the implementation of any PSA raises its own practical challenges and
in order to achieve phase-sensitive amplication, the phase relationship between the
PSA pump(s), signal and any idlers present needs to be maintained [47]. The chal-
lenges are mainly as follows: (i) phase-encoded signals usually have a suppressed
carrier eld (i.e. they are carrier-less) and (ii) even if the carrier were extractable, it
would generally still contain a component of the phase and amplitude noise of the
transmitted data. Proof-of-principle PSA demonstrations have overcome this prob-
lem by ensuring that both the signal and the pump(s) originate from a common laser
source, e.g. through the generation of an optical frequency comb [56, 57]. However,
in a real transmission link, any pumps would have to be generated locally in the
regenerating node to be truly independent of the data signal. An additional challenge,
common in many parametric amplication congurations, is that strong narrow-
linewidth continuous wave (CW) laser signals need to be used as PSA pumps. When
the nonlinear element used for the implementation of the PSA is an optical ber, a
strong limitation on the amount of pump power that can be launched into the ber is
imposed by the onset of stimulated Brillouin scattering (SBS).
In the remaining part of this section we review work to date on PSA regener-
ation, mainly in chronological order, and through the description of the experi-
mental demonstrations used, we will show how the above challenges can be
overcome to realize a phase regenerator for use in real-world transmission links.
5.3.1 Modulation Format Specic PSA Regenerators
The rst successful experimental demonstration of a degenerate dual pump scalar

PSA regenerator operating both on phase and amplitude of a BPSK signal was
reported by Croussore and Li [57]. In this instance the phase-locking between
pumps and signal waves was guaranteed by generating the pumps from the same
CW-laser used to produce the data signal (through comb lines generated by 40 GHz
amplitude modulation). SBS was avoided by using a bismuth oxide HNLF offering
an enhanced Kerr nonlinearity to SBS gure of merit. Noise was added by
additional amplitude and phase modulation at a chosen single frequency to the data
signal. For amplitude regeneration, saturation of the FWM process was exploited.
A major step toward the realization of a practical phase sensitive regenerator was
achieved in [58, 59, 66] where the issues of phase locking the various waves,
coping with real-world impairments (wideband noise) and the use of narrow-line-
width CW pumps were addressed.
The schematic of the corresponding experimental conguration of the degen-
erate PSA-based regenerator of BPSK signals is shown in Fig. 5.6 (details of the
experiment can be found in [66]). The incoming data signal is combined with
locally generated sources in order to synthesize a set of suitably phase-locked
waves that can be used to drive the PSA, allowing black-box operation. In order
to achieve this, the data signal was rst combined with a free-running narrow
linewidth CW source (noted as Pump 1 in the gure) in a conventional parametric
amplier, producing the spectral trace shown in the inset of Fig. 5.6. This spectral
trace shows that the generated idler was narrowband, i.e. any modulation existing in
the data signal was stripped from it. This is a consequence of the phase relation
between the idler idler and the BPSK data signal , i.e. idler = 2 pump1, through
which any phase jumps existing in the BPSK data are erased in the idler. An
additional consequence of this relation is that the phases of the group of three waves
(data, pump1 and idler) were now locked relative to each other, thus allowing for
phase-sensitive interaction in a subsequent parametric amplier. Furthermore any
noise present in the data that was transferred through to the idler as part of this
parametric process is largely removed via optical injection locking of a semicon-
ductor laser to the noisy idler [67]. Indeed, injection locking inherently allows
regeneration of signal amplitude [26], provides narrowband (few GHz) rejection of
Fig. 5.6 Experimental set-up of a degenerate PSA-based regenerator for BPSK signals (DE)MUX
De/multiplexer; EDFA Erbium-doped ber amplier; PZT piezo-electric ber stretcher. Insets
Spectral traces obtained at the output of HNLF1 showing the generation of the narrowband (BPSK
data-free) idler and at the output of HNLF2 for maximum gain (red, dashed) and de-amplication
(black, dotted). Adapted from [59, 66]
any high-frequency noise present on the idler, whilst effectively amplifying the
power of the injected light. The resulting phase-locked pumps and signal then
provide the inputs for the degenerate dual-pump scalar PSA.
The nonlinear medium was a 180-m sample of a fully-spliced, high SBS-
threshold HNLF (HNLF 2 in Fig. 5.6) with an Al2O3 doped core (instead of
germanium, which is the normal dopant of choice in silica-based HNLFs) and a
linear strain gradient applied along its length (ranging from 400 to 20 g) to broaden
the SBS gain bandwidth and thereby to reduce the peak gain (corresponding SBS
threshold was about 29 dBm) [68]. Furthermore, to ensure regeneration of the
signal amplitude as well as the phase, the PSA was operated in deep saturation [58,
59, 66].
The robustness of the system was tested rstly by emulating broadband phase
noise [59]. Some results are presented in Fig. 5.7a, where constellation diagrams at
40 and 56 Gbit/s as well as eye diagrams (at 40 Gbit/s only) of the original signal
(with and without phase noise added) and the signal at the output of the regenerator
are shown, illustrating the bit rate transparency of the scheme. Secondly, the
regenerator was installed in the middle of an 800-km section of the UKs Aurora
dark ber link [66], and the transmission performance of a 40 Gbit/s DPSK signal
(coupled together with another 36 WDM channels operating at the same data rate)
was tested in both the presence and absence of the PSA-based regenerator
(Fig. 5.7b).
Following this work, a polarization diversity scheme based on such a PSA for
polarization insensitive phase regeneration of single and dual polarization based
phase modulation formats was proposed in [69], where the HNLF was used in a
bi-directional conguration to process the two signal polarizations independently.
Fig. 5.7 a Constellation diagrams of the back-to-back, of the initial signals with added phase
noise (labelled as not regenerated) and of regenerated BPSK signals for two data rates (40 and 56
Gbit/s) and the corresponding demodulated eyes at 40 Gbit/s ([59]). b BER curves (and eye
diagrams) of the DPSK signal after transmission over 400 and 800 km of ber, with and without
PSA-based regeneration. Adapted from [59, 66]
A reduction in polarization sensitivity for BPSK and quadrature phase shift keyed
(QPSK) signal regeneration was shown through simulations, however to date only
preliminary experimental results have been presented [69].
The rst characterization of the degenerate vector PSA in terms of gain prole as
a function of input signal phase was presented in [70], demonstrating lower gain as
compared to the scalar case for the same total launched pump power, but a much
more compact bandwidth occupation since the vector scheme is much less affected
by higher-order FWM of the pump and signals.
From (5.1), it can be noted that to achieve a binary step-like phase response (i.e.
for BPSK signals), the strength of the idler must be comparable to that of the signal
allowing for a large phase sensitive extinction ratio, PSER, which is dened as the
difference between the maximum phase sensitive (PS) gain and the maximum PS
de-amplication. Large symmetric (i.e. maximum PS gain equal to maximum PS
de-amplication) PSERs are conventionally achieved only at high pump powers
resulting in the generation of other FWM components across a wide spectral
bandwidth. However, such sideband components, undesired in many instances, can
play an important role in inducing larger PS de-amplication (asymmetric PSER) at
limited pump powers. In fact, a large asymmetric PSER of 25 dB at a nonlinear
phase shift (NPS) of 0.8 radians in the scalar degenerate dual-pump PSA provided
efcient phase squeezing of DPSK signals [71], and QPSK phase squeezing
(de-multiplexing) along the in-phase and quadrature components [72].
The rst extension of the PSA regenerator scheme to allow for regeneration of,
in principle, any MPSK signal was proposed and experimentally demonstrated by
Kakande et al. in [6264], based on a modication of the PSA set-up presented
previously for BPSK signals. In the rst part of the black-box MPSK regenerator
phase multiplication and pump recovery (by injection locking the M-th harmonic,
which is modulation stripped for an MPSK input) occurs and in the second part of
the system the conjugated idler (bearing the temporal phase modulation M 1/)
is coherently added to the signal, thereby regenerating the phase, as discussed
previously. Figure 5.8 reports some of the corresponding results for QPSK signals
at 20 and 112 Gbit/s [64]. The phase quantization was accompanied by a sinusoidal
amplitude response, which is undesirable for regeneration. However, this phase-to-
amplitude conversion can be eliminated: (i) by operating the PSA in saturation as
was described above in the case of BPSK signals (even though it is to be appre-
ciated the higher the number of phase harmonics involved the more challenging the
task), (ii) subsequently regenerating the signal amplitude in a high dynamic range
limiting optical amplier, such as an injection-locked semiconductor laser [26], or
(iii) involving a larger number of phase harmonics in the PSA process so that the
phase-to-amplitude conversion can be avoided in the rst place as proposed in [73]
and subsequently experimentally demonstrated in [74]. This can be achieved by
mixing the (M 1)-th and the (M + 1)-th phase harmonics together with the
original signal, altering (5.1) as follows [73]:
Fig. 5.8 Constellation diagrams at the input/output of the regenerator for a QPSK signal a at 10
Gbaud with only phase noise added, b at 10 Gbaud with phase and amplitude noise added, c at 56
Gbaud without noise, d at 56 Gbaud with phase noise only. 2 is the phase error variance, 2rho
is the normalised amplitude noise variance. Adapted from [64]
Aeiht / ei/t meiM1/t eiM1/t ; 5:2
where m is the phase harmonic weight.

Following this initial ber-based work different nonlinear materials/processes
have been investigated using the above congurations. For example, cascaded
quadratic nonlinear processes in periodically poled lithium niobate (PPLN) wave-
guides (which offer a number of potential advantages, such as broadband operation,
low-latency, no intrinsic frequency chirp and SBS resilience) were successfully
exploited to achieve phase regenerative amplication with a wide phase sensitive
dynamic range of 20 dB for 40 Gb/s DPSK signals, as well as to implement the in-
line PSA as a repeater amplier in a 160 km ber link [75]. Albuquerque et al. also
investigated the feasibility of using a single PPLN waveguide, used in a bidirec-
tional conguration, for both the generation of phase correlated signals and phase-
sensitive regeneration and injection-locking for carrier phase recovery for BPSK
signals [76]. Cascaded PPLN waveguides were also used to achieve in line PSA for
40 Gbit/s QPSK signals [77] and some of the corresponding results are reported in
Fig. 5.9, where the constellation diagrams at the input (a) and output (b) of the PSA
when only phase noise was articially added to the signal, top row, and at the
output of the PSA when amplitude noise was added and different values of signal
power at the input of the PPLN were chosen (3.7 and 11.8 dBm, respectively,
Fig. 5.9 Top row Constellation diagrams at the input (a) and output (b) of the PSA when phase
noise was added to the signal. Bottom row Constellation diagrams at the output of the PSA when
amplitude noise was added to the signal and different signal powers were chosen, 3.7 dBm (a) and
11.8 dBm (b), respectively at the input of the PPLN. Reproduced with permission from [77]
(Copyright 2013 IEICE)
bottom row). It can be seen that as the input power into the nonlinear media
increases the amplitude noise becomes smaller highlighting the saturation regime.
Semiconductor optical ampliers were also investigated numerically as possible
nonlinear devices, conrming a suppression of phase variation in the dual-pump
degenerate four wave mixing conguration [78]. An experimental implementation
for two-channel regeneration using an SOA was reported in [61].
An alternative approach towards MPSK regeneration was proposed in [79]. It is
based on a dual-pump degenerate PSA structure which employs bi-directional
signal/conjugate generation in the rst instance to generate two pumps, whose
phases are conjugated as compared to the transmitted signal ((M/2 1)), fol-
lowed by a dual-pump FWM process for phase regeneration of potentially any
MPSK signal [79]. As compared to the solution proposed by Kakande et al. [62],
this PSA requires lower-order signal conjugates (conjugates [(M/2 1)] of the
input MPSK signal [] instead of [(M 1)] and [M], respectively), implying
that it may have high applicability to higher level PSK. Simulation results showed
the effective reduction of phase deviation for QPSK and 8PSK formats up to 50
Gbaud. However, it is worth pointing out that several practical issues need to be
resolved before this scheme can be experimentally demonstrated. Firstly, any
amplitude noise present in the original signal would transfer to the generated pumps
and would compromise the system performance, implying that an amplitude
regenerator has to precede such a scheme. Secondly, the challenge of phase-locking
of the pumps to the carrier-less signal will need to be addressed prior regeneration.
Finally, complex MPSK signals can always be thought of as the combination of
simpler modulation formats and as such their regeneration simplied by the parallel
regeneration of several simpler modulation format signals. For example, QPSK
signals can be regenerated by arranging two BPSK phase-sensitive ampliers in
parallel to separately regenerate the in-phase and quadrature components of the input
signal. These regenerated streams can then be combined in parallel in a coupler,
thereby re-synthesizing a cleaner QPSK signal, as proposed and simulated by Zheng
et al. [80]. Note that for the correct implementation of this scheme it is essential that
each individual regenerator also suppresses signal amplitude noise. Similarly, Webb
et al. have proposed and numerically demonstrated the simultaneous conversion of
the two orthogonal phase components of a signal to separate output frequencies, using
SOAs, thus converting a QPSK signal to two BPSK outputs [81, 82], which could
then be regenerated afterwards in simpler BPSK regenerators.
5.3.2 Modulation Format Transparent PSA
As discussed previously, the Modulation Format Transparent PSA scheme requires

the signal and the idler (generated in the copier) at different wavelengths to be
propagated along the transmission link together with the pump. After passing the
waves through a lossy medium (such as a long length of optical ber) to guarantee
uncorrelated noise at the signal and idler wavelengths, the cascaded copier-PSA
pair using HNLFs can provide low-noise phase-sensitive amplication [83]. The
rst experimental demonstration of amplication of three channels (100 GHz
spacing), 10 Gbd DQPSK signals with improved sensitivity compared to an EDFA-
based system was reported in [54]. Similar results were also demonstrated when the
emulated link loss was replaced by a > 80 km ber span carrying up to 10 Gbd
DQPSK data in the linear propagation regime [84] and such an improvement may
be maintained even when operating in the nonlinear transmission regime [85]. Note
that in real systems, low-noise pump recovery (for example based on injection
locking) is necessary to avoid noise-performance degradation, as well as polari-
sation trimming, and phase and dispersion/delay control to carefully align the rel-
ative temporal positions of the pump, signal and idler beams.
Fig. 5.10 ad Constellation diagrams at the input/output of the PSA for the four channels.
e Corresponding spectral trace at the output of the PSA. Reproduced with permission from [88] (
2012, Optical Society of America)
The potential of such a conguration in PPLN devices was highlighted in [86]

and subsequently, this concept was also employed in [87] for the rst time to
demonstrate the squeezing of the phase noise of a 10 GBit/s DPSK signal. Inter-
estingly, this demonstration also considered the generation of the phase-locked
signal-idler pair in a second PPLN waveguide, further miniaturizing the system.
The same authors then also demonstrated multi-channel operation of the scheme
with up to four signals, demonstrating negligible impact on the phase squeezing
characteristics of the adjacent channels when the PPLN was operated in the linear
regime [88]. Some of the corresponding experimental constellation diagrams and
spectral traces at the output of the PSA are shown in Fig. 5.10. Similar schemes
were also implemented by Umeki et al. to experimentally demonstrate for the very
rst time PSA regeneration of a 10 Gbaud multi-carrier polarisation multiplex of
QAM signal as will be discussed in the following section [89].
The same conguration was also used to demonstrate PSA in a 196 m slow-
light dispersion-engineered, silicon photonic crystal providing a PSER of 11 dB
[90], and in a 65 mm-long dispersion engineered chalcogenide chip obtaining a
PSER of 11.6 dB [91] using pulsed pumps.
5.4 Regeneration of Amplitude and Phase Encoded Signals
Recently, there have been a few reports in the literature on how to regenerate
amplitude and phase encoded signals, generally using regenerators of phase-only
encoded signals as the basic building blocks.
Following the same approach as in Sect. 5.3, we divide the existing PSA-based
schemes between those that are modulation format specic (requiring only the
signal to be propagated through a communication channel), and those that are
modulation format transparent (requiring signal and its complex conjugated copy to
be simultaneously propagated).
5.4.1 Modulation Format Specic PSA Regenerators
Three main approaches have been investigated so far for the regeneration of
amplitude and phase encoded signals. The rst one is based on using phase
regenerators to remove only the phase noise on the amplitude/phase encoded signal
[73, 92, 93]. In the second approach, the amplitude/phase modulated signals are
rst demultiplexed into simpler phase-only modulated tributaries, e.g., 16QAM is
demultiplexed into two QPSK signals [94], before parallel regeneration of each
individual tributary and subsequent coherent recombination, similarly to [8082]
for QPSK input signals. The last approach takes advantages of the periodic-like
transfer function of the amplitude in a nonlinear amplifying loop mirror (NALM) in
conjunction with a modulation format specic PSA to simultaneously allow
amplitude and phase regeneration [95].
In [73] the regenerative performances of the modied MPSK phase regenerator
scheme proposed in [62] to avoid phase-to-amplitude noise conversion were
numerically studied for three different modulation formats when the regenerator
was placed in the middle of a transmission span (overall length of about 1200 km).
The modied regenerator was found to offer a reduction of 50 % in bit error rate
(BER) for a 8-PSK signal as compared to the regenerator proposed in [62] and a
20 % reduction for a star 8-QAM signal (as for this modulation format the main
source of bit errors comes from the amplitude noise that remains uncorrected), see
Fig. 5.11 [73].
In [92], a two level star 8QAM signal with only phase noise added was pro-
cessed by the QPSK regenerator described in Sect. 5.3.1 [62]. Due to the presence
of two amplitude levels, the regenerator had to operate close-to-the-linear regime,
limiting the performance of the regenerator. Some experimental results are shown in
Fig. 5.12, where different signal-to-idler ratios at the input of the PSA were
considered.
In [93], various modulation formats were investigated theoretically in a recir-
culating loop considering an MPSK phase regenerator working in the linear regime
[62]. The results for various 4-bit (16-level) modulation formats are shown in
Fig. 5.13. As expected, the main conclusion is that single-amplitude-level phase
encoded signals outperform other modulation formats of identical bits/symbol
density, when a MPSK phase regenerator is used, due to the phase-to-amplitude
noise conversion associated with this scheme and the uncorrected accumulated
amplitude noise. This emphasizes the fact that phase regenerators work the best
with phase-only encoded signals.
A more complex approach for 16QAM regeneration has been suggested recently
in [94], where the 16QAM signal is demultiplexed into two QPSK data streams as
shown in Fig. 5.14a, which can then be individually regenerated in simpler
regenerator schemes and recombined in order to form a high delity 16QAM
signal. The scheme is based on the 4-level quantizer proposed in [64]. The required
enhanced quality of the quantisation process was guaranteed by cascading two
PSAs and the choice of the relative phase between the signal and the pumps in the
Fig. 5.11 BER curves and constellation diagrams at the end of a transmission link when no
regenerator was present (black circles and (i)), when the regenerator proposed in [62] (blue circles
and (ii)) and the modied regenerator proposed in [73] (red circles and (iii)) were placed in the
middle of the transmission link, respectively, for 8-PSK (a) and star 8QAM (b) signals [73].
Figure courtesy of G. Hesketh
Fig. 5.12 Constellations before and after phase quantization without (top row) and with additional
phase noise (PN) (bottom row). The results after the quantization are achieved at different power
ratios which relate the average power of the harmonic to the average power of the signal and
directly influence the optimization parameter m. The ratios are stated in logarithmic values [92].
Figure courtesy of T. Richter
Fig. 5.13 Phase symmetric constellations diagrams for 16 symbols before (leftmost) and after
transmission in linear channels (index 0) and nonlinear regenerative channels cascading 20 phase
regenerators (index 0 and 20 respectively) for the xed value of signal-to-noise ratio of 18 dB [93].
Figure courtesy of M. Sorokina
second PSA stage. As can be seen in Fig. 5.14a, in the rst step, one QPSK
tributary is extracted by squeezing the four points in each of the four quadrants of
the constellation into single points. In the second step, the other tributary is obtained
by coherent subtraction of the original 16QAM signal and the rst tributary. The
corresponding BER curves are shown in Fig. 5.14b, where it can be seen that the
second QPSK tributary suffers from signicant noise. It is worth mentioning that
Fig. 5.14 a Conceptual scheme of 16QAM optical demultiplexing into two QPSK tributaries
(b) and simulated BER curves of the 16QAM signal and its demultiplexed QPSK tributaries [94].
Figure courtesy of A. Bogris
Fig. 5.15 A schematic diagram of the NALM with PSA (a). OBPF- optical bandpass lter, (DE)
MUX(de)multiplexer, HOI- high-order idler. b Amplitude (left) and phase (right) transfer
characteristics of the NALM with PSA for a typical set of parameters. The signal states of a star-
8QAM signal are marked [95]. Figure courtesy of T. Rthlingshfer/ MPL
the scheme can, in principle, be extended to any format that can be divided up into
QPSK tributaries (e.g., 64QAM and 256QAM) [94]. Practical considerations for
this scheme include preservation of the coherence among the different paths, and
Bogris et al. believe that only photonic integrated solutions could offer the required
stability [96].
The last approach proposed and only simulated so far combines the character-
istics of a nonlinear amplifying loop mirror (NALM), already demonstrated as a
phase preserving multi-level amplitude regenerator in [24], and a directional PSA
[62], Fig. 5.15a, to obtain a step-like transfer function in both the amplitude and
phase domains, Fig. 5.15b, allowing for simultaneous amplitude and phase
regeneration of two amplitude levels and four phase states (8-star QAM) [95]. The
scheme was also compared to the cascading of a PSA, acting as a phase regenerator
only, and a subsequent NALM, acting as a phase-preserving amplitude regenerator,
showing a slight difference in noise conversion for similar transfer characteristics.
Fig. 5.16 QAM regeneration in the presence of single-tone phase noise (a) and ASE noise (b).
Reproduced with permission from [89] ( 2014, Optical Society of America)
5.4.2 Modulation Format Transparent PSA Schemes
As discussed in Sect. 5.3.2, the operational principle of this approach is based on

the fact that noise experienced by the signal and its copy during the propagation is
uncorrelated, while the signal and copy are fully correlated. When dealing with
multiple-amplitude-level signals, amplitude linearity with a high dynamic range is
very important as is the generation of a high quality phase-conjugated idler. Noise
reduction for a single 40 Gbit/s 16QAM signal has been experimentally demon-
strated for the very rst time in [89], where noise was articially induced using
either a phase modulator adding a single tone or using amplied spontaneous
emission (ASE) noise to both the signal and idler, see Fig. 5.16. Also, simultaneous
amplication of up to three carries and of two orthogonally polarized signals (via a
polarisation diversity scheme, using four highly efcient PPLN waveguides) was
demonstrated in [89]. This work addresses most of the challenges of optical
regenerators; it operates multi-channel, it is in principle capable of processing any
arbitrary modulation format, it is demonstrated with polarization-multiplexed sig-
nals and it operates as a black-box. However, it is worth noting that it requires the
signal to propagate over two wavelength channels (signal and complex conjugated
copy) and these need to be perfectly phase synchronized (i.e., with zero delay in
propagation) before the PSA. It is still a hotly debated question as to this is an
acceptable scenario in modern telecommunication networks.
5.5 Conclusions
In this chapter we have briefly reviewed the need, general principles and techno-
logical approaches to the regeneration of various forms of optical signal including
both OOK signals and phase encoded signals of differing levels of complexity. We
have discussed the evolution of the eld and have presented the current state-of-the-
art both in terms of the demonstrations of novel device concepts that extend the
range of functionality that can be achieved (e.g. range of advanced modulation

format signals that can be addressed) and enhance practicality (e.g. in providing
black-box operation, WDM performance and possible routes to integration). From
what we have written it should be apparent that optical regeneration remains a
vibrant and potentially technologically important eld of research, with signicant
advances made in recent years and enabled in many instances by progress in the
underpinning technologies e.g. highly nonlinear bers and high performance nar-
row-linewidth lasers. It should however be equally clear that many challenges
remain which will need to be overcome if all-optical regeneration is ever to make it
from the laboratory to commercial deploymentnot least in terms of enabling
signicant levels of WDM capability and the development of compact, power
efcient variants. Much work remains to be done and one can anticipate consid-
erable innovation and interesting results ahead, with many other application
opportunities (e.g. within test and measurement systems) opening up along the way.
References
1. S. Namiki, T. Kurosu, K. Tanizawa, J. Kurumida, T. Hasama, H. Ishikawa, T. Nakatogawa,

M. Nakamura, K. Oyamada, Ultrahigh-denition video transmission and extremely green
optical networks for future. IEEE Sel. Top. Quantum Electron. 17(2), 446457 (2011)
2. R.S. Tucker, Green optical communicationspart II: energy limitations in networks. IEEE
Sel. Top. Quantum Electron. 17(2), 261274 (2011)
3. C. Xu, X. Liu, L.F. Mollenauer, X. Wei, Comparison of return-to-zero differential phase-shift
keying and on-off keying in long-haul dispersion managed transmission. IEEE Photonics
Technol. Lett. 15, 617619 (2003)
4. A. Demir, Nonlinear phase noise in optical-ber-communication systems. J. Lightwave
Technol. 25, 20022032 (2007)
5. J.P. Gordon, L.F. Mollenauer, Phase noise in photonic communication system using linear
ampliers. Opt. Lett. 15, 13511353 (1990)
6. H. Kim, A.H. Gnauck, Experimental investigation of the performance limitation of DPSK
systems due to nonlinear phase noise. IEEE Photonics Technol. Lett. 15, 320322 (2003)
7. M. Jinno, M. Abe, All optical regenerator based on nonlinear bre Sagnac interferometer.
Electron. Lett. 28(14), 1350 (1992)
8. J.K. Lucek, K. Smith, All optical signal regenerator. Opt. Lett. 18(15), 1226 (1993)
9. W.A. Pender, P.J. Watkinson, E.J. Greer, A.D. Ellis, 10 Gbit/s all optical regenerator.
Electron. Lett. 31(18), 1587 (1995)
10. G. Raybon, Y. Su, J. Leuthold, R.-J. Essiambre, T. Her, C. Joergensen, P. Steinvurzel, K.D.K.
Feder, 40 Gbit/s pseudo-linear transmission over one million kilometers, in OFC, 1722
March 2002, pp. FD10-1FD10-3
11. S. Watanabe, F. Futami, R. Okabe, Y. Takita, S. Feber, R. Ludwig, C. Schubert, C. Schmidt,
H.G. Weber, 160 Gbit/s optical 3R regenerator in a ber transmission experiment, in OFC
(2003), p. PD16
12. P.V. Mamyshev, All-optical data regeneration based on self-phase modulation effect, in ECOC
(1998), p. 475
13. M. Nakazawa, E. Yamade, H. Kubota, K. Suzuki, 10 Gbit/s soliton data transmission over one
million kilometers. Electron. Lett. 27, 12701272 (1991)
14. O. Leclerc, B. Lavigne, E. Balmefrezol, P. Brindel, L. Pierre, D. Rouvillain, F. Seguineau,
Optical regeneration at 40 Gb/s and beyond. J. Lightwave Technol. 21, 27792790 (2003)
15. S. Fischer, M. Dlk, E. Gamper, W. Vogt, E. Gini, H. Melchior, W. Hunziker, D. Nesset, A.D.
Ellis, Optical 3R regenerator for 40 Gbit/s networks. Electron. Lett. 35(23), 20472049 (1999)
16. J. Leuthold, G. Raybon, Y. Su, R. Essiambre, S. Cabot, J. Jaques, M. Kauer, 40 Gbit/s
transmission and cascaded all-optical wavelength conversion over 1000000 km. Electron. Lett.
38(16), 890892 (2002)
17. B. Lavigne, P. Guerber, P. Brindel, E. Balmefrezol, B. Dagens, Cascade of 100 optical 3R
regenerators at 40 Gbit/s Based on all-active Mach Zehnder interferometers, in ECOC, vol. 3,
paper We.F.2.6 (2001), p. 290
18. Q. T. Le, L. Bramerie, S. Lobo, M. Gay, M. Joindot, J.-C. Simon, A. Poudoulec, M. Van der
Keur, C. Devemy, D. Massoubre, J.-L. Oudar, G. Aubin, A. Shen, J. Decobert, WDM
compatible 2R regeneration device based on eight-channel saturable absorber module.
Electron. Lett. 43(23), 13051306 (2007)
19. A. Bogoni, X. Wu, S.R. Nuccio, A.E. Willner, 640 Gb/s all-optical regenerator based on a
periodically poled lithium niobate waveguide. J. Lightwave Technol. 30(12), 18291834 (2012)
20. R. Salem, M.A. Foster, A.C. Turner, D.F. Geraghty, M. Lipson, A.L. Gaeta, Signal
regeneration using low-power four-wave mixing on silicon chip. Nat. Photonics 2, 338 (2008)
21. V.G. Taeed, M. Shokooh-Saremi, L. Fu, I.C.M. Littler, D.J. Moss, M. Rochette, B.J. Eggleton,
R.Y. Ruan, B. Luther-Davies, Self-phase modulation-based integrated optical regeneration in
chalcogenide waveguides. IEEE Sel. Top. Quantum Electron. 12(3), 360370 (2006)
22. M. Matsumoto, Fiber-based all-optical signal regeneration. IEEE Sel. Top. Quantum Electron.
18(2), 738752 (2012)
23. K. Cvecek, G. Onishchukov, K. Sponsel, A.G. Striegler, B. Schmauss, G. Leuchs,
Experimental investigation of a modied NOLM for phase-encoded signal regeneration.
IEEE Photonics Technol. Lett. 18, 18011803 (2006)
24. M. Hierold, T. Roethlingshoefer, K. Sponsel, G. Onishchukov, Bernhard Schmauss, G. Leuchs,
Multilevel phase-preserving amplitude regeneration using a single nonlinear amplifying loop
mirror. IEEE Photonics Technol. Lett. 23(14), 10071009 (2011)
25. M. Matsumoto, A ber-based all-optical 3R regenerator for DPSK signals. IEEE Photonics
Technol. Lett. 19, 273275 (2007)
26. A. Fragkos, A. Bogris, D. Syvridis, R. Phelan, Amplitude noise limiting amplier for phase
encoded signals using injection locking in semiconductor lasers. J. Lightwave Technol. 30(5),
764771 (2012)
27. A. Fragkos, A. Bogris, D. Syvridis, R. Phelan, Colorless regenerative amplication of constant
envelope phase-modulated optical signals based on injection-locked fabry perot lasers. IEEE
Photonics Technol. Lett. 24(1), 2830 (2011)
28. L. Quang Trung, L. Bramerie, N. Hoang Trung, M. Gay, S. Lobo, M. Joindot, J.L. Oudar, J.C.
Simon, Saturable-absorber-based phase-preserving amplitude regeneration of RZ DPSK
signals. IEEE Photonics Technol. Lett. 22, 887889 (2010)
29. L. Quang Trung, L. Bramerie, M. Gay, M. Joindot, J. Simon, A. OHare, N. Hoang Trung,
J. Oudar, All-optical phase-preserving amplitude regeneration of 28-Gbaud RZ-DQPSK
signals with a microcavity saturable absorber in a recirculating loop experiment, in OFC/
NFOEC (2011), pp. 13
30. K. Cvecek, K. Sponsel, C. Stephan, G. Onishchukov, R. Ludwig, C. Schubert, B. Schmauss,
G. Leuchs, Phase-preserving amplitude regeneration for a WDM RZ-DPSK signal using a
nonlinear amplifying loop mirror. Opt. Express 16, 19231928 (2008)
31. N. Chi, L. Xu, K.S. Berg, T. Tokle, P. Jeppesen, All-optical wavelength conversion and
multichannel 2R regeneration based on highly nonlinear dispersion-imbalanced loop mirror.
IEEE Photonics Technol. Lett. 14(11), 15811583 (2002)
32. C. Kouloumentas, L. Provost, F. Parmigiani, S. Tsolakidis, P. Petropoulos, I. Tomkos, D.
J. Richardson, Four-channel all-bre dispersion managed 2R regenerator. IEEE Photonics
Technol. Lett. 20(13), 11691171 (2008)
33. N.S.M. Shah, M. Matsumoto, 2R Regeneration of time-interleaved multiwavelength signals
based on higher order four-wave mixing in a ber. IEEE Photonics Technol. Lett. 22(1), 2729
(2010)
34. P.G. Patki, M. Vasilyev, T.I. Lakoba, Multichannel all-optical regeneration, in Photonics
Society Summer Topical Meeting Series (2010), pp. 172173
35. M. Vasilyev, T.I. Lakoba, All-optical multichannel 2R regeneration in a bre-based device.
Opt. Lett. 30(12), 14581460 (2005)
36. C. Kouloumentas, P. Vorreau, L. Provost, P. Petropoulos, W. Freude, J. Leuthold, I. Tomkos,
All-berized dispersion-managed multichannel regeneration at 43 Gb/s. IEEE Photonics
Technol. Lett. 20(20), 18541856 (2008)
37. T. Ohara, H. Takara, A. Hirano, K. Mori, S. Kawanishi, 40-Gb/s x 4- channel all-optical
multichannel limiter utilizing spectrally ltered optical solitons. IEEE Photonics Technol. Lett.
15(5), 763765 (2003)
38. F. Parmigiani, L. Provost, P. Petropoulos, D.J. Richardson, W. Freude, J. Leuthold, A.D. Ellis,
I. Tomkos, Progress in multichannel all-optical regeneration based on ber technology. IEEE
Sel. Top. Quantum Electron. 18(2), 689699 (2012)
39. L. Provost, F. Parmigiani, P. Petropoulos, D.J. Richardson, K. Mukasa, J. Takahashi,
J. Hiroishi, M. Tadakuma, Investigation of four-wavelength regenerator using polarization-
and direction- multiplexing. IEEE Photonics Technol. Lett. 20, 16761678 (2008)
40. G. Zarris, E. Hugues-Salas, N. Amaya-Gonzalez, R. Weerasuriya, F. Parmigiani, D. Hillerkuss,
P. Vorreau, M. Spyropoulou, S.K. Ibrahim, A.D. Ellis, R. Morais, P. Monteiro, P. Petropoulos,
D.J. Richardson, I. Tomkos, J. Leuthold, D. Simeonidou, Field experiments with a grooming
switch for OTDM meshed networking. J. Lightwave Technol. 28(4), 316327 (2010)
41. D. Levandovsky, M. Vasilyev, P. Kumar, Amplitude squeezing of light by means of a phase-
sensitive ber parametric amplier. Opt. Lett. 24, 984986 (1999)
42. K. Croussore, G.F. Li, Phase and amplitude regeneration of differential phase-shift keyed signals
using phase-sensitive amplication. IEEE Sel. Top. Quantum Electron. 14(3), 648658 (2008)
43. M.E. Marhic, C.H. Hsia, J.M. Jeong, Optical amplication in a nonlinear bre interferometer.
Electron. Lett. 27, 210211 (1991)
44. R.E. Slusher, B. Yurke, Squeezed light for coherent communications. J. Lightwave Technol.
8, 466477 (1990)
45. J.A. Levenson, I. Abram, T. Rivera, P. Grangier, Reduction of quantum noise in optical
parametric amplication. J. Opt. Soc. Am. B 10, 22332238 (1993)
46. W. Imajuku, A. Takada, Y. Yamabayashi, Low-noise amplication under the 3 dB noise
gure in high-gain phase-sensitive bre amplier. Electron. Lett. 35, 19541955 (1999)
47. R. Tang, P. Devgan, V.S. Grigoryan, P. Kumar, Inline frequency non- degenerate phase-
sensitive bre parametric amplier for bre optic communication. Electron. Lett. 41, 10721074
(2005)
48. R. Tang, J. Lasri, P.S. Devgan, V. Grigoryan, P. Kumar, M. Vasilyev, Gain characteristics of a
frequency nondegenerate phase sensitive ber-optic parametric amplier with phase self-
stabilized input. Opt. Express 13, 1048310493 (2005)
49. J. Kakande, C. Lundstrm, P.A. Andrekson, Z. Tong, M. Karlsson, P. Petropoulos, F.
Parmigiani, D.J. Richardson, Detailed characterisation of a ber-optic parametric amplier in
phase-sensitive and phase-insensitive operation. Opt. Express 18(5), 41304137 (2010)
50. Z. Tong, C. Lundstrom, A. Bogris, M. Karlsson, P. Andrekson, D. Syvridis, Measurement of
sub-1 dB noise gure in a non-degenerate cascaded phase-sensitive bre parametric amplier,
in ECOC (2009), pp. 12
51. C. McKinstrie, S. Radic, Phase-sensitive amplication in a ber. Opt. Express 12(20), 4973
4979 (2004)
52. R. Tang, P. Devgan, P.L. Voss, V.S. Grigoryan, P. Kumar, In-line frequency nondegenerate
phase-sensitive ber-optical parametric amplier. IEEE Photonics Technol. Lett. 17, 18451847
(2005)
53. M. Vasilyev, Distributed phase-sensitive amplication. Opt. Express 13(19), 75637571
(2005)
54. Z. Tong, C. Lundstrm, P.A. Andrekson, C.J. McKinstri, M. Karlsson, D.J. Blessing, E.
Tipsuwannakul, B.J. Puttnam, H. Toda, L. Grner-Nielsen, Towards ultrasensitive optical
links enabled by low-noise phase-sensitive ampliers. Nat. Photonics 5, 430436 (2011)
55. A. Bogris, D. Syvridis, RZ-DPSK signal regeneration based on dual-pump phase-sensitive

amplication in bers. IEEE Photonics Technol. Lett. 18(20), 21442146, (2006) Oct. 15
56. K. Croussore, G. Li, Phase Regeneration of NRZ-DPSK Signals Based on Symmetric-Pump
Phase-Sensitive Amplication. IEEE Photonics Technol. Lett. 19(11), 864866 (2007)
57. K. Croussore, G. Li, BPSK phase and amplitude regeneration using a traveling-wave phase-
sensitive amplier, in IEEE/LEOS Winter Topical Meeting Series (2008), pp. 4546
58. R. Slavik, F. Parmigiani, J. Kakande, C. Lundstrom, M. Sjodin, P.A. Andrekson, R.
Weerasuriya, S. Sygletos, A.D. Ellis, L. Gruner- Nielsen, D. Jakobsen, S. Herstrom, R. Phelan,
J. OGorman, A. Bogris, D. Syvridis, S. Dasgupta, P. Petropoulos, D.J. Richardson, All-
optical phase and amplitude regenerator for next-generation telecommunications systems. Nat.
Photonics 4, 690695 (2010)
59. R. Slavik, A. Bogris, F. Parmigiani, J. Kakande, M. Westlund, M. Skold, L. Grner-Nielsen,
R. Phelan, D. Syvridis, P. Petropoulos, D.J. Richardson, Coherent all-optical phase and
amplitude regenerator of binary phase-encoded signals (Invited Paper). IEEE Sel.
60. S. Sygletos, P. Frascella, S.K. Ibrahim, L. Grner-Nielsen, R. Phelan, J. OGorman, A.D. Ellis,
A practical phase sensitive amplication scheme for two channel phase regeneration. Opt.
Express 19, B938B945 (2011)
61. S. Sygletos, M.J. Power, F.C. Garcia Gunning, R.P. Webb, R.J. Manning, A.D. Ellis,
Simultaneous dual channel phase regeneration in SOAs in ECOC, Tu.1.A.2 (2012)
62. J. Kakande, R. Slavik, F. Parmigiani, A. Bogris, D. Syvridis, L. Gruner-Nielsen et al.,
Multilevel quantization of optical phase in a novel coherent parametric mixer architecture.
Nat. Photonics 5, 748752 (2011)
63. J. Kakande, A. Bogris, R. Slavik, F. Parmigiani, D. Syvridis, P. Petropoulos, D.J. Richardson,
First demonstration of all-optical QPSK signal regeneration in a novel multi-format phase
sensitive amplier, in ECOC (2010), pp. 13
64. J. Kakande, A. Bogris, R. Slavik, F. Parmigiani, D. Syvridis, P. Petropoulos, D. Richardson,
M. Westlund, M. Skold, QPSK phase and amplitude regeneration at 56 Gbaud in a novel idler-
free nondegenerate phase sensitive amplier, in OFC/NFOEC, paper OMT4 (2011)
65. T. Umeki, H. Takara, Y. Miyamoto, M. Asobe, 3-dB signal-ASE beat noise reduction of
coherent multicarrier signal utilizing phase sensitive amplication. Opt. Express 20(22),
2472724734 (2012)
66. R. Slavik, A. Bogris, J. Kakande, F. Parmigiani, L. Gruner-Nielsen, R. Phelan, J. Vojtch,
P. Petropoulos, D. Syvridis, D.J. Richardson, Field-Trial of an all-optical PSK regenerator/
multicaster in a 40 Gbit/s, 38 channel DWDM transmission experiment. J. Lightwave Technol.
30, 512520 (2012)
67. R. Weerasuriya, S. Sygletos, S.K. Ibrahim, R. Phelan, J. OCarroll, B. Kelly, J. OGorman, A.
D. Ellis. Generation of frequency symmetric signals from a BPSK input for phase-sensitive
amplication, in OFC, Paper OWT6 (2010)
68. L. Gruner-Nielsen, S. Dasgupta, M.D. Mermelstein, D. Jakobsen, S. Herstrom, M.E.V. Pedersen,
E.L. Lim, S. Alam, F. Parmigiani, D. Richardson, B. Palsdottir, A silica based highly nonlinear
bre with improved threshold for stimulated brillouin scattering, in ECOC (2010), pp. 13
69. J.-Y. Yang, M. Ziyadi, Y. Akasaka, S. Khaleghi, M.R. Chitgarha, J. Touch, M. Sekiya,
Investigation of polarization-insensitive phase regeneration using polarization-diversity phase-
sensitive amplier, in ECOC (2013)
70. A. L. Riesgo, C. Lundstrm, M. Karlsson, P. A. Andrekson, Demonstration of degenerate
vector phase-sensitive amplication. in ECOC, We.3.A.3 (2013)
71. M. Gao, T. Kurosu, T. Inoue, S. Namiki, Efcient phase regeneration of DPSK signal by
sideband-assisted dual-pump phase-sensitive amplier. Electron. Lett. 49(2), 140141 (2013)
72. M. Gao, T. Kurosu, T. Inoue, S. Namiki, Efcient quadrature squeezing of QPSK signals by
sideband-assisted dual-pump phase sensitive amplier, in OFC, OW3C.1 (2013)
73. G. Hesketh, P. Horak, Reducing bit-error rate with optical phase regeneration in multilevel
modulation formats. Opt. Lett. 38, 53575360 (2013)
74. K.R.H. Bottrill, G. Hesketh, F. Parmigiani, P. Horak, D.J. Richardson, P. Petropoulos, An

optical phase quantiser exhibiting suppressed phase dependent gain variation, in OFC (2014)
75. T. Umeki, M. Asobe, H. Takenouchi, In-line phase sensitive amplier based on PPLN
waveguides. Opt. Express 21(10), 1207712084 (2013)
76. A.A.C. Albuquerque, B.J. Puttnam, M.V. Drummond, . Szab, D. Mazroa, S. Shinada, N.
Wada, R.N. Nogueira, Investigation of black-box phase regeneration using single
bi-directional PPLN waveguide, in OFC, OW4C.2 (2013)
77. M. Asobe, T. Umeki, H. Takenouchi, Y. Miyamoto, In-line phase-sensitive amplier for
QPSK signal using multiple QPM LiNbO3 waveguide, PD23, OECC/PS (2013)
78. K. Saito, H. Uenohara, Analytical investigation of possibility of a phase sensitive amplier
based on four wave mixing in a semiconductor optical amplier, in OECC (2012)
79. J.-Y. Yang, Y. Akasaka, M. Sekiya, Optical phase regeneration of multi-level PSK using dual-
conjugate-pump degenerate phase-sensitive amplication, in ECOC, P3.07, 2012
80. Z. Zheng, L. An, Z. Li, X. Zhao, X. Liu, All-optical regeneration of DQPSK/QPSK signals
based on phase-sensitive amplication. Opt. Commun. 281, 27552759 (2008)
81. R.P. Webb, J.M. Dailey, R.J. Manning, A.D. Ellis, Phase discrimination and simultaneous
frequency conversion of the orthogonal components of an optical signal by four-wave mixing
in an SOA. Opt. Express 19(21), 2001520022 (2011)
82. R.P. Webb, M. Power, R.J. Manning, Phase-sensitive frequency conversion of quadrature
modulated signals. Opt. Express 21(10), 1271312727 (2013)
83. Z. Tong, C. Lundstrm, P.A. Andrekson, M. Karlsson, A. Bogris, Ultralow noise, broadband
phase-sensitive optical ampliers, and their applications, IEEE J. Sel. Top. Quantum Electron.
18(2), 10161032 (2012)
84. B. Corcoran, S.L.I. Olsson, C. Lundstrom, M. Karlsson, P.A. Andrekson, Phase-sensitive
optical pre-amplier implemented in an 80 km DQPSK link, in OFC, PDP5A.4 (2012)
85. S.L.I. Olsson, B. Corcoran, C. Lundstrm, M. Sjdin, M. Karlsson, P.A. Andrekson, Phase-
sensitive amplied optical link operating in the nonlinear transmission regime, in ECOC, Th.2.
F.1 (2012)
86. K.J. Lee, F. Parmigiani, S. Liu, J. Kakande, P. Petropoulos, K. Gallo, D. Richardson, Phase
sensitive amplication based on quadratic cascading in a periodically poled lithium niobate
87. B.J. Puttnam, D. Mazroa, S. Shinada, N. Wada, Phase-squeezing properties of non-degenerate
PSAs using PPLN waveguides. Opt. Express 19, B131B139 (2011)
88. B.J. Puttnam, . Szab, D. Mazroa, S. Shinada, N. Wada, Multi-channel phase squeezing in a
PPLN-PPLN PSA, in OFC, OW3C.6 (2012)
89. T. Umeki, O. Tadanaga, M. Asobe, Y. Miyamoto, H. Takenouchi, First demonstration of high-
order QAM signal amplication in PPLN-based phase sensitive amplier. Opt. Express 22(3),
24732482 (2014)
90. Y. Zhang, C. Husko, J. Schrder, S. Lefrancois, I. Rey, T. Krauss, B.J. Eggleton, Record
11 dB phase sensitive amplication in sub-millimeter silicon waveguides, in CLEO-PR (2013)
91. Y. Zhang, R. Neo, J. Schrder, C. Husko, S. Lefrancois, D.-Y. Choi, S. Madden, B. Luther-
Davies, B.J. Eggleton, Pump-degenerate phase sensitive amplication in chalcogenide
waveguides. CLEO-PR, ThB2-2 (2013)
92. T. Richter, R. Elschner, C. Schubert, QAM Phase-regeneration in a phase-sensitive ber-
amplier, in ECOC, We.3.A.2 (2013)
93. M. Sorokina, S. Sygletos, A.D. Ellis, S. Turitsyn, Optimal packing for cascaded regenerative
transmission based on phase sensitive ampliers. Opt. Express 21(25), 3120131211 (2013)
94. A. Bogris, D. Syvridis, All-optical signal processing for 16-QAM using four-level optical
phase quantizers based on phase sensitive ampliers, in ECOC, We.3.A.6 (2013)
95. T. Roethlingshoefer, G. Onishchukov, B. Schmauss, G. Leuchs, All-optical simultaneous
multilevel amplitude and phase regeneration. IEEE Photonics Technol. Lett. (2013)
96. J. Matres, G.C. Ballesteros, P. Gautier, J-M Fdli, J. Mart, C.J. Oton, High nonlinear gure-
of-merit amorphous silicon waveguides. Opt. Express 21(4), 39323940 (2013)
Chapter 6
Photonic Signal Processing for Logic
and Computation
Antonella Bogoni and Alan Willner
Abstract Photonic signal processing and computing take advantage of the sub-
picosecond response time of nonlinear optical phenomena to perform high speed
operations either on analog and digital optical signals directly in the optical domain,
thus avoiding any electro-optic conversion. This chapter rst provides a theoretical
explanation of the main nonlinear processes and the key enabling technologies
exploited in all-optical signal processing applications. In the last two section of the
chapter a description of several schemes, used for signal processing, all-optical
logic operation and computation, with the respective experimental characterization,
are reported.
6.1 Introduction
In this chapter, after a brief description of the main nonlinear processes exploited
for all-optical signal processing, a description of the main technologies used for
implementing logic and computation functionalities is provided. In last two chap-
ters the state of the art for logic functions and computing are reported.
A. Bogoni (&)
CNIT, National Laboratory of Photonic Networks,
Via Moruzzi 1, 56124 Pisa, Italy
e-mail: antonella.bogoni@cnit.it
A. Willner
Department of Electrical and Computer Engineering,
University of Southern California, Los Angeles, CA 90089, USA

158 A. Bogoni and A. Willner
6.2 Photonic Logic and Computation Functions

for Multi-Format Data Communication and Storage
Telecommunication transport networks are facing with an impressive increase in

the internet trafc. Statistics show 566 % [1] worldwide increase of internet usage
in the period 20012012. The same statistics became much more impressive if
considering single world regions. In the same period Africa and Middle East
experienced a growth, respectively, of about 3600 and 2600 %. This increase in the
internet connection demand, pushed also by the spread of always connected smart
portable devices, and together with the increased demand for high bandwidth ser-
vices, forces researchers to constantly push the transmission capacity limit. In this
scenario where optical links, with the use of processing hungry complex multi-
level modulation formats (e.g., orthogonal frequency division multiplexing
(OFDM), quadrature phase shift keying (QPSK), M-symbol quadrature amplitude
modulation (M-QAM)), are able to provide Tbit/s channel capacity [2], opto-
electronic conversion and electronic based signal processing became the system
bottleneck. Femtosecond response time of optical nonlinearities makes possible
process signals over bandwidths beyond THz, thus enabling photonic devices as a
promising alternative [3], or even an assisting technology [4, 5] to overcome the
speed limitation of electronic devices.
With respect to electronic solutions, optical processing does not requires optical-
electro-optical conversions, thus performing the required processing directly in the
optical domain and potentially allowing to reduce the system power consumption
[6, 7]. A single photonic element can transparently process a data channel
regardless of its data rate or modulation format, allowing for efcient scaling of
network resources [5, 811]. Moreover, a single photonic element can simulta-
neously operate on multiple data channels, thus strongly reducing the need of
redundant parallel processing structures.
Many are the materials and devices suitable for optical signal processing and
photonic logic implementation, as highly nonlinear bers, lithium niobate wave-
guides, silicon waveguides, and many others. All these have been largely exploited,
allowing to implement all-optical logic gates [3, 5, 12], flip-flops [13, 14], buffers
[15]. More complex all-optical logic functions and computation systems are also
currently under investigation, as all-optical eld-programmable gate arrays
(FPGAs) [16] and routers [17, 18].
6.3 Overview of Nonlinear Processes
The optical nonlinear effects commonly exploited in all-optical signal processing

can be divided in two categories:
wave mixing,
phase modulation.
6 Photonic Signal Processing for Logic and Computation 159
Wave mixing and phase modulation can be studied by considering the material
polarization induced by an externally applied electric eld. When an electric eld
E(t) is applied on a material, a polarization eld P(t) is induced. The total electric
eld in the medium Em t can be described as the superimposition of the external
excitation and the induced polarization [19]:
Pt
Em t E t 6:1
e0
The polarization P(t) depends on the strength of the external excitation E(t), it can
be modelled by Taylor series expansion as:

Pt e0 v1 E t v2 E 2 t v3 E 3 t ; 6:2
where e0 is the electrical permittivity of vacuum, v1 is the linear electric sus-

ceptibility of the material. v2 and v3 , that are zero in linear materials, are the
second and third-order nonlinear susceptibility. If a nonlinear medium has cen-
trosymmetric crystal structure, the second-order nonlinear coefcient v2 vanishes,
while the third-order nonlinear coefcient can be non-zero either in centrosym-
metric and noncentrosymmetric materials.
6.3.1 Wave Mixing
In wave mixing multiple optical waves, at different frequencies, interact with each
other and generate an idler signal at a new frequency. Wave mixing can be caused
by either second (v2 ) or third-order (v3 ) nonlinear interactions. As it will be seen,
in second-order nonlinear materials two waves can interact, thus generating a third
wave (the idler), for this they are also called three wave mixing processes [20]. In
third-order nonlinear materials three waves can mix, thus generating a fourth wave,
this process is called four wave mixing (FWM) [21].
Let us consider a second-order nonlinear medium (v2 6 0), the polarization
vector eld is proportional to the square of the applied eld:
Pt / v2 E2 t 6:3
If the external excitation is given by the sum of two waves E1 t and E2 t, with
respective angular frequency 1 and 2, and phase 1 and 2, the polarization
vector eld can be expressed as:
Pt / v2 E1 t E2 t2 6:4
The mixing of the two input signals generates new waves at frequency:
xDFG1 x1 x2 ; 6:5
xDFG2 x1 x2 ; 6:6
xSHG1 2x1 ; 6:7
xSHG2 2x2 : 6:8
Amplitude and phase of the generated waves are, respectively:
ESFG t / E1 tE2 t
6:9
uSFG u1 u2
EDFG1 t / E1 tE2 t
6:10
uDFG1 u1 u2
EDFG2 t / E1 tE2 t

6:11
uDFG2 u1 u2
ESHG1 t / 2E12 t
6:12
uSHG1 2u1
ESHG2 t / 2E22 t
6:13
uSHG1 2u2
where * denote phase conjugation. The processes responsible of the generation of

the sum, difference and double frequency/phase components are called, respec-
tively, Sum Frequency Generation (SFG), Difference Frequency Generation (DFG)
and Second Harmonic Generation (SHG).
In many applications the two input signals are in the same frequency band (e.g.,
C-band 1550 nm 193 THz), SHG/SFG term are thus generated at 775 nm
(386 THz). Often there is an interest in keeping the generated signals in the same
frequency band as the input ones, this is usually obtained by cascading two of
nonlinear effects, for example SFG and DFG (cSFG/DFG) or SHG and DFG
(cSHG/DFG), as shown in Fig. 6.1, respectively.
If we consider a third-order nonlinear medium, the polarization vector can be
now expressed as:
Pt / v3 E3 t: 6:14
Fig. 6.1 cSFG/DFG (a), cSHG/DFG (b)
Fig. 6.2 Non degenerate FWM
Under the phase-matching condition if three optical waves E1 t, E2 t and

E3 t, with respective angular frequency and phase 1, 2, 3, 1, 2 and 3, are
injected into the material, the polarization vector eld can be expressed as:
Pt / v3 E1 t E2 t E3 t3 6:15
The signals, interacting three at a time generate many mixing components, among
the generated signal shown in Fig. 6.2, idlers (the fourth wave) with angular
frequencies:
xFWM1 x1 x2 x3 ; 6:16
xFWM2 x1 x2 x3 ; 6:17
xFWM3 x1 x2 x3 ; 6:18
Their amplitudes and phases are given by:
EFWM1 t / E1 tE2 tE3 t

6:19
uFWM1 u1 u2 u3

6:20
uFWM1 u1 u2 u3

6:21
uFWM1 u1 u2 u3
are generated. This process, in which a fourth wave is generated by the nonlinear
interaction of three input waves is called non-degenerate four wave mixing.
FWM can occur also if only two input signals are applied, thus having the so
called degenerate FWM. The polarization eld is described by:
Pt / v3 E1 t E2 t3 6:22
Thus generating idler signals at frequency (Fig. 6.3):
xFWM1 2x1 x2 ; 6:23
xFWM2 x1 2x2 : 6:24
Electric eld and phase of the converted signals are thus given by:
EFWM1 t / E12 tE2 t;

6:25
uDFG1 2u1 u2 ;
EFWM2 t / E1 tE22 t;

6:26
uDFG1 u1 2u2 :
In the degenerate FWM, as the converted signals is proportional to the square of

the input signal eld, it neither conserves the phase or intensity information of the
original signals.
The described wave mixing processes are often exploited in optical signal
processing and logic operation, as in [3, 22], in which all-optical XOR of high
speed amplitude and phase modulated signals, have been implemented by means of
cascaded three wave mixing processes in periodically poled lithium niobate (PPLN)
waveguides.
Fig. 6.3 Degenerate FWM

6.3.2 Phase Modulation
If an high intensity signal is injected into a nonlinear medium, the refractive index,
experienced by the optical signal, can change. This modulation of the refractive
index instantaneously changes the effective optical path length thus causing a
nonlinear phase shift on the optical signal. The phase shift is given by [21]:
!
X
uNL cLeff Psignal t 2 Pi t ; 6:27
i6signal
in which is the nonlinear parameter, Leff is the effective length of the medium, and
Pi denotes the power of the i-th co-propagating wavelength. The phase modulation
induced by nonlinear phase shift can be split in two terms:
uSPM cLeff Psignal t; 6:28

X
uXPM 2cLeff Pi t: 6:29
i6signal
The rst term, (6.28), corresponds to the nonlinear phase shift related to the
signal itself, for this reason it is called self phase modulation (SPM). The second
term, (6.29), takes into account the nonlinear phase shift induced by the other co-
propagating signals, for this reason it is called cross phase modulation (XPM).
Figure 6.4 depicts an example where an amplitude modulated data signal and a CW
pump are injected into a third-order nonlinear medium. SPM causes spectral
broadening on the input signal, while the CW pump experiences XPM. Thus, the
information of the data signal is transferred to the CW pump through XPM and can
be recovered by appropriate side-band ltering.
Because nonlinear phase shift is induced as a result of the power variations of the
input signals, XPM and SPM are used in the processing of amplitude-modulated
signals (e.g. OOK).
Fig. 6.4 Cross-phase modulation process

6.4 Enabling Technologies
Many are the nonlinear materials and devices available for implementing logic
functions. v3 nonlinear phenomena are shown by highly nonlinear bers (HNLF),
semiconductor optical ampliers (SOAs) or silicon devices, while lithium niobate
are used for v2 based logic schemes. Depending on the application each material is
able to meet specic system requirements, as large and flat bandwidth or integra-
bility, just to mention some of them. In this section a brief overview on the main
technologies used in all-optical signal processing is reported.
6.4.1 Optical Fiber
Optical bers are transparent bers made of silica. This material is a third-order
nonlinear medium, thus showing FWM, XPM and SPM effects. However standard
bers have a low nonlinear parameter, which leads to low efciency, thus making
standard bers inconvenient to be used as an optical nonlinear medium. By properly
doping the silica, i.e. with germanium dioxide, or by making chalcogenide (e.g.
Ag2Se3) or bismuth-oxide-based (Bi2O3) ber, it is possible to strongly increase the
v3 parameter of the ber, thus making it highly nonlinear. This kind of bers are
called highly nonlinear bers (HNLF). HNLF-based logic schemes take advantage
of the large phase matching bandwidth of the ber, allowing the nonlinear inter-
action of multiple signals in a single device. On the other hand, HNLFs based
processing usually require meter or even hundreds of meters of ber, thus pre-
venting any possibility of integration.
HNLF have been widely exploited for demonstrating all-optical logic functions,
as NOLM-based (nonlinear optical loop mirror) 40 Gbit/s NAND, OR, NOR, XOR,
and XNOR logic gates [23].
6.4.2 Semiconductor Devices
Semiconductor optical ampliers (SOAs) are active optoelectronic devices manu-

factured, when designed for the 1550 nm window (C-band), with IIIV compounds,
as indium phosphyde (InP) or indium gallium arsenide based, as (InGaAsP). If an
high intensity signal is injected into an SOA it undergoes v3 nonlinearities, as for
the HNLF [24, 25]. With respect the this last one, however, SOAs are compact in
size, integrable and provide optical gain. As a drawback, the gain recovery
dynamics of the SOA, typically in the order of tens of ps [26], limits the maximum
working data-rate. Another common semiconductor structure is the semiconductor
ring laser (SRL), which is a ring shaped laser cavity [14]. Devices based on this
structure exploit the Cross Gain modulation (XGM) effect, which is a nonlinear
effect where the optical gain experienced by a signal propagating into an active
medium depends also on the intensity of the all co-propagating signals.
6.4.3 Photonic Crystals
Photonic crystals (PCs) are nanostructures in which the refractive index of the
medium is characterized by periodic variations, in the order of the light wavelength,
thus affecting the propagation properties of the medium. Proper engineering of the
photonic crystal structure can enable to nely control the light propagation, e.g.
negative refractive index, self-collimation etc. When applied to a nonlinear med-
ium, PCs can strongly enhance the nonlinear behavior of such material. Among the
different materials, exploiting photonic crystal structures, the most used are pho-
tonic crystal bers (PCF) [27] and photonic crystal waveguides [28].
6.4.4 Periodically Poled Lithium Niobate Waveguides
Lithium niobate is a non-centrosymmetric crystal exhibiting second-order optical

nonlinearity, this would make lithium niobate waveguides suitable for nonlinear
signal processing. However, the frequency dependency of the phase velocity makes
the phase relation between the interacting signals varying during the propagation
through the crystal, thus causing a periodic succession of constructive and
destructive interaction that makes the maximum efciency of the nonlinear process
extremely low. To solve this problem the quasi-phase matching (QPM) technique
[19] is used in the waveguide designs. QPM consist of a periodic sign change in the
nonlinear susceptibility of the medium, this operation is called poling, thus com-
pensating for the velocity mismatch of the interacting waves [20] and effectively
increasing the nonlinear efciency. This device is called Periodically poled lithium
niobate (PPLN) waveguide. The relatively low propagation loss, and ease of fab-
rication, makes this device suitable for the implementation of advanced all-optical
signal-processing functions at bandwidths beyond THz [20]. As in [3] and [22], in
which PPLN waveguides have been exploited to implement XOR operation among,
respectively, OOK and QPSK modulated signals up to 640 Gbit/s.
6.4.5 Silicon Devices
Silicon devices for optical signal processing applications are attracting lots of
research efforts. Silicon photonic technology, based on CMOS compatible fabri-
cation processes [29], open the way for a mass production and cheap monolithic
integration between optical and electronic components. Common silicon devices

are based on the silicon-over-insulator (SOI) technology, which thanks to the high
refractive index allows the design of ultra-small silicon wire waveguides. Silicon
material shows third-order nonlinear effects, as for optical bers, thus allowing the
implementation of devices based on FWM, SPM or XPM effects [30].
6.5 State of the Art for Logic
6.5.1 Overview of Logic Functions and Achieved Results
Exactly as in the electrical domain, logic operations are basic elements for enabling
digital-signal-processing. However, performing such kind of operations in the
optical domain can dramatically improve system performances, both for single
network operation, e.g. for OTDM demultiplexing [5, 11], or to implement more
complex system, as optical routers or complex processing devices [1618].
In Table 6.1 a brief list of optical logic functions are reported, together with the
maximum achieved data rate and modulation format (in brackets). The last column
reports the exploited nonlinear effect.
6.5.2 OOK 640 Gbit/s Logic Operations
PPLN waveguides can be effectively used for high speed logic gates, as in [3], in
which 640 Gbit/s A and B and A and B logic gates have been implemented by
exploiting pump depletion in a PPLN waveguide.
The proposed scheme of principle is shown in Fig. 6.5a.
Two signals A and B with respective wavelength (A and B) placed symmet-
rically to the Quasi-Phase Matching (QPM) wavelength Fig. 6.5b of the PPLN
waveguide can nonlinearly interact through SFG nonlinear effect. As a conse-
quence, if both input signals are high the two wavelengths undergo signal depletion,
while no depletion occurs if one of the two signals is low. In this way, at the PPLN
Table 6.1 Examples of state of the art all optical Logic functions and their data rate
Logic operation Data rate Nonlinear effect
40 Gbit/s (OOK) FWM in HNLF
1.28 Tbit/s (OOK) FWM in ChG waveguide [5, 11]
160 Gbit/s (DPSK) FWM in ChG waveguide [12]
AB 640 Gbit/s (OOK) Pump depletion in PPLN waveguides [3]
Half-adder 40 Gbit/s Cascaded SFG/DFG in PPLN waveguide [31]
Optical buffer 130 s
(a) (b)
640 Gb/s Logic gate SFG
Signal A A and B
Pump
Depletion
A Pump BPF A
t
Depletion t QPM
B in PPLN BPF B
t
A and B
t QPM A B
640 Gb/s
Signal B 2
Fig. 6.5 Scheme of principle (a), pump-depletion (b)
Input Output OR/XOR functions

signals signals
A OR B = B + AB = A + AB
A B A B A XOR B = AB + AB
0 0 0 0
0 1 0 1 Half-subtractor
1 0 1 0
DIFF = A XOR B = AB + AB
1 1 0 0
Borrow (A-B) = AB
AB AB Borrow (B-A) = AB
Fig. 6.6 Truth table
output, A and B or A and B logic operation can be selected by ltering, respectively,

a or B, as also reported by Fig. 6.6.
Moreover, as summarized in Fig. 6.6, coupling together the two logic function
outputs, the XOR logic operation can be accomplished. XOR output represents the
difference between A and B. Thus the output signals at A and B individually give
the Borrow (B A) and Borrow (A B) functions, respectively, thus allowing to
obtain an half-subtractor. Finally, coupling together the input signal B and the
output signal at A or the input signal A and the output signal at B, the A OR B
function can be obtained.
Experimentally, by exploiting this principle, two 640 Gbit/s OTDM signals have
been optically processed. As shown in [32], the pump depletion effects can not be
simultaneously maximized for both pumps, since simultaneous optimization will
decrease the system performance. Therefore, when the pump depletion is optimized
for signal A, A and B logic function is performed, otherwise A and B logic operation
is obtained. The use of two PPLN waveguides will eventually allow to simulta-
neously optimize both logic gates. Note that pump depletion introduces little dis-
tortion on the pump signals. Figure 6.7 shows the experimental setup.
In order to generate the two input signals a 40 GHz mode locked laser (MLL),
producing *1 ps pulses at C = 1542 nm, is modulated by a 40 Gbit/s PRBS
sequence, and then split into two branches. The rst branch is directly multiplexed,
40-to-640 Gbit/s, by using an optical ber-based multiplexer. The second branch is
used to generate a 40 Gbit/s supercontinuum spectrum through propagation in an
HNLF. The supercontinuum spectrum is then ltered to obtain the 40 Gbit/s data
signal at s = 1560 nm. Another 40-to-640 Gbit/s optical ber-based multiplexer
ODL SC-Based 640 Gb/s

Wavelength Converter signal A
BPF DCF
40 GHz MUX
BPF 40:640
MLL
ODL MZM Gb/s
DCF
MUX
40 GHz Pattern 40:640 BPF ODL
clock generator 40 GHz Gb/s 640 Gb/s
Clock
signal B
BPF
PPLN-Based Optical Logic Gate
Receiver BPF BPF BPF BPF PPLN
DEMUX
Fig. 6.7 Experimental setup
640 Gb/s 640 Gb/s

A and B A and B
Signal A Signal B
1.56 [ps/Div]
Fig. 6.8 Input and output eye diagrams
produces the second 640 Gbit/s signal. Polarization controllers are needed in the two
signal paths in order to align the two signal polarizations. At the output of the PPLN
waveguide, two cascaded tunable band-pass lters (BPFs) and a low noise EDFA are
used to perform the desired logic function. Finally, a 640-to-40 Gbit/s nonlinear
optical loop mirror (NOLM)-based optical demultiplexer is used to test the perfor-
mance of the involved signals. Depending on the wavelength of the signal under test,
a supercontinuum-based wavelength converter is used to tune the clock wavelength.
In Fig. 6.8, the eye diagrams of the involved signals at the PPLN input and
output shows as the pump depletion effect introduces little amplitude distortions on
both the OTDM frames. The slightly broadening of the involved signals pulsewidth
is ascribed to chromatic dispersion.
6.5.3 PSK 160 Gbit/s Logic Functions
Phase modulated signals are an attracting solution for future optical networks, for this
reason many efforts are spent for the study of all-optical logic gates able to process
high speed PSK signals. In [12] an all-optical XOR an among 160 Gbit/s return-to-
zero (RZ) DPSK signals is performed by exploiting four-wave mixing (FWM) in a
dispersion-engineered, highly nonlinear chalcogenide (ChG) planar waveguide.
ChG planar waveguides are optical devices showing third-order nonlinear

behavior. If two Differentially encoded Phase Shift Keying (DPSK) data streams
with respective wavelength 1 and 2 (phases 1 and 2) and a continuous wave
(CW) at p (with a constant phase p) are injected, in a co-propagating way, injected
signals undergo non-degenerate FWM. As a result of such interaction, an optical
idler at wavelength i and phase i = 1 + 2 p is generated. As illustrated in
Fig. 6.9b, showing the phase of the FWM idler at the output of the ChG chip, if the
two input DPSK channels are out of phase the phase of the FWM idler is while
it is 0 if they are in phase. At the chip output the idler signal can be separated by
means of an optical band pass lter (BPF). The XOR logic operation is veried by
demodulating both input data and idler signal.
ChG waveguides exhibits a large nonlinear coefcient * 104 W1 km1, thus
allowing high nonlinear interaction efciencies. The large ChG dispersion is offset
by tailored waveguide dispersion, thus yielding an anomalous dispersion of
*29 ps/nm/km. The low dispersion and short propagation length (6.8 cm) offers
ultra-low walk-off and efcient phase matching between the input signals and the
CW probe.
The 160 Gbit/s XOR gate has been experimentally implemented by the setup
shown in Fig. 6.10. A 40 GHz pulse train (s = 1555 nm) was spectrally broadened
to *20 nm by 400 m and a 100 mm HNLFs. The pulse train was then modulated
by a Mach Zehnder modulator and interleaved to produce a 160 Gbit/s DPSK
signal. A Finisar Waveshaper was used to create two replicas of the 160 Gbit/s
DPSK signal at different wavelengths.
The average power of each channel inside the waveguide was *10 mW,
yielding a relatively low energy per bit (*62.5 fJ/bit). The XOR product has
Fig. 6.9 a Working principle of the ChG chip based all-optical XOR gate and b a truth table of
the phase of a FWM idler (XOR gate)
Fig. 6.10 Experimental setup for a chip based XOR gate for DPSK signals
Fig. 6.11 a Eye diagrams of two 160 Gbit/s input channels and XOR idler. b waveforms and
c eye diagrams of two demultiplexed and demodulated 40 Gbit/s input data and a XOR product
been then extracted by means of BPFs, demultiplexed to 40 Gbit/s and nally

demodulated.
As illustrated by Fig. 6.11a, being the signal information encoded in the signal
phase, the two 160 Gbit/s input signals and XOR product show constant pulse
amplitude. XOR operation is conrmed by In Fig. 6.11b, c, where the demulti-
plexed and demodulated waveforms and eye diagrams of the two 40 Gbit/s DPSK
data and the XOR signal are reported.
6.5.4 Hexadecimal 16PSK Addition
In binary logic functions, the information bit-rate is the same as the signal switching
rate (baud rate). Optical signals can use multi-level modulation formats, such as
16-PSK, to encode multiple bits onto a symbol. Therefore, logic functions that
operate on multi-level modulated signals exhibit higher bit rates than switching
rates. Here, we present the work in [33] that demonstrates simultaneous addition
and subtraction on 10-GBd 16-PSK signals in an HNLF through non-degenerate
FWM. Figure 6.12a shows the schematic output spectrum of the HNLF. Signals
A and B and a CW pump P are sent to an HNLF with low dispersion slope close to
the ber zero dispersion wavelength (ZDW). Non-degenerate FWMs occur
simultaneously to produce signals proportional to EP EA tEB t, EA tEB tEP , and
EB tEA tEP . Because signals A and B are 16-PSK, the multiplication of their elds
results in the addition of their phases \A \B, and the multiplication of a signal
with the phase-conjugate of the other signal generates subtraction of their phases
(i.e., \A \B and \B \A).
The experimental constellation diagrams of the input and output signals are
illustrated in Fig. 6.12b, in which the error vector magnitude (EVM) of the addition/
subtraction output signals is on average 2.5 % higher than that of the input.
(a) (b) OSNR

Penalties at
Addition
BER of 2e-3:
16-PSK A+ B ( A + B)
Signal A Signal A
Nonlinear A B EVM: ~10.0% 5.0 dB
Device (Input)
16-PSK
Signal B B A EVM: ~7.3%
Subtraction
( A B)
CW
EVM: ~9.8% 4.3 dB
A Pump B
Signal B
A B A+ B B A
(Input)
EVM: ~7.6% Subtraction
Wavelength ( B A)
EVM: ~9.9% 4.8 dB
Fig. 6.12 a Concept of using nonlinearities (e.g., FWM) for high-base number addition and
subtraction on PSK signals. b Experimental results for 16-PSK addition and subtraction at 10 GBd
speed [33]
6.6 State of the Art for Computation
6.6.1 Overview of Tapped Delay Lines
A key building block for many signal processing applications is a tapped-delay-line

[3437]. As shown in Fig. 6.13, the incoming data signal is tapped at different time
intervals; each tap is multiplied by a complex coefcient; the taps (copies) are then
added together to form the output. The output, y(t), of the tapped-delay-line is a
weighted sum of the delayed copies of the input signal, x(t):
X
N1
y t hi xt Ti 6:30
i0
in which, N is the number of taps, hi and Ti are the complex coefcient and time
delay of the i-th tap, respectively. This equation represents an N-tap FIR lter that
can be used for various ltering applications. The function of the tapped-delay-line
(TDL) can be changed by varying its parameters.
6.6.2 Fundamental Tools to Enable Photonic TDL
An optical TDL can be realized by implementing an all-optical fan-out stage to

create multiple replicas of the original signal using (2) nonlinear processes in the
periodically-poled-lithium-niobate (PPLN) waveguide, conversion/dispersion delay
to induce relative delays on signal replicas [38] and coherent multiplexing in
Input, x(t)

Delay Delay Delay Delay
T0 T1 T2 TN-1
|h0|e j h0 |h1|e j h1 |h2|e j h2 |hN-1|e j hN-1

+ + + + y(t), Output
Fig. 6.13 A tapped-delay-line generic block diagram: Input signal is tapped at different time
intervals, copies are weighted by a coefcient and summed to form the output
Tunable Optical Tapped-Delay-Line

Copy 1 1 01 1 00 Copy 1 1 0 1 10 0
Copy 2 1 01 1 00 Copy 2 1 0 110 0
Copy 3 1 0 1 1 00 Copy 3 1 011 0 0
Time Time
1 N N 1
Input Optical Multicasting Optical Multiplexing Output
Signal
Nonlinear
Phase Filter
Frequency-Dependent Nonlinear Signal
N+1 Element Optical Delay Element
(LCoS Filter)
lasers (PPLN) (Chromatic Dispersion) Laser (PPLN)
SFG Mixing DFG Mixing

DFG Mixing SFG Mixing
Input Delay 2L Output
Power
Power
Delay
Various Tap Signal Delayed Copies Signal

Coefficients Signal Copies
P1 D3 D2 D1 C1 C2 C3 S P2 D3 D2 D1 C1 C2 C3 MUX
Frequency ( ) Frequency ( ) Frequency ( )
Fig. 6.14 Optical implementation of a TDL based on nonlinearities and conversion-dispersion-

based delays: N copies of the input optical signal are generated at different frequencies using
cascaded nonlinear wave mixings of SFG followed by DFG. Copies are sent into a chromatic
dispersive medium to introduce the tap delays. Delayed and weighted signal copies are sent to a
second nonlinear medium to be multiplexed and to create the output signal [36]
another PPLN waveguide [39]. The principle of operation of this tunable TDL is
depicted in Fig. 6.14. The incoming data signal is rst fanned-out to N signal copies
using cSFG-DFG in a PPLN waveguide. The fan-out scheme uses N independent
pumps to generate N copies. All signals and pumps are sent into a phase and
amplitude programmable lter based on liquid crystal on silicon (LCoS) technology
[40], in which the independent signal copies and their pumps are let through while
the original signal and pump are blocked. The LCoS lter can also be used to apply
the tap phases. Next, the signals propagate through a chromatic dispersive element
such as DCF in which the N signal copies are delayed with respect to each other.
The delayed signals are then sent into another PPLN waveguide to be multiplexed
to one wavelength channel coherently. Each copy represents a tap in the TDL;
therefore, the number of taps is the number of pump lasers, and tap amplitudes can
be independently varied by changing the power of the pumps that generate the
signal copies. The tap delays are based on conversion-dispersion delays; therefore,
can be widely tuned by changing the wavelengths of the pumps. Finally, the tap
phases can either be imposed by the LCoS lter or ne tuning of the pumps [41].
Therefore, the TDL can be completely tuned by varying the properties of the input
pump lasers.
According to the wave mixing rules presented in Section II, the equivalent
mathematical representation of this optical TDL, with input electric eld amplitude
Es(t) and output eld EMUX(t) is [36]
X
N
EMUX t / jhi jej]hi ES t Ti 6:31
i1
in which,
jhi j jEDi j2 6:32
]hi L b2 xDi xQPM 2 /LCoS

i ci 6:33
Ti L b2 xDi xD1 6:34
In (6.22)(6.24), EDi is the eld amplitude of the i-th CW pump, 2 is the group
velocity dispersion parameter of the chromatic dispersive medium of length L,
/LCoS
i is the phase that can be applied on i-th pump after the rst PPLN stage by the
LCoS lter, and ci represents a phase shift due to higher-order dispersion terms.
Alternatively, the phase of the i-th tap can be varied by ne frequency detuning of
the xDi dummy pump by dxDi . This detuning can change ]hi approximately by
2L b2 xDi xQPM dxDi . Because the tap delays also depend on the frequency of
the pumps, the ne-frequency detuning technique to apply tap phases can cause
deviation in the tap delays dTi =Ti dxDi =xDi xQPM . Therefore, care must be
taken in choosing pump wavelength separations and dispersion values to minimize
time error degrading effects.
The tunable optical TDL can be used to perform many different signal pro-
cessing functions, including FIR ltering [42], equalization, correlation [36], dis-
crete Fourier transform [43], D/A conversion, and synthesis of arbitrary optical
waveforms [44], all at the line rate of optical communications.
6.6.3 Optical 1D Correlation Results Using Nonlinearities

and On-Chip MZIs
A correlator can search for a pre-determined target pattern on an incoming data

stream at the line rate [39, 45]. Figure 6.15a shows the concept of pattern matching
to an incoming signal. A correlator can be implemented using the TDL structure.
(a) Signal Symbols:

OOK: 0,1 ; BPSK: -1,1 ; QPSK: Example: A, B, C, D
Input Data Stream
A C C B A A B D One time before
A C C B A A B D Matched time
A C C B A A B D One time after
Pattern for Full Match
Matching: CBAA C* B* A* A*
Threshold
Not
Matched
Four Taps
time
1 Match 4 Matches 2 Matches
- - A- C B A A C - - A
(b) Vector Addition Amplitude & Phase Intensity

I
Not
Tap 1 Tap 2 Tap 3 Tap 4 2 Matched
+ + + = Q time
Correlation
I 4 Full Match
Peaks
Tap 1 Tap 2 Tap 3 Tap 4
+ + + = Q time
Fig. 6.15 Concept of a correlator for phase-modulated signals: a tap coefcients are set equal to
the complex conjugate of the search pattern; so when the input data stream slides through the
tapped-delay-line, a high correlation peak occurs if the patterns match. b Complex-coefcient taps
enable correlation on QPSK signals by allowing vector addition of adjacent symbols by generating
higher-order QAM with a variable intensity [36]
Here, the tap coefcients are set as the phase-conjugate of the target pattern that we
want to search for. As the data stream slides through a set of taps, adjacent symbols
are multiplied by the tap coefcient and added to form the output. To identify the
appearance of particular consecutive values, CBAA, in a long pattern of QPSK
symbols, the correlator weights need to be set to the phase conjugate of the target
pattern (i.e., C*B*A*A*). An N-tap correlator on QPSK signal has an output of
(n + 1)2-QAM constellation. Figure 6.15b shows how the 25QAM constellation can
be generated by 4QPSK signals. The upper right corner of the output constellation
corresponds to the target pattern. One example of matched pattern (upper right
corner) and one example of a mismatched pattern are shown in Fig. 6.15b. Direct
detection of the correlator output can be used to search the pattern based on output
intensity. When the patterns fully match, a high intensity is created, while partial
matches result in lower amplitudes. Additionally, a thresholder can be used after the
correlator to identify full matches when the intensity exceeds a certain threshold.
Figure 6.15b also illustrates the corresponding intensity of two sample cases of a
full match (large amplitude output) and a mismatch [11]
(a) Fixed Bit Rate 8-bit Optical Correlator In [46], a recongurable silica planar-
lightwave-circuit optical delay line lter is used for pattern recognition on a
40-Gbit/s BPSK signal. Figure 6.16a shows the schematic of an 8-bit corre-
lator with tunable coefcients. The variable optical attenuators (VOA) set the
amplitude of the taps to be equal, and phase shifters program each tap phase to
be 0 or to choose the target pattern. The experimental results of the 8-bit
correlator are shown in Fig. 6.16b. Here, the target pattern is set to 0 0 0
0 0. Assuming that, of 8 bits, only m bits match, the output eld of the
correlator is (+1) m + (1) (8 m) = 2 m 8 (m {08}). The correlator
adds the optical elds; thus, when all 8 bits match, the eld has a level of 8.
However, the photodiode is a square law device; therefore, the amplitude of a
full match after the photodiode is 64, which is higher than the amplitude of 7
matches (i.e., 36). Figure 6.16b depicts the output eye diagram after the
photodiode. The optical waveform after the correlator is also shown with the
electronically thresholded pattern that features a pulse where a pattern match
occurs.
(b) Tunable Correlator The optical TDL can be used to perform optical corre-
lation on phase- and amplitude-modulated signals (BPSK, QPSK, and OOK)
[36]. Figure 6.16cf summarize examples using this approach on 20- and 40-
GBd signals. Output eye diagrams, optical correlation output waveforms, and
electronically thresholded waveforms are shown for three different modulation
formats and patterns. In Fig. 6.16c, an output with a 1 1 1 1 pattern search in
a 40-Gbit/s RZ-OOK signal is presented. Figure 6.16d, e illustrate the
experimental results for 40-Gbit/s RZ-BPSK and 80-Gbit/s RZ-QPSK signals,
in which the patterns 0 0 and 3/4 3/4 /4 are searched, respec-
tively. In Fig. 6.16f, the coherent detection of the correlator output is depicted
(a) All-Optical BPSK Correlator (c) (d) (e) (f)

40-Gbit/s RZ-OOK signal, 40-Gbit/s RZ-BPSK signal, 80-Gbit/s RZ-QPSK signal, 40-Gbit/s QPSK signal,
pattern: 1 1 1 1 pattern: 0 0 pattern: 3 /4 3 /4 /4 pattern: /4 3 /4 5 /4 /4
Eye Diagram
Correlation
Intensity
Optical
VOA: variable optical attenuator

Phase shifter: 0 and phases 0 50ps 0 50ps 0 50ps
Correlation
Waveform
(b)
Intensity
Optical
8-bit Correlation for 40-Gbit/s BPSK Signals
Correlation Output Optical Correlation Waveform

Eye Diagram 1
Peak Level (v)
Pattern: 500ps 500ps 500ps

Thresholded
0 0 0 0 0
Waveform
Waveform
Electrical
0.5
Electronically ThresholdedOutput
0
0 1000 2000 3000 4000
Time Time Time
Symbol
0 5 10 15
Time ( ps )
Fig. 6.16 a Structure of a silica planar lightwave circuit optical delay line lter for 8-bit
correlation of BPSK signals. b Experimental eye diagram, optical correlation waveform and
electronically thresholded output for searching pattern 0 0 0 0 0 [46]. All-optical tunable
correlation based on a tunable tapped-delay-line to search for (c) pattern 1 1 1 1 in a 40-Gbit/s
RZ-OOK signal, d pattern 0 0 in a 40-Gbit/s RZ-BPSK signal, and e pattern 3/4 3/4 /
4 in a 80-Gbit/s RZ-QPSK signal [36]. f Coherent detection of the correlation results on 40-Gbit/s
QPSK signal when /4 3/4 5/4 /4 is the target pattern
while the incoming 20-Gbaud QPSK data is searched for the pattern /4 3/4
5/4 /4. The correlation peaks that correspond to the upper right corner of
the output constellation are also shown. These peaks indicate the position of
the target pattern appearance in the rst 4096 incoming symbols.
6.6.4 Optical WDM Correlator and 2D Correlation
The optical tunable correlator performs pattern searching on a single incoming data
signal. In many applications, it is desired to have a single correlator that can process
multiple WDM data channels simultaneously and independently. This scheme can
increase the processing capacity dramatically by performing parallel computations
on independent incoming data channels.
Figure 6.17a shows a multi-channel correlator. The WDM input channels are
simultaneously wavelength converted in a nonlinear medium, which generates a
replica for each WDM channel at a new center frequency. Because each signal and
its replica are at different frequencies, a relative delay is induced between them after
passing through a chromatic dispersive medium. The replicas and delayed original
signals are then sent into a phase and amplitude programmable lter based on LCoS
technology that applies the complex tap coefcients to the replicas. Finally, the
delayed original signals and their weighted replicas are sent into another wave-
length converting stage that coherently copies the original signals onto the replicas
for a second time, which creates the desired two-tap processing. Thus, a same-delay
through independent two-tap OTDL can be implemented on each WDM channel. A
higher number of taps can potentially be implemented by cascading the two-tap
tunable building blocks (see Fig. 6.17a).
(a) (b)
Multiple S (t) SHG+DFG or FWM Independent 2-tap
1 S2(t) S3(t) Optical Spectrum 8-Channels
Input Original Correlators on 8 Channels
Phase -Conjugate Copies
Channels Input Signals
Delayed & Weighted
(1st Taps) Generation of 1st Taps, 26-Gbd QPSK
Ch-5
Ch-1
S1 S3 S1(t) S3(t) 8 Input Signals

1st WDM a1S1*(t-T 1)
Dispersive a3S3*(t-T 1) First Taps
-Conversion Device & P
(Nonlinear 9.3%, [1, 1] 10.8%, [1,x,1]
Device)
LCoS Filter C3 C1 P S1 S3 C8 C1
Delay = Dispersion
Ch-6
Ch-2
1st & 2nd Taps SHG+DFG

S1 S8
a3S3*(t-T 1) a1S1*(t-T 1) S1(t) S3(t) 1541 1551 1561

+ S1*(t) 9.0%, [1, j] 10.7%, [1,x,j]
2nd WDM Dispersive
+S3*(t)
Conversion Device & Generation of 2nd Taps, 26-Gbd QPSK
(Nonlinear
Ch-7
Ch-3
Device)
LCoS Filter C3 C1 P S1 S3 Original
Signals Output
1st, 2nd & 3rd Taps SHG+DFG (1st & 2nd Taps)
10.7%, [1, -1] 10.8%, [1,x,-1]
b1S1*(t-T 2)+ a1S1*(t-T 1)+ S1*(t) S1(t) S3(t)
3rd WDM
Ch-8
Ch-4
Dispersive
-Conversion Device &
(Nonlinear 1541 1551 1561
Device)
LCoS Filter C3 C1 P S1 S3
3 Taps Add Wavelength (nm) 10.0%, [1, -j] 13.0%, [1,x,-j]
Fig. 6.17 a Principle of operation of the WDM-OTDL, shown for a three-tap implementation:
First, taps are created by simultaneous wavelength conversion of all WDM channels in nonlinear
elements, with one such element for each tap. Second, a dispersive element induces a relative delay
between them. Then, tap coefcients are applied by an inline LCoS lter. Because the pump and
signals are reused, the weighted taps add coherently. b Spectra after the rst and the second
nonlinear stages and simultaneous two symbol pattern search on eight 40-GBaud QPSK WDM
channels, which results in an aggregate throughput of 416 Gbit/s
Figure 6.17b depicts the spectra of generating the rst tap and second tap for the
eight-channel 20-Gbaud QPSK data. The two-tap OTDL is in principle, an inter-
ferometer, thus destructive interference can be observed as power dips in the
converted signals of the second spectrum. A correlation of a two-symbol pattern
search is performed by applying various tap coefcients (see Fig. 6.17b). The
processed channels are detected using a coherent receiver [47].
Using the WDM correlator scheme, a 2-D correlator can be implemented to
perform pattern searching in 2-D images. The conceptual block diagram of the 2-D
OTDL is shown in Fig. 6.18. First, each image row is mapped to a WDM channel
by modulating the optical frequency comb ngers (phase locked sources) with
QPSK symbols that correspond to the color of the image pixels. Second, a WDM-
TDL is performed by generating the inner sums on each WDM channel with the
given coefcients (bi) independently. Subsequently, these processed WDM chan-
nelswhich are phase coherentare multiplexed together with column coefcients
(ai) to generate the outer sums.
The experimental setup for the 2-D correlator is depicted in Fig. 6.19a. A mode-
locked laser with a 10 GHz repetition rate and a DLI with FSR 20 GHz is used to
generate coherent comb ngers with 20 GHz frequency spacing. The 20 GHz comb
is then passed through an HNLF ber to generate a flat and broad spectrum. A Liquid
Crystal on Silicon (LCoS-1) lter is used to select and write complex weights on the
comb ngers and separate them into the signal path and pump path. A nested Mach-
Zehnder modulator is used to generate the 20-Gbaud QPSK data on the two comb
ngers and write the image pixels information on the signal path separated by
*1.6 nm. To make the two modulated signals as two consecutive rows in a 31 31
image, they need to be sent to DCF-1 with 12 km length to induce *969 ps/nm
dispersion (31-symbols of delay). The resulting signals are coupled with an amplied
CW pump and sent to an HNLF to produce the rst tap of the inner sum. All signals
then travel through DCF-2 and LCoS-2 to apply tap delays, phases and amplitudes.
Target Pattern (coefficients) Match
Threshold
Input Map Each Row WDM Weighted Multiplexing
to a WDM 1-D Correlator of WDM Channels
Image Channel (Inner Sums) (Outer Sum) time
Correlation over Rows Correlation over Columns

Correlator
I
peaks
Row 1 WDM Tapped-Delay-Line Weighted Multiplexing
WDM Wavelength Wavelength WDM Wavelength Wavelength WDM Q
Conversion Dependent Delay Conversion Dependent Delay Multiplexing
Row 2 f f
P1 Pump Laser Delay=Dispersion Delay=Dispersion P2 Pump Laser
Input Rows First Copies Second + first Copies 2D Correlator Output

S1(t) S2(t)
S1 S2 S1 S2 P c2 c1 S1 S2 P1 c2 c1 D1 D2 P2 c2 c1
Fig. 6.18 Concept of 2-D optical tapped-delay-lines: A 2-D correlation can be achieved by
cascading a WDM correlator with a multiplexing stage to realize the nested sum required for 2D
operation
(a) 0
1-tap Row
(b)
0
-20
Correlations
-20 22
-40
3131 Image (961 pixels)
Correlation
-40
-60
-60
Target
-80
-80
1542 1545 1548 1551 1554 1557 Pattern
1545 1550 1555 WDM TDL(Inner Sum)
31 Bits Delay
Flat 20-GHz Comb Generation 450m
EDFA
400m 5 nm
12km
LCoS
MLL Filter- 5 nm DCF-1 PC
BPF HNLF
DLI HNLF 1
5 nm 1 nm
20 Gbaud P1
70 m Correlation Peaks
P2
nm
5
LCoS
PPLN-1 Filter- Mismatched
12km 2 DCF-2
1 nm 5 nm 5 nm
Coherent
ATT PPLN-2
Receiver 1 nm 0
1 nm 1 nm DCF-3 2-tap Row
Weighted MUX (Outer Sum) -20 Correlations ~4
-40
Matches
-60
-80
1542 1545 1548 1551 1554 1557
Matched
Fig. 6.19 a Experimental setup of 2-D optical tapped-delay-lines. First appropriate comb ngers
are selected for both the signal and pump paths and the image pixel colors are mapped to the
QPSK symbols. Then, using rst and second wavelength conversions and a dispersive element, a
WDM correlator is realized on both rows. Finally, another nonlinear stage is used to coherently
combine the two processed rows with the outer sum coefcients. b Searching a 2 2 target pattern
in an image with 961 pixels. The dark squares correspond to the target patterns
The signals, their copies, and the CW pump laser are then sent to the rst PPLN
waveguide to create the second tap. The quasi-phase matching (QPM) wavelength of
the PPLN waveguide is temperature-tuned to the CW pump wavelength. These
signals, along with the comb ngers selected for pump path and another CW laser
pump, are sent to the second PPLN to perform coherent multiplexing of the two
processed rows. The multiplexed signal is ltered and sent to the coherent receiver,
after passing through another DCF of the same length to compensate for the induced
dispersion. In Fig. 6.19b, a 31 31 image (961 pixels) is searched for a 2 2 target
pattern. The experimental results exhibit 4 peaks (dark points) in output table at the
2 2 sections that matched the target patterns [48].
6.6.5 Discrete Fourier Transforms Using Nonlinearities

and On-Chip MZIs
Optical OFDM signals involve creating a wideband data channel by modulating

multiple orthogonal frequency subcarriers at a lower rate. The modulated subcar-
riers overlap in the frequency domain [49]. Traditionally, optical OFDM signals
have been generated in electronics using electronic inverse fast Fourier transform
(IFFT) techniques to calculate time-domain symbols from frequency-domain
symbols, which, after parallel-to-serial and digital-to-analog conversions are con-
ducted, create the I and Q electrical signals that modulate a laser [49]. The subcarrier
modulation speed is limited to the speed at which discrete Fourier transform (DFT)
or fast Fourier transform (FFT) can be implemented in electronics. Alternatively,

OFDM signals can be generated optically using an optical frequency comb source
as the optical subcarriers and independently modulate each comb line and combine
them to create an optical OFDM channel [43, 50]. An optical OFDM channel can be
demodulated using parallel electronic DFT or FFT. Alternative optical DFT/FFT
approaches can allow for the use of higher baud rates for subcarrier modulation by
performing the DFT/FFT at the line rate of optics [43, 50, 51].
An OFDM signal with N subcarriers separated by Df can be represented as
X
N1
S t xn tej2pf0 nDf t 6:35
n0
in which f0 and xn t are optical carrier frequency and nth subcarrier signal,
respectively [4]. Optical N-point DFT can extract the nth subcarrier signal from
St, according to
N 1
X
k
xn t S t ej2pnk=N 6:36
k0
Df N
which needs to be sampled at 1=Df time intervals to recover the transmitted

symbols. This demodulation equation follows the form of a tapped delay line in
(X). In this section, we review two examples of optical OFDM demodulation
involving a xed FFT approach and a baud-rate-tunable DFT approach based on the
tunable TDL presented in Sect. 6.1 [43, 50].
(a) Optical FFT Based on Delay Interferometers The concept of an optical FFT to
separate the subcarriers of an optical OFDM signal is shown in Fig. 6.20a [50].
The structure is based on cascaded delay interferometers and is capable of
performing serial-to-parallel conversion with optical FFT to decompose the
optical OFDM signal into its constituent subcarriers. The data of each sub-
carrier is then recovered by proper time-gating using electro-absorption
modulators. This conguration requires only N 1 passive delay interfer-
ometers to realize an N-point FFT. An 8-point FFT with xed delays is
demonstrated in [50]. Furthermore, in [50], 325 comb lines of a 12.5-GHz
frequency comb source are modulated by 10-GBd 16-QAM signals in two
polarizations to generate an OFDM channel at 26-Tbit/s. A simplied two
point FFT is performed to extract the subcarriers. Figure 6.20b shows two
constellation diagrams of the extracted 16-QAM subcarriers with the spectrum
and error vector magnitude (EVM) of all output subcarriers.
(b) Adjustable Bit-Rate Optical DFT The tunable optical TDL depicted in Fig. 6.14
has also been used to perform optical DFT with various numbers of taps and bit
rates [43]. Figure 6.21a illustrates the concept of using the tunable TDL to
perform DFT for extraction of OFDM subcarriers. According to OFDM
demodulation (6.27), which follows the form of the tunable TDL (6.21), to
Fig. 6.20 Optical OFDM demodulation using delay-interferometer-based fast Fourier transform:
a structure, and b constellations for two 10-GBd demultiplexed subcarriers of a 26-Tbit/s OFDM
signal (PM-16-QAM subchannels) [50]
(a) (b)
Multicasting
Input
OFDM Signal Multiplexing Stages Signal Pump
NL-Waveguide
SC1
NL-Waveguide
NL-Waveguide QPM
Multicasting Stage
NL-Waveguide 1541.7 1551.7 1561.7
NL-Waveguide SC2
NL-Waveguide
Multiplexing
NL-Waveguide Output
Subcarrier
NL-Waveguide P
SC3
NL-Waveguide
Multiple Wavelength Source QPM

SC4
1541.7 1551.7 1561.7
Wavelength (nm)
Fig. 6.21 Tunable optical OFDM demodulation using optical DFT enabled by the optical tapped-
delay-line: a concept, and b multicasting and multiplexing spectra and output eye diagrams of all
four subcarriers in a four 20-GBd BPSK subcarrier OFDM signal
extract the nth subcarrier from an N-subcarrier OFDM signal with Df separa-
tion, the tunable TDL needs to be congured to have N taps, each separated in
time by 1=N Df , with tap coefcients ej2pnk=N k 2 f0; :::; N 1g. In
Fig. 6.21b, the input OFDM channel is a 80-Gbit/s four-subcarrier BPSK signal
generated optically by modulating four optical comb lines separated by
20 GHz. The multicasting and multiplexing spectra for the implementation of a
4-tap TDL are shown in Fig. 6.21b, where the tap coefcients and tap delays are
congured to represent the coefcients of a 4-point DFT. The TDL coefcients
were recongured four times to demodulate each of the four 20-Gbit/s BPSK
subcarriers. The eye diagrams of the four extracted subcarriers are also shown
in Fig. 6.21b.
References
1. http://www.internetworldstats.com/stats.htm
2. L. Pot, G. Meloni, G. Berrettini, F. Fresi, M. Secondini, T. Foggi, G. Colavolpe, E. Forestieri,
A. DErrico, F. Cavaliere, R. Sabella, G. Prati, Casting 1 Tb/s DP-QPSK communication into
200 GHz bandwidth, in Proceedings of the International Conference ECOC, Amsterdam, 2012
3. A. Bogoni, X. Wu, Z. Bakhtiari, S. Nuccio, A.E. Willner, 640 Gbit/s photonic logic gates. Opt.
Lett. 35(23), 39553957 (2010)
4. F. Scotti, F. Laghezza, S. Pinna, P. Ghel, and A. Bogoni, High precision photonic ADC with
four time-domain-demultiplexed interleaved channels, in Proceedings of the International
Conference Photonic in Switching, paper TuO1-3, Kyoto, 2012
5. T.D. Vo, H. Hu, M. Galili, E. Palushani, J. Xu, L.K. Oxenlwe, S.J. Madden, D.-Y. Choi, D.
A.P. Bulla, M.D. Pelusi, J. Schrder, B. Luther-Davies, B.J. Eggleton, Photonic chip based
transmitter optimization and receiver demultiplexing of a 1.28 Tbit/s OTDM signal. Opt.
Express 18(16), 1725217262 (2010)
6. J. Marti, K. Williams, Is all-optical processing green? in presented at Workshop OFC 2010,
San Diego, CA
7. R.S. Tucker, A green internet, in Proceedings of the IEEE Lasers Electro-Optics Society
Annual Meeting (2008), pp. 45
8. P. Martelli, P. Bof, M. Ferrario, L. Marazzi, P. Parolari, R. Siano, V. Pusino, P. Minzioni, I.
Cristiani, C. Langrock, M.M. Fejer, M. Martinelli, V. Degiorgio, All-optical wavelength
conversion of a 100-Gbit/s polarization-multiplexed signal. Opt. Exp. 17(20), 1775817763
(2009)
9. M. Galili, J. Xu, H.C. Mulvad, L.K. Oxenlwe, A.T. Clausen, P. Jeppesen, B. Luther-Davies,
S. Madden, A. Rode, D.-Y. Choi, M. Pelusi, F. Luan, B.J. Eggleton, Breakthrough switching
speed with an all-optical chalcogenide glass chip: 640 Gbit/s demultiplexing, Opt. Exp. 17(4),
21822187 (2009)
10. T.D. Vo, M.D. Pelusi, J. Schroder, F. Luan, S.J. Madden, D.-Y. Choi, D.A.P. Bulla, B.
Luther-Davies, B.J. Eggleton, Simultaneous multi-impairment monitoring of 640 Gbit/s
signals using photonic chip based RF spectrum analyzer. Opt. Exp. 18(4), 39383945 (2010)
11. H. Ji, H. Hu, M. Galili, L.K. Oxenlwe, M. Pu, K. Yvind, J.M. Hvam, P. Jeppesen, Optical
waveform sampling and error-free demultiplexing of 1.28 Tbit/s serial data in a silicon
nanowire, in Presented at the OFC 2010, San Diego, CA, paper PDPC7
12. T.D. Vo, R. Pant, M.D. Pelusi, J. Schrder, D.-Y. Choi, S.K. Debbarma, S.J. Madden, B.
Luther-Davies, B.J. Eggleton, Photonic chip-based all-optical XOR gate for 40 and 160 Gbit/s
DPSK signals. Opt. Lett. 36, 710712 (2011)
13. L. Liu, R. Kumar, K. Huybrechts, T. Spuesens, G. Roelkens, E.-J. Geluk, T. de Vries,
P. Regreny, D. Van Thourhout, R. Baets, G. Morthier, An ultra-small, low-power, all-optical
flip-flop memory on a silicon chip. Nat. Photonics 4, 182187 (2010)
14. A. Trita, G. Mezsi, M. Sorel, G. Giuliani, All-optical toggle flip-flop based on monolithic
semiconductor ring laser. Phot. Tech. Lett. 26(1) (2014)
15. F. Scotti, G. Berrettini, G. Contestabile, A. Bogoni, A regenerative Variable optical buffer for
NRZ and RZ Packets. J. Lightw Technol. 30(9), 13661372 (2012)
16. S. Le Beux, Z. Li, C. Monat, X. Letartre, I. OConnor, Recongurable photonic switching:
towards all-optical FPGAs, in Proceedings of the 21st International Conference on VLSI-SoC
(2013)
17. H. Brahmi, G. Giannoulis, M. Menif, V. Katopodis, D. Kalavrouziotis, C. Kouloumentas,

P. Groumas, G. Kanakis, C. Stamatiadis, H. Avramopoulos, D. Erasme, On the fly all-optical
packet switching based on hybrid WDM/OCDMA labeling scheme. Opt. Commun. 312, 175
184 (2014)
18. J. Bannister, J. Touch, P. Kamath, A. Patel, An optical booster for internet routers, in
Proceedings of the 8th International Conference High Performance Computing (2001),
pp. 339413
19. R.W. Boyd, Nonlinear Optics, 2nd edn. (Academic, New York, 2003)
20. C. Langrock, S. Kumar, J.E. McGeehan, A.E. Willner, M.M. Fejer, All-optical signal
processing using _2 nonlinearities in guided-wave devices. J. Lightw. Technol. 24(7), 2579
2592 (2006)
21. G. Agrawal, Nonlinear Fiber Optics (Academic Press, New York, 2001)
22. E. Lazzeri, A. Malacarne, G. Serano, and A. Bogoni, Optical XOR for error detection and
coding of QPSK I and Q components in PPLN waveguide. 2258 PTL 24(24), 2258 (2012)
23. Y. Miyoshi, K. Ikeda, H. Tobioka, T. Inoue, S. Namiki, K. Kitayama, Ultrafast all-optical
logic gate using a nonlinear optical loop mirror based multi-periodic transfer function. Opt.
Exp. 16(4), 25702577 (2008)
24. G. Berrettini, A. Simi, A. Malacarne, A. Bogoni, L. Pot, Ultrafast integrable and
recongurable XNOR, AND, NOR, and NOT photonic logic gate. IEEE PTL 18(8) (2006)
25. Z. Li, G. Li, Ultrahigh-speed recongurable logic gates based on four-wave mixing in a
semiconductor optical amplier. IEEE Photonics Technol. Lett. 18(12) (2006)
26. P.P. Bavejaa, A.M. Kaplana, D.N. Maywarb, G.P. Agrawala, Pulse amplication in
semiconductor optical ampliers with ultrafast gain-recovery times, in Optical Components
and Materials VII Proceedings of SPIE, vol. 7598, 7598171 (2010)
27. P.A. Andersen et al., Wavelength conversion of a 40-Gb/s RZ-DPSK signal using four-wave
mixing in a dispersion-flattened highly nonlinear photonic crystal ber. IEEE Photonics
Technol. Lett. 17(9), 19081910 (2005)
28. A. Yariv, P. Yeh, Optical Waves in Crystals, vol. 5 (Wiley, New York, 1984)
29. A.C. Turner-Foster, M.A. Foster, J.S. Levy, C.B. Poitras, R. Salem, A.L. Gaeta, M. Lipson,
Ultrashort free-carrier lifetime in low-loss silicon nanowaveguides. Opt. Express 18, 3582
3591 (2010)
30. F. Li, T.D. Vo, C. Husko, M. Pelusi, D.-X. Xu, A. Densmore, R. Ma, S. Janz, B.J. Eggleton,
D.J. Moss, All-optical XOR logic gate for 40 Gb/s DPSK signals via FWM in a silicon
nanowire, in Proceedings of the IEEE Photonics Conference (2011), pp. 593594
31. J. Wang, J. Sun, Q. Sun, Single-PPLN-based simultaneous half-adder, half-subtracter, and OR
logic gate: proposal and simulation. Opt. Express 15(4), 16901699 (2007)
32. A. Bogoni, X. Wu, I. Fazal, A.E. Willner, 160 Gb/s time-domain channel extraction/ insertion
and all-optical logic operations exploiting a single PPLN waveguide. J. Lightw. Technol. 27,
42214227 (2009)
33. J. Wang, J.-Y. Yang, X. Wu, O.F. Yilmaz, S. R. Nuccio, A.E. Willner, 40-Gbaud/s (120-Gbit/
s) octal and 10-Gbaud/s (40-Gbit/s) hexadecimal simultaneous addition and subtraction using
8PSK/16PSK and highly nonlinear ber, in Proceedings of the Optical Fiber Communications
Conference (2011), pp. 13
34. J.G. Proakis, Digital Communications (McGraw-Hill, New York, 2000)
35. B. Moslehi, J.W. Goodman, M. Tur, H.J. Shaw, Fiber-optic lattice signal processing. Proc.
IEEE 72(7), 909930 (1984)
36. S. Khaleghi, O.F. Yilmaz, M.R. Chitgarha, M. Tur, N. Ahmed, S.R. Nuccio, I.M. Fazal, X.
Wu, M.W. Haney, C. Langrock, M.M. Fejer, A.E. Willner, High-speed correlation and
equalization using a continuously tunable all-optical tapped delay line. IEEE Photon. J. 4(4),
12201235 (2012)
37. M.R. Chitgarha, S. Khaleghi, O.F. Yilmaz, M. Tur, M.W. Haney, C. Langrock, M.M. Fejer,
A.E. Willner, Coherent correlator and equalizer using a recongurable all-optical tapped delay
line. Opt. Lett. 38, 22712273 (2013)
38. D.B. Hunter, R.A. Minasian, Programmable high-speed optical code recognition using bre
Bragg grating arrays. IEEE Electron. Lett. 35(5), 412414 (1999)
39. M.C. Hauer, J.E. McGeehan, S. Kumar, J.D. Touch, J. Bannister, E.R. Lyons, C.H. Lin, A.A.
Au, H.P. Lee, D.S. Starodubov, A.E. Willner, Optically assisted Internet routing using arrays
of novel dynamically recongurable FBG-based correlators. J. Lightw. Technol. 21(11),
27652778 (2003)
40. S. Frisken, G. Baxter, D. Abakoumov, H. Zhou, I. Clarke, S. Poole, Flexible and grid-less
wavelength selective switch using LCOS technology, in Proceedings of the Optical Fiber
Communications Conference (2011), Paper OTuM3
41. S. Khaleghi, M.R. Chitgarha, O.F. Yilmaz, M. Tur, M.W. Haney, C. Langrock, M.M. Fejer,
A.E. Willner, Experimental characterization of phase tuning using ne wavelength offsets in a
tunable complex-coefcient optical tapped-delay-line. Opt. Lett. 39, 735738 (2013)
42. S. Khaleghi, M.R. Chitgarha, O.F. Yilmaz, M. Tur, M.W. Haney, C. Langrock, M.M. Fejer,
A.E. Willner, Experimental performance of a fully tunable complex-coefcient optical FIR
lter using wavelength conversion and chromatic dispersion. Opt. Lett. 37, 34203422 (2012)
43. M.R. Chitgarha, S. Khaleghi, O.F. Yilmaz, M. Tur, M.W. Haney, C. Langrock, M.M. Fejer,
A.E. Willner, Demonstration of channel-spacing-tunable demultiplexing of optical orthogonal
frequency-division-multiplexed subcarriers utilizing recongurable all optical discrete Fourier
transform. Opt. Lett. 37, 39753977 (2012)
44. M.R. Chitgarha, S. Khaleghi, O.F. Yilmaz, J.-Y. Yang, A.E. Willner, Bit depth and sample
rate tunable digital to analog converter using conversion/dispersion based delays, in Presented
at the European Optical Communications, Geneva, Switzerland, 2011
45. M.S. Rasras, I. Kang, M. Dinu, J. Jaques, N. Dutta, A. Piccirilli, M.A. Cappuzzo, E.Y. Chen,
L.T. Gomez, A. Wong-Foy, S. Cabot, G.S. Johnson, L. Buhl, S.S. Patel, A programmable 8-
bit optical correlator lter for optical bit pattern recognition. IEEE Photon. Technol. Lett. 20
(9), 694696 (2008)
46. I. Kang, M. Rasras, M. Dinu, M. Cappuzzo, L.T. Gomez, Y.F. Chen, L. Buhl, S. Cabot, A.
Wong-Foy, S.S. Patel, C.R. Giles, N. Dutta, J. Jaques, A. Piccirilli, All-optical byte
recognition for 40-Gb/s phase-shift-keyed transmission using a planar-lightwave-circuit
passive correlator. IEEE Photon. Technol. Lett. 20(12), 10241026 (2008)
47. S. Khaleghi, M.R. Chitgarha, M. Ziyadi, W. Daab, A. Mohajerin-Ariaei, D. Rogawski, J.D.
Touch, M. Tur, C. Langrock, M.M. Fejer, A.E. Willner, Simultaneous and independent
processing of multiple input WDM data signals using a tunable optical tapped delay line. Opt.
Lett. 38, 42734276 (2013)
48. M.R. Chitgarha, M. Ziyadi, S. Khaleghi, A. Mohajerin-Ariaei, A. Almaiman, J.D. Touch, M.
Tur, C. Langrock, M.M. Fejer, A.E. Willner, Recongurable 2-D WDM optical tapped-delay-
line to correlate 20Gbaud QPSK data, in Presented at the European Conference Optical
Communication, Geneva, Switzerland, 2011
49. J. Armstrong, OFDM for optical communications. J. Lightw. Technol. 27(3), 189204 (2009)
50. D. Hillerkuss, R. Schmogrow, T. Schellinger, M. Jordan, M. Winter, G. Huber, T. Vallaitis, R.
Bonk, P. Kleinow, F. Frey, M. Roeger, S. Koenig, A. Ludwig, A. Marculescu, J. Li, M. Hoh, M.
Dreschmann, J. Meyer, S. Ben Ezra, N. Narkiss, B. Nebendahl, F. Parmigiani, P. Petropoulos, B.
Resan, A. Oehler, K. Weingarten, T. Ellermeyer, J. Lutz, M. Moeller, M. Huebner, J. Becker, C.
Koos, W. Freude, J. Leuthold, 26 Tbit s1 line-rate super-channel transmission utilizing all-
optical fast Fourier transform processing, Nat. Photon. 5, 364371 (2011)
51. J. Schrder, L.B. Du, M.A. Roelens, B. Eggleton, A.J. Lowery, Recongurable all-optical
discrete Fourier transform in a wavelength selective switch for optical OFDM demultiplexing,
in Proceedings of the Optical Fiber Communications Conference (2012), Paper OTh1G.6
Chapter 7
Wide-Band and Noise-Inhibited Signal
Manipulation in Dispersion-Engineered
Parametric Mixers
Bill P.-P. Kuo and Stojan Radic
Abstract Practical frequency generation and conversion requires highly efcient

parametric interaction across wide spectral band. To achieve high efciency, con-
ventional devices rely on optical enhancement in highly resonant cavity structures.
While optical cavity also inhibits noise generation and out-of-band amplication
during the parametric process, it inherently limits frequency tuning and any band
reconguration. Recognizing this fundamental impairment, we describe the new
class of cavity-free frequency generation in heterogeneous, traveling-wave para-
metric mixers. Driven by a continuous-wave seeds, these devices combine inher-
ently more stable lasers with distributed noise inhibition in dispersion-managed
parametric process. The operating principle, design methodology and performance
limits are outlined and discussed. The use of new technology is illustrated on
examples of signal multicasting and ultrafast channel processing.
7.1 Introduction
Recent developments of new optical devices capable of wide-band optical eld

generation and manipulation are motivated by three general trends. Firstly, the
adoption of coherent modulation schemes in terabit-per-second (Tbps) communi-
cation channels will soon require signal processing to be performed at rates exceeding
that of any feasible electronic backplane [1]. Secondly, communication and com-
putation systems relying on direct manipulation of photon states preclude, even in
principle, photon-electron conversion to perform subsequent electronic processing.
Finally, the need for general sensing has expanded outside the conventional bands
served by sensitive imagers (visible) and detectors (near-infraredNIR). Indeed, in
the immediate vicinity of NIR band, a short-wave IR (SWIR) offers a transparent
B.P.-P. Kuo S. Radic (&)

Department of Computer and Electrical Engineering, University of California San Diego,
9500 Gilman Drive, La Jolla, CA 92093-0407, USA
e-mail: sradic@ucsd.edu

186 B.P.-P. Kuo and S. Radic
atmospheric window, spurring the development of new LIDAR, communication and

imaging systems [2]. Unfortunately, the technological complement outside of stan-
dard (NIR) window does not include either source or detector with performance
comparable to the telecommunication devices. While SWIR laser and detector
(imager) can arguably be improved, basic physical mechanisms responsible for
1550 nm device performance is simply inferior in this range. Recognizing this lim-
itation, the alternative strategy can still use NIR sources and detectors by leveraging
parametric mixers capable of distant spectral translation for either emission or
reception of a coherent optical eld. In broader terms, the combined need for high
signal rates and direct, coherent eld control call for specialized set of functions such
as frequency referencing, beam switching and band mapping that do not rival elec-
tronics in its complexity. However, these are invariably fast and, in context of this
chapter, can be realized using specialty parametric mixing device.
In addition to this application-driven impetus, a set of important technological
advances has provided an additional motivation: it is now possible to design and
construct parametric devices using telecommunication derived components. In
contrast to conventional, crystalline parametric devices, the introduction and recent
maturing of high-connement ber has drastically reduced pump power required
for efcient frequency generation and light amplication [3]. It is no longer nec-
essary to consider high peak intensity sources such as mode-locked lasers (MLL) in
order to generate and convert signals over tens of THz: it is now possible to
accomplish the same using continuous-wave (CW), Sub-Watt scale telecommuni-
cation diodes, greatly reducing the complexity and reliability of general parametric
device. In addition to a great progress in reducing the loss of high connement ber
[4], advances in precise dispersion control and measurement [58] are of equal
importance. Indeed, combined advances in design, fabrication and characterization
of highly-nonlinear ber (HNLF) have led to the rst multistage, CW-pumped
parametric mixers possessing bandwidths in excess of 20 THz. New HNLF types
now offer the promise of highly stabilized local dispersion parameters [4, 911],
guaranteeing, for the rst time, spectral access to combined NIR/SWIR bands
approaching an octave span. While important for bandwidth engineering, localized
dispersion control also assumes critical roles in noise inhibition in distributed
parametric devices [12]. When combined with heterogeneous parametric design
[12, 13], new generation of mixers can provide spectral purity and bandwidth that is
simply not feasible by any alternative technology.
While it is not possible to accurately predict all mixer-enabled applications in the
future, one can state, with high condence, that coherent signal preprocessors,
advanced emitters and sensitive receivers have widespread near-future adoption. In
the rst of these categories (preprocessing), coherent sampling [14], digitization
[14] and spectral analysis [15] has led to performance level exceeding that of
comparable electronic architectures. In the second category (emitters) parametric
mixers have enabled frequency combs spanning combined C- and L-bands [16]
while possessing optical signal to noise ratio (OSNR) well beyond MLL technology
[17, 18]. Similarly, the new mixers were used to demonstrate fast tunable lasers
7 Wide-Band and Noise-Inhibited Signal Manipulation 187
operating over combined NIR/SWIR bands [19]. In the third category (receivers), it
is now possible to register few-photon packets at rates exceeding 100 GHz [20].
Consequently, this chapter aims to describe basic principles, impairment
mechanisms and the design principles used to generate and manipulate light over
wide bandwidth and in noise-inhibited manner.
7.2 Fundamentals of Parametric Mixers
Optical parametric mixing encompasses a collection of nonlinear processes in

which, the energies (as well as momenta) of the participating photons are re-dis-
tributed but not transferred to the medium. Due to the passive role of the medium,
parametric mixing is the only known mechanism that provides direct access to
participating photon characteristics (in terms of number, energy and polarization)
since the overall states of the photons are conserved in the process. This also means
parametric mixing can possess an arbitrarily-wide spectral coverage (bandwidth),
which is solely limited by the transparent window of and potential multi-photon
resonant processes in the medium. Furthermore, unlike stimulated-emission pro-
cesses in which the nite upper-state lifetime delays the mixing response, the
absence of (meta-)stable state in parametric mixing processes gives rise to its
ultrafast response.
Parametric mixers of interest in this chapter are based on four-photon mixing
(FPM) processes mediated by the third-order nonlinear response. A simple four-
photon mixer annihilates two photons, typically from a single or two frequency-
distinct pump waves, to create two photons that coincides the energy of the
remaining participating waves. This photon-exchange mechanism leads to a number
of important phenomena, including optical amplication, generation, and frequency
conversion [2123]. Specic wave dynamics of a parametric mixer are primarily
governed by the pump congurations and the corresponding phase-matching con-
ditions, as revealed in the following sections.
7.2.1 One-Pump Parametric Mixing
In the presence of a single pump wave possessing high intensity, the two photons
donated in the FPM process are both derived from this single source. The para-
metric mixing process driven by this pump conguration, as shown in the energy
diagram of Fig. 7.1, is known as one-pump or degenerate FPM.
In a one-pump FPM process, two photons from a single pump wave (where its
relevant variables are denoted by a subscript P) are annihilated to create a photon pair
known as signal and idler. Energy conservation demands that the sum of signal and
idler photon energies be equal to two pump energy quanta (hxS hxI 2hxP ),
which also means the frequencies of signal and idler must be symmetric with respect
Pump
- S
+ P
- I
Signal Idler
+ P
S P I
Fig. 7.1 Frequency conguration (left) and energy diagram (right) of one-pump parametric
mixing
to the pumps (xS xP xP xI X). In addition to energy conservation, the

momenta of the participating photons must also be matched in order to attain the
maximum mixing efciency [24]. The extent of momentum conservation, also known
as phase matching condition in parametric processes, is dictated by the propagation
constants of the pump, signal and idler waves [25]:
MI: Db bS bI 2bP 7:1
While the energy conservation is self-satised in absence of the idler at the input,
the distributed nature of parametric mixing implies that the momentum conservation
is being determined by the accumulated propagation conditions in the mixing
medium. Various levels of phase matching introduce different regimes of operation
distinguished by the power growth rate of signal and idler. Monotonic power growth
is feasible only when the linear phase mismatch Db is bound between 4cPP and
zero, where c and PP denote the nonlinear coefcient and pump power of the mixer
[25]. The signal experiences a gain scale to the square of the factor cPP L, referred to
as nonlinear gure of merits (NFOM) hereinafter, when the gain coefcient g is
considerably less than one. Meanwhile, zero phase mismatch, dened by the con-
dition Db 2cPP forces exponential signal power growth at a rate of exp(2cPP L)
[25]. The availability of highly-nonlinear bers with low attenuations and large
nonlinear coefcients enabled one-pump parametric ampliers providing up to 70-
dB gain [26]. This level of gain is considerably higher than those demonstrated by
erbium-doped ber ampliers (EDFA) [27], primarily due to the directivity of
parametric gain which circumvented lasing due to Rayleigh scattering [28].
Accompanying the signal amplication is the creation of an idler at an efciency
g G 1. Unlike amplication which mandates adequately small phase mis-
matching (4cPP \Db\0), idler creation occurs even with considerable phase
mismatch, albeit at reduced efciency. The idler created at a new wavelength carries
the signal eld envelope, but with complex conjugation [25]. This capability,
known as wavelength conversion, lays the foundation of many optical signal pro-
cessing functions that are unique to parametric mixers [29].
7.2.2 Two-Pump Parametric Mixing
In contrast to process driven by two photons with identical frequencies, a two-pump

interaction derives the source photons from two frequency-distinct waves. The mul-
tiple FPM processes seeded by the extra pump classies two-pump parametric mixers a
distinct type of devices, rather than an extension of the one-pump architecture.
A two-pump parametric mixer supports at least three basic FPM processes [30].
Originating from one-pump mixing, the modulation instability (MI) process anni-
hilates two photons from one of the two pumps, and creates a signal and idler
photon pair symmetrical to the source pump i.e. xS xI1 2xP1 (Fig. 7.2a).
Conversely, phase conjugation (PC) mixing results from the annihilation of pump
photons at distinct frequencies xP1 and xP2 , which generate signal and idler
photons at xS and xI2 (Fig. 7.2b). Energy conservation mandates the frequencies of
signal and idler be tied to those of the two pumps as xS xI2 xP1 xP2 . The
resulting frequency conguration of PC process is symmetric to the mean frequency
xm xP1 xP2 =2. While both MI and PC processes provide signal gain as a
result of photon addition at the signal frequency, the third process, namely Bragg
scattering (BS), attenuates signal by transferring its energy to the associated idler
(Fig. 7.2c). Energetically, BS process annihilates one pump photon (for instance,
xP1 ) and one signal photon to create a photon at another pump frequency xP2 and
an idler photon at xI3 xP1 xS xP2 . Note, however, that the amplication
and attenuation of a specic mode in the above description is ctitious for a
frequency conguration that allows all three processes to occur, since then the
idler of one process will be involved as signal in the other processes.
The coupling of three FPM processes in a two-pump parametric mixer dees a
simple model [31]. Nevertheless, the evolution of the four signal-idler sidebands
depends on the phase-matching conditions governing the individual FPM pro-
cesses. The phase-matching conditions for PC, MI and BS processes are as follows:
MI: Db bS bI1 2bP1 7:2a
(c)
(b)
- P1
(a)
- S + P2
+ P2
+ P1 -
MI PC BS S
- I3
- I2
+ S
+ + P1
P1 - I1
I1 P1 S I2 P2 I3
MI PC BS
Fig. 7.2 Frequency conguration (leftmost) and energy diagrams (right) showing a modulation
instability (MI), b phase conjugation (PC) and c Bragg scattering (BS) processes
PC: Db bS bI2 bP1 bP2 7:2b
BS: Db bS bP2 bI3 bP1 7:2c
The four-sideband dynamics can be solved in several notable phase-matching

regimes [31]. In absence of linear phase mismatches for all FPM processes, all
the sidebands exhibit power growth with quadratic scaling with pump powers and
propagation length. Meanwhile, perfect phase-matching in which the linear phase
mismatch fully compensates the nonlinear mismatch component leads to power
evolutions that approaches exponential growth with respect to distance and pump
power. These two regimes resemble the operation of a one-pump mixer with gain
coefcients g approaching zero and maximum respectively, albeit the signal power
growth is accompanied by creation of three idler sidebands in a two-pump mixer,
instead of only one in its one-pump counterpart. Inherent to the coherent nature of
FPM processes, the three idlers created in a two-pump mixer carry either identical
modulation of the input signal through BS process, or phase-conjugated copies
through MI and PC processes. The combination of three FPM processes extends the
simple wavelength converting function to wavelength multicasting, in which
multiple signal replicas of different wavelengths are generated [32].
7.2.3 Signal Processing Functions of Parametric Mixers
Apart from signal amplication, a unique capability of parametric mixers is the

transfer of modulation (information) to a new wavelength through idler generation,
namely, wavelength conversion. This attribute of parametric mixer operation is
sought for applications that include wavelength switching and routing [3234],
novel light generation [19, 35, 36] and pattern recognition [37]. A particularly
noteworthy phenomenon entailed by the conjugated wavelength conversion is
propagation effect reversal by phase conjugation [38, 39]. In principle, propagation
effects including dispersion and nonlinear phase rotation are fully reversible in
presence of a spectral inverter in the middle of the otherwise homogeneous link
[40]. While the temporal phase-conjugation offered by a parametric mixer does not
fully resemble the spectral inversion operation required for full reversal, the phase
conjugators were found effective for mitigating impairments for conventional
channels [39].
Departing from continuous-wave operation, parametric mixers driven by modu-
lated (pulsed) pumps offer an agile platform for time-domain signal manipulation
the quasi-instantaneous response inherent to parametric processes renders the per-
formance of signal processing in parametric mixers be time-invariant. In general,
resolving the FPM dynamics involving a pulsed pump require full consideration of
dispersion and frequency-dependent nonlinear terms in the couple-mode equations,
which will yield no closed-form solution [41]. In practice, however, parametric
mixers operating with pulsed pumps are typically constructed on nearly dispersion-
less medium in order to preserve the pump waveform and prevent excessive pulse
walk-off [42, 43]. Consequently, the gain model developed for continuous-wave
pumped parametric mixers are typically applicable to pulsed pumped counterparts,
except that the gain and conversion efciency follow the change in instantaneous
pump power. This class of parametric mixers is adopted for a number of applications
ranging from optical stroboscope [42], optical time-division multiplexed (OTDM)
signal switching [43], analog optical signal sampling [14], to more sophisticated
usage in OTDM-WDM conversion using time-frequency modulated pumps [44].
The presence of idler at the mixer input invokes a unique operating regime of
parametric mixerthe signal and idler couple and modify the dynamics in the
parametric mixer. Known as phase-sensitive (PS) operation, the gain (conversion
efciency) in the parametric mixer is now dependent on the initial phases of the
signal, idler, as well as the pump [45], where a mutual phase difference of 0 (or )
will amplify (or attenuate) the signal and idler. In PS amplication mode, the signal
and idler are theoretically devoid of excess noise due to the phase selectivity of the
process. This capability of amplication and manipulation of optical signals in
noise-less manner (i.e. with 0-dB noise gure) is unique to parametric mixers [45],
and has recently attracted great interest in various venues, such as optical link-reach
extension [46], signal regeneration [47], and wavelength multicasting [48]. Detailed
discussion of PS parametric mixing can be found in [49].
7.2.4 Effect of Chromatic Dispersion
Chromatic dispersion of a medium is characterized by the nonlinear relationship

between the phase constant and frequency. To parametric mixers, the presence of
chromatic dispersion changes the phase-matching conditions according to the fre-
quencies of the pumps, signal and resulting idlers, thereby affects the mixing ef-
ciency, bandwidth and noise performance of the mixer. In absence of chromatic
dispersion, phases of the participating waves are always balanced since the group
velocities remain constant. However, typical optical media possess nite chromatic
dispersion as a result of resonant (lossy) response of the material or the waveguide
[50]. The frequency-dependency of group delay d/d as a manifestation of
chromatic dispersion thus restricts the frequency range in which linear phase-
matching can occur. Such dependency, however, also allows perfect total phase
matching to happen by providing a nite linear phase mismatch to balance the
nonlinear phase shift terms for certain combinations of pump, signal (and idler)
frequencies. Consequently, chromatic dispersion signicantly shapes the spectral
response of parametric mixer, which would otherwise be virtually achromatic [25].
Chromatic dispersion engineering represents one of the most important aspects
in parametric mixer construction, since the intrinsic (material) dispersion charac-
teristics of many nonlinear optical materials are far from ideal for parametric
mixing. Standard single-mode ber (SSMF), for example, possesses a chromatic

dispersion in the lowest loss window at 1550 nm which allows phase-matched
mixing across a band narrower than 100 GHz [52]. This contrasts the 1000 THz
mixing window that the transparency window of fused silica would have provided
[53]. Alternation of the material dispersion often involves the use of waveguide
structures, in which waveguide dispersion is introduced to compensate for material
counterpart. Optical bers denote a particularly important class of waveguides in
which chromatic dispersion engineering is practiced. While dispersion-engineered
optical bers were traditionally sought after for mitigating signal propagation
impairments due to dispersion [54], the resulting dispersion-shifted waveguide
designs have also provided the low-dispersion environment suitable for wide-band
parametric mixing [55]. Indeed, rened waveguide structures have enhanced the
phase-matched mixing bandwidth due to precise control of higher-order dispersion
terms. In particular, fourth-order dispersion control in specially-tailored ber
waveguides, also known as highly-nonlinear bers, have enabled parametric mixing
over a near-octave bandwidth [10, 56]. Microstructure ber waveguides that
comprise photonic crystal or hollow guiding structures have further extended the
operating spectral region and range of parametric mixers to beyond the capability of
solid ber [57], by allowing guiding in novel optical materials [58, 59] and dis-
persion control in regions not feasible in solid ber waveguides [60].
Although waveguide cross-section design provides the vital dispersion control
for parametric mixing, the influence on dispersion by waveguide structure means
any deviation in the wavelength core size will exert signicant dispersion deviation
[6]. This effect signies challenges in producing repeatable mixing performance
from device to device, as well as in attaining the target performance due to the
deviation of local dispersion from the global dispersion [61]. In the next section, the
challenges and solutions for synthesizing dispersion-stable waveguides discussed in
depth.
7.3 Dispersion-Stable Waveguide Engineering

for Wide-Band Parametric Mixer Synthesis
Chromatic dispersion of the optical medium has profound effects on the efciency
as well as the attainable spectral bandwidth of a parametric mixer due to the phase
matching mandates [61]. While chromatic dispersion of an optical waveguide can
be arbitrarily tailored through waveguide geometry engineering, the local dispersive
behavior of the waveguide can deviates from its global characteristics by consid-
erable amount. The dependence of local dispersion for phase-matching, an unfor-
tunate attribute of distributed nature of typical parametric mixer, only permits the
dispersion to fluctuate within a small margin.
7.3.1 Wide-Band Parametric MixingThe Atomic-Scale

Challenge
Local dispersion fluctuation can be revealed from the mechanism of dispersion

tailoring by waveguide geometry design [54]. Waveguiding effect shifts the overall
dispersion from the material dispersion by modal area control [6264]. In general,
the modal spread of a guided optical eld increases with the wavelength, conse-
quently forcing a higher portion of the eld to the cladding region and decreasing
the effective index. This phenomenon gives rise to the wavelength dependency of
the effective index, and thereby to the shift in dispersion. Typical dispersion-shifted
bers utilize this wavelength-dependent guiding effect to create strong waveguide
dispersion for compensating the material dispersion [54]. Using a constricted core
with high index contrast, a relatively abrupt change in effective index will occur in
the proximity of a wavelength point which is determined by the core size. This
abrupt feature in the effective index curve induces sizable second-order dispersion
effect, which cancels out the large material dispersion in the region.
While the tight connement in the core can result in ideal dispersion-free
characteristics for wide-band parametric mixing [4], this waveguide geometry also
sensitizes the dispersion to geometry fluctuations since the dispersion shift is
controlled by the size of waveguide [6]. Microscopic geometry fluctuations are
inevitably introduced during waveguide fabrication, typically amount to 0.1 % of
the waveguide cross-sectional size [4]. These fluctuations stochastically deviate the
dispersion characteristics of the waveguide along its propagation axis. With a
typical highly-nonlinear ber (HNLF) that is commonly deployed as a parametric
mixing platform, the bers dispersion curve shown in Fig. 7.3 shifts by as much as
0.5 ps/nm/km for a 10-nm change to its 2 m core radius [6].
Even though the dispersion shift might be deemed negligible when compared to
the material dispersion (16 ps/nm/km), the extent is sufciently large to provoke
signicant penalty to parametric mixing. The quantitative impact of dispersion
fluctuations is revealed by a couple-mode model incorporating stochastic dispersion
variation [65]. As an illustration of dispersion tolerance for wide-band mixing, a
parametric amplier pumped at 1.55 m demands a dispersion accuracy better than
0.005 ps/nm/km in order to reach signal/idler pairs at 1.3 and 2.0 m [9]. This
requirement necessitates a radial control that is better than the length of a single
silicon-oxygen bond [9].
7.3.2 Post-Fabrication Dispersion Fluctuations Rectication
While enforcing atomic-scale waveguide fabrication accuracy is deemed imprac-

tical, the dispersion accuracy demand can possibly be addressed through resolving
the local dispersion characteristics and subsequently reversing the irregularities
Fig. 7.3 Effect of waveguide geometry variation on chromatic dispersion in HNLF. Core-size
fluctuations comparable to the elementary silica manifold are adequate to induce sizable dispersion
fluctuations
after fabrication. This method will require local dispersion measurement techniques
with fs/ps/km dispersion and meter-scale longitudinal resolutions [9, 61].
The evolution of FPM products depends on the aggregated local phase walk-off
between participating waves. While this presents a challenge in parametric device
construction, this mechanism also provides means to resolve the local dispersion
characteristics of the mixing medium.
7.3.2.1 Modied Optical Time-Domain Reflectometry
The rst class of methods relies on a modied optical time-domain reflectometry

(OTDR) [6668]. Owing to dispersion fluctuations along the waveguides propa-
gation axis, a pump-signal wavelength pair may experience phase-matching at a
certain region which consequently leads to amplication of the signal and creation
of an idler [67]. By detecting the photon echo of the pulsed signal produced by
Rayleigh scattering, the captured power evolution of the signal and/or created idler
along the waveguide can then infer the position of phase-matched region in the
waveguide. Subsequently, the complete dispersion map of the waveguide is gen-
erated by sweeping the pump and signal wavelengths in order to capture all possible
phase-matching wavelength combinations and their respective position in the
waveguide. The major drawback of this approach rests on its sensitivitydue to the
low Rayleigh scattering coefcients in typical waveguides (*40 dB for optical
bers), the detection of signal gain and idler generation typically requires long
signal pulse width (*s), as well as long averaging period [67]. The rst operating
requirement necessarily compromises the spatial resolution to 100-m level, while

the second imposes long measurement time for obtaining accurate results.
7.3.2.2 Space-Frequency Domain Retrieval
Alternatively, one may obtain the localized FPM behavior by resolving the eld
evolution from the spectral variation of wavelength conversion efciency [5]. With
two waves of similar power propagating in a nonlinear medium of interest, the
power of the FPM product is given by:
L 2
Z

PFPM
e jDbz/z
dz 7:3

0
The mean phase mismatch and its local fluctuations are denoted by the terms
Dbz and /z respectively. Particularly, if the mean phase mismatch is controlled
such that it spans a linear space by, for instance, changing the wavelength pair of
the two waves, the corresponding variation of FPM product will then resemble the
squared amplitude of the Fourier transform of the local phase mismatch evolution
exp[j/z]. In the spectral region close to the zero-dispersion wavelength (ZDW),
the linear phase-mismatch scaling can be performed by xing the wavelength
separation between the two input waves while the mean wavelength is being tuned
in linear manner, since the dispersion curve in this region is nearly proportional to
wavelength by its dispersion slope dD=dk.
By generating the power spectrum from the power of FPM products, the local
phase mismatch term /z is subsequently resolved by applying phase-retrieval
algorithms to the power spectrum of exp[j/z]. While a number of phase-retrieval
algorithm exists for problems of this kind [5, 69], the spatial resolution of this
approach can be estimated by regarding the mixing process as sampling of the
phase mismatch evolution exp[j/z] [5]. The spatial resolution, or equivalently,
the bandwidth of the sampler, is given by the maximum range of mean phase
mismatch Db one can introduce. Taking the tunable range of typical lasers into
consideration, the 100-nm tunable range will limit the spatial resolution to 40 m for
a HNLF with 0.02 ps/nm2/km dispersion slope. This resolution level remains
decient for capturing meter-scale fluctuations that are pertinent to wide-band
parametric mixing [61].
7.3.2.3 FPM Localization by Dispersive Pulse Walk-off
The third class of methods measures the local FPM evolution by utilizing spatial
walk-off between modulations on two distinct wavelengths [70]. The architecture
involves two pulsed sources at two tunable wavelengths, where one serves as the
pump and the other as the signal. The pump and signal pulses travel at different group
velocities in the dispersive medium and subsequently overlap (collide) at a certain
region of the medium, whereas the location of collision is controlled by the initial
delay between the pump and signal pulses. The collided pump and signal then create
an idler whose generation efciency is determined by the local dispersion (phase-
matching) properties of the medium. Consequently, the dispersion map of the med-
ium is aggregated by capturing the power of idler generated at various pump-signal
wavelength pairs and differential delays, which dene the dispersion value and
position of the dispersion map respectively. However, the reliance of the mediums
dispersion to localize pump-signal collision sets a compromise between the smallest
dispersion and the nest spatial resolution one can attain with this method. In essence,
the spatial resolution is proportional to the inverse of the dispersion, meaning that the
lower the dispersion, the coarser the spatial resolution will be. Therefore the adoption
of this approach is limited to highly dispersive waveguides such as standard single-
mode bers (SSMF), but is not applicable to low-dispersion media that are necessary
for wide-band parametric mixing.
7.3.2.4 Counter-Collision Method
Recognizing the compromise between dispersion and spatial resolution that plagues
conventional mapping approaches, a new approach was introduced in which the
spatial resolution is controlled by active means [6].
The key to achieving ne spatial resolution unrestricted by the dispersion of the
medium rests in local optical amplication. The architecture of the new approach,
as shown in Fig. 7.4, involves two wavelength-tunable pulse streams, referred to as
probe (P) and signal (S) hereinafter. The probe and signal pulses are both launched
into the waveguide-under-test at a low power, at which nonlinear effects are
Fig. 7.4 Schematic of Counter-Collision Method. a Low-power probe (P) and signal (S) are
launched on contra-directionally to the pump (); b, c collision at zC leads to amplication of the
probe; d idlers F are generated due to mixing between signal and amplied probe. e The
aggregated idler power evolution depicts the local phase matching conditions
negligible. Meanwhile, high-power pulses, denoted as pump pulses (), are

launched into the waveguide at the other end. The wavelength of the pump pulses is
controlled such that the pump pulse will create a Brillouin gain peak at the probe
wavelength, whereas the timing of the pump pulses controls the location of colli-
sion. When the forward propagating pulses collide with a pump pulse at location zC,
the probe pulse is selectively amplied through stimulated Brillouin scattering. The
probe pulse, now possessing signicantly higher power, interacts with the co-
propagating probe pulse and creates idler (F+ and F). The output power of the
idler is governed by the phase-matching condition from location zC to the output of
the waveguide L. Consequently, aggregation of idler power for various pump-gate
collision locations yield the phase mismatch with resolution only constrained by the
timing resolution of gate generation.
The combination of spatial and dispersion resolution gives rise to the new level
of mapping accuracythe scheme provided meter-scale length resolution and fs/
nm/km accuracy [6, 7]. This level of performance enabled measurement and sub-
sequent rectication of dispersion fluctuations in low-dispersion HNLF for the rst
time [6]. Using the dispersion map captured by a counter-collision setup, the dis-
persion fluctuations in the HNLFs were mitigated by splicing sections of similar
dispersion proles [74], or by applying strain to shift the local dispersion charac-
teristics [7, 8]. The reduction in dispersion fluctuations in the treated HNLFs was
manifested by the enhancement in both the gain and bandwidth of the parametric
amplier constructed upon [7, 71].
7.3.3 Waveguide Design Methods for Achieving Intrinsic

Dispersion Stability
Rather than post-fabrication correction, the required dispersion stability can

achieved by an alternative waveguide design that provides reduced dispersion
sensitivity to geometric fluctuations. This design-based approach offers an
unprecedented level of dispersion stability out of reach by post-fabrication recti-
cation, and eliminates the production that is limited by device-basis trimming.
7.3.3.1 Numerical Optimization-Based Design Approach
Early attempts to address dispersion stability issues in ber waveguide relied on

numerical optimization to search for an optimal refractive index prole. Particularly
noteworthy achievements in optimization-based ber design was reported by
Gabrielli et al., in which a genetic algorithm was used to search for refractive index
proles [72]. Using a tness function that comprised zero-dispersion wavelength
and higher-order dispersion terms (4 and 6) for the genetic evolution, the resulting
multi-layer index proles provided control over higher-order dispersion at up to the
sixth order, and improvement in zero-dispersion wavelength stability. However, the

extensive computation associated with repeated mode solving for evaluating pos-
sible index proles forbade consideration on certain important parameters (e.g.
nonlinear coefcient and 2 fluctuations). Consequently the optimization converged
to solutions that possessed low nonlinearity (34 W1 km1), and counterintu-
itively, high dispersion fluctuations (2 ps/nm/km/%) due to elevated dispersion
slope.
7.3.3.2 Semi-theoretic Design Approach
Apart from optimization-based design method, a semi-theoretical approach has

recently been developed for dispersion-stable waveguide synthesis [9]. This method
stemmed from the observation that the single-layered core structure deployed in
typical high-connement waveguides produces two distinct responses in waveguide
dispersion when the core is scaled radially. In conventional high-connement
waveguides such as HNLF, the eld connement and dispersion control are pro-
vided both by a single delta-like core structure. As shown in Fig. 7.5, scaling the
radius of the single-layer core causes the V-shaped normalized waveguide disper-
sion curve to shift laterally, thereby changing the dispersion in the low and high
frequency regions in opposite directions [9].
Recognizing this opposing behavior, a dual-layer core structure was developed to
leverage the dispersion shifts in two distinct core layers for overall dispersion change
cancellation. Instead of relying on one core layer to control eld connement and
dispersion, the additional layer in the new structure shown in Fig. 7.6 provides an
additional degree of freedom to manipulate the wavelength-dependent behavior of the
(a) 2.0m 1.2m (b)

2
Normalized Dispersion -V d (Vb)/dV
0
+3.4%
-0.5
V1
V2
2 2
-1
-1.5
V1 = 3.67
-2 (c)
-2.5
-3
1 2 3 4 5 6 7 -6.4%
Normalized Frequency V
V2 = 1.28
Fig. 7.5 Normalized waveguide dispersion of a conventional HNLF. Sub-graphs (b) and (c) show
the dispersion shift in the high frequency and low frequency regions respectively
n1
n2
nClad n3
0 r
a1 a2 a3
Fig. 7.6 Refractive index prole of the dispersion-stabilized waveguide design. Radii a1 and a2
denote the size of the inner and outer cores. Trench between {a2, a3} provides further control of
guiding condition
0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
0.2
(a)
0
2.0 m 1.2 m
-0.2
Normalized Dispersion -V 2d 2(Vb)/dV 2
-0.4
-0.6
-0.8
-1
0 0
-0.5 -0.5
-1 -1
-1.5 -1.5
(b) (c)
-2 -2
1 2 3 4 5 1 2 3 4 5
Fig. 7.7 Normalized waveguide dispersion of the new waveguide. a Optimal outer core
parameters results in complete cancellation of dispersion fluctuations; b, c outer core size and
refractive index affects overall waveguide dispersion in different manner
central core, thus enabling control of dispersion stability. The origin of dispersion
stability in the new core design is revealed in Fig. 7.7. In the waveguide dispersion
curve shown in Fig. 7.7a, the dual minima characteristic reflects two distinct guiding
regimes enacted by the dual-layer core. At low frequencies, the spread of the optical
eld over both cores implies the eld is guided by both cores and cladded by the
cladding layers. At high frequencies, on the other hand, the optical eld is primarily
conned to the central core, meaning that the outer core layer now partially behaves
as a cladding layer. The low-contrast guiding provided by the inner-outer core
structure for high-frequency elds subsequently introduces the additional dip in the
high-frequency regime. With proper selection of the outer core index and radius, the
opposite dispersion shift exerted by the two guiding regimes can eliminate the dis-
persion change in the wide-range saddle region, in which a parametric mixing can
operate. As a result, the overall dispersion of the waveguide remains stable in the
presence of geometrical fluctuations. HNLF as an example, the ber dispersion shift
due to core size scaling is reduced by 100-fold when compared to a conventional
single-core HNLF design.
HNLFs using the new waveguide design had demonstrated improved dispersion
stability while maintaining high eld connement. Experimental characterization
shows that the dispersion fluctuations were reduced by 92 % [10]. The improved
dispersion stability consequently led to enhancement in parametric mixing ef-
ciency, manifested by an increased parametric gain and spectral reach of the
amplier. Specically, the new HNLFs reduced the threshold pump power by 60 %
for ber parametric oscillators operating in the short-wave infrared (SWIR) band
[10], and allowed, for the rst time, tunable lasing from 1.2 to 2.15 m in a ber
parametric oscillator (Fig. 7.8a) [73]. In addition to parametric mixing efciency
and bandwidth enhancement, dispersion resilient to core deformation in the new
HNLFs also allowed stimulated Brillouin scattering (SBS) management by ber
stretching without compromising phase-matching (Fig. 7.8b) [11]. This attribute
enabled the rst continuous-wave (CW) pumped coherent and modulation-agnostic
wavelength conversion across 700 nm [11, 74].
(a) 20 (b) 4
10 900nm
Dispersion [ps/nm/km]
3
0
2
Power [dBm]
-10
1
-20
0
-30
-1
-40
-2 Untreated
-50
-3 Stretched
-60
-4
-70 1540 1560 1580 1600 1620 1640
1200 1400 1600 1800 2000
Wavelength [nm]
Wavelength [nm]
Fig. 7.8 a Tuning spectrum of a parametric oscillator constructed with dispersion-stabilized

HNLFs; b dispersion curves of dispersion-stabilized HNLFs under various level of longitudinal
stretch
7.4 Inhomogeneous Dispersion Engineering

for Noise-Inhibited Parametric Mixing
In a two-pump parametric mixer, combination of mixing processes replicate a

single input to multiple frequency-distinct locations. In addition to the pump-signal
interactions, the presence of two strong pump waves intuitively leads to efcient
pump-pump mixing which produces additional intense waves. These newly created
secondary pumps further duplicate the mixing process driven by the primary pumps
and consequently create more replica of the input. Known as self-seeded or cas-
caded parametric mixing, this class of cascaded interaction is actively sought in
recent years for low-noise signal processing as well as other advanced time-fre-
quency domain processing [13, 14, 43, 48, 49].
The conversion efciency, spectral reach and noise performances of a self-
seeded mixer are primarily determined by the secondary pump generation pro-
cesses. The following section reviews the principle of self-seeded mixing and
derives its limits.
7.4.1 Self-seeded Two-Pump Parametric

MixingHomogeneous Limit
In general, the wide-band and complicated generation dynamics of secondary

pumps require full account of the nonlinear interplay between FPM, dispersion and
loss using nonlinear Schrodinger models. However, a simple model can be derived
by knowing the optimal environment for homogeneous self-seeded mixing
[13]nonlinear media characterized by negligible chromatic dispersion are chosen
to facilitate bandwidth-uninhibited interaction between pumps [13, 43]. Under this
condition, the pump-pump mixing process is reduced to mere nonlinear phase
modulation due to the sinusoidal power evolution of the pump eld in time [13].
The number of secondary pumps created is consequently predicted by a modied
Carsons rule for phase modulation, in which the empirical law N 2dm 1e
states the tone count scales proportionally to the nonlinear gure of merit (NFOM)
m cPL [13]. This rule implies an important consequencethe number of tones
obtainable in a homogeneous mixer is pathetically bounded to a small count by
nonlinear power limits, dened by various mechanisms including Brillouin scat-
tering and material absorption [75, 76]. As an illustration, the state-of-the-art silica
HNLFs possess a Brillouin-scattering bounded NFOM of 0.5 [75], meaning the
maximum number of tone creatable is lower than 4. While the Brillouin-scattering
strength can be suppressed by various means [7779], the level of suppression is
inadequate to enable generation beyond four tones [13, 43]. In addition to NFOM
bound, high pump power demand for secondary pump generation will lead to
initiation of multiple noise generation processes, thereby severely degrades the
delity of the secondary pumps and the ensuing photon mixing processes [80].
7.4.2 Inhomogeneous Dispersion Engineering
Recognizing the limits of homogeneous mixer for cascaded parametric mixing, a

new mixer design based on stage-wise dispersion engineering was devised to
reduce the necessary pump power [11, 12]. The new mixer design, as shown in
Fig. 7.9, consists of at least two elementary blocksa seeding and a mixing stage
[11]. Two input pump waves are launched to the rst seeding stage, which com-
prises a section of nonlinear waveguide and followed by a dispersive element. In
this rst seeding stage, the pump waves create new tones through FPM in the
nonlinear waveguide section. In the time domain, the nonlinear interaction of
pumps is equivalent to phase modulation due to SPM, thus resulting in frequency
chirping of the pump eld. Consequently when the pump eld propagates in the
dispersion section with anomalous dispersion, the frequency-chirped pump eld
will experience temporal compression of its originally sinusoidal envelope to pulses
with high peak power [81]. Subsequent to the seeding stage, efcient cascaded
mixing occurs in the mixing stage composed of a nonlinear waveguide. Since
nonlinear interaction strength increases at higher peak power and spectral span [82],
the compressed pump eld from the rst stage experiences extensive SPM and
results in massive spectral broadening. The spectral broadening of the pump pulses
is manifested as generation of optical tones, with their frequency spacing strictly
following the input pump frequency separation.
The efciency of cascaded mixing and delity of the generated tones are solely
determined by the dispersion management in each of the mixers constituent stages.
The following sections will dene the dispersion criteria for optical mixing.
7.4.2.1 Inhomogeneous Mixer Design ConsiderationsSeeding Stage
The seeding stage converts relatively low intensity pump eld into high peak power
pulses. Proper dispersion and nonlinearity managements are required in this stage to
preserve tone delity, as well as to attain the ideal temporal and spectral pulse
characteristics for mixing in subsequent stages.
In the rst stage, the ideal mixer should comprise an achromatic nonlinear
section characterized by the absence of chromatic dispersion, which is followed by
a dispersive section with appropriate chromatic dispersion to convert the chirped
Pulse Compression Stage Mixing Stage

NLWG Disp. NLWG

t t t
Fig. 7.9 Multi-stage mixer architecture for achieving massive cascaded mixing. NLWG and Disp
denote nonlinear waveguides and dispersive element respectively
eld into transform-limited pulses. Following the derivation in [13], it is found that
the phase prole of the chirped eld obtained at the output of the rst nonlinear
section is largely quadratic. Consequently, a dispersive element providing appro-
priate negative second-order (anomalous) dispersion will be adequate to remove the
quadratic nature of the spectral phase prole, thereby converting the eld into
transform-limited pulses. In practice, the need for anomalous dispersion can be
fullled by standard SMFs. However, the spectral span at high phase-modulation
depth m becomes excessively wide, so that the higher-order dispersion terms of
standard SMFs will cause considerable pulse shape deviation from its transform-
limited form [83]. Nevertheless, the dispersion provided by SMFs can adequately
satisfy the need in the seeding (pulse compression) stage, over the practical oper-
ating condition constrained by SBS [75].
7.4.2.2 Inhomogeneous Mixer Design ConsiderationsMixing Stage
Subsequent wide-band generation in the second stage mandates a considerably more

stringent dispersion control in the nonlinear medium [12]. Practical mixing stage
construction deploys dispersion-flattened bers to provide low-dispersion environ-
ment that enables wide-band mixing. Indeed, the combination of intense peak power
in the mixing stage and complex dispersion characteristics of a dispersion-flattened
waveguide guarantees rich nonlinear dynamics and tight dispersion requirements
[12]. Depicted by the nonlinear Schrodinger model shown in Fig. 7.10, a miniscule
dispersion deviation of 0.1 ps/nm/km is adequate to overturn the tone delity pos-
sessed by a properly dispersion managed mixer. The spectral collapse observed in an
anomalous dispersion as low as 0.1 ps/nm/km is attributed to the perturbation of
Fig. 7.10 Output spectra of the mixing stage with anomalous (top) and normal (bottom)
dispersion
solitons by modulation instability [84, 85]. With the extreme power built-up by
soliton formation, vacuum fluctuations are vigorously amplied which consequently
leads to destruction of the soliton train [85]. In sharp contrast, a mixing stage with
strictly normal dispersion (Dpeak 0) produces pristine tones characterized by high
carrier-to-noise ratio. Perhaps more importantly, equalized tone power distribution is
possible only in the normal dispersion regime.
The low-noise, power-equalized tone generation is attributed to the existence of
similariton-like asymptotic state for pulse propagation in medium possessing nor-
mal parabolic dispersion prole [86]. Figure 7.11 shows the evolution of the
Gaussian pulse in the presence of only 4th-order dispersion, which emulates
the nonlinear dynamics in an optimal mixing stage shown in Fig. 7.10. Following
the initial dispersion-free propagation, the Gaussian pulse gradually evolves into an
arch-shape, featured with steep edges and a near-triangular top portion. Known as
optical shock-wave formation, the steepening of edges is responsible for the rapid
spectral broadening before the normalized position 40 due to Kerr-induced fre-
quency chirping [87]. The onset of wave breaking at 50 marks the transition of
pulse propagation regime to asymptotic (steady) state, as the pulse edges start to
collapse as a result of the continued action of the shear force by the chirp and
dispersion [88]. Further propagation beyond this point merely leads to pulse con-
verging into a rectangular shape. The rectangular portion of the pulse contains a
linear frequency chirp, of which the spectral width of the pulse equate the chirp
Fig. 7.11 Nonlinear evolution of a Gaussian pulse in the time and spectral domains. The position,
time and frequency are normalized to the nonlinear coefcient , pulse peak power P0 and
Gaussian pulse width T0. Simulation parameters: cP0 1 m1 ; zero second- and third-order
dispersion, and a fourth-order dispersion b4 b4 =cP0 T04 2:56 106
extent [87]. The mechanisms for low-noise and spectrally flattened generation
become apparent from this study. First, convergence to an asymptotic state implies
the propagation dynamics is immune to noise perturbations, contrary to the unstable
solitonic regime mediated by anomalous dispersion [85]. Second, the fourth-order
dispersion that produces parabolic dispersion prole forces the pulse to converge
into rectangular shape. The flat-top shape of the chirped pulse means the frequency
components of the pulse are equally distributed across its span, thus giving rise to
the equalized spectral shape.
The combination of multi-stage design and precise dispersion engineering for
low-noise mixing allowed generation of a wide-band, low-noise frequency comb
using two low-power continuous-wave pumps [11, 12]. Creation of high-OSNR
optical frequency tones further enables a plethora of applications in coherent
communications and generalized signal processing, as discussed in the next section.
7.4.3 Applications
7.4.3.1 Coherent Carrier Generation
The adoption of coherent modulation has radically changed the landscape of optical
transport. Full-eld recovery of signals by interdyne/intradyne detection using
optical local oscillators alleviates the need in physical mitigation of channel
impairments comprising chromatic dispersion, polarization mode dispersion (PMD)
and nonlinear penalties [8993]. By carrying information on both amplitude and
phase quadratures of a light-wave carrier using multi-level modulations, coherent
optical transmission allows expansion of conventional, direct-detection based sys-
tem capacity as a result of improved spectral efciency [94]. On the other hand,
however, coherent transmission challenges conventional optical carrier generation
approaches due to strict carrier-phase and frequency stability requirements [95]. As
an illustration of the challenge, transporting a 100 Gbit/s, 16QAM channel requires
an optical carrier with a linewidth below 30 kHz to guarantee optimal link budget
[95]. Distributed feedback (DFB) diode laser that is a mainstay of conventional
direct-detection systems, however, typically possess a linewidth exceeding 200 kHz
[96]. Worse, even if one is allowed to construct a transmission system solely based
on narrow linewidth sources, the emission frequencies of these individual sources
wander in random manner and, thus, requires power-consuming frequency control
on per-channel basis.
The low-noise optical frequency comb generated by multi-stage parametric
mixer provides a practical solution in phase noise and frequency stability man-
agement for high capacity optical links [16, 17, 94]. Using a three-stage mixer, an
optical frequency comb comprising 120 tones on 100-GHz grid or 1500-tones ultra-
dense WDM comb at 12.5 spacing were generated using a total optical pump of less
than 1 W. The mixer was seeded by a single narrow-linewidth source, where its
coherence was transferred to all generated tones characterized by a maximum
Fig. 7.12 a Spectra of comb spaced by 100 GHz; b sensitivity of coherent channels carried by the
comb at up to 64QAM
linewidth of 25 kHz. The 100-GHz comb in Fig. 7.12a shows the tones possessed
power of 5 dBm average and optical signal-to-noise ratio (OSNR) beyond 50 dB.
The low-noise comb consequently permitted transport of complex modulation at
performance near the theoretical bound (Fig. 7.12b). Low-noise parametric comb
generation further allowed coherent transport at an aggregated rate of 15.6 Tbps,
using complex modulations of up to 64QAM [94]. The absence of frequency
control for each carrier in parametric frequency comb rendered considerable
advantage in power consumption over individual emitter architecturewhile the
parametric comb generator consumed a total of 40 W, an equivalent emitter bank
easily consumes ten times higher power due to active temperature control.
7.4.3.2 High-Count Wavelength Multicasting
Due to the presence of three parametric mixing modes, a two-pump parametric

mixer is capable of delivering at least replicas of the input signal at three distinct
wavelength locations, resulting in a pair of side-bands adjacent to each pump
carrying modulations identical to that of the signal [98]. With the creation of
secondary pumps in a multi-stage mixer, the new pumps will duplicate the side-
bands of the primary pumps and thereby, creating more signal copies as the number
of pump tones grows [98]. This mixer architecture thus allows generation of
massive number of spectrally-distinct replicas which preserve both amplitude and
phase of the signal, and align to a frequency grid that is strictly dened by two
pump seeds. Such operation enables novel spectral analysis and time-frequency
domain signal processing [14, 15], alongside with the classical usages in data
stream multicasting [99]. Further description on spectral analysis and signal pro-
cessing applications will be given in the next section.
The concept of signal replication in multi-stage mixer has been demonstrated
with both amplitude and phase modulated signals [11]. Shown in Fig. 7.13, a 10-
dBm input signal to a mixer comprising two nonlinear stages produced nearly 70
copies, with positive conversion gain to a majority of the duplicates. The copies
were characterized by a nearly identical delity to the original signal, as depicted by
a 0.2 and 0 dB average sensitivity penalties for amplitude- and phase-modulated
inputs, respectively.
In additional to the classical, phase-insensitive mixing modality, the multi-stage
parametric mixer architecture further allows non-classical, noise-less interaction by
phase-sensitive mixing. Typical phase-insensitive optical amplication processes
cause a noise gure of at least 3 dB as a consequence of quantum noise ampli-
cation in both in-phase and out-of-phase signal quadratures [100]. Parametric
mixers are inherently phase-sensitive devices as a consequence of phase
Fig. 7.13 Multicast conversion efciencies in a two-stage dual-pump parametric mixer. The right
column shows the eye diagrams of the back-to-back signal and its replica on the short-wavelength
edge of the conversion spectrum
preservation in parametric processes. When operating in phase-sensitive regime by,

for instance, injection of signal and idler at the input or allowing signal and idler
frequencies to overlap, an ideal parametric mixer produces no excess noise since
only the in-phase component of the signal receives amplication, thereby resulting
in a zero noise gure.
Multi-stage parametric mixer extends the conventional phase-sensitive ampli-
cation concepts by generation of spectrally-distinct copies with no additional noise, in
addition to optical amplication localized to the injected input modes [48]. Using a
two-stage design with strictly normal dispersion in the mixing stage, the wavelength
multicasting mixer with near 0-dB noise gure managed to distribute signals to more
than 12 separate wavelength channels from its four-wavelength inputs [48]. Indeed,
the noise-less multicasting operation provided performance elevation over conven-
tional multicasting device comprising of power-splitting architecture. In a direct
comparison between the receiver sensitivity metrics of a phase-sensitive multicasting
mixer and power-splitting (attenuating) device, the phase-sensitive mixer improved
the overall receiver sensitivity by 11 dB, equivalent to at least three-fold increase in
copy counts in power-splitting architecture. The phase-sensitive multicasting archi-
tecture consequently provides a compelling means to extend conventional passive
optical network (PON) and generic signal processor performance, in which signal
replication plays critical roles [49].
7.4.3.3 Hybrid Time-Frequency Domain Signal Processing

and Recognition
The ability to produce massive count of signal replicas in parametric mixers permits
extensive parallelism for signal processing. This allows, for the rst time, optical
signal processors matching the of high-speed (THz-wide) channel.
Two general classes of signal processors relying on parametric signal replication
have been demonstrated to date. The rst kind enables real-time acquisition and
data conversion of fast analog or digital signals by parallelized sampling operations,
known as copy-and-sample-all (CASA) architecture [14]. This photonic-aided
analog-to-digital conversion (ADC) processing scheme, as shown in Fig. 7.14,
comprises two parametric mixers to perform parallel (flash) sampling. The rst
mixer is driven by two pumps to multicast the input signal into a number of color-
distinct copies. The copies are subsequently sampled by the second parametric
mixer driven by a pulsed pump, in which all copies at distinct frequencies mix with
the pump pulses to create corresponding idlers. An array of electronic digitizers
receives the idlers carrying temporal snapshots of the copies at a rate equal to the
pump repetition frequency, thus producing the digital discrete-time equivalence of
the input signal. The maximum signal bandwidth is ultimately tied to the number of
copies at which the mixer chain can generate and sample. While single-stage mixers
generating ten copies have satisfactorily fullled the demand for 40 GHz capture
[14], reception of THz-wide signals will require a copy-count in the order of
Fig. 7.14 Architectural schematic of CASA photonic processor
hundreds since commercial converter chips typically possess bandwidths of

10 GHz, thus rendering the use of multi-stage mixers mandatory.
Frequency traceability of signal replicas in parametric multicasters creates a new
signal-processing regime, in which signal manipulation occurs in the frequency
domain. This second class of signal processors, known as channelizer, granulates
signal contents into sufciently narrow frequency bands that permit low-latency,
low-rate signal recognition in the (electronic) reception back plane [15]. The gen-
eral concept illustrated in Fig. 7.15a utilizes spectral replication function of a two-
pump mixer to create a set of signal copies falling on a known periodic frequency
grid. Frequency-domain slices are subsequently extracted out of each copy by
resonant ltering [101] or digital lter bands synthesized by coherent eld capture
and digital signal processing [15]. The spectral positions of the slices relative to
their originating copies are determined jointly by the mixer pump separation and
free-spectral range of the lter. Consequently, reception of these spectrally-
constricted slices allows full reconstruction of the original signal using band-
restricted receivers. Beyond signal reception, channelizers naturally support
arbitrary channel response synthesis and, perhaps more importantly, instantaneous
spectral analysis. Conventional spectral analysis relies on frequency-scanned het-
erodyning or discrete Fourier transform (DFT), but both of these mechanisms
constrain the spectral acquisition rate due to fundamental limits in frequency source
tuning and digital computations [102]. On the other hand, single-step frequency
decomposition in a photonic channelizer is not limited by the aforementioned
barriers, thereby allowing acquisition rate reaching the fundamental uncertainty
bound [103]. Indeed, the advantages of photonic signal channelization would have
remained elusive without a scalable generation capacity for spectrally separated yet
Fig. 7.15 a Photonic signal channelizer principle; spectrum and spectrogram of a 100-GHz-wide
signal captured by a photonic channelizer are shown in (b) and (c) respectively
disciplined signal replicas, the ability uniquely provided by multi-stage mixers.

Validated in [15], a multi-stage mixer capable of creating 18 copies enabled spectral
analysis across 100 GHz in microsecond refresh rates. The high-speed, high
spectral resolution acquisition had made capture of fast but rare events possible
(Fig. 7.15b, c).
7.5 Conclusions
This chapter reviewed the physics, design practice and applications of wide-band
signal processors realized dispersion-engineered parametric mixers. While the
ubiquity of third-order nonlinear response implies that parametric mixing is feasible
in a plethora of nonlinear optical platforms, tight dispersion specications driven by
phase-matching conditions renders many of those unsuitable for signal processing
purposes when efciency and noise performance metrics are considered. Recog-
nizing the fundamental challenges in attaining the dispersion requirements for
synthesizing low-noise and efcient photon mixers, two classes of dispersion
engineering schemes were reviewed.
Based on waveguide cross-section geometry control, the rst class of dispersion
engineering techniques mitigates the dispersion fluctuations that impacts phase-
matching stability over an extended nonlinear interaction length. Chromatic disper-
sions of waveguides are inevitably subject to perturbation due to geometry variation
in atomic scales introduced during the fabrication processes. Instead of demanding
unphysical fabrication accuracy, the miniscule dispersion fluctuations can be mea-
sured with meter-level position resolution, thereby enabling post-fabrication reversal.
Alongside with post-fabrication corrections, dispersion stability can be guaranteed by
design. New waveguide design methodologies have now reduced dispersion fluc-
tuations level due to size variation by hundred fold, thus allowing nearly perfect
phase-matched parametric amplication and wavelength conversion over 900 nm for
the rst time.
Besides homogeneous dispersion control, multi-section dispersion and nonlin-
earity engineering provides a new two-pump parametric mixing regime in which
massive frequency tone generation is allowed. While homogeneous mixer prohibits
extensive pump-pump mixing due to the power demanded for this interaction type,
a tailored cascade of nonlinear and dispersive lines can scale the mixing efciency
by orders of magnitude due to temporal compression of pump elds. The conse-
quent massive generation of secondary pumps enabled a plethora of signal pro-
cessing applications, comprising of coherent transport, high-count spectral
replication, ultra-fast signal reception and instantaneous spectral analysis.
In summary, dispersion engineering techniques have redened the conventional
notions of inefciency and functional deciency commonly associated with para-
metric mixer technology, providing the viable path for bridging high-speed pho-
tonic signals with complex electronic processing. Further advances in dispersion
engineering, perhaps extension to material platforms other than silica, are expected
and will ultimately allow direct photon access in arbitrary spectral regions and
seamless interface with large-scale electronic backplane.
References
1. G. Raybon, A. Adamiecki, P.J. Winzer, S. Randel, L. Salamanca, A. Konczykowska, F. Jorge,

J.-Y. Dupuy, L.L. Buhl, S. Chandrashekhar, C. Xie, S. Draving, M. Grove, K. Rush,
R. Urbanke, High symbol rate coherent optical transmission systems: 80 and 107 GBaud.
2. M. Ebrahim-Zadeh, I.T. Sorokina (eds.), Mid-Infrared Coherent Sources and Applications
(Springer, Berlin, 2007)
3. S. Radic, C.J. McKinstrie, Optical amplication and signal processing in highly nonlinear
optical ber. IEICE Trans. Electron E88-C, 859869 (2005)
4. M. Hirano, T. Nakanishi, T. Okuno, M. Onishi, Silica-based highly nonlinear ber and their
application. IEEE J. Select. Topics Quant. Electron. 15, 103113 (2009)
5. I. Brener, P.P. Mitra, D.D. Lee, D.J. Thomson, High-resolution zero-dispersion wavelength
mapping in single-mode ber. Opt. Lett. 23, 15201522 (1998)
6. E. Myslivets, N. Alic, J.R. Windmiller, S. Radic, A new class of high-resolution
measurements of arbitrary-dispersion bers: localization of four-photon mixing process.
7. E. Myslivets, C. Lundstrm, J.M. Aparicio, S. Moro, A.O.J. Wiberg, C.-S. Brs, N. Alic,
P.A. Andrekson, S. Radic, Spatial equalization of zero-dispersion wavelength proles in
nonlinear bers. IEEE Photon. Technol. Lett. 21, 18071809 (2009)
8. M. Takahashi, M. Tadakuma, T. Yagi, Dispersion and Brillouin managed HNLFs by strain
control techniques. J. Lightwave Technol. 28, 5964 (2010)
9. B.P.-P. Kuo, S. Radic, Highly nonlinear ber with dispersive characteristics invariant to
fabrication fluctuations. Opt. Express 20, 77167725 (2012)
10. B.P.-P. Kuo, J.M. Fini, L. Grner-Nielsen, S. Radic, Dispersion-stabilized highly-nonlinear
ber for wideband parametric mixer synthesis. Opt. Express 20, 1861118619 (2012)
11. B.P.-P. Kuo, M. Hirano, S. Radic, Continuous-wave, short-wavelength infrared mixer using
dispersion stabilized highly-nonlinear ber. Opt. Express 20, 1842218431 (2012)
12. E. Myslivets, B.P.-P. Kuo, N. Alic, S. Radic, Generation of wideband frequency combs by
continuous-wave seeding of multistage mixers with synthesized dispersion. Opt. Express 20,
33313344 (2012)
13. B.P.-P. Kuo, E. Myslivets, N. Alic, S. Radic, Wavelength multicasting via frequency comb
generation in a bandwidth-enhanced ber optical parametric mixer. J. Lightwave Technol.
29, 35153522 (2011)
14. A.O.J. Wiberg, L. Liu, E. Myslivets, B.P.-P. Kuo, N. Alic, S. Radic, Demonstration of
40 GHz analog-to-digital conversion using copy-and-sample-all parametric processing, in
Proc. OFC/NFOEC 2012, paper OW3C.2
15. A.O.J. Wiberg, D.J. Esman, L. Liu, V. Ataie, E. Myslivets, B.P.-P. Kuo, N. Alic, S. Radic,
S. Zlatanovic, J. R. Adleman, E.W. Jacobs, Agile wideband microwave/millimeter-wave
analysis by coherent lterless channelizer, in Proc. 2013 IEEE International Topical Meeting
on Microwave Photonics (MWP), Post-deadline paper
16. B.P.-P. Kuo, E. Myslivets, V. Ataie, E.G. Temprana, N. Alic, S. Radic, Wideband parametric
frequency comb as coherent optical carrier. J. Lightwave Technol. 31, 34143419 (2013)
17. E.G. Temprana, V. Ataie, B.P.-P. Kuo, E. Myslivets, N. Alic, S. Radic, Low-noise
parametric frequency comb for continuous C-plus-L-band 16-QAM channels generation.
Opt. Express 22, 68226828 (2014)
18. A.H. Gnauck, B.P.-P. Kuo, E. Myslivets, R.M. Jopson, M. Dinu, J.E. Simsarian, P.J. Winzer,
S. Radic, Comb-based 16-QAM transmitter spanning C and L bands. IEEE Photon. Technol.
Lett. 26, 10411135 (2014)
19. B.P.-P. Kuo, N. Alic, P.F. Wysocki, S. Radic, Simultaneous NIR and SWIR wavelength-
swept generation over record 329-nm range using swept-pump ber optical parametric
oscillator. J. Lightwave Technol. 29, 410416 (2011)
20. R. Nissim, A. Pejkic, E. Mylivets, B.P.-P. Kuo, N. Alic, S. Radic, Ultrafast optical control by
few photons in nanometer scale tailored ber, in review
21. J. Hansryd, P.A. Andrekson, M. Westlund, J. Li, P. Hedekvist, Fiber-based optical parametric
ampliers and their applications. IEEE J. Select. Topics Quant. Electron. 8, 506520 (2002)
22. S. Radic, Parametric amplication and processing in optical bers. Laser Photonics Rev. 2,
498513 (2008)
23. M.E. Marhic, Fiber Optical Parametric Ampliers, Oscillators and Related Devices
(Cambridge University, Cambridge, 2007)
24. J.A. Armstrong, N. Bloembergen, J. Ducuing, P.S. Pershan, Interaction between light waves
in a nonlinear dielectric. Phys. Rev. 127, 19181939 (1962)
25. R.H. Stolen, J.E. Bjorkholm, Parametric amplication and frequency conversion in optical
bers. IEEE J. Quant. Electron. 18, 10621072 (1982)
26. T. Torounidis, P.A. Andrekson, B.-E. Olsson, Fiber-optical parametric amplier with 70-dB
gain. IEEE Photon. Technol. Lett. 18, 11941196 (2006)
27. R. Laming, M.N. Zervas, D.N. Payne, Erbium-doped ber amplier with 54 dB gain and
3.1 dB noise gure. IEEE Photon. Technol. Lett. 4, 13451347 (1992)
28. M.N. Zervas, R.I. Laming, Rayleigh scattering effect on the gain efciency and noise of
erbium-doped ber ampliers. IEEE J. Quant. Electron. 31, 468471 (1995)
29. S. Radic, Parametric signal processing. J. Selected Topics Quant. Electron. 18, 670680
(2012)
30. C.J. McKinstrie, S. Radic, A.R. Chraplyvy, Parametric ampliers driven by two pump
waves. IEEE J. Select. Topics Quant. Electron. 8, 538547 (2002)
31. C.J. McKinstrie, S. Radic, M.G. Raymer, Quantum noise properties of parametric ampliers
driven by two pump waves. Opt. Express 12, 50375066 (2004)
32. Q. Lin, R. Jiang, C.F. Marki, C.J. McKinstrie, R. Jopson, J. Ford, G.P. Agrawal, S. Radic,
40-Gb/s optical switching and wavelength multicasting in a two-pump parametric device.
IEEE Photon. Technol. Lett. 17, 23762378 (2005)
33. S. Radic, C.J. McKinstrie, R.M. Jopson, A.H. Gnauck, J.C. Centanni, A.R. Chraplyvy,
Multiple-band bit-level switching in two-pump ber parametric devices. IEEE Photon.
Technol. Lett. 16, 852854 (2004)
34. S. Watanabe, F. Futami, R. Okabe, R. Ludwig, C. Schmidt-Langhorst, B. Huettl, C. Schubert,
H.-G. Weber, An optical parametric amplied ber switch for optical signal processing and
regeneration. IEEE J. Select. Topics Quant. Electron. 14, 674680 (2008)
35. M.E. Marhic, K.K.-Y. Wong, L.G. Kazovsky, T.-E. Tsai, Continuous-wave ber optical
parametric oscillator. Opt. Lett. 27, 14391441 (2002)
36. M. Fiorentino, P.L. Voss, J.E. Sharping, P. Kumar, All-ber photon-pair source for quantum
communications. IEEE Photon. Technol. Lett. 14, 983985 (2002)
37. D.M.F. Lai, C.H. Kwok, K.K.-Y. Wong, All-optical picoseconds logic gates based on a ber
optical parametric amplier. Opt. Express 16, 1836218370 (2008)
38. S. Watanabe, G. Ishikawa, T. Naito, T. Chikama, Generation of optical phase-conjugate
waves and compensation for pulse shape distortion in a single-mode ber. J. Lightwave
Technol. 12, 21392146 (1994)
39. B.P.-P. Kuo, E. Myslivets, A.O.J. Wiberg, S. Zlatanovic, C.-S. Brs, S. Moro, F. Gholami,
A. Peric, N. Alic, S. Radic, Transmission of 640-Gb/s RZ-OOK channel over 100-km SSMF
by wavelength-transparent conjugation. J. Lightwave Technol. 29, 516523 (2011)
40. R.A. Fisher (ed.), Optical Phase Conjugation (Academic, New York, 1983)
41. A.V. Yulin, D.V. Skryabin, P.St.J. Russell, Four-wave mixing of linear waves and solitons in
ber with higher-order dispersion. Opt. Lett. 29, 24112413 (2004)
42. J. Li, J. Hansryd, P.O. Hedekvist, P.A. Andrekson, S.N. Knuden, 300-Gb/s eye-diagram
measurement by optical sampling using ber-based parametric amplication. IEEE Photon.
Technol. Lett. 13, 987989 (2001)
43. C.-S. Brs, A.O.J. Wiberg, B.P.-P. Kuo, J.M. Chavez-Boggio, C.F. Marki, N. Alic, S. Radic,
Optical demultiplexing of 320 Gb/s to 840 Gb/s in single parametric gate. J. Lightwave
Technol. 28, 434442 (2010)
44. E. Palushani, H.C.H. Mulvad, M. Galili, H. Hu, L.K. Oxenlowe, A.T. Clausen, P. Jeppesen,
OTDM-to-WDM conversion based on time-to-frequency mapping by time-domain optical
Fourier transformation. IEEE. J. Select. Topics Quant. Electron. 18, 681688 (2012)
45. C.J. McKinstrie, S. Radic, Phase-sensitive amplication in a ber. Opt. Express 12, 49734979
(2004)
46. Z. Tong, C. Lundstrom, P.A. Andrekson, C.J. McKinstrie, M. Karlsson, D.J. Blessing,
E. Tipsuwannakul, B.J. Puttnam, H. Toda, L. Gruner-Nielsen, Towards ultrasensitive optical
links enabled by low-noise phase-sensitive ampliers. Nat. Photon. 5, 430436 (2011)
47. R. Slavik, F. Parmigiani, J. Kakande, C. Lundstrom, M. Sjodin, P.A. Andrekson, R. Weerasuriya,
S. Sygletos, A.D. Ellis, L. Gruner-Nielson, D. Jakobsen, S. Herstrom, R. Phelan, J. OGorman,
A. Bogris, D. Syvridis, S. Dasgupta, P. Petropoulos, D.J. Richardson, All-optical phase and
amplitude regenerator for next-generation telecommunications systems. Nat. Photon. 4, 690695
(2010)
48. Z. Tong, A.O.J. Wiberg, E. Myslivets, B.P.-P. Kuo, N. Alic, S. Radic, Broadband parametric
multicasting via four-mode phase-sensitive interaction. Opt. Express 20, 1936319373
(2012)
49. Z. Tong, S. Radic, Low-noise optical amplication and signal processing in parametric
devices. Adv. Opt. Photon. 5, 318384 (2013)
50. M. DiDomenico Jr., Material dispersion in optical ber waveguides. Appl. Opt. 11, 652654
(1972)
51. D. Gloge, Dispersion in weakly guiding bers. Appl. Opt. 10, 24422445 (1971)
52. N. Shibata, R.P. Braun, R.G. Waarts, Phase-mismatch dependency of efciency of wave
generation through four-wave mixing in a single-mode optical ber. IEEE J. Quant. Electron
QE-23, 12051210 (1987)
53. H.R. Philipp, Silicon Dioxide (SiO2) (Glass), in Handbook of Optical Constants of Solids, ed.
by E.D. Palik (Academic, New York, 1985)
54. B.J. Ainslie, C.R. Day, A review of single-mode bers with modied dispersion
characteristics. J. Lightwave Technol. LT-4, 967979 (1986)
55. K. Inoue, Four-wave mixing in an optical ber in the zero-dispersion wavelength region.
56. J.M. Chavez-Boggio, S. Moro, B.P.-P. Kuo, N. Alic, S. Radic, Tunable parametric all-ber
short-wavelength IR transmitter. J. Lightwave Technol. 28, 443447 (2010)
57. J.E. Sharping, C. Pailo, C. Gu, L. Kiani, J.R. Sanborn, Microstructure ber optical parametric
oscillator with femtosecond output in the 1200 to 1350 nm wavelength range. Opt. Express
18, 39113916 (2010)
58. V.V. Ravi Kanth Kumar, A.K. George, J.C. Knight, PStJ Russell, Tellurite photonic crystal
ber. Opt. Express 11, 26412645 (2003)
59. L. Brilland, F. Smektala, G. Renversez, T. Chartier, J. Troles, T.N. Nguyen, N. Traynor,
A. Monteville, Fabrication of complex structures of holey bers in chalcogenide glass. Opt.
Express 14, 12801285 (2006)
60. P. Russell, Photonic crystal bers. Science 17, 358362 (2003)
61. F. Yaman, Q. Lin, S. Radic, G.P. Agrawal, Impact of dispersion fluctuations on dual-pump
ber-optical parametric ampliers. IEEE Photon. Technol. Lett. 16, 12921294 (2004)
62. P. Sansonetti, Modal dispersion in single-mode bres: simple approximation issued from
mode spot size spectral behaviour. Electron. Lett. 18, 647648 (1982)
63. K. Petermann, Constraints for fundamental-mode spot size for broadband dispersion-
compensated single-mode bres. Electron. Lett. 19, 712714 (1983)
64. C. Pask, Physical interpretation of Petermanns strange spot size for single-mode bres.
Electron. Lett. 20, 144145 (1984)
65. M. Farahmand, M. de Sterke, Parametric amplication in presence of dispersion fluctuations.
Opt. Express 12, 136142 (2004)
66. S. Nishi, M. Saruwatari, Technique for measuring the distributed zero dispersion wavelength
of optical bres using pulse amplication caused by modulation instability. Electron. Lett.
31, 225226 (1995)
67. L.F. Mollenauer, P.V. Mamyshev, M.J. Neubelt, Method for facile and accurate
measurement of optical ber dispersion maps. Opt. Lett. 21, 17241726 (1996)
68. J. Gripp, L.F. Mollenauer, Enhanced range for OTDR-like dispersion map measurements.
Opt. Lett. 23, 16031605 (1998)
69. J.B. Schlager, Zero-dispersion wavelength distribution in optical bers from CW four-wave
mixing. Lasers and Electro-Optics Society Annual Meeting, 1998. LEOS98. IEEE 2, 309310
(1998)
70. M. Eiself, R.M. Jopson, R.H. Stolen, Nondestructive position-resolved measurement of the
zero-dispersion wavelength in an optical ber. J. Lightwave Technol. 15, 135143 (1997)
71. S. Moro, E. Myslivets, J.R. Windmiller, N. Alic, J.M. Chavez-Boggio, S. Radic, Synthesis of
equalized broadband parametric gain by localized dispersion mapping. IEEE Photon.
Technol. Lett. 20, 19711973 (2008)
72. L.H. Gabrielli, H.E. Hernndez-Figueroa, H.L. Fragnito, Robustness optimization of ber
index proles for optical parametric ampliers. J. Lightwave Technol. 27, 55715579 (2009)
73. B.P.-P. Kuo, Recent advances in wide-band optical parametric mixer synthesis, in Proc.
OECC/PS 2013, paper TuS2-1
74. F. Gholami, B.P.-P. Kuo, S. Zlatanovic, N. Alic, S. Radic, Phase-preserving parametric
wavelength conversion to SWIR band in highly nonlinear dispersion stabilized ber. Opt.
Express 21(9), 1141511424 (2013)
75. J.H. Lee, T. Tanemura, K. Kikuchi, T. Nagashima, T. Hasegawa, S. Ohara, N. Sugimoto,
Experimental comparison of a Kerr nonlinearity gure of merit including the stimulated
Brillouin scattering threshold for state-of-the-art nonlinear optical bers. Opt. Lett. 30,
16981700 (2005)
76. H.K. Tsang, C.S. Wong, T.K. Liang, I.E. Day, S.W. Roberts, A. Harpin, J. Drake, M. Asghari,
Optical dispersion, two-photon absorption and self-phase modulation in silicon waveguides at
1.5m. Appl. Phys. Lett. 80, 416418 (2002)
77. N. Yoshizawa, T. Imai, Stimulated Brillouin scattering suppression by means of applying

strain distribution to ber with cabling. J. Lightwave Technol. 11, 15181522 (1993)
78. K. Shiraki, M. Ohashi, M. Tateda, Suppression of stimulated Brillouin scattering in a bre by
changing the core radius. Electron. Lett. 31, 668669 (1995)
79. E. Lichtman, R.G. Waarts, A.A. Friesem, Stimulated brillouin scattering excited by a
modulated pump wave in single-mode bers. J. Lightwave Technol. 7, 171174 (1989)
80. A. Cerqueira-Sodre Jr., J.M. Chavez Boggio, A.A. Rieznik, H.E. Hernandez-Figueroa, H.K.
Fragnito, J.C. Knight, Highly-efcient generation of broadband cascaded four-wave mixing
products. Opt. Express 16, 28162828 (2008)
81. D. Grischkowsky, A.C. Balant, Optical pulse compression based on enhanced frequency
chirping. Appl. Phys. Lett. 41, 13 (1982)
82. G. Yang, Y.R. Shen, Spectral broadening of ultrashort pulses in a nonlinear medium. Opt.
Lett. 9, 510512 (1984)
83. P.K.A. Wai, C.R. Menyuk, Y.C. Lee, H.H. Chen, Nonlienar pulse propagation in the
neighborhood of the zero-dispersion wavelength of monomode optical bers. Opt. Lett. 11,
464466 (1986)
84. K. Tai, A. Hasegawa, A. Tomita, Observation of modulational instability in optical bers.
Phys. Rev. Lett. 56, 135138 (1986)
85. M. Nakazawa, H. Kubota, K. Tamura, Random evolution and coherence degradation of a
higher-order optical soliton train in the presence of noise. Opt. Lett. 24, 318320 (1999)
86. B.P.-P. Kuo, Agile Optical Frequency Synthesis via Parametric Processes, Ph.D. Dissertation
(University of California, San Diego, 2011)
87. J.E. Rothenberg, D. Grischkowsky, Observation of the formation of an optical intensity
shock and wave breaking in the nonlinear propagation of pulses in optical bers. Phys. Rev.
Lett. 62, 531534 (1989)
88. W.J. Tomlinson, R.H. Stolen, A.M. Johnson, Optical wave breaking of pulses in nonlinear
89. M.G. Taylor, Coherent detection method using DSP for demodulation of signal and subsequent
equalization of propagation impairments. IEEE Photonics Technol. Lett. 16, 674676 (2004)
90. S.J. Savory, G. Gavioli, R.I. Killey, P. Bayvel, Electronic compensation of chromatic
dispersion using a digital coherent receiver. Opt. Express 15, 21202126 (2007)
91. E. Ip, J. Kahn, Digital equalization of chromatic dispersion and polarization mode dispersion.
92. D.S. Millar, S. Makovejs, C. Behrens, S. Hellerbrand, R.I. Killey, P. Bayvel, S.J. Savory,
Mitigation of ber nonlinearity using a digital coherent receiver. IEEE J. Select. Topics
Quant. Electron. 16, 12171226 (2010)
93. S. Savory, Digital coherent optical receivers: algorithms and subsystems. IEEE J. Select.
Topics Quant. Electron. 16, 11641179 (2010)
94. R. Essiambre, G. Kramer, P.J. Winzer, G.J. Foschini, B. Goebel, Capacity limits of optical
ber networks. J. Lightwave Technol. 28, 662701 (2010)
95. E. Ip, J.M. Kahn, Feedforward carrier recovery for coherent optical communications.
96. I. Fatadin, D. Ives, S.J. Savory, Differential carrier phase recovery for QPSK optical coherent
systems with integrated tunable lasers. Opt. Express 21, 1016610171 (2013)
97. V. Ataie, E. Myslivets, B.P.-P. Kuo, N. Alic, S. Radic, Spectrally equalized frequency comb
generation in multistage parametric mixer with nonlinear pulse shaping. J. Lightwave
Technol. 32, 840846 (2014)
98. C.-S. Brs, A.O.J. Wiberg, B.P.-P. Kuo, N. Alic, S. Radic, Wavelength-multicasting
of 320 Gb/s channel in self-seeded parametric amplier. IEEE Photon. Technol. Lett. 21,
10021004 (2009)
99. L.H. Sahasrabuddhe, B. Mukherjee, Light-trees: optical multicasting for improved
performance in wavelengthrouted networks. IEEE Commun. Mag. 37, 6773 (1999)
100. H.P. Yuen, Reduction of quantum fluctuation and suppression of the Gordon-Haus effect
with phase-sensitive linear ampliers. Opt. Lett. 17, 7375 (1992)
101. C.-S. Brs, S. Zlatanovic, A.O.J. Wiberg, J.R. Adleman, C.K. Huyuh, E.W. Jacobs,
J.M. Kvavle, S. Radic, Parametric photonic channelized RF receiver. IEEE Photon. Technol.
Lett. 23, 344346 (2011)
102. H. Sorensen, M. Heideman, C. Burrus, On computing the split-radix FFT. IEEE Trans.
Acoust. Speech Signal Process. 34, 152156 (1986)
103. W.J. Williams, M.L. Brown, A.O. Hero III, Uncertainty, information, and time-frequency
distributions, Proc. SPIE 1566, Advanced Signal Processing Algorithms, Architectures and
Implementations II, 144156 (1991)
Chapter 8
All-Optical Pulse Shaping for Highest
Spectral Efciency
Juerg Leuthold and Camille-Sophie Brs
Abstract Pulse shaping gives communications engineers another degree of freedom

in designing a link. It holds promise to allow extending transmission reach, achieve
optical multiplexing at highest spectral efciency or to limit nonlinear distortions. A
variety of pulse shapesrectangular, sinc, raised cosine to cite just a fewhave been
investigated but the important question is how optical transmitters can generate such
pulses at the necessary speed. Should the transmitter be realized in the digital domain,
the all-optical domain or can it be implemented as a hybrid? In this chapter, the
fundamentals for pulse shaping in transmitters and receivers are reviewed. A par-
ticular emphasis is on orthogonal frequency division multiplexing (OFDM) where
the systems data are encoded onto subcarriers with a rectangularly shaped impulse
response, and Nyquist pulse shaping where the symbols are carried by Nyquist
pulses. Electronic, digital and optical processors are described and recent experi-
mental demonstrations are reported.
8.1 Introduction
Pulse shaping has been used for many years in optical communications. However,
pulse shaping is an effort and therefore it was only used for medium and long-haul
transmission systems. In practice, conventional pulse shaping would be performed
this way: One would encode information in the form of an on-off keying signal onto
a cw laser. This way a non-return-to-zero (NRZ) encoded signal would be formed.
NRZ signals were good enough to transmit 10 Gbit/s signals over hundreds of
J. Leuthold (&)
Institute of Electromagnetic Fields, ETH-Zurich, ETZ K 81, Gloriastrasse 35,
CH-8092 Zurich, Switzerland
e-mail: JuergLeuthold@ethz.ch
C.-S. Brs
Institute of Electrical Engineering, Ecole Polytechnique Fdrale de
Lausanne (EPFL), Station 11, CH-1015 Lausanne, Switzerland

218 J. Leuthold and C.-S. Brs
kilometers. If longer reach was needed, one would carve the pulse by means of a
Mach-Zehnder modulator onto the NRZ signal. For long reach communications one
would typically use a 33 % return-to-zero pulse shape. In some instance, when
nonlinearities would be an issue one could as well opt for a 66 % carrier-suppressed
return to zero pulse shape.
More recently pulse shaping is not only used to extend the reach but rather as a
means to achieve optical multiplexing at highest spectral efciency. Two most
prominent examples are OFDM [110] and Nyquist [1113] multiplexed signals. And
indeed, a recent transmission experiment shows that signal encoding with 18 bit/s/Hz
and beyond is feasible [14]. Prerequisites for such multiplexing schemes are advanced
optical transmitters that encode the information as complex-valued, specially shaped
pulse envelopes onto optical carriers. The shapes are selected such that signals are
orthogonal in time and frequency. This allows for demultiplexing of appropriately
generated signals even if they overlap in time and/or in the frequency domain.
A variety of pulse shapes have been investigated: Rectangularly-shaped, sinc-
shaped and raised-cosine shaped pulses are just a few. The important question now
is how optical transmitters can generate such pulses at the necessary speed.
Pulse shaping is most easily performed by digital signal processing. Shaping
signals with digital signal processing (DSP) has become possible due to the
availability of massive parallel high-speed electronics. We now cannot only encode
any complex value onto an optical carrier, but we even have sufcient processing
bandwidth to shape it real-time within the time slot of a symbol with sufcient
precision. And indeed, the trend is towards software dened optical transmitters that
can adapt modulation formats, pulse shapes and symbol rates at the push of a button
[15]. However, electronic pulse shaping costs DSP signal processing resources and
comes at the price of a higher power consumption. Moreover, electronic pulse
shaping will always be limited to electronic signal processing speed. Conversely,
all-optical techniques can be extremely energy efcient and there is hardly any
speed limitation. Also, with all-optical techniques quantization and clipping of
pulses with high peak-power can be avoided. Generally, it is believed that all-
DSP pulse shaping approaches are most reasonable when lower-speed tributaries
are combined into a super-channel at a moderate aggregate data rate. In turn, all-
optical approaches are attractive when combining higher-bit-rate channels into
multiple, spectrally efcient Tbit/s superchannels. In practice, hybrid implementa-
tions, where transceivers are neither all-DSP nor all-optical might make most
sense for encoding superchannels.
8.2 Fundamentals
Transmitters (Tx) and receivers (Rx) employing pulse shaping techniques are
somewhat different from more conventional transceivers. In the following section
we follow and to some extend summarize a discussion recently published in [16],
where one may also nd a more extensive review.
8 All-Optical Pulse Shaping for Highest Spectral Efciency 219
(a) Tx/Rx with Pulse-Shaping In the Optical Domain

... ...
Tx Channel Rx
DSP E/O Opt Opt O/E DSP
Coh. Det.
IQ-Mod.
Shaping
Shaping
Split
...
...
Mux
* *
(b) Tx/Rx with Pulse-Shaping In the Digital Domain
Tx Channel Rx
DSP E/O Opt O/E DSP
Coh. Det.
IQ-Mixer
IQ-Mixer
IQ-Mod.
Shaping
Shaping
...
Split
...
Mux
* *
Fig. 8.1 Optical and electrical transmitter-receiver concepts for multiplexing at highest spectral
efciencies. In a the symbols are rst converted into the optical domain by IQ-modulators and then
optically multiplexed. In b the signals are modulated on different radio frequency carriers and
multiplexed electronically before being encoded onto an optical carrier (gure modied from [16])
Two implementations of a Tx/Rx pair using pulse shaping are depicted in Fig. 8.1.
In Fig. 8.1a pulse shaping is performed in the optical domain. Here the complex
information in the form of complex symbols are rst converted from the electronic to
the optical domain (E/O-box) by IQ-mixers (, in optical communications also
known as optical IQ-modulators). A particular pulse shape hs(t) is then carved onto
the optical signal. Finally, several tributaries may be wavelength multiplexed. The
optical transmitter scheme of Fig. 8.1a is best suited to generate Tbit/s superchannels
[8]. In Fig. 8.1b pulse shaping is implemented in the digital domain. In the digital
domain, pulse shaping, signal encoding and multiplexing are performed by a pro-
cessor. The whole electrical signal with all the subcarriers is then encoded by means
of an IQ-Modulator (E/O) onto a single laser. Such digital signal processing power
has only recently become available and in fact it is only recently, that 100 Gbit/s
OFDM [17] and Nyquist multiplexing [13] has been demonstrated. A key argument
for using pulse shaping is the possibility to increase the spectral efciency dramat-
ically. If the pulse shape in the transmitter and receiver are properly chosen record
spectral efciencies of as much 18 bit/s/Hz with a net spectral efciency of 15 bit/s/
Hz [14] have already been demonstrated. Such world records are only possible by
clever pulse shaping that allow multiplexing of subcarrier data without guard bands.
We now more closely review the pulse-shaping functions implemented in the
transmitter (Tx) and receiver (Rx). Pulse-shaping in the transmitter and ltering in
the receiver are interrelated. When the impulse response of the signal in the
transmitter is hs(t) then the received signal needs to be convolved with a lter
response function hr(t), where hs(t) and hr(t) need to be orthogonal, i.e. they need to
obey the orthogonality condition [16]
Z1
Ts hs t tm hr tm0 t dt dmm ; 8:1
1
where tm mTs with integers m; m0 and Ts the symbol duration.

Important functions that meet an orthonormality relations are e.g.
Z1 t t
1
rect m rect m0 dt dmm0
T T T
1
Z1 t t
1
sinc m sinc m0 dt dmm0
T T T
1
ToZT=2 8:2
1 t t
exp j2pm exp j2pm0 dt dmm0
T T T
To T=2
FZo F
1 pf f pf f
cos exp j2pm cos exp j2pm0 df dmm0
F 2F F 2F F
Fo F
In (8.2) the functions rectt=T and sinct=T are dened as follows

t
1 for jtj\T=2
rect 8:3
T 0 else
t
1 for t 0
sinc sinpt=T
else : 8:4
T pt=T
Similar to sinc-pulses, raised cosine pulses share the property that impulse
maxima and zeros of neighbouring pulses can be made to coincide, but such pulses
are not necessarily orthogonal in the sense of (8.2) and as such do not allow ISI-free
detection with spectral efciencies at the Nyquist limit.
Tx Channel Rx
Split
Sum
Shaping
Shaping
Sampling

* *
Fig. 8.2 Generic transmission system with transmitter (Tx), channel, and receiver (Rx). A Tx with
^
N branches v is shown. Each branch is fed by a time series of complex data coefcient c iv at time
intervals iT. They are convoluted by impulse-forming lter with impulse response hsv(t) and up
converted to the sub-channel frequency fvP by a complex (in-phase/quadrature, IQ) mixer with a
local oscillator (LO). The summing node v adds the up-converted signals siv t in time frame
i for all branches v. The resulting total signal s(t) consists of concatenated time frames and enters
the linear channel. On the right side we have a Rx with N branches v0 . The total received signal r
(t) that leaves the channel enters a splitting node sv and is distributed with equal amplitudes to the
receiver branches v0 . The received signal ri0 v0 t in time frame i0 and branch v0 is frequency down-
converted with an IQ mixer and a LO expj2pfv0 t. After a receiver lter with impulse response
^0
hrv0 t, a sampler extracts at time t i0 T tgv0 the complex data coefcient c i0 v0 (gure modied
from [16])
We now take a more mathematical approach to the implementation of a pulse-

^
shaping transmitter and receiver pair, see Fig. 8.2. In Fig. 8.2 the coefcients c iv
represent the complex baseband data signal in branch v at time t iT. A complex
mixer (in-phase/quadrature mixer, or a so-called IQ mixer) converts the baseband
signal to the frequency fv of a complex subcarrier represented by the analytic signal
exp (j2 fv t),
fv vFs fc ; v 0; 1; . . .; N 1; Fs 1=Ts ; channel frequency fc

8:5
As a result, a signal at a subcarrier frequency fv is generated. It enters the node

Sum in Fig. 8.2 where the subcarrier signals from all branches are summed
(or multiplexed) to form the signal si(t).
X
N 1
^
si t c iv Ts hsv t iT expj2pfv t 8:6
v0
Finally, the innitely many consecutive time frames i constitute the signal s
(t) that is sent across the channel
X
1
st si t 8:7
i1
The effect of the channel can be taken into account by delaying the signals
impulse response hsv(t) by the group delay tgv, and by adapting a retardation onto the
carrier phase. Of course, the signal will also be distorted by dispersion, which can be
undone in the receiver. The channel will further introduce some nonlinear distortions.
Some of which can be undone by clever DSP algorithms. Other nonlinear distortions
need to be minimized by e.g. reducing the launch power of the signal. To keep things
simple we only introduce the group delay tgv onto the received signal and indicate
potential distortions due to transmission by modifying the signal impulse response
hsv(t) into a modied response h0sv t. The received signal r(t) then reads
X
1 X
N 1
civ h0sv t iT tgv expj2pfv t uv : 8:8
^
r t ri t; r i t Ts
i1 v0
^
The ultimate goal of a receiver is the extraction of the data coefcients c iv . A schematic
of a generic receiver that performs this operation is shown at the right-hand side of
Fig. 8.2. Again, only one out of N branches v0 is depicted. First, the received signal
r(t) is split in a splitting node Split with equal magnitudes and phases to N receiver
branches v0 . In a second step the received signal ri0 t is then down converted in
branch v0 for each time frame i0 with a complex mixer and a LO expj2pfv0 t. In a
third step, the down-converted signal needs to be convolved with the receivers
impulse response hrv0 t. If sampling is performed at times t i0 T tgv0 then one can
^0
extract the coefcient ci0 v0 . The total operation can be expressed by
1 Z
1
X
ri0 t0 expj2pfv0 t0 hrv0 t t0 dt0 with t i0 T tgv0
^0
ci0 v0 8:9
i1
1
To nd the proper sampling time t i0 T tgv0 and frequency fv0 a timing, fre-
quency and phase estimation unit is needed, which is not shown in the simplied
receiver depicted in Fig. 8.2.
Inserting (8.8) into (8.9) gives
1 " #
1 Z
X X
N1 0 0
c iv Ts hsv t i T tgv expj2pfv t exp2pfv0 t0 hrv0 t t0 dt0
0
^0 ^
c i0 v0
i1 v0
1
X
1 X Z1
N1 0 0
hsv t i T tgv hrv0 t t0 expj2pfv f v0 t0 dt0

^
civ Ts
i1 v0
1
8:10
In the ideal case where the frequency of the local oscillator in the receiver fv0 fv
one can see from (8.10) that the exponential term reduces to a 1. Further, if the
pulse shape in the transmitter hsv and the lter response hrv0 in the receiver are
orthogonal and the sampling time in the receiver is chosen as t i0 T tgv0 , then
(8.10) reduces to
^0 ^
ci0 v0 c iv di0 i dv0 v 8:11
However, if frequencies or pulse shapes are not properly chosen there might be
interference. Equation (8.10) allows for interferences from sub-channels v 6 v0 or
inter channel interference (ICI) from channels other than the present receiving
branch v0 . Distortions might also come in through neighboring data inside the same
sub-channel v0 but belonging to different time slots i 6 i0 , i.e. inter-symbol inter-
ference (ISI).
8.3 Orthogonal Frequency Division Multiplexing (OFDM)
8.3.1 OFDM Tx and Rx Implementations
8.3.1.1 Pulse-Shaping Implementation
In orthogonal frequency division multiplexed (OFDM) systems data is encoded

onto subcarriers with a rectangularly shaped impulse response. The minimum pulse
duration is Ts, with Ts the symbol duration. In OFDM the pulse maintains its
constant value for the whole duration of the symbol time. The symbol period T, i.e.
the time it takes for the next symbol to start, however might be longer than the
symbol duration Ts, i.e. T > Ts. The symbol period denes the clock rate fT
1=T Fs at which the symbols are sent. In practical implementations the pulse at
best maintains its value not only for Ts but the whole period T. The extension of the
pulse shape beyond Ts is known as the cyclic extension. However, this extension
not necessarily needs to be used as a cyclic extension. As a matter of fact the pulse
shape may take on any value in the interval beyond Ts. In many cases this extension
is simply needed because it takes some time for a practical modulator to produce a
rectangularly shaped symbol with a certain value.
In frequency domain the rectangularly shaped time domain subcarriers of the
OFDM signal have sinc-shaped spectra and the sub-carriers are spaced at the
inverse of the symbol duration Ts.
Mathematically, we can describe an OFDM impulse response by
8
>
< 1=Ts for 0\t\Ts
hsv t unspecified for Ts \T 8:12
>
:
0 else:
The receivers lter impulse response needs to meet the orthogonality condition
(8.1) in order to guarantee interference-free reception with highest spectral
efciency. For this the lter needs to be chosen such that it is non-zero only in a
region 0 < t < Ts, see equation.
(
1 t 1 1=Ts for 0 \ t \ Ts
hrv t rect : 8:13
Ts Ts 2 0 else
To encode an OFDM signal we now only need to follow the recipe from (8.6).
We rst shape the pulse with the rectangularly shaped pulse from (8.12). To keep
things simple we disregard any cyclic redundancy and potential guard bands. We
then mix it with the subcarrier frequencies fv and sum all subcarriers. Finally, the
signal is mapped onto an optical carrier at frequency fc by means of a complex
mixer, i.e. an IQ-modulator. This leads us to the following expression for an OFDM
signal
X
N 1
^ t iTs
si t expj2pfc t c iv rect expj2pfv t 8:14
v0
Ts
We then notice, that the N subcarriers of an OFDM signal are equipspaced by Fs

and fv vFs + fc
N 1
t iTs X ^
si t expj2pfc t rect civ expj2pvFs t: 8:15
Ts v0
In literature, the rect expression is often dropped as people usually assume that
^
the coefcients civ are only dened for the duration of a symbol (Fig. 8.3).
The shape and spectrum of an OFDM signal with three subcarriers is visualized
in Fig. 8.3.
Next we built a pulse-shaping OFDM receiver. It is a hypothetical receiver as
this type of a receiver is not really implemented in practical systems. Stillit is
(a) (b)
Fig. 8.3 OFDM pulses in a time domain, siv t and b frequency domain ^siv f . The green box
marks the nite symbol time. a The left graph shows the real parts of three rectangularly shaped
subcarriers oscillating at a different carrier frequency each. The sum over all three carriers would
represent one time-domain OFDM symbol of duration Ts. b The right graph shows the
corresponding spectrum ^s0v f of three subcarriers with a spectral separation of jfv1 fv j Fs
1=Ts (gure derived from [16])
(a) Pulse-Shaping Tx
Sample Shaping Mix Sum E/O
(b) Pulse-Shaping Rx
O/E Split Mix Shaping Sample
delay by
Fig. 8.4 Possible implementation a OFDM transmitter and b receiver by means of the pulse-
shaping technique
quite educational to discuss it. Now, in this hypothetical receiver, the signal would
rst be down converted by the local oscillator frequency fc by means of an optical
hybrid from the optical to the electrical domain. Subsequently, the electrical signal
would be split onto N copies. Each of which would then be mixed by the subcarrier
frequency and integration over a symbol duration would be performed. The
mathematical process according to (8.9) leads to
i0 1Ts tgv0
Z
X
1
exp j2pfc i0 Ts tgv0 ri0 t0 expj2pfv0 t0 dt0
^0
ci0 v0 8:16
i1
i0 Ts tgv0
A graphical sketch of the OFDM transmitter and receiver is shown in Fig. 8.4.
The OFDM transmitter and receiver of Fig. 8.4 are electrical OFDM imple-
mentations since all the relevant processing is done in the electrical domain
except for the E/O and O/E up- and down conversion. The scheme represents a
more in-detail description of the concept mentioned by Fig. 8.1b. However, it
should be emphasized, that the electrical up- and down mixers could be replaced by
E/O and O/E converters, i.e. the electrical mixers could be replaced by lasers and
IQ-modulators. This would then lead to an optical implementation as Fig. 8.1a. We
cannot discuss any possible transceiver implementation here. However, we will see
such an implementation further below when we discuss our example.
8.3.1.2 Discrete Fourier Transform Implementation
We now take a closer look at the OFDM signal in (8.15). We notice that in practice
one usually works with digital signals and thus generates and detects signals by
time-discrete sampling at times tn iTs + nTs/N with n 0; 1; 2; . . . N 1.
Replacing t by tn in (8.15) leads to
N1
tn iTs X ^ nTs
si tn expj2pfc t rect c iv exp j2pvFs
Ts =N v0
N
XN 1 vn
tn iTs ^
8:17
expj2pfc t rect c iv exp j2p
Ts =N v0
N

tn iTs
expj2pfc t rect cin :
Ts =N
The last term thus happens to be the expression for the Inverse Discrete Fourier
Transform (IDFT) and the coefcients cin are thus the discrete time samples of the
^
Fourier coefcients c iv . Thus, an alternative way to encode the OFDM signal is by
^
applying the IDFT onto the coefcients c iv and to directly encode the signal si(tn) in
time domain onto an optical carrier with frequency fc.
An implementation of an IDFT transmitter is depicted in Fig. 8.5a. The coef-
^
cients civ are rst gated to a duration Ts/N before being fed into the IDFT processor.
^
The IDFT processer then performs the inverse DFT operation on the coefcients c iv
X
N 1 vn
^
cin c iv exp j2p 8:18
v0
N
Subsequently, the time samples cin of duration Ts/N are concatenated (a parallel-
to-serial conversion is performed) in order to produce an OFDM symbol of duration
Ts. Eventually, the signal is E/O converted by means of an IQ-modulator, see
Fig. 8.5a.
At the receiver, the inverse process is performed. The incoming OFDM data
stream r(t) is split onto N signals that are delayed such that N OFDM time samples
cin within a symbol Ts are simultaneously fed into the DFT processor (fed in parallel
^
into the DFT processor). The DFT processor generates the Fourier coefcients c iv .
^
The Fourier coefcients c iv are yielded only for the short duration of Ts/N, i.e. the
short time during which all N time samples have been fed in parallel into the
processor. We thus perform a gating operation at the output in order to suppress
output signals from the DFT processor that are based on wrong input signals.
(a) DFT Tx
Sample Gating IDFT P/S E/O
...

Delay by Identical LO
(b) DFT Rx
S/P DFT Gating Sample
E/O
...
delay by Sample at
identical LO
Fig. 8.5 Implementation of an a OFDM transmitter and b receiver based on the IDFT and DFT
8.3.2 Optical Fourier Transform Processors and Optical

OFDM
We just have seen that Discrete Fourier Transform (DFT) processors are important
building blocks. The generation and reception of OFDM is just one possible
application. DFT processors also play an important role in many applications where
spectral components of a signal are needed. Meanwhile there exist efcient elec-
tronic DFT processors that allow operation up to highest speed. And while there is
an impressive progress to the extent that 100 Gbit/s real-time processing has
become possible [18] the speed is ultimately limited by electronics. Fortunately,
unprecedented DFT processing speed can be obtained by optical means. We will
now discuss such optical realizations of the DFT operation.
8.3.2.1 FFT with Butterfly Topology and Optical OFDM
A fast Fourier Transform (FFT) processor yields the same input-output operation as
a DFT processor. Yet, a FFT processor performs the DFT with less complexity and
thus fewer elements. Since the number of elements should be minimized in an

optical implementation it makes sense to implement an FFT rather than a DFT.
More precisely, if the expression for a DFT of order N is given by
X
N 1 vn
^
c iv cin exp j2p ; 8:19
n0
N
where N 2p with an integer p then the equivalent expression can be decimated

into two interleaved DFTs of order N/2, where
(
^
Ev exp j2p v
N Ov if v\N=2
c iv 8:20
EvN=2 exp j2p vN=2
N OvN=2 if v N=2;
where the quantities Ev and Ov are the even and odd DFT of size N/2 for even and
odd inputs ci,2n and ci,2n+1 (n (0, 1, 2, , N/2 1)), respectively. The FFT of
(8.20) is then performed recursively to the end.
In Fig. 8.6 we have depicted the conguration for an optical OFDM receiver
with a built in optical 4 4 FFT processor. The overall OFDM implementation
closely follows the scheme outlined in Fig. 8.5b. The OFDM symbols ri are rst fed
into a serial-to-parallel converter. This can be obtained by splitting the signal into
four copies and delaying each of the copies by Ts/4 with respect to the previous
copy. This way the symbol is split in this example into four time samples. The
signals cin (still encoded on an optical carrier) are then launched into said optical
FFT processor. The FFT depicted here is a direct implementation of the mathe-
matical FFT expression (8.20). The optical FFT scheme shown here was rst
suggested by Marhic as early as 1987 [19]. A rst realization of a 2 2 FFT was
presented by Sanjoh et al. in 2002 [20] and a most impressive OFDM implemen-
tation with an 8 8 FFT was demonstrated in 2010 by Takiguchi et al. [21]. In the
optical implementation shown here, the down conversion into the electronic domain
(O/E conversion) by means of a coherent receiver is only performed at the end. This
S/P FFT Gating O/E
0 0 Gate
0 0 Gate
0 0 Gate
0 - /2 Gate
Fig. 8.6 Plot of an optical OFDM processor with an optical 4 4 FFT. The N 4 butterfly-type
FFT processor comprises of a serial-to-parallel converting delay stage, a FFT stage and a gating
stage. The unit processes signals with a symbol duration Ts. Rectangular sharp-edged boxes stand
for optical couplers with an (unphysical) amplitude split ratio of 1. Loops symbolize time delays by
fractions of the symbol duration Ts. Rounded boxes with numbers mark the respective phase shifts
u according to (8.18). See [10, 16]
is in contrast to an electrical FFT where the O/E conversion would be performed in

a rst stage.
With an FFT algorithm the DFT complexity of order CDFT N2 can typically be
reduced to a complexity of order CFFT N log2 N. Despite of this simplication the
direct FFT implementation still scales very badly with the order of the FFT. An
integrated version of order 8 is already quite a task and requires multiple couplers
and precisely tuned phase-shifter elements [21].
8.3.2.2 FFT with Cascaded Delay Interferometers and Optical OFDM
In 2010 Hillerkuss et al. demonstrated that the optical FFT could be performed with
a much simpler delay interferometer cascade [10]. Here we follow the line of
arguments in said paper where it is demonstrated how the cascaded delay inter-
ferometer approach can be obtained by applying some simple transformations on
the aforementioned optical FTT. In 2011, a record 26 Tbit/s OFDM signal was
encoded onto a single laser and the cascaded delay interferometer FFT approach
was used to demultiplex a total of 325 subcarriers [9].
The cascaded delay interferometer FFT in [10] is based on clever rearrangements
of both the S/P and FFT elements. To derive the new conguration we follow the
drawings in Fig. 8.7. In a rst step we rearrange the delays in the S/P conversion
stage as indicated in Fig. 8.7a and re-label the outputs accordingly. Please note that
we started with the very same conguration as in Fig. 8.6. After this rearrangement,
we end up with an optical FFT consisting of two parallel delay interferometers (DI)
with the same free spectral range (FSR) but different absolute delays, see Fig. 8.7b.
By moving the common delay Ts/4 in both arms of the lower DI to its outputs, one
obtains two identical DIs that are fed with the same input signal, Fig. 8.7c. This
redundancy can be eliminated by replacing the two DI with a single DI, and by
splitting its output, Fig. 8.7d. These simplication rules (a mathematical proof is
available [10]) can be iterated for FFTs of any size.
The new optical FFT processor consists of N 1 cascaded DI with a complexity
of only CFFTDI 2N 1 couplers and N 1 phase shifters. Thus, we not only
got rid of the S/P converter by integrating it into the FFT but also have now an
optical FFT that scales linearly with the order. The new FFT always requires less
complexity than the previous FFTexcept for N 2, where the complexity is
identical, since the two schemes by chance lead to an identical conguration.
Once the concept has been understood, one may go back to literature and will
nd, that cascaded delay-interferometers have already been studied by Li in 1998
for narrow spectral slicing. No gating elements were present for these so-called
Fourier lters though as these elements were mostly considered for narrow spaced
ltering. The gates, however are important for digital signal processing because
outside of the window one will not detect the proper coefcients. These lters were
also put in a wider context in connection with the wavelet transform by Cincotti
[22, 23] in 2002 and 2004 who also integrated the serial-to-parallel converter into a
network.
(a) Gate
0 0
0 0 Gate
0 0 Gate
0 - /2 Gate
S/P FFT Sampling
(b)
0 0 Gate
0 0 Gate
0 0 Gate
0 - /2 Gate
Combined S/P & FFT Sampling
(c)
0 0 Gate
0 0 Gate
0 0 Gate
0 - /2 Gate
(d)
Gate
0 Gate
0 Gate
- /2 Gate
Fig. 8.7 Simplication of optical FFT by rearranging and combining the S/P and all-optical FFT
elements illustrated on a four-point FFT circuit. a Traditional FFT implementation with separate S/
P conversion stage as discussed in Fig. 8.6. The arrows in red indicate as of how the coupler
elements can be rearranged without an effect on the output signals. b New combined S/P and FFT
structure consisting of two delay interferometers (DI) with the same differential delay. The red
arrow indicates that the additional Ts/4 delay can be moved out of the second DI. c The structure
comprises of two identical DI which can be replaced by a single DI followed by signal splitters.
d Resulting low-complexity scheme with combined S/P Conversion and FFT. In this picture,
rectangular sharp-edged boxes stand for optical directional couplers with an (unphysical) P
amplitude split ratio of 1. We follow the convention that the eld in the upper output arm
results from summing the complex amplitudes of the two coupler inputs, while the eld in lower
output arm D stands for the difference (scheme modied from [10, 16])
8.3.2.3 FFT Approximated by DI and Conventional Filters
The optical FFT schemes discussed above are perfect optical analog implementa-
tions of the FFT operation. However, in many instances an approximation would be
sufcient. The cascaded delay interferometer FFT as derived by Hillerkuss [10] will
lead us to approximations of the FFTs. We again follow [10, 16] to derive said FFT
approximations.
An approximation to an FFT is a lter that as closely as possible approximates the
ideal frequency response of the FFT. To nd an approximation to the FFT it is useful
to study effects of the delay interferometers (DI) in the FFT DI cascade. We consider
(a) Cascaded DI FFT T/8

Gate
0 Gate
T/4
0 Gate
T/8
p /2 Gate
T/2
0 Gate
T/8
p /4 Gate
T/4
p /2 Gate
T/8
3p /4 Gate
(b) Filter Response
Fig. 8.8 a Cascaded delay interferometer (DI) FFT of order N 8. b Frequency response of FFT.
The frequency response of the rst DI is plotted in red, the response of the second DI stage is
plotted in blue and the response of the third stage is plotted in black. The overall response of the
^ ^
output with response c i0 and c i4 is plotted in green. It can be seen that the rst DI stage has the
largest impact on the overall response, see [10]
the FFT of order N 8 in Fig. 8.8a. The frequency response of the 0th and 4th FFT
coefcients are plotted in Fig. 8.8b. From the gure one can see the effect of the
response of the rst DI stage (plotted in red), the second DI stage (plotted in blue) and
the third DI stage (plotted in black) and the overall response of all DI providing the
^ ^ ^
elements c i0 and c i4 . If we consider e.g. the response for the element ci4 , it can be seen
that the largest effect on the response is from the rst DI stage. The subsequent 2nd
stage still changes the overall response but it mainly serves as lter to suppress the
elements 2 and 6. Similarly the 3rd stage is needed to suppress the 0th element. One
could thus conclude that any lter combination that provides a sufciently good
approximation to a DI stage might serve as a good approximation to the FFT.
By the arguments of the discussion with Fig. 8.8 it follows that in most cases a
good approximation to an FFT of any order may be found by a simple single-stage
DI together with a bandpass lter and a subsequent sampling gate [10]. In the spirit
of the arguments given above and derived in [10] one can search for other lter
schemes with similar frequency responses. One will then nd that e.g. ring lters
[24] might provide a sufcient approximation as well.
The effect on the eye diagram and the signal quality as a function of the number
of DI stages is shown in Fig. 8.9. Calculations have been performed for a 20 GBd
OFDM signal with 8 subcarriers equispaced by 25 GHz.
In Fig. 8.9a we show the eye diagram and signal quality as a Q2-factor in dB.
One can see that the DI stage FFT provides an almost ideal eye diagram with close
(a) 8-FFT (b) 2-FFT with Filter (c) Filter only
Gauss Filter
3-Stage DI 1-Stage DI + Gauss Filter
8-FFT 2-FFT+Gauss Filter Gauss-Filter
8-FFT (no filter)

30
10 log10 (Q)
20
4-FFT
10 2-FFT
Gauss Filter
0
0 50 100 150 200
Optical filter bandwidth [GHz]
Fig. 8.9 Three different implementations to an 8 point FFT. a 8 point FFT with 3 delay
interferometers (DI) providing an ideal FFT operation providing a wide and open eye diagram and
a close to perfect signal quality. b Approximation to the FFT with a single stage DI and a
subsequent Gaussian lter to approximate stage 2 and 3 of the DIs. The eye is still open and the
signal quality is still close to 20 dB if the Gaussian bandpass lter is chosen with a proper
bandwidth. c The DI stages are replaced by a single Gaussian bandpass lter. The frequency
response is quite degradedeven if an optimum optical lter bandwidth is chosen. (Experiment
performed for a 20 GBd NRZ signal with a 25 GHz subcarrier spacing). The gure has been
modied from [10, 16]
to perfect signal quality. When the stages 2 and 3 of the DIs are replaced by a
Gaussian lter, the eye degrades and the signal quality drops, see Fig. 8.9b. Yet the
quality is still sufcient, particularly, if the optical lter bandwidth of the Gaussian
lter is properly chosen (around 50 and 75 GHz). If the DI stages are replaced by a
Gaussian lter the FFT still provides an eye diagram. However, the eye is degraded
and the signal quality is rather loweven at the optimum lter bandwidth around
25 GHz, see Fig. 8.9c.
8.3.2.4 Free Space Fourier Transform
A lens is a Fourier transform element on its own. For a lens to work as a Fourier
element the temporal samples of the input signal need to be serial-to- parallel (S/P)
converted and mapped onto an object plane x. The lens then maps the object into
the image plane, which represents a Fourier image of the object.
In Fig. 8.10 we have depicted the situation. The input time samples cin need to be
mapped (S/P converted) onto the positions xn in the focal plane at z 0 of the input
object. The lens with focal length l then maps the input points onto an image plane
at z 2l. Each image point in the image plane is the product of a coherent
superposition of all input image samples. The superposition of the discrete input
images happens to be the DFT as discussed above. A thorough mathematical
analysis has been given in [16]. Figure 8.10 has been repeated here for the sake of a
complete discussion of the topic.
Superposition of all
in
gives
. . .. . .
... ...
Fig. 8.10 Fourier transform lens with a focal length l mapping the input samples cin at positions xn
of the input image at z 0 onto output images at z 2l, where each image happens to be the
superposition of the input images with the phase weights of the DFT. The lens therefore acts as a
Fourier lens as long as the paraxial approximation holds. The gure has been modied from [16]
S/P DFT Gating
Outputs
Input
Time Delays
Input Slab Coupler
with Equal Phase Shifts Output Fourier Lens
Fig. 8.11 DFT based on an arrayed waveguide grating router (AWGR). The input slab coupler
with the arrayed waveguides acts as a S/P converter. The subsequent slab coupler at the output acts
as a Fourier lens that superposes the input coefcients according to the DFT expression (gure
from [16])
8.3.2.5 The Arrayed Waveguide Grating Router DFT
The function of a lens can be mimicked in integrated optics by means of an arrayed

waveguide grating router (AWGR) [25]. A discussion of the AWGR as a Fourier
lens for a discrete set of input samples has been given by Lowery [26].
In Fig. 8.11, the AWGR input slab coupler illuminates all grating waveguides with
equal amplitude and phase. The increasing lengths of grating waveguides provide time
delays incremented by Ts/N for providing a S/P conversion. This provides the N time
^
samples cin to specic locations of the second slab couplers. The slab coupler is then
designed such that it acts as a Fourier lens mapping and superposing the N input
^
samples cin onto the output plane of the slab coupler. If the locations are properly
chosen, the input samples will have equal weight with the phase coefcients of the
DFT. Consequently, one can extract the DFT coefcients at the output of the AWGR.
Experimental demonstrations based on the AWGR concept have been given by
various groups, so for instance by Lee et al. [27], Lowery et al. [28] and others. A
challenge of the AWGR concept is that all the N 1 phase shifters have to be
accurately adjusted relative to each other.
8.3.3 OFDM Tx and RxExperimental Implementations
In the previous sections we have studied the various possibilities to generate and
receive OFDM signals. OFDM signals can be generated and detected by either
using pulse-shaping techniques or by applying DFT techniques. Likewise we can
either encode or decode OFDM in the electronic domain or the optical domain.
8.3.3.1 Electronic OFDM Implementations
Electronic OFDM transmitters are preferred when many lower bit-rate subcarrier
signals are to be multiplexed to a total data-rate of a few 100 Gbit/s. A complete
real-time OFDM transceiver operating at 11 Gbit/s has already been shown [29] and
more recently the rst 256 Gbit/s OFDM transmission has been demonstrated [30].
It is expected that electronic approaches will reach Tbit/s capacity in the future.
Electronic implementations typically generate OFDM signals by carrying out a
serial-to-parallel conversion of the input data stream and then performing the IDFT
electronically as outlined in in the previous section. After performing the IDFT, the
signal is usually clipped before being passed on to digital-to-analogue (DAC)
converters. The real and the complex part of the signal (the so-called inphase and
quadrature part) are then separately fed to an optical IQ-modulator in order to
encode the electrical signal on an optical carrier. Electronic implementations may
flexibly adapt the order of the FFT and the symbol rate to changing channel
conditions and customer requirements. With such transmitters OFDM signals with
very few up to a 1024 of subcarriers have been implemented in optical commu-
nication systems [31]. On the other hand, electronic OFDM transmitters have a
speed limitation which is due to given electronic circuit speeds. Electronics also
scales not very favorable with power consumption, particularly at higher speed.
Conversely, optical OFDM concepts have advantages when data channel speed
exceed the speed offered by electronics. In this context they are most interesting in
view of a next generation Tbit/s superchannel implementation. Various concept
with 2, 4 or 8 OFDM subcarriers are currently discussed as possible candidates of a
1.2 Tbit/s superchannel. Likely scenario include the generation and reception of the
higher speed subcarriers in the optical domain whereas the subcarriers themselves
could carry lower speed data signals based on electronic OFDM or Nyquist mul-
tiplexed sub-sub carriers.
8.3.3.2 Optical OFDM Implementations
Optical OFDM subcarrier generation and reception however will only be pursued if
signals can be generated and received at highest quality with the least effort. We
have evaluated various all optical concepts such as discussed above and now pick
the one concept that in our opinion works most reliably with the least effort.
For the all-optical generation of an OFDM signal we have selected the OFDM
transmitter based on optical pulse shaping which follows the general concept
outlined in Fig. 8.4a. In this concept the OFDM signal is generated by applying
(8.6). In Fig. 8.12a we have depicted an example for an implementation of such an
optical OFDM transmitter. The transmitter comprises of lasers emitting at fre-
quencies f0, f1, , fN1 equispaced by Fs. The carriers are encoded with information
^ ^ ^
ci0; ci1 and ciN1 by means of an optical IQ-modulator. In this implementation, the
IQ-modulators serve two purposes. They encode information and carve a
(a) OFDM Tx based on Pulse-Shaping In the Optical Domain

LD Shaping and E/O
...
(b) OFDM Rx exploiting optical DFT

S/P and DFT Gating O/E
~
~
~ ci0
...
...
...
...
0
Fig. 8.12 Schematic of an optical OFDM transmitter and receiver. a Optical OFDM transmitter
based on the optical pulse shaping concept. b OFDM receiver relying on the cascaded delay
interferometer approach with a Gaussian lter to perform the FFT in the optical domain
rectangularly shaped pulse hs onto the carriers. Finally, the subcarriers are multi-
plexed e.g. by guiding them through an optical combiner.
The OFDM receiver on the other hand is based on an implementation of the
DFT. We have chosen to do so, because receiver pulse shaping in the optical
domain is not straight forward and digital signal processing at highest speed is not
really an option. Implementation of the all-optical DFT approach is however quite
straight forward and works at almost any speed. In Fig. 8.12b we show the
implementation that has rst been introduced in [10] and that ultimately was used to
process a 26 Tbit/s OFDM superchannel [9]. In this work a single stage DI followed
by a flat-top passpand has been used, see Fig. 8.9b. The advantage of this concept is
that this way a single DI is sufcient to rst demultiplex the OFDM signal into even
and odd channels. This DI also is the rst stage in the cascaded delay interferomter
FFT approach discussed in Fig. 8.9b. Subsequently, the second stage lter only
needs to suppress the subcarriers in every 2nd channel. A simple Gaussian bandpass
lter or flat-top ll can do it. Finally, gating with a Gating window of a length Ts/2
is applied and the optical signal is down converted into the electronic domain by
means of a coherent receiver.
8.3.3.3 Optical OFDM of a 26 Tbit/s Signal
To demonstrate the strength of the optical OFDM techniques we would like to

conclude this section with a brief discussion of the 26 Tbit/s optical OFDM gen-
eration, transmission and reception experiment published in [9].
The experimental setup is shown in Fig. 8.13. Rather than multiple lasers, a
single mode-locked laser (MLL) was used to generate an optical comb with a total
of 325 subcarriers and a frequency separation Df 12.5 GHz. The frequency comb
with its large amount of subcarriers has been generated by launching the MLL in a
highly nonlinear bre for broadening [32]. Subsequently, a programmable optical
lter equalizes the power of the subcarriers and separates them into even and odd
subcarriers. Both sets of subcarriers are then individually modulated with inde-
pendent 16QAM pseudo-random bit sequences (PRBS 215-1).
The OFDM signal with a power of 9.5 dBm is then launched into a single-mode
bre SMF-28 with a length of 50 km and a span loss of 10.2 dB. After transmission,
the signal is amplied and the dispersion of the link is compensated in a dispersion
compensation module. The total insertion loss of the module was 4.8 dB and the
residual dispersion of the link was 17 ps/nm.
The receiver is based on the optical FFT using a single delay interferometer
followed by a lter in the form of a wavelength-selective switch (Finisar
Waveshaper). The time gating was performed by an electro absorption modulator
(EAMCIP 40G-PS-EAM-1550) at 12.5 GHz. The subcarriers are then converted
to the electrical domain by a coherent receiver in the form of an OMA (Agilent
N4391A) with an electrical bandwidth of 16 GHz (Agilent Inniium
DSOX93204A). In the OMA, the signal is equalized with a 21 tap equalizer.
Typical constellation diagrams are depicted in Fig. 8.13c. To assess the quality of
the received signal in each polarization, the error vector magnitude is calculated and
averaged over 10 received sequences of 1024 symbols. The error vector magnitude
(EVM) is an important quality metric for complex signalssimilar to what the
quality factor is for on-off keying signals [33]. The EVM measurement was carried
out for all 325 subcarriers of the OFDM signal, see Fig. 8.13d. It can be seen that
the EVM for all subcarriers after transmission are between 11 and 16 % with an
average EVM of 14.2 %. The estimations indicate that all 325 subcarriers are below
the BER 1.9 102 third generation FEC limit with 20 % overhead [34]. The
demonstrated line rate of 26 Tbit s1 therefore corresponds to a net error-free data
rate of 21.7 Tbit s1. The achieved spectral efciency for the modulated subcarriers
is 6.3 bit/s/Hz.
8.4 Nyquist Pulse Shaping
The combination of several lower-rate channels with high spectral efciency into
Tbit/s super-channel can be achieved in the time domain by appropriate pulse
shaping. For Nyquist transmission with a maximum spectral efciency, symbols are
(a) All-Optical OFDM Transmitter

LO: Comb Generator
MLL
100 m
12.5 3 nm
GHz
HNLF WSS
OFDM Tx
Shaping & DAC IQ-Mixer Mux
FPGA DAC
Polarization Multiplex
90
5.3 ns
FPGA DAC
(b) Transmission Link (c) All-Optical OFDM Receiver

Rx S/P & OFFT Gating IQ-Mixer
SMF-28 DCF Optical
Modulation
Gate Analyzer
1
EAM
50 km
40 ps DI
WSS
1547.0 nm 1551.3 nm
EVM 14.0 % EVM 11.1 %

(d) 26 Tbit /s OFDM Transmission Result
Fig. 8.13 Setup of the single laser 26 Tbit/s OFDM experiment with a optical OFDM transmitter,
b transmission link, c all-optical FFT receiver and d a plot with error-vector magnitudes of each
and every subcarrier. Figure modied from [9, 16]
carried by Nyquist pulses [11] that meet the condition of (8.1). If the maximum
spectral efciency is not needed, the Nyquist criterion for intersymbol interference
(ISI) is good enough, according to which impulse maxima coincide with the zeroes
of adjacent pulses such that the communication channel is ISI free. Nyquist pulses
that are shaped to meet the Nyquist criterion but do not offer the maximum spectral
efciency have already been used to demonstrate some quite impressive data rate
transmission experiments [3538]. Compared to OFDM, Nyquist pulse shaping has
several unique advantages: reduction of the transmitter and receiver complexity
[3941], lower receiver bandwidths requirement [42], lower peak-to-average power
ratios [11, 40], reduced sensitivity to bre nonlinearities [41]. Nyquist signaling is
also known as an orthogonal time-division-multiplexing technique, because
although each symbol pulse spreads into many adjacent time slots, no ISI occurs if
the sampling is performed at the center of each symbol. The Nyquist criterion for a
channel impulse response h(t) satisfying the zero ISI implies that, for a particular
sampling period TS and any integer i:

1; i 0
hiTs 8:21
0; i
6 0
When such a signal is periodically sampled with a period TS, a nonzero value is
obtained only at the time of origin. In the frequency domain, the Nyquist criterion is
equivalent to

1 X 1
k
H f 1 8:22
Ts k1 Ts
with H(f) the Fourier transform of h(t). From (8.22) it is intuitively understood that
aliased versions of the frequency domain H(f) must add up to a constant. These two
concepts, time and frequency Nyquist criterion, are illustrated in Fig. 8.14a, b,
respectively. The pulse in Fig. 8.14a is at its maximum when t 0, while null at all
other sampling locations corresponding to t iTS, with i an integer. The maxima of
adjacent pulses coincide with zeroes of the initial pulses while the sum of all other
pulses at t 0 is equal to zero, thus leading to no ISI at this specic sampling time.
In the frequency domain, i/TS frequency shifted versions of the pulse spectrum add
up to a constant. From such concept, it can right away be inferred that the minimum
bandwidth Nyquist pulse would have a rectangular spectrum.
(a) h(t) (b) H(f)

H(f+1/Ts) H(f-1/Ts)
t f
-2Ts -Ts 0 Ts 2Ts -1/Ts 0 1/Ts
Fig. 8.14 a Raised cosine response (solid line) meets the Nyquist ISI criterion. The maximum is
at t 0 while h(t) is zero at all other sampling locations given by t iTS, i an integer. Consecutive
raised cosine impulses (dashed lines) demonstrate the zero ISI property: at t 0, the sum of all
other impulses is zero. b Aliased versions of the frequency domain impulse add up to a constant
The general expressions for the pulse waveform h(t) and corresponding spectrum
H(f) are given by the impulse response and transfer function of a raised cosine
Nyquist lter, respectively [43, 44]:

sin pt
Ts cos bpt
Ts
ht pt 2 8:23
Ts 1 Ts
2bt
8
>
< Tsh jfj 1b
h ii 2Ts
H f Ts
1 cos pTs
j f j 1b 1b
j f j 1b 8:24
>
: 2 b 2Ts 2Ts 2Ts
0 otherwise
The factor is known as the roll off factor [43], which is in the range 0 1.
Amongst the class of Nyquist pulses [11], the sinc-shaped pulse obtained for 0
is of particular interest owing to its rectangular spectrum. It is therefore the mini-
mum bandwidth Nyquist pulse and such feature allows minimizing the guard band
between optical channels. As the roll off increases, the required bandwidth increases
but a smoother time response is obtained, see Fig. 8.15.
(a) h(t)
=1
=0
=0.5 =0.3
-4 -3 -2 -1 0 1 2 3 4
t/TS
(b) H(f)
=0
=0.3
=0.5
=1
-1 -0.5 0 0.5 1
fTS
Fig. 8.15 a The pulse waveform h(t) is given by the impulse response of a raised cosine lter
shown here for different values of the roll-off factor . The time response becomes smoother as the
roll-off increases. b The corresponding spectrum H(f), also shown for different values of , shows
that the required bandwidth increases with while a rectangular spectrum is obtained for 0,
corresponding to the sinc-shaped pulse
Theoretically for a sinc-pulse Nyquist transmission, each symbol consists of a

time unlimited sinc-pulse. As a consequence of causality, the sinc function
sinct sinpt=pt is therefore only a theoretical construct. Instead, periodic
pulses are typically used in most experimental demonstrations of Nyquist pulse
transmission [13, 3942, 4446]. Several approaches for the generation of Nyquist
pulses have been suggested and can be divided into three main categories
depending on whether pulse shaping is performed in the electrical [38], digital [35]
or optical domain [40]. Electronic Nyquist processing consists in utilizing analog
electrical lters to form an appropriate output signal with an ideal rectangular
shaped spectrum. Digital signal processing is based on the ISI free shaping of a
baseband signal in a digital signal processor (DSP) that is part of the transmitter.
Finally, Nyquist pulses can be directly shaped in the optical domain by relying on
optical ltering or frequency comb synthesis. While at low symbol rates
(<25 Gbaud) all shaping methods are available, at higher symbol rates, only analog
or optical methods are at hand [47]. Additionally, optical pulse shaping can lead to
pulses with much shorter duration (in the picosecond or less). The three methods
are described in the following sections.
8.4.1 Electronic Nyquist Processing
Pulse shaping can be done in the electrical domain by solely relying on an electrical
lter to approximate sinc-shaped pulses by analog means. The electrical lter
shapes the electrical drive signals that are then fed to the optical modulator as
shown in Fig. 8.16 for a software dened transmitter.
This technique was applied to demonstrate the long haul transmission of
28 Gbaud 16QAM signals on a 50 GHz grid, achieving a spectral efciency of
4 bits/s/Hz [38]. Passive low pass electrical lters with bandwidths of 14 GHz were
used for the pulse shaping in order to reduce spectral overlap between adjacent
wavelength division multiplexed channels. For the generation of sinc-shaped pul-
ses, the electrical lter should have very steep slope, in order to get as close as
possible to a rectangular spectrum. However such lters are not easily available
while a complex transmitter would necessitate multiple of these analog lters.
Nyquist filter electrical

DSP DAC optical
Modulator Out
ECL
Fig. 8.16 Setup for Nyquist pulse shaping in the electrical domain using an electrical lter. The
electrical lter solely performs pulse shaping of the signals. The electrical signals are then fed to an
optical modulator which encodes the data onto an external cavity laser (ECL)
In [47], pulse shaping in the electrical domain was investigated in a setup

generating 20 GBd QPSK. In addition to Nyquist pulse shaping at the transmitter,
the receiver was optimized with a sophisticated equalization technique based on
Nyquist brick-wall ltering. The lter used in the experiment had 3 and 20 dB cut-
off frequencies of 12.3 and 12.5 GHz, respectively as from the simulated S-
parameters in Fig. 8.17a. A typical received and assembled averaged power spec-
trum of a 20 GBs QPSK signal at an optical signal to noise ratio of 30 dB before
and after the receiver DSP is shown in Fig. 8.17b, top and bottom respectively. The
individual pulse form was derived from the un-equalized signal spectrum.
The electrical pulse shaper produces sinc-typical side lobes but they decay rapidly
as seen in Fig. 8.17c.
(a)
0
S-Parameters [dB]
-10
-20
-30
-40
-50
0 4 8 12 16 20
Frequency [GHz]
(b) (c)
Power Spec. [W/10 MHz]
10-5
Amplitude [a.u.]
10-7
10-9
10-4
10-6
10-8
10-10
-20 -10 0 10 20 -500 0 500

Frequency [GHz] Time [ps]
Fig. 8.17 a Simulated S-parameters of the electrical low pass lter employed in [47]. The
measured cut-off frequency is 12.3 GHz and reflections are suppressed by at least 20 dB.
b Measured and assembled averaged spectrum of a 20 GBd electrically generated QPSK signal
before (top) and after (bottom) receiver DSP. As expected the overall bandwidth corresponds to
two times the electrical lter bandwidth. c Impulse response measured from the received un-
equalized spectrum. The electrical pulse shaper produces sinc-typical side lobes but with a faster
decay. Figure modied from [47]
8.4.2 Digital Signal Processing Based Generation
The use of electronics within the optical transceiver, such as eld programmable
gate arrays (FPGA), high speed digital-to-analog and analog-to-digital converters
(DAC, ADC), has enabled real time signal processing and has facilitated the use of
advanced modulation formats in optical communication systems. Such approach is
commonly applied to optical orthogonal frequency division multiplexing
(O-OFDM) for the demonstration of real time transmission with very compact
spectrum [9, 48]. Similarly, software dened optical transmitters acting as arbitrary
waveform generator (AWG) programs can be used for the real-time generation and
transmission of Nyquist pulses [13]. Nyquist pulses are generated by ISI free
shaping of a baseband signal in a digital signal processor (DSP) that is part of the
transmitter as shown in Fig. 8.18.
Electrically generated Nyquist pulses are shaped with a nite duration impulse
response (FIR) lter. While an elementary Nyquist shaped impulse with minimum
spectral width is a sinc-function extending innitely in time, real Nyquist pulses
need to be approximated by a nite length representation. In addition, the baseband
spectrum needs to be separated from its periodic repetitions (aliases) using realiz-
able lters. To such ends, oversampling by a factor q is needed. The standard
oversampling factor of q 2 saves FPGA resources since sampling points of
adjacent symbols fall onto the same time slot but limits the symbol rate to half of
the sampling frequency of DACs. Smaller oversampling factors, such as q 1.33,
have been demonstrated but at the cost of increased processing complexity [30].
An FIR lter of order R can be constructed by a sequence of R delay elements of
value Ts/q and R + 1 taps in between as shown in Fig. 8.19. The tapped signals are
weighted by the lter coefcients kr, also called tap weights, and summed up to
form the lter output. When R 0, the lter simply reproduces the input. As the
lter order gets larger, the closer the spectrum can approach a rectangle and thus
reaching roll-off factors close to 0.
The order R of the lter has a signicant effect on the spectrum. Signal gener-
ation with various FIR lter orders R and q 2 are shown in Fig. 8.20, see [30].
The left column shows the simulated impulse response of each lter. The effective
anti-alias
filter
Baseband
DAC
data signal
DSP Nyquist filter
Modulator Out
ECL
Fig. 8.18 Shaping (pre-ltering) of the baseband data signal for Nyquist pulse shaping in the
digital domain. The digital lter is part of the digital signal processing (DSP) in the transmitter.
Pulse shapes can be programmed offline. An off-the-shelf analog electrical lter is used after the
digital-to-analog conversion to remove spurious spectra. The electrical signal is then fed to an
optical modulator which encodes the data onto an external cavity laser (ECL)
tap 0 tap 1 tap 2 tap R

Filter Ts/q Ts/q Ts/q
input
k1 k2 kR
Filter
output
Fig. 8.19 A nite impulse response lter (FIR) of order R. The top part is the R stage delay line
with R + 1 taps. A delay of Ts/q is positioned between each tap. The tapped signals are weighted by
R tap weights kr and summed to form the lter output. Figure modied from [30]
windowing is indicated by a green rectangle. The corresponding simulated transfer

functions on a linear and logarithmic scale are shown in the second and third
column, respectively. An FPGA based DSP was implemented to realize such real
time digital Nyquist pulse transmitter. Within the FPGA, complex Nyquist pulses
are calculated from a 215-1 pseudo random bit sequence in real time and signals pre-
shaped by FIR lters with order 16 and 32 were implemented.
The convolution of the rectangular spectrum of an innitely extended temporal
sinc-pulse with the sinc shaped spectrum of the rectangular time window leads to
the power spectra shown in the two middle columns. As the lter order increases,
the spectrum evolves towards an ideal rectangle with a spectral width equal to the
Nyquist bandwidth. For R 32, a signicant increase in spectral efciency can be
already obtained in comparison to NRZ modulation. Experimentally, this method
can provide a very good roll-off factor of 0.0024 and has been applied for the
demonstration of 512QAM Nyquist sinc-pulse transmission at 54 Gbit/s in an
optical bandwidth of 3 GHz [14]. This method is however restricted by the speed of
the electronics because of the limited sampling rate and limited processor capa-
bilities, whereas the quality of the Nyquist pulse highly depends on the resolution of
the digital-to-analog converters.
8.4.3 Optical Processors
Compared to pulse shaping in the electrical or digital domain, optical pulse shaping
is not entirely limited by electrical components or processors capabilities. Optical
techniques can also lead to pulse sequences with much shorter duration which can
subsequently be multiplexed in time to reach an ultrahigh symbol rate. To optically
generate Nyquist pulses, liquid crystal spatial modulators have been used to shape
Gaussian pulses from a mode-locked laser into raised-cosine Nyquist pulses [44]. It
is also possible to generate Nyquist pulses using ber optical parametric ampli-
cation pumped by parabolic pulses in combination with a phase modulator to
compensate the pump-induced chirp [46]. Additionally, sequences of very high
Time Domain Frequency Domain: power spectrum

linear logarithmic
(a) Ts Fs 1
R =1
R=1
NRZ
0 t/Ts 0.5 0 0.5 f/Fs

(b)
R =16
R = 16
0 t/Ts 0.5 0 0.5 f/Fs

(c)
R = 32
R = 32
0 t/Ts 0.5 0 0.5 f/Fs

(d)
... ...
R = 1024
R=
1024
0 t/Ts 0.5 0 0.5 f/Fs

f/Fs
Fig. 8.20 Impulse responses and transfer functions of FIR lters with various orders R. The rst
column shows the simulated impulse responses. The second column shows plots of simulated
power spectra on a linear scale and the third column shows the same spectra on a logarithmic scale.
For a a single NRZ impulse shaped by a lter of order R 0. b An ideal sinc pulse truncated by a
lter with order R 16. c An increased order lter to R 32 resulting in a spectrum evolving
towards an ideal rectangular shape. d A lter with very high order R 1024 closely
approximating a rectangular shaped spectrum. Figure has been modied from [30]
quality Nyquist pulses ( close to 0) can be obtained by the direct synthesis of a flat
phase-locked frequency comb with high suppression of out-of-band components
[49]. In the following sections, ltering based pulse shaping and frequency comb
synthesis are discussed in more details.
8.4.3.1 Filtering Based Pulse Shaping
Filtering of the optical signal can be done in order to achieve the best possible
transmission quality. As for electrical lters, the optical lter must result in a
spectrum of rectangular shape by removing side lobes and flattening the region of
interest. An optical band pass lter can be directly applied to a conventionally
modulated signal [47]. In this conguration, the spectrum of the data signal is
reshaped as a bandwidth limited data signal and the obtained pulse width remains
close to identical to the allocated bit interval. Optical interleavers can also be used
for xed-frequency optical lters with a nearly rectangular frequency response. An
alternative for a more flexible system is the use of a programmable optical lter,
which comprises of general imaging optics, a diffraction grating and liquid crystal
on silicon (LCoS) cells for shaping the phase of the optical spectrum [50].
In view of the high potential of optical ltering methods for the generation of
ultra-short duration pulses, Nakasawa et al. proposed a slightly different approach
consisting in the synthesis of a Nyquist pulse train at a repetition rate Fs in which
the pulse width is set to be much shorter than the bit interval given by 1/Fs [44]. The
output of an initial optical pulses source, such as a modelocked laser, with a
repetition rate Fs and Gaussian pulses, is shaped by the liquid crystal spatial
modulators into raised-cosine Nyquist pulses. The concept is illustrated in Fig. 8.21.
For a spectrum reshaped around the optical carrier fc with a width of NFs, the
symbol period dened as the peak-to-rst-zero time interval is then given by
Ts 1/(NFs).
Optical Nyquist pulse generation based on this ltering technique produces
much higher roll-off factors compared with the electrical method because optical
(a) (b) (c)
Fig. 8.21 Schematic of pulse shaping of the output of a pulses source with Gaussian shaped
pulses into sinc-shaped Nyquist pulse train using an optical lter. a An optical pulse source with
repetition rate Fs outputs a train of Gaussian shaped pulses with repetition period 1/Fs and
corresponding optical spectrum comprising of several spectral lines spaced by Fs and centered at
the carrier frequency fc. b The optical lter reshapes line by line the spectrum of the pulse train
such that it is conned to a bandwidth of NFs, with N the number of spectral lines. The
corresponding pulse train shows sinc-like oscillations with a peak-to-rst-zero spacing of 1/NFs.
c The obtained pulse train can then be modulated by going through an optical modulator driven by
the electrical signal, here on-off-keying shown
lters only produce slopes with limited steepness. A programmable optical lter
like a Waveshaper has a minimum 3 dB bandwidth of around 12 GHz and a 30 dB
bandwidth of around 40 GHz. The roll-off factor also strongly depends on the
overall bandwidth of the signal and typically ranges between 0.15 and 0.5 [51, 52].
However the obtained pulses are characterized by a peak-to-rst zero-crossing
separation of 1/NFs for a 1/Fs repetition period, such that the Nyquist pulse stream
can be multiplexed in time domain to increase the overall transmission rate by a
factor of N. This is designated as orthogonal optical time-division multiplexing
(OTDM). The generated sequence is split into N channels, which are then delayed
and modulated to transport the channel corresponding data. This requires N mod-
ulators, with N being the number of branches or the number of time domain
channels. However compared with direct modulation, the baud rate of each mod-
ulator is N times reduced which drastically relaxes the requirements on modulators
and electronics (Fig. 8.22).
Lets dene the time interleaved optical Nyquist pulses as gi(t) h(t iTs), where
h(t) is given by (8.23), and the symbol data at t iTs as ci. For example, if we consider
on-off-keying modulation ci can take value 1 or 0. When 0, according P to the
sampling theorem, the time division multiplexed Nyquist data train st ci gi t
is equivalent to an analog signal s(t) given by the interpolation of ci
X
st siTs gi t 8:25
i
The relationships between s(t), gi(t) and ci s(iTs) are shown in Fig. 8.23, for the
case i 4. Time division multiplexing of Nyquist pulses can be viewed as a digital-
to-analog conversion from ci to s(t) in the optical domain. Based on this property,
the symbol data ci can be extracted from the time division multiplexed stream by
use of narrow optical sampling gate at t iTs, as illustrated by the grey box in
Fig. 8.23. Optical sampling gates based on nonlinear effects have been successfully
used for the demultiplexing of time division multiplex Nyquist pulses proving the
potential of such technique, despite the fact that the sampling gate cannot be an
ideal delta function thus leading to leakage from adjacent time channels. The simple
example drawn in the gure, also illustrates the orthogonality of the time
0 M0
Ts M1
2Ts M2
Time Time
(N-1)Ts MN-1
Fig. 8.22 Periodic sinc-pulse sequences can be split into N branches, each of which corresponds
to an independent channel. In the nth branch, the sequence is delayed by n times the interval
Ts 1/NFs, with n 0, 1 N 1. Each channel can be modulated independently with a
modulator M0, MN1 with any modulation formats. The shown multiplexing is carried out in the
time domain at one carrier wavelength
ci=s(iTs)
s(t)
gn(t)
1 2 3 4 5 6 7 8 9 10
Time (t/Ts)
Fig. 8.23 Illustration of orthogonal time division multiplexing of Nyquist pulse train. The
individual channels gi(t) (dashed lines) are time multiplexed with a delay multiple of Ts 1/NFs.
At times t iTs, the symbol data is given by ci (in this example 0 or 1). The time division
multiplexed Nyquist data train is s(t). The symbol data di can be extracted by optical sampling of s
(t) at t iTs. The narrow gate (shown here for i 4) and the orthogonal properties of the Nyquist
channels enable recovering of the data despite strong overlap between channels
interleaved optical Nyquist pulses in spite of the strong overlap with adjacent
pulses. Indeed the time interleaved optical Nyquist pulses gi(t) s(t iTs) satisfy
the orthogonality condition when 0
Z/
1 0 i 6 m
gi tgm tdt 8:26
TS 1 i m
/
For a general value of , the integral of (8.26) has generally a non-zero value for
i m and the orthogonality is not rigorously satised theoretically. However, the
integral converges to 0 for large |i m| such that the quasi-orthogonality property is
maintained even when 0.
8.4.3.2 Frequency Comb Based Nyquist Pulse Train Generation
Considering that owing to physical limitations the ideal sinc pulse with perfect
rectangular optical spectrum has not been demonstrated yet, a different optical
approach was proposed by Soto et al. [49]. The technique is a straightforward way to
realize sinc-shaped Nyquist pulses in the optical domain and is based on the time-
frequency duality described by Fourier domain. It is well known that a sinc pulse can
be represented by a rectangular spectrum in the Fourier domain, while the frequency
content of a train of sinc pulses corresponds to a flat comb with equally spaced
components within the bandwidth dened by the single-pulse spectrum (Fig. 8.24).
Therefore instead of shaping a single sinc pulse, a sequence of sinc pulses is
directly produced by the generation of an optical frequency comb having uniformly
(a) (b)
Nf
Time TS=1/(Nf) Frequency
(c) (d)
T=1/(f) f
Nf
Time Frequency
Fig. 8.24 Time (left) and frequency (right) representation of a single-sinc pulse (top) and a sinc-
pulse sequence (bottom). The gure shows the intensity in the time domain since the directly
observed quantity is proportional to the intensity. The Fourier domain representation of a sinc
pulse (a) is a rectangular function (b), while the spectrum of an unlimited sing-pulse sequence
(c) is a frequency comb with uniform phase under a rectangular envelope (d)
spaced components with narrow linewidth, equal amplitude and linear phased locked
phase, together with strong out-band suppression [26]. While properties similar to
that of the ltering based methods are expected, such as the possibility of utilizing
such generated Nyquist pulse train for orthogonal time division multiplexing, this
method leads to the generation of a sequence of very high quality Nyquist pulses with
an almost rectangular spectrum ( close to 0).
The time-domain representation of the optical eld of a frequency comb with
N lines, having the same amplitude E0/N and frequency spacing f around the
central frequency f0 can be expressed as
X X
N1 N1
2 2
E0 2ipf0 nDf ti/ E0
E t e e2ipf0 ti/ e2ipnDft
N N
nN1
2 nN1
2 8:27
sinpNDft 2ipf0 ti/
E0 e
N sinpDft
This mathematical demonstration if given for an odd number of frequency lines and
the derivation for an even number can be straightforwardly obtained following the
same procedure. For simplicity, it is assumed that all frequency components have
the same phase . It is however sufcient that all frequency components be locked
showing a linear dependence. From (8.27), the normalized envelope of the optical
eld is calculated to be sinpNDft=N sinpDft referred thereafter as periodic sinc
function. Using Fourier transform, the frequency domain representation is given by
8 N1 9
< X2 = X
N1
2
sinpNDft
= = e 2ipnDft
df nD 8:28
sinpDft : ;
nN1
2 nN1
2
By introducing the rectangular function Pn=N that is 1 for all integers n where
jnj N 1=2 and 0 elsewhere, (8.28) can be written as
X
2
N1 X
1
f
df nD P df nDf 8:29
N1
NDf n1
n 2
where the rectangular spectrum Pf =NDf , covering a bandwidth Nf, is repre-

sented in the time domain by a sinc-pulse Nf sinc(Nft). The temporal dependence
of the above expression can then be obtained by taking its inverse Fourier transform
and using the Poisson summation formula
( X )
f 1 X
1
n
1
= P df nDf N sincNDft d t 8:30
NDf n1 n1
Df
where denotes the convolution operation. Considering the right hand side of
(8.30)
X
1 X
1
n n
N sincNDft d t N sinc NDf t 8:31
n1
Df n1
Df
From (8.28)(8.31) we can conclude that
X
1
sinpNDft n
sinc NDf t 8:32
N sinpDft n1 Df
Following similar procedure, the envelope of the optical eld in the case of an even
number of frequency lines can also be expressed as a train of sinc pulses. The
generalized expression for the normalized envelope of the optical eld resulting
from a flat frequency comb, independent of the parity of N
X
1
sinpNDft N1n n
1 sinc NDf t 8:33
N sinpDft n1 Df
The (1)n factor appearing for even values of N comes from the absence of a
spectral line at the central frequency of the comb, eliminating the direct current
(DC) component on the optical eld envelope. The difference in the periodic sinc
function xt sinpNDft=N sinpDft for even and odd number of frequency
components N is depicted in Fig. 8.25. For odd values of N, all sinc pulses of the
(a) (b)
Normalized amplitude
Normalized amplitude
1 1
0 0
-1 -1
-1 0 1 2 3 4 5 -1 0 1 2 3 4 5
Time [t/T] Time [t/T]
Fig. 8.25 Normalized eld envelope of a frequency comb for a an odd (N 9) and b an even
(N 8) number of spectral lines. An odd number of lines leads to a sequence of in-phase sinc
shaped Nyquist pulses, while an even number results in a sequence of alternated -phase-
modulated pulses
pulse train show the same phase and x(ts) 1 at every sampling instant ts n/f for
all integer n. For even N, each pulse envelope is of opposite sign with its preceding
and following pulses and x(ts) (1)n. However, for either odd or even N the
optical intensity measured by a photodetector is the same and given by
I t j E t j 2 :
Consequently the eld envelope of the time domain representation of a fre-
quency comb of N identical and equally spaced lines corresponds to an innite
summation of sinc-shaped Nyquist pulses with period 1/f and symbol period 1/
Nf. Considering that the pulse repetition period Ts 1/f is a multiple of the time
interval T 1/Nf, the resulting time-domain envelope x(t) satises the following
condition for any integer m

xmr 1N1m=N m . . .; 2N; N; 0; N; 2N. . . 8:34
0 otherwise
The sequence of sinc pulses resulting from a phase-locked, rectangular fre-

quency comb satises the Nyquist criterion for free ISI within every pulse repetition
period Ts. The generated sinc-pulse sequence can be multiplexed in time without
ISI, as described in the previous section. In addition, since the time multiplexed
channel shows a sharp-edged spectrum, the next wavelength channel can be directly
adjacent to the previous with almost not guard band while being multiplexed in the
time domain the same way, reaching high temporal and spectral densities together.
There are several approaches for the generation of a frequency comb. They can
be obtained from conventional femtosecond lasers [32, 5355] or from continuous
wave optical source exploiting Kerr nonlinearities in optical resonators [56, 57],
employing a combination of intensity and phase modulation [58], intensity and
phase together with dispersive medium [59], chirped Bragg gratings [60] or highly
nonlinear bers [6163]. However not every comb does necessarily result in a
sequence of Nyquist pulses. The sinc pulse sequence can only be obtained under
specic conditions, namely line amplitudes as equal as possible, linear phase
dependence through all lines and strong out-of-band lines rejections. Close to ideal
rectangular shaped optical frequency combs can be produced from a non-optimal

frequency comb in combination with a spectral line-by-line manipulation of the
optical Fourier components. A simple method consisting of a cascade of intensity
modulator will be presented in the following section.
8.4.4 Implementations
In this section, we describe the implementations of optical pulse shaping using a

programmable lter [44, 52, 64, 65] and frequency comb synthesis [66].
The transmitter in [44] consists of a 40 GHz modelocked ber laser. The output
pulses of the source are generated at 1540 nm and are characterized by a Gaussian
shape with 1.8 ps width. The spectrum was initially broadened by self-phase
modulation (SPM) in a highly nonlinear ber in order to enable the generation of
short Nyquist pulse. The Nyquist pulses with a peak-to-rst-zero time Ts 1.56 ps
and roll-off 0.5 are generated from this pulsed source by manipulating the
spatial intensity and phase of the modelocked laser spectral components using a
liquid crystal spatial modulator (Waveshaper). The output of the modelocked laser
shown in Fig. 8.26a is therefore reshaped according to the optical lter transfer
function shape depicted in Fig. 8.26b. The resulting Nyquist pulse spectrum and
waveform, Fig. 8.26c, d, respectively, have the typical raised-cosine prole. Due to
the limited steepness of the optical lter, the spectrum deviates from the ideal
rectangular spectrum resulting in damped sinc-like oscillations in the time domain.
These pulses where multiplexed to a 1.28 Tbit/s polarization-multiplexed DPSK of
640 GBd Nyquist OTDM and transmitted over 525 km.
In addition to Nyquist OTDM, the technique proposed in [44] was also applied
for Nyquist OTDM-WDM [51, 52]. The output of a 43 GHz, 400 fs optical pulse
generator is modulated by a 43 Gbit/s PRBS sequence and multiplexed in time to
generate the OTDM stream. The multiplexing scheme can be varied with factor
N from 1 to 8 in order to obtain signals from 43 to 344 GBd. A Waveshaper is used
for Nyquist ltering with a raised cosine function of bandwidth (N + 1) 43 GHz to
shape the N 43 GBd signal. The extra bandwidth is used for the purpose of clock
recovery. The WDM channels are generated by applying the Nyquist ltering at
different wavelengths, using the same Waveshaper. In addition the signals are
polarization multiplexed to double the total capacity. The transmitter is shown in
Fig. 8.27.
The spectrum of the mixed bitrate Nyquist OTDM-WDM shown in Fig. 8.28a
reflects the typical raised-cosine ltering performed by the Waveshaper. As men-
tioned in Sect. 4.3.1, due to the limited resolution of the programmable lter, the
steepness of the ltering transfer function varies with the overall bandwidth. Wider
Nyquist pulses are therefore characterized by a larger roll-off than shorter ones and
the performance between the different OTDM channels slightly varies. The corre-
sponding waveforms show the typical sinc-oscillations and a good tting with the
raised-cosine approximation (Fig. 8.28b). Due to the non-ideal rectangular spectra,
(a) (b)
0
Attenuation (dB)
10 dB/div
-20
-40
-60
1530 1535 1540 1545 1550 1530 1535 1540 1545 1550
(c) (d) 1.2

1
Intensity (a.u.)
0.8
10 dB/div
0.6
0.4
0.2
0
1530 1535 1540 1545 1550 1.56 ps/div
Wavelength [nm]
Fig. 8.26 Generation of Nyquist pulse from shaping the output of a modelocked laser. a Optical
spectrum of 40 GHz Gaussian pulse after self-phase modulation induced spectral broadening.
b Transfer function of the programmable optical lter based on liquid crystal spatial modulators
used to shape the spectrum shown in (a). c Optical spectrum after the programmable lter and
d corresponding waveform of the generated optical Nyquist pulses characterized by a peak-to-rst-
zero of 1.56 and 25 ps repetition time (adapted from [64])
Fig. 8.27 Optical transmitter for Nyquist TDM-WDM. The output of 43 GHz optical pulse
generator is modulated by a PRBS sequence and time multiplexed by a factor N(N 18).
A Waveshaper performs the Nyquist ltering at various wavelengths simultaneously before
polarization multiplexing. The output consists of a Pol-mux Nyquist OTDM-WDM (adapted from
[27])
small guard bands are introduced between the WDM channels to avoid inter-
channel-interference. This obtained data stream was successfully used for the
transmission and pass-drop operations of Nyquist OTDM.WDM signal with bau-
drate of 43 GBd to dual polarization 344 GBd over 320 km of super large area
inverse dispersion ber (SLA-IDF) in combination with wavelength selective
switch nodes.
(a)
Optics power [10 dV/div] (b)
Normalized power
86GBd
= 0.4
344GBd
86G = 0.15
Nyquist
channel 172G Nyquist 377G Nyquist 688G Nyquist 172GBd
channels channels channel = 0.25
1552 1554 1556 1558 1560 1562 1564 -15 -10 -5 0 5 10 15
Wavelength [nm] Time [ps]
Fig. 8.28 Nyquist OTDM-WDM demonstration. a Optical spectrum of the dual-polarization

Nyquist OTDM-WDM signal with spectral arrangement of mixed baud rate from 43 to 344 GBd/
pol. b The waveforms captured on a high resolution (1 ps) optical sampling oscilloscope for the
different baud rates. The raised cosine tting reflects the decreasing roll-off factor as the
bandwidth of the pulse increases. Figure adapted from [52]
Fig. 8.29 Optical Nyquist pulse generation through frequency comb synthesis using two cascaded
Mach Zehnder modulators (MZM). An external cavity laser (ECL) generates a narrow linewidth
continuous wave light at 1550 nm. The rst MZM generates spectral lines separated by a
frequency f1 30 GHz. The second MZM re-modulates these seeding components with a
frequency f2 10 GHz. RF power and DC bias of both modulators are adjusted so that a
rectangular frequency comb is obtained. The RF generators are synchronized on a common time
base. The resulting output is amplied and observed on as optical spectrum analyzer (OSA) with a
0.01 nm spectral resolution and on an optical sampling oscilloscope with a 500 GHz bandwidth
The experimental setup for the generation of sinc-shaped Nyquist pulses through
frequency comb synthesis is shown in Fig. 8.29. The proof-of-concept experiment
relied on two cascaded Mach-Zehnder modulators. While the rst modulator, dri-
ven by a radio frequency (RF) f1 is adjusted to generate three seeding spectral
components, the second modulator re-modulates those lines using an RF signal at
f2. For instance to generate a comb with N 9 spectral lines, the condition f1 3f2
or f2 3f1 has to be satised, resulting in a frequency spacing f min (f1, f2). The
number of spectral lines can be easily varied by operating the modulators in carrier-
suppressed mode and/or driving the modulators with multiple RF signals such that
and many experimental variants can be implemented using a similar approach.
A high quality rectangular-shaped frequency comb can be obtained by tuning the

DC bias and the RF voltage amplitude of each modulator while adjusting the phase
delay between the different RF signals. The results for the case f1 30 GHz and
f2 10 GHz are shown in Fig. 8.30. A frequency comb spanning 90 GHz, com-
prising of 9 spectral lines, spaced by 10 GHz is obtained. An out-of-band rejection
of at least 27 dB and a power difference between the spectral lines of less than
0.2 dB was measured. The corresponding time domain trace, observed on an optical
sampling scope with 500 GHz bandwidth, conrms the resulting high quality sinc-
shape Nyquist pulses. The repetition period of 100 GHz and the peak-to-zero
crossing of 11.1 ps are in agreement with the spectral features. The power error with
respect to the theoretical curve was less than 1 %, the roll off of the experimentally
obtained pulse is close to zero. Close to ideal rectangular spectrum, such as seen in
g. (8.30a) can be compared to the ones obtained with the lter based approach and
a no guard-band Nyquist OTDM-WDM can be now envisioned.
(a) (b)
Normalised intensity [a.u.]
10 11.1 ps 100 ps
90 GHz 1
Optical power [dBm]
0 0.8
0.6
-10 27 dB
0.4
-20 0.2
0
-30
1549.4 1549.8 1550.2 1550.6 -150 -100 -50 0 50 100 150
Wavelength [nm] Time [ps]
(c)
40
Optical power [mW]
30
20
10
0 25 50 75 100
Time [ps]
Fig. 8.30 Sinc-shaped Nyquist pulse from frequency comb synthesis. a Spectrum and b respective
time-domain waveform of a comb generated with N 9 spectral components separated by 10 GHz
and expending over a 90 GHz bandwidth. A power difference between spectral components lower
than 0.2 dB and out-of-band rejection of more than 26 dB is obtained. The dashed box represents
the theoretical Nyquist bandwidth of the generated sinc pulses. c Color grade gure of the
experimentally obtained sinc-pulse. Measurements indicate a jitter of 82 fs and a signal-to-noise
greater than 40 dB [49]
(a) (b)
1 1
Intensity [a.u.]
Intensity [a.u.]
Repetition period: 100 ns Repetition period: 10 ns
0.8 Bandwidth: 90 MHz 0.8 Bandwidth: 900 MHz
FWHM: 9.8 ns FWHM: 0.98 ns
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-20 0 20 40 60 80 -2 0 2 4 6 8
Time [ns] Time [ns]
(c) (d)
1 1
Repetition period: 1 ns Repetition period: 100 ps
Intensity [a.u.]
Intensity [a.u.]
0.8 Bandwidth: 9 GHz 0.8 Bandwidth: 90 GHz
FWHM: 98 ps FWHM: 9.8 ps
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-200 0 200 400 600 800 -20 0 20 40 60 80
Time [ps] Time [ps]
Fig. 8.31 Sinc-shaped Nyquist pulses measured using a 500 GHz optical sampling oscilloscope.
The calculated waveforms (dashed lines) are compared with the measured pulses (black straight
lines) for different bandwidth conditions. Nyquist pulses are obtained from the generation of a
rectangular frequency comb with 9 phase-locked components spanning over a spectral width
between 90 MHz and 90 GHz, using modulating frequencies a f1 30 MHz and f2 Df f 10 MHz,
b f1 300 MHz and f2 Df 100 MHz, c f1 3 GHz and f2 Df 1 GHz, and d f1 30 GHz
and f2 Df 10 GHz. The maximum difference between measured pulses and theoretical ones
remained in all cases below 1 % showing the high flexibility of this method [49]
The last major advantage of this method is its flexibility which lies in the fact
that by changing the modulating signal frequencies different frequency combs can
be generated while maintaining the same quality in terms of pulse characteristics.
Figure 8.31 shows measured sinc pulses when the frequency spacing between the
spectral components of the comb was varied over many decades (10 MHz and
10 GHz). The generated pulses coincide very well with the ideal ones over four
frequency decades, showing the root mean square error below 1 % for all cases.
Acknowledgments Juerg Leuthold acknowledges contributions from the EU-Project FOX-C and
support by ETH Zurich, Switzerland. Juerg Leuthold is also very much indebted to Wolfgang
Freude from Karlsruhe Institute of Technology (KIT) with whom he collaborated in the past and
who contributed a lot to the OFDM section in this chapter. Camille Brs acknowledges support by
the EPFL Lausanne, Switzerland.
References
1. W. Shieh, C. Athaudage, Coherent optical orthogonal frequency division multiplexing, in

Electronics Letters (Institution of Engineering and Technology, 2006), pp. 587589
2. R.W. Chang, Synthesis of band-limited orthogonal signals for multichannel data transmission.
Bell Syst. Tech. J. 45, 17751796 (1966)
3. A.J. Lowery, L. Du, J. Armstrong, Orthogonal Frequency Division Multiplexing for Adaptive
Dispersion Compensation in Long Haul WDM Systems. in Optical Fiber Communication
Conference and Exposition and The National Fiber Optic Engineers Conference (Optical
Society of America, Anaheim, California, 2006), p. PDP39
4. I.B. Djordjevic, B. Vasic, Orthogonal frequency division multiplexing for high-speed optical
transmission. Opt. Express 14, 37673775 (2006)
5. N. Cvijetic, L. Xu, T. Wang, Adaptive PMD compensation using OFDM in long-haul 10 Gb/s
DWDM systems, in Optical Fiber Communication Conference and Exposition and The
National Fiber Optic Engineers Conference (Optical Society of America, Anaheim,
California, 2007), p. OTuA5
6. S.L. Jansen, I. Morita, T.C. Schenk, H. Tanaka, Long-haul transmission of16 52.5 Gbits/s
polarization-division- multiplexed OFDM enabled by MIMO processing (Invited). J. Opt.
Netw. 7, 173182 (2008)
7. E. Yamada, A. Sano, H. Masuda, E. Yamazaki, T. Kobayashi, E. Yoshida, K. Yonenaga, Y.
Miyamoto, K. Ishihara, and Y. Takatori, 1 Tb/s (111 Gb/s/ch 10ch) no-guard-interval CO-
OFDM transmission over 2100 km DSF, in Opto-electronics and Communications
Conference, Sydney (2008)
8. S. Chandrasekhar, X. Liu, B. Zhu, D. Peckham, Transmission of a 1.2-Tb/s 24-carrier no-
guard-interval coherent OFDM super channel over 7200-km of ultra-large-area ber, in ECOC
2009 (2009)
Bonk, P. Kleinow, F. Frey, M. Roeger, S. Koenig, A. Ludwig, A. Marculescu, J. Li, M. Hoh,
M. Dreschmann, J. Meyer, S. Ben Ezra, N. Narkiss, B. Nebendahl, F. Parmigiani,
P. Petropoulos, B. Resan, A. Oehler, K. Weingarten, T. Ellermeyer, J. Lutz, M. Moeller,
M. Huebner, J. Becker, C. Koos, W. Freude, J. Leuthold, 26 Tbit s(-1) line-rate super-channel
transmission utilizing all-optical fast Fourier transform processing. Nat. Photonics 5, 364371
(2011)
10. D. Hillerkuss, M. Winter, M. Teschke, A. Marculescu, J. Li, G. Sigurdsson, K. Worms, S. Ben
Ezra, N. Narkiss, W. Freude, J. Leuthold, Simple all-optical FFT scheme enabling Tbit/s real-
time signal processing. Opt. Express 18, 93249340 (2010)
11. H. Nyquist, Certain topics in telegraph transmission theory. Trans. Am. Inst. Electrical Eng.
(A. I. E. E.) 47, 617644 (1928)
12. G. Bosco, V. Curri, A. Carena, P. Poggiolini, F. Forghieri, On the performance of Nyquist-
WDM terabit superchannels based on PM-BPSK, PM-QPSK, PM-8QAM or PM-16QAM
subcarriers. J. Lightwave Technol. 29, 5361 (2011)
13. R. Schmogrow, M. Winter, M. Meyer, D. Hillerkuss, S. Wolf, B. Baeuerle, A. Ludwig, B.
Nebendahl, S. Ben-Ezra, J. Meyer, M. Dreschmann, M. Huebner, J. Becker, C. Koos, W.
Freude, J. Leuthold, Real-time Nyquist pulse generation beyond 100 Gbit/s and its relation to
OFDM. Opt. Express 20, 317337 (2012)
14. R. Schmogrow, D. Hillerkuss, S. Wolf, B. Bauerle, M. Winter, P. Kleinow, B. Nebendahl, T.
Dippon, P.C. Schindler, C. Koos, W. Freude, J. Leuthold, 512QAM Nyquist sinc-pulse
transmission at 54 Gbit/s in an optical bandwidth of 3 GHz. Opt. Express 20, 64396447
(2012)
15. R. Schmogrow, D. Hillerkuss, M. Dreschmann, M. Huebner, M. Winter, J. Meyer, B.
Nebendahl, C. Koos, J. Becker, W. Freude, J. Leuthold, Real-time software-dened multiformat
transmitter generating 64QAM at 28 GBd. IEEE Photonic Tech. L 22, 16011603 (2010)
16. J. Leuthold, W. Freude, Optical OFDM and Nyquist multiplexing, in Optical Fiber
Telecommunications VIB, ed. by I.P. Kaminow, T. Li, A.E. Willner (Academic Press, 2013)
17. R. Schmogrow, M. Winter, D. Hillerkuss, B. Nebendahl, S. Ben-Ezra, J. Meyer, M.
Dreschmann, M. Huebner, J. Becker, C. Koos, W. Freude, J. Leuthold, Real-time OFDM
transmitter beyond 100 Gbit/s. Opt. Express 19, 1274012749 (2011)
18. R. Freund, No, x, M. lle, C. Schmidt-Langhorst, R. Ludwig, C. Schubert, G. Bosco, A.
Carena, P. Poggiolini, Oxenl, x00F, L. We, M. Galili, H.C.H. Mulvad, M. Winter, D.
Hillerkuss, R. Schmogrow, W. Freude, J. Leuthold, A.D. Ellis, F.C.G. Gunning, J. Zhao,
P. Frascella, S.K. Ibrahim, N.M. Suibhne, Single- and multi-carrier techniques to build up Tb/s
per channel transmission systems, in Transparent Optical Networks (ICTON), 2010 12th
International Conference on (2010), pp. 17
19. M.E. Marhic, Discrete Fourier transforms by single-mode star networks. Opt. Lett. 12, 6365
(1987)
20. H. Sanjoh, E. Yamada, Y. Yoshikuni, Optical orthogonal frequency division multiplexing
using frequency/time domain ltering for high spectral efciency up to 1 bit/s/Hz, in Optical
Fiber Communication Conference and Exhibit, 2002. OFC 2002, 401402 (2002)
21. K. Takiguchi, M. Oguma, H. Takahashi, A. Mori, Integrated-optic eight-channel OFDM
demultiplexer and its demonstration with 160 Gbit/s signal reception, in Electronics Letters
(Institution of Engineering and Technology, 2010), pp. 575576
22. G. Cincotti, Fiber wavelet lters [and planar waveguide couplers for full-wavelength
demultiplexers]. IEEE J. Quantum Electron. 38, 14201427 (2002)
23. G. Cincotti, Full optical encoders/decoders for photonic IP routers. J. Lightwave Technol. 22,
337342 (2004)
24. M. Nazarathy, D.M. Marom, W. Shieh, Optical comb and lter bank (De)Mux enabling 1 Tb/s
orthogonal sub-band multiplexed CO-OFDM free of ADC/DAC limits, in Optical
Communication, 2009. ECOC 09. 35th European Conference on (2009), pp. 12
25. M.K. Smit, C. Van Dam, PHASAR-based WDM-devices: principles, design and applications.
IEEE J. Sel. Top. Quantum Electron. 2, 236250 (1996)
26. A.J. Lowery, Design of arrayed-waveguide grating routers for use as optical OFDM
demultiplexers. Opt. Express 18, 1412914143 (2010)
27. K. Lee, C.T.D. Thai, J.-K.K. Rhee, All optical discrete fourier transform processor for 100
Gbps OFDM transmission. Opt. Express 16, 40234028 (2008)
28. A.J. Lowery, L. Du, All-optical OFDM transmitter design using AWGRs and low-bandwidth
modulators. Opt. Express 19, 1569615704 (2011)
29. R.P. Giddings, X.Q. Jin, E. Hugues-Salas, E. Giacoumidis, J.L. Wei, J.M. Tang, Experimental
demonstration of a record high 11.25 Gb/s real-time optical OFDM transceiver supporting
25 km SMF end-to-end transmission in simple IMDD systems. Opt. Express 18, 55415555
(2010)
30. R. Schmogrow, M. Meyer, P.C. Schindler, B. Nebendahl, M. Dreschmann, J. Meyer, A.
Josten, D. Hillerkuss, S. Ben-Ezra, J. Becker, C. Koos, W. Freude, J. Leuthold, Real-time
Nyquist signaling with dynamic precision and flexible non-integer oversampling. Opt. Express
22, 193209 (2014)
31. B. Inan, S. Adhikari, O. Karakaya, P. Kainzmaier, M. Mocker, H. von Kirchbauer, N. Hanik,
S.L. Jansen, Real-time 93.8-Gb/s polarization-multiplexed OFDM transmitter with 1024-point
IFFT. Opt. Express 19, B64B68 (2011)
32. D. Hillerkuss, T. Schellinger, M. Jordan, C. Weimann, F. Parmigiani, B. Resan, K.
Weingarten, S. Ben-Ezra, B. Nebendahl, C. Koos, W. Freude, J. Leuthold, High-quality
optical frequency comb by spectral slicing of spectra broadened by SPM. IEEE Photonics J. 5,
7201011 (7201011 pp) (2013)
33. R. Schmogrow, B. Nebendahl, M. Winter, A. Josten, D. Hillerkuss, S. Koenig, J. Meyer, M.
Dreschmann, M. Huebner, C. Koos, J. Becker, W. Freude, J. Leuthold, Error vector magnitude
as a performance measure for advanced modulation formats. IEEE Photonic Technol. Lett. 24,
6163 (2012)
34. Fujitsu, Soft-Decision FEC Benets for 100G (Fujitsu Network Communications Inc., 2012)
35. Z. Xiang, L.E. Nelson, P. Magill, R. Isaac, Z. Benyuan, D.W. Peckham, P.I. Borel, K. Carlson,
PDM-Nyquist-32QAM for 450-Gb/s Per-Channel WDM Transmission on the 50 GHz ITU-T
Grid. J. Lightwave Technol. 30, 553559 (2012)
36. T. Kobayashi, A. Sano, A. Matsuura, Y. Miyamoto, K. Ishihara, Nonlinear tolerant spectrally-
efcient transmission using PDM 64-QAM single carrier FDM with digital pilot-tone.
37. C. Jin-Xing, C.R. Davidson, A. Lucero, Z. Hongbin, D.G. Foursa, O.V. Sinkin, W.W.
Patterson, A.N. Pilipetskii, G. Mohs, N.S. Bergano, 20 Tbit/s transmission over 6860 km with
sub-nyquist channel spacing. J. Lightwave Technol. 30, 651657 (2012)
38. A.H. Gnauck, P.J. Winzer, S. Chandrasekhar, X. Liu, B. Zhu, D.W. Peckham, Spectrally
efcient long-haul WDM transmission using 224-Gb/s polarization-multiplexed 16-QAM.
39. R. Schmogrow, R. Bouziane, M. Meyer, P.A. Milder, P.C. Schindler, R.I. Killey, P. Bayvel,
C. Koos, W. Freude, J. Leuthold, Real-time OFDM or Nyquist pulse generationwhich
performs better with limited resources? Opt. Express 20, B543B551 (2012)
40. T. Hirooka, P. Ruan, P. Guan, M. Nakazawa, Highly dispersion-tolerant 160 Gbaud optical
Nyquist pulse TDM transmission over 525 km. Opt. Express 20, 1500115007 (2012)
41. T. Hirooka, M. Nakazawa, Linear and nonlinear propagation of optical Nyquist pulses in
bers. Opt. Express 20, 1983619849 (2012)
42. G. Bosco, A. Carena, V. Curri, P. Poggiolini, F. Forghieri, Performance limits of Nyquist-
WDM and CO-OFDM in high-speed PM-QPSK systems. IEEE Photonics Technol. Lett. 22,
11291131 (2010)
43. J.G. Proakis, M. Salehi, Digital Communications (McGraw Hill, 2008)
44. M. Nakazawa, T. Hirooka, P. Ruan, P. Guan, Ultrahigh-speed orthogonal TDM
transmission with an optical Nyquist pulse train. Opt. Express 20, 11291140 (2012)
45. D. Hillerkuss, R. Schmogrow, M. Meyer, S. Wolf, M. Jordan, P. Kleinow, N. Lindenmann, P.
C. Schindler, A. Melikyan, X. Yang, S. Ben-Ezra, B. Nebendahl, M. Dreschmann, J. Meyer,
F. Parmigiani, P. Petropoulos, B. Resan, A. Oehler, K. Weingarten, L. Altenhain, T.
Ellermeyer, M. Moeller, M. Huebner, J. Becker, C. Koos, W. Freude, J. Leuthold, Single-
Laser 32.5 Tbit/s Nyquist WDM transmission. J. Opt. Commun. Netw. 4, 715723 (2012)
46. A. Vedadi, M.A. Shoaie, C.-S. Brs, Near-Nyquist optical pulse generation with ber optical
parametric amplication. Opt. Express 20, B558B565 (2012)
47. R. Schmogrow, S. Ben-Ezra, P.C. Schindler, B. Nebendahl, C. Koos, W. Freude, J. Leuthold,
Pulse-shaping with digital, electrical, and optical lters-A comparison. J. Lightwave Technol.
31, 25702577 (2013)
48. L. Xiang, S. Chandrasekhar, Z. Benyuan, P.J. Winzer, A.H. Gnauck, D.W. Peckham, 448-Gb/
s reduced-guard-interval CO-OFDM transmission over 2000 km of ultra-large-area ber and
ve 80-GHz-Grid ROADMs. J. Lightwave Technol. 29, 483490 (2011)
49. M.A. Soto, M. Alem, M. Amin Shoaie, A. Vedadi, C.-S. Brs, L. Thvenaz, T. Schneider,
Optical sinc-shaped Nyquist pulses of exceptional quality. Nat. Commun. 4 Article No. 2898.
(2013)
50. G. Baxter, S. Frisken, D. Abakoumov, Z. Hao, I. Clarke, A. Bartos, S. Poole, Highly
programmable wavelength selective switch based on liquid crystal on silicon switching
elements, in Optical Fiber Communication Conference, 2006 and the 2006 National Fiber
Optic Engineers Conference. OFC 2006 (2006), p. 3
51. H.N. Tan, K. Tanizawa, T. Inoue, T. Kurosu, S. Namiki, No guard-band wavelength
translation of Nyquist OTDM-WDM signal for spectral defragmentation in an elastic add
drop node. Opt. Lett. 38, 32873290 (2013)
52. T. Hung Nguyen, T. Inoue, T. Kurosu, S. Namiki, Transmission and pass-drop operations in
all-optical elastic network using Nyquist OTDM-WDM up to 2344 Gbaud/channel, in
OptoElectronics and Communications Conference held jointly with 2013 International
Conference on Photonics in Switching (OECC/PS), 2013 18th (2013), pp. 12
53. B. Washburn, R. Fox, N. Newbury, J. Nicholson, K. Feder, P. Westbrook, C. Jrgensen,

Fiber-laser-based frequency comb with a tunable repetition rate. Opt. Express 12, 49995004
(2004)
54. A. Ruehl, A. Marcinkevicius, M.E. Fermann, I. Hartl, 80 W, 120 fs Yb-ber frequency comb.
Opt. Lett. 35, 30153017 (2010)
55. A. Bartels, R. Gebs, M.S. Kirchner, S.A. Diddams, Spectrally resolved optical frequency
comb from a self-referenced 5 GHz femtosecond laser. Opt. Lett. 32, 25532555 (2007)
56. T.J. Kippenberg, R. Holzwarth, S.A. Diddams, Microresonator-based optical frequency
combs. Science 332, 555559 (2011)
57. P. DelHaye, A. Schliesser, O. Arcizet, T. Wilken, R. Holzwarth, T.J. Kippenberg, Optical
frequency comb generation from a monolithic microresonator. Nature 450, 12141217 (2007)
58. A.K. Mishra, R. Schmogrow, I. Tomkos, D. Hillerkuss, C. Koos, W. Freude, J. Leuthold,
Flexible RF-Based Comb Generator. IEEE Photonic Technol. Lett. 25, 701704 (2013)
59. Y. Yamamoto, T. Komukai, K. Suzuki, A. Takada, Multicarrier light source with flattened
spectrum using phase modulators and dispersion medium. J. Lightwave Technol. 27, 4297
4305 (2009)
60. T. Yamamoto, T. Komukai, K. Suzuki, A. Takada, Spectrally flattened phase-locked multi-
carrier light generator with phase modulators and chirped bre Bragg grating, in Electronics
Letters (Institution of Engineering and Technology, 2007), pp. 10401042
61. T. Yang, J. Dong, S. Liao, D. Huang, X. Zhang, Comparison analysis of optical frequency
comb generation with nonlinear effects in highly nonlinear bers. Opt. Express 21, 85088520
(2013)
62. V. Ataie, B.P.P. Kuo, E. Myslivets, S. Radic, Generation of 1500-tone, 120 nm-wide ultraflat
frequency comb by single CW source, in Optical Fiber Communication Conference/National
Fiber Optic Engineers Conference 2013 (Optical Society of America, Anaheim, California,
2013), p. PDP5C.1
63. Z. Tong, A.O.J. Wiberg, E. Myslivets, B.P.P. Kuo, N. Alic, S. Radic, Spectral linewidth
preservation in parametric frequency combs seeded by dual pumps. Opt. Express 20, 17610
17619 (2012)
64. K. Harako, D. Seya, T. Hirooka, M. Nakazawa, 640 Gbaud (1.28 Tbit/s/ch) optical Nyquist
pulse transmission over 525 km with substantial PMD tolerance. Opt. Express 21, 21062
21075 (2013)
65. H. Hu, D. Kong, E. Palushani, J.D. Andersen, A. Rasmussen, B.M. Srensen, M. Galili, H.C.
H. Mulvad, K.J. Larsen, S. Forchhammer, P. Jeppesen, L.K. Oxenlwe, 1.28 Tbaud Nyquist
signal transmission using time-domain optical fourier transformation based receiver, in CLEO:
2013 Postdeadline (Optical Society of America, San Jose, California, 2013), p. CTh5D.5
66. M.A. Soto, M. Alem, M.A. Shoaie, A. Vedadi, C.S. Brs, L. Thvenaz, T. Schneider,
Generation of Nyquist Sinc pulses using intensity modulators, in CLEO: 2013 (Optical Society
of America, San Jose, California, 2013), p. CM4G.3
Chapter 9
Energy-Efcient Optical Signal Processing
Using Optical Time Lenses
Leif Katsuo Oxenlwe, Michael Galili,

Hans Christian Hansen Mulvad, Hao Hu, Pengyu Guan,
Evarist Palushani, Mads Lillieholm and Anders Clausen
Abstract This chapter describes advanced functionalities for optical signal

processing using optical time lenses. A special focus is devoted to functionalities
that allow for energy-savings. In particular, we nd that optical signal processing,
where the processing is broadband and capable of handling many bits in a single
operation allows for sharing the processing energy by the many bits, and hence the
energy per bit is reduced. Such functionalities include serial-to-parallel conversion
in a single time lens, where a large number of parallel demultiplexers may be
substituted by a single time lens. Combining time lenses into telescopic arrange-
ments allows for more advanced signal processing, such as temporal or spectral
compression or magnication. A spectral telescope may for instance allow for
conversion of OFDM signals to DWDM-like signals, which can be separated
passively, i.e. without additional energy. This is opposed to the DFT OFDM
receivers otherwise suggested, where a temporal active gate is required for each
tributary. With the spectral telescope, only two active time lenses are required,
irrespective of how many tributaries are used. This chapter describes how optical
time lenses function and by showing examples of some advanced functionalities
points to future scenarios where energy consumption may be considerably reduced.
9.1 Introduction: Energy-Efcient Solutions Using

Nonlinear Optical Signal Processing
Some of the biggest challenges in optical communications today are the relentless
demands for more capacity and the accompanying demands for more energy-efcient
solutions. Current technologies cannot keep up with the bandwidth demands without
L.K. Oxenlwe (&) M. Galili H.C.H. Mulvad H. Hu P. Guan E. Palushani

M. Lillieholm A. Clausen
DTU Fotonik, Department of Photonics Engineering, Technical University of Denmark,
rsteds Plads 343, 2800 Kongens Lyngby, Denmark
e-mail: lkox@fotonik.dtu.dk

262 L.K. Oxenlwe et al.
resorting to extensive parallelism, where the number of components and their com-
bined power consumption, simply scale with the bandwidth. The actual trafc on the
internet grows approximately 50 % per year, pointing to the looming imminent
capacity crunch, where the trafc outgrows the available capacity [1, 2], and is
responsible for about 2 % of the global CO2 emissions [3] today. These are CO2
emissions comparable to that of the aviation industry, and it thus remains one of the
biggest technological challenges to nd sustainable solutions for the future internet.
There is therefore a strong need for groundbreaking developments in optical science
enabling novel breakthrough technologies that will allow for increased capacity with
a reduced energy consumption. There are indications that nonlinear optical signal
processing may in some cases offer the higher bandwidth at lower energy. However
one needs to be very careful with choosing the right functionalities to harvest the
benets from both electronics and optics to globally benet from what is possible
with electronics and what is possible with optics. For instance, in [4, 5], it is con-
cluded that optical signal processing may be appropriate for certain functionalities
where the number of signal processing elements is small, and where the signal
processing itself is simple. This could be the case for the ultrafast nonlinear optical
signal processing of ultrahigh bit rates in few but ultrafast devices, as will become
apparent in the following sections. Furthermore, this chapter will describe a large
number of advanced ultrafast functionalities that are achievable today using mostly
commercially available components.
9.1.1 Energy-Efcient Optical Signal Processing: Many Bits

per Operation
Certain functionalities are promising for low-energy signal processing, for instance
functionalities where many bits are processed per operation, so that many bits share
the used pump energy. But one has to carefully choose the right type of func-
tionalities and nonlinear effects. An example demonstrating reduced energy con-
sumption per bit is cross-phase modulation (XPM) based nonlinear signal
processing relying on the Kerr effect in e.g. highly nonlinear bre (HNLF). Here the
amount of phase modulation depends on the peak power of the pump signal as well
as the nonlinear coefcient and the length of the bre: Du / cLPpeak . This fact is
useful for designing energy-efcient switching schemes. Consider a train of pulses
with a certain width and repetition rate at a given average power resulting in a
certain peak power. If the repetition rate is now doubled, and the pulse widths are
correspondingly twice as narrow, then the peak power remains the same for the
same average power. If this pulse train is used as the pump pulses, then it is clear
E
that the switching energy per bit is inversely proportional to the bit rate: bit / 1=N,
where N is the multiplication factor on the pulse rate. So basically, a 20 Gbit/s data
signal could be switched with half the energy/bit compared to a 10 Gbit/s signal,
and when going to very high bit rates like 640 Gbit/s or 1.28 Tbit/s, this advantage
9 Energy-Efcient Optical Signal Processing 263
is multiplied manifold. One concrete example of this idea is the use of XPM-based
sideband generation on a CW wave [6]. A data pulse stream exerts XPM on the
CW, and every time a 1-bit, i.e. a pulse, arrives, a sideband occurs as opposed to
when a 0-bit arrives, where nothing happens. Hence by ltering out the sidebands,
one gets a wavelength converted, or in some cases an additionally regenerated, copy
of the original data stream. To create a certain sideband requires a certain peak
power, and as outlined above, the higher the bit rate, the less energy/bit is needed
for this wavelength conversion scheme.
In the following, we will address another scheme, where we will show in detail
that the switching energy per bit decreases with increased bit rates, ultimately
resulting in total system power reduction.
Figure 9.1 shows a scheme to perform serial-to-parallel conversion based on
optical time lenses. The technical principles of operation will be described in later
sections, and here we will simply discuss the implications on power consumption,
and for now just assume that it works. The nonlinear effect used for this scheme is
four-wave mixing (FWM). In FWM, two pump photons are used to create an idler
photon together with the signal photon. As long as the pump is not depleted, which
is most cases of any practical use, there are enough pump photons to create the
idlers, and hence for a given (undepleted) pump power, one may simply increase
the bit rate and thus use more of the pump and thereby reduce the pump energy/bit.
This is conrmed in [7] where detailed simulations of the serial-to-parallel
conversion scheme demonstrated that increasing the data bit rate from 160, over
320640 Gbit/s resulted in the same bit error rate for the same pump power, see
Fig. 9.1c.
In [9], power consumption for various data capacities is calculated for OTDM
and WDM point-to-point systems. No in-line functionalities other than amplica-
tion and passive dispersion compensation are considered. The total system power
(a) (b) (c)

1
Power Spectrum [dBm]

-log10 (Bit Error Rate)
3 2
Pump D
spectral
Si
Data D/2
OFT
64
time
Wavelength after FWM [nm] Average Pump Power [dBm]
Fig. 9.1 Serial-to-parallel conversion using time lens based time-domain optical Fourier
transformation (OFT). a Principle of serial-to-parallel conversion, in this sketch using a silicon
nanowire as nonlinear medium. b, c Simulation results of converting a 640 Gbit/s serial optical
data signal into 64 parallel DWDM data channels each at 10 Gbit/s, from [7]. b Output spectra of
the OFT showing the input broad-spectrum OTDM data signal, the flat-top-spectrum pump signal
and the resulting DWDM signal, in this case 50 GHz spaced. c Resulting BER performance,
revealing that for the same average pump power as used in the experimental demonstration in [8]
the bit rate is not important, i.e. for bit rates of 160, 320 and 640 Gbit/s, the same BER is obtained
for the same pump power, revealing that the pump energy per bit is inversely proportional to the
bit rate. The functionality may be carried out using a highly nonlinear bre or a silicon nanowire as
the nonlinear medium creating the optical phase modulation required in the time lens
System Power Consumption

Total System Power Consumption (without common equipment)
2510 10000
consumption [W]
consumption [W]
OFT Experiment OFT Experiment
System power
System power
2010 OTDM1 OTDM1

1000 WDM
1510 WDM
OFT OFT
1010
100
510
10 10
10 100 1000 10 100 1000
Capacity [Gbit/s] Capacity [Gbit/s]
Fig. 9.2 Total system power consumption comparison between WDM and OTDM scenarios,
including the case of OTDM using time lens based time-domain optical Fourier transformation
(OFT) for demultiplexing all OTDM channels simultaneously. Left Total system power
consumption including all components. Right Disregarding common equipment (data modulators,
in-line EDFA, and photodiodes) to more easily spot the differences between the scenarios
Table 9.1 Assumed power values for power consumption calculation

P (W) Cool (W)
CW 1 3 N WDM
Data mod 2.2 N WDM/OTDM
RZ mod 2.2 1 WDM
RZ laser 3.2 3 1 OTDM
Demux-laser 3.2 3 N OTDM
Line-EDFA 2.5 6 2 pump diodes 1
MUX-EDFA 2.5 6 2 pump diodes 1
Demux-EDFA 5 12 4 pump diodes N OTDM
PD 0.04 N WDM/OTDM
Values correspond to settings required for BER 1E-9, and are either taken from actual laboratory
settings or from data specications from manufacturers, cf. [9]
consumption is shown in Fig. 9.2, based on numbers derived for the results pre-
sented in [9], with the addition of a data point corresponding to an experimental
demonstration of OTDM-to-WDM conversion at 640 Gbit/s. The total system
power is calculated for a system performance yielding BER = 1E-9 for each
channel, including cooling of all laser devices, and electrical power to all ampliers,
lasers etc. [9]. The assumed values for power both for driving lasers a.o. and for
cooling lasers are realistic in the sense that they are taken from tabletop commercial
products actually used in the experimental demonstrations, see Table 9.1.
Looking at Table 9.1, one may easily derive how many of a certain component is
used in the classical WDM and OTDM setups. In this simple calculation, only a single
in-line EDFA is used, corresponding to only transmitting over a single span of bre.
Both the WDM and the OTDM setups require the same number of data modulators
(data mod in Table 9.1) and photodiodes (PD), which do not need to be
temperature controlled. In the WDM case, N parallel CW lasers are used, and a single
RZ modulator is assumed (RZ mod), which does not change much. In the OTDM
transmitter a single pulse source (RZ laser) is used, corresponding in power to a
CW laser and an RZ modulator. The OTDM setup also includes a booster amplier to
compensate for the loss in the OTDM split-and-delay-and-combine multiplexer
(MUX EDFA), N parallel demultiplexers based on N HNLFs and N control
pulse sources (demux laser) and N EDFAs to boost these control pulse sources
(demux-EDFA), corresponding to a worst case scenario. The demux-EDFA is
assumed to be a high-power EDFA requiring 4 pump laser diodes, where the other
EDFAs have 2 pump diodes. For cooling, it is assumed that only the lasers require
cooling and that each laser will require 3 W for coolingnumbers taken from
standard laboratory equipment. In addition to this, the serial-to-parallel scheme setup
is added, assuming two HNLF-based OFTs, sharing a single pump source and using
the pump power used in the actual experimental demonstration described in [8]. The
experimental demonstration at 640 Gbit/s is also added.
Figure 9.2 shows the outcome of these calculations. Generally WDM is less
power consuming due to the passive demultiplexing. The worst case is by far the N
parallel HNLFs and pump sources, being an order of magnitude worse than the
WDM case. However for serial-to-parallel conversion at high bit rates (above
640 Gbit/s) the OFT-OTDM system is almost an order of magnitude less power
consuming compared to the WDM system, and almost two orders of magnitude
better than the worst-case parallel HNLF receiver OTDM setup. Furthermore, this
benet increases for higher bit rates as can be seen in Fig. 9.2, right, which is
because the OFT-receiver scheme uses constant power irrespective of the bit rate.
This nding agrees well with the above postulation that processing more bits in
fewer components is where optical signal processing has its strongest advantage [4,
5]. For serial-to-parallel conversion, the energy per bit scales inversely with the bit
rate, as most power is used for the control signal. And irrespective of the total
OTDM bit rate, only two OFTs will be required, and to a large extent, the same
pump power will be needed, which ultimately leads to a better energy-efciency for
higher bit rates. It even outperforms the simple point-to-point WDM systems.
These calculations are realistic in the sense that they are based on extracted
parameters from the experimental demonstration. One should of course be cautious
with putting too much trust into calculations like thesethey are based on
assumptions of certain power values, and these may vary a lot, and the actual curves
representative of the different system scenarios will change correspondingly. This
sort of investigation is not hard science and the results should be seen as specu-
lative, and we mostly focus on the observed trends. In that respect, we can observe
that the use of time lenses for serial-to-parallel conversion has a tendency to
increase in system power more slowly than the WDM and classical OTDM sys-
tems. Where exactly the curves will cross each other is obviously strongly
dependent on the parameter values used.
In the following section, a detailed description of the time lens based OFT
scheme is given.
9.2 Time-Domain Optical Fourier Transformation/Time

Lens Principle
In [10] Kolner presents the space-time duality principle, which accounts for the
equivalence between diffractive imaging and dispersive propagation. Based on this
space-time duality formalism, temporal imaging becomes a simple and versatile
tool with many applications, such as scaling temporal waveforms using simple
systems, or indeed scaling of spectral shapes, as well as converting between tem-
poral and spectral domains.
9.2.1 Time Lens Principle and Time-Domain OFT
In the space-time duality description, it is shown how diffraction and dispersion

operate similarly on spatial objects and temporal waveforms, respectively. The
diffractive far eld image (Fraunhofer diffraction) is equivalent to a dispersed
waveform, in that the far eld image becomes the Fourier transform of the spatial
object, and similarly the dispersed waveform becomes the Fourier transform of the
Fig. 9.3 Equivalence of diffractive imaging and dispersive propagation. Top A plane wave
propagating through a slit in a wall, corresponding to a rectangular object is diffracted to
generate the Fourier transform of the object as the image in the far eld (propagation distance
z a, object size), known as Fraunhofer diffraction. A spherical lens will add a quadratic phase
modulation on the diffracted wavefront and may be used to manipulate the size and focal plane of
the image, e.g. by focusing the image down. Bottom Dispersive propagation of an optical pulse is
equivalent to diffractive imaging in the time domain. For large dispersion (D = 2L |t0|2), i.e.
much greater than the original pulse width (object size) squared, the transmitted pulse shape
(image) becomes the Fourier transform of the input pulse shape (object). A parabolic phase
modulation, corresponding to a linear chirp acts as a lens in time (time lens), and may be used to
focus the image to smaller extent, or more advanced temporal imaging [10]
input waveform. Figure 9.3 shows this basic equivalence. So diffraction and dis-
persion are equivalent. One may also include lenses. In spatial optics, a spherical
lens will result in a quadratic phase modulation of the diffracted wavefront, which
can be used to focus or enlarge an image. In the time domain, a temporal parabolic
phase modulation on the waveform will impose a linear chirp, which, when
balanced with proper amount of dispersion, can be used to focus or enlarge the
temporal waveform, and such a system is therefore called a time lens.
In the following, a derivation of the simple time lens based OFT system shown
in Fig. 9.3 (bottom) and in Fig. 9.4 is given. This derivation shows that a parabolic
phase modulation followed by appropriate amount of dispersion will result in an
output waveform, which is exactly equal to the Fourier transformation of the input
waveform, which again corresponds to the input spectrum [11]. For the reverse
system, i.e. dispersion followed by phase modulation, the solution may readily be
derived the same way, and result in an output where the spectral output is equal to
the input waveform, i.e. a temporal-to-spectral conversion.
In Fig. 9.4, the input and output electric eld amplitudes in the spectral and
temporal domains are shown. They relate to the optical power as

Ii t jAi tj2 ; Ii x A~ i x2
Figure 9.4 shows how the electric elds evolve through the different parts of the
set-up. In the phase modulator, a linear chirp, C, is added to the electric eld
yielding an electric eld waveform of
i 2
Ac 0; t A0 0; t: e2Ct
Note that in order to obtain a linear chirp, it is necessary to add a parabolic phase
modulation, as the chirp is the time derivative of the phase. The chirped spectrum is
the Fourier transform of the chirped waveform, i.e.
~ c 0; x F fAc 0; tg
A
The chirped waveform is now propagated through a length of dispersive med-

ium, such as a bre with accumulated dispersion D b2 L, where 2 is the GVD
parameter of the bre and L is the total length of the bre. The linear transfer
function for such a bre is H x e2b2 x L , see e.g. [12]. Thus, the spectral output
i 2
of the dispersive element becomes
OFT for frequency-to-time mapping

A0 (0,t) C Ac(0,t)=A 0 exp(iCt 2/2) A1 (L,t)
- mod
D
0 (0, ) c(0, ) = (A c) 1 (L, )= c(0, )H ()
Fig. 9.4 Principle sketch for deriving the time-domain optical Fourier transformation (OFT)
relations for the case of frequency-to-time conversion
~ 1 L; x A
A ~ c 0; x H x A
~ c 0; x e2i b2 x2 L
and the output temporal shape is just the inverse Fourier transform of this, i.e.
n o
A1 L; t F 1 A~ 1 L; x F 1 A ~ c 0; x e2i b2 x2 L
Z1
Ac 0; t0 ht t0 dt
1
where
Z1 r
1 1 i i t2
2b 2 x L ixt
i 2
h t F fH xg e e dx e2D
2p 2pD
1
i.e. the output temporal shape is the convolution of the chirped input pulse to the
dispersive element with the bre response function h(t). Now, assuming that the
accumulated dispersion is matched to the chirp added to the pulse (D = 1/C), and
using the transformation t/D = , the output waveform becomes
Z1
A1 L; t Ac 0; t0 ht t0 dt
1
r Z1
i i 02 i 0 2
A0 0; t0 e2Ct e2Dtt dt0
2pD
1
r Z1 r Z1
i i 2
0 t 0
0 i i t2 0
e2D t
A0 0; t e D dt
i t
e2D A0 0; t0 eixt dt0
2pD 2pD
1 1
r
i i 2
~ 0 0; x
e2Dt A
2pD
Finally, the intensity at the output of the OFT becomes
1 ~ 2 1
I1 t j A 1 t j 2 A0 x I0 x
2pD 2pD
i.e. the output temporal waveform is exactly given by the input spectral shape [13]
scaled with a constant depending on dispersion D.
Example: A supergaussian spectrum: It is often easier to create special shapes in
the spectral domain than in the time domain, and thus this OFT scheme can be seen as
a tool to transfer e.g. grating-based spectral shapes into temporal waveform shapes. If
2m
for instance a supergaussian spectrum is generated, I0 x K ex=Dx with
2m
amplitude K, then the temporal output pulse shape becomes I1 t 2pD 1
et=DDx ,
for K = 1, when using the aforementioned transformation t/D = . That is, exactly the
same shape as the spectral shape and with a tuneable 1/e-width given by
Dt1=e DDx. The width is tuneable, as D = 1/C is tuneable, or in other words, by
tuning the chirp rate (i.e. the quadratic phase modulation) of the time lens, the size of
the image can be controlled.
From the above equations, it is seen that it is also possible to compensate for the
temporal impairments induced by dispersion in transmission bres by ltering and
applying OFT, thus transforming the Gaussian unaffected spectrum to a clean
temporal waveform, as investigated extensively by Nakazawa et al. [11]. In [11],
m = 1, and the pulse is thus Gaussian. As a benchmark, the dispersion is set to be
equal to the original pulse width squared, i.e. jDj Dt02 , which leads to an output
pulse width after OFT of Dt1=e Dt02 Dx, which is equal to Dt0 for a Gaussian, and
hence the original pulse is obtained.
The above derivation is for frequency-to-time mapping, and a similar approach
will result in the formalism for time-to-frequency mapping [14], where the order of
dispersion and phase modulation is simply reversed, yielding

Ic x A~ c x2 2p jA1 0; tj2 2p I0 t
jCj jCj
Here again, t = D = /C, and C is seen to determine the scaling factor between
time and frequency. Thus a temporal spacing t will be mapped to a spectral
spacing of = C.
The phase modulation may be obtained in several ways, e.g. by electro-optic
phase modulation as sketched above, or by four-wave mixing (FWM) to obtain very
high chirp rates. In [15], a good comparison between different types of phase
modulation is given.
In the case of FWM, the dispersed waveform to be transformed in a time-to-
frequency mapping acts as signal Es(t) and the phase modulation is applied using
linearly chirped pump Ep(t) pulses, generated by propagation of transform limited
pulses in a dispersive medium. In the FWM process, the signal is converted to an
idler Ei(t), which combines the phases of both pump and signal Ei / Ep2 Es . In our
case, the signal is A0(L, t), and the idler is Ac(t). As a result of the FWM process,
the time information contained in the signal is mapped onto the power spectrum of
the generated idler, see more detailed example below for serial-to-parallel
conversion.
9.2.2 Important OFT/Time Lens References
Several groups worldwide have contributed enormously to the development of time

lenses and their use for different applications. It is beyond the scope of this chapter
to give a full account of all the work that has been done, but some of the most
important references that will be useful to learn more about time lenses are men-
tioned here. In [10], Kolner rst described the equivalence between spatial and
temporal optics as mentioned above. In [16], the time-to-frequency case is
described in detail, and in [17], the use of the space-time duality is described by
Azaa and Muriel in the context of the application spectral analysis. The ability of
a time lens to focus a waveform to a certain focal point in time is suggested by
Mollenauer and Xu to be used for timing jitter compensation [18]. Jalali and co-
workers describe the time lens usage for time-stretching in [19]. This scheme may
be used to investigate spectral proles/ngerprints in the time domain, by simple
Fourier transformation and photonic stretching. In [11], Nakazawa et al. details the
time-domain optical Fourier transformation (OFT) principle used for distortion-free
transmission, where a frequency-to-time conversion is utilized. Weiners group has
used the time lens concept for arbitrary waveform generation, e.g. for ultra-wide-
band communication [20]. In [21], a description of various ultra-fast applications of
time lenses are given, and in [22], the rst demonstration of time lenses based on
a silicon chip is presented. Finally, in [15], Salem, Foster and Gaeta give a
good overview of recent ultrahigh-speed optical signal processing applications
using time lenses.
9.3 Serial-to-Parallel Conversion
Serial-to-parallel conversion is a case of temporal to spectral conversion, i.e. where

dispersion is preceding phase modulation. One implementation of this is to trans-
form a serial OTDM data signal into separate parallel WDM channels as already
introduced in Fig. 9.1
9.3.1 640 Gbit/s OOK OTDM-to-DWDM Conversion
For an efcient solution, the electro-optic phase modulation is exchanged with an

optical phase modulation, which may be provided in various ways. Using four-
wave mixing (FWM) allows for very strong phase modulation [15]. Figure 9.5
shows a schematic of OTDM-to-WDM conversion using FWM.
The data and pump are rst dispersed in standard single mode bre (SMF) to
create a linear chirp on the pulses. The rectangular pump spectrum is sent through a
bre of dispersion D, starting at its Fourier transform position. After D, it will have
acquired a flat top waveform. To obtain OTDM-to-WDM conversion with spectral
compression of the individual WDM channels, allowing for DWDM creation, the
data pulses must traverse a bre with dispersion exactly equal to half of the pumps
dispersion, i.e. D/2. Now, these two signals are injected into a nonlinear medium,
e.g. a HNLF, where they interact via FWM. The pump and data signal create an
Data t
spectrum
Data linear
Pump
chirp (1/Ds)
spectrum
Pump linear
Idler chirp (1/Dp)
spectrum
s
-
p
=2
Chirp free
idler
i

Fig. 9.5 OTDM-to-WDM conversion using FWM. Spectrographic representation of the basic
principle. Dispersed waveforms interact in a four-wave mixing process yielding non-chirped idlers
at separate wavelengths
idler at the optical frequency i = 2p s = 2c/i. Because the pump is chirped

and covers all the OTDM channels, different channels will overlap in time with
different parts of the pump spectrum and thus give rise to idlers at different
wavelengths. And because the data is chirped with D/2 compared to the pump being
chirped by D, each OTDM pulse will be converted to an idler with a compressed
spectrum. This makes it possible to place the idlers very densely, i.e. to create a
DWDM signal. Each DWDM channel now originates from a specic OTDM
channel, and they may simply be ltered out by passive lters. This way, all OTDM
channels are demultiplexed simultaneously.
Figure 9.6 shows the basic schematic setup for OFT-based OTDM-to-WDM
conversion. In principle all OTDM channels can be demultiplexed in one device
[23], but in practice it will be very difcult to create the required sharp edged
rectangular pump pulses, and thus some data pulses may overlap with two neigh-
bouring pump pulses giving some impairment. It is therefore more practical to
assume that two of these OFT-based demultiplexers or OTDM-DWDM converters
in parallel are needed to demultiplex all channels. Still, this is potentially a sub-
stantial reduction in active components in an OTDM receiver. As described in the
introduction, this gives rise to an enormous power consumption reduction,
1

3 2
Pump Chirp (dispersion): D
spectral
HNLF
Data Chirp (dispersion): D/2
OFT
64
time
Fig. 9.6 Schematic setup for OTDM-to-DWDM conversion using the optical Fourier transfor-
mation technique (OFT) by four-wave mixing in a HNLF. In practice, two OFT units may be
preferred. From [8]
(a) -20 640 Gbit/s 10 GHz pump (b) -60

OTDM (OOK)
Power
[dBm]
-30 -65
Power [dBm]
-70
-40
-75
-50
Power penalty
idler: 15
Sensitivity
25 GHz DWDM -24

-60 12
[dBm]
[dB]
-28 9
-70 -32 6
32 ch's
-36 within 3.5 dB 3
-80
0
1540 1550 1560 1570 1580 1590 1572 1574 1576 1578 1580 1582 1584 1586
Fig. 9.7 Experimental demonstration of 640 Gbit/s OTDM-DWDM conversion by OFT using
FWM in HNLF. a Output spectrum from the HNLF showing the OTDM data, the flat-top pulse
and the DWDM idler channels. b Zoom in on the DWDM idler spectrum (top) and the receiver
sensitivity and power penalty at BER 109 for the demultiplexed channels (bottom)
rendering an OTDM point-to-point system less power consuming than a WDM

point-to-point system for high bit rates.
Figure 9.7 shows results from an experimental demonstration where a 640 Gbit/s
OTDM signal is converted into a DWDM signal with only 25 GHz spacing between
the DWDM channels [8, 24]. 40 of the 64 tributaries are simultaneously demulti-
plexed with a BER less than 1E-9. The channels to the sides of the spectrum,
correspond to the OTDM channels overlapping with the edges of the pump pulse,
and thus have lower conversion efciency, and ultimately lower OSNR, and
therefore higher penalty. The most extreme channels cannot get a BER less than
109. The receiver sensitivity at BER 109 is measured for the 40 centre-most
channels and 32 of these have a sensitivity spread within 3.5 dB. This clearly
demonstrates the effectiveness of this scheme. With only two of such OFT-units in
parallel using temporally detuned pumps, all OTDM channels should be attainable,
irrespective of the OTDM bit rate. In the example shown in Fig. 9.7, the non-linear
element was a HNLF but in [24], it was also shown that this scheme could be
carried out using a silicon nanowire as the nonlinear medium.
9.3.2 Serial-to-Parallel Conversion of Data with Advanced

Modulation Formats
Using FWM for phase modulation, one may benet from the phase-preserving
nature of FWM, i.e. that the idler retains the phase information of the signal, albeit
in conjugate form. This means that the above described serial-to-parallel conversion
scheme may also be used on phase-modulated data signals without losing any
information. Figure 9.8 shows experimental results from [25], where a 160 or 320
Gbaud OTDM signal carrying QPSK or 16QAM modulation is converted to
DWDM 10 Gbaud channels. In this experiment, it was conrmed that the 16 QAM
data could be successfully received to below the forward error correction (FEC)
limit of 3.8E-3, corroborating that this scheme is independent of the modulation
format.
9.3.3 Wavelength-Preserving Serial-to-Parallel Conversion
In the two cases mentioned above, degenerate FWM is used, where the two pump
photons are at the same wavelength, originating from the same source. In that case
the idler appears at the other side of the pump from the signal. However, nonde-
generate FWM may equally well be used. This adds more design freedom to where
the desired idler wavelength may be positioned. In [26], the pump photons are in
orthogonal polarisation and at different wavelengths.
This scheme allows one to choose the idler wavelength to be for instance exactly
where the input signal was. In [26], the orthogonally polarized pumps are sym-
metrically positioned around the input signal wavelength, and one pump is aligned
in polarisation with the input data signal. Owing to energy and momentum con-
servation, the idler will now be generated at the same wavelength as the data signal,
but in orthogonal polarisation to the data signal. One may thus lter the idler away
(a) (b) 1
B2B NRZ-16QAM, 1533.7 nm
-30 10 GHz B2B NRZ-16QAM, 1558 nm
Pump B2B RZ-16QAM, 1558 nm
160 GBd -> 10 GBd, 1530.4 nm
Power [dBm]
-40
-log(BER)
320 GBd -> 10 GBd, 1530.4 nm

2
idler:
-50 100 GHz DWDM 3.8E-3
3
320 GBd
-60 OTDM (QPSK) 4
5
160 GBd, OSNR = 19 dB
-70 6
1520 1530 1540 1550 1560 1570 12 14 16 18 20 22
Wavelength [nm] OSNR [dB]
Fig. 9.8 QPSK and 16QAM data on up to 640 Gbaud converted to DWDM 10 Gbaud channels.
From [25]
(a) DF-HNLF 10 Gbit/s PRBS 2^7-1

DCF 65 m
400 m Optical MUX x16
S=1555 nm 10 G
3 dB RX
WSS 1
2
ERGO com
3 HNLF
4
P2=1563.5 nm 20 nm 200 m 0.8 nm
1557 nm 25.7 dBm
t
DCF
150 m 28,5 dBm
P1=1544.5 nm 16 nm
t
DCF
150 m 25 dBm
(b) (c) 10
Pumps and signal into HNLF (d) WDM signal out of HNLF after PBS
0
0
Power (dBm)
-10 ch -4
-10
Power (dBm)
FWM -20 -20 ch 6
-30 -30
-40 -40
P1 S P2 P1
-50
-50 I P2
P1 S P2 P1 I P2 -60
1540 1550 1560 1570 1540 1545 1550 1555 1560 1565
Wavelength (nm) Wavelength (nm)
Fig. 9.9 Setup and principle of wavelength-preserving serial-to-parallel conversion using two
orthogonally polarized pumps symmetrically spaced about the data channel. a The setup: two
orthogonal pumps are merged with the data signal into a HNLF. b Principle of orthogonally
polarised idler. c Spectrum of the pumps and the signal into the HNLF. d Spectrum of the output of
the HNLF and after a polarisation beam splitter. From [26]
from the signal using e.g. a polarisation beam splitter. Figure 9.9 shows a schematic
setup of this scheme, where a 160 Gbit/s OTDM signal is converted to a DWDM
signal at the same wavelength as the original data. This can reduce the required
spectral bandwidth otherwise needed for FWM OFT, and shows that OFT is
wavelength transparent.
9.3.4 Nyquist-OTDM to OFDM-Like Signal Conversion
In addition to the pursuit of higher spectral efciency by the use of higher order
modulation, signal formats like orthogonal frequency division multiplexing
(OFDM) and Nyquist WDM are also very interesting. They both rely on shaping of
either the spectral or temporal domains to create sinc-function-like shapes and very
dense multiplexing. In a Nyquist channel, the temporal waveform is a sinc-function,
and each symbol goes to zero exactly at the neighbouring symbol position, resulting
in orthogonality. The spectral shape is rectangular, and it is therefore simple to put
WDM channels shoulder-by-shoulder and thus increase the spectral efciency
almost to 1 bit/Hz (for binary modulation). If one combines Nyquist channels with
the potential of ultra-high symbol rate OTDM, one may simply create ultra-high
symbol rate Nyquist channels by the use of rectangular optical ltering directly on
an OTDM signal. This is simple and straightforward, and additionally helps to
reduce spectral redundancy from guard bands that one would otherwise encounter
in low-rate Nyquist WDM channels [27, 28]. In fact, one may say that the higher
the symbol rate, the lower the waste of valuable spectrum. This is because optical
lters will give relatively steeper flanks of the lter compared to the width of the
lter shape for broader spectra. In the time domain, the pulses will become sinc-
function shaped, and if the lter has the width of the symbol rate, then each sinc-
function shaped pulse will exactly give a null in the neighbouring symbol. In [29], it
was shown that a 1.28 Tbaud Nyquist channel could be generated this way using
only 1.28 THz bandwidth, i.e. with no spectral redundancy within the 1.28 THz
bandwidth. This type of spectrally efcient serial data signal can also be demulti-
plexed using a time lens serial-to-parallel converter [29].
Figure 9.10 shows the experimental results for the 1.28 Tbaud Nyquist channel
generation and detection. It is observed that the spectral extent is dramatically
reduced to 1.28 THz, about 10 nm. The characteristic time-domain waveform is
observed in Fig. 9.10b, with the minimum interference points, corresponding to
where the sinc-shaped waveforms go to null. The serial-to-parallel conversion
spectra are shown in Fig. 9.10c. The N-OTDM spectrum is converted to an OFDM-
like spectrum for each channelthe individual spectral channels are not discernable
1.28 Tbit/s Nyquist ch. 100 km transmission

(a)
5 ps (b)
1.28 THz
(10 [nm])
1550 1560
2.56 Tb/s PDM-N-OTDM @ OSNR
N-OTDM to 10%-OFDM of 38.3dB
Power density (dBm/0.1nm)
2
-20
(c) BER = 3E-3
(d)
-30
-40 3
-log(BER)
-50
4
-60
-70 5
-80 6 TE mode
TM mode
1520 1530 1540 1550 1560 1570 7
1 16 32 48 64 80 96 112 128
Wavelength (nm)
Channel (#)
Fig. 9.10 Nyquist-OTDM serial-to-parallel conversion to OFDM-like signal. a Spectral prole

of the original 1.28 Tbaud RZ signal and the ltered Nyquist signal. b The generated temporal
prole of the 1.28 Tbaud Nyquist channel. c The output spectral prole of the time lens based
serial-to-parallel converter. d Measured bit error rate points for all 128 tributaries in two
polarisations, corresponding to 256 points in total yielding 2.56 Tbit/s transmission. From [29]
in this gure, due to the spectral overlap of the channels. The OFT is tailored such
that the generated OFDM-like signal has a large guard interval (around 90 %). This
enables extraction of the tributaries simply using passive ltering, as the associated
pulse broadening is accommodated for by the guard interval. The 1.28-Tbaud N-
OTDM signal is passed through a dispersion compensation bre (DCF) having
2LDCF = 1.243 ps2 and subsequently launched into a 100 m polarisation-main-
taining highly nonlinear bre (PM-HNLF, zero dispersion at 1545-nm and
S = 0.025 ps/nm2km). The desired chirp rate C = 1/(2L) = /t depends on the
temporal spacing and desired frequency spacing. In this work, the target was to map
a 1.28 Tbaud signal with t = 0.781 ps to a frequency spacing of 100 GHz. The
generated pump pulses having a super-Gaussian spectrum (FWHM of 1.6 THz)
centred at 1545 nm are linearly chirped with a chirp rate of C/2 by propagation
through some DCF length and merged with the data in the FWM process in the PM-
HNLF. Since the frequency spacing (100 GHz) is much larger than the baud rate
(10 Gbaud for each tributary), a tuneable OBF with a bandwidth of 40 GHz is used
to extract the subcarriers. The chosen bandwidth is a trade-off between minimisa-
tion of interchannel interference (ICI) and inter-symbol interference (ISI). In this
particular experiment, there was no attempt to obtain many simultaneous converted
channels, but nevertheless 14 tributaries at different wavelengths could still be
simultaneously received with a BER < 3E-3. As explained above, at least half of the
channels should be possible under the right circumstances. Finally it should be
mentioned that the N-OTDM signal was transmitted over 100 km SLA-IDF
transmission bre and that forward error correction coding (FEC) was employed,
resulting in a net 2.4 Tbit/s truly error free data rate after transmission, i.e. with zero
errors occurring in the measurement time.
In [30], it is furthermore shown that polarisation-independent serial-to-parallel
conversion is possible using a polarisation diversity scheme placing the above-
mentioned PM-HNLF in a polarisation diversity loop.
In summary, serial-to-parallel conversion using time lens based OFT has been
described in this section, and various demonstrations have been mentioned. In
particular, it has been experimentally demonstrated that this functionality is trans-
parent to symbol-rate, modulation format, data format, wavelength and polarisation,
and indeed that the nonlinear medium can successfully be a silicon nanowire as well
as a highly nonlinear bre. In addition to this, it should be mentioned that the
opposite process may equally be implemented, and a demonstration of WDM-to-
OTDM conversion has also been demonstrated by converting 16 WDM channels
into 16 OTDM channels [31]. This is a case of spectral-to-time conversion, and
other examples of this include distortion-free pulse transmission [11] and flat-top
pulse generation [12, 32]. Having control of both time-to-frequency and frequency-
to-time conversion opens up for a number of interesting applications, where more
advanced time lens systems may be created using more than one time lens. In the
following some such telescopic arrangements will be described.
9.4 Spectral Telescopes and Applications
Just like in spatial optics, where the combination of several lenses can make more
advanced imaging systems such as telescopes, the same is the case for time lenses.
Both temporal telescopes and spectral telescopes can be constructed [15]. In the
following we will focus on spectral telescopes used for optical processing of
spectrally parallel data signals such as WDM and OFDM signals.
9.4.1 WDM Nonlinear Optical Signal ProcessingWDM

Grid Manipulation
Nonlinear optical signal processing directly on WDM channels is traditionally

challenging due to the nonlinear intermixing of the WDM channels. Therefore
many schemes require a separation of the WDM channels followed by an individual
signal processing of each channel separately. This is not considered a power-
efcient and viable solution as discussed in Sect. 9.1, because one would not be
processing the entire WDM signal in one device. Using optical spectral telescopes
this may change. One could imagine converting a WDM signal to a TDM signal,
where the bits do not overlap in time, and then do some form of optical signal
processing on that TDM signal. Adding a second time lens stage could then bring
the OSPed signal back to the original WDM state or some manipulated form of the
WDM signal.
In this section, we will describe how to use two time lenses to construct a
spectral telescope to manipulate the grid spacing of a WDM signal. This may e.g.
nd use in flexible grid applications.
Figure 9.11 shows a principle spectrographic representation of the use of FWM
for WDM grid manipulation. This representation tool is a simple and illustrative
way to design such time lens systems. An incoming WDM signal, in this case
represented by three wavelengths lling up most of the available time slot and
shoulder-to-shoulder in wavelength, is aligned with a chirped pump pulse for phase
modulation.
FWM occurs between the pump and the signal and creates chirped idlers fol-
lowing the relation xi 2xp xs , with the indices i, p, s corresponding to the
idler, the pump and the signal (WDM inputs), respectively. In this mapping
schematic, each temporal component of the signal is mirrored on the spectral axis
(vertical) through the pump to a position on the other side of the pump at the same
spectral distance to the pump. Thus it is easy to predict the spectrographic prole of
each idler. The idlers are now dispersed, e.g. by transmission through bre with
accumulated dispersion D1 b2 L1 . This has the effect of temporally positioning
the individual frequency components of the idlers following the linear mapping
relation Ds x1 x2 b2 L1 . In this schematic spectrographic representation, the
linear mapping due to dispersion is illustrated by a dashed straight line in which
1 , p (FWM)
2
D2
D1

Dispersion:
FWM: i =2 p s
= (1 ) ( 2 ) = ( 1 2 ) 2L
Fig. 9.11 Spectrographic schematic of a spectral telescope, consisting of two time lenses, based
on FWM, and accompanying dispersion. In this schematic, 3 WDM channels are closely
positioned in optical frequency, , vertical axis and they each ll out their time slots (space
between dotted vertical lines). By FWM using a chirped pump, doubly chirped idler waves are
generated (lower left). These are then dispersed to compensate for the chirp to yield 3 separate
temporal channels. Thus a parallel-to-serial conversion has been obtained. Please note that they are
also slightly displaced in frequency, corresponding to the original spacing between the WDM
channels. 1 and D1 constitute the rst time lens stage. Now, the converted signal is sent through
the second time lens stage with a different focal length corresponding to a different phase (chirp
rate) and dispersion. In this case, 2 and D2 are a factor of two off from 1 and D1, yielding a
telescopic magnication factor of 2. After serial-to-parallel conversion in the second time lens
stage, a magnied WDM spectrum is obtained. The WDM channels are now twice as wide
spectrally compared to the original, and spaced twice as far from each other, and twice as narrow
in the time domain. This makes is possible to use a narrow lter, with the original WDM signal
lter bandwidth on each channel, and the result is a WDM signal but with twice the grid spacing
each spectral component of the generated idler is now simply mirrored hori-
zontally along the time axis to a point on the other side of the dispersion line with
the same distance to this line as the original idler point. This completes the rst step
from parallel to serial conversion, and the WDM channels are now completely
separated in time, i.e. being a serial signal. This rst time lens system is now
followed by a second time lens system, where the dispersion and chirp values may
be chosen freely.
They could be the same, which would lead back to the original signal, or as here,
differ by a factor of 2. Going through the second dispersion (note that the slope of
the D2-line is twice that of the D1-line) and phase modulation, 2, with the FWM
pump-2 being chirped twice as much as pump-1 for 1, and being spectrally twice
as broad, results in a mapping of the WDM channels to different frequencies and
with different pulse and spectral widths. Essentially, as D1/D2 = 2, and the spectral
width of pump-1 is half that of pump-2, the output spectra are twice as broad and
the pulse widths twice as narrow compared to the input. However, the magnication
factor can be dimensioned with a large degree of freedom and does not need to be
an integer number. Just like for a spatial telescope, where the magnication is given
by the ratio of the focal lengths of the objective lens (f1) and the eye piece lens (f2),
M = f1/f2, the magnication for a spectral telescope is given by the equivalent focal
lengths, i.e. the accumulated dispersions: M = D1/D2. In the above example, M = 2,
but can be tailored to a specic application.
For M < 1, the spectrum is compressed. In [33], an example of spectral com-
pression is given. There, a 100 GHz spaced WDM signal consisting of 8 wave-
lengths is compressed to 50 GHz spacing. Each wavelength carries 10 Gbit/s
DPSK, and thus already ts into a 50 GHz grid. The BER performance is very good
with only 1.4 dB sensitivity penalty at a BER = 1E-9.
Figure 9.12 shows an example of M = 2. In this case, 16 WDM channels spaced
at 50 GHz each carrying 10 Gbit/s DPSK data are spectrally magnied to a grid
spacing of 100 GHz. To simplify the setup, a folded time lens scheme is employed
the same nonlinear device (here an HNLF) is used for both time lenses operating
in opposite directions.
The input 16-channel 50 GHz spacing WDM signal enters the grid manipulator
from the top left through a circulator together with a chirped pump (Pump1).
Together they travel from left-to-right through the HNLF into a second circulator
sending the pump, signal and generated idler down into the loop. Here a lter
removes all but the idler, which is now transmitted through a dispersive medium,
either SMF or DCF (anyone can be chosen as long as the chirp is properly matched
to it). After the dispersive medium, the chirped idler is joined with Pump2 through a
coupler and send back through the right-most circulator and into the HNLF from
right-to-left. The now magnied WDM signal is extracted through the circulator
and an optical lter suppressing pump and idler from the loop. Figure 9.12 bottom
WDM + Pump1 Pump2

x
HNLF Chan. 01 - 1544.71 nm

Chan. 02 - 1545.43 nm
500 m Chan. 03 - 1546.16 nm
FWM idler 1
2 Chan. 04 - 1547.95 nm
Chan. 05 - 1547.78 nm
SMFi /DCF i
Chan. 06 - 1548.55 nm
FWM idler 2
Double lens Chan. 07 - 1549.35 nm

system ~ 3 Chan. 08 - 1550.15 nm
WDM
spectral Chan. 09 - 1550.91 nm
out
telescope Chan. 10 - 1551.68 nm
4
-log(BER)
Chan. 11 - 1552.47 nm
-30 -20
Pump 10 GHz Second FWM stage Chan. 12 - 1553.3 nm
-40
First FWM stage
-30 5 Chan. 13 - 1554.11 nm
Chan. 14 - 1554.9 nm
-50
16 WDM channels -40 6
Chan. 15 - 1555.65 nm
Power [dBm]
Power [dBm]
50 GHZ grid
50 GHz
-50 7 Chan. 16 - 1556.38 nm
-60 100 GHz
-60 8 B2B
-70 Idler first converter
-70
9
-80
Tuneable spacing 10
1545 1550 1555 1560 1565 1570 1575 1580
-80
1545 1550 1555 1560 1565 1570 1575 1580
11
-45 -44 -43 -42 -41 -40 -39 -38 -37 -36
Received power [dBm]
Fig. 9.12 WDM grid manipulation using spectral magnication. Left A folded counter
propagation conguration is used for the two time lenses. Bottom left Output of rst FWM
stage (time lens), showing the 16 WDM channels 50 GHz spaced together with Pump1 and the rst
idler. Bottom right Output of the second time lens showing the now magnied WDM signal with
16 channels 100 GHz spaced along with the output spectrum of the second pump and the rst idler.
Right Bit error rate performance. The back-to-back (B2B) is an average of all 16 original WDM
channels
left shows the spectra involved with the 16 WDM channels being 50 GHz spaced at
the input and the output being 100 GHz spaced. A detailed BER characterisation
shows that all channels are error free, Fig. 9.12, right. There is a roughly 2 dB
penalty at BER = 1E-9. By changing the chirp rate of the pump pulses, and
matching the dispersion to it, the magnication may be changed. This may nd use
in e.g. flexible grid networks, where data channels may be easily tted into
available bandwidths and according to required performance criteria.
In the following, an alternative application of spectral magnication is presented.
The idea here is to spectrally magnify an orthogonal frequency division multi-
plexing (OFDM) signal in order to be able to merely receive it by passive optical
ltering.
9.4.2 OFDM-to-DWDM Conversion by Spectral

Magnication
OFDM is a very spectrally efcient modulation format with sinc-function shaped

spectra with a null in the neighbouring spectral channel. These sinc-shaped spectra
are thus overlapping and it is not straightforward to separate these OFDM channels
spectrally, as it is for e.g. DWDM signals. One could say that both a spectral and
temporal ltering is most often required, either in the electrical or the optical
domain. Most work on all-optical OFDM demultiplexing has focused on the
Discrete Fourier transformation (DFT), where up to 26 Tbit/s OFDM demulti-
plexing has been achieved [34]. In optical DFT, the Fourier transformation is
accomplished by matched ltering using cascaded delay-interferometers as sug-
gested by Marhic [35] and subsequent optical gating in a narrow time window
where the interference is absent. This scheme can give very good performance [36,
37]. However, optical DFT requires phase-stabilisation of the optical paths in the
delay-interferometer, and an optical gate per subcarrierin other words, the
receiver complexity increases with the number of subcarriers, and as the optical
gating requires active switching and time alignment, the power consumption thus
also scales with the number of subcarriers.
If instead of DFT, one uses time-domain OFT in a single spectral magnier, the
OFDM signal may be spectrally magnied in a transform-limited way. As above
this means that the spectral components are spread out, whereas the temporal
components are squeezed tighter together within the available time slots. This now
makes it possible to apply bandpass ltering with a lter bandwidth equal to the
symbol rate and actually extract the data information without excessive inter-
symbol and inter-carrier interference (ISI and ICI), as also sketched in Fig. 9.11. In
Fig. 9.11, after spectral magnication, bandpass lters are applied, which broadens
the time waveform and narrows the spectral extent of each channel. As suggested in
Fig. 9.13, this corresponds to a conversion from OFDM to a DWDM-like signal,
which may be received merely by passive ltering, i.e. avoiding optical gating of
AWG
OFDM-
DWDM
Fig. 9.13 Reception of OFDM signals using a WDM-type receiver requires an OFDM-to-
DWDM data format conversion. This may be achieved with a spectral telescope
time time time time

time time
lens 1 lens 2
wavelength ()

Fig. 9.14 Principle of OFDM conversion to a WDM-like data signal using a spectral telescope
[38]. The two time lens stages will create a spectral magnication of the OFDM signal, allowing
for tight direct ltering using e.g. a WDM lter
each channel. As discussed in the introduction for the OTDM-case, there is almost a
two-order of magnitude power reduction when going from N parallel gates to a
single (or two) time lens based serial-to-parallel converters. Here a similar situation
is observed: many optical gates may be replaced by a single spectral magnier and
yield power savings. For AO-OFDM, the gates do not need to be as fast and power
consuming as for the OTDM scheme, so the actual power saving may not be as
dramatic as for OTDM, but the power and complexity scaling with the number of
tributaries can be eliminated this way.
Figure 9.14 shows a principle spectrographic sketch, similar to Fig. 9.11, of the
OFDM-to-WDM-like converter using spectral magnication and subsequent pas-
sive ltering.
Figure 9.14 shows an OFDM signal with its characteristic overlapping sinc-
shaped spectral channels and rectangular waveforms. This OFDM signal is sent
through the spectral magnier consisting of two time lenses separated by disper-
sion. After the rst time lens, the OFDM signal is basically converted to a Nyquist
OTDM signal with sinc-shaped waveforms. The second time lens converts the
signal back to an OFDM signal but with a magnication factor, in this sketch
roughly set to M = 2. Since the time domain waveforms do not ll out the time slots
and the spectra are broad, it is now possible to simply apply a lter with a band-
width equal to the symbol rate, i.e. Df timeslot1. This will broaden the wave-
form to ll out the time slot, but not extend into the neighbouring time slots, i.e. not
generate ISI, and at the same time lter away a considerable part of the spectra of
the neighbouring carriers to reduce the ICI. In principle, the larger the spectral
magnication factor M, the less ICI. However, for the same Df , the larger M is the
more spectral content is ltered away, and hence the less OSNR remains for useful
detection. Therefore, an optimum exists, and empirically M * 4 seems like a good
compromise.
Figure 9.15 shows a schematic setup for of 4 magnication of an OFDM signal
and simulation results in terms of eye diagrams. The spectral telescope consists of
two time lenses and dispersion in between. The simulation assumes 5 10 Gbaud
subcarriers, with 12.5 GHz sinc spectra and 12.5 GHz spacing with DQPSK
modulation. The lower part of Fig. 9.15 shows the case where a bandpass lter
(BPF) is applied directly to the original OFDM signal. The lter bandwidth is tuned
to yield the best possible eye opening under these circumstances. It is evident that
the eye opening is not very good even for an optimum lter bandwidth of 8 GHz.
Figure 9.15 top shows the magnication 4 results. After the second phase mod-
ulation, the spectrum is 4 times broader, and therefore it is seen to be possible to
apply a lter with a bandwidth close to the baud rate to strongly suppress the
neighbouring carriers. In this case the optimum is 15 GHz, and clear and open eye
diagrams are obtained. Hence only a single spectral magnier is required in con-
junction with bandpass ltering. The lter shapes are Gaussian in both cases.
Figure 9.16 shows experimental results in terms of spectra at two points within
the spectral telescope, after the rst time lens and after the second time lens. 10
carriers are used in this experiment, and they are 12.5 GHz spaced and carry
10 Gbit/s DPSK modulation at 10 Gbaud. The OFDM carriers are generated by
spectral sinc-function shaped ltering directly on the 10 Gbit/s DPSK continuous
spectrum, in a scheme adapted from the method proposed and demonstrated in [39]
Spectral magnification + bandpass filtering:

BPF
1 D1 D2 2
15 GHz
(opt)
BPF
Comparison with band -
pass filtering only:
8 GHz
(opt)
Fig. 9.15 Comparison of spectral magnication and bandpass ltering (top) to the case without
magnication (bottom). Simulations show a clear improvement of the eye diagram when spectrally
magnifying 4. From [38]
1 D1 D2 2
-20 Pump1 -20 Pump2 - 10 GHz

10 GHz
OFDM signal signal from
-30 12.5 GHz spacing -30 generated idler 2:
time-lens 1
Magnified
Power [dBm]
Power [dBm]
(idler 1)
-40 -40 OFDM signal
50 GHz spacing
generated idler 1
-50 -50
-60 -60
-70 -70
-80 -80
1545 1550 1555 1560 1565 1570 1575 1580 1545 1550 1555 1560 1565 1570 1575 1580
Fig. 9.16 OFDM-to-DWDM conversion for a 100 Gbit/s OFDM superchannel (10 10 Gbit/s
carriers 12.5 GHz spaced). Spectra after the rst time lens (left) and after the second time lens
(right) revealing the 4 broader OFDM spectrum
-40 after magnification x4

-33
Sensitivity [dBm]
Power [dBm]
-34 Sensitivities: B2B Magnified -50

-35
-36 -60
-37
-38
-39 -70
-40 odd channels
-41 10 Gbit/s DPSK reference (B2B, filtered, odd only) -80 even channels
-42
-4 -3 -2 -1 0 1 2 3 4 5 1546 1547 1548 1549 1550 1551 1552 1553 1554
Subcarrier ID Wavelength [nm]
Fig. 9.17 BER evaluation of spectral magnication of an OFDM signal. The sensitivities are
measured at BER 1E-9 and show a clear improvement with spectral magnication compared to
simply ltering on the unmagnied OFDM signal. The outer channels suffer somewhat due to
spectral broadening caused by imperfections in the time lens. From [38]
by Du et al. The OFDM signal is emulated by separately generating even and odd
subcarriers, each consisting of ve 12.5 GHz sinc functions with 25 GHz spacing,
using a wavelength selective switch (WSS). The sign is reversed between neigh-
bouring sinc subcarriers (for both even and odd), in order to overcome the limited
WSS resolution of *10 GHz and thus obtain the highest possible contrast ratio in
the generated sinc spectra.
The time lenses are based on FWM in HNLF and the 4 magnication is clearly
observed in Fig. 9.16.
Figure 9.17 right shows a zoom in on the magnied spectra. For visualisation
purposes the even and odd channels are shown individually, but they are magnied
as a combined signal for BER characterisation. Figure 9.17 left, shows the BER
performance for all 10 channels, before and after magnication and bandpass
ltering, in terms of receiver sensitivities at a BER of 1E-9. The subcarriers are

individually ltered out using an optical tuneable lter, with a Gaussian prole of
0.12 nm full-width at half maximum (FWHM) for the magnied case and 0.08 nm
FWHM for the unmagnied case. In nearly all cases the bandpass ltered magnied
spectra are better than the original signals after bandpass ltering. Note that as the
data signal used in this case is an emulated OFDM signal, as described above with
even and odd channels and every other carrier being out of phase with its
neighbour, there is a greater isolation between the carriers than in a real OFDM
signal. Therefore it is even possible to get very low BER values by simply bandpass
ltering directly on this emulated OFDM signal. This will not be the case for a real
OFDM signal. None the less, when applying spectral magnication to this signal,
the sensitivity is improved on average with about 2 dB. As seen above in Fig. 9.15,
the simulated eye diagrams would suggest a much greater improvement if used on
real OFDM signals, where one would not expect to be able to detect the bits, with
the eye being completely closed. The outer channels in Fig. 9.17 are worse for the
magnied case, which is due to small aberrations of the time lenses. These cause
distortions on the magnied spectra of the outer channels (see arrows in Fig. 9.17
right), and these then cause ICI with their neighbours. With better optimisation of
the time lenses, these aberrations are expected to disappear.
This application may become useful for AO-OFDM superchannels with many
subcarriers, where one may wish to eliminate the need for equally many optical
gates by introducing a single active element, namely the spectral telescope. The
telescope is in its current implementation more complex and power consuming than
some simple optical gates, like the eletroabsorption modulator used in [34] or low-
energy variants as described in [40]. But there may be potential for low-energy time
lenses based on e.g. amorphous silicon nanowires [41], which may indeed also hold
potential for ultra-wideband operation over several hundred nm [42, 43]. This
would allow for reduced power consumption and handling of ultra-broadband
superchannels.
Other applications of time lens telescopic arrangements include temporal tele-
scopes, as e.g. demonstrated in [44] for data packet compression. In that case, a
short data packet is compressed a factor 27 in time to bring the bit rate from 10 to
270 Gbit/s. In some applications, a smaller modication of the bit rate may be
desired, and in that case a simplied time lens system may be used, so instead of a
full telescope, a single lens may be used. In the following, such an example is
given.
9.5 Time-Domain Processing
One of the early communication applications utilizing time lenses was for timing
jitter compensation or retiming [18], which was basically based on an electro-optic
phase modulator with sinusoidal modulation and a dispersive element, as sketched
in Fig. 9.4. The same basic setup was also used for dispersion compensation [11]
and later pulse shaping [13, 32] and pulse position locking [45]. With this imple-
mentation, the sinusoidal modulation implies that one can only operate at a bit rate
at or below the rate of the modulation. This is in contrast to the FWM based time
lens using flat-top pulses. If the e/o-phase modulation could be truly parabolic, then
the situation would be similar to the FWM cases above. However, one of the
attractions of the sinusoidal modulation is its simplicity. It is very convenient to
obtain sinusoidal modulation.
Figure 9.18 shows an example where the time lens system is set to operate on an
Ethernet data packet. Ethernet data packets/frames are very challenging to handle
optically, as the packet length may vary from 64 to 1518 bytes, they arrive asyn-
chronously and they have a certain allowed repetition rate variation, which in the 10
GE WAN PHY standard can vary with up to 20 ppm of the nominal transmission
rate, i.e. 200 kHz frequency offset between transmitter and receiver for 10 Gbit/s
Ethernet packets. The idea in Fig. 9.18 is to use a time lens to synchronise the
10 Gbit/s Ethernet data packets in such a strict manner that they can be optically
time multiplexed together to form a Tbit/s multiplexed Ethernet data signal (TbE).
The requirements to Tbit/s timing jitter are very strict (*50 fs) and thus the
synchronisation has to be very effective. Furthermore, it needs to be very flexible to
accommodate for the varying types of data packets and include NRZ-to-RZ
conversion.
Figure 9.18 left shows the schematic setup. An NRZ data packet arrives to the
simple time lens system, where an additional amplitude modulator based on a
Mach-Zehnder modulator (MZM) is introduced. The MZM carves part of the NRZ
bit out, thus effectively converting from NRZ to RZ, and the phase modulator adds
the required chirp to the bit, and the following dispersive element aligns the now
RZ bit to the focal point within the time slot, i.e. at the peak of the sinusoidal
modulation, where it approaches a parabolic phase modulation and the chirp is
linear, being the derivative of the phase modulation. The phase modulator modu-
lation depth is 4, and it and the MZM are both driven by a master clock at
9.9536 GHz (200 kHz offset from the input data packet clock), and then launched
into a 400 m DCF. The 10 Gbit/s input asynchronous Ethernet packet with a data
Fig. 9.18 Time lens based Ethernet packet synchronisation for 1.29 Tbit/s data streams. From [46]
rate of 9.9534 Gbit/s is thus converted into a synchronized Ethernet packet with a
data rate of 9.9536 Gbit/s. At the same time, the Ethernet packet is format converted
into an RZ signal with a full width at half maximum (FWHM) of 6 ps, due to the
pulse carving of the MZM and the chirping. Additionally, the converted RZ signal
is further pulse compressed to a FWHM of 400 fs in a 500 m dispersion-flattened
highly nonlinear bre (DF-HNLF), not shown here [46]. Alongside this synchro-
nisation part of the demonstration setup, a standard 1.28 Tbit/s OTDM data signal is
generated, but with the time channels moved together a small fraction so as to make
room for an additional empty time slot. The now synchronised RZ data packet is
then multiplexed together with the 1.28 Tbit/s OTDM signal to form a 1.29 Tbit/s
aggregate signal, as seen to the right in Fig. 9.18. The Ethernet packet channel ts
into the vacant OTDM channel, and when demultiplexing and BER characterizing,
the performance is sufciently good to reach a BER of 1E-9.
Thus this time lens scheme does full the 10 GE WAN PHY requirements and is
able to operate on asynchronous standard Ethernet packets with 200 kHz offset, and
a maximum packet length of 1518 bytes to the extent that they may be multiplexed
to Tbit-Ethernet, TbE.
9.6 Summary
This chapter has addressed the use of optical time lenses for efcient optical signal
processing. We have identied and described scenarios where time lens based OSP
with its ultrafast potential can offer power-efciency. This goes for applications
where many bits are processed simultaneously, such as for serial-to-parallel con-
version. Other applications where all the data channels are processed simulta-
neously include the possibility of processing WDM signals to e.g. manipulate grid
spacing using a spectral telescope. The same approach may be used for spectral
magnication of OFDM signal in order to receive AO-OFDM signals directly using
WDM-like receivers relying on passive bandpass ltering. This, like serial-to-
parallel conversion, would break the scaling of complexity and power with number
of tributaries, since one would only need a single spectral telescope. Finally, we
showed how time lenses would enable the optical synchronisation and multiplexing
of standard Ethernet packets to create a new multiplexed type of Tbit/s Ethernet
with ultrahigh timing precision and format conversion.
All in all, we nd that time lens based optical Fourier transformation is a very
exciting and versatile tool for optical communications, allowing for a range of
optical signal processing applications, which may offer energy efciency as well as
advanced functionality otherwise unobtainable, relying on the ultrafast nature of
e.g. four-wave mixing and implementations offering transparency to data and signal
formats, wavelength, polarisation and bit rate. Ultra-broadband operation may also
be obtainable with nonlinear waveguides, such as nanowires, and with future
developments in this eld, ultra-low energy switching using time lenses may be
realistic.
References
1. M. Mauldin, The state of the global internet, Webinare (2011). www.telegeography.com

2. E.B. Desurvire, Capacity demand and technology challenges for lightwave systems in the next
two decades. J. Lightwave Technol. 24(12), 46974710 (2006)
3. http://www.redorbit.com/news/science/1112756793/internet-produces-greenhouse-gas-010313/
4. K. Hinton, G. Raskutti, P.M. Farrell, R.S. Tuckeret, Switching energy and device size limits on
digital photonic signal processing technologies. J. Sel. Top. Quantum Electron. 14(3), 938945
(2008)
5. R.S. Tucker, K. Hinton, Energy consumption and energy density in optical and electronic
signal processing. IEEE Photonics J. 3(5), 821833 (2011)
6. M. Galili, L.K. Oxenlowe, H.C.H. Mulvad, A.T. Clausen, P. Jeppesen, Optical wavelength
conversion by cross-phase modulation of data signals up to 640 Gb/s. J. Sel. Top. Quantum
Electron. 14(3), 573579 (2008)
7. M. Lillieholm, H.C.H. Mulvad, E. Palushani, C. Peucheret, P. Jeppesen, L.K. Oxenlwe, in
IEEE Photonics Conference (IPC). Numerical Investigation of Power Requirements for Ultra-
high-speed Serial-to-parallel Conversion, Burlingame, California, paper MM5, 2327 Sept
2012
8. H.C.H. Mulvad, E. Palushani, H. Hu, H. Ji, M. Lillieholm, M. Galili, A.T. Clausen, M. Pu, K.
Yvind, J.M. Hvam, P. Jeppesen, L.K. Oxenlwe, Ultra-high-speed optical serial-to-parallel
data conversion by time-domain optical Fourier transformation in a silicon nanowire. Opt.
Express 19(26), B825B835 (2011)
9. J. Xu, C. Peucheret, P. Jeppesen, in Proceedings of the ICTON 2010. Power Consumption
Comparison Between Point-to-point WDM and OTDM Systems, Munich, Germany, paper Th.
A1.1 (2010)
10. B.H. Kolner, Space-time duality and the theory of temporal imaging. J. Quantum Electron. 30
(8), 19511963 (1994)
11. M. Nakazawa, T. Hirooka, F. Futami, S. Watanabe, Ideal distortion-free transmission using
optical Fourier transformation and Fourier transform-limited optical pulses. Photonics
Technol. Lett. 16(4), 10591061 (2004)
12. G.P. Agrawal, Fiber-Optic Communication Systems (Wiley, New York, 1997). ISBN:
0471175404/0-471-17540-4
13. E. Palushani, L.K. Oxenlwe, M. Galili, H.C. Hansen Mulvad, A.T. Clausen, P. Jeppesen,
Flat-top pulse generation by the optical Fourier transform technique for ultrahigh speed signal
processing. J Quantum Electron. 45(11), 13171324 (2009)
14. E. Palushani, H.C. Hansen Mulvad, M. Galili, H. Hu, L.K. Oxenlwe, A.T. Clausen,
P. Jeppesen, OTDM-to-WDM conversion based on time-to-frequency mapping by time-
domain optical Fourier transformation. IEEE J. Sel. Top. Quantum Electron. 18(2), 681688
(2012)
15. R. Salem, M.A. Foster, A.L. Gaeta, Application of spacetime duality to ultrahigh-speed
optical signal processing. Adv. Opt. Photonics 5(3), 274317 (2013)
16. J. Azaa, Time-to-frequency conversion using a single time lens. Opt. Commun. 217(1), 205
209 (2003)
17. J. Azaa, M.A. Muriel, Real-time optical spectrum analysis based on the time-space duality in
chirped ber gratings. IEEE J. Quantum Electron. 36(5), 517526 (2000)
18. L.F. Mollenauer, C. Xu, Time-lens timing-jitter compensator in ultra-long haul DWDM
dispersion managed soliton transmissions. CLEO, paper CPDBl (2002)
19. Y. Han, B. Jalali, Photonic time-stretched analog-to-digital converter: fundamental concepts
and practical considerations. J. Lightwave Technol. 21, 30853103 (2003)
20. I.S. Lin, J.D. McKinney, A.M. Weiner, Photonic synthesis of broadband microwave arbitrary
waveforms applicable to ultra-wideband communication. IEEE Microwave Wireless Compon.
Lett. 15, 226228 (2005)
21. J. van Howe, Chris Xu, Ultrafast optical signal processing based upon space-time dualities.
J. Lightwave Technol. 24(7), 2649 (2006)
22. R. Salem, M.A. Foster, A.C. Turner, D.F. Geraghty, M. Lipson, A.L. Gaeta, Optical time lens
based on four-wave mixing on a silicon chip. Opt. Lett. 33(10), 10471049 (2008)
23. K.G. Petrillo, M.A. Foster, Scalable ultrahigh-speed optical transmultiplexer using a time lens.
Opt. Express 19(15), 1405114059 (2011)
24. H.C.H. Mulvad, E. Palushani, H. Hu, H. Ji, M. Galili, A.T. Clausen, M. Pu, K. Yvind, J.M.
Hvam, P. Jeppesen, L.K. Oxenlwe, in Proceedings of the ECOC 2011. Ultra-high-speed
Optical Serial-to-parallel Data Conversion in a Silicon Nanowire, Geneva, Switzerland, paper
Th.13.A.2 (2011)
25. E. Palushani, T. Richter, R. Ludwig, C. Schubert, H.C. Hansen Mulvad, A.T. Clausen, L.K.
Oxenlwe, in Proceedings of the OFC 2012. OTDM-to-WDM Conversion of Complex
Modulation Formats by Time-domain Optical Fourier Transformation, Los Angeles,
California, USA, paper OTh3H.2 (2012)
26. M. Galili, E. Palushani, H.C.H. Mulvad, H. Hu, L.K. Oxenlwe, Wavelength preserving
optical serial-to-parallel conversion, OFC/NFOEC technical digest. Opt. Soc. Am. OM2G.4
(2013)
27. M. Nakazawa, T. Hirooka, P. Ruan, P. Guan, Ultrahigh-speed orthogonal TDM transmission
with an optical Nyquist pulse train. Opt. Express 20, 11291140 (2012)
28. H.N. Tan, T. Inoue, T. Kurosu, S. Namiki, Transmission and pass-drop operations of mixed
baudrate Nyquist OTDM-WDM signals for all-optical elastic network. Opt. Express 21,
2031320321 (2013)
29. H. Hu, D. Kong, E. Palushani, J.D. Andersen, A. Rasmussen, B.M. Srensen, M. Galili, H.C.H.
Mulvad, K.J. Larsen, S. Forchhammer, P. Jeppesen, L.K. Oxenlwe, in CLEO Technical Digest.
1.28 Tbaud Nyquist Signal Transmission Using Time-domain Optical Fourier Transformation
Based Receiver, Optical Society of America, Postdeadline paper CTh5D.5 (2013)
30. H. Hu, D. Kong, E. Palushani, M. Galili, H.C.H. Mulvad, L.K. Oxenlwe, in ECOC 2013,
Detection of 320 Gb/s Nyquist OTDM Received by Polarisation-insensitive Time-domain
Optical Fourier Transformation, paper We.1.C.4 (2013)
31. H.C.H. Mulvad, H. Hu, M. Galili, H. Ji, E. Palushani, A.T. Clausen, L.K. Oxenlwe,
P. Jeppesen, in Proceedings of the ECOC 2011. DWDM-to-OTDM Conversion by Time-
domain Optical Fourier Transformation, Geneva, Switzerland, paper Mo.1.A.5 (2011)
32. L.K. Oxenlwe, M. Galili, A.T. Clausen and P. Jeppesen, in European Conference on Optical
Communication, ECOC 06. Generating a Square Switching Window for Timing Jitter
Tolerant 160 Gb/s Demultiplexing by the Optical Fourier Transform Technique, Cannes,
France, Paper We2.3.4, Sept 2006
33. E. Palushani, H.C. Hansen Mulvad, M. Galili, F.D. Ros, H. Hu, P. Jeppesen, L.K. Oxenlwe,
in OptoElectronics and Communications Conference, OECC 2013. Spectral Compression of a
DWDM Grid Using Optical Time-lenses, Kyoto, Japan, Paper ThO2-1, July 2013
Bonk, P. Kleinow, F. Frey, M. Roeger, S. Koenig, A. Ludwig, A. Marculescu, J. Li, M. Hoh, M.
Dreschmann, J. Meyer, S. Ben Ezra, N. Narkiss, B. Nebendahl, F. Parmigiani, P. Petropoulos, B.
Resan, A. Oehler, K. Weingarten, T. Ellermeyer, J. Lutz, M. Moeller, M. Huebner, J. Becker, C.
Koos, W. Freude, J. Leuthold, 26 Tbit/s line-rate super-channel transmission utilizing all-optical
fast Fourier transform processing. Nat. Photonics 5, 364371 (2011)
35. M.E. Marhic, Discrete Fourier transforms by single-mode star networks. Opt. Lett. 12(1), 6365
(1987)
36. H. Sanjoh et al., in Proceedings of the OFC 2009.Optical Orthogonal Frequency Division
Multiplexing Using Frequency/time Domain Filtering for High Spectral Efciency up to 1 bit/
s/Hz, ThD1 (2009)
37. A.J. Lowery, Design of arrayed-waveguide grating routers for use as optical OFDM
demultiplexers. Opt. Express 18, 14129 (2010)
38. E. Palushani, H.C.H. Mulvad, D. Kong, P. Guan, M. Galili, L.K. Oxenlwe, All-optical
OFDM demultiplexing by spectral magnication and band-pass ltering. Opt. Express, 22(1),
136144 (2014). (ISSN: 1094-4087) (doi:http://dx.doi.org/10.1364/OE.22.000136)
39. L.B. Du, J. Schroeder, J. Carpenter, B. Eggleton, A.J. Lowery, in Proceedings of the OFC13.
Flexible All-Optical OFDM using WSSs, PDP5B.9 (2013)
40. L.K. Oxenlwe, Optical communications: single-laser superchannel, news and views. Nat.
Photonics 5, 329331 (2011)
41. K.-Y. Wang, K.G. Petrillo, M.A. Foster, A.C. Foster, Ultralow-power all-optical processing of
highspeed data signals in deposited silicon waveguides. Opt. Express 20(22), 2460124606
(2012)
42. M. Pu, H. Hu, M. Galili, H. Ji, L.K. Oxenlwe, K. Yvind, P. Jeppesen, J.M. Hvam, in
Conference on Lasers and Electro-Optics, CLEO 11.Ultra-Broadband Tunable Wavelength
Conversion of Sub-Picosecond Pulses in a Silicon Nanowire, Baltimore, Maryland, U.S.A.,
Paper CMAA1, May 2011
43. N. Ophir, R.K.W. Lau, M. Menard, X. Zhu, K. Padmaraju, Y. Okawachi, R. Salem, M. Lipson,
A.L. Gaeta, K. Bergman, Wavelength conversion and unicast of 10-Gb/s data spanning up to
700 nm using a silicon nanowaveguide. Opt. Express 20(6), 64886495 (2012)
44. M.A. Foster, R. Salem, Y. Okawachi, A.C. Turner-Foster, M. Lipson, A.L. Gaeta, Ultrafast
waveform compression using a time-domain telescope. Nat. Photonics 3, 581585 (2009)
45. C.W. Chow, A.D. Ellis, F. Parmigiani, Time-division-multiplexing using pulse position
locking for 100 Gb/s applications. Opt. Express 17(8), 65626567 (2009)
46. H. Hu, J. Laguardia Areal, H.C. Hansen Mulvad, M. Galili, K. Dalgaard, E. Palushani, A.T.
Clausen, M.S. Berger, P. Jeppesen, L.K. Oxenlwe, in European Conference on Optical
Communication, ECOC11. Synchronisation, Retiming and OTDM of an Asynchronous 10
Gigabit Ethernet NRZ Packet Using a Time Lens for Terabit Ethernet, Geneva, Switzerland,
Paper Tu.3.K.4, Sept 2011
Chapter 10
Signal Processing Using Opto-Electronic
Devices
Mary McCarthy, Simon Fabbri and Andrew Ellis
Abstract All-optical signal processing is a powerful tool for the processing of

communication signals and optical network applications have been routinely con-
sidered since the inception of optical communication. There are many successful
optical devices deployed in todays communication networks, including optical
amplication, dispersion compensation, optical cross connects and recongurable
add drop multiplexers. However, despite record breaking performance, all-optical
signal processing devices have struggled to nd a viable market niche. This has
been mainly due to competition from electro-optic alternatives, either from detailed
performance analysis or more usually due to the limited market opportunity for a
mid-link device. For example a wavelength converter would compete with a rec-
ongured transponder which has an additional market as an actual transponder
enabling signicantly more economical development. Never-the-less, the potential
performance of all-optical devices is enticing. Motivated by their prospects of
eventual deployment, in this chapter we analyse the performance and energy
consumption of digital coherent transponders, linear coherent repeaters and mod-
ulator based pulse shaping/frequency conversion, setting a benchmark for the
proposed all-optical implementations.
10.1 Introduction
The demand for communication capacity has continued to increase exponentially

over the last 35 years [1] regardless of repeated projections in the intervening period
that this growth rate will slow down or stop. Since the introduction of optical bre
[2] the optical communications industry initially met this growth with symbol rate
increases, followed by with wavelength division multiplexing (WDM) [3]. As we
M. McCarthy (&) S. Fabbri A. Ellis (&)

Aston University, Aston Triangle, Birmingham B4 7ET, UK
e-mail: andrew.ellis@aston.ac.uk

292 M. McCarthy et al.
approach the (*5 THz) bandwidth limit of conventional amplication technology,

research has focused on enabling higher complexity modulation formats to access
bre capacity [4]. As total volume of data transported across the network has
grown, energy consumption, as well as increased capacity, have become a critical
drivers in deploying new technologies [5]. In addition to transponders at the end
terminals of an optical network, certain signal processing functions, such as
regeneration, format conversion, wavelength conversion and arbitrary waveform
generation are often proposed. As described extensively elsewhere in this book,
many of these intermediate functions may be performed all optically, with the
primary advantage of increased bandwidth and consequent resource sharing.
However these functions may be performed using the very optoelectronic devices
used in the transponders themselves, either including decision circuitry and/or FEC
or as linear media converters. In this chapter, we analyse the performance and
energy consumption of digital coherent transponders and modulator based pulse
shaping/frequency conversion, setting a benchmark for the proposed all-optical
implementations.
10.2 Coherent Transponders
After low loss single mode bres, the most transformative invention in optical
communications was the Erbium Doped Amplier (EDFA), enabling wavelength
division multiplexed (WDM) transmission over long distances [6]. This has fuelled
much of the increase in capacity up to the mid-2000s. In the rst coarse WDM
systems, where only a few wavelengths were amplied by each EDFA, the power
consumption of the system was dominated by the ampliers but as the information
spectral density of the systems have increased the link power budget has become
dominated by the transponders.
The power consumption of EDFAs should ideally be dominated by the current
required for the semiconductor un-cooled pumps (650 mW) with a simple electrical
to optical power conversion efciency of about 25 % achievable in practice.
However, in terrestrial systems, cooling, control loops and supervisory overhead
using robust technologies have come to dominate [7]. Terrestrial WDM ampliers
consume between 10 and 25 W of electrical power in order to amplify optical
signals up to 26 dBm. It is likely that, once the power consumption of transponders
has been considered and as the restrictions on power consumption become ever
more stringent, a trade-off between remote supervision of links and power con-
sumption of amplier nodes will be considered. Similar control and supervisory
overheads should be considered for any all-optical processing nodes, all of which
are likely to require at least one additional optical amplier to overcome device
insertion loss.
Once the available spectrum is allocated to its maximum permitted extent by a
combination of WDM and increased symbol rate, it is necessary to increase number
of bits per symbols, many different constellations are possible, but here we focus on
10 Signal Processing Using Opto-Electronic Devices 293
Fig. 10.1 Constellation diagrams for QPSK (left), 16QAM (centre) and 64QAM (right) with a
signal to noise ratio of 20
the well-known rectangular QAM family as shown in the constellation (amplitude

and phase) diagrams of Fig. 10.1 for an increasing number of constellation points
M, where each symbol coveys log2M bit per symbol giving a total bit rate per
wavelength Rb of.
Rb Rs log2 M 10:1
where Rs is the symbol rate. Increasing the number of constellation points that can
be encoded increases the required digital to analogue converter resolution and the
complexity of the receiver digital signal processing (DSP). This is mitigated by the
increased number of bits that are processed simultaneously. In what follows, we
optimise the constellation for minimum energy consumption.
Whilst dual quadrature formats were originally detected and simultaneously
demodulated using optical interferometers [8], coherent detection is currently used
to enable enhanced receiver sensitivity and digital impairment compensation. For a
coherent receiver, the requirement for polarisation diversity [9] implies that both
polarisation multiplexing and compensation of polarisation mode dispersion (PMD)
in the DSP are straightforward, doubling the bit rate shown in (10.1).
10.2.1 Power Consumption of Opto-Electronic Devices
Narrow linewidth lasers are required for coherent detection to both minimise phase
noise induced error floors and reduce the complexity of the phase recovery of the
coherent receiver digital signal processing. Whilst there is much development in
narrow linewidth widely tunable lasers, especially for use in integrated devices,
current state of the art lasers [10] have a power consumption of 3.55 W repre-
senting an approximately 10 % electrical to optical power conversion efciency.
Integration will reduce insertion loss by several dBs and consequently the required
output powers [11] and in certain circumstances one laser can be shared between
transmitter and receiver part of a coherent transponder module.
Optical modulators with bandwidths well in excess of 28 GHz have been

commercially available for many years. For binary systems they may be driven by
limiting ampliers which may operate with high power efciency. For coherent
systems, dual polarisation IQ modulators with low drive voltages are required and
are currently based on several key technologies including LiNbO3 (3.5 V) [12],
Silicon (3.25 V) [13], InP (*2.5 V) [14] and polymer (2.5 V) based photonics [15].
In production systems automatic bias control circuitry is required, but power
consumption is dominated by drive amplier power consumption (typically
determined by the 50 impedance of the connector standard selected) and any
DAC required for pulse shaping or pre compensation of channel impairments.
Ideally modulator drive voltages for these modulators should be reduced to match
the DAC output.
In order to implement modulation formats with high cardinality (16QAM and
above), linear ampliers are required. Whilst switched mode ampliers are useful
for narrowband wireless systems [20], for broadband ampliers the linearity
requirement results in a substantial increase in power consumption to minimise the
signal distortion. Figure 10.3 shows the power consumption of typical commer-
cially available drivers. For the limiting ampliers, the power consumption is
dominated by the amplier output power (3 V in 50 is *0.18 W) and conse-
quently is independent of the required bandwidth. This power consumption should
be at least doubled for linear drivers. Consequently we allocate 3.3 W per quad
drive amplier (one driver for each modulator in a of a PM-IQ modulator).
The Optical Internetworking Forum (OIF-DPC-RX-01.0) have agreed a stand-
ardised design for the coherent receiver [21]. It consists of a polarisation beam
splitter, two 90 Hybrids, four balanced photodiodes and a transimpedance
amplier (TIA) array as shown in Fig. 10.2. The main source of power dissipation
is the TIA. Figure 10.3 shows the power dissipation of currently available TIAs and
TIAs with appropriate automatic gain control circuitry, typically required to match
the output voltage the optimum range of the analogue to digital converters. Here the
output power is a negligible contribution to the power consumption, which illus-
trates the anticipated frequency dependence for devices without AGC. However, for
ease of implementation, we assume the use of AGC circuits, and consequently a
xed power consumption of around 0.8 W per TIA.
Fig. 10.2 Schematic diagram of a typical coherent transponder showing FEC, DSP, drivers,
passive and active optics. Framers, gearboxes and buffers are omitted for simplicity
Fig. 10.3 Left Power consumption of commercially available modulator drivers showing limiting
ampliers (blue) and analogue ampliers (red) [1619]. Right Power consumption of trans
impedance ampliers with (red) and without (blue) automatic gain control [19]
10.2.2 Energy Consumption of DSP
Consideration of the opto-electronic subsystems alone reveals that the power

consumption is largely independent of the symbol rate, suggesting that the highest
possible rate should be deployed as the appropriate technologies become available,
and indeed this was the case for much of the history of optical communications up
to systems operating at 40 Gbit/s. However, the generation and, in particular, the
detection of optical signals using coherent receivers requires the use of digital signal
processing which adds signicantly to the overall energy consumption. As shown
when coherent detection was rst investigated, DSP can clearly be avoided [22],
however this requires widely tuneable lasers with greater stability and limits the
opportunities to compensate for linear impairments. To account for the signal
processing, the signal processing and DAC/ADC requirements should be
considered.
Accurate design of CMOS DSP chips is a lengthy process requiring several steps
after optimised algorithms have been identied, and a full analysis is beyond the
scope of this book. Full details are design processes and the potential efciency
gains are now appearing following rigorous analysis, as shown in, for example [23].
Many detailed design choices are required, including choice of algorithms, paral-
lelism [24], layout, resolution, target latency and performance [25]. Whilst such
exact calculations are complex, estimation of power consumption through simple
counting of the required basic building blocks of digital signal processing (adders
and multipliers) is readily performed. The total resources need to take into account
both the number of operations presented in the algorithm, and also the precision to
which this operation is carried out. An adder with N bit precision requires N 1
stages and an N bit multiplier requires N 1 adders hence complexity of a mul-
tiplier scales as (N 1)2. This holds for both carry-save and Wallace Tree multiplier
architectures. As the precision N increases, multipliers will dominate in terms of
power consumption so we will further simplify our cost analysis to only include
complex multiplication which will require 3 real multipliers, two additions and
Boolean overhead per operation [26]. It has been demonstrated that the required
resolution for rectangular QAM constellations with M constellation points varies
slightly faster than the number of bits carried by the format [27, 28], is shown in
Fig. 10.4 and may be approximated as
p
NADC 2:8 log2 M 10:2
We assume that a radix-2 algorithm FFT requires N=2 log2 N complex multi-
plications and that trigonometric functions are estimated using CORDIC algorithm
with 3N adders [29]. Some commercially available digital coherent receivers
contain transmitter DSP and digital to analogue converters. These allow for the
generation of high cardinality modulation formats, pulse shaping (including OFDM
and Nyquist WDM), compensation of transmitter and/or a line impairments such as
dispersion and to limited extent nonlinearity [30]. DAC performance is limited by
the linearity of the circuit and quantization noise which is evaluated in terms of
effective number per bits (ENOB). ENOB is limited by integral nonlinearity, dif-
ferential nonlinearity, total harmonic distortion and noise [31] which limits both the
accuracy and speed of the DAC. An ENOB of at least 3.8 is required to generate
QPSK and at least 4.9 to generate 16QAM.
As can be seen from Fig. 10.5, DAC power consumption has remained relatively
constant with respect to sample rate. State of the art converters [33] consume
0.75 W and offer 8-bit physical resolution and sample rates up 65 GS/s which
should allow 2.3 oversampling for 28 Gbaud signals.
With regards to the transmitter DSP, some of the functions will be specic to the
transmitter hardware and can readily be implemented by a factory set look up table
(LUT) which will have relatively low power consumption but others will be link
dependent and may need to be tailored dynamically. These will take the form of
nite input response (FIR) lter whose length will be determined by the complexity
of the signal pre-conditioning required. The complexity of the lter for chromatic
Fig. 10.4 Minimum required ADC resolution for digital coherent receivers (points) [27],
approximation (10.2, solid line) and 1 bit margin for ENOB degradation (dashed line)
Fig. 10.5 Left Commercially available DAC power consumption as a function of sample rate [19,
32, 33]. Right Power consumption of ADCs [19, 32, 3438]
dispersion compensation will be discussed in the receiver digital signal processing

section and may be arbitrarily shared between transmitter and receiver with neg-
ligible implications for overall energy consumption. The most common signal
conditioning is currently Nyquist ltering and its variants, offering the highest
spectral efciency. Ideally the lter impulse response would be a Sinc function,
however this requires innite taps. In practice, a RRC lter gives a very good
approximation to a truncated Sinc function. Use of a RRC lters results in a lter at
the transmitter and a matched lter at the receiver.
t cospat
RRCt sinc T
; 10:3
T 1 4aT22t2
where T is the symbol period and is the roll-off factor. Typically the limiting
factor in the implementation of Nyquist lter will be the ENOB and sample rate of
the DAC rather than the number of taps in the FIR lter in the receiver so we will
limit the length of the lter considered in this work to four lters of sixty four taps
each for 28 GBaud signal [39]. All current systems employ some form of forward
error correction. The complexity of forward error correction encoding depends on
the latency constraints and the complexity of the code. However for RS(255,239)
encoding, this can be implemented using a shift register so the increase in trans-
mitter side power consumption is small compared to other functions.
Digital Signal Processing in the receiver is the major consideration for any
digital coherent receiver, allowing for polarisation mode dispersion compensation
and carrier frequency and phase error correction. Receiver DSP discussed in detail
in many, many publications and details of the algorithms may be found in [4042].
Its implementation can be divided in ve main areas in terms of power
consumption.
Optimisation of receiver DSP is currently an active area of research so we will
draw on recent published papers and bench mark examples with acknowledgement
that further, potentially substantial, improvement is expected. We consider each of
the ve major blocks individually (Fig. 10.6).
10.2.2.1 Analogue to Digital Conversion
High speed analogue to digital converters (ADCs) have been the single most
important key technology in the introduction of DSP to optics. There are many
architectures [27] which trade off speed, latency, NOB and implementation size. They
are generally specied in ENOB similar to DACs as reported in the previous section.
The main source of reduction of ENOB at high frequencies is clock jitter which
results in ADC being unable to sample at precise intervals. Figure 10.5 shows details
of power consumption for ADCs versus Sampling Rate over the last decade or so,
again the best available devices show negligible variation in power consumption with
sampling rate. The latest result demonstrates the availability of 56 Gbit/s ADC with
8 bit resolution at 1.2 W power consumption. This allows for 2 sampling at 28 GHz.
Progress in ADC performance usually tracks the development of successive
CMOS nodes, however the input referred noise has been increasing as the feature
sizes reduce below 90 nm and so it is possible that we are very close to the optimum
design for electronic ADCs and further improvement is ENOB and power con-
sumption will be incredibly challenging [31, 43]. This may result in a stagnation of
coherent DSP based systems at *28 Gbaud.
Coefficient initialisation and slow update
ADC
QSE DE
FOE
ADC FIR
CPE
FEC
ADC
FOE
FIR
CPE
QSE
CMA VV DD
ADC
Continuous coefficient update
Fig. 10.6 Schematic diagram of common DSP conguration including (1) analogue to digital converters
(ADC), (2) quasi static equaliser (QSE) for matched ltering, chromatic dispersion compensation and
PMD, (3) dynamic equaliser comprising time domain lters (FIR) in a butterfly conguration, (4)
frequency offset estimation (FOE) and carrier phase estimation (CPE) and (5) forward error correction
decoder (FEC)/symbol decisions. Also shows options for updating lters and estimators based on
constellation analysis (e.g. constant modulus algorithm (CMA)), nonlinearly processes signal analysis
(e.g. Viterbi and Viterbi phase estimation (VV)) and decision directed (DD) feedback
10.2.2.2 Quasi-static Equaliser [44]
Chromatic dispersion and quasi-static PMD can be compensated for using a

combination of a slowly updating (quasi static) lter and a rapidly updating
dynamic lter. In the quasi static lter, the expected chromatic dispersion is
equalised (with a memory length of sCD ) and the bulk of the polarisation mode
dispersion (typically up to three times the mean value of the differential group delay
3hsDGD i) are equalised with a long averaging time [45]. Fine tuning, particularly of
the polarisation mode dispersion, is performed in the dynamic equaliser (see
below). Hardware efcient clock recovery can also be implemented as part of this
structure [46].
The PMD tolerance that may be compensated in the quasi-static equalizer is a
variable design parameter depending on what is acceptable level of outage (typi-
cally 1015) when the dynamic equaliser is included. Thus it is possible to estimate
a conservative bound for the length of the lter, NCIR, to compensate the channel
impulse response (CIR) as
n
NCIR sCD 3hsDGD i 10:4
T
where n is the over sampling ratio and T is the symbol period. A time domain
equalizer to include some PMD compensation would require 4 such butterfly FIR
lters giving a total number of multiplications per bit

NCIR
cos tTDE 4 10:5
2 log2 M
However it has been shown extensively that for channels with realistic memory
lengths (10.4), the complexity of the time domain lter far out strips that of a
Fourier or frequency domain implementation [47, 48]. The complexity of frequency
domain algorithm is dominated by the FFT/iFFT operations. For a frequency
domain compensator for two polarisation, including the need for overlap and save
algorithm and assuming that CD DGD, the number of multiplications per bit is;
NFDE log2 NFDE 1n

cos tFDE 10:6
NFDE NCD 1log2 M
where NCD sCD n=T is the number of overlap bits required to ensure correct
frequency domain ltering. The choice of FFT size is an optimisation of NFDE
NCD 1 m where m is the number of new data per FFT. If data throughput is at
least 85 % NCD ,
n 1 1
NFDE [ NCD 10:7
T 2 log2 M 0:85
for 2 polarisations. FFT implementations are usually more efcient in multiples of

radix 24 in an ASIC implementation and implementation cost must be balanced
with the acceptable latency. Tap coefcients are at setup when a channel is rst
established, taking approximately 10NCIR multiplications for a zero-forcing algo-
rithm [49] and are updated infrequently (so that overhead is negligible) over the
lifetime of the system. Dynamic changes in the network will be compensated by the
subsequent dynamic equalizer.
10.2.2.3 Dynamic Equaliser
To compensate for channel parameters that change dynamically, it is necessary to

implement a short dynamic equaliser to compensate for residual CD and fast
changing PMD. It can be assumed that generally a channel is time invariant with a
period of 5 s [50]. As the length of the equaliser is short it is more power efcient
to implement it as a time domain butterfly FIR lter consisting of four individual
lters [41]. The taps must be updated continuously or quasi-continuously to track
the signal polarisation and other fluctuations in the link transfer function. Many
algorithms can be used to update the taps [4042]. The Constant Modulus algo-
rithm [51] has been widely applied and to illustrate the required power consump-
tion, we assume CMA, and limit the length of the individual lters to NCMA = 20
taps [40]. For the time domain equalizer, we require 4 FIR lters in a butterfly
conguration, with the number of multiplications per bit given by the FIR lter
length, plus an additional multiplication per symbol and per tap in order to update
the tap weight, giving a total multiplication per bit of
filter update
z}|{
z}|{
NCMA NCMA
cos tTDE 4 2 10:8
2 log2 M log2 M
10.2.2.4 Phase Estimation
After the mixing of the signal and the local oscillator (LO), it is almost certain that
there will be a frequency and phase offset which results in rotation of the con-
stellation of the incoming signal. There are several possible methods of phase
estimation including maximum a posteriori (MAP), decision directed, power law-
Wiener lter and Viterbi-Viterbi estimation [52, 53]. The VV algorithm are con-
sidered to have the lowest computation complexity. Here we consider the Viterbi-
Viterbi algorithm where an estimate, k, averaged over k symbols of the actual
relative phase between signal and local oscillator is given by
!
1 X k
/k arg xi wixi 4 10:9
4 ik
where w is an optimised weighting function dependent on the relative impact of

additive noise and laser phase noise and xi the received signal sample. As the
modulation format complexity increases, it is possible to only consider Class 1
symbols those that lie at modulation angles of (p4 f0; 1; 2::g p2). The different
symbol levels are detected using amplitude discrimination. However this would
result in only utilising a reducing percentage of the received signals being used as
even some Class 1 signals must be neglected as their modulus is very close to other
symbols. Several methods have been suggested for partitioning higher modulation
formats into sets of QPSK [54] which can then be processed in stages. Another
method [55] partitions by considering that sets of Class 2 (symbols with rotated by
f0; 1; 2::15g p8 4 ) which is only suitable to 16QAM. Another method is to raise
16QAM Class 2 symbol by eight [56] and including outer most 64QAM Class 1
symbols to generate the phase estimate. CPE is an active area of research as
performance is traded off against complexity. In Table 10.1, there is an estimate of
the minimum number of multiplications to implement CPE consisting of dual po-
larisation, number of QPSK decomposition and updates.
Using these approximations, it is possible to generate an estimate of operations
to
cos tVV 2 k 7:2 loge M 3:35 10:10
The complexity scales logarithmically with M up to 64. It is assumed that the

same number of symbols will be used for each modulation format. This may result
in different maximum combined line-width symbol duration product for different
modulation format as a smaller percentage of the symbols are suitable for pro-
cessing with VV at 64QAM. It is likely that, other more computationally intensive
methods will be necessary to track for higher modulation formats in practice. Also
each of these methods will have to implement a method to prevent cyclic slips
either using differential encoding or an intelligent phase unwrap. Also performance
can be improved by use of a maximum likelihood estimation stage.
It is assuming that the arg(*) operation is carried out efciently but with latency
cost using a CORDIC algorithm, lookup table or equivalent. A correctly congured
phase estimator may also have the benecial effect of cancelling some SPM and
XPM effects in WDM transmission environments. Such benecial effects are
neglected here.
Table 10.1 Estimation of CPE

cos tVV QPSK [54] 16QAM [55] 64QAM [56]
Complexity 24 1 1 k 2 3 4 1 1 k 2 20
64 kf4 1 1 8 1 1g
10.2.2.5 Forward Error Correction (FEC)
Whilst research results demonstrate the clear value of soft decision codes such as
low density parity check codes (LDPC), Reed-Solomon (RS) codes are typically
deployed in optical communications either as standalone codes or as concatenated
or turbo codes. The RS decoder consists of:
Syndrome computation block (SC)
Key equation Solver (KES)
Chien search error evaluation block (CSEE)
The implementation of these algorithms has received much effort over the years
but the challenge of implementing them for optical communication speeds is an on-
going concern. It is often a trade-off between area and latency and iterations to
allow optimum performance. For minimum latency the complexity is proportional
to (t = (N K)/2 = 8 for RS(N,K) [57])
CFEC ME 4 logM 50 2t 10:11
Using a systolic modied Euclidean algorithm which has a latency of 80 clock

cycles or for minimum complexity but with a latency of 288 clock cycles
CFEC BM 4 logM 6 2t 10:12
using the inversion-less Berlekamp-Massey algorithm. FEC decoding is imple-

mented at one sample per bit.
10.3 Link Power Consumption
Having established the power consumption of the essential transponders and optical
ampliers, estimation of the exact power consumption of an optical network
requires detailed knowledge of the node locations and link length and properties.
Using a closed form expression for the nonlinear Shannon limit [5861] with a 3 dB
additional operating margin as a network planning tool, we estimate the physical
layer power consumption of an arbitrary network with a total capacity of 1 Pbit/s,
requiring both multiple bres and, for modulation formats with high cardinality,
multiple regenerators. For fully coherent systems, optimum link congurations are
using non dispersion shifted single mode bre (ideally ultra-low loss) without in-
line dispersion compensation, since dispersion may be compensated for without
additional noise penalty in the digital domain. We use the formalism of [58], but the
results are equally valid for all forms.
Figure 10.7 illustrates the energy consumption of the 1 Tbit/s point to point link,
illustrating that the expected monotonic rise with transmission distance is domi-
nated by the addition of regenerator sites. We observe that the energy consumption
Fig. 10.7 Energy consumption of an optical system transporting 1 Pb/s data between two points
using coherently detected PM-QAM formatted signals spaced at the Nyquist rate, 100 km amplier
spacing (bre with 0.2 dB/km loss, 16 ps/nm/km dispersion and 1.4/W/km nonlinearity) and 5 dB
noise gure ampliers with 5 THz bandwidth. Showing QPSK (red), 8QAM (orange), 16QAM
(green), 32QAM (blue) and 64QAM (purple) homogenous trafc for hard decision FEC with an
FEC overhead of 7 % (left) and an FEC overhead of 25 % (right), both assuming a 3 dB signal to
noise ratio margin, as a function of link length at a spacing of 33 GHz (bottom) and symbol rate at
a link length of 1000 km (top). Thin lines in top right hand gure show similar calculations for a
10 Tbit/s transported capacity
also increases dramatically with the cardinality of the modulation format; this is
effect is dominated by the choice of phase estimator. We also observe a variation in
energy consumption as a function of symbol rate, associated with the inefcient
utilisation of inherent energy consumption individual channels such as modulator
drivers and individual lasers. Low symbol rates and low cardinality formats sig-
nicantly increase the number of transponders and hence total energy consumption.
For higher symbol rates, the reduced granularity in networks of low overall capacity
(thin solid lines in Fig. 10.7) may result in inefcient occupancy and slightly
enhanced energy consumption implying an optimum symbol rate in the region of
30 Gbaud. However, the optimum value appears to favour the highest imple-
mentable symbol rate. Overall, we anticipate that the link energy consumption of a
digital coherent link will be less than *0.2 pJ/bit/km for ultra-long links and a few
nJ/b for shorter links. Links employing coherently detected single quadrature
modulation formats (e.g. m-ASK, or m-PSK) will have similar transmission char-
acteristics, but would typically require twice the number of transponders as their
QAM counterparts (m2QAM) signicantly increasing the overall energy
consumption. Two alterative modulation schemes also bear attention from an

overall power consumption point of view. Firstly, transmission with a pilot tone and
associated all optical pilot tone recovery system [6264], where we neglect the
energy consumption associated with the phase recovery block, replacing this with a
xed 5 W energy consumption for the all optical phase recovery system and adding
a 10 % guard band to accommodate the pilot tone.
For terrestrial distances, the optimum energy consumption switches from a low
order format to the highest available modulation format, whilst for transoceanic
distances the difference in overall energy consumption becomes negligible with a
net consumption of *0.1 pJ/b/km for ultra-long haul links. Secondly we consider a
direct detection ASK system spaced at twice the Nyquist rate, signicantly higher
signal to noise ratios, in-line dispersion compensation (requiring dual stage
ampliers with 1 dB higher noise gures) and complex optical receiver lters. It
might be considered that these signicant disadvantages would be offset by the
reduced transponder power consumption since the DSP required for coherent
detection is omitted. The resulting energy consumption is shown in Fig. 10.8. Total
energy consumptions are signicantly greater than 0.5 pJ/b/km (at 1000 km) for all
Fig. 10.8 Energy consumption of an optical system transporting 10 Pbit/s gross data rate over
1000 km as a function of symbol rate (at 1000 km) and distance (at 33 Gbaud) using (left) pilot
assisted PM-QAM with a 10 % guard band, all other parameters as above and (right) directly
detected binary PM-ASK formatted signals spaced at the Nyquist rate with 80 km amplier
spacing (bre with 0.2 dB/km loss, 1.6 ps/nm/km dispersion and 1.4/W/km nonlinearity) and 6 dB
noise gure ampliers with 5 THz bandwidth for homogenous trafc for hard decision FEC with a
25 % overhead. Colours for solid lines represent the same number of constellation points as shown
in Fig. 10.7, dashed lines represent; pink (2-ASK), brown (6 ASK), cyan (10 ASK)
formats, except simple on-off keyed signals at symbol rates above 20 Gbaud, where
polarisation mode dispersion would become a key limitation. Furthermore, the
above calculations have been performed assuming a 65 nm CMOS processing
node, with an energy consumption of 1.5 pJ per multiplication [65]. With 14 nm
CMOS processes currently entering production, enhanced CMOS energy con-
sumption is perfectly possible; although not without difculties as smaller gates
appear to have increased noise properties.
The dominance of digital signal processing functions suggests that energy
reduction through ever more efcient CMOS processing will be of benet. The
level of improvement which may be expected is shown in Fig. 10.9, which
quanties the level of performance gain which may be achieved for a given
improvement in CMOS gate efciency. We have also estimated the required CMOS
node required to achieve three specic efciency enhancements: consistent with the
issues identied in the International Technology Roadmap for Semiconductors [66],
the reduction in total power consumption per gate is beginning to saturate [67].
Note that Fig. 10.9 neglects the increase in ADC noise (decrease in effective
number of bits) associated with small CMOS feature sizes [68]. Nevertheless, the
potential energy saving of the order of 50 % will be welcome, and will result in
similar savings in optical networks transporting high cardinality modulation for-
mats. The benet will be lower for PM-QPSK based networks.
Many signal processing functions, such as wavelength conversion, format con-
version (aggregation) etc. may be readily achieved using a transponder with the
receiver electrical output connected to the electrical input to the transmitter resulting
again in energy consumptions of the order of 1 nJ/b for a full functionality tran-
sponder. This reduces to around 200 pJ/b for a pure transparent wavelength converter
or phase conjugator [6971] without digital signal processing. The performance,
versatility and low energy consumption of the transparent opto-electronic wavelength
converter sets a benchmark for alternative schemes. To match the available perfor-
mance of conventional systems, it is essential that any alternative system is
Fig. 10.9 Impact of DSP processing power on overall link energy consumption for full digital
coherent system (left) and pilot aided coherent system (right) with 7 % FEC overhead. All other
parameters as specied in Fig. 10.7
compatible with high cardinality coherently detected polarisation multiplexed sys-

tems and should offer lower energy consumption than the conventional scheme
(0.21 nJ/b per operation).
10.4 Optical Signal Modulation Using Electro-Optic

Modulators
This section will cover the waveform processing capabilities of electro-optic

devices in the context of optical frequency translation and optical pulse shaping.
The section will identify the performance limitations, optical and electrical power
efciencies as a function of frequency shift or pulse width. We will consider two
types of electro-optic modulator for these applications; the phase modulator and its
direct derivatives of the Mach Zehnder modulator and IQ modulators, and the
electro-absorption modulator. The simplest mean to generate an optical waveform
using an electro-optic modulator is to simply prepare a drive waveform of the
required temporal shape, amplify to a suitable level and apply to the modulator.
This works well up to a large faction of the modulator bandwidth, and is utilised
extensively for basic signal generation such as data encoding. We consider here the
capabilities of such modulators to generate signals with optical bandwidths which
exceed the modulator bandwidth, focussing on pulse generation and optical comb
generation.
10.4.1 Pulse Generation Using Electro-Absorption

Modulators
Electro-absorption modulators may be readily used for pulse generation by

applying a simple sine wave drive [7275]. They generate pulses whose temporal
intensity prole closely approximates a sech2 shape, and the same conguration
may be readily employed for optical gating or demultiplexing. A pulse width as low
as 7 % of the drive frequency may be readily achieved using appropriately designed
devices.
A typical absorption characteristic is shown in Fig. 10.10, and for this example,
the packaged device performance included, typically, an additional 8.4 dB insertion
loss [76]. To rst order, the modulation characteristic may be expressed by con-
sidering an idealised relationship between the absorption (in dB) and the applied
reverse bias (V);
0 V V0

dB = k(V V0 ) , ... V0 < V < V1 10:13
XR = k(V V ) V1 V
1 0
30
25
Absorption (dB)
20
15
10
0
-1 0 1 2 3 4 5 6
Reverse Bias (V)
Fig. 10.10 left Typical absorption characteristic of an electro-absorption modulator (after [76])
showing exponential loss characteristic for reverse bias between 2 and 4 V. Right Predicted pulse
widths for a 10 GHz drive frequency as a function of available drive power for a modulator driven
at 10 GHz with an absorption slope of 11.8 dB/V (solid) and at 20 GHz with an absorption slope of
8 dB/V (grey)
where k represents the gradient of the absorption characteristic in dB/V, V0 is the

maximum reverse bias for zero absorption and V1 is the reverse bias at which
maximum absorption is achieved. XR represents the maximum extinction ratio (dB),
although for high values this may be neglected (V1 ). For a sinusoidal drive
signal, with a peak voltage of V1 below V0, the output of the modulator is given by;
E t E 0 e
2 logk
10 e
Vb Vpp cos2p:t=2V0 10:14
From which is it is easy to show [77] that for Vb > V0 + Vpp/2 the resultant pulse
width is given by;
1
2 1 2 log10 2 2 1:2 2
sFWHM cos 1 10:15
x kVpp x kVpp
where Vpp represents the peak to peak amplitude of the drive signal of angular
frequency . From this equation, we may deduce that the shortest pulses are pro-
duced for the highest drive frequencies, and for the steepest absorption character-
istics. The suitability of sine wave driven EAMs as a compact, stable, low jitter
source of short pulses and polarisation insensitive time gates suitable for use in high
bit-rate OTDM systems has been conrmed in an 80 Gbaud all electro-absorption
modulator system [78] and in system experiments at 160 Gbaud.
For Vb > V0 + Vpp/2, part of the sinusoidal drive falls within the low loss region of
the absorption characteristic (V < V0). This has the effect of flattening the peak of the
generated pulse (or switching window) as shown in Fig. 10.11a. This mode of
operation has several advantages, including improved de-multiplexer jitter tolerance
[79], and for devices able to withstand high forward bias voltages, channel drop [80].
(a) (b)
Fig. 10.11 a Theoretical switching window of electroabsorption modulator as a function of

reverse bias. b Concept of add drop multiplexer with simultaneous channel detection, taken from
two separate experiments [81]
A further advantage of this operating scheme is that the absorbed signal generates a
photocurrent in the device, which may be used as a channel selective receiver. Thus a
device driven with a high RF amplitude and low DC bias may simultaneously clear an
OTDM channel from a data stream and allow the detection of the dropped channel
(see Fig. 10.11b).
10.4.2 Pulse Generation and Optical Sampling Using Mach

Zehnder Modulators
Pulse generation using optical devices with sinusoidal transfer functions [82], and
in particular Mach Zehnder modulators has been considered by several authors
[8387]. The technique relies on the sinusoidal transfer function of the modulator,
when driven by sine wave signals. It is well known that the output of a Mach
Zehnder modulator (Eout) is governed by the following equation [88];

p V V0
Eout Ein cos 10:16
2 VAM
and illustrated in Fig. 10.10 in terms of the power transmission as a function of

normalised applied voltage. This results in a temporal output of the modulator of

p Vb V0 Vpp sinx:t=2
Eout Ein cos 10:17
2 VAM
where we have expressed the minimum on-off switching voltage of the modulator
(often referred to as V) as VAM. In general the modulator bias and (Vb) and drive
amplitude (Vpp) may be set independently, however there are a number of common
alignment strategies as shown in Fig. 10.12. In these strategies the modulator is
biased to such that, one extreme of the sine wave corresponds to maximum or
minimum transmission, quadrature, or to give frequency doubling. In three of these
Fig. 10.12 a Normalised power transfer characteristic of Mach Zehnder modulator along with
b drive voltages and c output waveforms for different bias conditions and a drive voltage of 0.6
VAM corresponding to the following bias conditions: minimum bias (green Vb V0 = Vpp/2),
maximum bias (red Vb V0 = VAM Vpp/2), quadrature bias (orange Vb V0 = VAM/2), carrier
suppressed (purple Vb V0 = 0), and frequency doubled (blue Vb V0 = VAM). Carrier
suppressed and frequency doubled modes result in pulse trains a twice the drive frequency
conditions if the bias is set such that the sine wave extreme or crossing corresponds
with minimum transmission, then the drive amplitude determines the pulse width
and amplitude (left hand panels of Fig. 10.13). For the other conditions, pulses of
acceptable extinction ratio are only generated for specic drive amplitudes (right
hand panels of Fig. 10.13). Achievable pulse widths, as a function of drive power
assuming VAM = 4 V are shown in Fig. 10.14.
The shortest pulse width is obtained for the frequency doubled mode (maximum
transmission corresponding to the crossing point of the sine wave drive), but which
obviously results in a frequency doubling operation. The fundamental frequency
may be achieved either by cascading with a second modulator, driven with
Vpp = VAM (at the expense of increased excess loss, typically around 4 dB) [89], or
by halving the drive frequency (at the expense of doubling the pulse width from 16
to 32 % of the target frequency). The most rectangular pulse is achieved using VAM
drive, and whilst increasing the drive amplitude slightly reduced the extinction
ratio, it results in a more rectangular pulse shape, which may be advantageous for
demultiplexing operations [90]. Note that in addition to sampling and pulse carving
operations described above, Mach Zehnder modulators may also be used to carve
out dark pulses for various applications such as serial TDM multiplexing and add-
drop multiplexing [91, 92].
Fig. 10.13 Variation of Mach Zehnder modulator switching window with RF amplitude for
minimum bias (green VAM/2 < Vpp < VAM), maximum bias (red Vpp = VAM), quadrature bias
(orange Vpp = VAM), carrier suppressed (purple VAM/2 < Vpp < 2 VAM), and frequency doubled
(blue Vpp = 2 VAM)
Fig. 10.14 Variation of pulse width (lled circles), expressed as a fraction of the cycle period of
the sine wave drive and peak transmission loss (open squares) as a function of applied drive power
(for a modulator VAM = 4 V) for minimum bias (green) maximum bias (red), carrier suppressed
(purple) and frequency doubled (blue) modes, and for a modulator driven with a pulsed drive
signal obtained from a frequency multiplier (brown, right). Power and drive voltages assume a
VAM of 4 V
10.4.3 Optical Comb Generation
Optical combs have a wide variety of uses, ranging from various forms of
metrology, frequency translation and carrier generation for multi-carrier transmis-
sion schemes including orthogonal and conventional frequency division multi-
plexing (OFDM and FDM respectively) [9395]. A wide variety of comb
generation schemes are possible, including mode-locked solid state [96], bre [97,
98] and semiconductor ring [99] lasers, semiconductor lasers [100102], non-linear
beat frequency generation [103106], or gain switching [107] some of which are
discussed in Chaps. 2 and 6 of this book. In this section, we will analyze the
frequency manipulation capabilities of sine wave driven modulators, showing that
direct analysis in the frequency domain allows optimized comb generation
performance.
We consider rst a simple sine wave driven phase modulator, where we observe
that the output spectrum may be obtained by considering [108];

p
Eout jEin j cos x:t #in fxi vi sinDx:t ui g 10:18
Vp
X
1
p p
Eout jEin j Jn vi cos xt #in xi nDxt nui 10:19
n1
Vp Vp
where represents the input optical carrier frequency, #in its phase, V the voltage
required to modulate the phase by [109], xi the dc bias applied to phase modulator
section i, vi the drive amplitude of an applied sinusoidal signal of frequency and
phase i. For a continuous light source input the amplitudes of each frequency
component are given by a series of Bessel functions (10.19, Jn represents the nth
order Bessel function). The amplitudes of each component are dependent only on
the relative drive amplitude whilst the phases are dependent on the input optical and
electrical phases, the dc bias voltage and the line number. Two typical intensity
spectra from such a comb is shown in Fig. 10.16a with drive amplitudes of 0.6 and
1.9 V. It is impossible to achieve an identically flat comb spectrum for more than
three comb lines, the characteristic spectrum being two wings with a central section
with strongly varying amplitude in between. A common enhancement to a phase
modulator based comb intended to improve the width and/or flatness of the comb is
to cascade multiple phase modulators in series [110] or parallel respectively
[111115] or combinations of both [116, 117]. For brevity, we consider here an
arbitrary 2 dimensional array of phase modulators, as shown in Fig. 10.15.
A cascade of phase modulators would be analysed by considering a single row
of phase modulators, whilst a single Mach Zehnder or IQ modulator would be
analysed by considering 2 or 4 parallel paths respectively, each comprising a single
stage of phase modulation. The performance of more complex combinations such
as cascaded MZMs may be calculated by an appropriate combination of modula-
tors, with dened relationships between the drive signals. To calculate the response
of an arbitrary conguration of modulators, a number of rows equal to the number
of possible independent paths through the structure are selected, and a number of
columns equal to the maximum number of independent drive voltages which may
be experienced on these paths. The corresponding drive voltages are assigned to
each of the modulator sections. For the pth phase modulator cascade, the overall
response may be emulated by a single phase modulator with a drive signal of
V1,1 V1,2 V1,S

V2,1 V2,2 V2,S
Phase modulator Phase modulator Phase modulator

1:N Coupler
N:1 Coupler
Ein
Eout
VP,1 VP,2 VP,S
Fig. 10.15 A two dimensional array of phase modulator segments in a simple interferometric
structure, illustrating the approach taken for the calculation of various combinations of phase and
amplitude modulators. Cascaded phase modulators would use, for example, the diagonally hashed
modulator segments, dual drive Mach Zehnder two parallel segments ( filled grey). For cascaded
amplitude modulators, four rows of modulator sections are used to represent the four possible
paths through the structure, with two columns allowing the different drive voltages to be applied
with appropriate signs
amplitude Vp, dc bias Xp and phase p, where the parameters are calculated by
summing phasors for each modulator section and are given by;
( )
X
NS
p 2 X
NS XNS

Xp ks;p xs;p Vp2 2 v2s;p 2 vs;p vt;p cos us;p ut;p
s1
Vp s1 ts1
! 10:20
XNS
X NS

Up ArcTan2 vs;p cos us;p ; vs;p sin us;p lp maxs ks;p
s1 s1
Having reduced the 2 dimensional arrays of phase modulators to a column of

individual phase modulators, the outputs are summed to a single virtual modulator,
again using phasor summation, with amplitude V and phase given by
v
u NP ( )
uX X NP

V t
0
lp Jn2 Vp 2 lp lq Jn Vp Jn Vq cos Xp nUp Xq nUq
p1 qp1
!
X
NP
XNP

0
U ArcTan2 lp Jn Vp cos Xp nUp ; lp Jn Vp sin Xp nUp
k1 p1
10:21
such that the overall output of the combination of modulators is given by
1 X1
Eout jEin j V 0 cosxt #in nDxt U0 10:22
2 NP n1
where NP represents the number of modulators in parallel, NS the number in series,

xs,p, vs,p, and s,p the dc voltage, amplitude and phase of the RF signal applied to the
phase modulator in the pth row and sth column of the modulator array, s,p a binary
flag representing the presence (1) or absence (0) of the phase modulator section, and
ArcTan2 the two argument inverse tangent function. ArcTan2 is used to return the
appropriate quadrant for the computed angle. The outputs of ve alternative comb
generators are also shown in Fig. 10.16. The spectra are the optimised outputs,
subject to either a maximum drive amplitude to each electrode of 0.6 V (red), or to
a maximum total drive power of 1.6 W, assuming a V of 4 V for each individual
phase modulator (blue). Analysing these results, we can clearly see that the number
of comb lines with signicant amplitudes increases both with the number of stages,
and the available drive power. On the other hand the uniformity (flatness) of the
comb is enhanced by additional parallel paths, with identically flat spectra available
with 5 lines from a MZM, and 7 for a cascade of 2 MZMs.
Fig. 10.16 Amplitude of optical comb line versus frequency shift (in multiples of applied drive
frequency) for a maximum drive amplitudes per modulator section of 0.6 V (lled discs) or a
maximum total RF power of 1.6 W (open circles) for top left phase modulators, top centre single
balanced Mach Zehnder modulator, top right dual drive Mach Zehnder modulator, bottom left
cascaded phase modulators, bottom centre two cascaded Mach Zehnder modulators, bottom right
an IQ modulator (four phase modulators in parallel). Indicative drive powers are shown assuming a
modulator V of 4V
It is straightforward to show that provided all modulators are driven at the same
frequency and phase, a cascade of phase modulators results in the same optical
comb as a single phase modulator, but with the argument of the Bessel function in
(10.19) replaced with the sum of the drive amplitudes of the modulator cascade (see
10.20). This results in an enhancement to the electrical power efciency in direct
proportion to the number of phase modulators cascaded. However, in practice,
cascading phase modulators results in a trade-off between optical output power and
conversion efciency due to the excess loss of each modulator, as illustrated in
Fig. 10.17. The optimum number of stages may be readily determined to be 10
log10(e)/ILdB where ILdB is the static insertion loss of the modulator in decibels,
implying that a series of modulators represents the most energy efcient solution
provided the modulator insertion loss is less than 3 dB. For higher modulator
insertion losses, the most efcient strategy is to use the smallest number of stages
allowed by the phase modulators power handling limits, usually determined by
bond wires and/or on chip electrical terminations. Use of optical ampliers and
arbitrary cascades of phase modulators is clearly useful, and is optimally imple-
mented in a ring conguration [118].
Different denitions of comb flatness have been proposed by various authors.
However, since it is likely that line amplitudes will be adjusted by programmable
optical lters, the most valid measure is the maximum optical insertion loss
encountered by any given comb line. This particular measure is shown in
Fig. 10.17b, where we may observe that the additional flatness offered by the dual
drive MZM excess loss (over the theoretical optimum of sharing the input power
between comb lines) of less than 5 dB, with drive powers of up to 1.5 W (as noted
previously, the direct power consumption associated with the drive signal should be
Fig. 10.17 a Comb line conversion efciency (minimum power per line divided by applied input
optical and electrical powers) for 3 (red), 5 (purple), and 7 (green) line combs generated by a
cascade of phase modulators, each of 1.5 dB insertion loss. b Modulation loss per line for a
number of different modulator based comb generators showing the congurations offering either
the lowest loss or the lowest total RF drive power as a function of the number of comb lines
generated. Solid line represents an ideal comb generator which shares the available optical power
equally between lines. Large coloured dots represent modulator conguration: (blue MZM
followed by a phase modulator, red phase modulator, brown dual drive modulator). Small red dots
represent cascade of 2 phase modulators with 4 dB additional insertion loss
doubled or tripled to take into account the efciency of electrical power ampliers).
Signicantly lower power requirements are found for single and cascaded phase
modulators with a power consumption of *2/3 W for 21 comb lines (Fig. 10.17).
State of the art devices offer bandwidths of up to 30 GHz, suggesting that a total
frequency shift of 300 GHz (450 GHz/W) could be achieved using such modulator
based comb sources and optical lters. Operation beyond the modulator bandwidth
is also possible, albeit at the expense of increasing the power consumption. As
travelling wave modulators typically have a linearly decreasing frequency response,
doubling the operating frequency would result in double the required drive power.
To avoid using optical lters, an IQ modulator may be biased to give single side
band frequency shifts of f with a drive power of *4 mW and an excess insertion
loss of 6 dB (for a 24 dB extinction ratio of the next highest harmonic). As seen in
Sect. 3.2.1, MZM may also be used for frequency doubling, and such properties
translate directly to comb generation, with balanced and dual drive MZM allowing
frequency doubled operation (by setting DC bias to the carrier suppressed point),
and IQ modulators offer frequency multiplication of 2.n.f where n is an integer. It
was recently shown that 4 [119] and 6 [120] fold frequency multiplication (2 sub
carriers spaced at 6.f) could be readily achieved using commercial IQ modulators
with less than 1 W drive per modulator (6 V V) and that 12 fold frequency
multiplication was theoretically possible with drive voltages of 6.5 V.
Comb generation is also possible using multiple drive frequencies, either slightly
offset to avoid eld addition in modulator cascades [121] or with larger offsets [114,
122127]. Following the same analysis as described above, the output eld of a Mach
Zehnder modulator driven by a number of different RF drive signals is given by;
N N
P
1 P
1 PNd Pd Qd pbi
Eout Ein 1Nd ... cos a p2qi cos qi Dxi /i Jqi Vp
q1 1 qN 1
i1
i1 i1
/i 1 i 1i p2
10:23
Fig. 10.18 Comb line spectra using high and low power electrical drive signals. a Four equally
spaced equal amplitude drive signals of low power, red 34 mW, blue 0.1 W, purple 0.4 W, brown
1.5 W. b Large signal drive combined with its third harmonic (1.6 W), (powers calculated
assuming phase aligned drive signals)
where Ndi represents the number of drive signals, a the dc bias of the modulator and
I, I, and bi are the frequency phase and amplitude of the ith drive signal
respectively. To calculate the amplitude of a particular frequency harmonic , it is
PNd
necessary to sum all terms where i1 qi Dxi Dx which is equivalent to
P d
asserting the condition that q1 Dx1 Dx Ni2 qi Dxi . Typically the innite
summations are restricted to a few times the expected number of optical comb lines
without signicant loss of accuracy. Two strategies are typically used to generate an
optical frequency comb from a multi-tone drive signal; low signal amplitudes to
avoid unwanted intermodulation products (Fig. 10.18) [124] or with optimised large
signal amplitudes to exploit the intermodulation products to generate new
frequencies.
In a similar way to the linear power reduction associated with modulator cas-
cades, additional line generation using a multi-frequency drive typically reduces the
required electric drive power without sacricing optical insertion loss. However
unlike cascaded phase modulators, it does not extend the frequency range of the
generated comb.
10.5 Conclusions
In this chapter, we have considered the implementation of various signal processing

functions in an optical communications network using either electro-optic modu-
lators or a combination of coherent detection and digital signal processing. Digital
coherent transmission systems, using a 90 nm processing node, offer fully regen-
erated signal processing including FEC for an energy consumption of less than 4 nJ/
b for modulation formats up to 64QAM. This is dominated by carrier phase
recovery, and the sacrice of a small level of spectral efciency to enable the use of
pilot tones reduces this to less than 1 nJ/band unsurprisingly increasing FEC
overheads is most effective for longer reach systems. Optimisation of the nonlinear
performance is benecial primarily by reducing the number of transponders/
regenerators in a link, and the contribution of optical ampliers to the overall
energy consumption of these systems is negligible. Due to the dominance of DSP
energy consumption, the benets associated with improved energy consumption of
particular CMOS technology nodes translate almost directly to benets in physical
layer energy consumption. However, this also opens the way for optical techniques,
such as phase recovery based on pilot tones to enable signicant reductions in
energy consumption.
We nd that pulse generation with duty cycles down to *16 (7) % of the
maximum drive frequency are readily achievable using Mach Zehnder (electro-
absorption) modulators, whilst optical frequency comb generation with an excess
modulation loss of less than 6 dB are possible, with total drive powers less than
1 W. Excess insertion loss is typically in the region of 48 dB.
References
1. A.D. Ellis, N.J. Doran, Are few mode bres a practical solution to the capacity crunch? in
ICTON 2013, paper Tu.C2.1 (2013)
2. Optical ber link for police computer. Electronics Weekly, Sept. 24, 3 (1975)
3. I.P. Kaminov, T. Li, A.E. Wilner, Optical Fiber Telecommunications V (Academic Press,
Burlington, 2008)
4. R. Essiambre, G. Kramer, P.J. Winzer, G.J. Foschini, B. Goebel, Capacity limits of optical
ber networks. J. Lightwave Technol. 28(4), 662 (2010)
5. R.S. Tucker, Green optical communications-Part I: energy limitations in transport. J. Sel.
Top. Quantum Electron. 17(2), 245 (2011)
6. Information Gatekeepers Inc., Lucent technologies helps BT increase network capacity and
offer enhanced services. Fiber Opt. Wkly Update 17(39), 1 (1997)
7. P.J. Winzer, Power consumption of SDM systems, in ECOC Workshop on Integration of
Optical Devices for SDM Transmission, ECOC 2013, http://www.ecoc2013.org/docs/peter-
winzer.pdf (2013)
8. A.H. Gnauck, P.J. Winzer, Optical phase-shift-keyed transmission. J. Lightwave Technol.
23, 115 (2005)
9. T. Okoshi, K. Kikuchi, Heterodyne type optical communications. J. Opt. Commun. 2(3), 82
(1981)
10. http://www.emcore.com/ber-optics/telecom-tunable-dwdm/itla-integrable-tunable-laser-
assembly/ (July 2014)
11. B. Collings, Photonics for metro and client 100G and beyondwants in 100G stream!, in
Next Generation Optical Networking 2014, paper day 3, stream 2, 9:40 (2014)
12. http://jp.fujitsu.com/group/foc/en/services/optical-devices/100gln/
13. P. Dong, L. Chen, Y.K. Chen, High-speed low-voltage single-drive push-pull silicon Mach-
Zehnder modulators. Opt. Express 20(6), 61636169 (2012)
14. N. Kikuchi, R. Hirai, Y. Wakayama, High-speed optical 64QAM signal generation using
InP-based semiconductor IQ modulator, in Optical Fiber Communication Conference, M2A-
2 (2014, March)
15. D. Korn, R. Palmer, H. Yu, P.C. Schindler, et al., Silicon-organic hybrid (SOH) IQ
modulator for 16QAM at 112 Gbit/s, in The European Conference on Lasers and Electro-
Optics (p. CK_9_2) (Optical Society of America, 2013, May)
16. Quad-Channel 32 Gbps Linear MZ Modulator Driver, http://www.macom.com/products/
product-detail/MAOM-003408 (July 2014)
17. Inphi IN3212SZ, 32 Gbps Single-Ended Mach-Zehnder Driver. http://www.inphi.com/
products/in3212sz.php (July 2014)
18. Gigoptix 4x32Gb/s MZ Linear Modulator Driver, http://www.gigoptix.com/products/
gx62472/ (July 2014)
19. M. Mller, High-speed electronic circuits for 100 Gb/s transport networks, in OFC, (2010)
20. B. Berglund, J. Johansson, T. Lejon, High efciency power ampliers. Ericsson Rev. 6, 92
(2006)
21. W. Forysiak, D.S. Goven, Progress towards 100 G digital coherent pluggables in InP-based
photonics. J. Lightwave Technol. doi:10.1109/JLT.2014.2329754 (2014). Alternative would
be W. Forysiak, Progress in InP-based photonic components and sub-systems for digital
coherent systems at 100Gbit/s and beyond, in ECOC 2013, Mo.3.C.2, Vol. 32, 16, pp. 2925
2934 (2014)
22. D.J. Malyon, Digital bre transmission using optical homodyne detection. Electron. Lett. 20
(7), 281 (1984)
23. Y. Audzevich, P.M. Watts, A. West, A. Mujumdar, S.W. Moore, A.W. Moore, Power
optimised transceivers for future switched networks. IEEE Trans. Very Large Scale Integr.
VLSI Syst. 22(10), 20812092 (2014).doi:10.1109/TVLSI.2013.2283300
24. M. Horowitz, E. Alon, D. Patil, S. Naffziger, K. Rajesh, K. Bernstein, Scaling power and the
future of CMOS, in Electron Devices Meeting, 2005. IEDM Technical Digest. IEEE
International (2005), pp. 7, 15
25. M. Kuschnerov, T. Bex, P. Kainzmaier, Energy efcient digital signal processing, in OFC
2014, paper Th3E.7 (2014)
26. D. Kudithipudi, E. John, Implementation of low power digital multipliers using 10 transistor
adder blocks. J. Low Power Electron. 1, 1 (2005)
27. C. Laperle, M. OSullivan, Advances in high speeds DACs, ADCs and DSP for optical
coherent transceivers. J. Lightwave Technol. 32(4), 629 (2014)
28. T. Pfau, S. Hoffmann, R. Noe, Hardware-efcient coherent digital receiver concept with
feedforward carrier recovery for M-QAM Constellations. J. Lightwave Technol. 27(8), 989
(2009)
29. H. Dowid, H. Meyr, CORDIC algorithms and architectures, in Digital Signal Processing for
Multimedia Systems (1999), pp. 623655
30. M. OSullivan, C. Laperle, A. Borowiec, K. Farley, A 400G/1T high spectral efciency
technology and some enabling subsystems, in OFC 2012, paper OM2H1 (2012)
31. B.E. Jonsson, A/D-converter performance evolution, in Converter Passion, ADM Design
(2012)
32. C. Laperle, M. OSullivan, Advances in high-speed DACs, ADCs, and DSP for optical
coherent transceivers. J. Lightwave Technol. 32(4), 629643 (2014)
33. M. Nagatani, H. Nosaka, K. Sano, K. Murata, K. Kurishima, M. Ida, A 60-GS/s 6-Bit DAC
in 0.5-im InP HBT technology for optical communications systems, in Compound
Semiconductor Integrated Circuit Symposium (CSICS) (2011)
34. Y. Greshishchev, et al., A 40GS/s 6b ADC in 65 nm CMOS, in ISSCC 2010 (2010),
pp. 390392
35. D. Crivelli, M. Hueda, H. Carrer, J. Zachan, V. Gutnik, M. Del Barco, R. Lopez, G. Hatcher,
J. Finochietto, M. Yeo, A. Chartrand, N. Swenson, P. Voois, O. Agazzi, A 40 nm CMOS
single-chip 50 Gb/s DP-QPSK/BPSK transceiver with electronic dispersion compensation for
coherent optical channels, in ISSCC 2012 (2012), pp. 328330
36. F. Lang, T. Alpert, D. Ferenci, M. Grozing, M. Berroth, A 6 bit 25 GS/s flash interpolating
ADC in 90 nm CMOS technology. in Ph.D. Research in Microelectronics and Electronics
(PRIME), 2011 7th Conference on. IEEE, 2011 (2011), p. 117
37. Fujitsu 56GSa/s 8-bit analogue-to-digital converter Factsheet, http://www.fujitsu.com/
downloads/MICRO/fme/documentation/c45.pdf (July 2014)
38. T. Pfau, R. Peveling, V. Herath, S. Hoffmann, C. Wrdehoff, O. Adamczyk, M. Porrmann, R.
No, Towards real-time implementation of coherent optical communication, in OFC 2009,
paper OThJ4 (2009)
39. D. Hillerkuss, R. Schmogrow, M. Meyer, S. Wolf, M. Jordan, P. Kleinow, N. Lindenmann,
P.C. Schindler, A. Melikyan, Y. Xin S. Ben-Ezra, B. Nebendahl, M. Dreschmann, J. Meyer,
F. Parmigiani, P. Petropoulos, B. Resan, A. Oehler, K. Weingarten, L. Altenhain, T.
Ellermeyer, M. Moeller, M. Huebner, J. Becker, C. Koos, W. Freude, J. Leuthold, Single-
laser 32.5 Tbit/s Nyquist WDM transmission. Opt. Commun. Netw. 4(10), 715723 (2012)
40. E. Ip, J.M. Khan, Digital equalisation of chromatic dispersion and polarisation mode
dispersion. J. Lightwave Technol. 25(8), 2033 (2007)
41. S.J. Savory, Digital coherent receivers: algorithms and subsystems. IEEE J. Sel.
Top. Quantum Electron. 16(5), 1164 (2010)
42. K. Kikuchi, Digital coherent optical communication systems: fundamentals and future
prospects. IEICE Electron. Express 8(20), 1642 (2011)
43. B.E. Jonsson, On CMOS scaling and A/D-converter performance, in 28th IEEE CAS Norchip
Conference 2010, 1516 Nov 2010, Tampere, Finland, Tu8.1.1 (2010), p. 284
44. B. Spinnler, Equalizer design and complexity for digital coherent receivers. IEEE J. Sel.
Top. Qunatum. Electron. 16(5), 1180 (2010)
45. K. Roberts, M. OSullivan, K.T. Wu, H. Sun, A. Awadalla, D.J. Krause, C. Laperle,
Performance of dual-polarization QPSK for optical transport systems. J. Lightwave Technol.
27(16), 35463559 (2009)
46. H. Sun and K.-T. Wu, A novel dispersion and PMD tolerant clock phase detector for
coherent transmission systems, in OFC, Paper OMJ4 (2011)
47. J.C. Geyer, C.R. Fludger, T. Duthel, C. Schulien, B. Schmauss, Efcient frequency domain
chromatic dispersion compensation in a coherent Polmux QPSK-receiver, in Optical Fiber
Communication Conference (Optical Society of America, 2010), p. OWV5
48. M.S. Faruk, K. Kikuchi, Adaptive frequency-domain equalization in digital coherent optical
receivers. Opt. Express 19(13), 1278912798 (2011)
49. R.A. Soriano, F.N. Hauske, N.G. Gonzalez, Z. Zhang, Y. Ye, I.T. Monroy, Chromatic
dispersion estimation in digital coherent receivers. J. Lightwave Technol. 29(11), 1621
(2011)
50. T. Pfau, C. Wrdehoff, R. Peveling, S.K. Ibrahim, S. Hoffmann, O. Adamczyk, S. Bhandare,
M. Porrmann, R. No, A. Koslovsky, Y. Achiam, D. Schlieder, N. Grossard, J. Hauden, H.
Porte, Ultra-fast adaptive digital polarization control in a realtime coherent polarization-
multiplexed QPSK receiver, in OFC2008, paper OTuM3 (2008)
51. D.N. Godard, Self-recovering equalization and carrier tracking in two-dimensional data
communication systems. IEEE Trans. Commun. 28, 11 (1980)
52. M.G. Taylor, Phase estimation methods for optical coherent detection using digital signal
processing. J. Lightwave Technol. 27(7), 901 (2009)
53. Y. Gao, A.P.T. Lau, S. Yan, C. Lu, Low-complexity and phase noise tolerant carrier phase
estimation for dual-polarization 16-QAM systems. Opt. Express 19(22), 2171721729
(2011)
54. S.M. Bilal, G. Bosco, P. Poggiolini, A. Carena, Low-complexity linewidth-tolerant carrier
phase estimation for 64-QAM systems based on constellation transformation, in ECOC
(2013), p. 966
55. I. Fatadin, I. David, S.J. Savory, Laser linewidth tolerance for 16-QAM coherent optical
systems using QPSK partitioning. Photonics Technol. Lett. 22(9), 631633 (2010)
56. S.M. Bilal, A. Carena, C. Fludger, G. Bosco, Dual stage CPE for 64-QAM optical systems
based on a modied QPSK-partitioning algorithm. IEEE Photonics Technol. Lett. 26(3), 267
(2014)
57. H.-C. Chang, C.B. Shung, C.-Y. Lee, A reed-solomon product-code (RS-PC) decoder chip
for DVD applications. IEEE J. Solid-State Circuits 36(2), 229238 (2001)
58. P.P. Mitra, J.B. Stark, Nonlinear limits to the information capacity of optical bre
communications. Nature 411, 10271030 (2001)
59. X. Chen, W. Shieh, Closed-form expressions for nonlinear transmission performance of
densely spaced coherent optical OFDM systems. Opt. Express 18, 19039 (2010)
60. P. Poggiolini, Modeling of non-linear propagation in uncompensated coherent systems, in
OFC 2013, paper OTh3G1 (2013)
61. A.D. Ellis, M.A. Sorokina, S. Sygletos, S.K. Turitsyn, Capacity limits in nonlinear ber
transmission, in Proceedings of Asia Communications and Photonics, Paper AW4F.1 (2013)
62. Y. Koizumi, K. Toyoda, M. Yoshida, M. Nakazawa, 1024 QAM (60 Gbit/s) singe-carrier
coherent optical transmission over 150 km. Opt. Express 20(11), 12508 (2012)
63. S. Adhikari, S. Sygletos, A.D. Ellis, B. Inan, S.L. Jansen, W. Rosenkranz, Enhanced self-
coherent OFDM by the use of injection locked laser, in OFC 2012, paper JW2A.64 (2012)
64. B.J. Puttnam, R.S. Lus, J.M.D. Mendinueta, J. Sakaguchi, W. Klaus, Y. Kamio, M.
Nakamura, N. Wada, Y. Awaji, A. Kanno, T. Kawanishi, T. Miyazaki, Self-homodyne
detection in optical communication systems. Photonics 1(2), 110 (2014)
65. S.J. Savory, Digital signal processing for coherent systems, in OFC 2012, paper OTh3C7
(2012)
66. International Roadmap for Semiconductors, http://www.itrs.net/Links/2013ITRS/Home2013.
htm (June 2014)
67. D.C. Kilper, G. Atkinson, S.K. Korotky, S. Goyal, P. Vetter, D. Suvakovic, O. Blume, Power
trends in communication networks, IEEE J. Sel. Top. Quantum Electron. 17(2), 275 (2011)
68. B.E. Jonsson, A/D converter performance evolution, ADMS Design AB, Doc. No.: T-13:001
(2013), http://converterpassion.wordpress.com/articles/ad-converter-performance-evolution/
69. E.F. Mateo, X. Zhou, G. Li, Electronic phase conjugation for nonlinearity compensation in
ber communication systems, in OFC, Los Angeles, CA, 2011, Paper JWA025 (2011)
70. B.-E. Olsson, C. Larsson, J. Martensson, A. Alping, Experimental demonstration of electro-
optic mid-span spectrum inversion for mitigation of nonlinear ber effects, in ECOC 2012,
paper Th.1.D (2012)
71. R. Asif, G. Shabbir, A. Akram, Hybrid mid link spectral inverter to compensate for linear and
nonlinear distortions for efcient transmission performance in 224 Gbit/s DP-16QAM
system. Opt. Eng. 52(9), 096103 (2013)
72. M. Suzuki, H. Tanaka, K. Utaka, N. Edagawa, Y. Matsushima, Transform limited 14 ps
optical pulse generation with 15 GHz repetition rate by InGaAsP electroabsorption
modulator. Electron. Lett. 28(11), 1007 (1992)
73. D.G. Moodie, A.D. Ellis, A.R. Thurlow, M.J. Harlow, I.F. Lealman, S.D. Perrin, L.J. Rivers,
M.J. Robertson, Multiquantum well electroabsorption modulators for 80 Gbit/s OTDM
systems. Electron. Lett. 31(16), 1370 (1995)
74. K. Wakita, K. Sato, I. Kotaka, M. Yamamoto, M. Asobe, Transform limited 7 ps optical
pulse generation using a sinusoidally driven InGaAsP/InGaAsP stained multiple quantum
well DFB laser/modulator monolithically integrated. IEEE Photonics Technol. Lett. 5(8), 899
(1993)
75. G. Raybon, M.G. Young, U. Koren, B.I. Miller, M. Chien, M. Zirngibl, C. Dragone, N.M.
Froberg, C.A. Burrus, Five channel WDM soliton pulse generation using sinusoidally driven
electroabsorption modulators in 16 1 laser/modulator array. Electron. Lett. 31(14), 1147
(1995)
76. D.G. Moodie, A.D. Ellis, C.W. Ford, Generation of 6.3 ps optical pulses at a 10 GHz
repetition rate using a packaged electroabsorption modulator and dispersion decreasing bre.
Electron. Lett. 30(20), 1700 (1994)
77. D.G. Moodie, Electroabsorption modulators and their impact on future telecommunication
networks, in MSc Thesis, Electronic and Electrical Engineering Department (University
College London, London, UK, 1994)
78. D.D. Marcenac, A.D. Ellis, D.G. Moodie, 80 Gbit/s OTDM using electroabsorption
modulators. Electron. Lett. 34(1), 101103 (1998)
79. L.F. Mollenauer, E. Lichtman, G.T. Harvey, M.J. Neubelt, B.M. Nyman, Demonstration of
error-free soliton transmission over more than 15000 km at 5 Gbit/s, single-channel, and over
more than 11000 km at 10 Gbit/s in two-channel WDM. Electron. Lett. 28(8), 792 (1992)
80. I.D. Phillips, A. Gloag, D.G. Moodie, N.J. Doran, I. Bennion, A.D. Ellis, Drop and insert
multiplexing with simultaneous clock recovery using an electroabsorption modulator.
Phonics Technol. Lett. 10(2), 291293 (1998)
81. A.D. Ellis, All optical networking beyond 10 Gbit/s, in PhD Thesis, School of Engineering
and Applied Science (Aston University, Birmingham, UK, 1997)
82. T.S. Kinsel, R.T. Denton, Terminals for a high speed optical pulse code modulation
communication system: II optical multiplexing and demultiplexing. Proc. IEEE 56(2),
146154 (1968)
83. H.A. Haus, S.T. Kirsch, K. Mathysek, F.J. Leonberger, Picosecond optical sampling. IEEE
J. Quantum Electron. 16(8), 870 (1980)
84. J.J. Veselka, S.K. Korotky, P.V. Mamyshev, A.H. Gnauck, G. Raybon, N.M. Froberg, A
soliton transmitter using a cw laser and an NRZ driven Mach-Zehnder modulator. IEEE
85. P.A. Krug, D.M. Spirit, L.C. Blank, Transmission study of optical pulses at 6.2 Gbit/s
generated using LiNbO3 modulators to externally modulate cw lasers, in Proceedings of 17th
ACOFT (1992)
86. H. Haga, M. Izutsu, T. Sueta, An integrated 1x4 high speed optical switch and its
applications to a time demultiplexer. J. Lightwave Technol. 3(1), 116 (1985)
87. G. Ishikawa, H. Ooi, Y. Akiyama, S. Taniguchi, H. Nishimoto, 80 Gbit/s (2x40 Gbit/s)
transmission experiments over 667 km dispersion shifted bre using Ti:LiNbO3 OTDM
modulator and demultiplexer, in Proceedings of ECOC96, Paper ThC.3.3 (1996)
88. F. Koyama, K. Iga, Frequency chirping in external modulators. J. Lightwave Technol. 6(1),
87 (1988)
89. A.D. Ellis, T. Widdowson, X. Shan, G.E. Wickens, D.M. Spirit, Transmission of a true single
polarisation 40 Gbit/s soliton data signal over 205 km using a stabilised erbium bre ring
laser and 40 GHz electronic timing recovery. Electron. Lett. 29(11), 990 (1993)
90. A.D. Ellis, C.W. Chow, Serial OTDM for 100 Gbit-Ethernet applications. Electron. Lett. 42
(8), 485486 (2006)
91. T. Matsumoto, M. Jinno, K. Kimura, K. Noguchi, Studies on optical digital cross-connect
systems for very-high-speed optical communications networks, in SUPERCOMM/ICC 94,
IEEE International Conference on Serving Humanity Through Communications, vol. 2
(1994), p. 1060
92. A.D. Ellis, T. Widdowson, X. Shan, D.G. Moodie, A 3 Node 40 Gbit/s OTDM network
experiment using electro optic switches. Electron. Lett. 30(16), 1333 (1994)
93. V. TorresCompany, A.M. Weiner. Optical frequency comb technology for ultrabroadband
radiofrequency photonics. Laser Photonics Rev. 8(3), 368 (2014)
94. J. Pfeifle, V. Brasch, M. Lauermann, Y. Yu, D. Wegner, T. Herr, K. Hartinger, P. Schindler,
J. Li, D. Hillerkuss, R. Schmogrow, C. Weimann, R. Holzwarth, W. Freude, J. Leuthold, T.
J. Kippenberg, C. Koos, Coherent terabit communications with microresonator Kerr
frequency combs. Nat. Photonics 8, 375 (2014)
95. N.R. Newbury, Searching for applications with a ne-tooth comb. Nat. Photonics 5(4), 186
(2011)
96. H.R. Telle, G. Steinmeyer, A.E. Dunlop, J. Stenger, D.H. Sutter, U. Keller, Carrier-envelope
offset phase control: A novel concept for absolute optical frequency measurement and
ultrashort pulse generation. Appl. Phys. B 69(4), 327332 (1999)
97. Y. Fukuchi, K. Hirata, H. Ikeoka, Wavelength-tunable and bandwidth-variable ultra-flat
optical frequency comb block generation from a bismuth-based actively mode-locked ber
laser. IEEE Photonics J. 6(1), 19 (2014)
98. A.D. Ellis, R.J. Manning, I.D. Phillips, D. Nesset, 1.6 ps pulse generation at 40 GHz in a
phase locked ring laser incorporating highly nonlinear bre for application to 160 Gbit/s
OTDM networks. Electron. Lett. 35(8), 645 (1999)
99. M. Foster, J. Levy, O. Kuzucu, K. Saha, M. Lipson, A. Gaeta, Silicon-based monolithic
optical frequency comb source. Opt. Express 19, 1423314239 (2011)
100. C.A. Williamson, M.J. Adams, A.D. Ellis, A. Borghesani, Mode locking of semiconductor
laser with curved waveguide and passive mode expander. Appl. Phys. Lett. 82(3), 322 (2003)
101. A. Akrout, A. Shen, R. Brenot, F. Van-Dijk, O. Legouezigou, F. Pommereau, F. Lelarge, A.
Ramdane, D. Guang-Hua, Separate error-free transmission of eight channels at 10 Gb/s using
comb generation in a quantum-dash-based mode-locked laser. Photonics Technol. Lett. 21
(23), 17461748 (2009)
102. S. Gee, F. Quinlan, S. Ozharar, P.J. Delfyett, J.J. Plant, P.W. Judoawlkis, Optical frequency
comb generation from modelocked diode laserstechniques and applications, in Digest of
the LEOS Summer Topical Meetings (2005), pp. 7172
103. D.J. Richardson, R.P. Chamberlin, L. Dong, D.N. Payne, A.D. Ellis, T. Widdowson, D.M.
Spirit, Demonstration of 205 km transmission of 35 GHz, 5 ps pulses generated from a diode
driven, low jitter, beat signal to soliton conversion source. Electron. Lett. 31(6), 470 (1995)
104. J.J. Veselka, S.K. Korotky, A multiwavelength source having precise channel spacing for
WDM systems. Photonics Technol. Lett. 10(7), 958960 (1998)
105. Z. Tong, A. Wiberg, E. Myslivets, B. Kuo, N. Alic, S. Radic, Spectral linewidth preservation
in parametric frequency combs seeded by dual pumps. Opt. Express 20, 1761017619 (2012)
106. R.P. Scott, N.K. Fontaine, J.P. Heritage, B.H. Kolner, S.J.B. Yoo, 3.5-THz wide, 175 mode
optical comb source, in Optical Fiber Communication and the National Fiber Optic
Engineers Conference, 2007. OFC/NFOEC 2007. PaperOWJ3 (2007)
107. R. Zhou, S. Latkowski, J. OCarroll, R. Phelan, L. Barry, P. Anandarajah, 40 nm wavelength
tunable gain-switched optical comb source. Opt. Express 19, B415B420 (2011)
108. M. Kourogi, K.I. Nakagawa, M. Ohtsu, Wide-span optical frequency comb generator for
accurate optical frequency difference measurement. J. Quantum Electron. 29(10), 26932701
(1993)
109. C. Cox, E. Ackerman, H. Roussell, P. Staecker, R. Osgood, T. Izuhara, D. Djukic, N. Jain, N.
Design of a broadband electro-optic modulator with very low V, in Technical Paper,
Photonic Systems (2003)
110. M. Fujiwara, M. Teshima, J. Kani, H. Suzuki, N. Takachio, K. Iwatsuki, Optical carrier
supply module using flattened multicarrier generation based on sinusoidal amplitude and
phase hybrid modulation. IEEE J. Lightwave Technol. 21, 27052714 (2003)
111. J.J. OReilly, P.M. Lane, R. Heidemann, R. Hofstetter, Optical generation of very narrow
linewidth millimetre wave signals. Electron. Lett. 28, 2309 (1992)
112. T. Sakamoto, T. Kawanishi, M. Izutsu, Widely wavelength-tunable ultra-flat frequency comb
generation using conventional dual-drive Mach-Zehnder modulator. Electron. Lett. 43(19),
10391040 (2007)
113. I.L. Gheorma, G.K. Gopalakrishnan, Flat frequency comb generation with an integrated dual-
parallel modulator. Photonics Technol. Lett. 19(13), 1011 (2007)
114. A.K. Mishra, R. Schmogrow, I. Tomkos, D. Hillerkuss, C. Koos, W. Freude, J. Leuthold,
Flexible RF-based comb generator. Photonics Technol. Lett. 25(7), 701704 (2013)
115. W. Li, J. Yao, Investigation of photonically assisted microwave frequency multiplication
based on external modulation. Trans. Microwave Theory Tech. 58(11), 32593268 (2010)
116. T. Healy, F.C. Garcia.Gunning, A.D. Ellis, J. Bull, Multi-wavelength source using low drive-
voltage amplitude modulators for optical communications. Opt. Express 15(6), 29812986
(2007)
117. Q. Chang, J. Gao, Y. Su, Generation of optical comb frequency signal with high spectral
flatness using two cascaded optical modulators, in Optical Fiber Communication &
Optoelectronic Exposition & Conference 2008, paper SaB.4 (2008)
118. K.P. Ho, J.M. Kahn, Optical frequency comb generator using phase modulation in amplied
circulating loop. Photonics Technol. Lett. 5(6), 721725 (1993)
119. C.T. Lin, P.T. Shih, J. Chen, W. Xue, P.C. Peng, S. Chi, Optical millimeter-wave signal
generation using frequency quadrupling technique and no optical ltering. Photonics
Technol. Lett. 20(12), 10271029 (2008)
120. C. ORiordan, S.J. Fabbri, A.D. Ellis, Variable optical frequency comb source using a dual
parallel Mach-Zehnder modulator, in ICTON 2011, Paper Tu.B1.5 (2011)
121. J. Zhang, J. Yu, N. Chi, Z. Dong, X. Li, Y. Shao, J. Yu, L. Tao, Flattened comb generation
using only phase modulators driven by fundamental frequency sinusoidal sources with small
frequency offset. Opt. Lett. 38, 552554 (2013)
122. T. Saitoh, M. Kourogi, M. Ohtsu, Expansion of span-width of an optical frequency comb
using a higher harmonic wave modulation. Photonics Technol. Lett. 8(10), 13791381
(1996)
123. J. Zhang, J. Yu, L. Tao, Y. Fang, Y. Wang, Y. Shao, N. Chi, Generation of coherent and
frequency-lock optical subcarriers by cascading phase modulators driven by sinusoidal
sources. J. Lightwave Technol. 30(24), 39113917 (2012)
124. S. Fabbri, S. Sygletos, A. Ellis, Multi-harmonic optical comb generation, in ECOC 2012,
paper Mo2A2 (2012)
125. Y. Lu, Y. Xing, Y. Dong, Equal-amplitude optical comb generation using multi-frequency
phase modulation in optical bers. Chin. Opt. Lett. 8, 316319 (2010)
126. S. Ozharar, F. Quinlan, I. Ozdur, S. Gee, P.J. Delfyett, Ultraflat optical comb generation by
phase-only modulation of continuous-wave light. Photonics Technol. Lett. 20(1), 3638
(2008)
127. M. Yamamoto, Y. Tanaka, T. Shioda, T. Kurokawa, K. Higuma, Optical frequency comb
generation using dual frequency optical phase modulation, in Integrated Photonics Research
and Applications/Nanophotonics for Information Systems, Technical Digest (Optical Society
of America, 2005), paper ITuF5 (2005)
Chapter 11
Optical Information Capacity Processing
Mariia Sorokina, Andrew Ellis and Sergei K. Turitsyn
Abstract The never-stopping increase in demand for information transmission

capacity has been met with technological advances in telecommunication systems,
such as the implementation of coherent optical systems, advanced multilevel
multidimensional modulation formats, fast signal processing, and research into new
physical media for signal transmission (e.g. a variety of new types of optical bers).
Since the increase in the signal-to-noise ratio makes ber communication channels
essentially nonlinear (due to the Kerr effect for example), the problem of estimating
the Shannon capacity for nonlinear communication channels is not only concep-
tually interesting, but also practically important. Here we discuss various nonlinear
communication channels and review the potential of different optical signal coding,
transmission and processing techniques to improve ber-optic Shannon capacity
and to increase the system reach.
11.1 Introduction: Information Capacity in Optical

Communications
The capacity of global and local communication systems has increased dramatically
over past decades, with recent laboratory experiments approaching the order of
Pbits/s data rates. However, the data demands rise with time at an even higher rate
at the moment calling for new technology advances. The efciency of current
transmission systems are limited by a number of physical effects, with optical noise
and nonlinearity being the major challenges [115]. It is well recognized nowadays
that current ber-optic communication systems and technologies are facing very
specic hurdles due to the nonlinear properties of optical ber channels, problems
not existing in wireless and other linear channels. In optical ber, the dominant
M. Sorokina (&) A. Ellis S.K. Turitsyn

Aston University, Aston Triangle, Birmingham B4 7ET, UK
e-mail: sorokinm@aston.ac.uk

326 M. Sorokina et al.
nonlinearity is induced by the Kerr effect, manifesting itself as an intensity

dependent refractive index. Intensities in an optical ber are enhanced by the
connement of the optical eld in a small core area and by long interaction dis-
tances during propagation. As a result, the output of ber communication channel
with increasing signal power is distorted not only by noise, as in linear channels,
but also by nonlinearity originating from the Kerr effect, specic manifestations
include: self-phase modulation (SPM)nonlinear effect at the same frequency;
cross-phase modulation (XPM)nonlinear interactions between two waves having
different frequencies; and four-wave mixing (FWM)nonlinear interaction
between three or four waves [1620]. This signicantly reduces the efciency of
ber transmission channel, and a variety of mitigation techniques have been pro-
posed and demonstrated [2139]. Nevertheless there remains strong interest in the
development of signal processing techniques that would take into account specic
nonlinear properties of ber-optic channels.
Considering the impact of nonlinearity on a transmitted signal, one should
distinguish between deterministic nonlinear effects, which, in principle, can be fully
compensated, e.g. by digital backward propagation (even if at the moment some of
those techniques might appear impractical) and noise-mixed nonlinear distortion
(including nonlinear signal-to-noise beating) that are non-recoverable. Thus, the
transmission efciency is limited by noise and nonlinearity, but also by practical
conditions and constraints, impacting the ability to compensate deterministic non-
linearity [32], such as availability, complexity, cost and power consumption of
signal processing elements and stochastic effects, such as polarization mode dis-
persion [40]. In this chapter we will discuss and overview both the practical and
fundamental limits imposed by nonlinearity on ber-channel capacity [115]. The
denition of nonlinear limits is not yet well established, and we hope that this
chapter will contribute to the denition of nonlinear capacity limits and a better
understanding of the required signal processing techniques to provide for highest
system capacity.
First of all, it is important to distinguish between use of a daily engineering term
system capacity and the strictly dened mathematical term Shannon channel
capacity [41]. The former used to describe the total available throughput (typically
at some very low bit error rate level) of the communication systems, measured
either in bits/second for the whole bandwidth used, or in bits/Hz/second for spectral
efciency. The Shannon channel capacity denes the maximum error-free trans-
mission rate achievable in communication channels obtained by the maximization
of the mutual information over all possible input signal distributions (modulation
formats and error-correcting codes), and is also measured in bits/second (total
capacity over certain bandwidth) or in bits/Hz/secondShannon capacity per unit
of bandwidth.
In ber-optic channels it was observed and reported in many publications that
the ber nonlinearity limits the spectral efciency at high enough signal power.
Though the exact Shannon capacity of nonlinear ber channels has not yet been
calculated, due to the lack of appropriate analytical channel models, it was shown in
[42] that in case of full nonlinear compensation and signal optimization over the
11 Optical Information Capacity Processing 327
whole used bandwidth (single-user channel model analysis) and neglecting para-
metric noise amplication, the Shannon capacity should grow monotonically with
signal power [4, 42]. However, the calculation of the Shannon capacity in a multi-
user system, for instance, with independent spectral channels in wavelength-divi-
sion multiplexed (WDM) transmission is more difcult. The currently favoured
approach in the optical communication community is to consider such multi-
userchannel system from the perspectives of a selected channel, assuming that the
other customer only impose a noise-like interference on the channel under con-
sideration. This approach originates from the fact that the full compensation of the
deterministic impact of nonlinearity is not always technically feasible, and in
multi-channel networks that use of the full bandwidth for post-processing is not
possible in practice. In this case, one can dene practical nonlinear system capacity
limits that occur due to decay of spectral efciency with signal power ([2, 9] and
references therein). In particular, the conventional multiplicative noise approxi-
mation [2, 9, 10, 43] shows that for uncompensated un-optimized link the spectral
efciency is a convex function of signal power.
Studies based on multiplicative noise typically calculate a so-called nonlinear
signal-to-noise ratio and determine predictions of Shannon capacity or bit error
rate (BER) of a given format, by substitution of the SNR with the nonlinear SNR in
the appropriate linear equation. However, a demonstrable difference between the
achievable rate (no optimization), the constrained capacity (limited optimization),
and the Shannon capacity (optimized over possible signal distributions) exists,
highlighting the importance of analytical calculation of the optimized mutual
information functionalShannon capacity. Note that the channel capacity of most
nonlinear channels is still unknown, but upper and lower bounds can be estimated.
We would also like to note that there are a vast number of various nonlinear ber
channels, for example any specic signal power distribution along the ber link or
any additional in-line elements may change the performance and constitute a new
nonlinear channel. Therefore, the nonlinear Shannon limit remains an open and
acute problem.
Approaching the Shannon channel capacity requires a range of efforts in multiple
research directions: advances in amplication schemes and forward error-correction
codes, development of new signal transmission and processing techniques,
including new methods of nonlinearity mitigation or, indeed, positive use of non-
linearity, progress in optical signal regeneration, as well as, investigation of addi-
tional degrees of freedom for more advanced data multiplexing. The latter includes
combining spectral multiplexing with polarization and spatial multiplexing. Modern
ber-optic communication systems already employ spectral-, time-, and polariza-
tion-division multiplexing to increase data rates through conventional single-mode
ber. As these degrees of freedom become saturated, new directions, such as space-
division multiplexing (SDM) with multicore [44] and multimode [45] bers come
into play. Indeed, it was shown that SDM technology enables several fold capacity
increase [46]. Note, however, that installation of novel ber base assumes an
enormous cost of new infrastructure, and such spatially-demultiplexed systems may

be also affected by the nonlinear crosstalk between the cores/modes and may
require digital signal processing (DSP) with similar complexity as digital nonlinear
compensation.
Overall, signal processing plays a critical role for further progress in this eld.
This includes both digital signal processing and linear/nonlinear analog all-optical
processing [2132]. Using digital signal processing one can pre- or post-compen-
sate deterministic transmission distortions. The immense challenge of a system
capacity crunch, coupled with the availability of new materials and advances in
integrated optics has re-ignited interest in previously studied methods in new
context of coherent optical communications. These include: optical phase conju-
gation [4749], phase-conjugated twin waves [30], optical regeneration [5052].
Recent experiments have shown that employing phase-conjugation one can ef-
ciently mitigate nonlinearity and dispersion [49, 5355], though it reduces the
flexibility of the system due to required symmetry conditions. This was addressed
by proposing different system congurations [56, 57]. In [30] it was proposed to
transmit a pair of symmetric phase-conjugated twin waves with the additional
requirement of dispersion-symmetry along the transmission path, resulting the
canceling of nonlinearity impairments by adding the recovered twin waves at the
receiver. This enabled demonstration of 400 Gbit/s super-channel with a record
distance of 12,800 km. This, however, reduces system capacity by at least a factor
of two, since two polarizations were exploited carrying the same data [22].
Currently, the most widely studied approach to compensate for nonlinearities is
digital back propagation [58], which enables full compensation of the determin-
istic nonlinear and dispersive effects within the compensation bandwidth. Digital
backward propagation requires extensive signal processing and optimization of
computational efforts is under development right now. One of the most important
impacts of nonlinearity is that it mixes otherwise orthogonal modes affecting
multiplexing. In some special designs ber channels (e.g. [59, 60]) that can be
approximated by the lossless nonlinear Schrdinger equation (NLSE) and the so-
called Manakov equations with additive noise, there is an interesting possibility of
signal multiplexing on an alternative basis that is not affected by the Kerr nonlin-
earity. Recently, digital signal processing based on the inverse scattering transform
[6163] demonstrated that it is possible to expand signal on a special basis, within
which the propagation of individual nonlinear modes is effectively linear. The
proof of this is dependent on the integrability of NLSE along similar bases [64].
The change of basisthe inverse scattering transform technique is a nonlinear
analog of the Fourier transform. All these techniques offer efcient nonlinearity
compensation, though differing in computational expends and complexity.
The nonlinearity is an essential factor in the design of advanced ber commu-
nication systems, but it is often shunned by engineers in view of its complexity.
Here we will review different nonlinear optical techniques and their impact on
Shannon capacity.
11.2 Fundamentals
We will start discussion of the ber channel capacity with a brief reminder of the
well known classical results of the information theory [41]. Note that we do not aim
to provide in this section a comprehensive overview of all relevant information
theory, but rather refresh well known facts, referring readers for more details to the
original seminal work [41] or to an excellent recent review paper [8].
11.2.1 Shannon Channel Capacity
Shannon showed that any noisy channel can still be used for reliable communi-
cation at non-zero rate (when the input and output are correlated) and introduced the
expression for the maximum possible error-free transmission rate. Let us consider a
communication channel as a system linking two random variables x and y repre-
senting the channel input and output. A communication channel is characterized by
a probability distribution function Pyjx that denes the probability of receiving
symbol y of the output alphabet given the transmitted symbol x of the input
alphabet . Further random variables and their deterministic outcomes are denoted
by upper Y and lower y letters correspondingly. The channel can be considered as
memoryless, when the output probability is dened by the input at that particular
time and is independent on previous channel inputs and outputs. Shannon formu-
lated the following theorem for the Shannon channel capacity C of the memoryless
discrete noisy channel.
Shannon stated [41] that It is possible to send information at the rate C through
the channel with as small a frequency of errors or equivocation as desired by proper
encoding. This statement is not true for any rate greater than C.
Here information channel capacity C (in bits per degree of freedom) for a me-
moryless channel involves maximizing the mutual information I [41]:
Z
pyjx
C sup IX; Y max dxdypxpyjx log2 R ; 11:1
px px dxpxpyjx
over all valid input probability distributions px subject to the power constraint
R
dx px jxj2 S. The capacity calculation requires knowledge of the statistical
properties of communication channel that are given by the conditional input-output
probability density function (PDF) pyjx. Shannon capacity analysis is not possible
without a full statistical description of the transmission for all possible signal
inputs.
Achievability of the capacity through appropriate coding and signal modulation
is covered by Shannons theorem 12 [41].
Let us denote by nT; q the maximum codebook size of duration T and error
probability q (i.e. the maximum number of signals selected from the alphabet with
probability of error less or equal to q), then the channel capacity is dened as
follows:
log2 nT; q
C lim : 11:2
n!1 T
It means that one can reliably decode about CT bits in time interval T, when T is
sufciently large.
This theorem states that even in the presence of noise one can recover the
original signal from the corrupted channel output. The construction of capacity
achieving codebooks over time interval T is a challenging problem involving
various approaches to modulation, coding and digitizing of an analog signal.
In case of channels with memory, the capacity is dened as [65]
1
C lim sup IX1n ; Y1n ; 11:3
n!1 n
with optimization over all joint distributions of X1 ; ::; Xn subject to the power
R
constraint dx px jxn1 j2 nS.
11.2.2 Numerical Computation of Shannon Capacity
The analytical derivation of the Shannon capacity is possible only for very limited
number of communication channels. However, one can compute Shannon capacity
by numerically maximizing the mutual information over all valid input signal
distributions for a discrete memoryless channel with xed input and output
alphabets. The Shannon capacity has a unique local maximum which is nite for a
nite alphabet. For complex channels, numerical computations is the only practical
way to obtain the Shannon capacity. The following key numerical methods are
typically used for capacity calculations:
Constrained maximization using the Kuhn-Tucker conditions [66, 67]. The
method generalizes the method of Lagrange multipliers for inequality
constraints
The gradient search algorithm of Frank-Wolfe [68], an iterative rst-order
optimization algorithm for constrained convex optimization, also known as the
conditional gradient method, reduced gradient algorithm and the convex com-
bination algorithm. It is based on a linear approximation of the objective
function in each iteration and moves slightly towards a minimum of this linear
function (taken over the same domain).
The Arimoto-Blahut algorithm [69, 70], an iterative method, which maximizes MI
of arbitrary nite input/output alphabet sources. The algorithm was modied for
faster convergence using natural-gradient-based or accelerated Blahut-Arimoto
algorithm [71]. A number of algorithms were proposed for discrete alphabet [72]
or continuous: using computation of a sequence of nite sums [73] or particle
based Blahut-Arimoto algorithm [74]. In the latter, the a particles xk are moved to
increase the relative entropy while keeping the output probability xed.
For discrete-input and discrete-output channels with memory an efcient tech-
nique for calculation of transmission rate was proposed [7577], later general-
ized for continuous channels in [78].
11.3 Linear Additive White Gaussian Noise Channel
11.3.1 Capacity of Linear Additive White Gaussian Noise

Channel
The linear Gaussian noise channel is a basic model of a communication channel that
has a number of applications. In particular, it is used in radio and satellite, wired
and wireless communications. A ber-optic channel can be approximated as a linear
Gaussian channel in the low signal power regime.
In the linear additive white Gaussian noise (AWGN) channel the output y and
input x are random variables (here both are real functions, but generalizations to
complex variables and the multi-dimensional case are straightforward):
Y Xg 11:4
Often, it is convenient to consider a time discrete linear Gaussian channel: yi

xi gi where the channel output and input at the time i are denoted by yi and xi ,
whereas the channel is disturbed by a zero-mean additive white Gaussian noise with
a variance N. The additive noise is independent of the signal. Though, the origin of
the noise may be different, using the central limit theorem it can be shown that the
cumulative effect of a large number of small random effects, each with a well-
dened expected value and well-dened variance, very often can be well approx-
imated by a normal distribution [79].
If the noise variance is zero or there are no constraints on the input signal, the
capacity is innite. However, in practice one has input constraints. In particular,
constraint on the average input power S results in the following:
hjxj2 i S 11:5
The information capacity of power constrained system is given by:
C max IX; Y 11:6

px ;hjxj2 i S
For the linear AWGN power-constrained channel the Shannon capacity can be
found analytically:
S
C log2 1 11:7
N
This is one of the most celebrated formula in the information and communication
theory. It can be generalized further for more complex channel models with various
noise distributions.
11.3.1.1 Constant-Intensity Modulation with Coherent

Detection [8082]
A channel using constant-intensity modulation formats with coherent detection is a

linear AWGN channel:
Y X g; 11:8
with complex x and y and the additional constraint jxj2 S
h p i/ 2 i
p 1 jy Se 0 j
pyjx Sei/0 exp 11:9
phjgj2 i hjgj2 i
where the input phase /0 is used for coding. In [79] Shannon capacity was cal-
culated numerically and the asymptotic expression was found for high SNR:
1 4p
lim C log2 SNR 11:10
SNR!1 2 e
As pointed out in [80] the intuitive explanation lies in the fact that the phase
difference between input and output at high SNR can be approximated as a
Gaussian random variable with variance 2SNR1 , then its entropy is given by
(11.10).
11.3.1.2 Capacity of Bandlimited Channels
The possibility of using discrete alphabets (digitized analog signals) to approach the
capacity of noisy communication channels is based on the observation that in
practice any communication channel has a limited bandwidth. In bandwidth limited
communication a continuous-time and amplitude/phase signal xt can be fully
characterized by a discrete-time and discrete amplitude/phase signal xn . This is
possible due to fundamental mathematical results that have been developed in a
communication context [8387].
Shannon-Nyquist Sampling Theorem
Any function f t bandlimited in frequency to W Hz can be completely determined

by samples f 2W
n
taken at the Nyquist rate of 2W samples per second. Taking the
sum of the samples, one reconstruct the original time-continuous signal using:
X1 n sinp2Wt n
f t f 11:11
n1
2W p2Wt n
It is easy to check that the spectrum of the reconstructed function lies in the band W
and is zero outside. The Shannon-Nyquist theorem [41, 83, 84] denes that a
bandlimited function has 2W degrees of freedom per second.
In the bandlimited Gaussian noise channel each sample can be considered as an
independent, identically distributed Gaussian random variable. If the noise has
power spectral density N0 =2 and occupies the same bandwidth as the signal, the
noise power is N0 W. The expression for Shannon capacity of discrete-time linear
AWGN channel:
S
Cbits per modulation symbol log2 1 11:12
N
Given that there are 2W samples per second and N N0 W, one obtains:
S
Cbits per second 2W log2 1 11:13
N0 W
There is a number of well developed codes which allow the Shannon capacity of
the linear AGWN channel to be approached (see e.g. [88] and references therein).
11.4 Nonlinear Fiber Channel
11.4.1 Basic Models
The communication channel using optical ber cable as a transmission medium is

signicantly different from linear AWGN due to nonlinear dependence of the
refractive index on the signal power [89]. In general, the dynamics of an optical eld
Ez; t in silica ber is governed by the nonlinear Schrdinger equation (NLSE):
@E b @2E a
i 2 2 E icEjEj2 ; 11:14
@z 2 @t 2
Here 2, a, and c denote dispersion, attenuation, and nonlinear coefcients

correspondingly.
This equation governs only key deterministic linear (dispersion and attenuation)
and nonlinear effects. In standard single mode ber the attenuation coefcient is 0.2
dB/km in the 1550-nm wavelength region [8], consequently, optical amplication is
required (in the absence of opto-electronic regeneration) for long transmission
distances. This induces optical noise, that depends on the amplication scheme and
includes amplitude spontaneous emission (ASE). The noise can be incorporated
into the model as an additive Langevin-noise source term g that accounts for the
stochastic perturbations of the signal with the ASE being the main contributor. It
has been shown that ASE can be represented as a circularly symmetric additive
Gaussian noise [90, 91], which is fully dened by an autocorrelation function [8,
89]. Without going into details (such as redenition of the nonlinear coefcient) that
have been discussed already in [8] we dene the principal ber channel model as:
@E b @2E
i 2 2 icEjEj2 0: 11:15
@z 2 @t
The (11.15) apart from being the principal master channel model, can also be
directly derived in two practically important periodic amplication schemes:
(i) periodically spaced discrete point ampliers (e.g. Erbium-doped ber ampli-
ers). In this case, a path-average model (11.15) occurs as a result of the averaging
over periodic gain and loss variation (valid when the amplier spacing is much
smaller compared with the length scale of nonlinear effects) leading to effectively
conservative signal dynamics governed by the re-normalized NLSE (11.15) (the
nonlinear coefcient is modied due to the averaging), see e.g. [92, 93]; and (ii)
quasi-lossless distributed Raman amplication where the Raman gain continuously
compensates for the ber loss and the signal average power remains a constant
along the entire transmission span [7, 59, 60]. The papers [7, 8] provide a detailed
account of how noise intensity is related to the parameters of the line.
The similar vector NLSE model that takes into account propagation polarization
effects is also applied, especially in polarization-based modulation formats. How-
ever, in general, ber polarization properties vary at the scale of several hundred
meters, therefore, full modeling, accounting for these fast rotations of polarizations
with propagation over long distances is time consuming. The model that is used
instead, is an average system of equations (here averaging is over fast changes of
polarization) was derived by Manakov [94] and subsequently veried in [9597]:
@E1 a b @ 2 E1 8c
E1 i 2 i jE1 j2 jE2 j2 E1 ; 11:16
@z 2 2 @t 2 9
@E2 a b @ 2 E2 8c
E2 i 2 i jE1 j2 jE2 j2 E2 11:17
@z 2 2 @t2 9
Note that NLSE and Manakov equations are integrable by inverse scattering
method [92, 98].
11.4.2 Simplied Nonlinear Channel Models
The nonlinear propagation equations (NLSE and Manakov system) presented above
describe complex nonlinear continuous-time communication channels. Corre-
sponding conditional discrete-time probabilities are not known in the general case
for these channels. Instead we consider simplied models with particular applica-
tions in which capacity can be estimated or computed. The continuous-time model
can be transformed into discrete-time model by expanding the continuous-time
signal (complex input Xt and output Yt) over a complete set of orthogonal
functions [99]. This results in discrete-time complex variables Xk and Yk . Further
we omit indexes for memoryless channels, except Sects. 11.4.2.4 and 11.4.2.5.
Input X and output Y can take deterministic values x rin eiuin and y rout eiuout
chosen from the alphabets X and Y correspondingly.
11.4.2.1 Intensity-Modulated Direct-Detection (IMDD)

Channel [100, 101]
The simplest communication systems are intensity-modulated direct-detection

(IMDD), where signal power is modulated and detected. A photo-detector responds
to the square of the incident electric eld, and so the output and input are connected
as follows:
Y jX gj; 11:18
The conditional pdf for this model was found to be the Rice distribution [100]:
!
y y2 jxj2
pyjx exp 11:19
phjgj2 i hjgj2 i
where I0 is zero-order modied Bessel function of the rst kind. It was shown in
[100, 101], that the capacity-achieving distribution has a number of innite mass
points, also lower and upper bounds on capacity were presented, including the
asymptotically tight lower bound [102]:
1
IMDD
Cub log2 1 SNR 0:5 11:20
2
where the square root operation required to recover the output from the detected
photo-current eventually translates to an offset of approximately 1=2 log1=2.
Thus, the Shannon capacity of the IMDD channel is a monotonically increasing
function of SNR and the capacity-achieving distribution has an innite number of
mass points and does not contain an innite set of mass points on any bounded
interval [101].
11.4.2.2 Partially Coherent AWGN [101104]
The discrete-time partially coherent (phase detection is applied) AWGN channel

can be introduced as:
Y XeiH g; 11:21
here the additive noise g is as above a complex Gaussian random variable with zero
mean. Variable H mimics the nonlinear phase noise and can be modelled as real-
valued wrapped AWGN with variance r2U as in [103, 104].
The conditional pdf for such channel can be also found:
Zp h jy xeih j2 i X
1 h h 2pk2 i
dh 1
pyjx 2
exp 2
p exp 11:22
phjgj i hjgj i 2pr2U k1 2r2U
p
If a phase tracking device (such as phase-locked loop) is employed, than H

might be modelled by Thikhonov distribution [101] with positive parameter q:
eq cosh
ph 1 11:23
2pI0 q
The conditional pdf for such channel was derived as:
rout h r 2 r 2 i I m
0
pyjx exp out in
11:24
pN N I0 q
r
r2 r2 rout rin
m 4 out 2 in 4q cosuin uout q2
N N
The capacity-achieving distribution was shown to be circularly symmetric [105]

and (similarly to the result in the previous subsection) it does not contain an innite
set of mass points on any bounded interval [101].
In the case when (x) is not a random variable [106] (describing deterministic
nonlinear phase noise, e.g. due to SPM effect) the conditional pdf is simplied to:
1 h jy xeihx j2 i
pyjx exp 11:25
phjgj2 i hjgj2 i
In [106] it was demonstrated that the Shannon capacity (11.1) with the conditional
pdf given by (11.26) is identical to linear Shannon limitShannon capacity of a
linear AWGN channel log2 1 SNR. This is because is deterministic, so there is
no loss of information. Following optimization procedure on can nd the optimum
input pdf as:
1 h @hx @hx i h jxj2 i

px 1 ic x x exp 11:26
pS @x @x S
If, however, (11.26) is approximated by Gaussian distributing (drop rst bracketed

term), the obtained result (mutual information with non-optimal px) is lower than
the channel capacity.
11.4.2.3 Nonlinear Phase Noise Channel [4, 107111]
In systems predominantly limited by a nonlinear phase noise channel, such as

systems operating with low path average dispersion, the output is given by:

Y X g eiUNL 11:27
Here g is the total AWGN accumulated on the transmission path. The equation
models NLSE (11.15) with zero dispersion b2 0 and periodically inserted ASE
and describes the phase noise caused by the SPM effect. The nonlinear phase noise
is given as [107]:
cL X
K
UNL jx gi j2 11:28
K i1
here summation is performed over the ASE noise accumulated at each of i

1; . . .; K ber segments: gi N1 Ni (Ni denotes ASE noise aroused at i-th
span) and L is the propagation distance. The conditional pdf for such channel (after
proper normalization described in [4, 107, 111]) reads:
prout jrin 1 X 1
py rout eiuout jx rin eiuin Re eimuout uout Km rout ; rin
2p p m1
11:29
where prout jrin is the Ricean pdf of the received power:
2rout r 2 r 2 2r r
out in
prout jrin exp out in
I0 11:30
N N N
and the analytical expressions for the coefcients Cm rout ; rin :

h i
Km r rout bm exp am rout
2
rin
2
Im 2bm rout rin 11:31
p p
imc
am coth imcr2 z
r
p
imc 1
bm p
r sinh imcr2 z
here Im is the modied Bessel function.

The impact of optimization of symbol error probability over amplitude phase-
shift keying constellations was studied in [107110]. The lower bound on channel
capacity was derived in [4]:
1
C log SNR 11:32
2 2
11.4.2.4 Innite-Memory Gaussian Noise Channel Model

[1, 3, 810, 32, 43, 112, 113]
The general channel model NLSE can be rewritten in the case of wavelength-
division multiplexing (WDM) in a form that shows the evolution of the signal in
each WDM channel (the optical eld of k-th channel) with account of SPM and
XPM effects:
!
@Ek b @ 2 Ek X
i 2 2 icEk jEk j2 2jEj j2 gt; z; 11:33
@z 2 @t j6k
here b2 and c denote dispersion and nonlinearity coefcients and ideal compen-
sation of ber losses is assumed, with gt; z being an amplier spontaneous
emission noise, which results from the signal amplication required to compensate
ber losses during propagation.
This presentation allows one to introduce another simplied channel model by
omitting intra-channel effects. This may be justied in the regime when the inter-
channel effects are dominant nonlinear effects or when self-phase modulation is
compensated [114116]. In this case the resulted model equation has the form:
@Ek b @ 2 Ek
i 2 2 icEk Vk gt; z; 11:34
@z 2 @ t
where the effective potential Vk is

X
Vk 2jEj j2 11:35
j6k
In dispersion uncompensated transmission the potential Vk can be modelled

(approximately) as a Gaussian stochastic process short-range correlated in space
and time (assuming high dispersion and moderate to no spread in symbol rates).
This leads to the following simplied channel model with multiplicative Gaussian
noise:
Yi AYi BHij Xj g 11:36
where the output spectrum is denoted as Yi E ~ s xi ; z, whereas the input signal is

Xi E~ s xi ; 0. The inter-channel interference is given by the matrix Hij , which is
assumed to be a zero-mean, Gaussian independent and identically distributed
complex random matrix with variance hHik Hjl i m1 dij dkl , here m is the dimen-
sionality of the signal. Factor B denes the strength of signal-signal interactions and
A is a normalization parameter for conservation of signal power. The conditional
pdf of such model is given in [2]:
1 2
ejyAxj =r
2
pyjx 11:37
2pr2 m
where variance is given by:
r m1 hjgj2 i B2 jxj2 11:38
Various authors have followed a similar generic approach which shares the same
assumptions but include a variety of different physical effects. These produce
slightly different lower bounds on capacity (including impairments from SPM,
XPM, and FWM effects), which can be generalized by the following formula:

S
CLB log2 1 11:39
N lS3 mS2 N
which summarizes the a number of results for a variety of communication channels,

such as OFDM or WDM, with arbitrary dispersion, distributed or lumped ampli-
cation, for each of which the coefcients are found taking into account of different
nonlinearities: l reflects signal-signal interactions, such as FWM and XPM ([1, 3,
810, 43]) and m arises from signal-noise interactions ([32, 113]).
Equation (11.39) is important for systems with low signal to noise ratio or for
systems employing nonlinearity compensation (such that l is small) [32, 117], and,
as originally suggested in [2], signal depletion should be taken into account by
subtracting lS3 from the numerator of (11.39). It has been shown to have limita-
tions for coded transmission [118] and for QPSK with low accumulated dispersion
[108, 119], which may correspond to a dominance of nonlinear phase noise.
11.4.2.5 Finite-Memory Gaussian Noise Channel Model [118]
In the innite-memory Gaussian noise channel model it was implied that the input
is independent and identically distributed (i.i.d.). Under this assumption one can
derive the channel model where nonlinear signal-signal interactions are modelled as
Gaussian noise with variance dependent on average input signal power, so that the
noise affecting the k-th symbol is given by:
v
u 3
u 1 kXM p
gIMGN gGN tN C lim
jXi j2 gGN N CS3 11:40
M!1 2M 1
ikM
where gGN is i.i.d. zero-mean unit-variance circularly symmetric complex Gaussian

random variable, is nonlinear parameter, and M is memory. This simplied
channel model yields simple and experimentally veried expression for the lower
bound on channel capacity (see 11.39).
However, in practice channels have a nite memory dened by physical prop-
erties of the channel, such as dispersion, nonlinearity, etc. The nite memory
Gaussian noise channel can be modelled with its nonlinear noise given by [118]:
v
u 1 3
u kX
M
gFMGN gGN tN C jXi j2 11:41
2M 1 ikM
The model enables the study of non i.i.d. non-stationary inputs, where the channel
capacity denition given by (11.3) can be applied. For the case of uncoded signals
the proposed channel model yields similar results as the innite memory GN
channel, whereas, the case of coded signals new improved lower bounds on
Shannon capacity saturate at high signal powers to a non-zero constant independent
on channel memory M [118]. Though, the model is simplied and does not take
into account a number of effects, such as signal-noise interactions or effects in
multichannel systems, it highlights the importance of channel coding and analytical
modelling for capacity estimations, it clearly demonstrates that it is possible to
transmit information at high signal power via coded transmission.
11.4.2.6 Regenerative Channel [120, 121]
Consider a model transmission system of xed total length that consists of R spans
with linear AWGN channels interleaved with nonlinear regenerators. The ideal
regenerators assign each transmitted symbol to the closest element of the given
alphabet without signal decoding. We assume that the regenerator performs this
transformation independently for each signal quadrature, therefore the problem for
multidimensional signal can be reduced to one-dimensional independent lattices.
We write down formulae for the one-dimensional signal, but they can be gen-
eralized in a straightforward manner to the multi-dimensional case.
The ideal regenerator assigns diffused point x0 (originated from the input point xl )
to the closest neighbor xk in the decision area Sk . Thus, the conditional probability for
the output of regenerator to be a symbol xk given the input signal xl is the probability
that the distorted point x0 falls within the decision boundary Sk . Therefore, the con-
ditional pdf of such a system is dened through the matrix elements [120]:
Z x0 x 2
1
dx0 pG x0 jxl Wkl ; pG x0 jxl p exp
l
Py xk jxl
2pN=R 2N=R
Sk
11:42
It may readily be shown by considering the probability of insertion and omission

errors for each constellation point that the transmission matrix is given by
r
1

R
Wkl erfDkl erfDkl ; Dkl xk xk1 2xl ; 11:43
2 8N
where delta is a statistical parameter describing the spread of the constellation

p
points and is normalized the closest neighbor distance as follows D d R=8N .
Due to the Markovian nature of the stochastic system (the regenerative transfor-
mations are statistically independent), the overall transition matrix after R regen-
erative segments reads as M WR .
If, instead of an ideal transfer function, the regenerator has a smooth transfer
function T, the output of the n-th regenerator can be expressed through the
recurrence relation:
Yn1 TYn gn ; Y1 x: 11:44
Here Ty is a one-dimensional projection of the nonlinear transfer function (see

Fig. 11.1a), which is required to satisfy the following [121, 122]:
Tx x; T 00 x 0; jT 0 xj\1 11:45
The rst expression implies that an undistorted input point (alphabet), x, is unaf-
fected by transformation, the second ensures that the alphabet point is in the center
of the regenerative region. The solution of the rst two expressions identies the
alphabet pointsthe set of points, which are further used for the construction of
modulation format. Finally, the third expression means that the distortion is
effectively suppressed and it determines the flexibility in the transfer function
parameters. When the rst derivative is equal to zero, the alphabet is superstable
and one can observe plateau around the alphabet points (the solution of T 0 x 0 is
the best parameter choice), however plateau is not necessary as long as jT 0 xj\1
(suboptimal parameters choice).
(a) (b)
Fig. 11.1 a Example of a regenerative TF: y x a sinbx and TF for the ideal regenerator;
b the regenerative limitShannon capacity of the channel with ideal regenerators plotted
alongside Shananon capacity of regenerative channel with smooth TF: y x a sinbx plotted
for different number of regenerators R 10 and 20 and different choice of parameters: superstable
T 0 x 0 (corresponds to ab 1) and stable T 0 x 1=2 (ab 1=2), the value of T 0 denes
parameter a, whilst b is subject of optimization of MI in (11.1)
The conditional pdf for the output at the n-th regenerator for each quadrature
yn1 given the input yn is found to be
h y 2i
1 n1 Tyn
pyn1 jyn p exp Nn \jgn j2 [ 11:46
2pNn 2Nn
Because of the Markovian property of the process, the conditional pdf of the
received signal after propagation through R links, yR , given the input, x, is
expressed by a product of single-step conditional probabilities
Z
pyR1 jy1 x dyR ::dy2 pyR1 jyR ; . . .; py2 jy1 x 11:47
Using the conditional pdf given in (11.47), one can calculate Shannon capacity
using (11.1). The Shannon capacity analysis of the system with the ideal regen-
erators denes the upper bound of regeneration efciencymaximum Shannon
capacity that can be achieved by implementation of regeneration, which was cal-
culated in [121] (Fig. 11.1b). Shannon capacity calculations for a regenerative
channel with smooth transfer functions is plotted in Fig. 11.1b. The benet of using
regeneration becomes more pronounced when the number of regenerators is
increased (results are plotted for R 10 and R 20). The gure demonstrates that
the regenerative elements need to be properly optimized using aforementioned

optimization procedure, which denes the optimum parameter values and subop-
timal ones (within the operating margins of the element)compare the Shannon
capacity for two cases T 0 x 0 (TF with plateaus) and T 0 x 0:5. The results
illustrate that by suppression of signal distortions one can achieve Shannon
capacity higher than a system without regenerators operating with the same signal
launch power and with the same power of noise added to the signal during trans-
mission [121].
11.5 Optical Signal Processing to Improve Signal

Transmission
11.5.1 Introduction
Knowledge of the optical phase (in addition to eld intensity) after coherent detection
has dramatically changed the whole concept of compensation and management of
impairments in ber-optic communication. In particular, linear effects such as dis-
persion and polarization-mode dispersion are now routinely compensated entirely in
the coherent receiver. Electronic signal processing using back propagation has also
been applied to the compensation of nonlinear ber impairments. However, the
drawback of the methods of reconstruction of the transmitted data from the received
signal based on nonlinear back propagation is that they rely on computationally-
intensive techniques. Real time nonlinear digital backward propagation is still a
challenging problem especially for SNR gains of more than a DB.
In this section we will review methods, which have potential to either increase
the system capacity or to simplify DSP.
11.5.2 Quasilossless Transmission
As channel capacity is limited by noise, for long haul communication the ampli-
cation scheme is an important part of the system design. Since 1987 the Erbium
doped ber amplier [123] enabled high capacity transmission in a broad region of
wavelengths, which led to the development of WDM systems [123].
Distributed Raman amplication via stimulated Raman scattering (which
enables energy transfer from the pump to the signal) was proposed [124] to create
distributed gain. Moreover, it was experimentally demonstrated [59, 60] that ultr-
along ber Raman lasers enable quasilossless long-distance transmission. As was
indicated in [59, 60] the distributed gain is achieved for a broad spectral bandwidth.
The resulted OSNR improvement (due to low noise) and, corresponding, capacity
improvement was estimated in [7, 8]. Furthermore, the lossless NLSE is a nonlinear
integrable equation, which simplies DSP and offers potential for a number of
techniques, including soliton transmission and employment of inverse scattering
method. Also, quasilossless transmission offers symmetric power variation which is
of particular importance for the technique based on optical phase conjugation.
11.5.3 Optical Phase Conjugation
To compensate nonlinear impairments optically without extensive DBP simulations

a number of methods were proposed. In particular, in 1979 [47] it was shown that
performing optical phase conjugation (OPC) in the middle of the link one can
achieve effective mitigation of deterministic nonlinear and dispersive impairments
with improved system performance over longer transmission distances. Since then a
number of experimental and theoretical works were performed [48, 49]. The OPC
technique sets stringent requirements for the symmetry of the system and near
perfect phase conjugation of the wave near the middle of the link. As a result, a
formal time reversal is achieved in the NLSE and deterministic nonlinear signal-
signal distortions are cancelled during propagation the second half-link. The gen-
eralized condition for effective compensation was given in [125]:
b2 z1 b2 z2
11:48
cz1 Sz1 cz2 Sz2
which means that the ratio of the dispersion and nonlinearity has to be equal at the
corresponding positions z1 and z2 . However, the stringent symmetry requirements
result in complexity of system design and high sensitivity to symmetry violations
(in particular, third order dispersion in highly nonlinear regime). To achieve
symmetric power map Raman amplication scheme was used [126]. Recently, it
was experimentally demonstrated that OPC accompanied with second-order ultra-
long Raman ber laser enables to increase system reach by 30 % by compensating
70 % of the nonlinear distortions. To avoid mid-link OPC, an optical tunable
compensation method was proposed [56], where the nonlinear phase shift (sign-
reversed to that acquired by propagation through the link) is added to the signal
with consequent OPC at before transmission.
Removal of deterministic distortions increases the achievable rate [32, 127, 128]:
!
S
OPC
CLB log2 1 11:49
N 2nOPC
ns
1 mS N
2
where ns and nOPC (assumed to be odd) are the numbers of ampliers and OPCs
correspondingly, assuming OPC compensates entirely deterministic distortions.
11.5.4 Phase-Conjugated Twin Waves
To compensate nonlinear distortions without adding an OPC element, but

exploiting the symmetry available through the use of polarization, the twin wave
model was proposed, where the signal and its conjugated copy are transmitted
simultaneously on either different polarization [30] or different wavelength [129].
Using the symmetric properties of Manakov equation and a symmetric dispersion
map: Dz DL z (this can be achieved by applying pre-dispersion at the
transmitter), the nonlinear perturbations are correlated and, therefore, the original
wave can be restored at the receiver by adding the wave and its conjugated twin.
Compared to DBP, which requires a large number of additional multiplications per
bit [130], PCTW only requires one multiplication and addition per bit. Alterna-
tively, it may be implemented optically using a PSA [131]. Thus, the system
requirements are relaxed compared to other schemes (e.g. OPC or DBP), which
increases the flexibility and reduces cost, also no additional DSP is required.
As a result, it was demonstrated that the nonlinear distortions were reduced by
more than 8:5 dB with the high speed 400 Gbit/s over 12,800 km. The technique
offers greater communication reach when compared to existing schemes. However,
loading two polarizations with the same information affects the system capacity. As
was indicated in [22], as polarization multiplexing is impossible, the information
capacity of the proposed model will be reduced (unless, as can readily be shown
using (11.49) with nOPC 1 and accounting for the doubling of the SNR arising
from the transmission of duplicate information, the capacity of the equivalent dual
polarization system is less than 1.7 dB).
The effective improvement was observed for a wide range of parameters,
however, is limited by crosstalk induced by polarization mode dispersion polari-
zation-dependent loss. As indicated by the authors, the effect is analogous to
nonlinear noise squeezing [30], which demonstrates the benecial interplay of
nonlinearity and dispersion.
The technique offers a simple and energy efcient scheme (without requirement
of additional mid-link devices) for fast ultra long-haul transmission, however it
reduces spectral efciency of the channel by the factor of 2 due to the exploitation
of the spectrum or polarization to carry a conjugated copy of the same signal.
Recently, dual PCTW technique was proposed [132], where two sets of PCTW
were combined a single carrier. Compared to single PCTW with transmitted signal
Ex E and Ey E and recovered Ex Ey , dual PCTW two signals E1 and E2 are
combined at the transmitter as Ex E1 E2 and Ey E1 E2 and the recovered
signals are E1 Ex Ey and E2 Ex Ey . This enables nonlinearity mitigation
(demonstrated Q-factor improvement of 1:2 dB compared to conventional
DP-QPSK performance.
The idea to use phase conjugation to reduce nonlinear impairments was also
implemented in coding [133] and Q-factor improvement up to 1:5 dB was achieved
for QPSK CO-OFDM signals with phase-conjugated subcarrier coding (even

number of OFDM subcarriers is required), when pairs of neighboring subcarriers
were encoded as Sk Ek Ek1 and Sk1 Ek Ek1
with decoding at the

receiver: Rk Ek Ek1 =2 and Rk1 Ek Ek1 =2.
11.5.5 Optical Regeneration
As described above only regenerative channels can reduce the accumulation of

amplied spontaneous emission noise and introduce contractive effects that can
combat ber nonlinear impairments. Regeneration may take many forms, ranging
from Decode-and-Forward model (allowing the use of FEC in each stage), hard
decision (without FEC) [134], transformation with smooth nonlinear transfer
functions (TF), which results in noise squeezing [51] and distributed pulse and
noise shaping [135]. System capacities for Decode-and-Forward model can be
readily calculated from the capacity of a single link channel [136], whilst another
approach is required for hard decision and smooth TF based regenerators, these
were discussed in Sect. 11.4.2.6.
Optical regeneration enables partial removal of the optical noise and, conse-
quently, increases channel capacity. We would like to stress that impact of
regeneration on capacity does not require using the Decode-and-Forward method
that effectively means that the whole link presents a chain of optical transmitter and
receivers. Instead, regenerative effect can be achieved by passive distributed non-
linear ltering with signal regeneration without requiring a hard decision. The
regeneration is based on the constructive use of nonlinear signal transformations
which result in noise squeezing. A number of promising all-optical regenerators
have been reported recently, falling into the class of nonlinear transformations
based regenerators: black-box WDM [137], 4-phase-shift keying (PSK) [117, 138
142], multilevel modulation formats 6-PSK [50], 8-PSK [143], and 16-quadrature
amplitude modulation (QAM) [144], which prove the possibility to remove noise
from both quadratures of the signal. Another important feature of regeneration is
cascadability, as being placed in cascades along the line regeneration enables
accumulation of noise squeezing effect and, as a result, improves a received signal.
Thus, optical regeneration in coherent communication systems has emerged as fast
developing and promising technique (e.g. recently demonstrated cascaded PPLN
based phase sensitive ampliers [145]).
Recently, the impact of regeneration on Shannon capacity has been studied [120,
121] proving that regenerative channels have Shannon capacity higher than a
system without regenerators operating with the same signal launch power and with
the same noise per amplier. Recent advances opened regeneration for coherent
communication, e.g. multilevel phase regeneration [142, 146]. Nevertheless,
implementation of these techniques faces a number of challenges as regenerators
are designed to operate with isolated pulses, so dispersion compensation and
demultiplexing of WDM channels is required. Although a number of works were

proposed for multichannel regeneration (for the case coherent communication see
[147]), it still remains an open problem.
11.6 Conclusion
This chapter gives an overview of optical signal processing in the context of the
Shannon channel capacity. The most important and challenging limit on ber
channel spectral efciency is imposed by ber nonlinearity. Therefore, exploitation
of inherently linear techniques in a nonlinear communication ber channel is not
optimal. The increase in the signal power to improve the signal-to-noise ratio that is
the standard procedure in a linear channel, leads to power-dependent nonlinear
transmission distortions. Though, the Shannon capacity for the nonlinear ber-optic
channel was not yet determined, numerous techniques have been proposed for
increasing achievable rates.
Here we reviewed basic analytical models of communication channels, which
allow to estimate bounds on Shannon capacity. We discussed various techniques to
increase the achievable transmission rate. It was shown that quasilossless trans-
mission enables communication with lower amplication noise, OPC and PCTW
efciently suppress nonlinearity mitigation and might be considered as all-optical
competitor to DBP, whilst all-optical regeneration offers possibility to reduce both
destructive nonlinearity and stochastic noise, however it requires additional efforts
towards practical implementability. There is a clear challenge to develop radically
new approaches to coding, transmission and processing of information in ber
communication channels that will take into account the nonlinear properties of the
transmission medium.
Acknowledgments We are grateful to S. Sygletos and K.S. Turitsyn for useful discussions. This
work has been supported by the EPSRC project UNLOC (Unlocking the capacity of optical
communications) EP/J017582/1.
References
1. A. Splett, C. Kurtzke, K. Petermann, Ultimate transmission capacity of amplied optical ber

communication systems taking into account ber nonlinearities, in Technical Digest of
European Conference on Optical Communication (1993), paper MoC2.4
2. P.P. Mitra, J.B. Stark, Nonlinear limits to the information capacity of optical bre
communications. Nature 411, 10271030 (2001)
3. J. Tang, The Shannon channel capacity of dispersion-free nonlinear optical ber
transmission. J. Lightw. Technol. 19(8), 11041109 (2001)
4. K.S. Turitsyn, S.A. Derevyanko, I.V. Yurkevich, S.K. Turitsyn, Information capacity of
optical ber channels with zero average dispersion. Phys. Rev. Lett. 91, 203901 (2003)
5. B. Wu, E. E. Narimanov, Information Capacity of Nonlinear Fiber-Optical Systems, in

Optical Fiber Communication Conference and Exposition and The National Fiber Optic
Engineers Conference, Technical Digest (CD) (Optical Society of America, 2005), paper
OME58
6. E.B. Desurvire, Capacity demand and technology challenges for lightwave systems in the
next two decades. J. Lightwave Technol. 24, 46974710 (2006)
7. R.-J. Essiambre, G. Foschini, G. Kramer, P. Winzer, Capacity limits of information transport
in ber-optic networks. Phys. Rev. Lett. 101, 163901 (2008)
8. R.J. Essiambre, G. Kramer, P.J. Winzer, G.J. Foschini, B. Goebel, Capacity limits of optical
ber networks. J. Lightwave Technol. 28, 662701 (2010)
9. A.D. Ellis, J. Zhao, D. Cotter, Approaching the non-linear Shannon limit. J. Lightwave
Technol. 28, 423433 (2010)
10. X. Chen, W. Shieh, Closed-form expressions for nonlinear transmission performance of
densely spaced coherent optical OFDM systems. Opt. Express 18, 1903919054 (2010)
11. D.J. Richardson, Filling the pipe. Science 330, 327 (2010)
12. R.I. Killey, C. Behrens, Shannons theory in nonlinear systems. J. Mod. Optics 58, 110
(2011)
13. A. Mecozzi, Limits to the long haul coherent transmission set by the Kerr nonlinearity and
noise of in-line ampliers. J. Lightwave Technol. 12, 19932000 (1994)
14. A. Mecozzi, R.-J. Essiambre, Nonlinear Shannon limit in pseudo-linear coherent systems.
15. K.S. Turitsyn, S.K. Turitsyn, Nonlinear communication channels with capacity above the
linear Shannon limit. Opt. Lett. 37, 36003602 (2012)
16. D.J. Malyon, T. Widdowson, E.G. Bryant, S.F. Carter, J.V. Wright, W.A. Stallard, Electron.
Lett. 27(2), 120121 (1991)
17. H.J. Thiele, R.I. Killey, P. Bayvel, Influence of ber dispersion and bit rate on cross-phase-
modulation-induced distortion in amplied optical ber links. Electron. Lett. 34(21), 2050
2051 (1998)
18. A.D. Ellis, W.A. Stallard, Four wave mixing in ultra long transmission systems incorporating
linear ampliers. IEE Colloquium 159 (1990), http://ieeexplore.ieee.org/xpl/freeabsall.jsp?
arnumber=190875
19. D.A. Cleland, A.D. Ellis, C.H.F. Sturrock, Precise modelling of four wave mixing products
over 400 km of step-index bre. Electronics Letrers 28, 11711172 (1992)
20. R.-J. Essiambre, B. Mikkelsen, G. Raybon, Intra-channel cross-phase modulation and four-
wave mixing in high-speed TDM systems. Electron. Lett. 35(18), 15761578 (1999)
21. X. Liu, D.A. Fishman, A fast and reliable algorithm for electronic pre-equalization of SPM
and chromatic dispersion, in OFC (Optical Society of America, 2006), paper OThD4
22. E. Ip, J. Kahn, Compensation of dispersion and nonlinear impairments using digital
backpropagation. J. Lightwave Technol. 26, 34163425 (2008)
23. M. Nazarathy, J. Khurgin, R. Weidenfeld, Y. Meiman, P. Cho, R. Noe, I. Shpantzer, V.
Karagodsky, Phased-array cancellation of nonlinear FWM in coherent OFDM dispersive
multi-span links. Opt. Express 15, 1577715810 (2008)
24. X. Li, X. Chen, G. Goldfarb, E. Mateo, I. Kim, F. Yaman, G. Li, Electronic post-
compensation of WDM transmission impairments using coherent detection and digital signal
processing. Opt. Express 16(2), 880888 (2008)
25. D.S. Millar, S. Makovejs, C. Behrens, S. Hellerbrand, R.I. Killey, P. Bayvel, S.J. Savory,
Mitigation of ber nonlinearity using a digital coherent receiver. IEEE J. Sel. Topics
Quantum Electron. 16, 12171226 (2010)
26. E. Ip, Nonlinear compensation using backpropagation for polarization-multiplexed
transmission. J. Lightwave Technol. 28(6), 939951 (2010)
27. L.B.Y. Du, A.J. Lowery, Pilot-based XPM nonlinearity compensator for CO-OFDM
systems. Opt. Express 19, B862B869 (2011)
28. E.F. Mateo, X. Zhou, G. Li, Improved digital backward propagation for the compensation of
inter-channel nonlinear effects in polarization-multiplexed WDM systems. Opt. Express 19,
570583 (2011)
29. L.B. Du, M.M. Morshed, A.J. Lowery, Fiber nonlinearity compensation for OFDM super-
channels using optical phase conjugation. Opt. Express 20, 1992119927 (2012)
30. X. Liu, A.R. Chraplyvy, P.J. Winzer, R.W. Tkach, S. Chandrasekhar, Phase-conjugated twin
waves for communication beyond the Kerr nonlinearity limit. Nat. Photonics 7, 560568
(2013)
31. H. Cheng, W. Li, Y. Fan, Z. Zhang, S. Yu, Z. Yang, A novel ber nonlinearity suppression
method in DWDM optical ber transmission systems with an all-optical pre-distortion
module. Opt. Comm. 290, 152157 (2013)
32. D. Raque, A.D. Ellis, Impact of signal-ASE four-wave mixing on the effectiveness of digital
back-propagation in 112 Gb/s PM-QPSK systems. Opt. Express 19, 34493454 (2011)
33. S. Shieh, I. Djordjevic, OFDM for Optical Communications (Academic Press, Amsterdam,
2010)
34. W. Shieh, H. Bao, Y. Tang, Coherent optical OFDM: theory and design. Opt. Express 16,
841859 (2008)
35. A.D. Ellis, J.D. Cox, D. Bird, J. Regnault, J.V. Wright, W.A. Stallard, 5 Gbit/s soliton
propagation over 350 km with large periodic dispersion coefcient perturbations using
erbium doped ber amplier repeaters. Electron. Lett. 27(10), 878 (1991)
36. I. Morita, K. Tanaka, N. Edagawa, M. Suzuki, Impact of the dispersion map on long-haul
40 Gbith single-channel soliton transmission with periodic dispersion compensation, in
Proceedings of OFC99, San Diego, Paper FD1, 1999
37. P.V. Mamyshev, L.F. Mollenauer, Pseudo-phase-matched four-wave mixing in soliton
wavelength-division multiplexing transmission. Opt. Lett. 21(6), 396398 (1996)
38. N.J. Smith, N.J. Doran, Modulational instabilities in bers with periodic dispersion
management. Opt. Lett. 21(8), 570572 (1996)
39. E. Pincemin, A. Tan, A. Bezard, A. Tonello, S. Wabnitz, J.-D. Ania-Castanon, S. Turitsyn,
Robustness of 40 Gb/s ASK modulation formats in the practical system infrastructure. Opt.
Express 14(25), 1204912062 (2006)
40. T. Tanimura, T. Hoshida, S. Oda, T. Tanaka, C. Ohsima, J.C. Rasmussen, Systematic
analysis on multi-segment dual-polarisation nonlinear compensation in 112 Gb/s DP-QPSK
coherent receiver, in Technical Digest of European Conference on Optical Communication
(2009), paper 9.4.5
41. C.E. Shannon, A mathematical theory of communication. Bell Syst. Tech. J. 27(379423),
623656 (1948)
42. E. Agrell, M. Karlsson, WDM channel capacity and its dependence on multichannel
adaptation models, in Optical Fiber Communication Conference and Exposition and the
National Fiber Optic Engineers Conference, OFC/NFOEC (2013)
43. P. Poggiolini, A. Carena, V. Curri, G. Bosco, F. Forghieri, Analytical modeling of non-linear
propagation in uncompensated optical transmission links. IEEE Photon. Technol. Lett. 23,
742744 (2011)
44. J. Sakaguchi, Y. Awaji, N. Wada, A. Kanno, T. Kawanishi, T. Hayashi, T. Taru, T.
Kobayashi, M. Watanabe, Space division multiplexed transmission of 109-Tb/s data signals
using homogeneous seven-core ber. J. Lightwave Technol. 30, 658665 (2012)
45. R. Ryf, S. Randel, A.H. Gnauck, C. Bolle, A. Sierra, S. Mumtaz, M. Esmaeelpour, E.C.
Burrows, R.-J. Essiambre, P.J. Winzer, D.W. Peckham, A.H. McCurdy, R. Lingle, Mode-
division multiplexing Over 96 km of few-mode ber using coherent 6 6 MIMO processing.
46. A. Mecozzi, C. Antonelli, M. Shtaif, Nonlinear propagation in multi-mode bers in the
strong coupling regime. Opt. Express 20, 1167311678 (2012) (York, 2002)
47. A. Yariv, D. Fekete, D.M. Pepper, Compensation for channel dispersion by nonlinear optical
phase conjugation. Opt. Lett. 4(2), 5254 (1979)
48. R.A. Fisher, B.R. Suydam, D. Yevick, Optical phase conjugation for time-domain undoing of
dispersive self-phase-modulation effects. Opt. Lett. 8, 611613 (1983)
49. I. Phillips, M. Tan, M.F. Stephens, M. McCarthy, E. Giacoumidis, S. Sygletos, Exceeding
the nonlinear-shannon limit using raman laser based amplication and optical phase
conjugation, in Optical Fiber Communication Conference, M3C. 1
50. J. Kakande, R. Slavik, F. Parmigiani, A. Bogris, D. Syvridis, L. Gruner-Nielsen, R. Phelan,
P. Petropoulos, D.J. Richardson, Multilevel quantization of optical phase in a novel coherent
parametric mixer architecture. Nat. Photonics 4(12), 748752 (2011)
51. M. Matsumoto, Fiber-based all-optical signal regeneration. IEEE J. Sel. Top. Quantum
Electron. 18(2), 738752 (2012)
52. M. Vasilyev, T.I. Lakoba, Opt. Lett. 30, 1458 (2005)
53. R.A. Fisher, B.R. Suydam, D. Yevick, Optical phase conjugation for time-domain undoing of
dispersive self-phase-modulation effects. Opt. Lett. 8, 611613 (1983)
54. S. Watanabe, T. Chikama, G. Ishikawa, T. Terahara, H. Kuwahara, Compensation of pulse
shape distortion due to chromatic dispersion and Kerr effect by optical phase conjugation.
IEEE Photon. Technol. Lett. 5, 12411243 (1993)
55. X. Chen, X. Liu, S. Chandrasekhar, B. Zhu, and R. W. Tkach, Experimental demonstration
of ber nonlinearity mitigation using digital phase conjugation, in Optical Fiber
Communication Conference (OFC), paper OTh3C.1 (2012)
56. M.D. Pelusi, B.J. Eggleton, Optically tunable compensation of nonlinear signal distortion in
optical ber by end-span optical phase conjugation. Opt. Express 20, 80158023 (2012)
57. D. Raque, A.D. Ellis, Nonlinearity compensation in multi-rate 28 Gbaud WDM systems
employing optical and digital techniques under diverse link congurations. Opt. Express 19,
1691916926 (2011)
58. X. Li, X. Chen, G. Goldfarb, E. Mateo, I. Kim, F. Yaman, G. Li, Electronic post-
compensation of WDM transmission impairments using coherent detection and digital signal
processing. Opt. Express 16(2), 880888 (2008)
59. J.D. Ania-Castan, T.J. Ellingham, R. Ibbotson, X. Chen, L. Zhang, S.K. Turitsyn,
Ultralong Raman ber lasers as virtually lossless optical media. Phys. Rev. Lett. 96, 023902
(2006)
60. S.K. Turitsyn, J.D. Ania-Castan, S.A. Babin, V. Karalekas, P. Harper, D. Churkin, S.
Kablukov, A. El-Taher, E. Podivilov, V.K. Mezentsev, 270-km ultralong Raman ber laser.
Phys. Rev. Lett. 103, 133901 (2009)
61. E.G. Turitsyna, S.K. Turitsyn, Digital signal processing based on inverse scattering
transform. Opt. Lett. 38, 41864188 (2013)
62. J.E. Prilepsky, S.A. Derevyanko, S.K. Turitsyn, Opt. Express 21, 24344 (2013)
63. S.K. Turitsyn, Capacity-achieving techniques in nonlinear channels, in Optical
Communication (ECOC 2014), 39th European Conference and Exhibition on Optical
Communications, paper Mo4.3.5
64. V.E. Zakharov, A.B. Shabat, Sov. Phys. JETP 34, 62 (1972)
65. S. Verdu, T.S. Han, A general formula for channel capacity. IEEE Trans. Inf. Theory 4,
11471157 (1994)
66. W. Karush, Minima of Functions of Several Variables with Inequalities as Side Constraints.
M.Sc. Dissertation. Department of Mathematics, University of Chicago, Chicago, Illinois
(1939)
67. H. W. Kuhn, A.W. Tucker, Nonlinear programming, in Proceedings of 2nd Berkeley
Symposium (University of California Press. Berkeley, 1951), pp. 481492
68. M. Frank, P. Wolfe, An algorithm for quadratic programming. Naval Res. Logistics Q. 3, 95
(1956). doi:10.1002/nav.3800030109
69. S. Arimoto, An algorithm for computing the capacity of arbitrary discrete memoryless
channels. IEEE T. Inform Theory 18, 1420 (1972)
70. R. Blahut, Computation of channel capacity and rate-distortion functions. IEEE T. Inform
Theory 18, 460473 (1972)
71. G. Matz, P. Duhamel, Information geometric formulation and interpretation of accelerated

Blahut-Arimoto-type algorithms, in Proceedings of the 2004 IEEE Information Theory
Workshop, San Antonio, TX, USA, 2429 Oct 2004
72. N. Varnica, X. Ma, A. Kavcic, Capacity of power constrained memoryless AWGN channels
with xed input constellations, in Proceedings of the IEEE Global Telecommunications
Conference (GLOBECOM), Taipei, Taiwan, China, 2002, pp. 13391343
73. C.-I. Chang, L.D. Davisson, On calculating the capacity of an innite-input nite (innite) -
output channel. IEEE Trans. Inf. Theory 34, 10041010 (1988)
74. J. Dauwels, Numerical computation of the capacity of continuous memoryless channels, in
Proceedings of the Symposium on Information Theory in the Benelux, Brussels, Belgium,
2005, pp. 221228
75. D. Arnold, H.-A. Loeliger, On the information rate of binary-input channels with memory, in
Proceedings of the 2001 IEEE International Conference on Communications, Helsinki,
Finland, 1114 June 2001, pp. 26922695
76. V. Sharma, S.K. Singh, Entropy and channel capacity in the regenerative setup with
applications to Markov channels, in Proceedings of the 2001 IEEE International Symposium.
Information Theory, Washington, DC, USA, 2429 June 2001, p. 283
77. H.D. Pster, J.B. Soriaga, and P.H. Siegel, On the achievable information rates of nite-
state ISI channels, in Proceedings of the 2001 IEEE Globecom, San Antonio, TX, 2529 Nov
2001, pp. 29922996
78. J. Dauwels, H.-A. Loeliger, Computation of information rates by particle methods. IEEE
Trans. Inf. Theory 54, 406409 (2008)
79. J. Rice, Mathematical Statistics and Data Analysis, 2nd edn. (Duxbury Press, ISBN 0-534-
20934-3, 1995)
80. Certain topics in telegraph transmission theory, AIEE Trans. 47, 617644 (1928)
81. J.M. Geist, Capacity and cutoff rate for dense M-ary PSK constellations, in Military
Communications Conference, vol. 2 (1990) pp. 768770
82. J.P. Aldis, A.G. Burr, The channel capacity of discrete time phase modulation in AWGN.
IEEE Trans. Inf. Theor. 39(1) (1993)
83. K. Ho, J.M. Kahn, Channel capacity of WDM systems using constant-intensity modulation
formats, in Optical Fiber Communications Conference, ed by A. Sawchuk, vol. 70 of OSA
Trends in Optics and Photonics (Optical Society of America, 2002), paper ThGG85
84. E.T. Whittaker, On the functions which are represented by the expansions of the interpolation
theory. Proc. R. Soc. Edinburgh 35, 181194 (1915)
85. V.A. Kotelnikov, On the carrying capacity of the ether and wire in telecommunications, in
Material for the First All-Union Conference on Questions of Communication, Izd. Red. Upr.
Svyazi RKKA, Moscow, 1933 (Russian)
86. H. Nyquist, Certain factors affecting telegraph speed. Bell Syst. Tech. J. 3, 324352 (1924).
http://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=989875
87. D. Gabor, Theory of communication. J. Inst. Elec. Eng. 93, 429457 (1946)
88. T.M. Cover, J.A. Thomas, Elements of Information Theory (Wiley, New York, 1991)
89. S. Verdu, Fifty years of shannon theory. IEEE Trans. Inf. Theor. (Special Commemorative
Issue) 44(6), 20572078 (1998)
90. G.P. Agrawal, Fiber-Optic Communication Systems, 3rd edn. (Wiley, New York, 2002)
91. J.P. Gordon, W.H. Louisell, L.R. Walker, Quantum fluctuations and noise in parametric
processes II. Phys. Rev. 129, 481485 (1963)
92. J.G. Proakis, Digital Communications (McGraw-Hill, New York, 2001)
93. A. Hasegawa, Y. Kodama, Solitons in Optical Communications (Oxford University Press,
Oxford, 1995)
94. E. Iannone, F. Matera, A. Mecozzi, M. Settembre, Nonlinear Optical Communication
Networks (Wiley, New York, 1998)
95. S.V. Manakov, On the theory of two-dimensional stationary self-focusing of electromagnetic
waves. Sov. Phys. JETP 38, 248253 (1974)
96. D. Marcuse, C.R. Menyuk, P.K.A. Wai, Application of the Manakov-PMD equation to
studies of signal propagation in optical bers with randomly varying birefringence.
97. C. Menyuk, Application of multiple-length-scale methods to the study of optical ber
transmission. J. Eng. Math. 36, 113136 (1999)
98. I.V. Kolokolov, K.S. Turitsyn, Effective equation of nonlinear pulse evolution in a randomly
anisotropic medium. J. Exp. Theor. Phys. 98(2), 348351 (2004)
99. J.P. Gordon, L.R. Walker, W.H. Louisell, Quantum statistics of masers and attenuators. Phys.
Rev. 130, 806812 (1963)
100. V.E. Zakharov, A.B. Shabat, Sov. Phys.-JETP 34, 62 (1972)
101. Keang-Po Ho, Exact evaluation of the capacity for intensity-modulated direct-detection
channels with optical amplier noises. IEEE Photonics Technol. Lett. 174, 858860 (2005)
102. M. Katz, S. Shamai (Shitz), On the capacity-achieving distribution of the discrete-time
noncoherent and partially coherent AWGN channels. IEEE Trans. Inf. Theor. 50(10), 2257
2270 (2004)
103. A. Lapidoth, Capacity bounds via duality: a phase noise example, in Proceedings of the 2nd
Asian-European Workshop on Information Theory, Breisach, Germany, 2629 June 2002,
pp. 5861
104. B. Goebel, R.-J. Essiambre, G. Kramer, P.J. Winzer, N. Hanik, Calculation of mutual
information for partially coherent Gaussian channels with applications to ?ber optics. IEEE
Trans. Inf. Theor. 57(9), 57205736 (2011)
105. K. Ho, Phase-modulated Optical Communication Systems (Springer, New York, 2005)
106. P. Leoni, S. Calabro, B. Lankl, Constellation expansion for 100G transmission. IEEE Photon.
Technol. Lett. 25, 19041907 (2013)
107. E. E. Narimanov, P. Mitra, The channel capacity of a ber optics communication system:
perturbation theory. J. Lightw. Technol. 20, 3 (2002)
108. P. Hou, B. Belzer, T.R. Fischer, Shaping gain of partially coherent additive white Gaussian
noise channel, in Proceedings of the Conference Information Sciences and Systems,
Princeton, NJ, 1517 Mar 2000, pp. wp617wp622
109. P. Hou, B. Belzer, T.R. Fischer, Shaping gain of the partially coherent additive white
Gaussian noise channel. IEEE Commun. Lett. 6, 175177 (2002)
110. A.P.T. Lau, J.M. Kahn, Signal design and detection in presence of nonlinear phase noise.
J. Lightw. Technol. 25(10), 30083016 (2007)
111. C. Hager, A. Graell i Amat, A. Alvarado, E. Agrell, Constellation optimization for coherent
optical channels distorted by nonlinear phase noise, in Global Communications Conference
(GLOBECOM) (2012), pp. 28702875
112. A. Mecozzi, C.B. Clausen, M. Shtaif, S.-G. Park, A.H. Gnauck, Cancellation of timing and
amplitude jitter in symmetric links using highly dispersed pulses. IEEE Photon. Technol.
Lett. 13, 445447 (2001)
113. K. Kikuchi, Electronic post-compensation for nonlinear phase fluctuations in a 1000-km 20-
gbit/s optical quadrature phase-shift keying transmission system using the digital coherent
receiver. Opt. Express 16(2), 889896 (2008)
114. M.I. Youse, F.R. Kschischang, On the per-sample capacity of nondispersive optical bers.
IEEE Trans. Inf. Theor. 57(11), 75227541
115. J.B. Stark, P. Mitra, A. Sengupta, Information capacity of nonlinear wavelength division
multiplexing ber optic transmission line. Opt. Fiber Technol. 7, 275288 (2001)
116. X. Liu, X. Wei, R.E. Slusher, C.J. McKinstrie, Improving transmission performance in
differential phase-shift-keyed systems by use of lumped nonlinear phase-shift compensation.
Opt. Lett. 27, 16161618 (2002)
117. M. Nakazawa, E. Yamada, H. Kubota, K. Suzuki, 10 Gbit/s soliton data transmission over
one million kilometres. Electron. Lett. 27(14), 12701272 (1991)
118. L. F. Mollenauer, A. Grant, X. Liu, X. Wei, C. Xie, I. Kang, C. Doerr, Demonstration of 109
X 10G dense WDM over more than 18,000 km using novel, periodic-group-delay
complemented dispersion compensation and dispersion managed solitons, in Proceedings of
ECOC 03, Rimini, Post-deadline Paper Th4.3.4, 2003
119. E. Agrell, A. Alvarado, G. Durisi, M. Karlsson, Capacity of a nonlinear optical channel with
nite memory. J. Lightwave Technol. 16, 28622876 (2014)
120. R. Dar, M. Shtaif, M. Feder, New bounds on the capacity of the nonlinear ber-optic
channel. Opt. Lett. 39, 398401 (2014)
121. M. Shtaif, R. Dar, A. Mecozzi, M. Feder, Nonlinear interference noise in WDM systems and
approaches for its cancelation, in Optical Communication (ECOC 2014), 39th European
Conference and Exhibition on Optical Communications, paper We1.3.1
122. K.S. Turitsyn, S.K. Turitsyn, Nonlinear communication channels with capacity above the
linear Shannon limit. Opt. Lett. 37(17), 36003602 (2012)
123. R.J. Mears, L. Reekie, I.M. Jauncey, D.N. Payne, Low-noise Erbium-doped bre amplier at
1.54 m. Electron. Lett. 23, 10261102 (1987)
124. R.H. Stolen, E.P. Ippen, Raman gain in glass optical waveguides. Appl. Phys. Lett. 22, 276
(1973)
125. M.A. Sorokina, S.K. Turitsyn, Regeneration limit of classical Shannon capacity. Nat.
Commun. 5, Article number: 3861 (2014)
126. M.A. Sorokina, S. Sygletos, S.K. Turitsyn, Optimization of cascaded regenerative links
based on phase sensitive ampliers. Opt. Lett. 38, 43784381 (2013)
127. S. Watanabe, M. Shirasaki, Exact compensation for both chromatic dispersion and Kerr effect
in a transmission ber using optical phase conjugation. J. Lightwave Technol. 14(3), 243
248 (1996)
128. I. Brener, B. Mikkelsen, K. Rottwitt, W. Burkett, G. Raybon, J. B. Stark, K. Parameswaran,
M.H. Chou, M.M. Fejer, E.E. Chaban, R. Harel, D.L. Philen, S. Kosinski, Cancellation of all
Kerr nonlinearities in long ber spans using a LiNbo3 phase conjugator and Raman
amplication, in Optical Fiber Communication Conference, OSA Technical Digest Series
(Optical Society of America, 2000), paper PD33
129. D. Raque, A.D. Ellis, Nonlinearity compensation via spectral inversion and digital back-
propagation: a practical approach, in Optical Fiber Communication Conference, OSA
Technical Digest (Optical Society of America, 2012), paper OM3A.1
130. S.L. Jansen, D. van den Borne, B. Spinnler, S. Calabro, H. Suche, P.M. Krummrich, W.
Sohler, G.-D. Khoe, H. de Waardt, Optical phase conjugation for ultra long-haul phase-shift-
keyed transmission. J. Lightwave Technol. 24, 54 (2006)
131. Yue Tian, Yue-Kai Huang, Shaoliang Zhang, Paul R. Prucnal, Ting Wang, Demonstration of
digital phase-sensitive boosting to extend signal reach for long-haul WDM systems using
optical phase-conjugated copy. Opt. Express 21, 50995106 (2013)
132. B. Corcoran, S. Olsson, C. Lundstrm, Mitigation of Nonlinear Impairments on QPSK Data
in Phase-Sensitive Amplied Links, in 39th European Conference on Optical
Communication and Exhibition, ECOC 2013, London, 2226 Sept 2013
133. X. Liu, S. Chandrasekhar, P.J. Winzer, A.R. Chraplyvy, R.W. Tkach, B. Zhu, T.F. Taunay,
M. Fishteyn, D.J. DiGiovanni, Scrambled coherent superposition for enhanced optical ber
communication in the nonlinear transmission regime. Opt. Express 20, 1908819095 (2012)
134. T. Yoshida, T. Sugihara, K. Ishida, T. Mizuochi, Spectrally-efcient dual phase-conjugate
twin waves with orthogonally multiplexed quadrature pulse-shaped signals, in Optical Fiber
Communication Conference, OSA Technical Digest (online) (Optical Society of America,
2014), paper M3C.6
135. S.T. Le, E. Giacoumidis, N. Doran, A.D. Ellis, S.K. Turitsyn, Phase-conjugated subcarrier
coding for bre nonlinearity mitigation in CO-OFDM transmission, in Optical
Communication (ECOC 2014), 39th European Conference and Exhibition on Optical
Communications, paper M3C.6
136. T. Widdowson, J.P. Hueting, A.D. Ellis, D.J. Malyon, P.J. Watkinson, Global bre
transmission using optically amplied regenerators for maximised repeater spacing. Electron.
Lett. 30, 20562057 (1994)
137. O. Leclerc, U.E. Desurvire, O. Audouin, Synchronous WDM soliton regeneration: toward
80160 Gbit/s transoceanic systems. Opt. Fiber Technol. 3(2), 97116 (1997)
138. A.D. Ellis, J. Zhao, Channel capacity of non-linear transmission systems, in Impact of
Nonlinearities on Fiber Optic Communications, ed. by S. Kumar. (New York, Springer,
2011), ch. 13, pp. 507538
139. S. Sygletos, P. Frascella, S.K. Ibrahim, L. Gruner-Nielsen, R. Phelan, J. OGorman, A.D.
Ellis, A practical phase sensitive amplication scheme for two channel phase regeneration.
Opt. Express 19(26), B938B945 (2011)
140. R.P. Webb, J.M. Dailey, R.J. Manning, A.D. Ellis, Phase discrimination and simultaneous
frequency conversion of the orthogonal components of an optical signal by four-wave mixing
in an SOA. Opt. Express 19, 2001520022 (2011)
141. S. Watanabe, S. Kaneko, T. Chikama, Long-haul ber transmission using optical phase
conjugation. Opt. Fiber Technol. 2(2), 169178 (1996)
142. G. Goldfarb, M.G. Taylor, G. Li, Experimental demonstration of ber impairment
compensation using the split-step innite impulse response method. Digest of the IEEE/
LEOS Summer Topical Meetings 2008, 145146 (2008)
143. D. Raque, J. Zhao, A.D. Ellis, Digital back-propagation for spectrally efcient WDM 112
Gbit/s PM m-ary QAM transmission. Opt. Express 19, 52195224 (2011)
144. R. Slavik, F. Parmigiani, J. Kakande, C. Lundstrom, M. Sjodin, P. Andrekson, R.
Weerasuriya, S. Sygletos, A.D. Ellis, L. Gruner-Nielsen, D. Jakobsen, S. Herstrom, R.
Phelan, J. OGorman, A. Bogris, D. Syvridis, S. Dasgupta, P. Petropoulos, D. Richardson,
All-optical phase and amplitude regenerator for next generation telecommunications systems.
Nat. Photonics 4(10), 690695 (2010)
145. G. Hesketh, P. Horak, Reducing bit-error rate with optical phase regeneration in multilevel
modulation formats. Opt. Lett. 38, 53575360 (2013)
146. A. Bogris, D. Syvridis, All-optical signal processing for 16-QAM using four-level optical
phase quantizers based on phase sensitive ampliers, in Proceedings of European
Conference in Optical Communications (ECOC 2013), London, paper We.3A.6, (2013)
147. T. Umeki, M. Asobe, H. Takara, Y. Miyamoto, H. Takenouchi, Multi-span transmission
using phase and amplitude regeneration in PPLN-based PSA. Opt. Express 21, 1817018177
(2013)
Chapter 12
Nonlinear Optics for Photonic Quantum
Networks
Alex S. Clark, Lukas G. Helt, Matthew J. Collins, Chunle Xiong,

Kartik Srinivasan, Benjamin J. Eggleton and Michael J. Steel
Abstract By harnessing quantum mechanical effects we can create technology

with greatly improved functionality including enhanced sensing, exponentially
faster computing, the simulation of previously inaccessible quantum systems, and
the secure transfer of information. In our ever more connected world, it is important
that we can communicate securely, and as technology develops into the quantum
regime it will become important for distant quantum processors to exchange
information over a quantum network. Here we discuss how quantum information
can be encoded and transferred around a such a network. We concentrate on the use
of nonlinear optics to generate and manipulate photons, and present schemes for
fully secure quantum communication.
A.S. Clark (&) M.J. Collins C. Xiong B.J. Eggleton

Centre for Ultrahigh bandwidth Devices for Optical Systems (CUDOS),
Institute of Photonics and Optical Science (IPOS), School of Physics,
University of Sydney, Camperdown, NSW 2006, Australia
e-mail: a.clark@physics.usyd.edu.au
M.J. Collins
e-mail: mcoll@physics.usyd.edu.au
C. Xiong
e-mail: chunle@physics.usyd.edu.au
B.J. Eggleton
e-mail: egg@physics.usyd.edu.au
L.G. Helt M.J. Steel
MQ Photonics Research Centre and CUDOS, Department of Physics and Astronomy,
Macquarie University, North Ryde, NSW 2109, Australia
e-mail: luke.helt@mq.edu.au
M.J. Steel
e-mail: michael.steel@mq.edu.au
K. Srinivasan
Center for Nanoscale Science and Technology, National Institute of Standards
and Technology (NIST), Gaithersburg, MD 20899, USA
e-mail: kartik.srinivasan@nist.gov

356 A.S. Clark et al.
12.1 Introduction: Photonic Quantum Networks
Photonics is an enabling technology for global communication systems and is of

profound importance in the evolution of society as we know it. In classical com-
munication networks many optical pulses of light, containing billions of photons,
are sent through optical bres and routed to different locations using a variety of
technologies. In classical networks these optical pulses can be amplied and fre-
quency converted with relative ease, allowing extremely high transmission rates for
information. A quantum network is intrinsically different, in that data is encoded in
an individual quantum system such as a single photon, which obeys inherently
different physics. It is in fact quantum physics that guarantees fully secure com-
munication schemes when using these quantum states. A quantum network is
therefore useful for secure communication and to move quantum information
between quantum information processing nodes. Unfortunately these quantum
states cannot be easily shifted in frequency due to the addition of noise to the
original and converted channels, and unknown arbitrary quantum states can de-
nitely not be amplied due to the famous no-cloning theorem developed by
Wootters and Zurek [1], and separately Dieks [2], in 1982. This has limited current
demonstrations of quantum communication to around 200 km. There are a number
of methods currently suggested to extend this distance, most of which require small
scale quantum information processing (QIP). These include quantum repeaters,
which store quantum information until it is ready to be used, and quantum relays,
which allow for the teleportation of quantum information.
Nonlinear optical processes and devices are used in many different forms in
quantum information experiments and technology. One of the most intuitive is the
use of a nonlinear material to allow a single photon to impart a p phase shift on
another single photon. This, however, has not been realised to date due to the
limited strength of nonlinear media meaning that the device would have to extend
for thousands of kilometres. There have, however, been a number of proposals that
show that quantum information processing schemes could be possible with weak
nonlinearities [3]. Such schemes rely on a single photon imparting a measurable
cross-phase shift on a high intensity beam of light. Experiments with optical bres
[4] and gas-lled hollow core photonic bandgap bres [5] have shown promising
routes to achieving such phase shifts. In parallel to these developments there have
been arguments against such nonlinear quantum computing schemes which state
that if a photon imparts an appreciable phase shift then it must also undergo self-
phase modulation which can destroy the encoded quantum state and cause
unwanted noise in the system [6]. The debate over whether such schemes can work
continues. Another more recently developed QIP scheme employing nonlinear
processes is that of coherent photon conversion (CPC) [7]. In this scheme a third-
order nonlinear medium is pumped with a high intensity laser in a particular mode
which creates an effective second-order nonlinearity which acts on a further three
modes. This can be thought of as a four-wave mixing interaction that creates a
tunable second-order nonlinearity. Theoretically, this process enables the creation
12 Nonlinear Optics for Photonic Quantum Networks 357
of numerous QIP components including a controlled phase gate, photon-doubling

for efcient detection of photons, and clearing higher-order noise terms from
photon sources; however current demonstrations have been limited to very weak
effective nonlinearities. These nonlinear QIP schemes are in their infancy and will
not be discussed in this chapter. Instead we focus on how nonlinear processes are
already in use in many QIP experiments across the world and will be integrated into
future quantum networks to provide enhanced functionality and agility. In the
following sections we will present advances in the use of nonlinear optics for the
generation of photons, rst looking at the theory behind such sources before turning
to experimental implementations with both second- and third-order nonlinear
media. We will discuss how to characterise these photon sources using simple
measurements which provide us with measures of spectral purity, noise metrics, and
photon statistics. We then describe how many of these spontaneous, probabilistic
sources can be actively multiplexed together to create a deterministic source of
photons. An integral part of a classical communications network is wavelength
division multiplexing and the ability to convert between different channels. In a
quantum network this wavelength conversion cannot be performed using stimulated
four-wave mixing. We describe methods for performing quantum frequency con-
version and review the state-of-the-art in this eld. Finally, we describe how a
quantum communications network can be built and how nonlinear optical processes
are integrated into such networks to enhance functionality and improve perfor-
mance. We use the remainder of this section as a short introduction to what a
quantum bit, or qubit, is, how such qubits can be manipulated and how they can be
made to interact to form two qubit operations.
12.1.1 Photonic Qubits, Gates and Algorithms
12.1.1.1 Qubits
Classical computation uses bits with two possible values, 0 and 1, to encode and
process information. This is commonly the off or on state of a switch or transistor.
As the physical size of this switch decreases quantum mechanical effects will come
into play. At this point we are left with one quantum mechanical object, in this case
a single electron, causing the switch to close, however it becomes difcult to
distinguish between the on and off states or make measurements on such objects.
This need not necessarily be a hindrance to computation. Such quantum mechanical
systems may in fact allow computational speed-up for certain operations.
In quantum computation the bit is replaced with what is termed a qubit. This is a
two level quantum mechanical system which may take the value j0i or j1i, where
we use the Dirac notation to show that these are states of a quantum system. This
qubit may also be placed in any superposition of j0i and j1i such that the most
general state jWi is
Fig. 12.1 A schematic of the

Bloch sphere mapping all
possible qubit states
h h
jWi cos j0i ei/ sin j1i ; 12:1
2 2
where h 2 0; p and / 2 0; 2p. We can also write this arbitrary state in the fol-
lowing vector format

cos h2
jWi ! ; 12:2
ei/ sin h2
which we may nd useful when performing transformations of qubits from one

state to another.
Here h and / dene a point on a sphere, known as the Bloch sphere, as seen in
Fig. 12.1. One can think of the surface of the Bloch sphere as a map of states that a
qubit can take. j0i and j1i are at the poles of the sphere, and therefore an equal
superposition of j0i and j1i can be found on the equator, with the phase / varying
around the equator. Other positions on the sphere can be reached by changing the
values of h and / through qubit transformations. Superposition is one of the
properties of qubits that allow the speed up of some algorithms compared to
classical methods.
12.1.1.2 Entanglement
Another property of qubits that is often hailed as the most counter-intuitive is that of
entanglement. Let us take two qubits which can each take the state j0i or j1i. These
two qubits can therefore be in the states j00i, j01i, j10i or j11i (where jabi is
equivalent to jaijbi) but they could also be in an arbitrary superposition state with
variable amplitudes aij where i and j can take the value 0 or 1. The state is then
jWi a00 j00i a01 j01i a10 j10i a11 j11i ; 12:3
for which there are special cases that we will now discuss. For particular amplitudes
we can create states which demonstrate entanglement, where one photon of the pair
will be correlated in some way to the other photon of a pair. The states we will
consider are known as the Bell states after John S. Bell presented them in 1964 to
use as a test of non-locality [8].
1
jW i p j01i j10i ; 12:4
2
1
jW i p j01i j10i ; 12:5
2
1
jU i p j00i j11i ; 12:6
2
1
jU i p j00i j11i : 12:7
2
The Bell states are maximally entangled states in that they cannot be factorised
into a state of the form a1 j0i b1 j1i1 a2 j0i b2 j1i2 where the subscripts
represent two separate quantum particles 1 and 2. It is this property of entanglement
that underlies much research into quantum communication and cryptography where
the correlations observed when measuring an entangled qubit can be used to build
up a key to securely transfer information [9, 10].
There are many avenues currently being explored to nd which two level sys-
tem, or qubit, is most suited to the building of a quantum computer. Some sug-
gested routes are the use of the electron levels of cooled trapped atoms [11] or ions
[12] or the electron spins in colour centres in diamond [13]. We will concentrate on
the use of photons as carriers of quantum information. This provides many
advantages such as the longevity of the storage of information due to the low
interaction of photons with their environment, characterised by their long coherence
times, and the ease of application of single qubit operations, performed with the use
of birefringent crystals or wave plates.
12.1.1.3 Encoding and Measurement
There are many different two level systems available to a photon which can be used
to encode a qubit. One of the most common is the use of path encoding shown in
Fig. 12.2a, where one path, say the upper path, can be labeled j0i whilst an alternate
path, the lower in this case, is labeled j1i. This encoding scheme is especially useful
(a) PBS
HWP HWP
PBS
(b) (c)
Fig. 12.2 Possible qubit encoding schemes using photons including a path encoding, b time-bin
encoding, and c polarisation encoding
in media where polarisation is either not discriminated or not preserved, such as

monolithic waveguide structures [14]. A scheme that is often used for long distance
quantum communication, where only one channel exists between the two users is
that of time-bin encoding [15], shown in Fig. 12.2b. In this scheme a photon that
travels through the short arm of the Mach-Zehnder interferometer is in the state j0i
whilst a photon that has travelled through the long arm is in the state j1i. These
photons are then sent and, as there is no dependence upon polarisation, are rela-
tively free from decoherence over long distances in optical bre. The reverse
process can be performed at the receiving end and analysis can be performed by
measuring the arrival times of the photons. Whilst this encoding is very robust
against decoherence it suffers from the fact that is hard to construct gates between
multiple qubits, making it more suited to communication schemes rather than
computation, although some recent schemes do exist [16].
For this discussion we will use one of the most common and well researched
encoding schemes, namely polarisation encoding. In this scheme, shown in
Fig. 12.2c, j0i and j1i are represented by horizontally and vertically polarised
photons, jHi and jVi respectively. Other polarisation states represent superpositions
of these states and can be easily generated through the use of birefringent plates or
wave plates, which will be discussed in the next section. It can be seen from
Fig. 12.2a that it is relatively easy to convert between polarisation encoding and
path encoding, and vice versa, with the use of polarising beamsplitters and wave
plates. Wave plates allow us to perform single qubit operations on polarisation
encoded photons. This birefringent media formed into plates causes light travelling
through in one polarisation to travel faster than the orthogonal polarisation so a
phase is gained between them. It is this extra phase that allows the wave plate to
change the polarisation of the light as it propagates. There are commonly two types
of wave plates used, the half-wave plate and the quarter-wave plate, so named for
the proportion of a wavelength of the light that they retard one polarisation com-
pared to the other.
Using the vector notation for polarisation states as in (12.2) we can write the
matrix representations for quarter- and half-wave plate operators as

d cos2u sin2u
HWPu ; 12:8
sin2u cos2u

d 1 1 i cos2u i sin2u
QWPu p ; 12:9
2 i sin2u i cos2u
where u is the angle the fast axis of the waveplate has been rotated from horizontal
in both cases, which is the same as the Jones matrices used in classical polarisation
optics. The NOT operation is performed by setting the angle of a half-wave plate to
u 45 while the Hadamard operation corresponds to a setting of u 22:5 . With
a combination of three wave plates, namely a quarter-, half-, quarter-wave plate,
any arbitrary rotation of the state of the qubit can be performed (Fig. 12.3).
Quantum information processing (QIP) requires more than just single qubit
operations. We also have the need for an interaction between qubits, known as two
qubit operations. We often refer to these operations as controlled gates as the state
of one qubit often decrees what will happen to the second qubit in the operation.
One of the simplest of such gates is the controlled-Phase gate, often shortened to
CZ gate. This operation applies a phase shift of p to one qubit, the target, condi-
tional on the other qubit, the control, being in the j1i state. To perform such
operations using photons as qubits we require two photons to interact with one
another; an inherently difcult task. The interaction associated with the CZ gate is
equivalent to a cross-phase modulation of p on one photon, induced by another.
There has been much research into nonlinear materials that may allow the presence
of one photon in a medium to have an effect on another photon present, but
currently no material has been found that could provide the full p phase shift
required to perform a CZ or CNOT gate in a sensible amount of media. In the
absence of such a strong nonlinearity one may think that photonic quantum gates
are not realizable, but a publication in 2001 by Knill et al. [17] suggested a feasible
route. This paper shows that by using ancilla (meaning simply extra) qubits,
(a) (b)
QWP
HWP HWP
QWP
Fig. 12.3 The use of wave plates to perform a a Hadamard operation and b the generation of an
arbitrary qubit state
detection, and linear optical networks a measurement-induced nonlinearity is

achieved, allowing the CZ and CNOT to be realised in a probabilistic manner.
At the heart of this scheme is the injection of two photons into a beam splitter at
which point they interfere with one another in a non-classical way. If we take a
beam splitter with a reflectivity of 0.5 and input one photon into each of its input
ports then, if the photons are completely indistinguishable in all degrees of freedom
including spectral, temporal, spatial and polarisation modes, they will both exit
through the same output port. This is what is known as quantum interference and it
arises from the fact that when we are using quantum particles we must calculate
their probability amplitudes for undergoing particular operations and not just the
probabilities.
Experiments to verify this effect involve taking a particular mode of one of the
photons and adjusting it in order to make it distinguishable from the other photon
involved and then scanning this through the point where the modes of the two
photons match. This effect can be observed by monitoring output ports and
counting the number of coincident events, conrming that there are none at the
point of complete indistinguishability. The rst such experiments were performed
by Hong et al. in 1987 [18] and demonstrated independently by Rarity and Tapster
[19] in which they varied the temporal mode by delaying the arrival at the beam
splitter of one photon compared to the other. As the delay is reduced a drop in
coincidences is observed which should reach zero at the point of zero delay if all
other photon modes are indistinguishable. As the delay is then increased again the
photon coincidences will rise. This characteristic dip in counts is often referred to as
the Hong-Ou-Mandel (HOM) dip and is at the heart of many linear optical QIP
schemes.
Whether we wish to perform some small scale quantum information processing
using linear optics, or investigate other nonlinear optical schemes, we will always
require sources of single photons which often rely on nonlinear interactions. In the
next section we will discuss how such photon sources can be constructed and how
the use of different nonlinear media can lead to differences in the generated photons.
12.2 Nonlinear Optics for Single Photon Generation
12.2.1 Single Photon Sources and the Heralding of Single

Photons
In the previous section, we have established the importance of developing high

quality sources of single photons. We now turn to the methods by which this might
be accomplished and the realistic description of both the ideal states and what is
possible today.
12.2.1.1 Ideal Single Photons
The ideal single photon should typically have a specied spatial and polarisation
state, and a temporal/spectral prole that is near Fourier-transform limited. Clas-
sically we might represent such a state by an electric eld wave packet written in
the frequency domain as
Z1
Er; t A dx f xEx r eixt c:c; 12:10
0
where f x is the spectral envelope, Ex is the spatial mode prole and c:c: denotes
the complex conjugate. The normalisation A would be chosen so that the energy of
the wave packet corresponds to hx where x is the centre frequency of the pulse.
However, many of the applications [17, 20, 21] discussed within Sect. 12.1 can
only be described in terms of the fully quantised electromagnetic eld, and as we
see below, the generation of few photon states by nonlinear optical processes must
also be described in this language. Therefore we introduce our ideal single photon
state using the formalism of eld quantisation and boson creation and annihilation
operators. A suitable single photon state can thus be written
Z1
jwa i y jvaci;
dx f xâax 12:11
0
y
where âax is the creation operator for the mode a at frequency x, and satises the
y
commutation relation âax ; âbx0 dab dx x0 , generalising the states jH i and
jV i of Sect. 12.1.1.3. Further jvaci is the vacuum state, and the photon wave
function f x is normalised according to
Z1
dx jf xj2 1: 12:12
0
For the remainder of Sect. 12.2, we are concerned with understanding and
developing physical techniques for creating light elds that approach the ideal of
(12.11).
12.2.1.2 Approximating Single Photons
There are three methods commonly used to produce approximations to this

idealisation.
Two-level System Excitation
Imperfect
collection
Fig. 12.4 Schematic illustration for the generation of photons using atomlike sources
Classical pulsed pump Poissonian photon statistics
Attenuator
Probability single photon per pulse < 0.1
Fig. 12.5 Schematic illustration for the generation of photons using attenuation
The rst approach is triggered or on-demand sources (see Fig. 12.4), which
rely on the spontaneous decay of atom-like systemsusually atoms, ions, colour
centres or quantum dots in a cavityto truly emit one photon at a time. Examples
of such systems include quantum dots [22, 23] and nitrogen vacancy colour centres
[24, 25]. The challenge here is that the systems available often require cryogenic
conditions and have poor emission directivity. System stability and reproducibility
of the emission wavelength can also be a problem, though they are rapidly
improving. Nevertheless while triggered sources represent an ultimate goal, it
remains highly challenging to produce indistinguishable photons from distinct
atom-like systems [26].
Alternatively, an approximation to a single photon source can be created by
strongly attenuating a laser, leaving only a small probability that a single photon
exits in a given time window (see Fig. 12.5). The likelihood of two or more photons
exiting the laser is even lower. However, as a laser eld corresponds to a coherent
state in quantum optics [27], there is a nonzero probability of emitting n photons,
determined by the Poisson distribution
ln
PPoisson n; l el ; 12:13
n!
where l is the average number of emitted photons per pulse. The ratio of multi to
single photon emission probabilities PPoisson [ 1; l=PPoisson 1; l l=2 can only
be reduced by lowering the net flux, limiting the usefulness of these sources.
As neither of these two classes depend on nonlinear optical effects we will not
discuss them further here. An excellent review may be found in [28].
The nal method is to use probabilistic or heralded sources, in which a pair of
photons is generated together and one member of the pair detected to herald the
presence of the other (see Fig. 12.6). This approach usually relies on the production
of photon pairs via spontaneous parametric downconversion (SPDC) [2936], or
spontaneous four-wave mixing (SFWM) [3742]. The concept of a single photon
state described by (12.11) should now be replaced by the bi-photon state
Classical pulsed pump

Poissonian/thermal pair statistics
Nonlinear
Media
Probability a photon pair per pulse < 0.1
Fig. 12.6 Schematic illustration for the generation of photons using photon pair generation in a
nonlinear medium
Z1 Z1
jwII i dxs y ây jvaci;
dxi /ab xs ; xi âax 12:14
s bxi
0 0
characterised by the bi-photon wavefunction /ab xs ; xi . Apart from a normali-

sation condition, the function /ab xs ; xi is in principle unconstrained, and indi-
cates that the generated bi-photon state may have strong and non-trivial frequency
and time correlations between the two photons. The correct quantum description of
the heralded single photon is found by tracing over the detected photon by a
procedure described in Sect. 12.2.5.3.
While the probabilistic nature of these sources is perhaps their greatest imped-
iment, we see in Sect. 12.2.6 that schemes for enabling near-deterministic behav-
iour are emerging rapidly [43, 44]. We therefore focus on probabilistic sources
throughout the rest of this chapter, focusing on theory in the next Sect. 12.2.2,
before turning to practice in Sects. 12.2.3 and 12.2.4. We then delve more deeply
into the quantum description of the generated states in Sect. 12.2.5.
12.2.2 Photon Pair Sources and Photon Statistics
The two nonlinear processes of SPDC and SFWM are actually quantum analogs of
familiar classical nonlinear wave-mixing effects [45, 46]: difference frequency gen-
eration and four wave mixing. In the classical processes, photons are transferred in
wavelength from one or two pump elds onto a weak input signal eld and a
second idler channel, the frequency of which is determined by energy conserva-
tion. The spontaneous quantum processes are essentially identical except that there is
no input signal eldthat role is taken by random quantum fluctuations. It is for this
reason that we require a quantum formalism to describe their operation [47].
In SPDC a second-order (v2 ) nonlinear optical material is used to probabilis-
tically convert an incident pump photon at xp into a photon at each of xs and xi
where xp xs xi . Similarly, in SFWM a third-order (v3 ) nonlinear material is
used to probabilistically convert two incident pump photons at xP into a photon at
each of xs and xi where 2xp xs xi (see Fig. 12.7). However, since there is no
pre-existing signal eld, in both cases the generated state is actually a superposition
SPDC SFWM
P S
P
2 S 3
I I
P=S+I 2P=S+I
Fig. 12.7 Schematics of the processes of spontaneous parametric downconversion (SPDC) and
spontaneous four-wave mixing (SFWM)
Fig. 12.8 Typical photon 1

pair production probabilities
per pump pulse as given by a
thermal distribution. The vast 0.8
majority of events produce no
pairs, and the multiple pair
Probability
0.6
events cannot be eliminated
0.4
0.2
0
0 1 2 3 4
No. of pairs
over all frequency pairs xs and xi which satisfy energy conservation. This leads to
frequency correlations between the two photons as we explore in detail in
Sect. 12.2.5.1. Note that as with classical four-wave mixing, it is also possible to
achieve degenerate photon pair production by using two pump elds satisfying
xp1 xp2 2xs .
Unfortunately a heralded photon source is still far from the ideal of being able to
produce a single photon on demand, for a number of reasons we will explore in
later sections. Perhaps the most fundamental is that we cant be sure of obtaining a
single pair (just as the attenuated laser cannot guarantee a single photon). In fact,
the simplest description of the output of such a source is as a two-mode squeezed
vacuum. In that approximation, the probability of obtaining n pairs per pump pulse
is given by the thermal distribution [48]
ln
PThermal n; l ; 12:15
1 ln1
where l is now the average number of pairs per pulse (see Fig. 12.8). This
description assigns only a single frequency mode each to the signal and idler, a
deciency we amend below. However, it gives useful insight into the photon
statistics.
From (12.15), it is easily found that the relative likelihood of generating two or
more pairs is given by PThermal [ 1; l=PThermal 1; l l, and so to keep this rate
small, the mean photon flux must be limited to l 1. As a result, the most likely
event for any given pump pulse is for the source to produce no photons at all!
Moreover, it is impossible to predict which pump pulses will produce pairs. What
then is the advantage over the attenuated laser? The utility of heralded sources is
that when photons are produced, they always appear in pairs.
Thus following the separation of signal and idler photons along different spatial
paths (or one of the different encoding schemes of Sect. 12.1.1.3), the detection of a
single photon serves as a good indication that another is present (subject to
absorption and scattering losses etc.). We cannot predict when a single photon from
this type of source is coming, but at least we get a warning. In Sect. 12.2.6, we see
that this advance notice is a crucial tool for turning an unpredictable spontaneous
source into a quasi-deterministic one.
We now turn to the laboratory characterisation of many of the photon statistics
concepts including the impact of losses and detector imperfections. Reference [49]
also discusses a number of these ideas in detail. The characterisation of sources
begins with the photon statistics: the rate of pairs, coincidence statistics and effects
of noise. This is the realm of the second order correlation function and the coin-
cidence-to-accidental ratio or CAR, with which we begin.
12.2.2.1 Noise in Photon Pair Sources and the Coincidence-to-

Accidental Ratio (CAR)
An ideal heralded pair source would produce a steady succession of equally spaced
single photon pairs, all identical. As weve seen however, both SPDC and SFWM
actually give rise to squeezed vacuum states with a nonzero rate of multiple pairs
and characterising this rate is important. Moreover in addition to noise from multi-
pair production, we must consider contributions from spurious photons generated
by spontaneous Raman scattering (especially in amorphous materials), and
imperfections of the measurement system including channel losses, dark counts,
after-pulsing and detector efciency [49].
The standard tool for this kind of characterisation is a coincidence measurement
setup, illustrated in Fig. 12.9. The output photons from the pair source under study
are ltered to remove residual pump light, and then separated into distinct output
channels, typically optical bres, labelled as signal and idler modes (based on
frequency, wavevector or polarisation as the case may be). In practice, this sepa-
ration may be accomplished with a dichroic mirror, arrayed waveguide grating
(AWG) or polarising beam splitter. Each output channel may contain additional
ltering elements such as Bragg gratings to remove as much of the residual pump
light as possible (a total of 120 dB pump suppression is common for SFWM
sources). Finally, each channel is connected to a single photon detector with one
Bright pump Photon pair generation by Block Wavelength Single photon

pulse SFWM pump Separation detection
AWG
SPDs &
Fig. 12.9 Schematic of a coincidence setup for CAR measurement. The output of a pair source is
ltered to remove excess pump light, and then passed through a device to separate signal and idler
channels. The two channels are detected separately with single photon detectors connected to a
time-interval analyser or other device that identies detection coincidences within some time
window
arm containing a variable time delay. The two detectors are connected to a time-
interval analyser which records coincidence events when the two detectors re
within a given interval, usually on the order of a few nanoseconds. The experiment
measures the number of coincidence events as the time delay is varied, either
continuously or discretely depending on whether a CW or pulsed laser is used.
If there were perfect separation of signal and idler, complete extinction of the
pump, negligible loss, no other sources of stray photons, and noiseless detectors of
perfect efciency, one would see both detectors click in coincidence when a pair is
present, and an absence of clicks (whether in coincidence or as singles ring in
only one channel) otherwise. In practice, most of these assumptions are invalid: the
pump ltering does not remove every pump photon from the signal and idler bands,
particularly for SFWM sources where the pump-signal separation may be only a
few THz; scattering or absorption loss in waveguides may be considerable; spon-
taneous Raman scattering can introduce a signicant level of uncorrelated photons,
particularly in amorphous materials; out-coupling efciencies from on-chip devices
may be only a few percent; and while single photon detectors are improving rap-
idly, it is routine to work with efciencies in the range of = 525 %, especially in
the telecom band. Finally, there is the fact that the photon statistics guarantee that
some events are multiple pairs.
To quantify these imperfections, a natural signal to noise metric is the ratio of the
rate of true coincidences, due only to the desired single pair events, to the
coincidence rate from all other causes: dark counts and after-pulsing, multiple pairs
and noise photons. Operationally, this motivates the introduction of the coincidence
to accidental ratio
Ct C A
CAR ; 12:16
A A
where Ct is the true rate of coincidences due to single pairs, C is the measured rate
of coincidences due to all causes, and A is the expected rate of accidental
coincidences due to all other causes, such as say the coincidence of a photon
generated by spontaneous Raman scattering and a detector dark count. We have no
way of identifying the good coincidences from the bad, so the accidentals rate
is determined by a statistical trick in which the coincidence rate measurement is
repeated with a time delay inserted into one arm. For a pulsed pump, this delay
corresponds to one or more multiples of the pump repetition time sp . We explain
why this works shortly.
Expressions for the CAR can be obtained with different levels of rigor [48, 50].
Suppose for a moment that pair creation events are entirely uncorrelated (a point to
which we return in a moment) and occur according to Poisson statistics with a mean
rate l per pump pulse. Ignoring dark counts and assuming l 1, the coincidences
are given by Ct lgs gi , where gs ; gi are the total efciencies including coupling
losses etc. of each detector. The rate for accidental coincidences in adjacent pump
pulses is A Ns Ni where the singles rate of clicks at each detector individually
are Ns lgs and Ni lgi . For uncorrelated creation events, this accidentals rate is
precisely the same as the rate for two pair creation events to occur in the same pulse
window. Thus the CAR obeys a simple 1=l law indicating that multiple pair events
are more likely at higher pair rates (see Fig. 12.10). Observe that while there is no
particular interest in the coincidences between successive pump pulses per se, in the
Poisson photon statistics regime, the accidentals rate is a convenient estimator of
the contribution of multiple pairs and other imperfections to the overall coincidence
rate, a contribution we cannot observe directly.
In practice, dark counts and other spurious effects such as spontaneous Raman
noise must be included, and a modied expression [50] is commonly used:
lgs gi
CAR r; 12:17
lgs ds lgi di
where ds ; di are the rate of dark counts per pump pulse of each detector, and r is a
correction factor we set as r 0 for now. Dark counts for near infra-red detectors
typically range from a few hundred Hz for superconducting devices to a few kHz
for InGaAs based avalanche photodiodes. As might be expected, the dark counts
suppress the CAR when the effective singles rate lg approaches the dark count rate
from above. Indeed assuming the detectors have identical characteristics, there is a
Fig. 12.10 Coincidence-to-

300
accidental ratio CAR
according to (12.17) for
efciency = 0.1, and dark
count rate d = 104/pulse 200
(solid blue line), and the
CAR
limiting case 1/ for

negligible dark counts
(dashed red) 100
0
0 0.01 0.02 0.03 0.04

maximum CAR of g=4d at lmax d=g (see solid blue line in Fig. 12.10). In
practice, researchers work in a range where l lmax and the 1=l fall due to
multipair generation (dashed red line in Fig. 12.10) is the dominant effect. It is
apparent that detector inefciency, dark counts and other imperfections play a key
role in determining the CAR which is in some ways a strange metric: it is as much a
measure of the detectors as the source itself.
The argument above regarding the accidentals relied on the generation of
uncorrelated events with Poisson statistics. How does the identication of the ac-
cidentals with the rate of uncorrelated events in successive pulses stand up when the
photon pair statistics are not Poissonian? Note that correlated pairs can be generated
with different numbers of independently correlated frequency modes. With just a
single pair of modes, the distribution of the number of generated pairs per pulse is
thermal according to (12.15). A more careful analysis of the photon statistics
leading to (12.17) has been performed by Takesue and Shimizu [48], and shows
that for l lmax the same equation holds with a correction r 2. On the other
hand, with a large number of frequency modes, while the distribution of pairs within
each mode is thermal, there is no correlation between different modes and as
multiple pair events are much more likely to involve two different modes rather than
the same mode, the total distribution of pairs approaches a Poisson distribution. In
this case, the correction r 1. For useful sources, researchers seek to obtain CARs
in the range 100 or larger [51], and the difference in the expected CAR between the
two regimes is unimportant, if even measurable.
Note that the question of whether a source is in the regime of a large or small
number of frequency modes, is very important, because of the connection to the
purity of the heralded photon, but this is best characterised by other measurements:
see Sect. 12.2.5.5.
12.2.3 Pair Generation Processes in Second-Order

Nonlinear Media
Today, the workhorse process for pair generation is spontaneous parametric

downconversion (SPDC) in v2 materials, still usually in bulk optics congura-
tions. In SPDC, one pump photon is annihilated to produce two daughter photons at
lower frequencies, (which may or may not be degenerate). As with the corre-
sponding classical process of difference frequency generation, for efcient SPDC
energy and momentum must be conserved (or phase-matched), so that
xp xs xi ;
k p k s ki : 12:18
Here xp;s;i and kp;s;i are the frequencies and wave vectors of the pump, signal
and idler photons, respectively. As usual, jkj nxx=c, where nx is the
refractive index of the nonlinear medium at frequency x and c is the speed of light
in vacuum. Such equations are familiar from the classical processes of difference
and sum frequency generation. In fact, as the nonlinear medium has nite length L
and the pump has nonzero bandwidth dxp , these conditions are more accurately
expressed as inequalities of the form

jDxj xp xs xi \ dxp ;

p
jDkj jkp ks ki j. : 12:19
L
Equation (12.19) show that the bandwidth of generated photons is determined by

the pump bandwidth and chromatic dispersion.
As the refractive index at xp is usually much higher than those at xs and xi , and
different values of chromatic and group velocity dispersion can be found at these
frequencies, it is non-trivial to achieve the required phase-matching and the non-
linear devices and experimental apparatus must be carefully designed to do so.
Depending on the design, the two generated photons can be emitted either collin-
early or at separated angles to the pump beam, and be either at different frequencies
or degenerate. With non-collinear generation in bulk crystals, the formal repre-
sentation of the state generalising (12.14) has the rather complex form
Z1 Z1 Z Z
y y
jwII i dxs dxi d2 kt;s d2 kt;i Uxs ; xi ; kt;s kt;i âxs kt;s âxi kt;i jvaci;
0 0
12:20
where the bi-photon correlations may depend on the transverse wavevectors of the
two photons as well as their frequencies [52]. In our later theoretical discussion, we
restrict to the notationally simpler case of collinear emission.
Photon pairs generated through SPDC have found numerous applications in
quantum photonic information processing such as entanglement generation [53],
quantum logic gates [14], quantum walks [54] and quantum simulation [55]. In the
following, we present schemes for their production in both bulk nonlinear crystals
and waveguide structures.
12.2.3.1 SPDC in Bulk Crystals
SPDC in bulk nonlinear crystals is the most often used approach to generate single
photons or entangled photons for quantum information processing. In this section
we briefly describe SPDC in bulk crystals such as beta-barium borate (BBO)
crystals. We discuss the phase matching angle, emission in cones and the different
types of SPDC. Reference [52] provides a recent and very detailed review of this
topic.
BBO is a negative uniaxial crystal, i.e., it has only one optic axis and no [ ne ,
where no and ne are the ordinary and extraordinary refractive indices respectively.
Based on this, there are two possible types of phase matching: type-I (e ! o + o)
and type-II (e ! o + e). In type-I SPDC, the two emitted photons have the same
polarisation which is orthogonal to that of the pump; in type-II SPDC, one of the
emitted photons has the same polarisation as the pump.
A typical type-I SPDC process in a BBO crystal is schematically shown in
Fig. 12.11a. The incident pump beam is generally normal or slightly tilted to the
surface of the crystal, and the generated signal and idler photons are emitted along
cones centered on the pump. To meet the phase-matching condition as shown in
Fig. 12.11b, the photons of each pair must lie on opposite sides of the pump beam.
Through the cut of the crystal surface, the angle hpm between the pump beam and
the crystal optic axis can be varied to control the phase-matching condition. By
adjusting hpm , the photons may be chosen to be degenerate and thus are emitted on
a single cone, as indicated by the squares in Fig. 12.11a. Alternatively they can
have different frequencies in which case they are emitted along two different cones,
as shown by the circles and triangles in Fig. 12.11a. The precise geometry of
emission is beyond the scope of this chapter and the details can be found in [56].
When hpm is adjusted properly, one can also have type-II phase matched SPDC
processes, in which the down-converted photons are emitted into two cones, one
ordinarily polarised, the other extraordinarily polarised. By adjusting the phase
matching angle, the relative orientation of the two cones can be adjusted. If they are
(a) (b)
Pump ks
pm
ki
i s
( p, k p)
kp
BBO crystal, type I Momentum conservation
(c)
extraordinary
Pump
( p, kp)
BBO crystal, type II ordinary
Fig. 12.11 The diagram for SPDC processes in BBO. a Type-I SPDC, b momentum conservation
for phase matching, and c type-II SPDC for polarisation entanglement generation
made to overlap [52], (Fig. 12.11c) the photons emitted along the intersection
directions can be described in the simplied picture of Sect. 12.1 by the entangled
state [53]:
p
jWi jHs Vi i eia jVs Hi i= 2; 12:21
where H and V indicate horizontal (extraordinary) and vertical (ordinary) polari-

sation, respectively, and the subscripts denote the signal and idler directions. The
relative phase a derives from the crystal birefringence. Using an additional bire-
fringent phase shifter, a can be set as desired, for example, to the values of 0 or p to
produce standard Bell states.
In experiments, the entanglement generation is complicated by the walk-off
between the ordinary and extraordinary photons since they have different velocities
inside the birefringent crystal. Two crystals identical to the down-conversion crystal
but only half as long, may be used to compensate the walk-off and remove the
temporal distinguishability of the ordinary and extraordinary photons [53].
Other nonlinear crystals that are often used to generate heralded single photons
and entangled photons are BiBO (BiB3 O6 ) and potassium titanyl phosphate
(KTiOPO4 or KTP). Compared with BBO, BiBO exhibits two times higher non-
linearity and has two optic axes. As a biaxial crystal, phase matching in BiBO
becomes more complicated because it depends on the azimuthal pump angle in
addition to the polar pump angle. However such increased complexity offers more
flexibility to tune the phase-matching condition and thus allows broadband angle
tuning [57, 58]. The exceptionally high nonlinearity and versatile phase matching
make BiBO very attractive for multi-photon entanglement generation. In contrast to
BBO and BiBO, which can only be used as bulk crystals, KTP is usually period-
ically poled to provide quasi-phase-matching for efcient SPDC [59, 60] and
waveguides can be formed in this material through titanium indiffusion or soft
proton exchange. The details will be covered in next section.
12.2.3.2 SPDC in Integrated Structures
The bulk nonlinear crystals just described have played a very important role in
generating heralded single photons and entangled photons for quantum experi-
ments. However there are intrinsic limitations that constrain the performance of the
sources based on bulk crystals. For example, the length of crystals is usually limited
to a few mm or sub-mm to avoid severe walk-off between generated photons [53,
57, 58]. The crystal length and damage threshold set the bottleneck of photon
source brightness. The other limiting factor is the need to use many free-space
optical components, which are environmentally sensitive and scale poorly for more
complicated quantum operations. To overcome these hurdles, integrated waveguide
devices based on both second-order and third-order nonlinear materials for heralded
single photon and entangled photon generation have attracted intensive research.
In waveguides, where all elds propagate collinearly, new strategies for

achieving phase-matching are required. These may include designing waveguides
that support highly distinct mode-proles at different wavelengths as has been done
with Bragg reflection waveguides in AlGaAs structures [6163]. Another elegant
approach is to use ring resonator structures [64] in which the variation of the
orientation of the crystal axes and eld polarisation guarantees phase-matching.
However, the most common approach is to adopt quasi-phase matching (QPM) [65]
in periodically poled second-order nonlinear waveguides such as periodically poled
KTP (PPKTP) [59, 60], lithium niobate (PPLN) [31, 66, 67] and lithium tantalate
(PPLT) [68].
QPM operates essentially identically for spontaneous processes as for classical
frequency conversion. Figure 12.12, shows a homogeneous single crystal (a) and a
periodically poled waveguide (b), and the evolution (c) of the eld amplitude of the
generated wave along the propagation direction z for three different phase-matching
conditions. It can be seen from Fig. 12.12c that in the case of perfect phase-
matching, the amplitude of the generated wave grows linearly with propagation
distance; while in the situation of no phase-matching, the amplitude oscillates with
distance. If the poling period K is set equal to twice the coherence length Lcoh of the
nonlinear interaction, each time the eld amplitude of the generated wave is about
to begin to decrease as a consequence of the phase-mismatch, a reversal of the sign
of the nonlinear coefcient occurs. This allows the eld amplitude to continue to
(a) (b)
(i)
(c) (i) Perfect phase-matching
(ii) Quasi-phase-matching
(ii)
(iii) No phase-matching
Field amplitude
0
(iii)
0 2 4 6 8
z/L coh
Fig. 12.12 Schematic of a second-order nonlinear optical material in the form of a a

homogeneous single crystal and b a periodically poled waveguide in which the axis of the
nonlinear material alternates in orientation with period so that its nonlinear coefcient changes
sign periodically. c Comparison of the evolution of the eld amplitude of the generated wave along
the propagation direction z for three different phase matching conditions: (i) perfect phase-
matching, (ii) quasi-phase-matching and (iii) no phase-matching. Lcoh indicates the coherent
buildup length [65]
grow monotonically and is equivalent to adding an additional term 2p=K to (12.18)

to compensate the phase-mismatch. In practice, the smallest poling period is limited
to a few lm due to fabrication constraints and this determines the maximum phase-
mismatch that can be compensated.
Compared with the perfect phase-matching via birefringence in bulk crystals,
QPM leads to a lower SPDC efciency per unit device length if the nonlinear coef-
cient is the same: the effective nonlinearity is reduced by a factor of 2=p [65].
However, QPM makes it possible to use the same polarisation direction for all
interacting waves, and this often allows a larger element of the nonlinear tensor to be
accessed. Furthermore, QPM allows SPDC in waveguide structures which conne all
the interacting waves in the same spatial waveguide mode and thus increases their
nonlinear coupling. As a result, the SPDC efciency in poled waveguides can be
signicantly higher than that in bulk crystals. For example, the SPDC efciency in a
PPKTP waveguide is more than 50 times higher than that in a bulk crystal [59, 60].
Moreover, a poled waveguide can be more than ten times longer than a bulk
crystal without decreasing the SPDC efciency, further enhancing the source
brightness. QPM in poled waveguides makes phase-matching so flexible that it
allows not only SPDC through three-wave mixing [31, 67], but also four-wave
mixing type SPDC via cascaded second-order nonlinear processes [66, 68]. This
makes it possible to generate photons in the telecom-band using readily available
telecom-band pump lasers. Finally, variation in the poling period can be introduced
to engineer the precise bi-photon state emitted from the structure. This can be used
to improve the purity of the resulting heralded single photon [69].
12.2.3.3 Cavity Enhanced SPDC
Since the bandwidth is determined by the chromatic dispersion, as was previously

mentioned, photons generated via SPDC in either bulk crystals or periodically poled
waveguides usually have a spectral bandwidth on the order of THz. This is signi-
cantly beyond the working bandwidth of Hong-Ou-Mandel quantum interference
experiments [18] and available quantum memories (MHz to GHz) [70], both of which
are at the heart of schemes for linear optical quantum computation [17]. The use of
SPDC sources in these applications requires narrow-band frequency lters. In par-
ticular, to store SPDC photons directly in atomic memories, the required lters will be
extremely narrow and introduce large losses [71]. To solve this problem, cavity
enhanced SPDC has been proposed and demonstrated [70, 72, 73].
The idea is illustrated in Fig. 12.13. A second-order nonlinear device, either a
bulk crystal or periodically poled waveguide, is placed in a Fabry-Perot cavity. The
cavity is designed such that both input and output mirrors transmit the pump, while
in the down-conversion band, the input mirror is highly reflective, but the output
mirror is only partially reflective. The bandwidth of the down-converted photons is
determined by Dm FSR F , where FSR and F are the free spectral range and nesse
of the cavity, respectively.
Cavity
Pump
Nonlinear device
Fig. 12.13 A second-order nonlinear device is placed in a Fabry-Perot cavity for narrow-band
SPDC. Both input and output mirrors of the cavity transmit the pump, while the input and output
mirrors highly and partially reflect the down-converted photons, respectively
12.2.4 Pair Generation Processes in Third-Order

Nonlinear Media
We now turn to the corresponding third-order (v3 ) nonlinear process, spontaneous

four-wave mixing (SFWM), and its applications in quantum photonics. The third
order nonlinearity is far weaker than the second order one, but is attractive because it
is present in all platforms including bre, semiconductor waveguides and planar glass
technologies. In contrast to SPDC in which only three photons are involved, SFWM
is a four-photon interaction: two pump photons are annihilated to produce a signal
photon and an idler photon at new frequencies. The signal and idler photons gener-
ated in this way are correlated and can be used for entanglement generation and
heralded single-photon sources in the same fashion as pairs from SPDC. Efcient
SFWM requires energy conservation and phase-matching according to [46]:
xp1 xp2 xs xi ; 12:22a
kp1 kp2 ks ki DkNL : 12:22b
Here xp1 ;p2 ;s;i and kp1 ;p2 ;s;i are the frequencies and wave vectors of pump, signal
and idler photons, respectively; DkNL is the nonlinear phase shift introduced by self-
phase modulation and cross-phase modulation. Like (12.18), these conditions are
more accurately represented by inequalities of the form of (12.19). Depending on
the polarisation states of the four waves and whether the two pump waves have the
same frequency, the form of DkNL varies. When xp1 6 xp2 and all four waves are
linearly co-polarised, DkNL cP1 P2 , where P1 and P2 are the pump power,
c n2 x=cAeff is the nonlinear coefcient, n2 is the nonlinear refractive index at
frequency x and Aeff is the effective mode area of pump waves in a nonlinear
waveguide. When xp1 6 xp2 and two pumps are linearly co-polarised but signal
and idler photons are orthogonally polarised with respect to pumps,
DkNL 53 cP1 P2 . In the case of xp1 xp2 and pump power of P, DkNL
2cP when all waves are linearly co-polarised, and DkNL 23 cP when the pump is
linearly polarised while signal and idler photons are polarised orthogonally to the
pump polarisation.
(a) Input SFWM
Output
p i p s i p s
(b) Input SFWM
Output
p1 p2 p1 s,i p2 p1 s,i p2
Fig. 12.14 Schematic of pump frequency a degenerate and b non-degenerate SFWM processes
The case of xp1 xp2 is usually referred to as pump frequency degenerate

SFWM, and the case of xp1 6 xp2 as non-degenerate SFWM. Both are schemat-
ically shown in Fig. 12.14. In the pump degenerate SFWM process shown in
Fig. 12.14a, the signal and idler photons appear at higher and lower frequency than
the pump, respectively. For non-degenerate pumps, as shown in Fig. 12.14b, the
signal and idler photons can be generated at the same frequency between two pump
frequencies. In this case the two pump frequencies are usually very close, so the
frequency dependency of c can be ignored [46]. In the following sections, we
discuss the implementation of these two types of SFWM in optical bres, integrated
waveguides and resonant structures.
12.2.4.1 SFWM in Optical Fibres
Here we describe previous single or entangled photon generation experiments using

SFWM in standard optical bres and photonic crystal bres (PCFs). Most work has
been done with pump degenerate SFWM, so we focus on this type of SFWM and
then briefly introduce the demonstration of non-degenerate pump SFWM. For a
recent review of pair generation in bres, see [74].
As mentioned above, for a single pump we have xp1 xp2 , and (12.22a,
12.22b) become
2xp xs xi ; 2kp ks ki BcP: 12:23
Here B 2 when all waves are linearly co-polarised while B 23 when the
polarisation of signal and idler photons is orthogonal to that of the linearly polarised
pump. For the case of all elds co-polarised, expanding kp;s;i in a Taylor series
about the pump frequency xp and keeping the rst few terms, the phase-matching
equation becomes
2 2
b2 X2 b X4 b6 X6 BcP 0; 12:24
4! 4 6!
n
where bn ddxkn is the nth-order dispersion parameter of an optical bre and
xp
X xs xp xp xi . When the signal and idler are polarised orthogonally to
the pump Db includes two contributions: phase mismatch dbD induced by dis-
persion and phase mismatch dbB induced by modal birefringence such that
dbD dbB BcP 0 12:25
It can be seen that phase-matching in optical bres is determined by the chromatic

and polarisation dispersion of the bre mode. Using (12.23), (12.24) and (12.25),
phase-matching curves calculated for a weakly birefringent PCF are shown in
Fig. 12.15 [75]. There are several features in this plot: (1) phase-matching can be
achieved when the generated photons are either co-polarised or cross-polarised to
the pump photons; (2) for the co-polarisation phase-matching (ff!ff or ss!ss), the
wavelength of generated photons is very close to that of pump photons when
pumping in the anomalous dispersion regime, while the wavelength of generated
photons is well separated from that of pump photons when pumping in the normal
dispersion regime; (3) for the cross-polarisation phase-matching (ss!ff or ff!ss),
there is a zero slope region of the curves, which means the bandwidth of generated
photons is independent of the pump bandwidth; (4) the phase-matching for ss!ff
only occurs in the normal dispersion regime, while for ff!ss it occurs in both
normal and anomalous dispersion regimes and the zero slope region is in the
anomalous dispersion regime. These features have found many applications in
single and entangled photon generation.
1.8
ff ss
Signal and idler (mm)
1.6 Normal
dispersion
1.4
1.2
ss ss
Anomalous
ss ff dispersion
1 ff ff
0.8
zero slope region
0.6
0.4
0.66 0.68 0.7 0.72 0.74 0.76 0.78 0.8 0.82 0.84
Pump wavelength (mm)
Fig. 12.15 Calculated co-polarisation and cross-polarisation phase-matching curves for a weakly
birefringent PCF. ss!ss and ff!ff mean that both the signal and idler photons are in the same
slow/fast polarisation mode as the pump. ss!ff and ff!ss mean that both the signal and idler
photons are in the polarisation orthogonal to the pump [75]
Initial attempts at photon generation in optical bres concentrated on phase-

matching in the anomalous dispersion regime because bre design and fabrication
technology was not available to introduce enough higher-order dispersion b4 and b6
to compensate the phase-mismatching induced by the positive b2 term for normal
dispersion [7679]. As can be seen from Fig. 12.15, the wavelengths of the gen-
erated photons are very close to that of the pump photons. Since silica has a broad
Raman gain spectrum and many uncorrelated photons close to the pump wave-
length are also generated from spontaneous Raman scattering (SpRS), this intro-
duces a high level of noise to the photon sources. To reduce the SpRS noise, bres
have to be cooled using liquid nitrogen or helium [80, 81], which requires cryostatic
conditions and is undesirable for many applications.
Breakthroughs in this regard were made as PCF fabrication technology matured.
Compared with standard step-index bres, PCFs have periodically arranged air
holes in the cladding, which signicantly increase the index difference between the
bre core and cladding and thus introduce relatively large higher-order dispersion
over a broad wavelength range. This makes phase-matching possible when
pumping in the normal dispersion regime and many demonstrations have shown
low-noise single or entangled photon generation at room temperature in this regime
[38, 50, 8285]. In these demonstrations, the frequencies of generated photons are
typically over 70 THz away from that of pump. This is far beyond the Raman gain
band of silica and therefore the generated photons are uncontaminated by the SpRS
noise. The structure of a PCF is so flexible that a weak birefringence can be
introduced by making two holes symmetrically around the bre core slightly
smaller than the other holes. This is particularly interesting because cross-polari-
sation phase-matching can be achieved in such a bre, as shown in Fig. 12.15, and
the generated photons are naturally narrow-band without the use of narrow-band
lters [84, 85]. This signicantly reduces the loss introduced by narrow-band lters
and thus greatly increases the source brightness. More importantly the heralded
photons are in a pure single-photon quantum state (see Sect. 12.2.5.3).
Hong-Ou-Mandel quantum interference of indistinguishable photons lies at the
heart of linear optical quantum computation [14, 17, 18]. In the pump frequency
degenerate SFWM process, the signal and idler photons appear at different wave-
lengths and are therefore distinguishable. To perform Hong-Ou-Mandel quantum
interference, two heralded single-photon sources based on such a SFWM process
are required [8385]. This involves four-photon coincidence events, which are
signicantly less likely to happen and so experiments are much longer and more
challenging than those involving two-photon coincident events. There have been
previous demonstrations of pump frequency non-degenerate SFWM to generate
photons at the mean frequency of two pump frequencies in optical bres [86, 87].
Since pulsed pumps are required for efcient SFWM in optical bres, when the two
pulsed pumps are used, they must be synchronised temporally so that they can
interact. Fan et al. used a pump frequency degenerate four-wave mixing to generate
the two pump pulses [86], while Chen et al. spectrally sliced a femtosecond laser to
form the two pump pulses [87].
12.2.4.2 SFWM in Integrated Waveguides
Optical bre based photon sources are naturally compatible with current bre
communication systems and therefore they are very useful for implementing
quantum key distribution for secure communication. On the other hand, photon
sources based on photonic chips are desired for quantum computation because
scalability and miniaturisation are essential for this particular application. We next
describe single or entangled photon generation via SFWM in direct write silica
waveguides [42], chalcogenide ridge waveguides and nanowires [8890], and sil-
icon nanowires [9195].
The integrated platform most similar to optical bre is the femtosecond laser
direct write waveguide. The approach to generating heralded single photons in
direct write waveguides is similar to that in optical bres described in Sect. 12.2.4.1
as the waveguiding properties are rather similar. A 4 cm long waveguide was
fabricated by femtosecond laser writing in an undoped silica chip [42]. An elliptical
transverse mode was produced to introduce a birefringence so that cross-polarisa-
tion phase-matching similar to that in Fig. 12.15 was achieved in the zero slope
region. The single-photon purity (see Sect. 12.2.5.2) was measured to be 0.86 from
the source number statistics without narrow spectral ltering [42]. Although the
intrinsic nonlinearity of silica is very low, the low-loss feature of silica waveguides
and the ease of coupling generated photons to silica bres make this platform
promising.
Compared with silica, chalcogenide glasses have up to 1000 times higher
nonlinear refractive index and can be made into compact waveguides [96]. In
particular the waveguide dispersion can be engineered so that the waveguides
exhibit anomalous dispersion in the telecommunication 1550 nm band and these
glasses exhibit very low values of nonlinear losses such as two-photon absorption
(TPA) and free-carrier absorption (FCA). These features have found many appli-
cations in nonlinear optical signal processing [96]. In 2011, chalcogenide As2 S3
ridge waveguides were used for quantum-correlated photon pair generation [88].
Similar to the phase-matching in the anomalous dispersion regime of silica bres,
very high SpRS noise was observed in this As2 S3 -based photon source. Fortunately
As2 S3 glass has a low-Raman gain window at 7.4 THz detuning and this was
subsequently used for low-Raman noise photon generation [89]. Another chalco-
genide glass, namely Ge11:5 As24 Se64:5 , exhibits more than 10 times higher non-
linearity than As2 S3 and can be made into more compact photonic nanowires rather
than ridge waveguides [90]. The Ge11:5 As24 Se64:5 glass also has a low-Raman gain
window, but at 5.1 THz, which has allowed low Raman-noise photon generation
[90]. Due to the large index contrast between chalcogenide glasses (n = 2.4) and
silica (n = 1.5), the coupling between chalcogenide waveguides and silica bres
remains a challenge.
Silicon nano-photonic waveguides are the most widely used platform for single
and entangled photon generation via SFWM because of the mature silicon nano-
fabrication technology and unique optical properties of silicon [9195]. There are two
forms of silicon: amorphous and crystalline silicon, both exhibiting as high a
nonlinearity as the chalcogenide glasses. Amorphous silicon has a broadband Raman

gain spectrum and shows very low TPA and FCA losses. Crystalline silicon has a
very narrow Raman gain band (<1 THz) with its peak at a frequency detuning of more
than 15 THz. Therefore photon sources based on crystalline silicon [9194] have a
much higher quantum signal-to-noise ratio than those based on amorphous silicon
[95]. In the following sections, we focus on crystalline silicon unless stated otherwise.
Silicon can be made into nanowires [9194] and resonant structures including
micro-rings [93, 97], photonic crystal waveguides [98], coupled resonator optical
waveguides (CROWs) [41, 99], and micro-disks [100]. We address silicon
nanowires in this section and silicon resonant structures in next section. A typical
silicon nanowires cross section is 450 nm wide and 220 nm high. Such a small area
results in an extremely high nonlinearity and anomalous dispersion in the tele-
communication 1550 nm band. This enables efcient SFWM for photon generation
in waveguide devices of only a few mm [9194], which is much smaller than the
platforms based on other materials such as silica [42]. More importantly, silicon
fabrication techniques allow the addition of inverse tapers at both input and output
ends of a waveguide for bre pigtailing, which signicantly improve the coupling
efciency and stability [94].
12.2.4.3 SFWM in Resonant Structures
To further reduce the size of a nonlinear device while maintaining the SFWM
efciency without increasing the pump intensity, resonant structures are the natural
solution. Here we discuss the use of resonant structures, such as micro-rings [93,
97], photonic crystal waveguides [98, 101, 102], CROWs [41, 99] and micro-disks
[100] to enhance SFWM.
In a micro-ring or a micro-disk structure photons can travel along a closed loop
for many orbits and thus the nonlinear interaction between photons and the device
can be enhanced. This requires that all three photon frequencies satisfy the reso-
nance condition ni L mki , where ni is the effective mode index of the micro-ring
or micro-disk at vacuum wavelength ki , L is the circumference, c is the speed of
light in vacuum, and m is a different positive integer for pump, signal and idler.
A typical micro-ring or micro-disk structure is made from silicon and shown in
Fig. 12.16. The structure has a ring or disk and a bus waveguide. In a micro-ring
structure, the ring can be either a perfect circle as shown in Fig. 12.16a [97] or a
racetrack cavity [93]. Both the ring and bus waveguides are dispersion-engineered
silicon nanowires. The gap between the bus and ring waveguides can be optimised
so that maximum pump power is stored in the ring enhancing the nonlinear
interaction. This optimised gap is called the critical-coupling distance. In summary,
the nanowire dimension determines the basic SFWM properties such as phase-
matching bandwidth, and the ring circumference together with the ring-bus gap
determines the SFWM enhancement. Photons generated in these devices can have
high signal-to-noise values, showing CAR values over 600 [51], albeit after
sweeping out the free carriers generated through two-photon absorption.
(a) Micro-ring (b) Micro-disk
r r
Pump Idler Idler
Signal Signal
Fig. 12.16 A typical micro-ring (a) and micro-disk (b) structure for photon generation via
enhanced SFWM. Three frequencies: idler, pump, signal are all resonant in the structure, and are
equally spaced by the FSR
Similar to a micro-ring, a micro-disk can resonantly enhance the SFWM ef-

ciency by the resonant effect given optimisation of the disk circumference and disk-
bus waveguide gap. However a micro-disk differs from a micro-ring in several
aspects. Unlike a silicon ring that sits on a silica substrate, a silicon disk sits on a
silica pedestal. This means that the cladding is air, so that the index contrast is xed
and signicantly larger than that between silicon and silica. Therefore the disper-
sion is determined largely by the disk thickness. For a disk thickness of 260 nm, the
zero-dispersion wavelength is shifted to the telecommunication band around
1.55 m [100]. To access a micro-disk, a tapered optical bre is used as the bus
waveguide to couple light in and out through evanescent coupling [100]. Since both
ends of a tapered bre are just single mode bre, the input and output coupling
efciency can be very high, resulting in unprecedented spectral brightness [100].
Taking advantage of the resonantly enhanced SFWM, the pump power required for
photon generation in micro-rings and disks can be as low as tens of W [97, 100],
which is ve orders of magnitude lower than that required for silicon nanowires
[91, 92]. Another feature of micro-rings and disks is that the resonance bandwidth is
extremely narrow, so the resulting heralded photons can be obtained in a spectrally
pure quantum state without the use of narrow-band lters. As with ring resonators,
these devices can show very high CAR values, accentuated by the low-loss tapered
coupling technique, reaching values over 1000 [100].
Compared with silicon micro-rings and disks, a silicon photonic crystal wave-
guide (PhCW) lacks a cavity, but nevertheless is still able to increase interaction
times compared to straight waveguides. A PhCW has a triangular lattice of air holes
etched in a suspended silicon membrane with a row of holes missing along the CK
direction, as shown in Fig. 12.17a [98]. The row without holes acts as the wave-
guide core and the air-hole lattice as the cladding. The photonic crystal cladding
forms a photonic bandgap for guiding light. Near the bandgap edge, the dispersion
can be very large and the group index ng can be much greater than the native
material refractive index n0 .
This slow-light effect can be characterised by the slow-down factor
S ng =n0 . For efcient SFWM, the group index ng should be nearly constant
across a broad bandwidth. This is realised by laterally shifting the rst row of holes
on both sides of the waveguide core away from the core by tens of nanometres
[103]. By adjusting the lattice period, hole diameter and the shift of the rst row of
(a) Spatially compressed

(b)
Input pump pulse
S = ngs/ng0 = 10 Output
ngs
ng0
p
Fast light i p s
Slow light
Fig. 12.17 a Schematic of SFWM in a silicon slow-light PhC waveguide. b Group index and total
transmission of light in a 96 m long silicon PhC waveguide. The window (between dotted lines)
with a flat group index of 30 and slightly increased loss denes the slow-light regime. The pump,
signal, and idler bands are represented by green (middle), blue (left), and red (right) lines,
respectively
holes, both ng and the slow-light bandwidth can be varied [103, 104]. Figure 12.17b
shows the measured group index and transmission for a 96 m long silicon PhCW
with a lattice period of 404 nm, hole radius of 115 nm and rst row shift of 50 nm
[98]. The wavelength window (between dotted lines) with a flat group index of
around 30 and slightly increased loss denes the slow-light regime. The low loss is
achieved by silicon access waveguides which incorporate inverse tapers terminated
by wide polymer waveguides at both the input and output of the PhCW region.
Photon generation through SFWM in such a silicon PhCW structure with ng
30 is illustrated in Fig. 12.17a. When a coherent pulse of light enters the PhCW
region, the pump travels at a group velocity of c=30 and is strongly spatially
compressed by the slow-light effect. Compared with silicon nanowires, this effec-
tively enhances the nonlinear interaction per unit length per unit pump power, c,
approximately two orders of magnitude, and in turn the SFWM efciency, which is
proportional to c2 , by four orders of magnitude [105]. After accounting for changes
in losses and mode area, this enhancement enabled single photon generation in a
100 times shorter nonlinear device than silicon nanowires operated at similar pump
power [98], which now often achieve a CAR in excess of 200. It is well known that
nonlinear losses such as two-photon absorption (TPA) and TPA-induced free-car-
rier absorption (FCA) occurs at around 1550 nm in silicon when the optical power
reaches a certain level. These effects are also enhanced by slow light in a silicon
PhCW and have been observed as a limiting factor for photon generation when
operating at high power [102]. However, if the PhCW is made from a larger
bandgap material such as GaInP, the nonlinear losses in photon generation can be
eliminated [101]. The challenge with GaInP is to further improve fabrication
technology to reduce waveguide propagation losses and therein improve the CAR
from the previously reported value of 30.
Finally we consider photon pair generation in coupled resonator optical wave-
guides (CROWs), which consist of a sequence of coupled resonators with high quality
factors [106]. Each individual resonator forms a unit cell of the CROW and can be a
micro-ring [41], a micro-disk or a photonic crystal cavity (PhCC) [99]. Figure 12.18
shows an image of a micro-ring CROW and a schematic of a typical PhCC CROW.
Fig. 12.18 a Micro-ring CROW [41], and b PhCC CROW [99]. The micro-ring CROW is
accessed by a silicon bus waveguide, while the PhCC CROW is accessed by a standard PhCW
Light propagates through a micro-ring CROW via evanescent-eld coupling between

adjacent rings, and in a PhCC CROW through evanescent Bloch waves [106]. In both
cases the separation between two adjacent resonators is sufciently large that the
resonators are weakly coupled. Consequently, the eigenmode of the electromagnetic
eld in such a CROW structure is a linear combination of the high-Q mode of the
individual resonators and forms a collectively resonant supermode.
Compared with a single micro-ring, a micro-ring CROW exhibits much larger
group index and broader resonant bandwidth. In a heralded single photon genera-
tion experiment, a CROW consisting of N rings can offer an N2 enhancement in the
generation rate [41, 107]. At the same time, the wider passbands have the benets
of ease of experimental alignment and increased thermal stability. Compared with a
single PhCC, a PhCC CROW shows a broadband transmission rather than multiple
narrow band resonant peaks [99]. The transmission band behaves like a slow-light
band in a standard PhCW, but with smaller area and greater group index. The
nonlinear coefcient of a PhCC CROW can be as high as 13,000/Wm. This can
signicantly enhance the SFWM efciency. The limiting factor of a PhCC CROW
is that the slow-light bandwidth is narrower than a standard slow-light PhCW,
making the isolation of generated photons from the pump challenging, which thus
far have restricted CAR values to be below 10.
12.2.5 Quantum Description of Photon Pair States
Having explored a range of experimental structures for photon generation, we now

return to the development of the formal quantum description of pair states produced
by the SPDC or SFWM processes.
12.2.5.1 The Bi-photon Wave Function
We begin by constructing a more complete description of the quantum state that is

produced in spontaneous pair sources. Here we restrict ourselves to the discussion
of integrated sources which, as they use a collinear conguration, can be repre-
sented in the frequency domain, rather than working on the full wavevector version
described in (12.20). Ignoring multi-pair and parasitic effects such as spontaneous
Raman generation or scattering loss, the full state generated per pump pulse in a
waveguide structure in the undepleted pump approximation is [47, 108110]
0 1
E Z1 Z1

w m X y
y â H:cAjvaci;
gen exp@p dx1 dx2 /ab x1 ; x2 âax 1 bx2
2 a;b
0 0
12:26
where H.c. denotes the Hermitian conjugate. This complicated expression can be
seen as a continuous frequency multi-mode squeezed vacuum. In the limit of a low
probability of pair production, jmj 1, the full expression may be approximated by
E
w jvaci mjwII i; 12:27
gen
where the photon pair state
Z1 Z1
1 X y ây jvaci:
jwII i p dx1 dx2 /ab x1 ; x2 âax 1 bx2
12:28
2 a;b
0 0
Here Greek subscripts indicate the mode type of each photon (e.g. lowest order TE-
like mode), and /ab x1 ; x2 /ba x2 ; x1 is known as the bi-photon wave
function or joint spectral amplitude for the photon pair in modes and .
Provided we normalise the components of the bi-photon wavefunction according to
XZ Z1
1
2
dx1 dx2 /ab x1 ; x2 1; 12:29
a;b
0 0
then, in this limit, jmj2 is just the average number of pairs generated per pulse.
Note that (12.28) applies for both SPDC and SFWM. The differences in the
outputs of these processes sits in /ab , along with issues such as phase-matching and
power dependence. Following terminology of classical nonlinear optics, a type-I
process creates photon pairs in which both photons are of the same mode type and
is represented by a single aa term in the sums above. A type-II process creates
photon pairs in which each photon is of a different type and thus has an ab as well
as a ba term. An additional possibility involves the use of two type-I processes back
to back in a sandwich conguration [111] to produce an as well as a term.

Such schemes are useful for making polarisation-entangled states.
Also note that in (12.26) the integrals are over the frequencies of the generated
photon pair rather than the wavevectors. The latter is normally required to treat bulk
systems such as those discussed in Sect. 12.2.3.1. In waveguides, the requirement
of transverse wavevector conservation is eliminated by the spatial connement, and
the frequency representation is typically more convenient. Nevertheless, the general
theory that follows holds in the bulk case, with suitable transformations to a
wavevector expansion.
12.2.5.2 Frequency Correlations in Bi-photon States
The importance of the bi-photon wave function comes in understanding the fre-
quency correlations between the signal and idler photons. The implication of
(12.28) is that except for certain separable states (see Sect. 12.2.5.3), the two
photons are frequency-entangled and it is not meaningful to talk of the frequency
content of one or the other. Such a description is vital in quantum engineering, since
different quantum information applications demand different correlations. If the
/ab x1 ; x2 for all and in the sum of (12.28) are equal, then there is no possible
frequency measurement that can determine the mode type of either photon. For
photon pair states in modal superpositions (i.e. with more than one term in the sum
of (12.28)), such equality implies that the photons are maximally entangled in mode
type [112], and enables maximal violation of Bell-like inequalities [113, 114]. On
the other hand, high delity quantum computation tasks may be spoiled by any sort
of frequency correlation [115]. Thus, it is useful to be able to quantify the degree of
spectral correlations.
The most common measure for quantifying the degree of frequency correlations
is through a Schmidt decomposition of the bi-photon wave function [116]. As a
complex function of two variables, it can always be expressed in the form
p X p
Nab /ab x1 ; x2 pj Uaj x1 Ubj x2 ; 12:30
j
p R1 2
where Nab is a normalisation factor such that Nab 0 dx1 dx2 /ab x1 ; x2 1.
In addition, the complex Schmidt functions Uaj x and UP bj x form complete
orthonormal sets, the weighting coefcients pj are real, and j pj 1. If there is
only one nonzero Schmidt coefcient p0 , i.e. p0 1 with pj 0; 8j 6 0, then the
bi-photon wave function contains no spectral correlations and the state is termed
separable. If there is more than one Schmidt coefcient, the Schmidt number [116]
X1
K ; 12:31
j
p2j
2 2
K=12.5 K=8.8
1 1
Fig. 12.19 Example bi-photon wave functions and their Schmidt numbers
quanties the effective number of Schmidt modes involved in the decomposition.

Note that K > 1 when the bi-photon wave function contains spectral correlations,
and reaches a minimum of 1 when it contains no spectral correlations. It is also the
inverse of the heralded single photon spectral purity, in the limit of negligible
multipair production. By way of example, we consider two real-valued bi-photon
wave functions in Fig. 12.19, and note how K is reduced the more /ab x1 ; x2 is
aligned with the x1 x2 axes. Energy conservation and phase matching tend to
produce a bi-photon wave function that is rather narrow and runs from the top left
to the bottom right of x1 x2 space, but clever pump and source engineering can
change this and thus reduce K, as seen in the next section.
12.2.5.3 The Factorability of Bi-photon States
Separable bi-photon states are not easily produced, but they play a special role in
the design of photon sources. When a herald photon of a pair is detected, the
remaining photon is represented by a single photon state. However in general, if the
heralding photon is observed by a broadband detector, the single photon state that
remains is not represented by the ideal pure state form in (12.11). Instead, it is
projected onto a quantum mixed state that accounts for the classical ignorance of the
detected photons wavelength. In the quantum density operator formalism, we nd
the heralded state by tracing over the frequency expansion of the detected photon as
Z1

â
q ^ x b
dx xb q 12:32
0
where the photon pair density operator is

Z1

D

^2
q dx1 dx2 dx01 dx02 /ab x1 ; x2 /ab x01 ; x02 x1a ; x2b x01a ; x02b : 12:33
0
y
Here we have dened x1a ; x2b jx1a ix2b where jxh i âhx jvaci, and
p R 1
assumed our initial bi-photon state to be jwi 2 0 dx1 dx2 /ab x1 ; x2

x1a ; x2b for convenience. Performing the trace in (12.32) gives
Z1
â 2
q dx1 dx2 dx /ab x1 ; x/ab x2 ; xjx1a ihx2a j: 12:34
0
If the bi-photon wave function can be written in the separable form

p
2/ab x; x0 f xgx0 then (12.34) collapses to the pure state q
â jwa ihwa j
with
Z1
jwa i dxf xjxa i; 12:35
0
which has the form of (12.11). However, in general this is not possible and the
heralded photon occupies a mixed state. This is a practical problem, since most
quantum protocols involving single photons assume pure state inputs.
There are two approaches to resolve the problem. First, one can simply lter the
outgoing photons strongly. Filters with bandwidth comparable to the minor axis
length of the bi-photon wave function will generally produce near-pure states, but at
the cost of discarding most of the generated pairs. Alternatively, one can attempt to
engineer the bi-photon state produced by the source as follows.
In waveguide sources, the shape of the bi-photon wave function in the x1 x2
plane is largely governed by just two things: the spectral shape of the incident pump
pulse, and the dispersion and nonlinearity of the waveguide. In v2 structures this is
often expressed as
2 2
/ab x1 ; x2 /P x1 x2 Fab x1 ; x2 ; 12:36
2
where /P x is the pump spectral amplitude and Fab x1 ; x2 is determined by
material properties and the waveguide geometry. Similarly, in v3 structures we can
write
xZ
1 x2
3 3
/ab x1 ; x2 dx/P x/P x1 x2 xFab x1 ; x2 ; x; 12:37
0
3
where again Fab x1 ; x2 ; x describes the waveguide materials and geometry.
Thus, many attempts to produce frequency uncorrelated states (that is states with a
Schmidt number K = 1), have focused on pump pulse shaping and waveguide
property engineering.
Researchers have examined the group velocity mismatch, crystal length, and
pump bandwidth conditions necessary to achieve factorable bi-photon wave func-
tions created via type-II SPDC in bulk crystals [117], as well as via SFWM in
photonic crystal bre [118]. These studies were experimentally veried in [33] and
[119] respectively. More exotically, pump pulse shaping [120], and engineering of
the strength of an SPDC waveguide nonlinearity itself [69], to achieve factorability
have also been investigated, in addition to resonant effects during SFWM in mic-
roring resonators [121]. There is more than one way to skin Schrdingers cat.
12.2.5.4 Measuring the Joint Spectral Intensity
Although the bi-photon wave function is quite an important quantity, it has proven
difcult to measure directly. Nonetheless, there are a number of experimental
techniques available to measure a related quantity, the joint spectral intensity (JSI)
of generated photon pairs, dened as j/ab x1 ; x2 j2 . One of the simplest methods
is to separate the signal and idler photons at the output of the generation device and
scan a tunable lter in both the signal and idler arms. At each combined position
{xs , xi } one measures the number of coincident arrivals of photons, as shown
schematically in Fig. 12.20a. By plotting these values as xs versus xi one can
create a contour plot of the JSI. As previously discussed the shape of the JSI implies
different amounts of entanglement in the system and therefore conveys the purity
with which photons can be heralded. This ltering technique has been used both
manually [33, 42, 122] with a low resolution and more recently using an automated
liquid-crystal-on-silicon (LCOS) tunable lter [123] with a high resolution. An
example of a JSI measured using this method from a chalcogenide bre photon pair
source is shown in Fig. 12.21. The LCOS device can also be used to create arbitrary
lter functions which allows the use of different measurement acquisition types,
including Hadamard coding [124], which can reduce the measurement time
required to reconstruct a full JSI.
Another method of measuring the JSI replaces the tunable lters in the signal
and idler arms with dispersive media, such as long lengths of dispersive optical
bres, and detectors with a high resolution photon arrival time measurement [125],
as shown in Fig. 12.20b. As the photon propagates through the dispersive medium
it becomes elongated in time, with the time of arrival directly related to the fre-
quency of the input photon. The coincidences arrive with random timing infor-
mation, meaning the JSI is built up randomly over the total acquisition time. This
has the added advantage of having no active tuning or moving parts, however also
means that more complicated ltering functions are not available to reduce the total
integrated time required to characterise the JSI.
(a)
s
From Photon
Pair Source
&
i
(b)
s
From Photon
&
Pair Source
i
Fig. 12.20 Schematic of JSI measurement methods using a tunable optical ltering and
b dispersive media and photon arrival time measurements
Fig. 12.21 The measured JSI -0.47

from a chalcogenide bre
photon pair source measured
with a high resolution
automated method [123]
-0.58
-0.69
-0.80
-0.91
0.49 0.60 0.71 0.82 0.93
More recently there have been theoretical proposals for reconstructing the JSI
using only classical measurements in a process known as stimulated emission
tomography (SET) [126]. This method uses the same pumping characteristics as are
used in the standard coincidence measurements above, however here a weak probe
is also injected to perform the stimulated analogue of the spontaneous process, such
as difference frequency generation for SPDC and stimulated four-wave mixing for
SFWM. By scanning the probe and measuring the generated signal using an optical
spectrum analyser it is possible to reconstruct the JSI with a high resolution and
over a much shorter integration time as no coincidence measurements are required,
shown very recently for SPDC [127].
12.2.5.5 Second-Order Correlation Functions
There are a number of other useful means of characterising the state that provide
complementary information to direct measurement of the JSI or the full phase-
sensitive JSA. In particular, indirect methods exist to measure its properties based
on correlation functions [128].
Measurements of normalised correlation functions serve to quantify the coher-
ence of the electromagnetic eld produced by a particular source. While a rst
introduction to coherence generally deals with intensity fringes from the superpo-
sition of elds such as in a Michelson interferometer, this captures only rst-order
coherence. In fact important information can be extracted from the second order
correlation function and higher order coherences introduced by Glauber [129, 130],
following the invention of intensity interferometry by Hanbury Brown and Twiss
[131].
The classical second-order correlation function
hE r1 ; t1 E r2 ; t2 E r1 ; t1 E r2 ; t2 i
g2 r1 ; t1 ; r2 ; t2 ; 12:38
hE r1 ; t1 E r1 ; t1 ihE r2 ; t2 E r2 ; t2 i
is the intensity analogue of the rst order function, and in the limit of stationary
elds it depends only the difference in time s t2 t1 jr2 r1 j=c. Switching to
a simple quantum mechanical picture and replacing classical elds by corre-
sponding Heisenberg eld operators, g2 s can be written in terms of the number
of counts on a photon detector, which is proportional to the intensity, as
h^nA t^nB t si
g2 s ; 12:39
h^nA tih^nB t si
D y E
where h i now denotes a temporal average. Here nÂ;B t âA;B tâA;B t is
the average number of clicks on photo detector A or B over some integration time
T. While derived for stationary elds, it turns out that (12.39) also holds for pulsed
sources provided T encompasses a single pulse width and s is an integer number of
temporal pulse spacings. Written this way, g2 s can be interpreted as simply the
likelihood of detecting a photon at detector A in coincidence with the detection of a
second photon delayed by time s at detector B. As such, the second-order corre-
lation function is usually measured with a Hanbury Brown and Twiss (HBT) setup
of the form sketched in Fig. 12.22a.
(a) SPD (b)

H
2
50:50 SPD
A
g (2)(x)
1
B
Delay 0
SPD (x) & -1 -2 0 +1 +2
x
Fig. 12.22 a Illustration of a single photon HBT experiment for measuring g2 x. Idler photons
are detected to herald signal photons which are incident on a 50:50 beam splitter and are either
transmitted for reflected in two arms A and B. Single photon detectors (SPD) detect the photons
and a coincidence counter correlates photon arrival times. An electronic delay can shift the signal
from detector B by a pulse period, x. b Experimental g2 x for a heralded single photon source.
For x = 0, g2 x ! 0 showing non-classical statistics, while for x 6 0, g2 x
1, showing the
recovery of uncorrelated Poissonian statistics
Figure 12.22b shows a typical g2 s trace where s, the optical delay in one arm,
is varied while counting coincidences from two single photon detectors (SPDs) A
and B. When the path length is equal, g2 0 ! 0 for a single photon source. For a
coherent source, the spacing between each photon in time is random and follows
Poissonian statistics, resulting in g2 0 1. For a chaotic/thermal light source
g2 0 2, which can be interpreted as a signature of photon bunching. The HBT
experiment in a photon picture gives a direct measurement of the g2 s of a light
source.
As mentioned above g2 s is just one of many possible correlation functions.
To write them in a convenient form for calculations involving multimode squeezed
states, we rst note that (12.30) allows us to write a generated multimode squeezed
state as
Z Z !
E m X 1 1
w y y
gen exp p dx1 dx2 /ab x1 ; x2 âax1 âbx2 H:c jvaci
2 a;b 0 0
!
m X p ^ y ^ y
exp p pj Aaj Abj H:c jvaci
2 a;b;j
12:40
where
Z1
^y
A dxUabj x^
y
aabx : 12:41
abj
0
While the analysis proceeds similarly for many different states, we focus here on
the state produced by a type-II process
!
E X p y y
w exp m ^ A^
gen pj A aj bj H:c jvaci: 12:42
j
We note that this state contains an average number of

D E X
p
^ h wgen
l wgen N sinh2 jmj pj ; 12:43
j
P y
^h ^ ^
photon pairs, where N j Ahj Ahj and h a or b. Furthermore, for factorable bi-
photon wave functions with K = 1, (12.42) represents a two-mode squeezed vacuum
with a probability of n pairs expressed by (12.15). On the other hand, when K is so
large that there are a very large number of Schmidt coefcients, all equal, and not
more than one pair is produced in each Schmidt mode, the probability of n pairs is
instead expressed by (12.13). Using (12.42) we may write zero-delay, s 0,
unheralded auto-correlation functions as [128]
D E
^ n : wgen
wgen : N h
n
gUnheralded D En ; 12:44
wgen N ^ h wgen
where the operators between the colons are normal ordered. Similarly, we may
write zero-delay cross-correlation functions as [128]
D E
wgen : N ^ n : wgen
^ am :: N
b
m;n
gCross D Em D En : 12:45
^ a wgen
wgen N wgen N^ b wgen
2
In the low probability of pair production limit, we will see that gUnheralded can
1;1
give information regarding the Schmidt number, K, and that gCross can give
information regarding the average number of generated photons per pump pulse.
Following a bit of algebra, it can be shown that
P
p
2 sinh4 jmj pj
j
gUnheralded 1 hP
pi2 ; 12:46
j sinh 2
jmj pj
and
P
p
1;1 sinh4 jmj pj
j 1
gCross 1 hP
i2 P
p : 12:47
p sinh jmj pj
2
j sinh 2
jm j p j j
The low pairD production

probability
E limit implies that the number of photons
^ ^
produced, here wgen Na Nb wgen 1; and thus jmj2 1. In this limit
2
X 1
gUnheralded 1 p2j 1 ; 12:48
j
K
and
1;1
X 1 2 1
gCross 1 p2j gUnheralded D E: 12:49
j j mj 2 wgen Nh wgen
^
as initially suggested. Finally, the heralded auto-correlation function

R1
^2 : q
2 dxhxh j : N h ^ h jx h i
gHeralded
0R 1 2 ; 12:50
dx hx h j ^
N h ^
q h j x h i
0
can easily be shown to be zero, if a number-resolving scheme is used to ensure that

only a single a or b photon has been detected to herald the presence of its partner
and qh takes the form of (12.33). Otherwise, it will be larger, as we seen in the
experiments discussed throughout this chapter.
12.2.6 Multiplexing: A Route to On-demand Photons
In Sect. 12.2.1.2 we saw that all existing approaches to producing single photons
have technical issues, making it very difcult to obtain a single photon in a truly on-
demand fashion. This is a fundamental challenge in quantum optics. The non-
deterministic nature of photon generation limits the feasibility of next-generation
quantum processors that will require multiple simultaneous single photon inputs. In
particular we noted that both attenuated laser sources and heralded pair sources emit
a portion of their light as multiple photons, with very similar photon statistics (see
(12.13) and (12.15)). At rst sight then, the heralded sources have no clear
advantage over attenuated lasers, other than perhaps the ability to gate single
photon detectors using the idler detection.
In fact however, the concept of photon multiplexing can break the intrinsic limits
on the photon statistics expressed by the inverse relationship of the CAR on the
mean photon rate (see (12.17)). In this section, we explain these approaches and
discuss the demonstrations to date.
12.2.6.1 Spatial Multiplexing
Single photons generated by parametric processes have an advantage over attenu-

ated laser sourcesbecause a photon pair is generated, the detection of one photon
heralds the existence of the other. Using this knowledge, it is possible to deter-
ministically combine the output of several sources. This results in a higher single-
photon rate without a corresponding increase in multiple-photons. This occurs
because the rate of higher order pairs grows quadratically with the pair generation
probability l in each source. Distributing the mean rate across N channels means
the number of higher order pairs scales as Nl=N2 / 1=N for the same net single
pair rate. In principle then, the signal to noise level is multiplied by N.
Practically this is done by using the herald detections to control an N 1 low
loss optical switch, routing single-photons from multiple sources to a common
single mode output. This is shown in Fig. 12.23, where 8 nonlinear photon pair
sources (in this case silicon PhC waveguides) have their outputs combined by an
optical switch, triggered by a herald detection, to produce an enhanced single
photon output. The probability of a single photon reaching the output of the mul-
tiplexing system is then, in the limit of l 1
Fig. 12.23 Schematic illustration of integrated spatial multiplexing of 8 nonlinear photon pair
sources. In this example, the photonic crystal waveguides monolithically fabricated on a single
silicon chip are excited by a common pump [44]. PLZT refers to the switch material and is the
optoceramic lead lanthanum zirconium titanate

log2 N
PMUXOut 1 1 lgh N gdelay gcouple gfilter gswitch ; 12:51
where N (a power of 2) is the number of multiplexed sources, gh

gcouple gfilter gdetector is the efciency with which a heralding click occurs, gdelay is the
delay line efciency, gcouple the efciency of coupling the photons from the source
into the bre, gfilter the efciency of the ltering and gswitch the efciency of a single
switch. For large N and zero losses and assuming perfect detector efciency, the
probability tends towards an on-demand single photon source. In reality there will
always be loss and imperfect detection limiting the performance of any scheme.
Also for a larger number of sources, the probability of two or more sources gen-
erating a single pair increases, so there are diminishing returns beyond a certain
number of sources. The number of switches required to implement this scheme
scales linearly, with (N 1) switches needed to multiplex N sources.
This idea of spatial multiplexing was proposed by Migdall et al. in 2002 [132].
An alternate, potentially lower-loss, method using electro-optic polarisation con-
trollers and polarising beam splitters rather than 2 2 Mach-Zehnder interfero-
metric switches, was suggested in 2007 [133]. This was the scheme used in the rst
demonstration by Ma et al. [134]. Using a double pumped BBO crystal and free
space polarisation switches they achieved a four-fold increase in single photon rate
for xed g2 0 values, however the instability of bulk optics impeded scaling to
more components. Other publications have examined the limits of multiplexing [43]
nding that a deterministic source could be built from 17 ideal sources and photon
number resolving detectors, and the practical requirements of using such a source
for linear optical quantum computing [135], which found that some multiplexing
schemes are more efcient than others and there should be a viable route to an on-
demand multiplexed source which will be useful for some QIP protocols.
The rst demonstration of partially integrated spatial multiplexing was per-
formed by Collins et al. in 2013, using two photonic crystal waveguide photon
sources combined on a single silicon chip [44, 136] implementing the Migdall
scheme. In this paper the heralded photon rate was enhanced by 62.4 % for a xed
signal-to-noise ratio. Figure 12.24 shows a clear improvement to the count rate for
all CAR values (blue triangles) when compared to the two separate sources (red
squares and green circles).
Subsequently, Meany et al. demonstrated the multiplexing of four monolithic
periodically-poled lithium niobate waveguides which were butt-coupled to direct
laser-written silica wavelength division separation components [137]. This dem-
onstrated not only the integration of four sources, but also several of the key
components needed for multiplexing. In 2013 an alternate scheme with an asym-
metric architecture for multiplexing was proposed, and it is predicted that for certain
regimes this scheme will outperform the traditional symmetric switching architec-
ture [138].
40
30
CAR
20
10
0
1.0 2.0 3.0 4.0 5.0 6.0
Heralded single photon rate (Hz)
Fig. 12.24 Plot to show the enhancement of multiplexing two sources together. The single source
heralded photon rates are plotted against the coincidence-to-accidental ratio for the two separate
sources (red squares and green circles). After multiplexing the sources together the same metrics
are measured (blue triangles) showing a clear enhancement of the heralded photon rate for all
CAR values, corresponding to an enhancement of 62.4 % [44]
12.2.6.2 Temporal Multiplexing
In the strategy described in the previous section, multiple spatially-separated

sources are combined using multiplexing. Each source has a single spatial mode,
and these modes are combined deterministically using an N 1 optical switch. We
now turn to temporal multiplexing where one or more photon pair sources have
many temporal modes actively combined by routing the heralded photon to a
different length of delay line. Depending on the temporal mode in which a photon
pair is created, indicated by the timing of the herald detection, a different delay time
is applied. This idea was proposed theoretically in 2011 by Mower and Englund
[139], who envisaged the scheme being implemented with a sophisticated photonic
integrated circuit (PIC).
Figure 12.25 shows a sketch of the concept. A pulse train with period T pumps a
nonlinear device to produce photon pairs. The output is then divided, for this
example, into four time-bins ti , i 2 1; 4. A herald detection, in for example time-
bin t1 , triggers its partner single photon to be switched through 3T of relative
optical delay line moving it into time bin t4 . On the other hand, if a heralded
detection is made in time-bin t4 , zero relative delay is applied. The output has a
period now of 4T with 4 the probability of a single photon per clock cycle. A
possible setup for temporal multiplexing is shown in Fig. 12.26 with Mach-Zehnder
2 2 switches used to route the heralded photon though the 3T of relative delay.
The number of switches required to implement this scheme scales logarithmi-
cally, with log2 N 1 switches needed to multiplex N time-bins, equivalent to
t1 t2 t3 t4 t1 t2 t3 t4 t1 t2 t3 t4 time
T Pump pulses at period T
Signal photons are detected

Heralding the idlers time-bin
4T Idler delayed with period 4T
Fig. 12.25 Timing diagram of a four fold temporal multiplexing scheme
Single photon
t1 t2 t3 t4 detector
FPGA Delayed single
photon
Source AWG T MZ1 MZ2 MZ 3

T 2T Dump
Fig. 12.26 Schematic illustration of temporal multiplexing of four temporal modes using Mach-
Zehnder optical switches (MZi)
N sources. Temporal multiplexing is therefore much more resource efcient than

spatial schemes for a high number of multiplexed temporal modes. It is however
more challenging to implement, because switching triggered on temporal slots is
difcult, and to date there have been no experimental demonstrations. In the future
this idea will need to be extended to combine many more devices onto a single
optical chip. Undoubtedly the hybrid integration of several material platforms will
be required to realise all of the components necessary, including delay lines and fast
switches. Multiplexing will allow many single photons to be generated nearly on-
demand, enabling a new regime of complex quantum processing.
12.3 Nonlinear Optics for Quantum Frequency Conversion
As in classical communications, a range of controllable light sources is not the only

requirement for effective transfer of quantum information. Using the bre network
effectively requires the ability to transfer information from one wavelength band to
another, and this will increasingly be an important property of quantum
information.
A working denition for quantum frequency conversion (QFC) is the translation
of a quantum state of light from one frequency band to another. First discussed
explicitly in the literature nearly 25 years ago by Kumar and colleagues [140, 141]
in the context of making a tunable source of squeezed light, QFC may play an
important role in photonic quantum information processing networks as an interface
to link different components (operating at different wavelengths) together. One
application would be to map the emission from a single quantum emitter (e.g.,
InAs/GaAs quantum dots, single neutral alkali atoms, or nitrogen vacancy centers
in diamond) to the 1550 nm telecommunications band, for which single mode silica
optical bres show their lowest attenuation levels, and back to the original wave-
length for interaction with another single quantum emitter [142]. A second appli-
cation is in detection, by translating the wavelength of a quantum state for which
detection is currently difcult (e.g., due to limitations in detector quantum ef-
ciency, timing resolution, dark count rates, etc.) to one in which high performance
detectors have already been developed [143, 144]. More generally, complete and
efcient spectro-temporal control of quantum states of light would in principle
allow arbitrary quantum photonic components, individually optimised to achieve a
desired functionality, to be linked together as part of a high performance quantum
internet [145].
In this section, we briefly discuss the fundamentals of QFC in second- and third-
order nonlinear media, review the current status of frequency conversion devices,
and highlight recent experiments demonstrating frequency conversion of single
photon states of light. Finally, we discuss some perspective towards future direc-
tions with QFC, including recent progress using integrated device platforms and
proposals for achieving temporal wavepacket shaping using QFC.
12.3.1 Second-Order Nonlinear Media
Nonlinear media governed by the second-order electronic susceptibility v2 can be

used for sum- and difference-frequency generation, in which an input signal at
frequency xi is combined with a strong pump at frequency xp to generate light at
the sum and difference frequencies xo xi xp . Because optical nonlinearities
are weak, efcient conversion only occurs over an interaction region that is many
wavelengths in length. This places the requirement of phase-matching in a pre-
eminent position; that is, for efcient sum- or difference-frequency generation, the
wavevectors of the three elds must satisfy ko ki kp . Very much the same
phase-matching issues apply for frequency conversion as were discussed in
Sect. 12.2.3 for pair generation. Without phase matching, energy oscillates between
the three waves involved in the mixing process rather than being monotonically
transferred to the output eld, and the overall conversion efciency is low.
Improvements in conversion efciency were initially gained using birefringent
phase-matching, which takes advantage of the fact that orthogonally polarised light
beams often travel at different velocities within a birefringent material. By tuning
the angles of the input beams appropriately, phase matching can be achieved,
although the interaction regions may be limited in size, particularly if tight focusing
of the beams is used (which is often the case in order to achieve high intensities). As
a result, efcient conversion (on the order of tens of percent) is typically not
achieved in these systems without the application of extremely high pump powers.
A development that has increased the utility of frequency conversion devices and
been particularly benecial to subsequent QFC experiments is the use of quasi-
phase-matching in a waveguide-based geometry [146] (see Sect. 12.2.3.2).
Conversion efciencies (not including input/output coupling losses) in excess of
80 % have been achieved for both frequency upconversion and downconversion
[147, 148]. Periodically-poled lithium niobate (PPLN) waveguides are perhaps the
most common devices to have been used in QFC experiments to this point. The-
oretically, sum- and difference-frequency generation can be noise-free, suitable for
QFC all the way down to the level of single photon Fock states. In practice, one
might expect that scattering and fluorescence processes associated with the inter-
action of a strong pump eld with a solid material might lead to signicant noise
contributions. Raman scattering can be a dominant noise source if the pump and
input signal are within a couple hundred nanometres of each other [146]. Keeping
the pump on the long-wavelength side of the input signal [146] and increasing the
pump-signal separation to several hundred nanometres [147, 149] can increase
signal-to-noise levels to >100:1 (Fig. 12.27b). Spontaneous parametric downco-
version can also be of concern, and can be exacerbated due to fabrication errors in
the poling process [148].
The rst experimental demonstration of QFC using a non-classical input signal
was performed by Huang and Kumar [141] soon after Kumar introduced the
concept [140], with non-classical intensity correlations between a pair of elds
shown to be preserved after one of the elds was frequency converted. Over a
decade later, Tanzilli and colleagues [142] showed that entangled elds remain so
after one of the elds is frequency converted. This work using non-classical elds
came at approximately the same time that several groups began pushing the
(a) (b)
Fig. 12.27 a Schematic for quasi-phase-matching (QPM) in a v2 waveguide (here, sum-

frequency generation is shown). An input signal at frequency xi is combined with a strong pump
at frequency xp to generate an output signal at xo . Phase matching is achieved through periodic
poling, which generates a grating momentum 2p=K that compensates for wavevector mismatch.
b The strong pump eld can generate broadband Raman scattering in the QPM material. If this
Raman scattered light reaches the input signal frequency, it will be frequency converted along with
the signal and will thus be a limiting noise source. This source of noise can be limited by: (1)
Placing the input signal on the anti-Stokes (higher frequency) side of the pump; (2) Increasing the
pump-input signal separation as much as possible; (3) Narrowband spectral ltering around the
target conversion frequency
conversion efciency levels achievable in v2 nonlinear crystals towards 100 %, for

applications such as low light level detection of telecommunications-band photons
using visible wavelength single photon counters [143, 144, 146].
In recent years, a number of experiments have looked at frequency upconversion
of single photon Fock states using quasi-phase-matched v2 nonlinear crystals
(Fig. 12.28). For example, one experiment combined a 1300 nm triggered single
photon source based on an InAs/GaAs quantum dot with a 1550 nm pump in a
PPLN waveguide to generate 710 nm photons, with antibunched photon statistics
measured to explicitly conrm the single photon nature of the upconverted light
[150]. The utility of frequency conversion for studying the dynamics of telecom-
munications-band single quantum emitters was also highlighted, where conversion
to the visible improved the dynamic range of time-correlated single photon
counting measurements of the quantum dot excited state lifetime by a factor of 25,
due to the superior characteristics of Si single photon counters (sensitive at 710 nm)
relative to InGaAs devices (sensitive at 1300 nm). Subsequent work on upcon-
version of quantum dot single photon sources has included temporal shaping of the
photon wavepackets using an amplitude modulated 1550 nm pump [151] and
Before conversion (980 nm photons) After conversion (980 nm photons)

Photon correlation setup Photon correlation setup
Detector Detector
1 1
0 0
g(2)() g(2)()
Detector Detector
1.5 1.5
g (2)()
g (2)()
1.0 1.0
0.5 (2)
g (0) = 0.33 0.5 (2)
g (0) = 0.24
0 0
-100 -50 0 50 100 150 200 250 -100 -50 0 50 100 150 200 250
Delay Time (ns) Delay Time (ns)
Fig. 12.28 Photon statistics measured before and after frequency upconversion, as in [149]. (Top
left) 980 nm photons generated by a single InAs/GaAs quantum dot pumped with a continuous-
wave 780 nm laser are sent into a beamsplitter, and photons in each output port detected by a
single photon counter. A histogram of coincidence counts as a function of the difference in arrival
times between the two paths (bottom left) is generated, showing antibunching with a value
g2 0 0:33 0:03 see (12.38). (Top right) After frequency conversion to 600 nm using a
1550 nm pump in a PPLN waveguide, photons are directed into a beamsplitter, and photons in
each output port are detected by a single photon counter. The generated coincidence histogram
(Bottom right) shows anti-bunching (g2 0 0:24 0:04 \ 0:5), indicating that the single
photon nature of the input signal is preserved. The improved g2 value is likely due to the extra
spectral ltering provided by the quasi-phase-matching response of the PPLN waveguide. The
uncertainty in the g2 0 values is due to fluctuations in the detected photon count rate and
represents a one standard deviation value
experiments demonstrating that two-photon interference is preserved during fre-

quency conversion [149]. The latter work also considered the use of frequency
conversion to erase spectral distinguishability by converting the light from spec-
trally distinct sources to a common frequency. This work was quite analogous to an
experiment that preceded it by a few years [152], which converted signal and idler
photons from a correlated photon pair source to a common wavelength using sum
frequency generation in a pair of PPLN waveguides, and then demonstrated two-
photon interference of the now spectrally indistinguishable photons.
Frequency downconversion through difference frequency generation in a v2
medium has also been experimentally explored in the context of single photon input
elds. One experiment [153] used an entangled photon pair source and demon-
strated that entanglement between signal and idler photons was preserved after
downconversion of one of the two elds (essentially the downconversion equivalent
of the upconversion experiment of [142]), as well as explicitly showing that the
downconverted eld exhibited antibunched photon statistics (through a heralded
g2 measurement). Frequency downconversion has also been demonstrated with
triggered single photon sources based on a single semiconductor quantum dot, with
photon anti-bunching and coherence shown to be preserved [154] and downcon-
version to the 1550 nm band used as an integral part of a spin-photon entanglement
experiment [155].
12.3.2 Third-Order Nonlinear Media
Quantum frequency conversion using the third-order nonlinear susceptibility v3 is

achieved by the process of four-wave-mixing Bragg scattering [156], sometimes
referred to in the literature as wavelength exchange [157]. This is a non-degenerate
four-wave-mixing process in which two pump elds create an effective modulation
in the v3 nonlinearity, enabling the (in principle) noise-free translation of an input
signal to an output that is shifted by an amount equal to the difference in the two
pump frequencies (Fig. 12.29). The importance of using four-wave-mixing Bragg
scattering for QFC was highlighted by McKinstrie and colleagues [156]. They
outlined the noise considerations that arise when considering different four-wave-
mixing processes, including degenerately pumped processes, which can provide
wavelength conversion but will also be accompanied by spontaneous emission
noise (i.e., pump photons converted to the output wavelength band).
The dual pumps in four-wave-mixing Bragg scattering provide a great deal of
flexibility in the span over which frequency conversion can take place, with both
narrow and wide ranges for both upconversion and downconversion possible.
Figure 12.29b schematically shows some of the different congurations of pumps,
input signal, and output converted signal that can be utilised to achieve these
different functionalities. Interestingly, it should be pointed out that for each con-
guration, two possible idlers can be generated, as both are energy conserving.
(a) (b) Broadband downconversion
Narrowband upconversion
Fig. 12.29 Four-wave-mixing Bragg scattering. a Two strong pumps at frequencies xp1 and xp2
are combined with an input signal at xi to generate an output converted eld at xo which, in this
case, is shifted from the input by an amount xp2 xp1 . b Different congurations for four-wave-
mixing Bragg scattering. (Top) Broadband downconversion can be achieved by placing pumps xp1
and xp2 (xp1 [ xp2 ) in widely separated bands, with input signal xi at a wavelength closer to xp1
than xp2 ; broadband upconversion (not shown) can similarly be achieved, with input signal closer
to xp2 . (Bottom) Narrowband conversion can be achieved by placing pumps xp1 and xp2 relatively
close to each other in frequency. In all of these cases, two possible idlers can be generated,
consistent with energy conservation, and phase matching determines the efciency with which
each idler is produced
Phase-matching then determines which idler is more efciently generated (noting

that for a long interaction length, even a small phase mismatch can be enough to
select for one idler).
The most common medium for four-wave-mixing Bragg scattering has been
some form of an optical bre, including photonic crystal bre [158] and highly
nonlinear dispersion shifted bre [157, 159, 160]. Though the v3 nonlinearity is
generally much weaker than the v2 nonlinearity (e.g., comparing the values for
materials like photonic crystal or highly nonlinear bre with lithium niobate), the
combination of relatively localised eld connement, the low propagation losses,
and ability to fabricate optical bres uniformly over length scales of kilometres has
enabled four-wave-mixing Bragg scattering frequency converters to achieve a
conversion efciency of about 30 % for experiments at visible wavelength [158]
and nearing unity in the telecommunications band [156160] (typical length scales
are *30 m for photonic crystal bre and *1 km for highly nonlinear bre in
comparison to a few cm for lithium niobate). Phase-matching is achieved in these
optical bres by careful dispersion engineering, where waveguide dispersion (the
variation in lights phase velocity with wavelength, due to the wavelength-depen-
dent distribution of the optical eld within the bre core and cladding) can com-
pensate for material dispersion.
While noise-free in principle, just as in the case of sum-frequency-generation
and difference-frequency-generation in v2 media, four-wave-mixing Bragg
scattering in optical bres can be limited by a number of different possible noise

sources. Spontaneous Raman scattering, a known phenomenon in silica optical
bres, can be a dominant noise source when the converted output lies relatively
close in wavelength to the pumps. This has been the case in experiments in which
all four elds are near the telecommunications band, as explicitly discussed in
[160], where the Raman noise signal was of similar strength to the weak (single
photon level) coherent state input signal. For pumps that are more widely separated
from the input signal and frequency converted output, noise due to spontaneous
Raman scattering is more limited [158]. Other possible noise sources can include
undesired four-wave-mixing processes, which in some circumstances, could be
phase-matched (or close to phase matched). For example, this could include
spontaneous four-wave-mixing from one of the pumps (degenerate) or both of the
pumps (non-degenerate).
Quantum frequency conversion experiments in v3 media utilising non-classical
states of light have not been as widespread as they have in v2 materials, with the
rst experiment being performed a few years ago [158]. Here, the output of a
heralded single photon source produced by a photonic crystal bre was upconverted
by 24 nm (to 659 nm) in a second photonic crystal bre. With input and frequency
converted signals >150 nm on the anti-Stokes (higher frequency) side of the pumps,
spontaneous Raman scattering was limited and g2 0 was essentially unchanged
after frequency conversion.
12.3.3 Future Directions with Quantum Frequency

Conversion
New media and devices for quantum frequency conversion are an active area of
research, spurred in part by recent work in demonstrating nonlinear optical phe-
nomena in integrated and scalable nanophotonic geometries, including some of the
systems discussed earlier in this chapter. Within v2 media, efforts at producing
higher levels of device integration and more compact form factors within the
lithium niobate platform have shown promise, including recent work on developing
thin-lm lithium niobate devices [161] and integrating them with silicon-based
materials. Other v2 media have included GaAs, GaP, and GaN, for which there has
been work to demonstrate second harmonic generation and sum and difference
frequency generation in waveguide geometries [162, 163], as well as recent efforts
to establish second harmonic generation in nanophotonic cavity geometries [164
167]. Though these works are currently focused on the classical domain,
improvements in conversion efciency and an understanding of potential noise
mechanisms could enable QFC experiments, with future prospects including direct
integration with single photon emitters based on InAs/GaAs quantum dots, for
example.
Recently, four-wave-mixing Bragg scattering based on the v3 nonlinearity in

silicon nitride (Si3 N4 ) waveguides fabricated on a Si substrate has been demon-
strated [168, 169], with both narrowband (few nanometres) and wideband (few
hundreds of nanometres) upconversion and downconversion ranges shown in the
classical regime. Along with its broadband optical transparency and lack of two-
photon absorption for pump wavelengths >700 nm, Si3N4 does not appear to show
signicant amounts of spontaneous Raman scattering, so early indications suggest
that high signal-to-noise levels can be achieved in this system. Furthermore, the
combination of strong eld connement and the relatively high v3 in Si3N4 create
an effective nonlinearity that can be two orders of magnitude greater than that of
highly nonlinear bre. However, the relatively short Si3N4 waveguide length (1 cm)
is in stark contrast to the typical hundreds of metres used for QFC in bres, and
conversion efciencies have been limited to 5 %, with all experiments done strictly
in the classical regime. Chalcogenide materials such as As2S3, which also has a
broad optical transparency and limited two-photon absorption for pump wave-
lengths >800 nm, might also be a strong candidate for four-wave-mixing Bragg
scattering, particularly due to its high nonlinear refractive index. Silicon is also a
possibility for certain applications (e.g., telecom), particularly if pump wavelengths
>2.0 m can be used to avoid two-photon absorption, as has recently been dem-
onstrated for degenerate four-wave-mixing experiments [170, 171].
Over the past few years, there have been a number of works looking to extend
QFC from only shifting the color of quantum states of light to also providing
temporal wavepacket shaping [145], a necessary resource for many applications in
photonic quantum information science. For example, optimal storage of a single
photon in a quantum memory will generally require control of not only the
wavelength but also the spectro-temporal prole of the photon. Frequency con-
version provides a versatile route to wavepacket shaping in large part because the
nature of the process is such that the frequency converted eld has characteristics
that depend both on the input signal (a quantum eld) and the input pump eld(s)
(strong classical eld(s)). An example of this was demonstrated experimentally in
[151], where classical pump pulses were used to produce upconverted single
photons of a controllable duration (in the regime where the pump pulses were
shorter than the input single photon pulses). That work essentially applied a non-
linear time gating operation to the input single photon states, and was therefore not
lossless (even in principle). Fast nonlinear time gating of single photon pulses was
also recently used to eliminate which-path information in the photon energy in a
recent quantum dot spin-photon entanglement experiment [155].
There has been much progress on the theoretical front in developing lossless
protocols for full spectro-temporal shaping of quantum light elds. In [172], a two-
step process was proposed. The rst step consists of spectral broadening and
shaping of a single photon wavepacket through nonlinear mixing with a classical
pump that has been imprinted with a specied temporal phase. The output single
photon wavepacket has the desired frequency spectrum, and the spectral phase can
then be corrected (using classical pulse shaping approaches) to allow for full
temporal shaping. This approach could be used to strongly compress single photon
pulses produced by single quantum emitters, which often have nanosecond (or
longer) time constants. In [173], engineering of the group velocities of the pump
and input signal in a v2 medium was considered as a means to either compress or
stretch a single photon wavepacket, depending on whether the pump moves faster
or slower than the input signal. Single photon wavepacket shaping using four-wave-
mixing Bragg scattering has also been considered [174, 175], with a recent result
[175] showing an arbitrary input eld temporal prole can be mapped to an arbi-
trary output eld prole with unity efciency. In comparison to the previous work
[174], in which there was a tradeoff between selectivity (ability to map the input to
only one desired output) and efciency, [175] provides a prescription for providing
perfect (in principle) selectivity and efciency.
12.4 Nonlinear Optics for Quantum Communication
The rst real impact of quantum information research on wider society is in the area
of quantum communication and quantum cryptography: the eld of using quantum
laws to enhance privacy in optical communications. In this section we introduce
one of the key concepts of this eldquantum key distributionand describe how
it can be used to provide completely secure quantum communication. We also
provide state-of-the-art examples from the literature.
Fast and secure information transfer is vital to economic and social stability in
the modern internet enabled world. The current standard cryptographic algorithms
are not based on information-theoretic proofs, but on the assumption that some
mathematical calculations are hard to compute. If a fast classical algorithm for these
problems is discovered or a modestly sized practical quantum computer is devel-
oped, the integrity of these encryption protocols will be compromised. Quantum
key distribution (QKD) is one avenue that offers practical information-theoretic
security and is ideally realised using photonics platforms.
Typically photons are used, as their quantum properties remain intact over the
large distances required for useful communication. bre optics provides the ideal
channel for transmitting QKD photons, making use of the existing global bre
network. QKD differs fundamentally from other forms of encryption in the way it
provides security. If a photon is intercepted in flight by a malicious third party, and
the result is re-sent in the measured basis then the eavesdropper can be readily
detected and communication ceased [176]. Note that in most, though not all, cases a
key is distributed by quantum means rather than the whole message, simply because
the bandwidth of quantum information transfer is very much lower than current
classical optical communications.
However, in reality there are many challenges associated with the practical
implementation of QKD schemes and in some areas nonlinear photonics can play a
key role. Reliable photon sources such as those described earlier in this chapter can
be used in these networks, and the inherent entanglement in photon pair sources can
be exploited for communication. Quantum frequency conversion will allow dis-
parate users of a quantum network to be able to swap information and entangle-
ment, and for long distance quantum communication to become a reality resonant
and cavity based sources of photons may be able to reach the stringent requirements
to interface with quantum memories.
12.4.1 Single Photon Schemes
The most common QKD schemes [20, 177, 178] encode the key information using
one of the degrees of freedom of a single photon. Photons must be generated that
are completely indistinguishable in every degree of freedom except one, typically
the polarisation, but arrival time and orbital angular momentum are also common
choices. The polarisation state of a photon can take a continuum of values, so two
orthogonal values are selected to represent the states j0i and j1i. If one measures in
that orthogonal basis and and receives a photon in state j0i, one is guaranteed to
measure j0i. However, detection of quantum particles such as a single photon is
probabilistic and if an eavesdropper makes a measurement in an arbitrary polari-
sation basis, there is a nonzero probability that they will get the wrong answer.
Most importantly, when the eavesdropper makes a measurement in the incorrect
basis the photon state is modied increasing the number of errors and revealing the
presence of an attacker. Figure 12.30 shows a simple schematic of the BB84
protocol [20], the rst and best-known representative of this class of schemes.
The single photon source required is imperfectly realised with an attenuated
laser, having the advantages of a high clock speed, low timing jitter, robustness and
cheap manufacturing. The major drawback is that the photon emission from a laser
is described by a Poisson distribution and in this case the output is attenuated to the
point where the probability of emitting one photon is larger than the probability of
emitting two or more. A source that emits two or more photons is potentially
vulnerable to attacks such as the beam splitter attack [176]. This multi-photon
rate can be reduced by multiplexing many nonlinear sources according to the
strategies discussed in Sect. 12.2.6, or by using true single photon emitters such as
diamond N-V centres and quantum dots, but the technical challenges of these
sources makes their use impractical at present.
Current commercial products offered allow links over 100 km with key distri-
bution rates of several Mbit/s [179] using optical bres, with each link costing at
least USD $100,000. State of the art research has demonstrated real QKD channels
running in parallel down the same bre as secured 10 GBit/s bright classical sig-
nals. All channels were located in the telecommunication C-band of wavelengths,
with standard channel spacing of 100 and a 200 GHz spacing between the quantum
and classical signals [180, 181]. For ultimate security, the gold standard is a one-
time pad which does not allow repeated use of any random key, with the key size
equal to the message size. 100 kbps one-time pad video conferencing over 45 km
A B
Classical Channel
or or
Quantum Channel
Fig. 12.30 BB84 QKD scheme. Alice, at location A, sends a stream of photons to Bob, at location
B, with the photon polarisation randomly orientated between the orthogonal states of vertical (0)
and horizontal (1). Alice also randomly changes the basis to a diagonal basis while sending the
stream. Bob measures the incoming photons and records a 0 or 1 for vertical and horizontally
polarised photons respectively. Both users then broadcast which photon they sent/measured in
which basis, the normal or diagonal. When the bases are the same, they keep those 0, 1 values and
use them for their key. This is called key sifting. They can then encrypt messages and send them
securely across public classical channels
was demonstrated at the Tokyo QKD network in a 2011 joint effort by NICT, NEC,
Mitsubishi Electric and NTT [182].
In any optical bre communication link photons are lost when they scatter off
imperfections in the bre. In the case of conventional communication, the infor-
mation can be preserved by regenerating or amplifying the signal. In the quantum
case, the no-cloning theorem, which forbids the reliable copying of a quantum
state, prevents a quantum signal from being amplied [176], thus the quantum key
distribution rate drops with distance, eventually falling below a usable threshold.
In a recently developed technique called differential-phase-shift QKD [183] the
key is encoded onto the phase of a coherent pulse, which is attenuated as it travels
through the bre, ensuring only 1 or fewer photons remain for detection. This has
enabled impressive demonstrations of QKD over 100 km at 24 kbps by Japan NTT
corporation [184] and more recently 260 km QKD over standard optical bre was
shown by Chinas Anhui Asky Quantum Technology Corporation [185].
12.4.2 Entanglement-Based Schemes
In 1992 the Eckert scheme for QKD was published [9], which exploits another
quantum mechanical resource called entanglement. A pair of photons can be gen-
erated in such a way that the observable properties of one photon are inseparably
linked to the other, and describing the state of one such photon independently of the
other is both impossible and meaningless. Pairs of entangled photons can be gen-
erated using a nonlinear optical crystal, or from an atomic relaxation. If these
photons are generated at the mid-point of a QKD link, they can be sent down an
optical bre to Alice and Bob, potentially doubling the maximum distance of a
QKD link. When Alice and Bob happen to measure their photon in the same basis,
entanglement ideally guarantees that their results will be perfectly correlated. Using
A B
Classical Channel
or or
Entangled
Photon
Source
Fig. 12.31 Schematic of entanglement based QKD
this to generate identical keys allows secure communication over a public channel,
as shown in Fig. 12.31. Observing either photon in the pair destroys the entan-
glement and the key correlations immediately drop, flagging the presence of an
eavesdropper.
NuCrypt, [186] a spin-out from Northwestern University in the USA in 2003,
offers the only entanglement based QKD product with a rack mountable 2.5 Gbit/s
unit that sends the encrypted signal over the same bre as the quantum keys. The
entanglement is generated using highly nonlinear silica bre, with the option of
cooling in liquid nitrogen for increased performance. The European project SEC-
OQC [187] has used such entangled photon pairs in combination with several other
schemes, to connect 6 nodes securely. Toshiba demonstrated a 200 km bre optic
link using practical avalanche photo diode (APD) detectors [188]. Japans NTT
demonstrated a 100 km link using polarisation entangled photons generated in a
CMOS compatible silicon chip.
12.4.3 Long Distance Quantum Communication
Long distance QKD is currently limited by losses in real world communication

systems, whether they are based on free-space or optical bre links, limiting
communication to roughly less than 200 km [185, 189, 190]. As already mentioned,
when dealing with single photons the no-cloning theorem disallows deterministic
copying of quantum states, meaning that classical optical ampliers are rendered
useless. There are however a number of proposals to extend this distance. In free
space links the use of a satellite based relay, or a simple reflector, could link two
distant points on Earth, whereas in optical bre links one would require a quantum
repeater or quantum relay to forward on the photons carrying the quantum
information.
12.4.3.1 Space-Based Systems
There have been a number of feasibility studies undertaken to discern whether

satellite based QKD is a viable method to securely transfer information on a global
scale [191, 192]. There are several possible methods for communication via this
methodwhether the receiver is placed on the ground or on the satellite, or if the
satellite is a simple retro-reflectoreach with its own inherent advantages and
problems. One problem recently addressed was that of having a shared xed ref-
erence frame to make basis choices for the sender and receiver, a hard problem
when dealing with a satellite moving through space. Through the use of circular
polarisation and integrated optics, researchers were able to devise a method of
forming reference frame independent QKD protocols [193].
A seminal experiment in 2007 [189] demonstrated a 144 km free space link
between two of the Canary Islands. On accounting for the degree of atmospheric
scattering one would encounter between two points of equal height from sea level,
this is comparable to sending a beam into space as the atmosphere becomes thinner
with increasing height above ground level. Recently more advanced proposals have
been put forward by a large number of universities to collaborate with the European
Space Agency to implement satellite-based entanglement experiments and experi-
ments to form a quantum link between the International Space Station and a tele-
scope site on Earth [194]. One such experiment has been completed where a
satellite is used as a weak reflector for a classical laser resulting in a channel
containing single photons which is then experimentally veried through photon
counting statistics, however no QKD has been performed using this method to date.
12.4.3.2 Quantum Repeaters and Relays
When sending classical communication through long distance optical bre links
one has the ability to amplify the channel at various locations to ensure that the
signal-to-noise ratio is not degraded between users wishing to transfer data. Once
again, this is impossible for quantum communication due to the impossibility of
cloning states. Instead, proposed architectures for long distance quantum commu-
nication employ either a quantum repeater [195] shown in Fig. 12.32, which allows
two-way communication, or a quantum relay [196] shown in Fig. 12.33, for one-
way communication.
The aim of a quantum repeater is to supply one photon from an entangled pair to
the next node in the communication chain whilst storing the other until all links
have been set up. An entanglement swapping routine is then performed at each
repeater on the stored photons until the two photons at the very end of the com-
munication chain are themselves entangled. One can then perform a standard Eckert
style QKD protocol. To perform the entanglement swapping routine one requires a
small scale quantum information processing element, such as a controlled-NOT
A B
50:50
Entangled Entangled
Photon Photon SPDs
Source Source
&
50:50
Quantum Quantum
Memory Memory
SPDs
&
Fig. 12.32 A schematic of a quantum repeater. For Alice to communicate with Bob they must set
up some entanglement between them. This is done through the use of a quantum repeater, shown
in the purple circle, and joint quantum measurement, shown in the red circle. The quantum
repeater contains entangled sources of photon pairs which are further entangled through
measurement until entanglement swapping means that the photons at A and B are entangled. The
quantum memory allows the initial entanglement to be set up before the other swapping occurs
gate [197], and single photon detection. The most challenging component in a
quantum repeater is the quantum memory used to store the photons to ensure a
successful swapping of entanglement along the chain. Current memory proposals
[198] are either based on bulky and fragile atomic ensemble setups which are not
easily used outside the lab, low temperature devices based on quantum dots, or
diamond N-V centres which require signicant further engineering of cavities and
out-coupling before they can be viable as memories.
On the other hand, a quantum relay is a teleportation assisted device where an
incoming photon is jointly measured with one photon of a correlated pair, shown
schematically in Fig. 12.33. After the joint measurement the state of the input
photon is teleported to the state of the other photon of the correlated pair which can
then be transmitted. The joint measurement requires a form of quantum interference
[18] as well as single photon detection. A preliminary experiment was recently
published [196] demonstrating such a quantum relay device in a lithium niobate
integrated architecture, however imperfect quantum interference and losses limit the
usefulness of the device and no teleportation has been demonstrated thus far.
A B
Entangled
Photon
50:50 Source
SPDs
&
Fig. 12.33 A schematic of a quantum relay. A photon enters the relay device and is interfered
with a photon from an entangled pair in a quantum measurement, which teleports the quantum
state to the output photon for transmission. These can then be concatenated to extend the distance
12.5 Conclusion
It is clear that nonlinear photonics can provide enhanced functionality to a number of

quantum technologies including in a quantum network, where quantum information
can be transported securely when encoded to single photon states. There are many
challenges to be overcome before we see this quantum photonic hardware in everyday
use, however research into this area is fast paced and we expect to see quantum
enhanced technologies and networks become prolic in the coming decades.
Acknowledgments This work was supported in part by the Centre of Excellence (CUDOS,
project number CE110001018), Laureate Fellowship (FL120100029) and Discovery Early Career
Researcher Award (DE130101148 and DE120100226) programs of the Australian Research
Council (ARC). L.G.H. acknowledges support from a Macquarie University Research Fellowship
(MQRF). K.S. would like to thank Matthew Rakher, Serkan Ates, and Imad Agha for helpful
comments and contributions to work presented in Sect. 12.3.
References
1. W.K. Wooters, W.H. Zurek, A single quantum cannot be cloned. Nature 299, 802803
(1982)
2. D. Dieks, Communication by EPR devices. Phys. Lett. A 92, 271272 (1982)
3. W.J. Munro, K. Nemoto, T.P. Spiller, Weak nonlinearities: a new route to optical quantum
computation. New J. Phys. 7, 137 (2005)
4. N. Matsuda, R. Shimizu, Y. Mitsumori, H. Kosaka, K. Edamatsu, Observation of optical-

bre Kerr nonlinearity at the single-photon level. Nature Photon. 3, 9598 (2009)
5. V. Venkataraman, K. Saha, A.L. Gaeta, Phase modulation at the few-photon level for weak-
nonlinearity-based quantum computing. Nature Photon. 7, 138141 (2013)
6. J.H. Shapiro, Single-photon Kerr nonlinearities do not help quantum computation. Phys.
Rev. A 73, 062305 (2006)
7. N.K. Langford, S. Ramelow, R. Prevedel, W.J. Munro, G.J. Milburn, A. Zeilinger, Efcient
quantum computing using coherent photon conversion. Nature 478, 360363 (2011)
8. J.S. Bell, Speakable and Unspeakable in Quantum Mechanics (Cambridge University Press,
Cambridge, 1987)
9. A. Ekert, Quantum cryptography based on Bells theorem. Phys. Rev. Lett. 67, 661663
(1991)
10. A.K. Ekert, J.G. Rarity, P.R. Tapster, G.M. Palma, Practical quantum cryptography based on
two-photon entanglement. Phys. Rev. Lett. 69, 12931295 (1992)
11. J. Ye, D.W. Vernooy, H.J. Kimble, Trapping of single atoms in cavity QED. Phys. Rev. Lett.
83, 49874990 (1999)
12. J.I. Cirac, P. Zoller, Quantum computation with cold trapped ions. Phys. Rev. Lett. 74, 4091
4094 (1995)
13. S. Prawer, A.D. Greentree, Diamond for quantum computing. Science 320, 16011602
(2008)
14. A. Politi, M. Cryan, J. Rarity, S. Yu, J. OBrien, Silica-on-silicon waveguide quantum
circuits. Science 320, 646649 (2008)
15. J. Brendel, N. Gisin, W. Tittel, H. Zbinden, Pulsed energy-time entangled twin-photon
source for quantum communication. Phys. Rev. Lett. 82, 25942597 (1999)
16. P.C. Humphreys, B.J. Metcalf, J.B. Spring, M. Moore, X.-M. Jin, M. Barbieri, S.
Kolthammer, I.A. Walmsley, Linear optical quantum computing in a single spatial mode.
Phys. Rev. Lett. 111, 150501 (2013)
17. E. Knill, R. Laflamme, G.J. Milburn, A scheme for efcient quantum computation with linear
optics. Nature 409, 4652 (2001)
18. C.K. Hong, Z.Y. Ou, L. Mandel, Measurement of subpicosecond time intervals between two
photons by interference. Phys. Rev. Lett. 59, 2044 (1987)
19. J.G. Rarity, P.R. Tapster, Fourth order interference in parametric down conversion. J. Opt.
Soc. Am. B 6, 1221 (1989)
20. C. Bennett, G. Brassard, Quantum cryptography: public key distribution and coin tossing, in
Proceedings of the International Conference on Computers, Systems & Signal Processing
(1984), pp. 175179
21. M.A. Nielsen, I.L. Chuang, Quantum Computation and Quantum Information (Cambridge
University Press, New York, 2004)
22. C. Santori, D. Fattal, J. Vuckovi, G.S. Solomon, Y. Yamamoto, Indistinguishable photons
from a single-photon device. Nature 419, 594597 (2002)
23. S.V. Polyakov, A. Muller, E.B. Flagg, A. Ling, N. Borjemscaia, E. Van Keuren, A. Migdall,
G.S. Solomon, Coalescence of single photons emitted by disparate single-photon sources: the
example of InAs quantum dots and parametric down-conversion sources. Phys. Rev. Lett.
107, 157402 (2011)
24. A. Huck, S. Kumar, A. Shakoor, U.L. Andersen, Controlled coupling of a single nitrogen-
vacancy center to a silver nanowire. Phys. Rev. Lett. 106, 096801 (2011)
25. F. Jelezko, J. Wrachtrup, Single defect centres in diamond: a review. Phys. Status Solidi A
203, 32073225 (2006)
26. P. Gold, A. Thoma, S. Maier, S. Reitzenstein, C. Schneider, S. Hfling, M. Kamp, Two-
photon interference from remote quantum dots with inhomogeneously broadened linewidths.
Phys. Rev. B 89, 035313 (2014)
27. R. Loudon, The Quantum Theory of Light, 3rd edn. (Oxford University Press, New York,
2000)
28. S. Scheel, Single-photon sources-an introduction. J. Mod. Opt. 56, 141160 (2009)
29. S. Fasel, O. Alibart, S. Tanzilli, P. Baldi, A. Beveratos, N. Gisin, H. Zbinden, High-quality

asynchronous heralded single-photon source at telecom wavelength. New J. Phys. 6, 163
164 (2004)
30. T. Pittman, B. Jacobs, J. Franson, Heralding single photons from pulsed parametric down-
conversion. Opt. Commun. 246, 545550 (2005)
31. O. Alibart, D.B. Ostrowsky, P. Baldi, S. Tanzilli, High-performance guided-wave
asynchronous heralded single-photon source. Opt. Lett. 30, 15391541 (2005)
32. A. Soujaeff, S. Takeuchi, K. Sasaki, T. Hasegawa, M. Matsui, Heralded single photon source
at 1550 nm from pulsed parametric down conversion. J. Mod. Opt. 54, 467471 (2007)
33. P. Mosley, J. Lundeen, B. Smith, P. Wasylczyk, A. URen, C. Silberhorn, I. Walmsley,
Heralded generation of ultrafast single photons in pure quantum states. Phys. Rev. Lett. 100,
133601 (2008)
34. Z.H. Levine, J. Fan, J. Chen, A. Ling, A. Migdall, Heralded, pure-state single-photon source
based on a potassium titanyl phosphate waveguide. Opt. Express 18, 37083718 (2010)
35. M.D. Cunha Pereira, F.E. Becerra, B.L. Glebov, J. Fan, S.W. Nam, A. Migdall,
Demonstrating highly symmetric single-mode, single-photon heralding efciency in
spontaneous parametric downconversion. Opt. Lett. 38, 160911 (2013)
36. S. Ramelow, A. Mech, M. Giustina, Highly efcient heralding of entangled single photons.
Opt. Express 21, 67076717 (2013)
37. E. Goldschmidt, M. Eisaman, J. Fan, S. Polyakov, A. Migdall, Spectrally bright and broad
ber-based heralded single-photon source. Phys. Rev. A 78, 013844 (2008)
38. A.R. McMillan, J. Fulconis, M. Halder, C. Xiong, J.G. Rarity, W.J. Wadsworth, Narrowband
high-delity all-bre source of heralded single photons at 1570 nm. Opt. Express 17, 6156
6165 (2009)
39. C. Sller, O. Cohen, B.J. Smith, I.A. Walmsley, C. Silberhorn, High-performance single-
photon generation with commercial-grade optical ber. Phys. Rev. A 83, 031806 (2011)
40. L. Yang, X. Ma, X. Guo, L. Cui, X. Li, Characterization of a ber-based source of heralded
single photons. Phys. Rev. A 83, 053843 (2011)
41. M. Davano, J.R. Ong, A.B. Shehata, A. Tosi, I. Agha, S. Assefa, F. Xia, W.M.J. Green, S.
Mookherjea, K. Srinivasan, Telecommunications-band heralded single photons from a
silicon nanophotonic chip. Appl. Phys. Lett. 100, 261104 (2012)
42. J. Spring, P. Salter, B. Metcalf, P. Humphreys, M. Moore, N. Thomas-Peter, M. Barbieri, X.-
M. Jin, N. Langford, W. Kolthammer, M. Booth, I. Walmsley, On-chip low loss heralded
source of pure single photons. Opt. Express 21, 1352213532 (2013)
43. A. Christ, C. Silberhorn, Limits on the deterministic creation of pure single-photon states
using parametric down-conversion. Phys. Rev. A 85, 023829 (2012)
44. M.J. Collins, C. Xiong, I.H. Rey, T.D. Vo, J. He, S. Shahnia, C. Reardon, T.F. Krauss, M.
J. Steel, A.S. Clark, B.J. Eggleton, Integrated spatial multiplexing of heralded single-photon
sources. Nat. Commun. 4, 2582 (2013)
45. A. Yariv, Quantum Electronics, 2nd edn. (Wiley, New York, 1975)
46. G.P. Agrawal, Nonlinear Fiber Optics, 4th edn. (Academic Press, Burlington, 2007)
47. L.G. Helt, M. Liscidini, J.E. Sipe, How does it scale? comparing quantum and classical
nonlinear optical processes in integrated devices. J. Opt. Soc. Am. B 29, 21992212 (2012)
48. H. Takesue, K. Shimizu, Effects of multiple pairs on visibility measurements of entangled
photons generated by spontaneous parametric processes. Opt. Commun. 283, 276287
(2010)
49. M.J. Stevens, S. Nam, in Single-Photon Generation and DetectionPhysics and
Applications, ed. by J.F.A. Migdall, S. Polyakov, J.C. Bienfang. Photon Statistics,
Measurements, and Measurement Tools (Academic Press, Burlington, 2013)
50. J.G. Rarity, J. Fulconis, J. Duligall, W.J. Wadsworth, P.S.J. Russell, Photonic crystal ber
source of correlated photon pairs. Opt. Express 13, 534 (2005)
51. E. Engin, D. Bonneau, C.M. Natarajan, A.S. Clark, M.G. Tanner, R.H. Hadeld, S.N.
Dorenbos, V. Zwiller, K. Ohira, N. Suzuki, H. Yoshida, N. Iizuka, M. Ezaki, J.L. OBrien,
M.G. Thompson, Photon pair generation in a silicon micro-ring resonator with reverse bias
enhancement. Opt. Express 21, 2782627834 (2013)
52. A. Christ, A. Fedrizzi, H. Hbel, T. Jennewein, C. Silberhorn, in Single-Photon Generation
and DetectionPhysics and Applications, ed. by J.F.A. Migdall, S. Polyakov, J.C. Bienfang.
Parametric Down-Conversion (Academic Press, Burlington, 2013)
53. P. Kwiat, K. Mattle, H. Weinfurter, A. Zeilinger, A. Sergienko, Y. Shih, New high-intensity
source of polarization-entangled photon pairs. Phys. Rev. Lett. 75, 43374341 (1995)
54. A. Peruzzo, M. Lobino, J. Matthews, N. Matsuda, A. Politi, K. Poulios, X.-Q. Zhou, Y.
Lahini, N. Ismail, K. Wrhoff, Y. Bromberg, Y. Silberberg, M. Thompson, J. OBrien,
Quantum walks of correlated photons. Science 329, 15001503 (2010)
55. A. Aspuru-Guzik, P. Walther, Photonic quantum simulators. Nature Phys. 8, 285291 (2012)
56. P. Kwiat, Nonclassical Effects from Spontaneous Parametric Down-conversion: Adventures
in Quantum Wonderland, Ph.D thesis, http://research.physics.illinois.edu/QI/Photonics/
theses.html, 1993
57. A. Halevy, E. Megidish, L. Dovrat, H. Eisenberg, P. Becker, L. Bohat, The biaxial
nonlinear crystal BiB3O6 as a polarization entangled photon source using non-collinear type-
II parametric down-conversion. Opt. Express 19, 2042020434 (2011)
58. R. Rangarajan, M. Goggin, P. Kwiat, Optimizing type-I polarization-entangled photons. Opt.
Express 17, 1892018933 (2009)
59. M. Fiorentino, S. Spillane, R. Beausoleil, T. Roberts, P. Battle, M. Munro, Spontaneous
parametric down-conversion in periodically poled KTP waveguides and bulk crystals. Opt.
Express 15, 74797488 (2007)
60. T. Zhong, F. Wong, T. Roberts, P. Battle, High performance photon-pair source based on a
ber-coupled periodically poled KTiOPO4 waveguide. Opt. Express 17, 1201912030
(2009)
61. S.V. Zhukovsky, L.G. Helt, P. Abolghasem, D. Kang, J.E. Sipe, A.S. Helmy, Bragg
reflection waveguides as integrated sources of entangled photon pairs. J. Opt. Soc. Am. B 29,
25162523 (2012)
62. R.T. Horn, P. Kolenderski, D. Kang, P. Abolghasem, C. Scarcella, A.D. Frera, A. Tosi, L.G.
Helt, S.V. Zhukovsky, J.E. Sipe, G. Weihs, A.S. Helmy, T. Jennewein, Inherent polarization
entanglement generated from a monolithic semiconductor chip. Sci. Rep. 3, 2314 (2013)
63. A. Valls, M. Hendrych, J. Svozilk, R. Machulka, P. Abolghasem, D. Kang, B.J. Bijlani, A.
S. Helmy, J.P. Torres, Generation of polarization-entangled photon pairs in a Bragg reflection
64. Z. Yang, P. Chak, A.D. Bristow, H.M. van Driel, R. Iyer, J.S. Aitchison, A.L. Smirl, J.E.
Sipe, Enhanced second-harmonic generation in AlGaAs microring resonators. Opt. Lett. 32,
826828 (2007)
65. R. Boyd, Nonlinear Optics, 2nd edn. (Academic Press, Burlington, 2003)
66. M. Hunault, H. Takesue, O. Tadanaga, Y. Nishida, M. Asobe, Generation of time-bin
entangled photon pairs by cascaded second-order nonlinearity in a single periodically poled
linbO3 waveguide. Opt. Lett. 35, 12391241 (2010)
67. A. Martin, A. Issautier, H. Herrmann, W. Sohler, D. Ostrowsky, O. Alibart, S. Tanzilli, A
polarization entangled photon-pair source based on a type-II PPLN waveguide emitting at a
telecom wavelength. New J. Phys. 12, 103005 (2010)
68. M. Lobino, G. Marshall, C. Xiong, A. Clark, D. Bonneau, C. Natarajan, M. Tanner, R.
Hadeld, S. Dorenbos, T. Zijlstra, V. Zwiller, M. Marangoni, R. Ramponi, M. Thompson, B.
Eggleton, J. OBrien, Correlated photon-pair generation in a periodically poled MgO doped
stoichiometric lithium tantalate reverse proton exchanged waveguide. Appl. Phys. Lett. 99,
081110 (2011)
69. A.M. Braczyk, A. Fedrizzi, T.M. Stace, T.C. Ralph, A.G. White, Engineered optical
nonlinearity for quantum light sources. Opt. Express 19, 5565 (2011)
70. H. Zhang, X.-M. Jin, J. Yang, H.-N. Dai, S.-J. Yang, T.-M. Zhao, J. Rui, Y. He, X. Jiang, F.
Yang, G.-S. Pan, Z.-S. Yuan, Y. Deng, Z.-B. Chen, X.-H. Bao, S. Chen, B. Zhao, J.-W. Pan,
Preparation and storage of frequency-uncorrelated entangled photons from cavity-enhanced

spontaneous parametric downconversion. Nature Photon. 5, 628632 (2011)
71. C. Clausen, I. Usmani, F. Bussires, N. Sangouard, M. Afzelius, H. Riedmatten, N. Gisin,
Quantum storage of photonic entanglement in a crystal. Nature 469, 508511 (2011)
72. X.-H. Bao, Y. Qian, J. Yang, H. Zhang, Z.-B. Chen, T. Yang, J.-W. Pan, Generation of
narrow-band polarization-entangled photon pairs for atomic quantum memories. Phys. Rev.
Lett. 101, 190501 (2008)
73. E. Pomarico, B. Sanguinetti, C. Osorio, H. Herrmann, R. Thew, Engineering integrated pure
narrow-band photon sources. New J. Phys 14, 033008 (2012)
74. A. McMillan, Y. Huang, B. Bell, A.S. Clark, P. Kumar, J. Rarity, in Single-Photon
Generation and DetectionPhysics and Applications, chap. 12, ed. by A. Migdall, S.
Polyakov, J. Fan. Four-Wave Mixing in Single-Mode Optical Fibers (Academic Press,
Burlington, 2013), pp. 411465
75. C. Xiong, Nonlinearity in Photonic Crystal Fibres. Ph.D thesis, University of Bath, 2008
76. J. Sharping, M. Fiorentino, P. Kumar, Observation of twin-beam-type quantum correlation in
optical ber. Opt. Lett. 26, 367369 (2001)
77. M. Fiorentino, P. Voss, J. Sharping, P. Kumar, All-ber photon pair source for quantum
communications. IEEE Photon. Tech. Lett. 14, 983985 (2002)
78. J. Sharping, J. Chen, X. Li, P. Kumar, Quantum correlated twin photons from
microstructured ber. Opt. Express 12, 30863094 (2004)
79. X. Li, P. Voss, J. Sharping, P. Kumar, Optical-ber source of polarization-entangled photons
in the 1550 nm telecom band. Phys. Rev. Lett. 94, 053601 (2005)
80. H. Takesue, K. Inoue, 1.5-m band quantum-correlated photon pair generation in dispersion-
shifted ber suppression of noise photons by cooling ber. Opt. Express 13, 78327839
(2005)
81. S. Dyer, M. Stevens, B. Baek, S. Nam, High-efciency, ultra low-noise all-ber photon-pair
source. Opt. Express 16, 99669977 (2008)
82. J. Fulconis, O. Alibart, J. OBrien, W. Wadsworth, J. Rarity, Nonclassical interference and
entanglement generation using a photonic crystal ber pair photon source. Phys. Rev. Lett.
99, 120501 (2007)
83. J. Fulconis, O. Alibart, W. Wadsworth, J. Rarity, Quantum interference with photon pairs
using two micro-structured bers. New J. Phys. 9, 276 (2007)
84. M. Halder, J. Fulconis, B. Cemlyn, A. Clark, C. Xiong, W. Wadsworth, J. Rarity,
Nonclassical 2-photon interference with separate intrinsically narrowband bre sources. Opt.
Express 17, 46704676 (2009)
85. A. Clark, B. Bell, J. Fulconis, M. Halder, B. Cemlyn, A. Olivier, C. Xiong, W. Wadsworth,
J. Rarity, Intrinsically narrowband pair photon generation in microstructured bres. New
J. Phys. 13, 065009 (2011)
86. J. Fan, A. Dogariu, L. Wang, Generation of correlated photon pairs in a microstructure ber.
Opt. Lett. 30, 15301532 (2005)
87. J. Chen, K. Lee, C. Liang, P. Kumar, Fiber-based telecom-band degenerate-frequency source
of entangled photon pairs. Opt. Lett. 31, 27982799 (2006)
88. C. Xiong, G. Marshall, A. Peruzzo, M. Lobino, A. Clark, D.-Y. Choi, S. Madden, C.
Natarajan, M. Tanner, R. Hadeld, S. Dorenbos, T. Zijlstra, V. Zwiller, M. Thompson,
J. Rarity, M. Steel, B. Luther-Davies, B. Eggleton, J. OBrien, Generation of correlated
photon pairs in a chalcogenide As2S3 waveguide. Appl. Phys. Lett. 98, 051101 (2011)
89. M. Collins, A. Clark, J. He, D.-Y. Choi, R. Williams, A. Judge, S. Madden, M. Withford, M.
Steel, B. Luther-Davies, C. Xiong, B. Eggleton, Low raman-noise correlated photon-pair
generation in a dispersion engineered chalcogenide As2S3 planar waveguide. Opt. Lett. 37,
33933395 (2012)
90. J. He, C. Xiong, A. Clark, M. Collins, X. Gai, D.-Y. Choi, S. Madden, B. Luther-Davies, B.
Eggleton, Effect of low-Raman window position on correlated photon-pair generation in a
chalcogenide Ge11.5 As24 Se64.5 nanowire. J. Appl. Phys. 112, 123101 (2012)
91. J. Sharping, K. Lee, M. Foster, A. Turner, B. Schmidt, M. Lipson, A. Gaeta, P. Kumar,

Generation of correlated photons in nanoscale silicon waveguides. Opt. Express 14, 12388
12393 (2006)
92. K. Harada, H. Takesue, H. Fukuda, T. Tsuchizawa, T. Watanabe, K. Yamada, Y. Tokura, S.
Itabashi, Generation of high-purity entangled photon pairs using silicon wire waveguide.
Opt. Express 16, 2036820373 (2008)
93. S. Clemmen, K. Huy, W. Bogaerts, R. Baets, P. Emplit, S. Massar, Continuous wave photon
pair generation in silicon-on-insulator waveguides and ring resonators. Opt. Express 17,
1655816570 (2009)
94. K. Harada, H. Takesue, H. Fukuda, T. Tsuchizawa, T. Watanabe, K. Yamada, Y. Tokura, S.
Itabashi, Indistinguishable photon pair generation using two independent silicon wire
waveguides. New J. Phys. 13, 065005 (2011)
95. S. Clemmen, A. Perret, S. Selvaraja, W. Bogaerts, D. van, Thourhout, R. Baets, P. Emplit, S.
Massar, Generation of correlated photons in hydrogenated amorphous-silicon waveguides.
Opt. Lett. 35, 34833485 (2010)
96. B. Eggleton, B. Luther-Davies, K. Richardson, Chalcogenide photonics. Nature Photon. 5,
141148 (2011)
97. S. Azzini, D. Grassani, M. Strain, M. Sorel, L. Helt, J. Sipe, M. Liscidini, M. Galli, D.
Bajoni, Ultra-low power generation of twin photons in a compact silicon ring resonator. Opt.
Express 20, 2310023107 (2012)
98. C. Xiong, C. Monat, M. Collins, L. Tranchant, D. Petiteau, A. Clark, C. Grillet, G. Marshall,
M. Steel, J. Li, L. OFaolain, T. Krauss, B. Eggleton, Characteristics of correlated photon
pairs generated in ultra-compact silicon slow-light photonic crystal waveguides. IEEE J. Sel.
Topics Quantum Electron. 18, 16761683 (2012)
99. N. Matsuda, H. Takesue, K. Shimizu, Y. Tokura, E. Kuramochi, M. Notomi, Slow light
enhanced correlated photon pair generation in photonic-crystal coupled-resonator optical
100. W. Jiang, X. Lu, J. Zhang, O. Painter, Q. Lin, A silicon-chip source of bright photon-pair
comb. arXiv:1210.4455 (2012)
101. A. Clark, C. Husko, M. Collins, G. Lehoucq, S. Xavier, A. De, Rossi, S. Combri, C. Xiong,
B. Eggleton, Heralded single-photon source in a IIIV photonic crystal. Opt. Lett. 38, 649
651 (2013)
102. C. Husko, A. Clark, M. Collins, A. De, Rossi, S. Combri, G. Lehoucq, I. Rey, T. Krauss, C.
Xiong, B. Eggleton, Multi-photon absorption limits to heralded single photon sources. Sci.
Rep. 3, 3087 (2013)
103. J. Li, T.P. White, L. OFaolain, A. Gomez-Iglesias, T.F. Krauss, Systematic design of flat
band slowlight in photonic crystal waveguides. Opt. Express 16, 62276232 (2008)
104. T.F. Krauss, Slow light in photonic crystal waveguides. J. Phys. D: Appl. Phys. 40, 2666
2670 (2007)
105. J. Li, L. OFaolain, I.H. Rey, T.F. Krauss, Four-wave mixing in photonic crystal waveguides:
slow light enhancements and limitations. Opt. Express 19, 44584463 (2011)
106. A. Yariv, Y. Xu, R.K. Lee, A. Scherer, Coupled-resonator optical waveguide, a proposal and
analysis. Opt. Lett. 24, 711713 (1999)
107. R. Kumar, J. Ong, J. Recchio, S. Mookherjea, K. Srinivasan, Wide-span and tunable-
wavelength photon pairs around 1.55 m from a silicon nanophotonic chip, 1304:4989v1
(2013)
108. Z. Yang, M. Liscidini, J. Sipe, Spontaneous parametric down-conversion in waveguides: a
backward Heisenberg picture approach. Phys. Rev. A 77, 033808 (2008)
109. L.G. Helt, E.Y. Zhu, M. Liscidini, L. Qian, J.E. Sipe, Proposal for in-ber generation of
telecom-band polarization-entangled photon pairs using a periodically poled ber. Opt. Lett.
34, 21382140 (2009)
110. M. Liscidini, L.G. Helt, J.E. Sipe, Asymptotic elds for a Hamiltonian treatment of nonlinear
electromagnetic phenomena. Phys. Rev. A 85, 013833 (2012)
111. P. Kwiat, E. Waks, A. White, Ultrabright source of polarization-entangled photons. Phys.

Rev. A 60, 773776 (1999)
112. S.V. Zhukovsky, L.G. Helt, D. Kang, P. Abolghasem, A.S. Helmy, J.E. Sipe, Generation of
maximally-polarization-entangled photons on a chip. Phys. Rev. A 85, 013838 (2012)
113. J. Bell, On the Einstein Podolsky Rosen paradox. Physics 1, 195200 (1964)
114. J. Clauser, M. Horne, Proposed experiment to test local hidden-variable theories. Phys. Rev.
Lett. 23, 880884 (1969)
115. T. Humble, W. Grice, Effects of spectral entanglement in polarization-entanglement
swapping and type-I fusion gates. Phys. Rev. A 77, 022312 (2008)
116. C. Law, J. Eberly, Analysis and interpretation of high transverse entanglement in optical
parametric down conversion. Phys. Rev. Lett. 92, 127903 (2004)
117. W. Grice, A. URen, I. Walmsley, Eliminating frequency and space-time correlations in
multiphoton states. Phys. Rev. A 64, 063815 (2001)
118. K. Garay-Palmett, H. McGuinness, Photon pair-state preparation with tailored spectral
properties by spontaneous four-wave mixing in photonic-crystal ber. Opt. Express 15,
29152920 (2007)
119. O. Cohen, J. Lundeen, B. Smith, G. Puentes, P. Mosley, I. Walmsley, Tailored photon-pair
generation in optical bers. Phys. Rev. Lett. 102, 123603 (2009)
120. Z. Walton, A. Sergienko, B. Saleh, M. Teich, Generation of polarization-entangled photon
pairs with arbitrary joint spectrum. Phys. Rev. A 70, 052317 (2004)
121. L.G. Helt, Z. Yang, M. Liscidini, J.E. Sipe, Spontaneous four-wave mixing in microring
resonators. Opt. Lett. 35, 30063008 (2010)
122. B.J. Smith, P. Mahou, O. Cohen, J.S. Lundeen, I.A. Walmsley, Photon pair generation in
birefringent optical bers. Opt. Express 17, 2358923602 (2009)
123. I. Jizan, A. Clark, L. Helt, M. Collins, E. Mgi, C. Xiong, M. Steel, B. Eggleton, High-
resolution measurement of spectral quantum correlations in the telecommunication band.
Opt. Commun. 327, 4548 (2014)
124. J.M. Lukens, A. Dezfooliyan, C. Langrock, M.M. Fejer, D.E. Leaird, A.M. Weiner,
Biphoton manipulation with a ber-based pulse shaper. Opt. Lett. 38, 46524655 (2013)
125. M. Avenhaus, A. Eckstein, P.J. Mosley, C. Silberhorn, Fiber-assisted single-photon
spectrograph. Opt. Lett. 34, 28732875 (2009)
126. M. Liscidini, J.E. Sipe, Stimulated emission tomography. Phys. Rev. Lett. 111, 193602
(2013)
127. A. Eckstein, G. Boucher, A. Lematre, P. Filloux, I. Favero, G. Leo, J.E. Sipe, M. Liscidini,
S. Ducci, High-resolution spectral characterization of two photon states via classical
measurements. Laser Photonics Rev. 8, L76L80 (2014)
128. A. Christ, K. Laiho, A. Eckstein, K.N. Cassemiro, C. Silberhorn, Probing multimode
squeezing with correlation functions. New J. Phys. 13, 033027 (2011)
129. R.J. Glauber, The quantum theory of optical coherence. Phys. Rev. 130, 25292539 (1963)
130. R.J. Glauber, Coherent and incoherent states of the radiation eld. Phys. Rev. 131, 2766
2788 (1963)
131. R.H. Brown, R.Q. Twiss, Correlation between photons in two coherent beams of light.
Nature 177, 2729 (1956)
132. A. Migdall, D. Branning, S. Castelletto, Tailoring single-photon and multiphoton
probabilities of a single-photon on-demand source. Phys. Rev. A 66, 14 (2002)
133. J.H. Shapiro, F.N. Wong, On-demand single-photon generation using a modular array of
parametric downconverters with electro-optic polarization controls. Opt. Lett. 32, 26982700
(2007)
134. X.-S. Ma, S. Zotter, J. Kofler, T. Jennewein, A. Zeilinger, Experimental generation of single
photons via active multiplexing. Phys. Rev. A 83, 043814 (2011)
135. T. Jennewein, M. Barbieri, A.G. White, Single-photon device requirements for operating
linear optics quantum computing outside the post-selection basis. J. Mod. Opt. 58, 276287
(2011)
136. C. Xiong, T.D. Vo, M.J. Collins, J. Li, T.F. Krauss, M.J. Steel, A.S. Clark, B.J. Eggleton,
Bidirectional multiplexing of heralded single photons from a silicon chip. Opt. Lett. 38,
51765179 (2013)
137. T. Meany, L.A. Ngah, M.J. Collins, A.S. Clark, R.J. Williams, B.J. Eggleton, M.J. Steel, M.
J. Withford, O. Alibart, S. Tanzilli, Hybrid photonic circuit for multiplexed heralded single
photons. Laser Photonics Rev. 8, L42L46 (2014)
138. L. Mazzarella, F. Ticozzi, A.V. Sergienko, G. Vallone, P. Villoresi, Asymmetric architecture
for heralded single-photon sources. Phys. Rev. A 88, 023848 (2013)
139. J. Mower, D. Englund, Efcient generation of single and entangled photons on a silicon
photonic integrated chip. Phys. Rev. A 84, 052326 (2011)
140. P. Kumar, Quantum frequency conversion. Opt. Lett. 15, 14761478 (1990)
141. J. Huang, P. Kumar, Observation of quantum frequency conversion. Phys. Rev. Lett. 68,
21532156 (1992)
142. S. Tanzilli, W. Tittel, M. Halder, O. Alibart, P. Baldi, N. Gisin, H. Zbinden, A photonic
quantum information interface. Nature 437, 116120 (2005)
143. A.P. Vandevender, P.G. Kwiat, High efciency single photon detection via frequency up-
conversion. J. Mod. Opt. 51, 14331445 (2004)
144. M.A. Albota, F.N.C. Wong, Efcient single-photon counting at 1.55 microm by means of
frequency upconversion. Opt. Lett. 29, 14491451 (2004)
145. M.G. Raymer, K. Srinivasan, Manipulating the color and shape of single photons. Phys.
Today 65, 32 (2012)
146. C. Langrock, E. Diamanti, R.V. Roussev, Y. Yamamoto, M.M. Fejer, H. Takesue, Highly
efcient single-photon detection at communication wavelengths by use of upconversion in
reverse-proton-exchanged periodically poled LiNbO3 waveguides. Opt. Lett. 30, 17251727
(2005)
147. J. Pelc, L. Ma, C. Phillips, Q. Zhang, Long-wavelength-pumped upconversion single-photon
detector at 1550 nm: performance and noise analysis. Opt. Express 19, 2144521456 (2011)
148. J.S. Pelc, C. Langrock, Q. Zhang, M.M. Fejer, Influence of domain disorder on parametric
noise in quasi-phase-matched quantum frequency converters. Opt. Lett. 35, 28042806
(2010)
149. S. Ates, I. Agha, A. Gulinatti, I. Rech, M.T. Rakher, A. Badolato, K. Srinivasan, Two-photon
interference using background-free quantum frequency conversion of single photons emitted
by an InAs quantum dot. Phys. Rev. Lett. 109, 147405 (2012)
150. M. Rakher, L. Ma, O. Slattery, X. Tang, K. Srinivasan, Quantum transduction of
telecommunications-band single photons from a quantum dot by frequency upconversion.
Nature Photon. 4, 786791 (2010)
151. M.T. Rakher, L. Ma, M. Davano, O. Slattery, X. Tang, K. Srinivasan, Simultaneous
wavelength translation and amplitude modulation of single photons from a quantum dot.
Phys. Rev. Lett. 107, 083602 (2011)
152. H. Takesue, Erasing distinguishability using quantum frequency up-conversion. Phys. Rev.
Lett. 101, 173901 (2008)
153. R. Ikuta, Y. Kusaka, T. Kitano, H. Kato, T. Yamamoto, M. Koashi, N. Imoto, Wide-band
quantum interface for visible-to-telecommunication wavelength conversion. Nat. Commun.
2, 1544 (2011)
154. S. Zaske, A. Lenhard, C.A. Keler, J. Kettler, C. Hepp, C. Arend, R. Albrecht, W.-M.
Schulz, M. Jetter, P. Michler, C. Becher, Visible-to-telecom quantum frequency conversion
of light from a single quantum emitter. Phys. Rev. Lett. 109, 147404 (2012)
155. K. De Greve, L. Yu, P.L. McMahon, J.S. Pelc, C.M. Natarajan, N.Y. Kim, E. Abe, S. Maier,
C. Schneider, M. Kamp, S. Hfling, R.H. Hadeld, A. Forchel, M.M. Fejer, Y. Yamamoto,
Quantum-dot spin-photon entanglement via frequency downconversion to telecom
wavelength. Nature 491, 421425 (2012)
156. C. McKinstrie, J. Harvey, S. Radic, M. Raymer, Translation of quantum states by four-wave
mixing in bers. Opt. Express 13, 91319142 (2005)
157. K. Uesaka, K.K.-Y. Wong, Wavelength exchange in a highly nonlinear dispersion-shifted

bre: theory and experiments. IEEE J. Sel. Top. Quantum Electron. 8, 560568 (2002)
158. H.J. McGuinness, M.G. Raymer, C.J. McKinstrie, S. Radic, Quantum frequency translation
of single-photon states in a photonic crystal ber. Phys. Rev. Lett. 105, 093604 (2010)
159. A.H. Gnauck, R.M. Jopson, C.J. McKinstrie, J.C. Centanni, S. Radic, Demonstration of low-
noise frequency conversion by bragg scattering in a ber. Opt. Express 14, 89898994
(2006)
160. A.S. Clark, S. Shahnia, M.J. Collins, C. Xiong, B.J. Eggleton, High-efciency frequency
conversion in the single-photon regime. Opt. Lett. 38, 947949 (2013)
161. A. Guarino, G. Poberaj, D. Rezzonico, R. DeglInnocenti, P. Gnter, Electro-optically
tunable microring resonators in lithium niobate. Nature Photon. 1, 407410 (2007)
162. L. Lanco, S. Ducci, J.P. Likforman, M. Ravaro, P. Filloux, X. Marcadet, G. Leo, V. Berger,
Backward difference frequency generation in an AlGaAs waveguide. Appl. Phys. Lett. 89,
031106 (2006)
163. J. Han, P. Abolghasem, B.J. Bijlani, A. Helmy, Continuous-wave sum-frequency generation
in AlGaAs Bragg reflection waveguides. Opt. Lett. 34, 36563658 (2009)
164. K. Rivoire, Z. Lin, F. Hatami, W.T. Masselink, J. Vukovi, Second harmonic generation in
gallium phosphide photonic crystal nanocavities with ultralow continuous wave pump
power. Opt. Express 17, 2260922615 (2009)
165. C. Xiong, W. Pernice, K.K. Ryu, C. Schuck, K.Y. Fong, T. Palacios, H.X. Tang, Integrated
GaN photonic circuits on silicon (100) for second harmonic generation. Opt. Express 19,
1046210470 (2011)
166. Y. Ota, K. Watanabe, S. Iwamoto, Y. Arakawa, Nanocavity-based self-frequency conversion
laser. Opt. Express 21, 1977819789 (2013)
167. P.S. Kuo, J. Bravo-Abad, G.S. Solomon, Second-harmonic generation using quasi-
phasematching in a GaAs whispering-gallery-mode microcavity. Nat. Commun. 5, 3109
(2014)
168. I. Agha, M. Davano, B. Thurston, K. Srinivasan, Low-noise chip-based frequency
conversion by four-wave-mixing Bragg scattering in SiN(x) waveguides. Opt. Lett. 37,
29972999 (2012)
169. I. Agha, S. Ates, M. Davano, K. Srinivasan, A chip-scale, telecommunications-band
frequency conversion interface for quantum emitters. Opt. Express 21, 14761478 (2013)
170. X. Liu, R. Osgood, Y. Vlasov, W. Green, Mid-infrared optical parametric amplier using
silicon nanophotonic waveguides. Nature Photon. 4, 25 (2010)
171. S. Zlatanovic, J. Park, S. Moro, Mid-infrared wavelength conversion in silicon waveguides
using ultracompact telecom-band-derived pump source. Nat. Photonics 4, 561564 (2010)
172. D. Kielpinski, J.F. Corney, H.M. Wiseman, Quantum optical waveform conversion. Phys.
Rev. Lett. 106, 130501 (2011)
173. B. Brecht, A. Eckstein, A. Christ, H. Suche, C. Silberhorn, From quantum pulse gate to
quantum pulse shaperengineered frequency conversion in nonlinear optical waveguides.
New J. Phys. 13, 065029 (2011)
174. C.J. McKinstrie, L. Mejling, M.G. Raymer, K. Rottwitt, Quantum-state-preserving optical
frequency conversion and pulse reshaping by four-wave mixing. Phys. Rev. A 85, 053829
(2012)
175. D. Reddy, M. Raymer, C. McKinstrie, Efcient sorting of quantum-optical wave packets by
temporal-mode interferometry. Opt. Lett. 39, 29242927 (2014)
176. N. Gisin, G. Ribordy, W. Tittel, Quantum cryptography. Rev. Mod. Phys. 74, 145195
(2002)
177. C. Bennet, Quantum cryptography using any two nonorthogonal states. Phys. Rev. Lett. 68,
31213124 (1992)
178. V. Scarani, A. Acin, G. Ribordy, N. Gisin, Quantum cryptography protocol robust against
photon number splitting attacks for weak laser pulse implementations. Phys. Rev. Lett. 92,
057901 (2004)
179. ID-Quantique website, retrieved on 6. Aug 2012
180. P. Eraerds, N. Walenta, M. Legre, N. Gisin, H. Zbinden, Quantum key distribution and 1gbps
data encryption over a single ber. New J. Phys. 12, 063027 (2010)
181. I. Choi, R.J. Young, D. Townsend, Quantum key distribution on a 10Gb/s WDM-PON. Opt.
Express 18, 96009612 (2010)
182. NICT News, No. 401 Feb 2011 (2011)
183. K. Inoue, E. Wake, Y. Yamamoto, Differential-phase-shift quantum key distribution using
coherent light. Phys. Rev. A 68, 022317 (2003)
184. N. Namekate, H. Takesue, Y. Honjo, T. Tokura, S. Inoue, High-rate quantum key
distribution over 100 km using ultra-low-noise, 2-GHz sinusoidally gated InGaAs/InP
avalanche photodiodes. Opt. Express 19, 1063210639 (2011)
185. S. Wang, W. Chen, Z.-Q. Guo, Z.-Q. Yin, W.-H. Li, Z. Zhou, G.-C. Guo, Z.-F. Han, 2 GHz
clock quantum key distribution over 260 km of standard telecom ber. Opt. Lett. 37, 1008
1011 (2012)
186. NuCrypt website, retrieved on 6. Aug 2012
187. SECOQC, Business White Paper: Quantum Cryptography: An Innovation in the Domain of
Secure Information Transmission (2008)
188. J.F. Dynes, H. Takesue, Z.L. Yuan, A.W. Sharpe, K. Harada, K. Honjo, H. Kamada, O.
Tadanaga, Y. Nishida, M. Asobe, A.J. Shields, Efcient entanglement distribution over 200
kilometers. Opt. Express 17, 1144011449 (2009)
189. R. Ursin, F. Tiefenbacher, T. Schmitt-Manderbach, H. Weier, T. Scheidl, M. Lindenthal, B.
Blauensteiner, T. Jennewein, J. Perdigues, P. Trojek, B. Omer, M. Furst, M. Meyenburg,
J. Rarity, Z. Sodnik, C. Barbieri, H. Weinfurter, A. Zeilinger, Entanglement-based quantum
communication over 144 km. Nature Phys. 3, 481 (2007)
190. N. Namekate, H. Takesue, Y. Honjo, T. Tokura, S. Inoue, High rate, long-distance quantum
key distribution over 250 km of ultra low loss bres. Opt. Express 19, 1063210639 (2011)
191. J.G. Rarity, P.R. Tapster, P.M. Gorman, P. Knight, Ground to satellite secure key exchange
using quantum cryptography. New J. Phys. 4, 82 (2002)
192. J.E. Nordholt, R.J. Hughes, G.L. Morgan, C.G. Peterson, C.C. Wipf, Present and future free-
space quantum key distribution. Proc. SPIE 4635, 116 (2002)
193. A. Laing, V. Scarani, J.G. Rarity, J.L. OBrien, Reference-frame-independent quantum key
distribution. Phys. Rev. A 82, 012304 (2010)
194. R. Ursin, T. Jennewein, J. Kofler, J.M. Perdigues, L. Cacciapuoti, C.J. de Matos, M.
Aspelmeyer, A. Valencia, T. Scheidl, A. Fedrizzi, A. Acin, C. Barbieri, G. Bianco, C.
Brukner, J. Capmany, S. Cova, D. Giggenbach, W. Leeb, R.H. Hadeld, R. Laflamme, N.
Lutkenhaus, G. Milburn, M. Peev, T. Ralph, J. Rarity, R. Renner, E. Samain, N. Solomos,
W. Tittel, J. P. Torres, M. Toyoshima, A. Ortigosa-Blanch, V. Pruneri, P. Villoresi, I.
Walmsley, G. Weihs, H. Weinfurter, M. Zukowski, A. Zeilinger, Space quest: Experiments
with quantum entanglement in space. IAC Proceedings 08, A2.1.3 (2008)
195. H.-J. Briegel, W. Dr, J.I. Cirac, P. Zoller, Quantum repeaters: the role of imperfect local
operations in quantum communication. Phys. Rev. Lett. 81, 59325935 (1998)
196. A. Martin, O. Alibart, M.P. De Micheli, D.B. Ostrowsky, S. Tanzilli, A quantum relay chip
based on telecommunication integrated optics technology. New J. Phys. 14, 025002 (2012)
197. A.S. Clark, J. Fulconis, J.G. Rarity, W.J. Wadsworth, J.L. OBrien, All-optical-ber
polarization-based quantum logic gate. Phys. Rev. A 79, 030303 (2009)
198. C. Simon et al., Quantum memories. a review based on the European integrated project Qubit
Applications (QAP). Eur. Phys. J. D 58, 122 (2010)
Chapter 13
Biphoton Pulse Shaping
Joseph M. Lukens and Andrew M. Weiner
Abstract Recent research has demonstrated how classical optical signal-processing

techniques can be extended to nonclassical entangled-photon states, permitting
unprecedented control of the time-frequency correlations shared by these light
quanta. In this chapter, we introduce the basic theory behind such biphoton pulse
shaping and highlight several key experiments conducted in this eld. Our emphasis
is not only on how entangled photons benet from these traditionally classical
techniques, but also how their specically quantum properties produce interesting
effects not observable with classical elds. We consider both Fourier-transform pulse
shaping, which relies on programmable spectral ltering, and electro-optic modu-
lation, in which the temporal phase or amplitude of the biphoton is manipulated by an
electrical signal. Both avenues of research have facilitated new insights into biphoton
correlations and look to play an important role in the future of quantum state
manipulation and the next generation of quantum communication networks.
13.1 Introduction
The groundbreaking experiments of Hanbury Brown and Twiss in the 1950s [1, 2]
highlighted a grave need in the optical community to better understand the corre-
lations shared by multiple photons. Virtually all optical interference experiments
prior to that time had consisted of rst-order intensity measurements, which
quantum mechanically could be understood as the interference of a single photon
with itself. Yet Hanbury Brown and Twiss instead examined the correlations
between two photoelectric detectors, nding that a pair of photons from a thermal
light source are twice as likely to arrive together than at a specic nonzero delay,
J.M. Lukens (&) A.M. Weiner

Purdue University, West Lafayette, IN, USA
e-mail: jlukens@purdue.edu
A.M. Weiner
e-mail: amw@purdue.edu

424 J.M. Lukens and A.M. Weiner
an effect now termed bunching [3]. The surrounding debate [47], coupled with the
advent of the laser, prompted Glauber to formulate the rst fully quantized theory of
optical coherence in 1963 [8, 9], and the modern era of quantum optics was born.
Since then, a particularly influential nonclassical state of light has proven to be
the entangled photon pair, or biphoton [10]. Although such photons can be
entangled in many possible degrees of freedom, in this paper our focus is on time-
frequency entanglement. This is frequently produced through spontaneous para-
metric downconversion (SPDC) of a narrowband pump eld, in which a single
pump photon decays into two daughter photons traditionally denoted signal and
idler. Whereas individually these photons possess an essentially random creation
time, collectively their detection times are highly correlated, since they share the
same birth; likewise, their frequencies can assume any value from a potentially
wide bandwidth, but once the energy of one photon is measured, the value of its
partner is known with high precision. Figure 13.1 provides a sketch of the gener-
ation process and joint spectro-temporal distributions of such biphoton states. The
simultaneously narrow correlations in both time and frequency variables violate the
seemingly reasonable tenet of classical locality; although critiqued by even the likes
of Einstein [11], such entanglement has proven essential in foundational tests of
quantum mechanics [12] and even some forms of quantum cryptography [13].
As might be expected, a key tool for characterizing biphoton entanglement
derives from Glaubers coherence theory, particularly the second-order correlation
function in either temporal or spectral form. Such an entangled state displays
super-bunching, for the probability of simultaneous photon detections can be
radically higher than the 2:1 contrast observed for thermal light [3]. And while
biphoton characterization is certainly of great importance in itself, the last decade
has also witnessed the emergence of pulse-shaping techniques for biphoton
manipulation, which exploit methods previously developed in classical ultrafast
optics [14]. The purpose of this chapter is to review these recent advances in
biphoton pulse shaping, emphasizing both the impact they have already had on the
eld of quantum optics and their promise in future applications in quantum
(a) (b) (c)

Pump
SPDC + =
Idler Frequency
Idler Time
Signal Frequency Signal Time
Fig. 13.1 a Biphoton generation through downconversion of a narrowband pump eld.

b Resulting biphoton spectral correlations. c Joint temporal correlations. Even though individual
photons have a wide spread of spectro-temporal content, they are highly correlated with their
partner
13 Biphoton Pulse Shaping 425
information. Additionally, although pulse shaping is typically associated with

programmable ltering in the spectral domain to achieve shaping in time, for the
purposes of this work we will also generalize the term to include its Fourier dual:
modulation in the time domain to transform the spectrum. Both specializations
demonstrate how classical communication techniques can assume new roles in
quantum optics.
In Sect. 13.2 we introduce the basics of classical Fourier-transform pulse shaping
and temporal modulation, showing thereafter in Sect. 13.3 how these techniques
can be applied to biphotons. Many of the key experiments are summarized in
Sect. 13.4, with two specic examples taken up in detail in Sects. 13.5 and 13.6.
Finally, in Sect. 13.7 we provide an outlook for future work, with an emphasis on
applications in quantum cryptography.
13.2 Classical Pulse Shaping
Biphoton pulse shaping represents the union of two optical disciplines which his-
torically have remained largely independent of one another: the study of quantum
states of light and the development of pulse-shaping techniques for manipulation of
classical light sources. Since most of the biphoton shaping examples covered in this
chapter can be viewed as outgrowths of similar experiments on classical pulses, we
focus in this section on the classical foundations of both Fourier-transform pulse
shaping and electro-optic modulation, introducing the basic theory behind these
technologies which have proven so valuable in a variety of optical applications.
This discussion furnishes a springboard for their extension to entangled photons, a
task assumed in Sect. 13.3.
Following its development in the 1980s [15], Fourier-transform pulse shaping
has enabled major advances in an assortment of elds, ranging from optical arbi-
trary waveform generation to coherent control, single-cycle pulse generation to
microwave photonics [16]. And whereas synthesizing user-dened optical wave-
forms on the femtosecond timescale is impossible with the fastest available elec-
tronics, Fourier-transform pulse shaping permits the generation such arbitrary elds
through parallel manipulation of the frequency content. An archetypal pulse shaper
is shown in Fig. 13.2a. The frequency components of a broadband input eld are
spatially dispersed via a diffraction grating and focused to small spots at a plane
where a programmably controlled spatial light modulator (SLM) is placed. Due to
the physical separation of distinct frequencies, it is then possible to control the
amplitude and phase at each wavelength independently; after recombining the
shaped frequency components with a second lens and grating, the desired waveform
is produced at the output, whose complexity is limited only by the spectral reso-
lution and total bandwidth of the pulse shaper [16]. Mathematically, if we take the
complex input and output time-domain elds as Ein t and Eout twith associated
spectra Ein x and Eout xthe effect of the pulse-shaper mask H x is to pro-
duce the output
Fig. 13.2 a Schematic of a prototypical pulse shaper. b Example shaped power spectrum (left)
and temporal intensity (right) for a square pulse (Image in a from [18] and that in b from [15])
Z
Eout t dxEin xHxeixt ; 13:1
so that Fourier-transform pulse shaping can be viewed as a general method to

achieve nearly arbitrary and ultrafast linear transformations on optical elds. Fig-
ure 13.2b shows just one example from the early days of pulse shaping, in which a
xed spatial mask was used to generate a square pulse [15]. The measured spectro-
temporal properties are in good agreement with theory, even displaying the
expected overshoot resulting from truncation of the ideal sinc spectrum. The basic
model for pulse-shaper operation embodied in (13.1) proves sufcient for the
purposes of this chapter; for more detailed discussion of pulse-shaper theory and
technology, the reader is directed to the reviews in [14, 16, 17]. The key point here
is that Fourier-transform pulse shaping manipulates an optical eld through fre-
quency-domain multiplication by a user-dened lter.
The Fourier dual shaping process is effected through direct modulation of the
temporal phase or intensity of an incident optical eld. Although a range of
modulator technologies have been considered in practice, we specialize our dis-
cussion to devices based on the electro-optic effect, as these have been employed
exclusively in the biphoton modulation experiments described later in this chapter.
In electro-optic materials, the refractive index changes with an applied external
voltage, in turn permitting electrical control of the phase acquired by an optical eld
propagating through it. Figure 13.3a shows a cross-section of a modulator based on
lithium niobate [19], which possesses a particularly strong linear electro-optic
response along its z-axis. In the arrangement depicted here, an electric voltage
applied across the signal and ground electrodes induces a refractive index change
for light polarized along the z-direction and propagating along y in the embedded
waveguide. Furthermore, by splitting the input optical eld into two armseach of
the form of Fig. 13.3aand recombining them after different phase shifts, it is
possible to convert this phase modulation to intensity modulation, via the Mach-
Zehnder interferometer conguration of Fig. 13.3b [19]. And by utilizing traveling-
wave electrodes, in which the driving electrical signal propagates parallel to the
optical waveguide, extremely high modulation bandwidths exceeding tens of
gigahertz are possible, permitting rapid switching of either intensity or phase in
Fig. 13.3 a Cross-section of an x-cut lithium niobate phase modulator. b Arrangement for
combining two phase-modulation arms to produce an intensity modulator. c Example comb spectra
generated by cascading three phase modulators and one intensity modulator. The spectral line
spacing, from top to bottom, is 7, 10, and 17 GHz (Images in a and b from [19] and c from [20])
high-speed optical communication networks. Formally, we can cover both phase

and intensity cases through a general modulation function mt, so that the input-
output complex eld relationship can be expressed as
Eout t Ein tmt; 13:2
where, for example, sinusoidal phase modulation at frequency xm is represented by

a function of the form mt expid sinxm t /, with d signifying the modu-
lation index, a quantity which is proportional to the applied voltage amplitude.
In addition to their extensive use in transmission of data [21], electro-optic
modulators have also found use as time lensesdevices which can compress and
spread waveforms in time just as lenses do in space [22]and frequency-comb
generators. In the latter case, cascaded phase and intensity modulators driven by a
single microwave source can be used to convert a monochromatic input spectrum
into a series of discrete spectral lines; Fig. 13.3c provides an example in which one
intensity modulator and three phase modulators generate broad and flat frequency
combs with tunable line spacing [20]. Such optical sources are well-suited for
applications including line-by-line pulse shaping [23] and radio-frequency photonic
ltering [24].
In light of the maturity of these manipulation techniques for both spectral and
temporal control of classical elds, the motivating question behind biphoton pulse
shaping is: how can we apply such classical approaches to entangled photons? As
highlighted in the experiments discussed below, pulse shaping indeed reveals
interesting physics, with potential applications in quantum information. Yet the
quantum extension involves a more rened understanding of what is entailed by the
term pulse shaping when nonclassical sources are considered. Accordingly, we
tackle the fundamentals of biphoton shaping theory in the next section.
13.3 Biphoton Pulse Shaping: Theory
As noted in the introduction, SPDC of a monochromatic pump can be used to

generate entangled photons that share both high temporal and high spectral cor-
relation. In order to derive an explicit form for the resultant state, we invoke the
formalism of the interaction picture of quantum mechanics [25], in which the
quantum state evolves in time according to an interaction Hamiltonian (in our case a
second-order nonlinear interaction), and the eld operators according to the free
Hamiltonian (simply related to linear propagation of the corresponding classical
elds). For a pump frequency xp 2x0 , a perturbative calculation [25] yields the
entangled state
Z

jWi M vacis vacii dX/Xjx0 Xis jx0 Xii ; 13:3
where M * 1, vac denotes the vacuum state, s and i signify signal and idler
modes, and /X is a complex weight function determined by phase-matching
conditions. In this form we have neglected any spatial or polarization dependencies,
which prove unimportant for the experiments considered in this review. The vac-
uum term, with a near-unity weighting coefcient, signies the relative improba-
bility of this spontaneous process; downconversion efciencies of only up to about
105 have been reported in the experiments discussed below [26]. Of course the
more interesting term in (13.3) is the second one, which consists of a superposition
of signal and idler frequency combinations, subject to the constraint that they must
sum to 2x0 . Since /X can be extremely broadband, a wide range of frequency
pairs are possible, thereby producing strong entanglement. The degree of such
entanglement can be operationally dened as the ratio of the single-photon band-
width to that conditioned on a frequency measurement of its partner [3]; for a
separable (non-entangled) state, this value is one, whereas for an entangled state
with terahertz-scale bandwidth and perhaps megahertz pump linewidthtypical for
the experiments considered herethe ratio can be well over a million. Such
massive entanglement lies at the heart of the nonclassical behavior which these
states can display.
For the Fourier-transform pulse-shaping experiments, manipulation and mea-
surement of the time-frequency biphoton state follows the basic setup in Fig. 13.4a.
Signal and idler photons are sent through programmable spectral lters, Hs x and
Hi x, and coincidences between two detectors are measured as the signal delay is
varied. Assuming the detectors are ideali.e., possess innite timing resolution and
respond to all frequencies of the incoming eldthe coincidence rate is directly
proportional to the second-order Glauber correlation function [8]
^ i tE
G2 s hWjE ^ s t sE ^ i tjWi:
^ s t sE 13:4
(a)
D1
2
D2
(b) (c)
D1
2 1 2)
2
D2
Fig. 13.4 Basic experimental setups for biphoton manipulation. a Pulse shaping. Each photon of
an entangled pair is sent through a programmable spectral lter and detected on a photon counter.
The coincidences between these (ideal) detectors map out the second-order Glauber correlation
function G2 s. b Free-running electro-optic modulation. The photons are modulated in the time
domain, with spectral correlations determined through coincidences between monochromators.
This experiment represents the Fourier dual to that in a. c Modulation of a heralded single photon.
Now the detection of the idler photon is used to trigger modulation of the signal, with G2 s
representing the measured quantity, as in a
The positive-frequency electric eld operators E ^ s t s and E

^ i t are
associated with annihilation of a signal or idler photon at time t s or t, respec-
tively; the corresponding negative-frequency operators create photons at these
times. Since the monochromatic pump is statistically stationary, the correlation
function depends only on the delay s, making the choice of t arbitrary. Moreover,
because the positive- and negative-frequency operators are related through Her-
mitian conjugation, the classical expression for G2 s, in which the electric eld
operators become complex numbers, is of the form of an intensity correlation
between the signal and idler waveforms, so we can view G2 s as the general
quantum mechanical version of an intensity cross-correlation measurement.
When the quantum state contains no more than two photons, G2 s can be
conveniently expressed as the modulus squared of a biphoton wavepacket
wt s; t, dened as [10]
wt s; t hvacjE ^ i tjWi:
^ s t sE 13:5
For the quantum state in (13.3) and the setup in Fig. 13.4a, the wavepacket
assumes the form [26]
Z
ws dX /XHs x0 XHi x0 XeiXs ; 13:6
where we have neglected an unimportant overall constant and a t-dependence

which vanishes on taking the modulus G2 s jwsj2 . This expression repre-
sents the main relation for biphoton pulse shaping and well describes the spectral
ltering experiments in this review. Therefore, analogous to a classical pulse [16],

the signal-idler temporal correlations can be shaped through programmable lter-
ing, permitting synthesis of waveforms orders of magnitude too fast for even the
speediest electronics. Equation (13.6) also indicates why extremely broadband bi-
photons are so desirable in Fourier-transform pulse shaping; for a given pulse-
shaper spectral resolution, the wider the bandwidth of /X, the larger the number
of independent spectral features which can be programmed on the biphoton spec-
trum. As in classical pulse shaping, the ratio of total bandwidth to spectral reso-
lution provides a measure of the maximum complexity of the generated waveforms
[16], so high-bandwidth biphotons prove most amenable to spectral shaping.
However, unlike the classical case summarized in (13.1), the net transfer function is
the product of the complex lters applied to the signal and idler photons, evaluated
at each entangled frequency combination. Thus the ltering applied to the signal
can actually be undone by ltering the idler with the inverse; for example, even
when both photons are dispersively broadened individually, it is possible that
collectively their correlations remain unaltered, an effect known as nonlocal dis-
persion cancellation [27] and considered in more detail in Sect. 13.5.
The second class of experiments we examine here is essentially the Fourier dual
to the pulse-shaping arrangement above. A schematic of this general setup is
provided in Fig. 13.4b. Now the signal and idler photons are multiplied in the time
domain by the modulation functions ms t and mi t, and then the spectral corre-
lations are measuredi.e., the coincidences between two narrow frequency bins. In
the absence of modulation, the entangled state in (13.3) indicates that only fre-
quency pairs xs ; xi satisfying xs xi 2x0 will register coincidences; how-
ever, the presence of the modulators can introduce new frequency correlations,
quantied by the second-order cross-spectral density:

H2 x1 ; x2 WE^ i x2 E
^ s x1 E ^ i x2 W :
^ s x1 E 13:7
These operators are the Fourier transforms of those in (13.4), with negative- and
positive-frequency arguments signifying photon creation and annihilation, respec-
tively. Therefore H2 x1 ; x2 is proportional to the probability of detecting a
signal photon at frequency x1 and an idler at x2 . Additionally, in the same fashion
as the temporal case, this correlation function can be decomposed into the modulus
squared of a biphoton spectral amplitude, denoted by hx1 ; x2 :
^ s x1 E
hx1 ; x2 hvacjE ^ i x2 jWi: 13:8
The effect of temporal modulation is to produce convolution in the frequency

domain, and in general the full expression for H2 x1 ; x2 in the arrangement of
Fig. 13.4b can become quite complicated. Yet if we restrict ourselves to the case in
which the biphoton correlation time is much shorter than the scale over which the
modulators vary, we can treat the inverse Fourier transform of /X as an impulse,
giving for the biphoton spectrum
Z
hD dt ms tmi teiDt ; 13:9
where we have neglected irrelevant overall constants and dened

D x1 x2 2x0 , the deviation of the sum frequency from that of the nar-
rowband pump. In analogy with the spectral ltering case in (13.6), the effect of
signal and idler modulation is cumulative, with the potential for one modulator to
nonlocally cancel the other in the measured spectral correlation [28].
In the above arrangement, the signal and idler modulators are free-running in the
sense that they are not synchronized with photon emission times. As an alternative
for biphoton manipulation, also useful for temporal shaping, the detection of one
photon can be used to trigger modulation of the other. Figure 13.4c shows how an
experiment like this could work; detection of the idler photon prompts a voltage
signal to be applied to the signal modulator, and then the temporal correlation
2
function is measured. If G0 s represents the correlation function with no mod-
ulation, the modied function measured is [29]
2
G2 s jms sj2 G0 s; 13:10
where, as before, ms s is the complex modulation applied to the signal photon, but
now it is a function of the relative signal-idler delay s instead of some global time
variable, since modulation is triggered only by detection of the idler photon. While
causality limits modulator operation to delays satisfying s [ 0, this restriction can
be effectively mitigated by delaying the signal photon prior to modulation so that its
entire duration falls in positive delays. A similar capability is often tacitly exploited
in Fourier-transform pulse shaping as well; due to the physical delay of the pulse
shaper itself, supercially non-causal spectral lters can be implemented by simply
subtracting off the overall delay in dening the output time origin.
In the form of (13.10), one can view such manipulation as the shaping of a
heralded single photonthat is, a photon whose presence is announced through
detection of its entangled partner. We note that while the free-running modulator
setup of Fig. 13.4b requires biphotons with correlation times much shorter than the
modulation period for (13.9) to hold, this alternative detection setup relies on
biphotons with the opposite propertynamely, correlation times greatly exceeding
the modulator period. Under this condition, the modulator can signicantly alter the
shape of G2 s. Thus, just as broadband biphotons are optimal for Fourier-
transform pulse shaping in the setup of Fig. 13.4a, temporally long biphotons prove
advantageous for electro-optic modulation, as they maximize the number of inde-
pendent features which can be imprinted by the modulator. And so the theory here,
typied in (13.6), (13.9), and (13.10), has motivated several experimental tests of
Fourier-transform pulse shaping and high-speed electro-optic modulation of bi-
photon statesboth well-established techniques in ultrafast photonics and optical
communications, but only considered in the quantum optics community within the
last decade.
13.4 Biphoton Pulse Shaping: Important Experiments
13.4.1 Ultrafast Coincidence Detection
In the conceptual pictures of Fig. 13.4, we have envisioned correlating the pho-
toelectric pulses generated from two single-photon counters in order to determine
coincident photon arrivals. In the spectral measurement of Fig. 13.4b, such a
scheme poses no problems, for high frequency resolution is attainable with well-
chosen spectral dispersers; the timing resolution of the detector is not a limiting
factor. However, since most single-photon detectors suffer from timing jitters of
tens of picosecondsor morea realistic implementation of Fig. 13.4a is unable to
resolve biphoton correlations on the few- and sub-picosecond time scales. And
although there are several interesting examples of shaping experiments with such
slow detectors [3032], the full potential of programmable pulse shaping is only
realized on the ultrafast time scale. This timing resolution problem permeates
classical optics as well, where correlation techniques based on nonlinear optical
interactions have facilitated high-resolution temporal measurements of ultrashort
pulses [14]. Borrowing from these ideas, one naturally hopes for a nonlinear pro-
cess at the single-photon level which would likewise provide sufcient temporal
resolution to characterize sub-picosecond biphoton correlations.
Such a technique has indeed been implemented: biphoton sum-frequency gen-
eration (SFG). In this process, signal and idler photons are mixed on a second
nonlinear crystal; with some small probability, temporally coincident photons can
then recombine and generate a new photon at the sum frequency [33]. Since only
photons that overlap within a femtosecond timescale can mix, the rate of upcon-
version in a properly designed crystal is proportional to the Glauber correlation
function, and so by sweeping the delay of the signal with respect to the idler,
G2 s is directly obtained. In this way SFG enforces temporal gating optically,
instead of electronically, thereby enabling huge improvements in timing resolution.
Unfortunately, since the interacting elds have extremely low intensities, the
efciency of biphoton SFG is quite low, with a record upconversion rate of only
105 [26]. Yet such low efciencies can be somewhat compensated by generating
an ultrahigh flux of entangled photons [33]. As a general rule, in order to observe
quantum mechanical effects, one must ensure that only a single biphoton is located
within a given detection window; otherwise, there will result spurious coincidences
in which the two detections actually correspond to different photon pairs. Since
photons from separate pairs are not entangled, the desired quantum correlations are
degraded. This limiting timescale for separate detectorsas in Fig. 13.4ais
electronic in origin, which for 1-ns resolution, say, xes the maximum biphoton
generation rate at *109 s1. On the other hand, with ultrafast SFG which measures
the biphoton directly, we need only ensure that successive biphotons themselves do
not temporally overlap. With correlation times below 1 ps quite common, the
maximum allowable quantum flux can exceed 1012 s1, permitting upconversion
rates that are easily detectable even at such low SFG efciencies [33]. For more
details regarding the conditions under which biphoton SFG accurately reproduces
the Glauber correlation function, the reader is directed to the theoretical develop-
ment in [34].
Accordingly, the rst demonstration of successful biphoton SFG [33] provided
the necessary basis for biphoton pulse shaping, which itself was realized shortly
thereafter [35]. The original experimental setup is reproduced in Fig. 13.5a. Signal
and idler photons are emitted collinearly, with a prism sequence removing
remaining pump light and spectrally decomposing the biphoton; by placing a spatial
light modulator (SLM) in the path of the dispersed photons, the spectral phase of
the signal and idler can be programmably controlled. (Since signal and idler are
distinguished by frequency, both lters can be implemented on a single pulse
shaper.) The shaped biphotons are then recombined on a second nonlinear crystal,
and the SFG photons are measured by a single-photon counting module (SPCM).
Additionally, because a shift in time is equivalent to linear phase in the frequency
domain, the pulse shaper can also impose the relative photon delay, omitting the
need for a dedicated delay stage. This general Fourier relationship is readily derived
by considering the linear spectral phase lter Hx expixs in (13.1), which
gives
Z
Eout t dx Ein xeixts Ein t s; 13:11
so that, indeed, the output eld is simply a time-shifted version of the input. Thus
by applying similar linear phase terms to the pulse shaper, it is possible to directly
control the delay of one photon with respect to the other.
Figure 13.5b shows some results from this experiment [35]. With linear phase
swept on the pulse shaper, but no ltering otherwise, a narrow correlation peak is
found; by adding a p phase jump halfway through the signal spectrum, this single
correlation peak is transformed into a doublet. These seminal results represent the
foundation of biphoton pulse shaping, upon which subsequent researchers have
expanded with more sophisticated ltering or efcient detection.
Fig. 13.5 First demonstration of biphoton pulse shaping. a Experimental setup. b Shaping results.
In the left case, the pulse shaper sweeps the relative signal-idler delay, allowing measurement of a
narrow temporal correlation function. On the right, an additional p phase shift is added halfway
through the signal spectrum, turning this peak into a doublet (Images from [35])
13.4.2 Additional Biphoton Shaping Experiments
The rst demonstration of biphoton pulse shaping [35] utilized a relatively low-
resolution, phase-only pulse shaper, and succeeding work has focused at least in
part on enhancing these capabilities. In 2008, the rst biphoton pulse shaper which
controlled both amplitude and phase was implemented [36]. One interesting con-
sequence of independent phase and amplitude control is the ability to simulate a
Franson interferometer [37]. In the typical Franson arrangement, signal and idler
photons are sent through separate Mach-Zehnder interferometers (MZIs), identical
except for phase shifts in their long arms, with coincidences recorded between one
output port of each MZI. If the relative delay between the long and short arms of
each MZI exceeds the single-photon coherence time, but not that of the original
pump, the single detector rates show no interference while the coincidence rate
does, due to the interference of indistinguishable quantum paths [38]. Yet because
an MZI represents a linear, time-invariant system, it can instead be implemented in
the frequency domain, by programming the associated amplitude and phase on a
pulse shaper, as rst demonstrated in [36], indicating a new application of biphoton
shapingnamely, stable interferometer emulation. This basic setup was recently
employed for creating and characterizing frequency-entangled biphoton states with
programmable dimensions, through partitioning of the biphoton spectrum into
discrete bins with the pulse shaper [39].
Pulse-shaping capabilities were further broadened by our experiments utilizing a
high-resolution, commercial telecom pulse shaper based on liquid-crystal-on-silicon
technology [40]. In this work, we were able to demonstrate complex amplitude
coding for efcient verication of spectral entanglement, as well as additional
pulse-shaper impersonations of various interferometers [32]. Due to the ultrahigh
spectral resolution, much more complicated binning can be achieved; whereas
signal and idler were divided into up to four bins each in [39], length-40 amplitude
codes were readily demonstrated with the arrangement in [32]. In subsequent
experiments with this pulse shaper [41], we have also been able to demonstrate the
temporal Talbot effect with biphotons, in which a periodic correlation function is
replicated on propagating through discrete amounts of optical dispersion [42]; in
this example, the pulse shaper simultaneously creates and manipulates a periodic
correlation function through spectral amplitude and phase ltering. Although the
correlation trains generated in this way are limited in maximum duration by the
pulse-shaper time aperture (around 50 ps in this experiment), such an approach
nonetheless offers a means to yield periodic biphotons with extremely high and
tunable repetition rates, without the need for a stable optical cavity.
In parallel with these developments in Fourier-transform pulse shaping, electro-
optic modulation of biphoton sources was demonstrated rst in 2008 [29], using a
setup similar in form to Fig. 13.4c. The generation processbased on the phe-
nomenon of electromagnetically induced transparencyproduces biphotons with
long correlation times approaching a microsecond, well-suited for temporal mod-
ulation and direct measurement with electronic detectors. Accordingly, a range of
temporal shapes for the heralded single photon were demonstrated, including
square, Gaussian, and exponential pulses. With slight modications to this setup,
electro-optic modulation was also exploited to perform a Fourier-based measure-
ment of the temporal correlation function [43]. These examples, coupled with the
developments in Fourier-transform pulse shaping, provide just a sampling of the
unique potential of spectro-temporal biphoton shaping techniques. In the next
sections we give a more detailed discussion of two particularly interesting classes of
programmable shaping and modulation, which we hope will further highlight the
fascinating aspects of biphoton pulse shaping. The rst, cancellation of dispersion
or modulation, emphasizes the truly nonclassical properties afforded by entangled
states; the second, biphoton coding, shows how these properties may be exploited
in quantum information. Both cases rely on a combination of classical techniques
and quantum attributes.
13.5 Detailed Example I: Cancellation of Dispersion

or Modulation
The possibility for nonlocal cancellation of dispersionthe spreading of broadband

pulses on propagation through a mediumwas mentioned briefly in Sect. 13.3, as a
consequence of the combined effects of the signal and idler spectral lters. In
general, dispersion results from the variation of group velocity with frequency, so
that the spectral components of a short pulse travel through a dispersive medium at
different speeds, thereby separating on propagation and causing the pulse to spread.
Such spreading is closely related to the spectral phase imparted by the medium; if
we assume a phase-only transfer function of the form Hx expiUx, the
group delay (under our time-harmonic convention) follows via [16]
@U
sx ; 13:12
@x
so that any variation of the spectral phase which is second order or higher in x
produces a frequency-dependent delayi.e., the frequency components separate in
time. This concept extends to broadband entangled photons as well, ordinarily
causing the correlation function to spread. But as shown theoretically by Franson
[27], if the signal and idler photons travel through separate media with second-order
dispersion coefcients equal in magnitude but opposite in sign, the joint correlation
function is unaffected, even when the photons are arbitrarily far apart. Although some
aspects of this effect have been shown using electronic coincidence detection [44], it
was not until utilization of ultrafast biphoton SFG that dispersion cancellation could
be observed on a femtosecond scale [45]. With the variety of low-loss dispersive
media available, programmable pulse shaping is in no way required to exhibit this
effect; indeed, prism sequences alone proved sufcient in the rst observation of
ultrafast dispersion cancellation [45]. However, these initial experiments considered
only cancellation of second-order spectral phase, whereas control of higher orders

proves essential for compressing the correlations of ultrabroadband biphotons [46].
Cancellation of such arbitrary phase is virtually impossible with standard optical
media, but it is achievable with Fourier-transform pulse shaping.
The value of such programmable pulse shaping for high-order dispersion
compensation was rst demonstrated on classical pulses. In one experiment [47], a
pulse shaper was exploited to achieve distortion-free transmission of an approxi-
mately 500-fs pulse through *50 km of single-mode ber. Figure 13.6 shows the
results from this test; after propagation through the 50-km ber link followed by a
matched dispersion-compensating ber module, the pulse shows an additional tail
due to uncompensated higher orders, since the dispersion-compensating ber
cannot simultaneously match all second- and third-order phase. But by inserting a
pulse shaper to apply third-order phase, the original bandwidth-limited pulse is
recovered. This example stresses both the importance of high-order dispersion
management and the difculty in achieving it with standard optical materials and
components. Thus the experiments considering high-order dispersion cancellation
for biphotons can be viewed as quantum extensions of these classical tests, with the
fundamental difference that they can be achieved on separated photons.
To understand explicitly how arbitrary-order biphoton dispersion cancellation
ensues, we specialize (13.6) to phase-only lters Hs x expiUs x and
Hi x expiUi x, expressing each phase function as a Taylor series about x0 :
n
X
1
Us;i
Us;i x x x0 n : 13:13
n0
n!
Fig. 13.6 Using a pulse shaper to compensate for high-order dispersion. After propagation
through *50 km of single-mode ber and the matched length of dispersion-compensating ber,
the measured pulse is still distorted due to residual third-order spectral phase. Yet this is
compensated by the pulse shaper, retrieving the original *500-fs pulse (Result from [47])
Under these conditions, the measured biphoton wavepacket is

Z ( )
X1
1 h n n n
i
w s dX/X exp iXs i U 1 Ui X ; 13:14
n0
n! s
so that high-order dispersion cancellation is achieved whenever the signal-idler

n
expansion coefcients satisfy Usn 1n Ui 0 for n 2; 3; . . .. [The n 0; 1
cases do not impact the overall shape of G2 s.] Interestingly, opposite signs of
dispersion give cancellation for even orders only, as odd orders instead require
identical phase coefcients to achieve compression. We note that this cancellation
behavior stands in stark contrast to dispersion compensation with classical pulses,
where all orders must be flipped in the compensating medium. Yet nonetheless this
biphoton cancellation condition can be understood intuitively as well. Because of
the spectral correlation shared by these entangled photons (the xed sum frequency
of 2x0 ), each signal photon at x0 X is correlated with the idler at x0 X, so that
we would naturally expect the correlation function to remain unchanged when each
of these signal-idler frequency pairs are delayed the same amount, that is, when
ss x0 X si x0 X; 13:15
where here ss and si are the frequency-dependent group delays for the signal and
idler photons. And since these delays are related to the corresponding spectral
phases through the derivative operation of (13.12) (which flips the parity), can-
cellation should be achieved whenever the signal-idler spectral phases are anti-
symmetric with respect to each other:
Us x0 X Ui x0 X: 13:16
This equation leads immediately to the alternating-sign condition found previ-

ously through direct calculation of the biphoton wavepacket, justifying the validity
of this more heuristic interpretation.
Utilizing the pulse shaper from [32], we have been able to verify this fascinating
behavior for high-order spectral phase, making use of highly efcient periodically
poled lithium niobate waveguides for both SPDC and biphoton SFG [26]. The basic
setup mirrored that in Fig. 13.5a, although the prism sequence was replaced by
optical ber and a ber-pigtailed pulse shaper. The results for high-order dispersion
cancellation are reproduced in Fig. 13.7, showing examples for the orders
n 3; 4; 5. With the signal portion of the spectrum programmed with a particular
order of spectral phase, and nothing applied to the idler, a lowered and stretched
correlation function is measured (signal dispersed cases). Similarly, when the
complementary phaseequal in the odd-order cases and with a sign flip in the even
casesis applied to the idler, but nothing on the signal, another dispersed wave-
packet is observed (idler dispersed cases). These results can be interpreted as
individually spreading one photon or the other. However, when both photons are
5000
(a) (b) (c)
Both Both Both
Counts [s1]
Dispersed Dispersed Dispersed

2500
Signal Idler Signal Idler Signal Idler
Dispersed Dispersed Dispersed Dispersed Dispersed Dispersed
0
5 0 5 5 0 5 5 0 5
Signal-Idler Delay [ps] Signal-Idler Delay [ps] Signal-Idler Delay [ps]
Fig. 13.7 High-order dispersion cancellation results for a third-order, b fourth-order, and c fth-
order spectral phase. In all cases, dispersing both photons recovers a narrow correlation peak,
whereas classical correlations would only broaden further (Plots adapted from [26])
dispersed simultaneously, the original narrow correlation function is recovered,

verifying complete cancellation (both dispersed cases). These pulse-shaping
experiments therefore not only improve understanding of arbitrary biphoton phase,
but also make signicant strides in the quest for single-cycle biphotonsentangled
photons with correlation times equal to a single optical period [46]for likely only
through improved control of high-order dispersion cancellation will single-cycle
compression be possible.
As alluded to in Sect. 13.3, the electro-optic modulation setup of Fig. 13.4b
permits realization of the Fourier dual to biphoton dispersion cancellationnamely,
nonlocal cancellation of modulation [28]. Specically, considering equi-amplitude
sinusoidal phase modulation in (13.9), ms t expid sin xm t and
mi t expid sinxm t a, yields the complex joint spectrum
Z
hD dt expfiDt idsin xm t sinxm t ag: 13:17
Thus the modulators completely cancel each other when driven out of phase
(a p). As in the case of dispersion cancellation, this result can also be given a
more intuitive footing. If we consider the general phase-modulation functions
ms;i t exp ius;i t , analogously to the denition of group delay in (13.12), one
can view the modulation as imparting a time-dependent instantaneous frequency
shift to each of the photons, dened via the relation
@us;i
Dxs; i t : 13:18
@t
Because the arrival time of a given signal or idler photon is random, the actual
frequency shift is undened prior to measurement. However, due to the strong
temporal correlation (xing joint photon arrivals to time windows much smaller
than the modulator period), the frequency shift experienced by the signal photon
can be exactly cancelled by that imposed on the idler when
Dxs t Dxi t; 13:19
leaving a single spectral correlation peak at the original pump frequency. For the
specic case of sinusoidal modulation considered here, this condition is satised by
signal-idler modulator functions which are p radians out of phase, as found pre-
viously. Accordingly, one can view cancellation of electro-optic modulation as
shifting signal and idler frequencies by equal and opposite amounts.
We note that, classically speaking, such modulation cancellation can only be
achieved locally. In one recent example [48] (which actually appeared after the
quantum experiment described below), this more limited form of cancellation was
employed to achieve high-speed temporal cloaking, or hiding of events in time
through manipulation of a probe beam. With a combination of sinusoidal electro-optic
phase modulation and dispersion, a monochromatic probe was converted to a pulse
train with intensity gaps; the event, intensity modulation, operated only in these gaps
and therefore had no impact on the probing eld. Thus, subsequently propagating the
probe through matched dispersion and complementary modulation produced the
original continuous-wave eld at the output, thereby hiding the event from an
observer. Importantly, in this classical case cancellation of phase modulation was
obtainable only by sending the same photons through complementary modulators,
whereas the version utilizing entangled photons can achieve cancellation on sepa-
rated photons, an example of one of the uniquely nonclassical properties of such states
of light.
An experiment demonstrating this quantum behavior was carried in the Harris
group [49]; the key results are depicted in Fig. 13.8. Recalling that D x1 x2
2x0 represents the deviation of the sum frequency from the monochromatic pump, we
see that with no modulators running, a sharp correlation at D 0 is observed, with
two additional sidebands appearing when only one photon is phase-modulated.
Switching on the second modulator in phase with the rst widens the correlation
bandwidth even further, whereas running the second one 180 out of phase recovers
the original sharp spectral correlation, with no trace of modulation whatsoever. Taken
together, these dispersion and modulation investigations provide a solid foundation
for applying more sophisticated optical processing techniques to entangled photons,
an example of which is discussed in the next section.
13.6 Detailed Example II: Encoding and Decoding

of Biphoton Wavepackets
The previously described experiments with cancellation of polynomial spectral

phase or sinusoidal temporal modulation can be viewed as building blocks for a
wide range of complex coding schemes. For indeed, an arbitrary well-behaved
function can be expanded as a linear superposition of polynomials (Taylors the-
orem) or sinusoids (Fourier series), implying that many alternative phase
Fig. 13.8 Spectral correlations of temporally modulated biphotons, for a no modulation, b one
photon modulated, c both photons modulated in phase, and d both photons modulated out of
phase. Proper phasing allows the frequency spreading of one photon to be canceled by modulation
of the other (Results from [49])
combinations could also nd application in biphoton manipulation. In particular, we

focus here on binary spectral or temporal phase sequences, common in optical
code-division multiple-access communications (O-CDMA) [50]. In the classical
spectral case, a code of 0 and p phases is imposed on the frequency spectrum of an
ultrashort pulse, which spreads and lowers the waveform in time; only by applying
the conjugate code is the original pulse recovered, thereby allowing users with
different codes to communicate over the same time-frequency space. In the Fourier
dual, the phase code is temporally modulated onto a narrowband optical signal,
spreading the frequency content; as before, the original narrowband message is
recovered by demodulating with the correct temporal code. This latter process is an
example of spread-spectrum communication, which has proven itself adept at
covert transmission and jamming resistance [51]. In this section, we describe two
experiments which extend these classical communication techniques to entangled
photons, showing how such quantum states can be hidden and recovered in
either time or frequency.
First we consider frequency coding and time spreading, which is based on
applying appropriate codes to a programmable pulse shaper. One particularly rel-
evant approach, developed for classical elds, utilizes narrowband SFG to correlate
phase codes applied on two halves of the optical spectrum. In this version of O-
CDMA, the sender applies a phase code to one half of the optical spectrum, say all
frequencies greater than x0 ; then the receiver applies another code to the lower half
(x\x0 ) and sends the eld into a nonlinear crystal for SFG. If we denote the coded
spectrum by Ex and assume that phase-matching permits all frequency compo-

nents to combine, the resultant SFG eld at 2x0 is given by
Zx0
ESFG 2x0 dX Ex0 XEx0 X; 13:20
0
which for a sufciently flat input spectrum represents the inner product of the codes
applied to each half. Therefore, with codes chosen from an orthogonal family,
matched pairs give a high SFG yield, whereas mismatched codes ideally produce
nothing. Figure 13.9a provides a schematic of this coding process [52], with results
from an early experiment in Fig. 13.9b [53]. When the two codes are identical, a
strong upconverted signal is measured, with this dropping close to zero when the
codes are different.
In the above classical coding example, narrowband SFG is required to correlate
the two halves of the input spectrum; however, in the case of time-frequency
entangled photons, the desired spectral correlation is available automatically.
Examining the general biphoton shaping formula, (13.6), we observe that for /X
roughly flat over signal and idler passbands, and at s 0, the biphoton wavepacket
represents nothing more than the inner product of the spectral lters Hs x and
Hi x, mathematically equivalent to the classical example of (13.20). Therefore, if
these lters are codes chosen from some orthogonal family (such as Walsh-Had-
amard codes), matched codes will give a correlation peak, whereas mismatched
codes will generate a null. Thus while the classical version requires some form of
spectral post-selection to ensure that only frequency components equally offset
from x0 can recombine, no such post-selection is necessary in the quantum version,
for the biphotons possess the requisite correlations intrinsically. This connection
between classical narrowband SFG and time-frequency entangled photons has also
been utilized to derive interesting (but limited) classical analogues of Hong-Ou-
Mandel interference [54, 55] and dispersion cancellation [56]. And so in this light,
Fig. 13.9 Classical spectral coding. a Schematic of the coding process, in which phase codes are
applied to each half of the optical spectrum, followed by narrowband SFG. b SFG output
measured for matched codes (top) and mismatched ones (bottom) (Image in a from [52] and that
for b from [53])
our work on coding of entangled photons can be viewed as exploiting this con-
nection in the opposite direction; instead of looking for classical versions of
quantum phenomena, we have developed a quantum coding approach based on
ideas rst developed classically.
An outline of how this biphoton coding process works is provided in Fig. 13.10a.
The signal half-spectrum of a temporally narrow biphoton is encoded with a sequence
of 0 and p phases, which spreads the correlation function and produces a null at s 0.
Then the idler is decoded by applying a second Hadamard code to the lower half of the
biphoton spectrum. If the codes match in the sense that they are symmetric about x0 ,
the original correlation peak is restored, but if they differ, Hadamard orthogonality
ensures that the correlation function instead remains spread with a zero at s 0. As
an aside, we note that this symmetric cancellation condition in no way contradicts the
general antisymmetric phase relation expressed in (13.16) regarding dispersion
cancellation, but rather is a result of the binary codes; for truly antisymmetric phase,
the p phase chips on the signal would need to be accompanied by p chips on the
idler, but since p phase is physically indistinguishable from p, symmetric and
antisymmetric conditions are equivalent in this case.
With this coding approach, the biphoton can even be encoded and decoded
nonlocally, if the signal and idler are spatially separated, although in the proof-
(a) Encoding Signal

00
Idler Signal
0 0
Decoding Idler
00 00
Idler Signal
0
0 0 0 00
Idler Signal
0
(b)
10,000
Uncoded
Counts [s1]
Matched
Idler Code
5000
Mismatched
Encoded Idler Code
0
5 0 5 5 0 5
Signal-Idler Delay [ps] Signal-Idler Delay [ps]
Fig. 13.10 Orthogonal spectral coding of entangled photons. a Schematic of the basic process, in
which the signal photon is encoded and the idler decoded; only when the codes match is the
ultrashort biphoton correlation recovered. b Experimental results for length-40 Hadamard codes,
conrming the conceptual picture in a (Images adapted from [57])
of-principle demonstration discussed here, the photons are detected locally with
biphoton SFG. An experimental example with length-40 Hadamard codes is fur-
nished in Fig. 13.10b [57], using a setup identical to that in the previously discussed
high-order dispersion cancellation experiments [26]. When the signal is encoded
with a particular Hadamard sequence, the initially sharp correlation is temporally
spread. Applying the correct idler code then retrieves the original biphoton, aside
from a peak reduction due to scattering losses in the pulse shaper. On the other
hand, the wrong idler code yields a new, but still spread, biphoton state. These
results hint at the fascinating possibilities for complex spectral coding in quantum
cryptography, although there remain several key hurdles, addressed in Sect. 13.7.
An example of the Fourier dual, temporal phase coding, was demonstrated in
2010 [58]. This experiment employed extremely narrowband photons in order to
permit fractionally large spectral spreading with gigahertz modulators. Moreover,
although coding and decoding could indeed be implemented nonlocally with the
arrangement of Fig. 13.4b, the authors instead applied both modulators to the same
photon, with the entangled partner serving as a trigger for coincidence counting;
this can be viewed as a special case of Fig. 13.4c in which the net modulation ms
is the product of encoding and decoding modulators m1 s and m2 s. In this way
the signal photon is rst temporally encoded with a pseudorandom sequence of
p=2 phases, which frequency-spreads the 3.5-MHz photon to around 10 GHz.
Then a second modulator is applied, followed by a narrowband lter, before
detection. To guarantee a high probability of transmission, the second modulator
must undo the effects of the rst one and restore the narrow photon spectrum;
otherwise, the lter will most likely reject the frequency-spread photon. This pro-
vides an interesting way to hide single photons.
To demonstrate this capability, Belthangady et al. injected a laser eldat the
same frequency and with a flux over 1000 that of the original photonafter the
rst modulator but before the second. As shown in Fig. 13.11a, when both
Fig. 13.11 Hiding single photons. a With no temporal modulation, the single-photon eld is
completely hidden by a much more powerful laser beam. b But when a temporal encoding and
decoding process analogous to classical spread spectrum is applied, the laser is rejected and the
characteristic shape of the single photon is recovered (Results from [58])
modulators are turned off, the laser eld dominates the coincidence counts, with no
evidence of time correlation. But when the modulators are turned on, the shape of
the hidden single photon is observed (Fig. 13.11b). This ensues because the photon
is frequency-spread by the rst modulator and despread by the second, thereby
passing through the narrowband detection lter with ease; on the other hand, the
laser is transformed only by the second modulator, so its spectrum is spread right
before the lter and is therefore highly attenuated. As this experiment highlights,
even the thoroughly classical concept of spread spectrum can be applied to single-
photon signals.
13.7 Outlook
The examples of biphoton pulse shaping and modulation considered in this chapter
indicate the rapid development of this eld, which over the last decade has suc-
ceeded in combining these classical technologies with quantum states of light.
Nonetheless, there remain several important avenues of research to bring these
proof-of-principle experiments to more applied specializations, such as secure
quantum key distribution (QKD). In some ways SFG, which proved essential in the
above demonstrations of biphoton pulse shaping, now sits as a hindrance to further
development. For by requiring signal-idler recombination, any experiments utiliz-
ing biphoton SFG are intrinsically ill-suited for QKD implementations, in which the
two communicating parties must obviously be spatially separated for any useful
application. Moreover, the use of SFG in the previous demonstrations of ultrafast
biphoton dispersion cancellation [26, 45] means that a fully nonlocal implemen-
tationin which both photons are measured on separated detectorshas not been
realized with such high timing resolution. It is therefore desirable to nd alternative
high-time-resolution detection methods which could observe these effects nonlo-
cally. Pending further improvements in single-photon detector jitter to the pico-
second level [59], methods based on single-photon mixing with ultrashort [60] or
chirped [61] classical pulses seem particularly promising, and so nonlocal gener-
alizations of ultrafast dispersion cancellation and orthogonal coding may be pos-
sible with such techniques, although they have yet to be demonstrated.
Furthermore, high-dimensional time-frequency QKD protocols based on dispersion
[62] or temporal modulation [63] have recently been proposed, and it would be
interesting to investigate how sophisticated biphoton control could be exploited to
realize these schemes in practice.
In fact, it is their potential as information carriers that has motivated much of the
previous work on biphoton pulse shaping, and so here we briefly discuss some of
the basic ideas behind time-frequency entanglement for QKD. Such correlations
prove particularly promising for key distribution in ber networks, as they are well-
preserved through long propagation distances and can generate multiple bits of
information per biphoton. A rough schematic of how this process could work is
provided in Fig. 13.12. Signal and idler photons produced through SPDC are sent to
Time Signal Idler Time
Alice Frequency SPDC Frequency Bob
Fig. 13.12 Schematic of time-frequency QKD with entangled photons. Alice and Bob measure
either the arrival time or frequency of their respective photons, using results with the same bases to
construct a secure key
Alice and Bob, respectively, who measure randomly either the arrival time or
frequency of their received photon. Due to the strong spectro-temporal entangle-
ment, measurements in the same basis produce highly correlated results, so by
dividing frequency and time into discrete bins, it is possible to construct a secret
key through measurements on successive biphotons. And since results in two dif-
ferent bases are uncorrelatedas time and frequency are Fourier conjugatesthe
presence of an eavesdropper is revealed by publicly comparing some subset of the
measured bits and looking for an increased error rate.
The expanded information potential resulting from high-dimensional entangle-
ment can be quantied through the Schmidt decomposition [64, 65], in which the
biphoton spectrum is expressed as a sum of factorable two-photon states. Roughly
speaking, the number of modes with non-negligible coefcients can be estimated by
the Schmidt number K [65], a slighly more rigorous measure of entanglement than the
bandwidth ratio discussed briefly in Sect. 13.3. The larger K is, the greater the amount
of information obtainable from the biphoton. As a reference, we note that the ide-
alized biphoton state of (13.3) with a perfectly monochromatic pump has K 1; in
reality, the nite linewidth of our pump drops our realistic estimate of K to several
million or so. Yet actually extracting this information potential can prove extremely
difcult. For example, in our experiments [57], the maximum number of frequency
chips is limited by the 10-GHz spectral resolution of the pulse shapernot the *200-
kHz pump linewidth. Similar behavior has been shown for time binning [66] and
orbital angular momentum [67], in which there were observed maximum dimensions
beyond which the secure information capacity dropped. The ability to optimize both
the intrinsic and practical information potential of biphoton states will prove essential
in the experimental development of high-dimensional QKD.
Interestingly, electro-optic modulators have already been proposed for imple-
menting a time-frequency QKD system [63]. Conceptually, the protocol matches
Fig. 13.12; however, instead of directly measuring the time variable with a single-
photon detector, the authors propose to use a second spectrometer preceded by a
time-to-frequency converter. In this fashion, the arrival time of each photon is
converted to a particular frequency, so that a wavelength measurement is equivalent
to a temporal measurement. The time-to-frequency converter itself consists of a
dispersive element and an electro-optic phase modulator. An advantage of this
approach is the improved timing resolution over simple avalanche photodiodes; a
disadvantage is the increased dark count rate due to utilization of a detector array in
the spectrometer. As might be expected, the Fourier dual to this approach has the
reverse advantage/disadvantage relationship; as proposed in [62], the spectrometers
in the setup of Fig. 13.12 can be replaced by two frequency-to-time converters
(highly dispersive elements) with opposite signs of dispersion, followed by single-
photon detectors. Thus if both photons are measured in the dispersed-time basis, the
dispersion cancellation effect described in Sect. 13.5 ensures that their arrival times
are still synchronized. Intuitively, the connection with a frequency measurement
can be understood by the fact that dispersion cancellation is itself a consequence of
narrowband spectral entanglement. Although Fourier-transform pulse shaping was
not explicitly considered in this proposaland admittedly, pulse shapers cannot
apply dispersions as large as other optical devices such as ber Bragg gratingsthe
basic idea does lend itself to such pulse shaping. In fact, it would be interesting to
consider how the spectral coding demonstrated in [57] could be used for similar
time spreading and despreading in a form of code-based time-frequency QKD.
And so while experiments up to this point have already revealed new insights
into the behavior of entangled photons, numerous opportunities remain unrealized,
particularly in the context of quantum cryptography. Thus we expect the next
decade to witness even more advances in biphoton pulse shaping; we have only
begun to scratch the surface of the potential within such entangled quanta, and
technologies developed in classical optics will no doubt continue to nd unantic-
ipated uses in the quantum regime.
References
1. R. Hanbury Brown, R.Q. Twiss, Nature 177, 27 (1956)

3. M. Chekhova, Progress in Optics (Elsevier, New York, 2011), vol. 56, chap. 4
4. E. Brannen, H.I.S. Fergusen, Nature 178, 481 (1956)
6. E.M. Purcell, Nature 178, 1449 (1956)
7. P. Fellgett, Nature 179, 956 (1957)
8. R.J. Glauber, Phys. Rev. 130, 2529 (1963)
9. R.J. Glauber, Phys. Rev. 131, 2766 (1963)
10. Y. Shih, Rep. Prog. Phys. 66, 1009 (2003)
11. A. Einstein, B. Podolsky, N. Rosen, Phys. Rev. 47, 777 (1935)
12. A. Aspect, Nature 398, 189 (1999)
13. N. Gisin, G. Ribordy, W. Tittel, H. Zbinden, Rev. Mod. Phys. 74, 145 (2002)
14. A.M. Weiner, Ultrafast Optics (Wiley, Hoboken, NJ, 2009)
15. A.M. Weiner, J.P. Heritage, E.M. Kirschner, J. Opt. Soc. Am. B 5, 1563 (1988)
16. A.M. Weiner, Opt. Commun. 284, 3669 (2011)
17. A.M. Weiner, Rev. Sci. Instr. 71, 1929 (2000)
18. S.T. Cundiff, A.M. Weiner, Nat. Photon. 4, 760 (2010)
19. G.L. Li, P.K.L. Yu, J. Lightw. Technol. 21, 2010 (2003)
20. A.J. Metcalf, V. Torres-Company, D.E. Leaird, A.M. Weiner, IEEE J. Sel. Top. Quantum
Electron. 19, 3500306 (2013)
21. E.L. Wooten, K.M. Kissa, A. Yi-Yan, E.J. Murphy, D.A. Lafaw, P.F. Hallemeier, D. Maack,
D.V. Attanasio, D.J. Fritz, G.J. McBrien, D.E. Bossi, IEEE J. Sel. Top. Quantum Electron. 6,
69 (2000)
22. B.H. Kolner, IEEE J. Quantum Electron. 30, 1951 (1994)
23. Z. Jiang, C.B. Huang, D.E. Leaird, A.M. Weiner, Nat. Photon. 1, 463 (2007)
24. V.R. Supradeepa, C.M. Long, R. Wu, F. Ferdous, E. Hamidi, D.E. Leaird, A.M. Weiner, Nat.
Photon. 6, 186 (2012)
25. L. Mandel, E. Wolf, Optical Coherence and Quantum Optics (Cambridge University Press,
Cambridge, UK, 1995)
26. J.M. Lukens, A. Dezfooliyan, C. Langrock, M.M. Fejer, D.E. Leaird, A.M. Weiner, Phys.
Rev. Lett. 111, 193603 (2013)
27. J.D. Franson, Phys. Rev. A 45, 3126 (1992)
28. S.E. Harris, Phys. Rev. A 78, 021807 (2008)
29. P. Kolchin, C. Belthangady, S. Du, G.Y. Yin, S.E. Harris, Phys. Rev. Lett. 101, 103601
(2008)
30. B. Dayan, Y. Bromberg, I. Afek, Y. Silberberg, Phys. Rev. A 75, 043804 (2007)
31. E. Poem, Y. Gilead, Y. Lahini, Y. Silberberg, Phys. Rev. A 86, 023836 (2012)
32. J.M. Lukens, A. Dezfooliyan, C. Langrock, M.M. Fejer, D.E. Leaird, A.M. Weiner, Opt. Lett.
38, 4652 (2013)
33. B. Dayan, A. Peer, A.A. Friesem, Y. Silberberg, Phys. Rev. Lett. 94, 043602 (2005)
34. B. Dayan, Phys. Rev. A 76, 043813 (2007)
35. A. Peer, B. Dayan, A.A. Friesem, Y. Silberberg, Phys. Rev. Lett. 94, 073601 (2005)
36. F. Zh, M. Halder, T. Feurer, Opt. Express 16, 16452 (2008)
37. J.D. Franson, Phys. Rev. Lett. 62, 2205 (1989)
38. L. Mandel, Rev. Mod. Phys. 71, S274 (1999)
39. C. Bernhard, B. Bessire, T. Feurer, A. Stefanov, Phys. Rev. A 88, 032322 (2013)
40. M.A.F. Roelens, S. Frisken, J.A. Bolger, D. Abakoumov, G. Baxter, S. Poole, B.J. Eggleton,
J. Lightw. Technol. 26, 73 (2008)
41. J.M. Lukens, O. Odele, C. Langrock, M.M. Fejer, D.E. Leaird, A.M. Weiner, Opt. Express 22,
9585 (2014)
42. V. Torres-Company, J. Lancis, H. Lajunen, A.T. Friberg, Phys. Rev. A 84, 033830 (2011)
43. C. Belthangady, S. Du, C.S. Chuu, G.Y. Yin, S.E. Harris, Phys. Rev. A 80, 031803 (2009)
44. S.Y. Baek, Y.W. Cho, Y.H. Kim, Opt. Express 17, 19241 (2009)
45. K.A. ODonnell, Phys. Rev. Lett. 106, 063601 (2011)
46. S.E. Harris, Phys. Rev. Lett. 98, 063602 (2007)
47. Z. Jiang, S.D. Yang, D.E. Leaird, A.M. Weiner, Opt. Lett. 30, 1449 (2005)
48. J.M. Lukens, D.E. Leaird, A.M. Weiner, Nature 498, 205 (2013)
49. S. Sensarn, G.Y. Yin, S.E. Harris, Phys. Rev. Lett. 103, 163601 (2009)
50. J.P. Heritage, A.M. Weiner, IEEE J. Sel. Top. Quantum Electron. 13, 1351 (2007)
51. R. Pickholtz, D. Schilling, L. Milstein, IEEE Trans. Commun. 30, 855 (1982)
52. Z. Zheng, A.M. Weiner, Opt. Lett. 25, 984 (2000)
53. Z. Zheng, A.M. Weiner, K.R. Parameswaran, M.H. Chou, M.M. Fejer, IEEE Photon. Technol.
Lett. 13, 376 (2001)
54. R. Kaltenbaek, J. Lavoie, D.N. Biggerstaff, K.J. Resch, Nat. Phys. 4, 864 (2008)
55. R. Kaltenbaek, J. Lavoie, K.J. Resch, Phys. Rev. Lett. 102, 243601 (2009)
56. R. Prevedel, K.M. Schreiter, J. Lavoie, K.J. Resch, Phys. Rev. A 84, 051803 (2011)
57. J.M. Lukens, A. Dezfooliyan, C. Langrock, M.M. Fejer, D.E. Leaird, A.M. Weiner, Phys.
Rev. Lett. 112, 133602 (2014)
58. C. Belthangady, C.S. Chuu, I.A. Yu, G.Y. Yin, J.M. Kahn, S.E. Harris, Phys. Rev. Lett. 104,
223601 (2010)
59. R.H. Hadeld, Nat. Photon. 3, 696 (2009)
60. O. Kuzucu, F.N.C. Wong, S. Kurimura, S. Tovstonog, Phys. Rev. Lett. 101, 153602 (2008)
61. J.M. Donohue, M. Agnew, J. Lavoie, K.J. Resch, Phys. Rev. Lett. 111, 153602 (2013)
62. J. Mower, Z. Zhang, P. Desjardins, C. Lee, J.H. Shapiro, D. Englund, Phys. Rev. A 87,
062322 (2013)
63. J. Nunn, L.J. Wright, C. Sller, L. Zhang, I.A. Walmsley, B.J. Smith, Opt. Express 21, 15959
(2013)
64. C.K. Law, I.A. Walmsley, J.H. Eberly, Phys. Rev. Lett. 84, 5304 (2000)
65. C.K. Law, J.H. Eberly, Phys. Rev. Lett. 92, 127903 (2004)
66. I. Ali-Khan, C.J. Broadbent, J.C. Howell, Phys. Rev. Lett. 98, 060503 (2007)
67. J. Leach, E. Bolduc, D.J. Gauthier, R.W. Boyd, Phys. Rev. A 85, 060304 (2012)
Chapter 14
Harnessing Nonlinear Optics
for Microwave Signal Processing
David Marpaung, Ravi Pant and Benjamin J. Eggleton
Abstract Harnessing nonlinear optical effects in a photonic chip scale has been
proven useful for a number of key applications in optical communications.
Microwave photonics (MWP) can also benet from the adoption of such a tech-
nology, creating a new concept of nonlinear integrated microwave photonics. Here,
we look at the potential of using nonlinear optical effects in a chip scale to enable
RF signal processing with enhanced performance. We review a number of recent
results in this eld, with particular focus on the creation of frequency agile and high
suppression microwave bandstop lters using on-chip stimulated Brillouin scat-
tering (SBS). We also discuss the future prospect of nonlinear integrated MWP to
enable a general purpose, programmable analog signal processor, as well as com-
pact, high performance active microwave lters with enhanced energy efciency.
14.1 Microwave Photonics
Microwave photonics (MWP) is a eld that deals with the use of optoelectronic
devices and systems for enhanced radio frequency (RF) signal manipulation,
including the generation, distribution, processing, and analysis of high-speed
microwave signals [11, 44, 45, 53]. MWP technologies have also enabled various
functionalities which are not feasible to achieve only by traditional microwave
D. Marpaung R. Pant B.J. Eggleton (&)

Centre for Ultrahigh Bandwidth Devices for Optical Systems (CUDOS),
School of Physics, Institute of Photonics and Optical Science (IPOS),
e-mail: egg@physics.usyd.edu.au
D. Marpaung
e-mail: david.marpaung@sydney.edu.au
R. Pant
e-mail: rpant@physics.usyd.edu.au

450 D. Marpaung et al.
(a)
RF in Modulation Photodetector RF out
Device
Optical Fiber
(b) Functionalities
- time delay
- phase shift
Modulation - filtering
RF in Photodetector RF out
Device - frequency
conversion
- etc.
Fig. 14.1 Schematics of a an MWP link and b a simple MWP system. The MWP link basically
consists of a modulation device for E/O conversion and a photodetector for O/E conversion. Such
an MWP link with added functionalities between the conversion will make an MWP system. From
[28]
techniques, such as the generation of ultra-wide bandwidth waveforms, or ltering

with recongurable shape and central frequency.
The basis of any MWP system is a so-called MWP link [1618]. This link
consists of a modulation device to convert RF signals to modulated optical signals
(E/O conversion) connected by an optical ber to a photodetector that restores the
RF signal back from the optical domain (O/E conversion), as shown in Fig. 14.1a
The majority of MWP links employ intensity modulation-direct detection (IMDD)
scheme, with a high speed electro-optic modulator (EOM) as the modulation device
and a p-i-n photodiode as the detector.
Initially, the main use of MWP links was to replace long-length coaxial-cable
links which were very lossy for high RF frequencies. Optical bers offer low
attenuation for virtually any microwave frequency, as well as massive weight
reduction, increased flexibility, and immunity to electromagnetic interference
(EMI). Subsequently, more efforts were directed to use these links to process the
modulated RF signals beyond only signal transport. This was the birth of MWP
systems, which established by means of adding functionalities between the two
conversions, i.e. signal processing in the optical domain (Fig. 14.1b).
MWP has a wide range of applications in the processing of high speed RF
signals. The generation of high purity microwave signals and high complexity ultra-
broadband waveforms are the latest development in MWP. The main attraction for
signal generation using MWP technique is the large frequency tunability and the
potential of reaching very high frequencies (up to the THz region) using relatively
simpler technique compared to the traditional microwave/electronic approach.
Moreover, the distribution of such high frequency signals using extremely low-loss
optical bers is also attractive, which would have been very lossy using coaxial
14 Harnessing Nonlinear Optics for Microwave Signal Processing 451
cables [43]. For waveform generations MWP techniques offer broad bandwidth
and the full recongurabilities of the phase and amplitude of the RF waveforms
[21, 54].
For microwave signal processing, MWP techniques have enabled ltering
[12, 31], tunable true time delay [41, 55, 61] and wideband phase shifting of
microwave signals [8, 15, 23]. The MWP concept added value are the operation
bandwidth and the potential of fast and agile recongurabilities of these function-
alities. Combining these basic functionalities lead to the realization of MWP pro-
cessors for optical beamforming and phase array antenna systems [30, 34, 60].
14.1.1 MWP System Performance
To achieve optimum functionalities from MWP systems, the underlying MWP link
needs to reach sufcient performance, which is limited by the losses, noise, and
nonlinear distortions introduced by the E/O and O/E conversions [17]. Their per-
formance can be described by a set of parameters, namely the link gain, the noise
gure, the intercept points, and the spurious-free dynamic range (SFDR) [16, 18].
The link gain describes the RF to RF power signal transfer in the MWP link or
system. Due to the limited conversion efciencies in the modulation device and
photodetector, it is common that the MWP link shows negative link gain in the
decibel scale, i.e. a net loss. For externally modulated MWP link, the link gain is
proportional to the square of the modulated optical power. Thus, a common tech-
nique to improve the link gain is to pump more optical power to the EOM. The
technique has been effectively used to demonstrate MWP links with net gain (i.e.
positive link gain) instead of loss, where the value as high as +44 dB has been
demonstrated [49].
The noise gure (NF) is a useful metric that measures the signal-to-noise ratio
(SNR) degradation in the system, expressed in decibels. The noise gure is pro-
portional to the total noise power in the link, and is inversely proportional to the
link gain. The total link noise is usually dominated by laser relative intensity noise
(RIN) and the photodetector shot noise. State-of-the-art MWP links can exhibit NF
below 10 dB [1, 29], which can be achieved using an ultra-low half-wave voltage
EOM in the order of 1 V, and a very high input power.
The E/O and O/E conversions in the MWP link also add nonlinear distortions to
the output RF signal. Due to the nonlinear response of the link (i.e. components like
the EOM or the photodetector), input RF tones will generate new frequency
components called the intermodulation distortions (IMDs). The second-order
intermodulation (IMD2) is generated due to the quadratic nonlinearity in the link
and the frequency components appear at the sum and the difference of the modu-
lating frequencies, while the third-order intermodulation (IMD3) is generated by
cubic nonlinearity in the link and appear at the sum and the difference of twice of
one frequency with the other frequency.
The gure of merit that incorporates the effect of noise and nonlinearity in the
MWP link is the spurious-free dynamic range (SFDR). The SFDR is dened as the
ratio of input powers where, on the one hand, the fundamental signal power is equal
to the noise power and, on the other hand, the nth-order intermodulation distortion
(IMDn) power is equal to the noise power. In terms of output powers, this can be
interpreted as the maximum output SNR that can be achieved while keeping the
IMDn power below the noise floor. The SFDR generally governs the lowest and the
strongest signal power that can be accommodated in the link.
14.1.2 Integrated Microwave Photonics
For the last 25 years, MWP systems and links have relied almost exclusively on
discrete optoelectronic devices and standard optical bers. These congurations are
bulky, expensive and power-consuming while lacking in flexibility. A second
generation of MWP systems, known as integrated microwave photonics (IMWP)
[28], aims at the incorporation of MWP components/subsystems in photonic inte-
grated circuits. With photonic integration, one can achieve a reduction in footprint,
inter-element coupling losses, packaging cost as well as power dissipation.
IMWP clearly promises a number of advantages compared to conventional ber-
based approach. The potential of integrating a number of different signal processing
functionalities in a single chip will lead to signicant reduction in size, weight,
power consumption, and cost. Moreover, the ability to integrate various functions
like modulation, passive signal processing, and detection in a single chip will lead
to reduction in insertion losses, which is the key for MWP signal processor to
compete with RF signal processor in terms of performance.
14.1.3 Nonlinear Integrated Microwave Photonics
Thus far, IMWP has largely been limited to linear optical processing. It is inter-
esting, however, to examine the potential of nonlinear optics in integrated platform
for microwave signal processing. Optical nonlinearities (e.g., v2 and v3 [2], such
as sum and difference-frequency generation (SFG/DFG), cross-phase modulation
(XPM), degenerate/non-degenerate four-wave mixing (FWM), and stimulated
Brillouin scattering (SBS), are suitable candidates for MWP signal processing. In
the context of optical signal processing, these nonlinear optics effects have ef-
ciently been harnessed, both in bers and in integrated devices, for a number of
functionalities such as all-optical delay [35, 47], all-optical switching and multi-
plexing [20], and wavelength multicasting [52].
Here, we look at the potential of harnessing nonlinear optical effects in a chip
scale to enable RF signal processing with enhanced performance. As an example,
we focus on stimulated Brillouin scattering (SBS) for MWP signal processing,

namely for frequency agile and high suppression microwave bandstop lters,
continuously tunable phase shifter, and tunable slow light delay lines.
14.2 Stimulated Brillouin Scattering
Stimulated Brillouin scattering (SBS) is a third-order optical nonlinearity, which

results from three way interaction between light and acoustic vibrations in the
medium. The process is characterized by the inelastic scattering of a pump beam
from an acoustic vibration in the medium to generate a Stokes wave [2]. For the
pump-probe conguration, the SBS process leads to a gain resonance centered at
the Stokes frequency
where the width of the resonance is determined by the phonon
lifetime sp . Typical phonon lifetime in commonly used optical materials is 10 ns,

which results in narrow resonance bandwidth 1=sp of tens-of-MHz. The band-
width of the SBS resonance, however, can be controlled by tailoring the pump
prole, enabling broadband (GHz) gain response. The Brillouin scattering process
can therefore provide wide bandwidth tunability, ranging from MHz to GHz.
Further, the centre frequency of the gain resonance can be tuned by tuning the
pump frequency xp , which enables SBS gain realization at any wavelength.
The flexibility of tailoring the SBS process parameters has enabled a number of
applications in MWP [39], including tunable delay lines [14, 35, 38, 47], phase
shifters [15, 23], and ltering [9, 26, 32, 51, 56, 57]. In the SBS process, a signal
centered at the Stokes frequency experiences gain whereas absorption resonance
occurs for a signal centered at the anti-Stokes frequency. Both the gain and
absorption resonance exhibit a phase response, obtained using the Kramers-Kronig
relation, which can be used to realize slow light induced tunable delay and also to
control the phase of a signal. The stimulated Brillouin scattering process is therefore
emerging as a potential platform for microwave photonics, having been utilized in
long length (km) optical bers for a number microwave applications. Integrated
MWP signal processing however, requires on-chip realization of SBS [40].
14.2.1 On-Chip SBS
Recently, on-chip SBS was demonstrated using a travelling wave geometry in a

photonic-chip fabricated using chalcogenide (As2S3) glass. Chalcogenide glasses
are known to have large refractive index, which results in large nonlinearity [20].
The strong connement of the optical and acoustic modes (see Fig. 14.2) in a
chalcogenide photonic-chip provides a small effective area and a large overlap
between acoustic and optical modes, which, when combined with large refractive
index of chalcogenide glass, resulted in large Brillouin gain coefcient and thus
(a)
Photonic chip Optical mode Acoustic mode
(b)
Fig. 14.2 a Schematic of the chalcogenide photonic chip showing strongly conned optical and
acoustic modes, critical for realizing on-chip stimulated Brillouin scattering. b Measured gain
prole for the Stokes signal showing a Lorentzian response with narrow linewidth of 34 MHz.
From [40]
large Brillouin gain in a 7 cm long rib waveguide using moderate pump power
(100 mW).
The measured frequency shift for the chip was mB 7:7 GHz, and the SBS gain
prole, measured in a pump-probe experiment, is plotted in Fig. 14.4 as a function
of frequency detuning from the Stokes frequency. It has a Lorentzian line shape,
with a full-width at half-maximum (FWHM) of 34 MHz. The narrow bandwidth of
the SBS resonance and the large dispersion associated with it is the key for enabling
a number of applications in microwave photonics.
14.3 Recongurable Microwave Filters Using SBS
RF lters are fundamental signal processing components that play a critical role in
determining the systems ultimate performance. The ultimate goal of a lter is to
improve the signal-to-noise ratio of the signal of interest either by rejecting
unwanted signals (i.e. band-stop lter) or by allowing only the desired signal to
propagate to the receiver (i.e. band-pass lter). To operate in the increasingly
crowded radio spectrum, RF receivers require frequency agility to operate in dif-
ferent frequency bands. A smart, recongurable lter that operates over a wide
frequency range is a critical technology to achieve this agility.
Fig. 14.3 a Measured response of a photonic-chip based MWP bandpass lter showing wide lter
tunability of 212 GHz. From [9]. b Measured response of a photonic-chip based MWP notch
lter based on single sideband modulation and SBS absorption. From [32]
The gain (absorption) responses of the SBS process can be used as a band-pass
(band-reject) optical lter, where the lter extinction depends on the G parameter.
Microwave photonic lters with shape reconguration and widely tunable pass-
band and notch have recently been demonstrated exploiting SBS in several km long
optical bers [51, 56, 57]. While high-Q MWP lters (1000) with wide tunability
(120 GHz) and large extinction 40 dB were reported, the use of long lengths in
these demonstrations prohibits photonic integration.
Recently on-chip band-pass and band-reject MWP lters were demonstrated
exploiting SBS in a chalcogenide photonic-chip discussed in Sect. 14.2.1. For the
bandpass response, on-chip SBS gain was used to break the amplitude symmetry of
Fig. 14.4 Principle of a MWP notch lter using a single side band modulation and SBS
absorption b double side band modulation and SBS absorption and c double side band modulation
and SBS gain. From [26]
the upper and lower side bands, obtained by phase modulating an optical carrier
with a radio-frequency (RF) signal (see Fig. 14.7). In the absence of SBS gain,
equal amplitude and p phase shift between the two sidebands causes destructive
interference at the microwave frequency on opto-electronic conversion, resulting in
no RF signal. By applying SBS gain, the part of the microwave spectrum that sees
the gain is ltered with the ltered spectrum having the shape and bandwidth of the
SBS gain prole. In [9], the MWP lter frequency was tuned from 2 to 12 GHz by
tuning the pump frequency (Fig. 14.3a) while maintaining stable amplitude (20 2
dB) and 3 dB bandwidth (f3dB 23 2 MHz).
Additionally, in the demonstration of Byrnes et al. [9], the 3 dB bandwidth and
lter shape factor S, dened as the ratio of the 20 to 3-dB bandwidth, were rec-
ongured using pump prole tailoring. The 3-dB bandwidth and shape factor S
were improved from f3dB 20 MHz and S 3:5 for the single-pump case to
f3dB 40 MHz and S 2 for the dual pump conguration. For a photonic-chip
based MWP lter, an extremely large Q factor of 520 was obtained using the lter
response at 12 GHz.
As in the case of bandpass lter, SBS absorption can be used as a band-stop lter
to realize a notch response in microwave domain. Single-side band (SSB)
Fig. 14.5 a Experimental results of the MWP notch lter bandwidth tuning achieved by means of
tuning the SBS pump power. b Experimental results of ultra-wide frequency tuning of the MWP
notch lter, preserving a narrow isolation bandwidth of 13 2 MHz and ultrahigh rejection of
60 2 dB. From [26]
modulation was used to encode the microwave signal to the optical domain, where a
SBS pump was used to create a band-stop response at the desired frequency. The
frequency of the notch was tuned by tuning the pump frequency. For this cong-
uration, the depth of the notch depends on the pump power via the Brillouin
absorption parameter. A photonic-chip based MWP notch lter with a tuning range
of 28 GHz (see Fig. 14.3b) was realized using this scheme [32]. The measured
notch suppression of 19 2 and 3 dB bandwidth (126 7 MHz) in this demon-
stration showed good stability in the lter operation. Although MWP notch lter
based on SSB modulation and SBS operation in absorption regime provides an easy
to implement scheme, it is highly inefcient and it requires high loss for large
suppression, resulting in large power consumption. Achieving large suppression
(>60 dB) using SSB modulation scheme with on-chip SBS is therefore highly
inefcient.
Recently, a remarkable progress has been made regarding MWP tunable notch
lters based on SBS. A long standing problem in RF electronic lters is to achieve
wide frequency tuning simultaneously with high resolution ltering. For example,
state-of-the-art absorptive band-stop RF lters are capable of a high peak attenu-
ation (>50 dB) and high resolution (<10 MHZ 3 dB isolation bandwidth measured
from the passband) but have limited notch frequency tuning range in the order of
1.4 GHz [25]. MWP notch lters, on the other hand, are capable of multi-gigahertz
tuning, but are often limited in peak attenuation (<40 dB) and resolution (gigahertz
bandwidth instead of tens of megahertz). This limitation stems from the traditional
way of implementing notch lters using SSB modulation and an optical lter. In
this scheme, the resolution and peak suppression of the MWP lter are limited by
those of the optical lter. Unfortunately, for some optical lters, there is a trade-off
between resolution and peak suppression. Hence, the resolution and peak sup-
pression of these MWP lters cannot be optimized simultaneously.
In [26], a solution to this limitation was proposed in the form of a new class of
MWP notch lter that operates by a novel concept of optical sidebands coherent
control using an electro-optic (EO) modulator and an optical resonant lter. This
scheme allows the creation of an anomalously high rejection MWP notch lter from
virtually any kind of optical resonance, irrespective of its type (gain or absorption),
or its magnitude. Instead of SSB modulation, this scheme relies on double-side
band (DSB) modulation to control the relative amplitude and phase response of
both modulation side bands, as shown in Fig. 14.4. This amplitude and phase
control was achieved using a dual-parallel Mach-Zehnder modulator (DPMZM)
driven using a quadrature hybrid coupler. Voltage biases to the DPMZM modulator
were used to tune the phase and amplitude of the two sidebands. The amplitude and
phase response of an optical resonance, which can be passive e.g. ring resonator or
active e.g. SBS, was then used to make the amplitude of two side bands equal and
out-of-phase by p. The destructive interference between the signals obtained by
beating of two sidebands with the carrier results in anomalously large suppression.
Microwave photonic notch lters with large suppression (>60 dB) were dem-
onstrated by implementing this scheme using SBS resonance in optical bers [26].
The lter provides large suppression (>60 dB) with wide frequency tuning
(30 GHz), and wide bandwidth tuning from 1065 MHz, achieved using optical
pump power tuning for both SBS gain and loss (see Fig. 14.5). For this scheme,
large suppression is achieved irrespective of the SBS gain or loss. The bandwidth of
the notch lter depends on the bandwidth of the SBS gain (loss). The SBS gain
(loss) bandwidth reduces (increases) with increase in the pump power. Bandwidth
tuning over a wide range can therefore be realized by varying the pump power.
Although excellent performance was achieved by exploiting SBS gain and loss, the
ber-based demonstration is incompatible with photonic integration.
A widely tunable (20 GHz) microwave photonic notch lter with large sup-
pression (>55 dB) using on-chip SBS in combination with DSB modulation was
recently reported [24]. The 3 dB bandwidth of the notch was tuned from 31 to
77 MHz by varying the SBS gain parameter from 12.5 to 7.6 dB. In contrast to the
notch lter based on SSB modulation + band-stop response of the SBS process,
where a suppression of 55 dB would require a Brillouin gain of 55 dB (with a pump
power of 600 mW), this scheme realizes a notch depth >55 dB using only 0.8 dB of
gain (pump power of tens of mW), which results in highly efcient operation.
14.4 SBS Tunable Delay Line and Phase Shifter
Recongurable optical delay lines and wideband tunable phase shifters have pri-
mary importance in a number of MWP signal processing applications like optical
beamforming [30, 61], and MWP ltering. The simplest way for generating delay in
the optical domain is through physical length of optical bers. However, this can
become rather bulky. For this reason, integrated photonic solutions are used.
14.4.1 On-Chip SBS Tunable Delay Lines
SBS can be harnessed as tunable delay lines with flexible operation bandwidth and
wavelength. For the SBS process, the delay is given by:
gB Pp L
Td ; 14:1
Aeff CB
where Pp is the pump power, Aeff is the effective mode area, CB is the full-width at
half-maximum (FWHM) linewidth of the Brillouin gain, L is the medium length,
and gB is the Brillouin gain coefcient. This suggests that the time delay can be
tuned by varying the pump power. Further, for a device that provides large gB and
small Aeff , large delay can be obtained using small power.
Many of the earlier SBS delay line demonstrations [14, 15] used silica based
optical bers where small gB and large mode area Aeff , necessitated the use of long
length (km) to achieve large delay using small pump power (100 mW).
Recently, SBS based slow- and fast-light was demonstrated in a 7 cm long
chalcogenide photonic chip with a delay of 23 ns which was obtained using a gain
of 23 dB. This was obtained from a pulse delay measurement [38] for an input of
100 ns long nearly Gaussian pulse. For a 25 ns pulse, a pulse delay of nearly one
pulse width was obtained with a gain of 23 dB. Figure 14.6 show a comparison of
the measured delay with the theoretical delay calculated using (14.1), showing good
agreement between theory and experiments.
On-chip SBS enabled optically controlled continuously tunable delay is a step
towards realizing tap-delay based microwave photonic lters. The magnitude of the
delay can be varied from ps to ns range by tailoring the pump prole to vary the
SBS gain bandwidth [15].
14.4.2 On-Chip SBS Phase Shifter
Radio frequency (RF) phase shifters are a fundamental building block for analog
signal processors [8], and are widely employed in phased array antenna systems.
The characteristics of an ideal phase shifter include a continuously tunable, full-
360 phase shift, constant over a wide, multi-GHz frequency band. In addition, it is
desirable to have low insertion loss over the whole band of operation, with a flat
amplitude response across the entire phase tuning range. These specications
however cannot be met by conventional electrical phase shifters, which can operate
over only a narrow frequency range, and cannot achieve continuous phase tuning.
Fig. 14.6 Comparison of the

measured and theoretical
tunable time delay from SBS
in a 7 cm chalcogenide
waveguide. From [38]
The most promising approaches for ultra-wideband phase shifter operation rely on
optical carrier processing [8, 15], in which a phase shift imparted on the optical carrier
is converted by square-law photodetection to a phase shift in the electrical domain,
with no theoretical electrical bandwidth limitations. In optical carrier processing, it is
highly benecial to use a narrowband optical effect to shift the phase of the optical
carrier, since this does not affect the optical sidebands and as a consequence, does not
restrict the phase shifter to high-frequency operation. Due to its low threshold power
and narrow bandwidth, SBS is ideal for optical carrier processing.
In 2006, Loayssa and Lahoz [23] proposed a SBS phase shifter topology with
two SBS pumps at frequencies up and down-shifted from the optical carrier, as
shown in Fig. 14.7. The upper frequency pump will induce a gain response at the
carrier frequency, which will be exactly cancelled by the loss response induced by
the lower frequency pump. However, due to the prole of the SBS phase response,
the phase shifts introduced by the two pumps will not cancel, but will in fact add
up. In this way, the carrier amplitude will not be affected by the SBS process, while
the carrier phase will be shifted by twice the SBS phase response. Tuning of the
phase shift experienced by the carrier is achieved by changing the frequency of the
SBS pumps relative to the carrier.
Recently, SBS MWP phase shifter using the dual-pump technique was demon-
strated in a chalcogenide photonic chip [36]. The phase shifter exhibited a maximum
of 2.5 dB amplitude fluctuations as the phase was continuously tuned over the range
of 210 (Fig. 14.8). This results were achieved using 25 dBm of pump power.
Extension to a full 360 range can be obtained by increasing the pump power by
3.6 dB, or by reducing the insertion loss of the photonic hip by the same amount.
14.5 FWM and XPM for MWP Signal Processing
On-chip ultrafast Kerr nonlinearity has also been exploited in the context MWP
signal processing, for example for ultra-wideband (UWB) signal generation [48]
and recongurable multi-tap ltering [13]. In [48], Tan et al. combine the effects of
cross-phase modulation (XPM) and birefringence in an As2S3 rib waveguide to
Fig. 14.7 SBS tunable MWP phase shifter using dual-pump SBS. The SBS magnitude responses
(dashed lines) cancel out at the carrier frequency, while the phase responses add up. From [37]
Fig. 14.8 Experimentally measured frequency response of the SBS phase shifter. a Magnitude
response with 2.5 dB amplitude fluctuations. b Phase response with 210 continuous tuning
range. From [37]
generate polarity-inverted UWB monocycles with a single optical carrier. The high
Kerr-nonlinearity of ChG in a chip platform enables efcient XPM in a short length
of 7 cm. The combination of XPM and birefringence essentially acts as an all-
optical differentiator that converts a train of Gaussian pulses into a train of
monocycle electrical pulses after photo-detection process. Using this technique, Tan
et al. demonstrated the generation of a wide variety of UWB pulses, including
polarity-inverted monocycles and doublets. A key advantage in exploiting XPM
was the chirp erasure of the input optical pulses, because XPM depends solely on
the pulses intensity. Thus, shaped UWB optical pulses have a good tolerance to
dispersion over ber and are more suitable for long-distance transmission in UWB
over ber communication systems.
Recently, Chen et al. [13] have reported a step towards the integration of a
recongurable MWP lter concept in [30], by exploiting an FWM process in a
silicon nanowire, instead of using HLNF. The silicon waveguide (h 220 nm,
w 650 nm, length = 12 mm) was dispersion engineered to yield broadband
FWM. They generated 2 idlers with measured conversion efciency of 25 dB.

These wavelengths taps were then modulated and amplitude controlled using a
waveshaper. They used 4 km of single mode ber (SMF) as the dispersive medium
required to generate time delay between the taps. They measured the response of a
4-tap MWP lter and demonstrated that the 12 mm silicon nanowire can be used to
replace 1 km of silica highly nonlinear ber (HNLF) without compromising the
quality of the MPF response.
14.6 Future Directions
In this section we discuss the potential of nonlinear IMWP and identify two key
concepts that we believe will shape the eld in the near future.
14.6.1 General Purpose Analog Processor
The exponential growth in bandwidth demand of radio communications has put

unprecedented challenges in the RF signal processing chain. Traditionally, strong
emphasis in such a chain is put on digital signal processing (DSP). However, for
increasing operating frequencies and larger bandwidths, a reliable analog signal
processor (ASP) that can match the flexibility of a DSP is desired [10, 25]. The aim
here is to create a robust analog signal processing engine, which can be flexibly
tailored in terms of group delay and amplitude responses, that works in millimeter
wave or terahertz frequency range where DSP based systems are inefcient or
inapplicable.
Due to its high bandwidth and potential recongurability, IMWP processor is an
attractive candidate for such a flexible ASP. However, current state-of the-art IMWP
processors were developed only for very specic tasks, for example beamforming
and ltering, frequency discrimination, or pulse shaping. Thus, there is a lack of
generality and multi-functionality of these processors. To emulate DSP, the IMWP
processor has to be application agnostic, multi-purpose and programmable.
The incorporation of optical nonlinearities in an IMWP processor will open up
the path towards a general purpose ASP. A large number of MWP signal processing
functions can be generalized as processing of a number copies of RF modulated
optical signals using recongurable optical lters. Nonlinear optical processes that
create new optical wavelengths such as FWM, SFG, or DFG are attractive candi-
dates for generating such signal copies. In fact, this concept of wavelength multi-
casting has recently been implemented for MWP signal processing techniques such
as channelization [5, 6] and recongurable ltering [50]. In the view of the general
processor, these wavelength copies will subsequently be processed by a recong-
urable optical lter. The typical signal processing tasks carried out by these lters
include tunable time delay, carrier phase tuning, and complex (amplitude and
Fig. 14.9 a Illustration of the envisioned general purpose analog signal processor architecture
combining ultrafast nonlinear optics for signal copies generation (for example wavelength
multicasting via on chip FWM) and recongurable linear optical ltering (for example Vernier
ring resonators tunable using thermo-optical tuning). b Possible implementations of high
performance RF notch lter using sideband cancellation and SBS hybrid/monolithically integrated
in a CMOS compatible photonic chip that consists of electro-optic modulator, tunable optical
coupler for pump tuning, a nanophotonic waveguide with high SBS gain, and a photodetector.
From [25]
phase) ltering. These processed optical signals will then be recombined or directed
to separate dedicated outputs for a photodetection process.
The key challenge here is to create a monolithic photonic chip that hosts the
nonlinear optics for wavelength multicasting and the recongurable linear optical
ltering. Figure 14.9a depicts an illustration of a possible design of such a photonic
chip in Si3N4 technology [42]. The chip consists of three main sections: two long
spiral waveguides for wavelength multicasting and multiplexing based on cascaded
FWM processes, and a recongurable optical lter based on a network of ring
resonators that can be tuned using thermo-optics effect. This ring network might
host a mixture of all pass ring lters, add-drop rings, as well as Vernier (cascade of
non-identical) rings which have been recently exploited for tunable delay lines in a
multi-wavelength beamformer [58]. The programmability of the ring network will
allow the chip to be recongured in real time, to synthesize different responses
according to the user demand.
Other material platforms such silicon can also be good candidates to construct
this ASP. However, Si3N4 waveguides might be more suited for this purpose due to
the following characteristics: ultra-low propagation losses (0.1 dB/cm [59]) which
is very important for MWP systems, moderately high optical nonlinearity (10 times
of silica) [33] and free from nonlinear losses such as two-photon absorption (TPA)
and free-carrier absorption (FCA). On top of this, Si3N4 waveguides have shown
compatibility with efcient thermo-optics tuning and is a CMOS compatible
material. While silicon also ticks some of the boxes and successful wavelength
multicasting has been demonstrated in this platform [3], the propagation loss of a
typical SOI nanowires is relatively high (23 dB/cm) [4] and they suffer from TPA
and FCA. These losses might be prohibitive for creating an efcient nonlinear
IMWP processor.
14.6.2 Highly Integrated Tunable RF Filter
Harnessing stimulated Brillouin scattering (SBS) in nanophotonic devices has

recently attracted signicant interest due to its great potential for realizing tunable
slow light, narrow-linewidth laser, optical frequency comb, and RF photonic signal
processor in a compact footprint. Observation of SBS in a CMOS-compatible
platform has recently been reported in a hybrid Si-Si3N4 photonic/phononic
waveguide [46], but the amount of realized SBS gain was only in the order of 1 dB,
which is far from sufcient for any actual application. While high gain and power
efcient SBS nanophotonic devices are ultimately desired, elegantly harnessing the
low SBS gain from current on-chip low-power devices to function in real-life
applications represents a technological breakthrough.
As discussed previously, using the novel notch lter scheme one can create a
high performance RF notch with an ultra-high suppression using less than 1 dB of
on-chip SBS and very low a pump power. The ultimate goal for such lter is,
however, to integrate more functionalities in a single photonic chip, together with
the optical waveguides where SBS is induced. This vision in integration is illus-
trated in Fig. 14.9b.
The realization of this integrated notch lter can go through two different paths:
monolithic integration, or hybrid integration. Silicon for example, stands out as an
ideal platform for monolithic integration of this lter. The fact that high speed
modulators and photodetectors are readily available in silicon is, indeed, a big
advantage. However, thus far there is still no observation of SBS in silicon, due to a
number of effects such as acoustic mode leakage [19]. Thus, the realization of
monolithic integration of this SBS notch lter will depend strongly on the efforts in
efciently inducing SBS in silicon, or other CMOS compatible materials. It is worth
to mention that monolithic integration can still be achieved with OFs other than
SBS, such as silicon ring resonators [27], or silicon integrated Bragg-gratings [7].
The hybrid integration route, on the other hand, has recently seen some
encouraging progress. hybrid integration of chalcogenide and lithium niobate
waveguides has recently been explored [22]. This is a very promising step to
combine currently two best materials for SBS on chip (chalcogenide) and high
speed electro-optic modulator (lithium niobate).
References
1. E.I. Ackerman, G.E. Betts, W.K. Burns, J.C. Campbell, C.H. Cox, N. Duan, J.L. Prince,
M.D. Regan, H.V. Roussell, in Proceedings of the IEEE MTT-S International Microwave
Symposium (IMS 2007) (Honolulu, HI, 2007) pp. 5154
2. G. Agrawal, Nonlinear Fiber Optics, 3rd edn. (Academic Press, London, 2001)
3. A. Biberman, B.G. Lee, A.C. Turner-Foster, M.A. Foster, M. Lipson, A.L. Gaeta, K. Bergman,
Wavelength multicasting in silicon photonic nanowires. Opt. Express 18(17), 1804718055
(2010)
4. W. Bogaerts, P. De Heyn, T. Van Vaerenbergh, K. De Vos, S.K. Selvaraja, T. Claes,

P. Dumon, P. Bienstman, D. Van Thourhout, R. Baets, Silicon microring resonators. Laser
Photonics Rev. 6(1), 4773 (2012)
5. C.S Bres, S. Zlatanovic, A.O.J. Wiberg, J.R. Adleman, C.K. Huynh, E.W. Jacobs, J.M. Kvavle,
S. Radic, Parametric photonic channelized rf receiver. IEEE Photonics Technol. Lett. 23(6),
344346 (2011)
6. C.-S. Brs, S. Zlatanovic, A.O.J. Wiberg, S. Radic, Recongurable parametric channelized
receiver for instantaneous spectral analysis. Opt. Express 19(4), 35313541 (2011)
7. M. Burla, L.R. Corts, M. Li, X. Wang, L. Chrostowski, J. Azaa, Integrated waveguide
Bragg gratings for microwave photonics signal processing. Opt. Express 21(21), 2512025147
(2013)
8. M. Burla, D. Marpaung, L. Zhuang, C. Roeloffzen, M.R. Khan, A. Leinse, M. Hoekman,
R. Heideman, On-chip cmos compatible recongurable optical delay line with separate carrier
tuning for microwave photonic signal processing. Opt. Express 19(22), 2147521484 (2011)
9. A. Byrnes, R. Pant, E. Li, D.-Y. Choi, C.G. Poulton, S. Fan, S. Madden, B. Luther-Davies,
B.J. Eggleton, Photonic chip based tunable and recongurable narrowband microwave
photonic lter using stimulated Brillouin scattering. Opt. Express 20(17), 1883618845 (2012)
10. C. Caloz, S. Gupta, Q. Zhang, B. Nikfal, Analog signal processing: a possible alternative or
complement to dominantly digital radio schemes. IEEE Microwave Mag. 14(6), 87103
(2013)
11. J. Capmany, D. Novak, Microwave photonics combines two worlds. Nat. Photonics 1(6),
319330 (2007)
12. J. Capmany, B. Ortega, D. Pastor, S. Sales, Discrete-time optical processing of microwave
signals. J. Lightwave Technol. 23(2), 702 (2005)
13. L.R. Chen, J. Li, M. Spasojevic, R. Adams, Nanowires and sidewall Bragg gratings in silicon
as enabling technologies for microwave photonic lters. Opt. Express 21(17), 1962419633
(2013)
14. S. Chin, L. Thvenaz, Tunable photonic delay lines in optical bers. Laser Photonics Rev. 6
(6), 724738 (2012)
15. S. Chin, L. Thvenaz, J. Sancho, S. Sales, J. Capmany, P. Berger, J. Bourderionnet, D. Dol,
Broadband true time delay for microwave signal processing, using slow light based on
stimulated Brillouin scattering in optical bers. Opt. Express 18(21), 2259922613 (2010)
16. C. Cox, E. Ackerman, R. Helkey, G.E. Betts, Techniques and performance of intensity-
modulation direct-detection analog optical links. IEEE Trans. Microw. Theory Tech. 45(8),
13751383 (1997)
17. C.H. Cox, Analog Optical Links: Theory and Practice (Cambridge University Press,
Cambridge, 2004)
18. C.H. Cox, E.I. Ackerman, G.E. Betts, J.L. Prince, Limits on the performance of RF-over-ber
links and their impact on device design. IEEE Trans. Microw. Theory Tech. 54(2), 906920
(2006)
19. B.J. Eggleton, C.G. Poulton, R. Pant, Inducing and harnessing stimulated Brillouin scattering
in photonic integrated circuits. Adv. Opt. Photon. 5(4), 536587 (2013)
20. B.J. Eggleton, T.D. Vo, R. Pant, J. Schr, M.D. Pelusi, D.Y. Choi, S.J. Madden, B. Luther-
Davies, Photonic chip based ultrafast optical processing based on high nonlinearity dispersion
engineered chalcogenide waveguides. Laser Photonics Rev. 6(1), 97114 (2012)
21. M.H. Khan, H. Shen, Y. Xuan, L. Zhao, S. Xiao, D.E. Leaird, A.M. Weiner, M. Qi,
Ultrabroad-bandwidth arbitrary radiofrequency waveform generation with a silicon photonic
chip-based spectral shaper. Nat. Photonics 4(2), 117122 (2010)
22. J. Kim, W.J. Sung, O. Eknoyan, C.K. Madsen, Linear photonic frequency discriminator on
as2s3-ring-on-ti:linbo3 hybrid platform. Opt. Express 21(21), 2456624573 (2013)
23. A. Loayssa, F.J. Lahoz, Broad-band rf photonic phase shifter based on stimulated Brillouin
scattering and single-sideband modulation. IEEE Photonics Technol. Lett. 18(1), 208210
(2006)
24. D. Marpaung, B. Morrison, R. Pant, D.Y. Choi, S. Madden, B. Luther-Davies, B.J. Eggleton,
A tunable rf photonic notch lter with record 55 db suppression using sub-1 db on-chip
Brillouin gain. in Frontiers in Optics 2013 Postdeadline. Opt. Soc. A., p. FW6B.9 (2013)
25. D. Marpaung, M. Pagani, B. Morrison, B.J. Eggleton, Nonlinear integrated microwave
photonics. J. Lightwave Technol. 32(20), 34213427 (2014)
26. D. Marpaung, B. Morrison, R. Pant, B.J. Eggleton, Frequency agile microwave photonic notch
lter with anomalously high stopband rejection. Opt. Lett. 38(21), 43004303 (2013)
27. D. Marpaung, B. Morrison, R. Pant, C. Roeloffzen, A. Leinse, M. Hoekman, R. Heideman,
B.J. Eggleton, Si3n4 ring resonator-based microwave photonic notch lter with an ultrahigh
peak rejection. Opt. Express 21(20), 2328623294 (2013)
28. D. Marpaung, C. Roeloffzen, R. Heideman, A. Leinse, S. Sales, J. Capmany, Integrated
microwave photonics. Laser Photonics Rev. 7(4), 506538 (2013)
29. J.D. McKinney, M. Godinez, V.J. Urick, S. Thaniyavarn, W. Charczenko, K.J. Williams, Sub-
10-dB noise gure in a multiple-GHz analog optical link. IEEE Photonics Technol. Lett. 19
(7), 465467 (2007)
30. A. Meijerink, C.G.H. Roeloffzen, R. Meijerink, L. Zhuang, D.A.I. Marpaung, M.J. Bentum,
M. Burla, J. Verpoorte, P. Jorna, A. Hulzinga, W. van Etten, Novel ring resonator-based
integrated photonic beamformer for broadband phased array receive antennaspart i: design
and performance analysis. J. Lightwave Technol. 28(1), 318 (2010)
31. R.A. Minasian, Photonic signal processing of microwave signals. IEEE Trans. Microw.
Theory Tech. 54(2, Part 2), 832846 (2006)
32. B. Morrison, D. Marpaung, R. Pant, E. Li, D.-Y. Choi, S. Madden, B. Luther-Davies,
B.J. Eggleton, Tunable microwave photonic notch lter using on-chip stimulated Brillouin
scattering. Opt. Commun. 313, 8589 (2014)
33. D.J. Moss, R. Morandotti, A.L. Gaeta, M. Lipson, New CMOS-compatible platforms based on
silicon nitride and Hydex for nonlinear optics. Nat. Photonics 7(8), 597607 (2014)
34. W. Ng, A.A. Walston, G.L. Tangonan, J.J. Lee, I.L. Newberg, N. Bernstein, The rst
demonstration of an optically steered microwave phased array antenna using true-time-delay.
IEEE Trans. Microw. Theory Tech. 9(9), 11241131 (1991)
35. Y. Okawachi, M.S. Bigelow, J.E. Sharping, Z. Zhu, A. Schweinsberg, D.J. Gauthier,
R.W. Boyd, A.L. Gaeta, Tunable all-optical delays via Brillouin slow light in an optical ber.
Phys. Rev. Lett. 94, 153902 (2005)
36. M. Pagani, D. Marpaung, D.Y. Choi, S.J. Madden, B. Luther-Davies, B.J. Eggleton, On-chip
wideband tunable rf photonic phase shifter based on stimulated Brillouin scattering. in Proc.
OptoElectronics and Communications Conference (OECC), pp. 5154 (2014)
37. M. Pagani, D Marpaung, B Morrison, B.J. Eggleton, Bandwidth tunable, high suppression rf
photonic lter with improved insertion loss. in CLEO: 2014, p. STu2G.7. Opt. Soc. Am.
(2014)
38. R. Pant, A. Byrnes, C.G. Poulton, E. Li, D.-Y. Choi, S. Madden, B. Luther-Davies,
B.J. Eggleton, Photonic-chip-based tunable slow and fast light via stimulated Brillouin
scattering. Opt. Lett. 37(5), 969971 (2012)
39. R. Pant, D. Marpaung, I.V. Kabakova, B. Morrison, C.G. Poulton, B.J. Eggleton, On-chip
stimulated Brillouin scattering for microwave signal processing and generation. Laser
Photonics Rev. 8(5), 653666 (2014)
40. R. Pant, C.G. Poulton, D.-Y. Choi, H. Mcfarlane, S. Hile, E. Li, L. Thevenaz, B. Luther-
Davies, S.J. Madden, B.J. Eggleton, On-chip stimulated Brillouin scattering. Opt. Express 19
(9), 82858290 (2011)
41. M.S. Rasras, C.K. Madsen, M.A. Cappuzzo, E. Chen, L.T. Gomez, E.J. Laskowski, A. Grifn,
A. Wong-Foy, A. Gasparyan, A. Kasper, J. Le Grange, S.S. Patel, Integrated resonance-
enhanced variable optical delay lines. IEEE Photonics Technol. Lett. 17(4), 834836 (2005)
42. C.G.H. Roeloffzen, L. Zhuang, C. Taddei, A. Leinse, R.G. Heideman, P.W.L. van Dijk,
R.M. Oldenbeuving, D.A.I. Marpaung, M. Burla, K.J. Boller, Silicon nitride microwave
photonic circuits. Opt. Express 21(19), 2293722961 (2013)
43. J.E. Roman, L.T. Nichols, K.J. Wiliams, R.D. Esman, G.C. Tavik, M. Livingston, M.G. Parent,
Fiber-optic remoting of an ultrahigh dynamic range radar. IEEE Trans. Microw. Theory Tech. 46
(12), 23172323 (1998)
44. A.J. Seeds, K.J. Williams, Microwave photonics. J. Lightwave Tech. 24(12), 46284641
(2006)
45. A.J. Seeds, Microwave photonics. IEEE Trans. Microw. Theory Tech. 50(3), 877887 (2002)
46. H. Shin, W. Qiu, R. Jarecki, J.A. Cox, R.H. Olsson, A. Starbuck, Z. Wang, P.T. Rakich,
Tailorable stimulated Brillouin scattering in nanoscale silicon waveguides. Nat. Commun. 4
(2013). doi:10.1038/ncomms2943
47. K.Y. Song, M.G. Herrez, L. Thvenaz, Observation of pulse delaying and advancement in
optical bers using stimulated Brillouin scattering. Opt. Express 13(1), 8288 (2005)
48. K. Tan, D. Marpaung, R. Pant, F. Gao, E. Li, J. Wang, D.-Y. Choi, S. Madden, B. Luther-
Davies, J. Sun, B.J. Eggleton, Photonic-chip-based all-optical ultra-wideband pulse generation
via xpm and birefringence in a chalcogenide waveguide. Opt. Express 21(2), 20032011
(2013)
49. V.J. Urick, M.S. Rogge, F. Bucholtz, K.J. Williams, Wideband (0.0456.25 GHz) 40 km
analogue bre-optic link with ultra-high (>40 dB) all-photonic gain. Electron. Lett. 42(9),
552553 (2006)
50. B. Vidal, J. Palaci, J. Capmany, Recongurable photonic microwave lter based on four-wave
mixing. IEEE Photonics J. 4(3), 759764 (2012)
51. B. Vidal, M.A. Piqueras, J. Mart, Tunable and recongurable photonic microwave lter based
on stimulated Brillouin scattering. Opt. Lett. 32(1), 2325 (2007)
52. A.E. Willner, O.F. Yilmaz, J. Wang, X. Wu, A. Bogoni, L. Zhang, S.R. Nuccio, Optically
efcient nonlinear signal processing. IEEE J. Sel. Top. Quantum Electron. 17(2), 320332
(2011)
53. J. Yao, Microwave photonics. IEEE Photonics Technol. Lett. 27(3), 314335 (2009)
54. J. Yao, Microwave photonics arbitrary waveform generation. Nat. Photonics 4(2), 7980
(2010)
55. J. Zhang, A.N. Hone, T.E. Darcie, Limitation due to signal-clipping in linearized microwave-
photonic links. IEEE Photonics Technol. Lett. 19(14), 10331035 (2007)
56. W. Zhang, R.A Minasian, Widely tunable single-passband microwave photonic lter based on
stimulated Brillouin scattering. IEEE Photonics Technol. Lett. 23(23), 17751777 (2011)
57. W. Zhang, R.A Minasian, Ultrawide tunable microwave photonic notch lter based on
stimulated Brillouin scattering. IEEE Photonics Technol. Lett. 24(14), 11821184 (2012)
58. L. Zhuang, M. Hoekman, W. Beeker, A. Leinse, R. Heideman, P. van Dijk, C. Roeloffzen.
Novel low-loss waveguide delay lines using Vernier ring resonators for on-chip multi-
microwave photonic signal processors. Laser Photonics Rev. 7(6), 9941002 (2013)
59. L. Zhuang, D. Marpaung, M. Burla, W. Beeker, A. Leinse, C. Roeloffzen, Low-loss, high-
index-contrast si3n4/sio2 optical waveguides for optical delay lines in microwave photonics
signal processing. Opt. Express 19(23), 2316223170 (2011)
60. L. Zhuang, C.G.H. Roeloffzen, A. Meijerink, M. Burla, D.A.I. Marpaung, A. Leinse,
M. Hoekman, R.G. Heideman, W. van Etten, Novel ring resonator-based integrated photonic
beamformer for broadband phased array receive antennaspart ii: Experimental prototype.
J. Lightwave Technol. 28(1), 1931 (2010)
61. H. Zmuda, R.A. Soref, P. Payson, S. Johns, E.N. Toughlian, Photonic beamformer for phased
array antennas using a ber grating prism. IEEE Photonics Technol. Lett. 9(2), 241243
(1997)
Chapter 15
Ultrafast Optical Techniques
for Communication Networks and Signal
Processing
Bhavin J. Shastri, John Chang, Alexander N. Tait,

Matthew P. Chang, Ben Wu, Mitchell A. Nahmias
and Paul R. Prucnal
Abstract Wireless communications, for data services in particular, have witnessed

an exponential growth, and wireless spectrum shortages necessitate increasingly
sophisticated methods to use spectrum efciently. The backhaul of nearly all
wireless data networks is ber-optic. Analog optical signal processing techniques,
or microwave photonics, provides an ideal platform for processing wireless infor-
mation before it is transported to data aggregation centers by bers. It is in this
context that we present recent advances in optical signal processing techniques for
wireless radio frequency (RF) signals. Specically, this chapter is devoted to the
discussion of photonic architectures for wideband analog signal processing,
including RF beamforming, co-channel interference cancellation, and physical layer
security. Photonics offers the advantages not only of broadband operation, but
reduced size, weight, and power, in addition to low transmission loss, rapid re-
congurability, and immunity to electromagnetic interference.
15.1 Introduction
The use of wireless communication is growing exponentially. In June 2012, more

than 5.6 billion subscribers had access to and were using a wireless device, nearly
80 % of the total world population of 7.02 billion [1]. By the end of 2017, more
than 90 % of the worlds population is expected to have access to mobile broadband
3G devices. This statistic demonstrates the importance of mobile wireless com-
B.J. Shastri and J. Changequal contribution.
B.J. Shastri J. Chang A.N. Tait M.P. Chang B. Wu M.A. Nahmias P.R. Prucnal (&)
Princeton University, Princeton, NJ 08544, USA
e-mail: prucnal@princeton.edu

470 B.J. Shastri et al.
munication in the current decade and beyond. By the end of 2015, mobile trafc is
expected to double. Nearly 11 exabytes will be transmitted per month in 2016, more
than four times the quantity transmitted in the last 2 years. Mobile broadband,
growing with a compound rate of 80 %, is shaping to become one of the most
successful and fastest growing industries in history. This growth in the mobile
market has been spurred by two key factors: fast, powerful, and ubiquitous mobile
networking, and high-performance smartphones with a million-plus mobile appli-
cations [2]. The rapid shift from traditional headsets to smartphones has been
attributed to increased performance and lower energy consumption.
As wireless communication continues to grow, the radio frequency spectrum
will become increasingly scarce. The climb in the demand for smartphones and
faster, large-coverage next-generation mobile networks will be paralleled by a
demand for available bandwidthbut this demand comes with an equal increase in
wireless spectrum use. There will be two technological barriers that will result from
these increased demands: rst, wireless spectrum is nitely limited and is con-
strained, and spectral efciency use must be carefully managed as bandwidth begins
to grow. Second, a heavily saturated spectrum means interference from neighboring
devices and frequency bands could negatively impact data rates and signal
bandwidth.
One way of addressing both these escalating problems is through the use of
interference cancellation. Although noise-ltering and other post-processing
methods could reduce interference, a more effective technique would involve
cancelling interference directly in the radio frequency (RF) domain. This technique
reduces interference and increases bandwidth availability. As a result, the appli-
cation of RF interference cancellation could have a signicant impact on the alle-
viation of overcrowding in the wireless domain.
Figure 15.1 shows a typical scenario seen by a wireless communication system
(such as a radar, for instance). There is a clear target, whose information we wish to
receive. Along with that information, however, is environmental clutter and other
interfering signals. Clutter results from the local environment, which can be
exacerbated by complex terrain or weather conditions. Noise and signals from other
sources also contribute to interference. Our goal is to design a system to extract a
target signal, or signal of interest (SOI), from this noise.
Adaptive arrays in electronics (Fig. 15.2) have been used to increase signal-to-
noise ratio (SNR) in the presence of dynamic interference and noise in wireless
communication systems since the 1960s and are a perfect solution for the problem
outlined above [3]. Arrays, instead of singular antennas, give the advantages of
higher gain (through multiple antennas), electronic beam forming (steering of the
beam-pattern without physical movement), and adaptive cancellation (by taking
advantage of beamforming capabilities). Because these adaptive arrays are both
tunable and easily recongurable, they are more reliable than conventional antennas
and can be used in multiple scenarios across different elds. However, electronic
systems suffer from two major decits: beam squinting and limited bandwidth.
Beam squinting arises from the use of phase shifters, which degrades an antenna
arrays beam-pattern with frequency, resulting in different responses for signals of
15 Ultrafast Optical Techniques for Communication Networks 471
Fig. 15.1 Typical clutter in wireless environment [5]
Fig. 15.2 a An antenna array uses phase information to detect the direction of arrival of an
incident wave. b Adaptive array coverage: a representative depiction of a main lobe extending
toward a user with null directed toward two co-channel interferers [3]
different frequencies. Bandwidth limitations result from the limited speed of elec-
tronic devices. This limits the use of electronic systems to narrow-band signals.
Furthermore, losses increase for electronic systems past 100 MHz, limiting systems
that wish to operate in the tens of GHz.
While wireless networks have seen a rapid increase their bandwidth capabilities,
optical networks have been popular for a decade and their capabilities greatly over-
shadow even the fastest wireless standard. Figure 15.3 compares wireless and optical
networks. Optical network growth follows that of wireless networks and is at least
10 faster at each step. Because of the inherent bandwidth superiority of optics,
wireless data is always aggregated and backhauled to data centers by the way of
optical networks. This serves as the perfect backdrop for an optical cancellation
system sandwiched in-between the wireless base stations and the optical backhaul
network. Wireless signals are modulated onto the optical domain and then directly
processed using a photonic cancellation system before being sent over an optical
network. This eld is known as microwave photonics, which is an interdisciplinary
area that studies the interaction between microwave and optical signals [4].
Fig. 15.3 Advances in wireless and optical capacity [1]
Optical systems bring advantages inherent to the physics of photonics, which

include low loss, high bandwidth, immunity to electromagnetic interference (EMI),
tunability, and recongurability [5]. Most importantly, optical systems solve both
the beam squinting and bandwidth limitations of electronic systems. By using
optical delay lines, true-time delay (TTD) antenna arrays eliminate beam squinting
[4]. Electronic systems oftentimes use multiple pieces of equipment to cover a wide
bandwidth. In contrast, because photonic systems keep the signal in the optical
domain, they can utilize the tremendous bandwidth of optics (up to THz) to process
fast-varying RF signals without degrading bandwidth using a single set of optical
equipment. This gives optical systems savings in size, weight, and power (SWAP),
especially for situations where compactness and light weight are important (i.e.
aviation). Furthermore, optical interference cancellation in conjunction with optical
backhaul networks makes it possible to centralize signal processing in one shared
location. Centralization allows for equipment sharing, dynamic allocation of
resources, simplied system operation and maintenance [5]. Different operators can
share resources using the same optical backhaul network to minimize cost and
energy.
Microwave photonics lters (MPFs) are an alternative photonic technology and
have an enormous amount of bandwidth selectivity. They can be employed in
optical systems for either channel rejection or channel selection to cancel interfering
signals that are picked up by antennas [4, 6, 7]. Unique to photonics lters is the
rejection of signals in the optical domain using optical signal processing techniques.
MPFs can be tuned to cancel a selected frequency band across a large bandwidth
from MHz to hundreds of GHz [5, 8].
This chapter investigates an optical method for wireless interference cancellation
systems using optical signal processing techniques. We marry the eld of adaptive
arrays and microwave photonics. Beam-steering is done by adaptive antenna arrays
with processing done all-optically by MPFs. By utilizing the broad bandwidth and
high capacity of optics to address growing problems in mobile communication

market, we hope to create a photonic interference cancellation system with wide-
ranging impact, both economically and scientically.
15.1.1 Scenario of Interest
This chapter is interested in a scenario with dynamically changing noise and

interference in the presence of a SOI. We assume that we have a receiver either in a
dense urban environment, such as in a mobile phone setting, or a crowded indoor
environment, such as in a WIFI network. We wish to follow a mobile transmitter as
it is broadcasting the SOI, so that its location, and possibly frequency, is constantly
changing. To make the problem even more challenging, we assume that we cannot
know anything about the SOI, other than its frequency, and have no information of
the content of this signal.
We make several additional assumptions: we must cancel an interfering signal to
receive the SOI. The SOI is much smaller in power than the interference. We also
lack knowledge of both the location and frequency of the interfering signal. The
interference is both dynamic and spontaneous, with multiple interferers appearing
and disappearing with shifting frequencies. We also do not know the relative power
or direction between the SOI and the interferers.
Since we are in a crowded environment (either urban or indoors), we can assume
high levels of dynamically changing noise. Transmitters and other users may appear
or disappear spontaneously and change locations, which is in addition to the
background noise that permeates throughout the wireless spectrum. Interference in
the form of multipath fading will be detrimental to signal quality. In the indoor
scenario, near-eld signals dominate, with wavefronts that may not be planar or
uniform with respect to the receiver.
The problem requires a two-part solution. First, lters for both space and fre-
quency are required to rid the interference and noise from the SOI. Secondly, an
adaptive method is needed to quickly keep track of moving targets and rapidly
changing interference. However, we do not have access to either the SOI or the
interference and we cannot use a pilot signal. Unique to our scenario is that the
signal incident on the receiverthe input signalis inaccessible and only a signal
processed by the lter is usable. This scenario nullies the use of traditional
adaptive algorithms, which typically require pre-steering, training, or an input
signal. Multipath effects are also signicant in a dynamically noisy environment,
which fluctuate and are highly unpredictable, preventing a device from distin-
guishing between the loss of signal (LOS) or some nth order multipath. As a result,
we cannot make traditional direction-of-arrival (DOA) estimates. Because broad-
band noise can fluctuate wildly in frequency and direction, stochastic gradient
method are also unusable; even if signals remain the same, no two measurements
will be the same as a result of noise. This chapter investigates solutions to the
scenario introduced above.
We rst discuss the basic theory of beamforming and the unique challenges of
broadband interference cancellation. Next, we present the requirements and chal-
lenges for building optical nite impulse response (FIR) lters, and review several
innovative MPF designs including optical tunable delays and weighting schemes.
We also review the state of the art optical beamformers recently proposed and
experimentally demonstrated in literature. Furthermore, we detail a highly scalable
photonic beamforming architecture designed for a particularly non-stationary,
interfering environment. Finally, we introduce a specic application for the pho-
tonic beamformer namely, physical layer security in optical backhaul networks.
15.2 Primer on Antenna Arrays and Beamforming
Antennas arranged in some physical geometryarrayexploit the spatial rela-

tionship between the antenna elements to either measure or manipulate the spatial
properties of RF signals which is also called beam steering. Typically, several
antennas are located with uniform spacing of scale similar to the frequency of
interest. If the signal source is far away from the antenna array, the beam can be
considered as a plane wave with a wavefront orthogonal to the direction of prop-
agation. This wavefront arrives at the antenna array at some angle, and as it
propagates across the array it impinges upon each antenna with a delay experi-
encing a phase shift. The delay is related to both the angle of incidence of the beam
and the physical layout of the array. The antenna array is able to resolve the signals
DOA by calculating the time differences between the responses of each antenna.
This information can then be used to steer the beampattern. A signal processor can
follow each antenna. The signal processor can be used to manipulate the signal in a
useful way; that is, to suppress interferers and enhance the SOI.
More formally, consider an array of M antenna elements arranged in an xy plane.
The DOA denoted by a monochromatic plane wave incident on the array is denoted
as h; / where h is the elevation angle measured from the z axis and / is the
azimuth angle measured from the x axis. The received signal at antenna m is
transformed by the steering vector ah; / given by
am h; / e jxsm h;/ ; m 0; 1; . . .; M 1 15:1
where x is the signal carrier frequency, sm is the time delay associated with antenna
m characterized by the array geometry. To implement beam steering, the array
elements are weighted and summed. The incident signal st produces an output
signal yt
yt w ah; / st: 15:2
where w is the weight matrix vector which comprises of the complex coefcients
wm , and inner product w ah; / is called the array factor AF h; / [9].
To generalize, if the antenna array is receiving L signals s0 t; . . .; sL1 t with

different DOAs h0 ; /0 ; . . .; hL1 ; /L1 , then the M L steering matrix is
Ah; / ah0 ; /0 . . . ahL1 ; /L1 15:3
and the system output is given by
yt wT Ah; / st: 15:4
If some of the signals are considered noise, the adaptive system becomes an
optimization problem over the space of the M antenna weights. The weights can be
chosen in such a way as to emphasize a signal coming from one direction while
attenuating signals from other directions.
Consider the linear and circular antenna arrays shown in Fig. 15.4 which we use
to illustrate the example of beam steering. The linear array consists of M antenna
(a)
(b)
Fig. 15.4 a Left Linear antenna array for beam steering. Phase difference is found from distance
csm . Right Response pattern of an unweighted linear array to a signal from different directions.
b Left A circular antenna array. Right Response pattern of an unweighted circular array
elements equally spaced distance d apart arranged in a straight line along the y axis.
If the rst antenna element has phase 0, then the time delay to antenna m can be
calculated by trigonometry to be
d
sm h; / m sin / sin h 15:5
c
where c is the speed of light. The linear array with weights wm then has an array
factor
X
M 1
wm e jxmc sin / sin h
d
AF h; / 15:6
m0
which is comparable in form to the frequency response of an M-tap FIR lter:

P jmX
F X M m0 wm e . Hence, selecting weights for a particular spatial response
is equivalent to selecting the weights of an FIR lter. The more popular circular
antenna array can steer a beam in any direction. Here, the M antenna elements are
arranged uniformly around a circle of radius r, with antenna 0 at / 0. If the
center of the circle has phase 0, then the time delay at each antenna element m is
given by

r 2pm
sm h; / cos / sin h 15:7
c M
which gives a response of
X
M 1
wm e jxc cos/ M sin h
r 2pm
AF h; / 15:8
m0
15.2.1 Narrowband Beamforming
To use the antenna array to steer a signal, the complex weights at each antenna are
adjusted to achieve the desired response. In the case of a narrowband signal, the
weights can be simple phase shifts chosen to replicate a plane wave in the desired
direction. These weights are pure phase delays, and can be realized with analog
phase shifters attached to each antenna or with digital signal processing (DSP).
Such a system is known as a phased array. For steering a beam with the linear and
circular arrays, the weights are
wm ejxmc sin / and wm ejxc cos/ ; respectively:

d r 2pm
M 15:9
Figure 15.5 illustrates an example of directional response of a steered circular

phase array. Phase shift steering provides control over the direction of the main lobe
which is centered on the look direction, but not over the relative placement of the
smaller side lobes on other directions or nulls which are the minima of the response;
signals from these directions are rejected by the array. To change the shape of the
response, variable weights are applied to the phase shifted signals as shown in
Fig. 15.6. An arbitrary phase shift is realized by adjusting the ratio between the
w
original signals and a quarter-shifted copy of the signal [10]: Du arctan wm;1 m;0
.
By controlling both the amplitude and phase of each antenna, this lter design
implements weights as complex numbers. The complex weight vector w can be
selected to solve the complex linear systems of (15.4) to achieve high gain of the
SOI and suppression of the interference signals. The equation can be solved for up
to M 1 interference signals which means the null steering array of M antennas
can place up to M 1 nulls [3]. If there are fewer than M 1 interference signals,
the additional degrees of freedom can be used to rene the gain levels and shapes of
the side lobes. However, if there are more than M 1 interference signals, then the
array cannot cancel them perfectly, but can still suppress them. In this case adaptive
algorithms can be used to nd the optimal weight vector for the incident signals.
These standard optimization algorithms include gradient estimation, digital least
mean square (LMS), and asynchronous LMS [11].
Filtering using phase shifts is only effective for narrowband signals, where the
phase shift corresponds with the time delay between the antennas. For signals with
different frequencies, the time delay between the antennas s d=c is constant, but
the phase shift Dum xsm changes with frequency. A phased array therefore has a
different response for each frequency and this is referred to as beam squint [4]. As a
result, a broadband signal can get distorted after passing through the lter. To
receive signals in a broad spectrum, the phase shifts can be replaced with TTDs
Fig. 15.5 Directional

response of a circular phase
array steered to angle 0.
Power is depicted in linear
scale
(a)
(b)
Fig. 15.6 a A weighted phased array beamformer applies both phase shifts and amplitude weights
to alter the shape of the response pattern. b A delay-and-sum beamformer uses true time delays to
receive a wideband signal
[12], as shown in Fig. 15.6b. A TTD can implement a phase shift greater than 2 for
some frequencies, as required by extremely broadband signals. Then the spatial
response is nearly the same for all frequencies.
Systems that use time delays for ltering are called delay-and-sum arrays.
Digital receivers can trivially shift input signals using hardware shift registers or
software signal processing. Optics is the perfect medium for creating TTD systems
for broadband signals. Since optical beamforming is based on using ber optics or
other types of discrete optical components that already exist in current broadband
optical networks, they are by nature broadband. TTD lters work well for receiving
broadband signals, but they only lter in the spatial domain. For interference signals
at different frequencies, it would be useful to control the frequency response as well
wideband beamforming. In this chapter, we will focus on photonic broadband
beamformers.
15.2.2 Wideband Beamforming
For wideband arrays, it is desirable to control not only the spatial response but also
the frequency response. To achieve an arbitrary frequency response, the weights wm
are replaced by linear lters hm t with frequency response Hm x. The combined
spatial and frequency of the system is the sum of the ltered array factors from
(15.8)
X
M 1
Hm xe jxc cos/ sin h :
r 2pm
Px; h; / M 15:10
m0
A signal whose frequency and direction of arrival falls near the peak of (15.8)
will be amplied whereas the one near null will be rejected by the lter. In the
adaptive array proposed by Widrow et al. in 1967 [10], the lters were analog
tapped delay lines (TDLs).
TDL lters, also known as transversal lters [13], are the continuous-time
analogue of the well-known discrete-time FIR lter. As shown in Fig. 15.7, the
input signal is delayed by successive time intervals, called taps. These delayed
signals are then weighted and summed. For an N-tap lter with weights wn and a
constant delay Td between each tap, the output is given by
X
N 1
y t wn txt nTd 15:11
n0
and the frequency response is
X
N1
H x wn ejxnTd : 15:12
n0
Note that both the frequency response of the TDL lter (15.12) and the spatial
response of the phase array (15.8) have a similar form to an FIR lter. Hence, in the
most general terms, the physical formation of the antennas provides spatial ltering
while the lters provide frequency ltering. By changing the weights of the FIR
lters, one can manipulate a beamformer to act as a spatial and frequency lter and
change the sensitivity of the beampattern to different frequencies and angles.
Mathematically, if an N tap lter is used for each of the M antenna elements in the
array, then there are MN total weights and delays in the system. The overall
response of the array is found by substituting (15.12) into (15.10):
X
M 1 X
N 1
wm;n ejxnTd c cos/ M sin h :
r 2pm
Px; h; / 15:13
m0 n0
The task of beamforming is to select the optimal set of MN weights to achieve

the desired shape of Px; h; /. Figure 15.8 shows the simulation response of a
Fig. 15.7 Antenna array with

tapped delay line lters
Fig. 15.8 Simulated spatial

and frequency response of a
tapped delay line lter using
arbitrary weights
wideband 4-antenna, 8-tap circular beamformer. The beam pattern for a system with
SOI of 500 MHz at 90o and interferences at 900, 1000, and 300 MHz at 120o, 70o,
and 60o respectively demonstrates both the spatial ltering on the x-axis and the
frequency ltering on the y-axis. We notice that nulls occur at the interferers while
there is a peak at the SOI frequency and angle.
15.3 Microwave Photonic Filters
The previous section introduced the FIR lter as a TDL architecture. This section
details the requirements and challenges for building optical FIR lters. We also
review several innovative MPF designs including optical tunable delays and
weighting schemes proposed and experimentally demonstrated in literature.
Optical lters or microwave photonic lters, are photonic subsystems speci-
cally designed with the objective of carrying out the same functions as those of
electronic or digital lters within the microwave range in an RF system or link [5].
The unique properties of MPFs offer many advantages, including high bandwidth,
low loss across the entire bandwidth, recongurability, and immunity to EMI. Thus,
there has been a considerable interest in the eld of photonic signal processing for
microwave ltering applications. Traditional RF and electronic approaches cannot
practically handle wide bandwidths in the GHz range, whereas processing in the
optical domain takes advantage of the broadband capabilities of optical delay
schemes. Optical lters require only a single set of equipment to be able to cancel
across a huge band of frequencies. Consequently, optical systems can potentially
offer SWAP advantages, that is, savings through size, weight, and power. This is
particularly critical for eld-tested military or aviation applications where size and
power are of key importance.
In this chapter, we are specically interested in the possible application of MPFs
in photonic phased array antennas, where they can provide the capability of steering
without any physical movement, and offer the ability for precise spatial and fre-
quency control of broadband signals. The key-motivating factor is that MPFs are
easily tunable and recongurable in real time, allowing for ltering with adaptive
capability. One can leverage these properties when designing such a system.
Moreover, the eld of photonic TTD units is essential for transversal, FIR lters (in
phased array antenna systems) as it allows for wide bandwidth signal processing
with little or no distortion or pulse broadening without being limited to a design
frequency. MPFs are the perfect optical complement for frequency ltering to the
spatial ltering of RF antenna arrays.
15.3.1 Requirements for MPFs
From (15.11), an FIR lter is also a TDL lter. In order to replicate the equation, we
need to be able to complete the same mathematical operations using optics. Each
tap of an FIR lter is a weighted delaymethods for optical weighting and optical
delay lines are necessary. The weighted delays are summed together to complete the
FIR lter, so an optical method of adding the symbols together is also needed. All
optical (and even analog) based lters require these three components: optical
weights, optical delay lines, and a summation method. Traditional optical fused
couplers are built specically for the purpose of combining optical signals (and
their powers), and since they are used widely in any optical communication system,
we will focus on optical methods for delays and weights here.
Figure 15.9 shows a general reference layout of an MPF. In all optical lters, the
RF signal must be converted to an optical signal. Generally, a continuous-wave
(CW) laser source modulates the RF signal onto the optical domain using modu-
lators such as electro-optic modulators (EOMs) like the Mach-Zehnder modulator
(MZM), or electro-absorption modulators (EAMs). The optical signal is then fed
Fig. 15.9 General reference layout of a microwave photonic lter showing the relevant electrical
and optical signals [5]
into a photonic circuit that delays and weights the signal on parallel taps and then
combines them with a fused coupler (or other method). The optical signal can be
converted back to an electrical signal using a photodetector.
15.3.2 Popular MPF Architectures
Many optical tunable delays and innovative weighting schemes have been proposed
and experimentally demonstrated in literature. We review some of the popular
techniques and present two unique MPF architectures.
Popular optical delaying techniques include lters that use ber-Bragg gratings
(FBGs) to delay lter taps [14]. Delays are created by forcing the optical signals to
propagate along the FBG until reflected by a wavelength-specic grating. FBGs are
popular since a whole array can be fabricated onto a single piece of optical ber.
Fiber delay lines have also been proposed for TTD in MPFs [15, 16], but are bulkier
than FBG-based delays when lter taps become large.
Fully tunable and programmable weighting using free space methods such as
spatial light modulators [17] and multi-port programmable wavelength processors
[18] have been demonstrated. Unfortunately, typical optical systems are generally
limited to positive tap coefcients, as light intensities are intrinsically positive.
Therefore, the shapes of their transfer functions are severely constrained and can only
be used to implement low-pass lters (LPFs). The addition of negative coefcients in
lter schemes allows for a practical lter with passband capabilities [5].
There are many ways to create negative optical weights. Recent advances in
negative coefcient weighting have used cross-phase modulation (XPM) in an
semiconductor optical ampliers (SOA)-MZ interferometer [19] and self-wave-
length cross gain modulation (XGM) in an SOA [20], but are suitable only for
realizing a few taps. Another technique involves using multiple phase inversion in
an SOA-based XGM wavelength converter [21]. An innovative method involving
sinusoidal group delay gratings, which use FBGs to achieve positive/negative
weighting and delays was proposed in [22].
Weight and delay are coupled, and it is difcult to have freely tunable atten-
uation without affecting the delays. Other optical weighting schemes involve
polarization modulation and optical polarizers [19, 23]. While these approaches are
simple, scalable, and tunable, the method relies heavily on mechanical polarization
controllers, which cannot rapidly change the polarity of the taps (to sub millisec-
ond-length time).
Simpler techniques entail using the negative bias slope of MZM to -shift the
output to create negative coefcients [2427]. However, some techniques neces-
sitate using two MZMs, which can be bulky when scaled [24]. Others using one
MZM rely on the dependence on wavelength but positive and negative coefcient
outputs are of different power. A novel technique using a specially designed inte-
grated 2 1 MZM to achieve -shifting was introduced in [26]. On the other hand,
a simple technique using a 1 2 dual output MZM to achieve negative weighting

by using phased-inversed dual outputs was proposed in [27].
Figure 15.10 shows an example of MPF using a spectrally-sliced free-space
conguration [5]. The most distinctive and important part of the setup is the optical
tapping through a sliced broadband source. A broadband laser is sliced by a dif-
fraction grating into 10 different wavelengths. The slicing into individual wave-
lengths allows summing to be done incoherently. Use of spatial light modulators
achieves the weighting and delaying of the samples. Fast tunability can be achieved
as individual spatial light modulators (SLMs) can be turned ON and OFF, which are
used to eliminate specic spectrum slices.
Figure 15.11 shows another realization of a MPF employing a discrete time
optical transversal ltering scheme [28]. The unique characteristic of this system is
the pulsed laser source being used. This means that the optical pulses sample the RF
Fig. 15.10 Transversal MPF using spectrally-sliced free-space optics [5]
Fig. 15.11 Discrete time optical transversal MPF [28]

input data (discretely). There is no need to have a separate analog-to-digital con-

verter (ADC) as the laser source serves that role. The delay is achieved through
ber delay lines. In order to avoid coherent summing, the delays are chosen so that
no optical pulses from one tap overlaps with pulses from another tap. Optical
attenuators provide weighting. There are no negative coefcients in this scheme.
15.3.3 Optical Technologies for MPFs
We now present the state of the art optical components and techniques that are
employed in MPFs. As stated earlier, MPFs need an optical delay scheme and also
an optical weighting scheme complete with negative coefcients. For an adaptable
lter, tunability of (at least) the weights is required which need to be electronically
programmable so that they can be quickly tuned. Table 15.1 summarizes some of
the novel optical components used for optical delays or weights in MPFs from
recent literature.
FBG arrays can be used for creating compact but xed delay lines [29]. The
drawback is that the xed delays prevent recongurability of the lter bandwidth.
Furthermore, this approach employs a spectrally sliced broadband amplied
spontaneous emission (ASE) noise source. While slicing this source will create
multiple optical lter taps, the ASE source is fundamentally noisy. An improvement
Table 15.1 State of the art optical technologies for microwave photonic lters
References Functionality Technology Advantages Limitations
[29] Optical FBG arrays No need to use bulky Spectrally sliced
delays and slowly tunable broadband ASE noise
ber delay lines source; hard to tune and
recongure
[30] Optical WDM lter WDM system, No tunability/
delays with FBG implementation of programmability of delays
arrays negative coefcients
[31] Optical WDM lter WDM system, No negative coefcients;
delays with recongurable lacks easy tunability/
chirped bandwidth programmability of
FBG arrays weights
[17] Optical Spatial Programmability of Free space optical setup
weights light weights (not compact); spectrally
modulators slice a noisy ASE source
[18] Optical Liquid Programmability of Free space optical setup
weights crystal on weights (not compact); spectrally
silicon slice a noisy ASE source
[27] Optical 1 2 MZM Simple Spectrally slice a noisy
weights and implementation of ASE source; lack easy
delays negative weights, tunability/programmability
delay using FBG of weights
arrays
can be achieved by using a wavelength-division multiplexing (WDM) scheme

paired with FBG delay lines [30]. This technique uses an array of distributed
feedback (DFB) lasers for the optical taps. While this gives the lter more dynamic
range, it sacrices simplicity and also lacks tunable delay lines. In contrast, using
chirped FBGs with tunable DFB lasers allows controllably variable delay lines [31].
However, this technique does not incorporate negative coefcients.
Similarly, many technologies have been developed for optical weighting
schemes. Free-space weighting schemes have recently garnered popularity
including the weighting technique based on SLM [17] and liquid crystal on silicon
(LCOS) [18]. Both these methods are extremely precise, can scale to lter orders of
hundreds, and are easily programmable. However, they still have a couple of
drawbacks. Firstly, they are free space optical techniques that are bulky and occupy
a much larger footprint than discrete optics. Secondly, they lack the ability to create
negative coefcients. A simple and compact method for creating negative coef-
cients can be integrated into the modulators that convert the electrical signal to the
optical signal [27]. However, this negative weighting technique must be combined
with another set of tunable optical weights for programmability.
15.4 Photonic Adaptive Beamformers
In this section, we rst review the state of the art optical beamformers recently
proposed and experimentally demonstrated in literature. We also detail a highly
scalable photonic beamforming architecture designed for a particularly non-sta-
tionary, interfering environment. Finally, we introduce a specic application for the
photonic beamformer namely, physical layer security in optical backhaul networks.
Beamformers have attracted signicant interest because of their wide range of
applications from radar, communication, and sensing. Beamforming is a technique
that utilizes an array of antennas as a spatial and frequency lter to manipulate the
beampattern to maximize the SNR of the SOI while cancelling/suppressing inter-
ferers and noise without a priori knowledge [32]. For instance, a beamformer can
spatially separate two signals that have overlapping frequencies but originate from
two different spatial locations [33]. Beamformers use optical FIR lters to process the
signals temporally with thermo-optic optical attenuatorscontrolling the signal
amplitudes and delays from each antenna elementto adaptively and rapidly adjust
the beampattern; that is, the whole array can act in unison to steer the beam pattern.
Meanwhile, the geometric conguration of the antennas allows the signals to be
processed spatially. This makes beamforming attractive for overcoming the direc-
tivity problems of a single antenna while offering a higher gain [3]. Adaptive
beamformers are particularly crucial in applications involving a highly non-stationary
target environment. Compared to conventional RF beamformers which have limited
narrowband performance due to their reliance on electrical phase shifters, the
wideband nature of optics offers a clear advantage. The architecture also offers the
distinct advantage of scalability to hundreds of antennas, as needed for practical

systems.
The rst adaptive beamformers were used as self-phased antennas that reradiated
signals in the direction in which they were received without prior signal knowledge.
In the 1960s, Howell and Applebaum introduced their versions of adaptive antennas
for interference nulling [3]. Widrow followed closely with self-optimizing adaptive
algorithms [10]. Applebaum and Widrow focused on beamforming specically for
applications in the eld of sonar and radar signal reception. Seismic array devel-
opment was especially popular in the 1960s. Recently, the major area of interest is
radar and communication systems for interference suppression [3]. Direction
nding in severe interference environments has also been a popular topic, as is
scanning for high angular resolution imaging. We will focus on interference can-
cellation techniques in this section.
15.4.1 State of the Art Photonic Beamformers
This section provides a review of the recently demonstrated state of the art optical
beamformers in literature. As explained previously, a photonic beamformer is an
array of antennas with a set of optical signal processors attached to each antenna. In
the previous section we stipulated the requirements needed for an FIR lter: a
method for creating both optical delays and optical weights. The same requirements
are necessary for a photonic beamformer. Moreover, optical delays with TTD
capability is essential for a beamformer, as explained in Sect. 15.2. However, since
a beamformer consists of multiple FIR lters (one for each antenna), scalability
with optical components is very important as well. We focus the literature review
on methods for TTD optical beamformers with an emphasis on scalability.
Figure 15.12 shows a TTD beamformer architecture based on FBG prisms. FBG
arrays are created in a prism format in which different sets of delays (ve as
depicted in the gure) are selected for a specic wavelength of a laser. The FBG
prism is built so that the delays create a pre-steered beamformer [34]. However, this
architecture cannot be fully tunable, as the number of pre-steered directions is
limited by the number of FBGs etched in the array.
A similar approach but that is tunable is shown in Fig. 15.13. In this scheme,
instead of FBGs, the beamformer uses dispersive bers that can be precisely tuned
by a nely tunable laser [35]. However, this method, while more precise, is
inherently slowed by the tuning speed of the laser.
Figure 15.14 depicts a scheme for a TTD beamformer based on WDM demul-
tiplexer (demux) delays [36]. The architecture consists of a single antenna, a tunable
laser with 8 s, and a corresponding WDM demux which selects one of eight
possible delays. The limitation in its resolution depends on the tunable laser and the
amount of wavelengths supported by the arrayed-waveguide (AWG) demux.
Fig. 15.12 Beamforming architecture based on FBG prism [34]
Fig. 15.13 Beamforming architecture based on dispersive ber prism [35]
Furthermore, the number of lasers required scales linearly with the number of
antennas, which can become bulky with large beamforming systems that necessitate
hundreds of antennas.
Fig. 15.14 Beamforming architecture based on WDM demux delay lines [38]
Fig. 15.15 SLM delay lines [37]
Fig. 15.16 Bit-switched delay lines [41]
Antenna arrays that use SLMs to perform optical delays have also been proposed
[37]. By apply an external electrical eld as shown in Fig. 15.15, dielectric mol-
ecules in the SLM form different polarizations. The polarization changes the index
of refraction of the delay line which affects the length of time the light travels
through the device.
Photonic crystal ber and XGM in SOAs have also been used to create TTD
lines in photonic beamformers for multi-transmit systems [38, 39]. On the other
hand, bit-switched delay lines are also popular [40, 41]. Figure 15.16 shows an
example of bit-switched delay line that selects a particular delay from a chain of
FBG arrays using a tunable laser.
A beamforming architecture based on opto-VLSI is shown in Fig. 15.17 [42].

The opto-VLSI processor is an array of liquid crystal (LC) cells driven by a Very-
Large-Scale-Integrated (VLSI) circuit. Delays can be created for each antenna by
slicing a broad-band optical source (ASE-based source) and routing specic sliced
wavebands through the Opto-VLSI processor to a high-dispersion ber [42].
A beamforming architecture based on high dispersion ber has also been
reported [43]. Here the beamformer uses an array of lasers that each experiences a
different delay based on passing through a wavelength-dependent high dispersion
ber. Figure 15.18 depicts the architecture of a single antenna. By using two -
shifted EOMS, negative coefcients can also be implemented. However, this
technique requires a laser array for each antenna array.
Recently, the integration of the photonic components into an integrated beam-
former-on-chip has been explored [44, 45]. Next generation systems investigate a
scalable integrated photonic beamformer that can be electronically controlled for
adaptive interference cancellation. The possibility of creating a beamformer-on-chip
has become a question of not if but when. Figure 15.19a illustrates an integrated
beamformer. Figure 15.19b shows a 16-antenna beamformer integrated on a chip,
roughly the size of a 20-cents of an euro coin. These systems show the potential for
extreme savings in SWAP along with full broadband processing ability.
Fig. 15.17 Beamforming architecture based on Opto-VLSI [42]

Fig. 15.18 Beamforming architecture based on high dispersion ber [43]
15.4.2 Highly Scalable Adaptive Photonic Beamformer
The key characteristic of a recongurable and tunable photonic beamformer is

scalability. Thus far, most of the beamformer architectures reviewed, are not
scalable to large arraysthe number of lasers required by these systems increases
linearly with the number of antennas. In this section, we present a highly scalable
photonic beamformer [46], unique to our lab, which is specically designed for a
particularly non-stationary, interfering environment. Using optical transversal lters
for each antenna element and thermo-optic optical attenuators, the array is capable
of both spatial beamforming and frequency-domain ltering while adaptively and
rapidly adjusting the beampattern. Our architecture offers the distinct advantage of
scalability to hundreds of antennas, as needed for practical systems, by using a
novel single-mode to multimode (SM-MM) combiner, and the same set of laser
wavelengths can be used for every antenna in the system. By eliminating coherent
effects, our system uses the same xed set of optical wavelengths for each antenna
in the system, resulting in a simple and compact architecture. We present experi-
mental results to show proof-of-concept and demonstrate the proposed adaptive
beamformer performance.
Fig. 15.19 a Potentially integrated beamforming network with EAM array [44]. b Schematic of
16 1 beamforming network with chip mask layout and photograph of a fabricated chip [45]
15.4.2.1 System Overview
Figure 15.20 shows the architecture of a wideband photonic beamformer that we

recently demonstrated experimentally. The key component of the beamformer is the
optical transversal lter which is driven by two eight-channel DFB laser arrays. The
rst array of wavelengths 18 corresponds to the positive coefcients whereas the
second array of wavelengths 916 corresponds to the negative coefcients. The
optical sources are inserted into a compact 16-channel thermal-optic attenuator for
easy adaptive control of the weights through a computer or voltage source. The
attenuators have a response time of 10 s per 0.1 and a 20 dB range.
The weighted taps are then combined using an AWG multiplexer (mux). The RF
signal to be processed is modulated onto the optical carrier using a dual output
electro-optic MZM. The modulated signals of the outputs are biased at the inverse,
-shifted, parts of the modulator transfer function. We use this complimentary
output to implement negative coefcients. Both outputs have equal insertion losses
of 3.7 dB. The weighted signals exit from both the positive and negative outputs of
the MZMs. The complementary outputs are launched in FBG arrays that only
reflect and delay the wavelengths assigned to the respective coefcients, via an
optical circulator (OC).
The coefcients encounter FBGs with the same delays, but at different wave-
lengths. As a result, each delay has both a positive and negative tap, and the
attenuators are used to switch on the tap and weight each tap by enabling/disabling
a certain wavelength. In this way, our 16-wavelength lter provides eight positive/
negative taps. Since time delays and lter bandwidth are inversely proportional,
fabricating FBGs with closer spacing and shorter delays can increase bandwidth.
There is a total optical insertion loss of *19.5 dB for each lter. If each DFB
laser has 13 dBm of power, each lter receives 9.4 dBm of power after splitting to
four antennas and outputs 10.1 dBm into the combiner. Assuming a four-antenna
beamformer, 4.6 dBm of power nally reaches the high speed MM photodetector,
well above the 25 dBm limit.
The outputs of the optical lters are summed using a special SM-MM optical
combiner, as in Fig. 15.20. The architecture is a blind adaptive approach, in which
the adaptive algorithm only has access to the output of the system. This requires a
single conversion to RF at the output, whereas traditional systems require an ADC
for each antenna element, which is impractical for large antenna systems. As a
result, conventional criteria such as minimum mean square error (MMSE) cannot be
used and analytical Wiener solutions cannot be found. Instead, blind algorithms rely
on correlating the processed signal with some known characteristic of the signal of
interest (such as frequency), and applying a gradient-based algorithm.
Our main advantage is that the same set of 16 wavelengths is used for each
antenna, reducing complexity and increasing scalability. Typically, when signals of
the same optical wavelength are combined, beat noise from coherent summing will
occur and severely degrade the performance. Therefore, without the use of a SM-
MM combiner the architecture would require 16 lasers for each antenna. To re-use
the same wavelengths, the SM-MM combiner is used. The combiner couples sig-
nals from several individual single-mode bers to distinct modes inside a multi-
mode ber. The combiner offers the advantage of phase-insensitivity and coupling
without optical interference. In-depth information and experimental data demon-
strating operation can be found in [47].
The architecture scales by simply adding optical splitters and ampliers up to the
limit imposed by the ASE of the ampliers. The optical weights, which are inte-
grated sixteen per chip and electrically controlled, do not limit the scalability of this
Fig. 15.20 Highly scalable photonic beamformer architecture (N = 16, M = 4)

architecture, nor does the addition of FBGs. A 100 m multimode ber can accept
up to 113 inputs (or antennas).
15.4.2.2 Proof-of-Concept Experimental Results
We experimentally demonstrate an 8-tap lter with tap weights [1.0914 1.0617

1.0715 1.0814 1 0.9683 0.9795 0.0594] and delays incremented by 4 ns. The
dark blue dotted curve in Fig. 15.21 shows the theoretical values and the thick light
green curve shows the measured magnitude response. We are able to achieve a
maximum extinction ratio of *40 dB. The bandwidth and the depth and placement
of the notches depend on the precision of both the delays and weights. Moreover,
our optical system adds no additional noise to the processed signal, as seen in
Fig. 15.21.
The data shows a 4.5 dB loss associated with the architecture. RF systems in the
low GHz range exhibit typical losses around 1 dB. RF systems in the high GHz
range (60 GHz), an area in which photonic systems are expected to excel, exhibit
*35 dB loss with one lter and a *17 dB loss with a four-lter beamformer [48].
RF signal losses in optical systems in general originate from the electrical-to-optical
conversion efciency of the modulators and the modulation depth of the signal
during this conversion. To reduce loss further, removal of the optical dc level of the
processed signal would allow the signal to be optically amplied further, reducing
the system insertion loss.
We experimentally demonstrate an adaptive single-antenna beamformer con-
sisting of 8 fully tunable taps with xed delays of 400 ps and a bandwidth of
2.5 GHz. We employ a modied version of the LMS algorithm called block LMS
with a block size of 256 and stepsize of 512. The adaptive results are preliminary.
The authors only had access to a single 5 Gs/s ADC card, instead of the two
required for the LMS algorithm.
A signal generator is fed into two narrowband tones at 200 MHz and 1 GHz to
the lter. The algorithm is programmed to cancel the 1 GHz interference and pass
the 200 MHz signal. Our workaround involves only digitizing the ltered output
signal. The input signals are replicated digitally and are used to calculate the error
signal. The adaptive algorithm then calculates optical weights that are sent to the
attenuators. We observe an SIR (signal-to-interference ratio) improvement of
*20 dB after *60 iterations at the output of the lter. We are able to drop the
interference to just 5 dB above the noise floor. The results are good and can be
easily improved with a second digitizer. In the future, we plan to apply a blind
adaptive technique as described previously. Limited resolution associated with the
optical attenuators degrades performance.
Future work will include the construction of multiple transversal lters to
complete true spatial beamforming capability.
Fig. 15.21 Measured and

predicted magnitude response
of eight-tap FIR lter
15.4.3 Photonic Beamforming for Physical Layer Security
We introduce a specic application for using the photonic beamformerphysical

layer security in optical backhaul networks. The purpose of this section is to
demonstrate the applicability of the beamformer in commercial systems and to test
the capabilities of the beamformer.
Emerging mobile services, such as mobile banking and desktop-to-mobile
applications, require both increasingly high data rates and high data security. To
address the rising bandwidth demand, strong emphasis is being placed on ber-
optic back- and fronthaul of mobile data [4951]. While optical architectures that
can satisfy the required data rate and latency requirements have matured [5258],
enhancing data security via optical-layer techniques is still in its infancy [5966].
The potential of such physical-layer security is tremendous. For example, while
higher-layer approaches, such as encryption or steganography, can be cracked by
malicious eavesdropping, optical-layer techniques can prevent eavesdroppers from
physically receiving the signal in the rst place. The highest form of security thus
resides in the physical layer, rendering it attractive for data-sensitive mobile
fronthaul applications.
To implement security on the physical-layer, photonic beamforming can be used
in the optical fronthaul network to cancel the signal (i.e. create a signal null) in the
physical direction of an eavesdropper. For the downlink, this can enable sensitive data
to only be received by an intended user within a meter-scale radius, based on the
intended users spatial location and signal carrier frequency. Given the users spatial/
frequency coordinates, a photonic beamformer can create the desired signal null by
effectively manipulating a beampattern that propagates through the optical fronthaul
network. This is achieved by the beamformers array of RF antennas and a series of
adaptive optical FIR lters. The physical formation of the antennas provides spatial
ltering while the FIR lters provide frequency ltering. By changing the weights of
the FIR lters, an adaptive beamformer that can respond to dynamically-changing/
noisy environments can be implemented. Similarly, physical-layer security can also
be provided on the uplink side, with potential application in a shared, multi-operator
environment. In this case, various signals picked up by remote antennas can jointly
propagate through a shared fronthaul infrastructure and be securely separated on a

per-operator basis via adaptive photonic beamforming at a centralized processing
site. It is noted that the proposed optical-layer security approach is also transparent to
the modulation format of the underlying wireless signal.
To the best of our knowledge, we have proposed and demonstrated the rst
experimental adaptive photonic beamforming technique for physical-layer security
in optical fronthaul of mobile trafc, achieving up to 43 dB signal cancellation of an
undesired eavesdropper and 30 dB power budget over 8 km standard single mode
ber (SSMF) fronthaul distance. The proposed approach is attractive for secure
optical fronthaul of advanced mobile services and is transparent across heteroge-
neous mobile technologies and standards.
15.4.3.1 Experimental Setup, Results, and Discussion
The proposed fronthaul architecture using photonic beamforming in the downlink

and the uplink is shown Fig. 15.22. In both cases, the beamformer acts as an optical
spatial and frequency lter that preserves optical signals arising from desired
physical orientations and carrier frequencies, while canceling signals from unde-
sired physical directions and frequencies. In the downlink, mobile data signals from
a centralized site are applied to parallel optical adaptive FIR lters, which optically
modulate and lter the input. The processed optical signals are then transmitted
over parallel SSMF optical fronthaul links to an array of remote antennas. The
optical FIR lters and remote antenna array together form the beamformer. Spe-
cically, by proper assignment of the beamformers lter coefcients, the remote
antennas are turned into a distributed spatial lter that directs the output only to
intended users. The conguration of Fig. 15.22a can thus be used to securely
deliver data to users 1 and 3 while producing a null in the direction of user 2, or vice
versa. In the uplink in Fig. 15.22b, wireless signals from multiple users are rst
detected by the remote antenna array, and optically processed by the FIR lters
such that, after SSMF transmission and photodetection, only the signal from
intended users (e.g. user 1 and user 3) is delivered to the centralized site, while
undesired interferers are suppressed (e.g. user 2).
For downlink experiments, 2.5 and 5 GHz sinusoids are used as the SOI to be
directed by the photonic beamformer to an intended user. After SSMF transmission
and photodetection, the received SOI power is measured at both intended user and
eavesdropper locations, mutually separated by 0.5 m. Downlink beamformer per-
formance is thus measured in terms of the signal strength ratio (SSR) at different
user locations. The beamformer is adjusted for maximum cancellation. Figure 15.23
shows the SSR for the 2.5 and 5 GHz signal. Similarly, for uplink measurements,
2.5 or 5 GHz sinusoids are used as the SOI, while a sinusoid at 900 MHz emulates
an interferer to be suppressed. The two wireless RF signals are generated at different
locations (0.5 m separation), picked up by the antenna array, processed by optical
FIR lters, propagated over parallel SSMFs, combined using a SM-MM coupler,
and detected by a multimode PD. In the uplink, the SSR of the SOI versus the
Fig. 15.22 Proposed architectures a Downlink fronthaul. b Uplink fronthaul
interfering signal is measured at the same location. Figure 15.24 shows the SSR for
the 2.5 and 5 GHz SOI.
Maximum SSR of 33 and 43 dB is achieved in the downlink and uplink,
respectively, over 8 km SSMF fronthaul distance. The 10 dB SSR gain for the
uplink is attributed to the fact that physical-layer signal cancellation physically
occurs in stable passive optical components, while in the downlink, it occurs in the
wireless multi-path channel, resulting in additional undesired interference. Down-
link SSR results in Fig. 15.23 are achieved for intended user versus eavesdropper
separations from 0.1 to 0.5 m, conrming efcacy for small-cell scenarios, with
similar results also measured for signals at 5 GHz, verifying performance for
various frequency bands of prominent wireless standards.
The SSR curves of Figs. 15.23 and 15.24 feature a flat plateau in the 010 dBm
optical power range, followed by a linear decrease versus optical power governed
by PD sensitivity. At SSR = 10 dB, power budgets of 28 and 30 dB are supported
by the downlink and uplink, respectively. Finally, from the experiments it may be
observed that there is no SSR penalty for increasing fronthaul distance compared to
optical back-to-back results. Consequently, so long as minimum PD sensitivity
requirements are met, increasing fronthaul reach will only result in optical FIR
coefcient scaling that is proportional to aggregate received power, without

degrading SSR integrity.
To assess the bit error rate (BER) of the fronthaul link, downlink and uplink
experiments are repeated with a baseband 6 Gbit/s non-return-to-zero (NRZ)
pseudorandom binary sequence (PRBS) signal as the SOI. Results for the downlink
and uplink architectures shown in Fig. 15.25, reveal a maximum transmission
distance of 8 and 18 km SSMF for the downlink and uplink, respectively. The
longer uplink transmission distance is attributed to the higher sensitivity of a
multimode PD compared to a single mode PD, conrming that both SSR and BER
performance are primarily governed by received optical power.
In summary, this investigation experimentally veries adaptive photonic
beamforming for advanced physical-layer security in optical fronthaul of mobile
trafc. Up to 33 and 43 dB signal cancellation to undesired eavesdroppers are
experimentally demonstrated over 8 km SSMF, with 28 and 30 dB power budgets
achieved in downlink and uplink transmission scenarios, respectively. No power
penalty is observed by our system. By enhancing physical-layer security with
modulation format transparency, the proposed approach is promising for secure
optical fronthaul of emerging mobile services.
(a) (b)
Fig. 15.23 Downlink signal strength ratio. a 2.5 GHz SOI. b 5 GHz SOI
(a) (b)
Fig. 15.24 Uplink signal strength ratio. a 2.5 GHz SOI. b 5 GHz SOI
(a) (b)
Fig. 15.25 BER versus optical signal power as a function of fronthaul distance. a Downlink
architecture. b Uplink architecture
15.5 Summary and Concluding Remarks
In the past half-decade, wireless communication for data services has witnessed an
exponential growth, with the advent of smart phones, WIFI, and wireless bluetooth
devices. Mobile apps and computing power has spurred the use of iPhones and
Android-powered devices, and it seems a wireless device occupies every pocket.
People crave the ability to access anything anywhere at anytime, but, unfortunately,
this popularity has caused both signicant network burdens and a wireless spectrum
crunch. As the growth of mobile Internet trafc continues, the need for an easily
implementable method for processing all of this data is essential. Efcient use of the
wireless spectrum also creates its own unique requirements.
The growth of wireless bandwidth mirrors the proliferation of optical network
bandwidth a decade before, and the backhaul of nearly all wireless data networks is
ber-optic. Analog optical signal processing techniques, or microwave photonics,
provides an ideal platform for processing wireless information before it is trans-
ported to data aggregation centers by bers. Photonics offer the advantages not only
of broadband operation, but reduced SWAP, in addition to low transmission loss,
rapid recongurability, and immunity to electromagnetic interference. It is in this
context that we presented recent advances in optical signal processing techniques
for wireless RF signals. This chapter was devoted to the discussion of photonic
architectures for wideband analog signal processing, including RF beamforming
and physical layer security.
Beamforming is a technique that utilizes an array of antennas as a spatial and
frequency lter to manipulate the beampattern for cancelling unwanted signals. The
beamformer uses optical FIR lters to process the signals temporally with thermo-
optic optical attenuators to adaptively and rapidly adjust the beampattern. Mean-
while, the geometric conguration of the antennas allows the signals to be pro-
cessed spatially. Compared to conventional RF beamformers which have limited
narrowband performance due to their reliance on electrical phase shifters, the
wideband nature of optics offers a clear advantage. The architecture also offers the
distinct advantage of scalability to hundreds of antennas, as needed for practical

systems. Such an integrated system could potentially form a frontend of an
emerging all-optical signal processing techniques such as photonic neuromorphic
processing allowing massive network integration [6776].
Interference cancellation plays a crucial part in increasing capacity as it enables
full duplexing. The ability to reject interference, while receiving an SOI at the same
frequency, is referred to as co-channel interference cancellation. Further, a system
needs to be adaptive to react to a highly non-stationary environment. Photonic
implementations of interference cancellation systems exhibit signicant capabilities
beyond what is possible with electronic processing.
An emergence of markets relating to mobile applications, such as mobile
banking, mobile-to-desktop, and mobile-to-cell towers, will require a high degree of
security. Data should only be received by their intended users, and eavesdroppers
should not have access to this data. The ultimate form of security is in the physical
layer, and it is easiest to place a null in the direction of an eavesdropper. Unlike the
software layer, such as encryption or steganography, which can be cracked,
physical layer security prevents eavesdroppers from even detecting the presence of
data transmission in the rst place. This creates a unique application of photonic
beamformers for physical layer security in optical fronthaul of mobile trafc.
In summary, this chapter discussed the basic theory of beamforming and the
unique challenges of broadband interference cancellation. Next, we presented the
requirements and challenges for building optical FIR lters, and reviewed several
innovative MPF designs including optical tunable delays and weighting schemes.
We also reviewed the state of the art optical beamformers recently proposed and
experimentally demonstrated in literature. Furthermore, we detailed a highly scal-
able photonic beamforming architecture designed for a particularly non-stationary,
interfering environment. Finally, we introduced a specic application for the pho-
tonic beamformer namely, physical layer security in optical backhaul networks.
We hope that this chapter will contribute to stimulating research in the cross-
disciplinary areas of photonics and optics, microwave engineering, wireless com-
munications, and signal processing, from fundamental principles to practical
applications.
References
1. Mobile broadband explosion, White Paper, Rysavy Research/4G Americas, Aug 2012
2. H. Bauer, F. Grawert, S. Schink. Semiconductors for wireless communications: growth engine
of the industry, McKinsey & Company on Semiconductors, Technical Report (2012)
3. R.A. Monzingo, R.L. Haupt, T.W. Miller, Introduction to Adaptive Antennas (SciTech
Publishing, Raleigh, 2011)
4. J. Yao, Microwave photonics. J. Lightwave Technol. 27(3), 314335 (2009)
5. J. Capmany, B. Ortega, D. Pastor, A tutorial on microwave photonic lters. J. Lightwave
Technol. 24(1), 201229 (2006)
6. J. Chang, M.P. Fok, J. Meister, P.R. Prucnal, A single source microwave photonic lter using a
novel single-mode ber to multimode ber coupling technique. Opt. Express 21(5), 55855593
(2013)
7. J. Chang, Y. Deng, M.P. Fok, J. Meister, P.R. Prucnal, A photonic microwave FIR lter using
a spectrally sliced supercontinuum source. Appl. Opt. 51(19), 42654268 (2012)
8. J. Chang, P.R. Prucnal, A novel analog photonic method for broadband multipath interference
cancellation. IEEE Microwave Wirel. Compon. Lett. 23(7), 377379 (2013)
9. R. Haupt, Antenna Arrays: A Computational Approach (Wiley, Hoboken, 2010)
10. B. Widrow, P.E. Mantey, L.J. Grifths, B.B. Goode, Adaptive antenna systems. Proc. IEEE
55, 21432159 (1967)
11. B. Widrow, S.D. Stearns, Adaptive Signal Processing (Prentice-Hall, Englewood Cliffs, 1985)
12. L.V. Blake, M.W. Long, Antennas: Fundamentals, Design, Measurement (SciTech
Publishing, Daryaganj, 2009)
13. S.O. Haykin, Adaptive Filter Theory (Prentice-Hall, Englewood Cliffs, 2002)
14. D.B. Hunter, R.A. Minasian, Microwave optical lters using in-ber Bragg grating arrays.
IEEE Microwave Guided Waveguide Lett. 6(2), 103105 (1996)
15. X. Yi, R.A. Minasian, New spectrum-sliced microwave photonic lter for high- frequency
signal processing. IEEE Photonics Technol. Lett. 21(4), 230232 (2009)
16. T. Chen, X. Yi, T. Huang, R.A. Minasian, Spectrum sliced microwave photonic signal
processor with tunablity and recongurablity, in Proceeding of the OptoElectronics and
Communications Conference (OECC), pp. 12, July 2009
17. J. Capmany, J. Mora, D. Pastor, B. Ortega, High-quality online-recongurable microwave
photonic transversal lter with positive and negative coefcients. IEEE Photonic Technol.
Lett. 17(12), 27302732 (2005)
18. X. Yi, T.X.H. Huang, R.A. Minasian, Microwave photonic lter with tunability,
recongurability and bipolar taps. Electron. Lett. 45(16), 840841 (2009)
19. M.D. Manzanedo, J. Mora, J. Capmany, Continuously tunable microwave photonic lter with
negative coefcients using cross-phase modulation in an SOA-MZ interferometer. IEEE
20. E.H.W. Chan, R.A. Minasian, Coherence-free equivalent negative tap microwave photonic
notch lter based on delayed self-wavelength conversion. IEEE Trans. Microw. Theory Tech.
5(11), 31993205 (2010)
21. J. Mora, A. Martinez, M.D. Manzanedo, J. Capmany, B. Ortega, D. Pastor, Microwave
photonic lters with arbitrary positive and negative coefcients using multiple phase inversion
in SOA based XGM wavelength converter. Electron. Lett. 41(16), 5354 (2005)
22. X. Yi, R.A. Minasian, Novel multitap, flat-top microwave photonic lter based on sinusoidal
group delay gratings. J. Lightwave Technol. 26(15), 25782583 (2008)
23. C.-K. Oh, T.-Y. Kim, C.-S. Park, Recongurable photonic microwave band-pass lter with
negative coefcients based on polarisation modulation. Electron. Lett. 43(11), 639641 (2007)
24. S. Mansoori, A. Mitchell, K. Ghorbani, Photonic recongurable microwave lter with
negative coefcients. Electron. Lett. 40(9), 541543 (2004)
25. B. Vidal, J.L. Corral, and J. Marti, Multi-tap all-optical microwave lter with negative
coefcients based on multiple optical carriers and dispersive media, in Proceeding of the
International Topical Meeting on Microwave Photonics (MWP), pp. 201204, Oct 2005
26. D. Pastor, J. Capmany, B. Ortega, A. Martinez, L. Pierno, M. Varasi, Recongurable RF
photonic lter with negative coefcients and flat-top resonances using phase inversion in a
newly designed 2 1 integrated Mach-Zehnder modulator. IEEE Photonics Technol. Lett. 16
(9), 21262128 (2004)
27. D.B. Hunter, Incoherent bipolar tap microwave photonic lter based on balanced bridge
electro-optic modulator. Electron. Lett. 40(12), 856858 (2004)
28. L.A. Bui, K.S. Dayaratne, A. Mitchell, Discrete time microwave photonic transversal lter, in
Proceeding of the International Topical Meeting on Microwave Photonics (MWP), pp. 13,
Oct 2009
29. X. Yi, R.A. Minasian, Dispersion induced RF distortion of spectrum-sliced microwave-

photonic lters. IEEE Trans. Microw. Theory Tech. 54(2), 880886 (2006)
30. J. Mora, S. Sales, M.D. Manzanedo, R. Garcia-Olcina, J. Capmany, B. Ortega, D. Pastor,
Continuous tuning of photonic transversal lter based on the modication of tapped weights.
IEEE Photonic Technol. Lett. 18(15), 15941596 (2006)
31. J. Capmany, D. Pastor, B. Ortega, New and flexible ber-optic delay-line lters using chirped
Bragg gratings and laser arrays. IEEE Trans. Microw. Theory Tech. 47(7), 13211326 (1999)
32. W. Liu and S. Weiss. Wideband Beamforming: Concepts and Techniques (Wiley, Chichester,
2010)
33. B.D. Van Veen, K.M. Buckley, Beamforming: a versatile approach to spatial ltering. IEEE
ASSP Mag. 5(2), 424 (1988)
34. H. Zmuda, R.A. Soref, P. Payson, S. Johns, E.N. Toughlian, Photonic beamformer for phased
array antennas using a ber grating prism, IEEE Photonics Technol. Lett. 9(2), 241243 (Feb
1997)
35. M.Y. Frankel, P.J. Matthews, R.D. Esman, Two-dimensional ber-optic control of a true time-
steered array transmitter. IEEE Trans. Microw. Theory Tech. 44(12), 26962702 (1996)
36. L. Yaron, R. Rotman, S. Zach, M. Tur, Photonic beamformer receiver with multiple beam
capabilities. IEEE Photonics Technol. Lett. 22(23), 17231725 (2010)
37. L. Jofre, C. Stoltidou, S. Blanch, T. Mengual, B. Vidal, J. Marti, I. Mckenzie, J.M. del Cura,
Optically beamformed wideband array performance. IEEE Trans. Antennas Propag. 56(6),
15941604 (2008)
38. H. Subbaraman, M.Y. Chen, R.T. Chen, Simultaneous dual RF beam reception of an X-band
phased array antenna utilizing highly dispersive photonic crystal ber based true-time-delay,
in Proceedings of the Asia Optical Fiber Communication and Optoelectronic Exposition and
Conference (AOE), pp 13, 2008, paper SaJ2
39. W. Xue, J. Mork, Microwave photonic true time delay based on cross gain modulation in
semiconductor optical ampliers, in Proceeding of the OptoElectronics and Communications
Conference (OECC), pp. 202203, July 2010
40. M.A. Piqueras, G. Grosskopf, B. Vidal, J. Herrera, J.M. Martinez, P. Sanchis, V. Polo, J.L.
Corral, A. Marceaux, J. Galiere, J. Lopez, A. Enard, J.-L. Valard, O. Parillaud, E. Estebe, N.
Vodjdani, M.-S. Choi, J.H. den Besten, F.M. Soares, M.K. Smit, J. Marti, Optically
beamformed beam-switched adaptive antennas for xed and mobile broad-band wireless
access networks. IEEE Trans. Microw. Theory Tech. 54(2), 887899 (2006)
41. Y. Liu, G. Burnham, G. Jin, J. Zhao, Wideband multi-beam photonics-based RF beamformer,
in Proceedings of the IEEE International Symposium on Phased Array Systems and
Technology (ARRAY), pp. 581585, Oct 2010
42. B. Juswardy, F. Xiao, K.E. Alameh, Opto-VLSI-based RF beamformer for space division
multiple access network, in Proceedings of the High-Capacity Optical Networks and Enabling
Technologies (HONET), pp. 222226, Dec 2010
43. H. Zmuda, E. N. Toughlian, Broadband nulling for conformal phased array antennas using
photonic processing, in Proceedings of the International Topical Meeting on Microwave
Photonics (MWP), pp. 1719, Sept 2000
44. D. Marpaung, L. Zhuang, M. Burla, C. Roeloffzen, B. Noharet, Q. Wang, W.P. Beeker, A.
Leinse, R. Heideman, Photonic integration and components development for a Ku-band
phased array antenna system, in Proceedings of the International Topical Meeting on
Microwave Photonics (MWP), pp. 458461, Oct 2011
45. L. Zhuang, D. Marpaung, M. Burla, C. Roeloffzen, W. Beeker, A. Leinse, P. van Dijk, Low-loss
and programmable integrated photonic beamformer for electronically-steered broadband phased
array antennas, in Proceedings of the IEEE Photonics Conference (PHO), pp. 137138, 2011
46. J. Chang, M.P. Fok, R.M. Corey, J. Meister, P.R. Prucnal, Highly scalable adaptive photonic
beamformer using a single mode to multimode optical combiner. IEEE Microwave Wirel.
Compon. Lett. 23(10), 563565 (2013)
47. M.P. Fok, Y. Deng, K. Kravtsov, P.R. Prucnal, Signal beating elimination using single-mode
ber to multimode ber coupling. Opt. Lett. 36(23), 45784580 (2011)
48. S. Lin, K.B. Ng, K.M. Luk, S.S.Wong, A. Poon, A 60 GHz digitally controlled RF
beamforming array in 65 nm CMOS with off-chip antennas, in Proceedings of the Radio
Frequency Integrated Circuits Symposium (RFIC), pp. 14, June 2011
49. D. Bojic, E. Sasaki, N. Cvijetic, T. Wang, J. Kuno, J. Lessmann, S. Schmid, H. Ishii, S.
Nakamura, Advanced wireless and optical technologies for small-cell mobile backhaul with
dynamic software-dened management. IEEE Commun. Mag. 51(9), 8693 (2013)
50. N. Cvijetic, A. Tanaka, Y.-K. Huang, M. Cvijetic, E. Ip, Y. Shao, T. Wang, 4-G mobile
backhaul over OFDMA/TDMA-PON to 200 cell sites per ber with 10 Gb/s upstream burst-
mode operation enabling 1 ms transmission latency, in Proceedings of the Optical Fiber
Communication Conference and Exposition (OFC/NFOEC), 2012, paper PDP5B.7
51. C. Liu, K. Sundaresan, M. Jiang, S. Rangarajan, G.-K. Chang, The case for re-congurable
backhaul in cloud-ran based small cell networks, in Proceedings of the IEEE INFOCOM,
pp. 11241132, 2013
52. B.J. Shastri, D.V. Plant, Scaling technologies for terabit ber optic transmission systems, in
Proceedings of the SPIE, vol. 7942, (San Francisco, CA, Feb 2011), paper 794206
53. B.J. Shastri, P.R. Prucnal, D.V. Plant, 20-GSample/s (10 GHz 2 clocks) burst-mode CDR
based on injection locking and space sampling for multiaccess networks. IEEE Photonics J. 4
(5), 17831793 (2012)
54. B.J. Shastri, D.V. Plant, Truly modular burst-mode CDR with instantaneous phase acquisition
for multiaccess networks. IEEE Photonics Technol. Lett. 24(2), 134136 (2012)
55. B.J. Shastri, D.V. Plant, 5/10-Gb/s burst-mode clock and data recovery based on semiblind
oversampling for PONs: theoretical and experimental. IEEE J. Sel. Top. Quantum Electron. 16
(5), 12981320 (2010)
56. B.J. Shastri, Y.B. MSallem, N. Zicha, L.A. Rusch, S. LaRochelle, D.V. Plant, Experimental
study of burst-mode reception in a 1300 km deployed ber link. J. Opt. Commun. Networking
2(1), 19 (2010)
57. B.J. Shastri, Z.A. El-Sahn, M. Zeng, N. Kheder, L.A. Rusch, D.V. Plant, A standalone burst-
mode receiver with clock and data recovery, clock phase alignment, and RS(255, 239) codes
for SAC-OCDMA applications. IEEE Photonics Technol. Lett. 20(5), 363365 (2008)
58. Z.A. El-Sahn, B.J. Shastri, M. Zeng, N. Kheder, D.V. Plant, L.A. Rusch, Experimental
demonstration of a SAC-OCDMA PON with burst-mode reception: local versus centralized
sources. J. Lightwave Technol. 26(10), 11921203 (2008)
59. B. Wu, B.J. Shastri, P.R. Prucnal, Secure communication in ber-optic networks, in Emerging
Trends in Information and Communication Technologies Security, ed. by B. Akhgar, H. Arabnia
(Elsevier, Waltham, 2013), ch. 11, pp. 173183
60. B. Wu, Z. Wang, Y. Tian, M.P. Fok, B.J. Shastri, D.R. Kanoff, P.R. Prucnal, Optical
steganography based on amplied spontaneous emission noise. Opt. Express 21(2), 20652071
(2013)
61. B. Wu, Z. Wang, B.J. Shastri, M.P. Chang, N.A. Frost, P.R. Prucnal, Temporal phase mask
encrypted optical steganography carried by amplied spontaneous emission noise. Opt.
Express 22(1), 954961 (2014)
62. B. Wu, M.P. Chang, B.J. Shastri, Z. Wang, P.R. Prucnal, Analog noise protected optical
encryption with two-dimensional key space. Opt. Express 22(12), 1456814574 (2014)
63. B. Wu, B.J. Shastri, P.R. Prucnal, System performance measurement and analysis of optical
steganography based on amplier noise. IEEE Photonics Technol. Lett. 26(19), 19201923
(2014)
64. B. Wu, M.P. Chang, Z. Wang, B.J. Shastri, P.R. Prucnal, Optical encryption based on
cancellation of analog noise, in Proceedings Conference on Lasers and Electro-Optics
(CLEO), San Jose, CA, June 2014, paper AW3P.5
65. B. Wu, Z. Wang, B.J. Shastri, Y. Tian, P.R. Prucnal, Phase mask encrypted optical
steganography based on amplied spontaneous emission noise, in Proceedings of the IEEE
Photonics Conference (IPC), Seattle, Sept 2013, paper MG3.3, pp. 137138
66. B. Wu, Z. Wang, B.J. Shastri, Y. Tian, P.R. Prucnal, Two dimensional encrypted optical
steganography based on amplied spontaneous emission noise, in Proceedings of the
Conference on Lasers and Electro-Optics (CLEO), San Jose, June 2013, paper AF1H.5
67. A. N. Tait, M.A. Nahmias, Y. Tian, B.J. Shastri, P.R. Prucnal, Photonic neuromorphic signal
processing and computing, in Nanophotonic Information Physics, ser. Nano-Optics and
Nanophotonics, ed. by M. Naruse, Springer, Heidelberg, 2014, pp. 183222. http://dx.doi.org/
10.1007/978-3-642-40224-1_8
68. B.J. Shastri, A.N. Tait, M.A. Nahmias, P.R. Prucnal, Photonic spike processing: ultrafast laser
neurons and an integrated photonic network. IEEE Photonics Soc. Newsl. 28(3), 411 (2014)
69. M.A. Nahmias, B.J. Shastri, A.N. Tait, P.R. Prucnal, A leaky integrate-and-re laser neuron
for ultrafast cognitive computing, IEEE J. Sel. Top. Quantum Electron. 19(5), 1800212 (Sept
Oct 2013)
70. A.N. Tait, M.A. Nahmias, B.J. Shastri, P.R. Prucnal, Broadcast and weight: an integrated
network for scalable photonic spike processing. J. Lightwave Technol. 32(21), 34273439
(2014)
71. B.J. Shastri, M.A. Nahmias, A.N. Tait, P.R. Prucnal, Simulations of a graphene excitable laser
for spike processing. Opt. Quant. Electron. 46(10), 13531358 (2014)
72. A.N. Tait, B.J. Shastri, M.A. Nahmias, M.P. Fok, P.R. Prucnal, The DREAM: an integrated
photonic thresholder. J. Lightwave Technol. 31(8), 12631272 (2013)
73. A.N. Tait, M.A. Nahmias, B.J. Shastri, P.R. Prucnal, Broadcast-and-weight interconnects for
integrated distributed processing systems, in Proceedings of the IEEE Optical Interconnects
Conference (OI), Coronado Bay, CA, May 2014, paper WA3, pp. 108109
74. B.J. Shastri, M.A. Nahmias, A.N. Tait, Y. Tian, B. Wu, P.R. Prucnal, Graphene excitable laser
for photonic spike processing, IEEE Photonics Conf. (IPC), Sept 2013, pp. 12. http://dx.doi.
org/10.1109/IPCon.2013.6656424
75. M.A. Nahmias, A.N. Tait, B.J. Shastri, P.R. Prucnal, An evanescent hybrid silicon laser
neuron, in Proceedings of the IEEE Photonics Conference (IPC), Seattle, WA, Sept 2013,
paper ME3.4, pp. 9394
76. B.J. Shastri, M.A. Nahmias, A.N. Tait, Y. Tian, M.P. Fok, M.P. Chang, B. Wu, P.R. Prucnal,
Exploring excitability in graphene for spike processing networks, in Proceedings of the IEEE
Numerical Simulation of Optoelectronic Devices (NUSOD), Vancouver, Canada, Aug 2013,
paper TuC5, pp. 8384
Index
A Bandpass ltering, 284

Adaptive array, 470, 472, 479 BB84, 407
Adaptive single-antenna beamformer, 493 Beamformer, 479, 493
ADC performance, 298 Beamformer-on-chip, 489
Adder, 295, 296 Beamforming, 474, 479, 485487, 489, 490,
Adjustable Bit-Rate Optical DFT, 179 493, 497499
Algorithm, 406 Beam steering, 475
All-optical DFT, 236 Berlekamp-Massey algorithm, 302
All-optical fan-out, 171 Bessel function, 311, 314
All-optical OFDM demultiplexing, 280 Bi-photon, 364, 365, 371, 375, 385389, 393
All-optical switching, 4, 5 Binary phase shift keying, 136
Amorphous silicon, 11, 1315, 35, 38, 42, 43, Biphoton correlation, 430, 442
45 Biphoton wavepacket, 429, 437, 441
A-Si:H nanowire, 43 Biphotons, 430434, 436, 441
Amplitude discrimination, 301 Birefringence, 7274, 76
Amplitude-modulated signals, 163 Bismuth-oxide-based (Bi2O3) ber, 164
Amplitude programmable lter, 176 Bit error rate (BER), 497
Amplitude regeneration, 131, 132, 134, 137, BPSK signal, 138
139 Bragg scattering, 189, 403
Analog-to-digital conversion (ADC), 208, 243 Brillouin gain, 453, 454, 458, 459
Analogue to digital converter, 243, 294, 298, Broadband amplier, 294
484
Analog lter, 241 C
Antenna array, 470, 472, 474476, 481, 488, Capacity, 292, 302, 303
489, 495 Carrier phase estimation (CPE), 298
Antenna elements, 474 Carrier recovery, 20, 21
Arbitrary optical waveforms, 173 Carrier suppressed, 309, 315
Arbitrary waveform generation, 270, 425 Cascaded delay interferometer, 229
Arbitrary waveform generator (AWG), 243 Cascaded delay interferometer FFT, 230
Array of phase modulators, 311 Cascaded FWM, 57
Arrayed-waveguide (AWG) demux, 486 Cascaded mixing, 202
Arrayed waveguide grating router (AWGR), Cascaded nonlinear wave mixing, 172
234 Cascaded parametric mixing, 202
Available spectrum, 292 Cavities, 411
AWG multiplexer, 491 Chalcogenide (ChG), 164, 453455, 459, 460,
464
B Chalcogenide chip, 145
Bandlimited Channels, 332 Chalcogenide glass, 11, 16, 37
Bandpass lter, 282, 455 Chalcogenide nanowire, 9, 10

Springer Series in Optical Sciences 194, DOI 10.1007/978-3-319-14992-9
506 Index
Chalcogenide planar waveguide, 168 D

Channel impulse response (CIR), 299 2-D correlator, 177
Channelizer, 209 DAC performance, 296
Chirped pulse, 205 DAC power consumption, 296
Chirped spectrum, 267 Data-sensitive mobile fronthaul, 494
Chromatic dispersion, 191, 192, 194, 201, 202, Decode-and-Forward model, 346
205, 210 Degenerate FWM, 162
Chromatic dispersion engineering, 191, 192 Degenerate PSA, 135, 139, 143
Circular beamformer, 480 Delay lter taps, 482
Classical analogues, 441 Delay interferometer, 179, 229, 237
Clock recovery, 252 Demultiplexing, 306, 309
CMOS compatibility, 36, 38, 45 Detection penalty, 23, 26
CMOS DSP chip, 295 DFT processor, 226
CMOS energy consumption, 305 Difference Frequency Generation, 160
CMOS gate efciency, 305 Differentially encoded, 169
CMOS node, 298 Digital-to-analog, 243
Coherent Carrier Generation, 205 Digital-to-analog conversion (DAC), 243, 247
Coherent continua, 112, 122 Digital back propagation, 328
Coherent detection, 332, 343 Digital coherent link, 303
Coherent eld capture, 209 Digital least mean square, 477
Coherent receiver, 228, 237 Digital Nyquist pulse transmitter, 244
Coherent transponder, 293, 294 Digital signal processing (DSP), 293, 328, 476
Coincidence-to-accidental ratio (CAR), 367, Digital signal processor (DSP), 243
368370, 381384, 395397 Digital to analogue converter, 296
Comb, 311, 313315 Digital to analogue converter resolution, 293
Comb flatness, 314 Direct detection, 304
Comb generation, 311, 315, 316 Direction of arrival, 478
Comb generator, 313 Discrete Fourier transform, 173, 209
Comb line, 316 Discrete Fourier Transform, 178, 226, 227
Comb source, 315 Discrete Fourier transformation, 280
Communication capacity, 291 Dispersion, 434437, 439, 444
Constellation, 292, 293, 296, 298, 300, 341 Dispersion cancellation, 430, 435, 436, 438,
Constellation diagram, 137, 170 441, 444, 446
Conversion efciency, 6, 8, 14, 15, 21, 23, 25, Dispersion compensation bre, 276
41, 47, 48, 50 Dispersion control, 192
Copier-PSA, 144 Dispersion engineering, 38, 39, 50, 53, 55, 56,
Correlation, 173, 359, 365, 366, 371, 386 202, 205, 210, 211
Correlation function, 367, 391, 394 Dispersion-flattened highly nonlinear bre
Counter-Collision Method, 196 (DF-HNLF), 286
Coupled-mode theory, 85, 86, 96 Dispersion fluctuations, 193, 194, 197200,
Coupled resonator optical waveguide, 381 210
Coupled waveguides, 94, 96 Dispersion fluctuations rectication, 193
Coupler, 8694, 9699 Dispersion stability, 197, 199, 200, 210
Cross-phase modulation, 72, 90, 262, 452, 460 Dispersive Pulse Walk-off, 195
Cross-spectral density, 430 Distributed feedback (DFB) diode laser, 205
Cross gain modulation, 18, 165 Downlink, 495
Cross phase modulation (XPM), 18, 163 DPSK signals, 142
Crosstalk, 22 DQPSK modulation, 282
Crystalline silicon, 9, 11, 13 Drive signal, 316
Cyclic extension, 223 DSP algorithm, 222
Index 507
DSP energy consumption, 316 Figures of merit (FOM), 2, 306

Dual-layer core structure, 198 Filter, 296, 297, 299, 300, 304
Dual polarisation IQ modulator, 294 Filter impulse response, 297
Dynamic equaliser, 299, 300 Finite-difference time-domain, 84
Finite impulse response (FIR) lters, 173, 244,
E 296, 297, 299, 300, 474, 476, 479
E, 311 Flat comb spectrum, 311
Eavesdropper, 497 Flatness, 311, 313
Eckert, 408, 410 Flat-top pulses, 285
Effective number of bits, 305 Flexible grid network, 280
Electro-absorption, 316 Forward error correction (FEC), 276, 297, 302
Electro-absorption modulator, 306, 307, 481 Four-photon mixing, 187
Electro-optic effect, 426 Four-wave mixing (FWM), 1, 8, 2123, 26, 61,
Electro-optic modulation, 423, 425, 431, 434, 63, 72, 159, 162, 168, 263, 269272
435, 438, 439 Four-wave-mixing Bragg scattering, 402406
Electro-optic modulators, 306, 316, 481 Fourier transform, 233, 239, 249, 250, 268
Electro-optic phase modulation, 269, 270 Fourier-transform pulse shaping, 425, 426,
Electromagnetic interference, 450 431, 446
Electronic ADC, 298 Fourth-order dispersion, 111
Electronic Nyquist processing, 241 FPM processes, 190
Electronic OFDM, 235 Franson interferometer, 434
Electronic pulse shaping, 218 Fraunhofer diffraction, 266
Encoding, 359, 360, 367 Free-carrier absorption, 463
Encryption, 494, 499 Free carriers, 36
Energy-Efcient Optical Signal Processing, Free Space Fourier Transform, 233
262 Free spectral range, 53
Energy-efcient switching, 262 Frequency comb, 46, 50, 5357, 60, 65, 205,
Energy consumption, 291293, 302306, 316 245, 249252, 255, 256
Energy Consumption of DSP, 295 Frequency comb synthesis, 241, 245, 254
Entanglement, 359, 371373, 376, 389, 402, Frequency-dependent group delays, 437
405, 407411 Frequency division multiplexing, 158
Equaliser, 300 Frequency-to-time conversion, 267, 270, 276
Equalization, 173 Frequency-to-time mapping, 269
Equalized spectral shape, 205 Frequency-Tunable pulses, 117
Error correction, 297 Fronthaul, 496
Error vector magnitude, 170, 179, 237
Ethernet data packet, 285 G
Ethernet packet, 285, 286 Gates, 357, 360, 361
External cavity laser, 241 Gaussian, 331
Extreme Events, 119 Gaussian noise, 331, 333, 334, 338340
Gaussian stochastic process, 338
F Generation, 311
Factorability, 387, 389 Geometrical fluctuations, 200
Fast Fourier transform (FFT), 179, 180, 227 Glauber correlation function, 428, 429
FBG array, 484, 486, 488, 491 Graphene, 90100
FBG-based delay, 482 Graphene coupler, 90, 93
FBG delay line, 485 Grid manipulator, 279
FFT processor, 227 Guard bands, 224
Fiber amplier, 112 Guided acoustic wave Brillouin scattering
Fiber-Bragg gratings, 482 (GAWBS), 135
Fiber delay line, 482
Fiber laser, 114 H
Fibre response function, 268 Hanbury Brown and Twiss, 423
Field programmable gate arrays, 243 Heralded single photon, 431, 435
508 Index
Hexadecimal 16-PSK Addition, 170 Link power budget, 292

High-order dispersion cancellation, 436438, Link Power Consumption, 302
443 Link transfer function, 300
Highly-nonlinear ber (HNLF), 134, 158, 164, Liquid crystal (LC) cells, 489
186, 193, 262 Liquid crystal on silicon, 172, 177, 485
Hybrid integration, 464 Liquid crystal spatial modulator, 246
Hydex device, 48, 50 Local oscillator, 222
Hydex glass, 35, 40, 49, 65 Logic operation, 162, 166, 167
Loss of signal, 473
I Low density parity check codes (LDPC), 302
IDFT transmitter, 226
Immunity to electromagnetic interference, 469, M
472, 498 Matched lter, 297
Immunity to EMI, 480 Mach Zehnder, 309, 311, 316
In OTDM-WDM conversion, 191 Mach Zehnder modulator (MZM), 177, 285,
Indirect gap material, 3, 4 458, 481, 306, 308, 315
Indium gallium arsenide, 164 Mamyshev regenerator, 131
Indium phosphyde, 164 Manakov equation, 345
Information, 325 Memory, 299
Information capacity, 331, 345 Michelson interferometer, 2
Inhomogeneous Dispersion Engineering, 201, Microresonator, 38, 53, 57, 60
202 Micro-ring resonator, 46, 49, 53
Insertion loss, 292, 293, 306, 314316 Microwave photonic lter, 472, 480
Instabilities, 120, 121 Microwave photonics, 452
Instantaneous frequency, 438 Mid-link device, 291
Integrated photonic beamformer, 489 Millers rule, 39, 42
Integration, 293 Minimum mean square error (MMSE), 492
Intensity-modulated direct-detection, 335 Minimum on-off switching voltage, 308
Interchannel interference, 276 Mobile applications, 470
Inter-symbol interference, 223, 276 Mobile broadband, 470
Intermodulation, 316 Modelocked ber laser, 252
Intersymbol interference, 238 Modelocked laser, 60, 237
Inverse Discrete Fourier Transform (IDFT), Modulation, 305, 306
226 Modulation cancellation, 439
Inverse fast Fourier transform, 178 Modulation format, 22, 304
IQ-modulator, 219, 225, 306, 311, 315 Modulation format transparent PSA, 135, 136,
144
J Modulation instability, 50, 52, 57, 112, 189,
Jacobi elliptic function, 75 204
Joint spectral intensity (JSI), 389391 Modulator, 294, 303, 307309, 311, 313316
Modulator cascade, 311, 314316
K Modulator insertion loss, 314
Kerr effect, 11, 21, 22, 325 Momentum conservation, 188
Kerr nonlinearity, 4, 13, 111, 251, 461 MPSK phase regenerator, 146
Kramers-Kronig relation, 2, 18 MPSK regenerator, 141
Kuznetsov-Ma soliton, 122 MPSK signal, 132, 137, 141, 143, 144
M-symbol quadrature amplitude modulation,
L 158
Laser phase noise, 301 Multicasting, 180
LCoS technology, 176 Multicasting device, 208
Lens, 233 Multichannel, 22
Link, 302304, 316 Multi-channel correlator, 176
Link conguration, 302 Multi-level modulated signal, 170
Link gain, 451 Multipath effects, 473
Index 509
Multiple-amplitude-level signals, 150 O

Multiplier, 295 O/E conversion, 228
Multiplexing, 357, 395398, 407 OFDM, 218, 219, 296
Multi-stage mixer, 207, 209, 210 OFDM-to-DWDM Conversion, 280
Multi-tone, 316 OFDM data stream, 226
MWP lter, 456, 457, 461, 462 OFT-OTDM system, 265
MWP link, 450452 OFDM receiver, 224, 236
MWP system, 450 OFDM signal, 223, 224, 226, 229, 232,
MZI, 7779 234237
OFDM subcarriers, 179
N OFDM superchannel, 283, 284
Narrow linewidth laser, 293 OFDM symbol, 226
Network, 494 OFDM transmitter, 225
Next-generation mobile networks, 470 On-demand, 364, 394, 396
NLDC, 72, 7476, 82, 83, 90 One-pump parametric mixing, 187, 188
Noise gure, 451 OOK signal, 130, 132, 134, 150
Noise-Inhibited Parametric Mixing, 201 OPO, 49, 50
Nondegenerate FWM, 273 Optical, 205
Nonlinear amplifying loop mirror, 146, 149 Optical 1D Correlation, 173
Nonlinear Channel, 335 Optical backhaul, 471
Nonlinear coupler, 82 Optical beamformer, 485
Nonlinear coupling, 72 Optical beamforming, 478
Nonlinear distortions, 451 Optical circulator, 491
Nonlinear ber channel, 333 Optical code-division multiple-access, 440
Nonlinear gure of merit, 201 Optical comb, 237, 310
Nonlinear optical loop mirror (NOLM), 75, 78, Optical comb generation, 306, 310
79, 8, 164 Optical correlation, 175
Nonlinear phase noise, 130, 336, 337 Optical delay line, 481
Nonlinear phase shift, 141 Optical delay line lter, 175
Nonlinear polarisation rotation (NPR), 18, 21, Optical FFT, 179
22 Optical FFT processor, 228
Nonlinear Schrdinger (NLS) equation, 106, Optical lter, 315
333 Optical ltering, 241, 246
Nonlinear SNR, 327 Optical Fourier transformation, 263, 264, 270,
Nonlinear susceptibility, 159, 165 272, 286
Notch lter, 455458, 463, 464 Optical frequency comb, 177, 248
NRZ modulation, 244 Optical fronthaul, 494, 495, 497, 499
N-tap correlator, 174 Optical interleaver, 246
Nyquist, 218 Optical modulator, 294
Nyquist bandwidth, 244 Optical Nyquist pulse, 247
Nyquist channel, 274, 275 Optical OFDM demodulation, 179
Nyquist criterion, 238, 239, 251 Optical OFDM receiver, 228
Nyquist lter, 297 Optical parametric mixing, 187
Nyquist limit, 220 Optical phase conjugation, 328, 344
Nyquist multiplexing, 219 Optical pulse shaping, 241, 244, 252
Nyquist OTDM, 252, 253, 281 Optical regeneration, 151, 328, 346, 347
Nyquist OTDM-WDM, 255 Optical resonator, 251
Nyquist pulse, 239241, 243, 246, 252, 255 Optical-signal-to-noise-ratio (OSNR), 138, 206
Nyquist pulse shaping, 242 Optical shock-wave, 204
Nyquist pulse train, 246, 248, 249 Optical soliton, 83
Nyquist signal, 239 Optical time-domain reflectometry (OTDR),
Nyquist sinc-pulse, 244 194
Nyquist WDM, 274, 296 Optical time division multiplex, 133, 191
510 Index
Optical tunable correlator, 176 Phase shift steering, 477

Optical tunable delay, 480 Phase shifter, 459461
Orthogonal frequency division multiplexed Phase squeezing, 136
(OFDM), 223 Phased lock, 249
Orthogonal frequency division multiplexing, Phasor, 312
274, 280 Photon pairs, 364, 385, 409, 411
Orthogonal optical time-division multiplexing Photonic Adaptive Beamformer, 485
(OTDM), 247249 Photonic beamformer, 495
Orthogonality condition, 248 Photonic beamforming, 474, 485, 494, 495
Orthonormality, 220 Photonic cancellation system, 471
OTDM channel, 286 Photonic crystal, 2, 12, 17, 71, 72, 100, 165
OTDM-to-DWDM Conversion, 270 Photonic crystal waveguides, 381, 395
OTDM-to-WDM conversion, 264, 270, 271 Photonic phased array antenna, 480
OTDM-to-WDM conversion using, 270 Photonic quantum networks, 356
OTDM demultiplexing, 166 Photonic signal channelization, 209
OTDM point-to-point system, 272 Physical layer security, 469, 474, 485, 494,
498, 499
P Physical-layer signal cancellation, 496
Parabolic waveforms, 110 Pilot tone, 304
Parallel demultiplexer, 265 PM-IQ modulator, 294
Parametric frequency comb, 206 Poincar sphere, 73, 74
Parametric gain, 50, 57, 188, 200 Point to point link, 302
Parametric mixer, 186192, 201, 205, 207, Point-to-point WDM system, 265
208, 210, 211 Poissonian, 392
Parametric mixing, 186 Poisson summation formula, 250
Parametric multicaster, 209 Polarization, 7277, 84, 90, 92, 159162, 168,
Parametric oscillator, 51 360363, 367, 372, 374378, 396,
Passive demultiplexing, 265 407410
Path, 359, 360, 392 Polarization diversity, 140
Peak-to-rst zero-crossing separation, 247 Polarization modulation, 482
Peregrine soliton, 121, 122 Power consumption, 262264, 271, 280, 284,
Periodically-poled-lithium-niobate (PPLN), 1, 292298, 300, 302, 304, 305, 314, 315
4, 6, 7, 131, 142, 171, 396 Power law-Wiener lter, 300
Periodically Poled Lithium Niobate PPLN waveguide, 145, 150
Waveguide, 165 Programmable optical lter, 314
Phase-locked signal-idler pair, 145 Pulse compression, 113
Phase-locked wave, 139 Pulse shaping, 105, 106, 108, 112, 117, 122,
Phase-matching condition, 9, 10 123, 217220, 235, 237, 239, 241, 242,
Phase-sensitive amplier, 144 244, 245, 296
Phase conjugation, 189, 190 Pulse train, 251
Phase estimation, 300 Pump depletion, 166168
Phase matching, 372374, 387, 399, 403 Pure phase delays, 476
Phase matching condition, 188
Phase modulation, 158, 159, 163, 311 Q
Phase modulator, 311315 Q factor, 456
Phase noise, 140 QAM regeneration, 146
Phase regeneration, 134, 135, 137, 140, 143, QAM signal, 145, 146, 150
146, 149 Quadratic phase modulation, 269
Phase sensitive amplier, 135 Quadrature phase shift keying (QPSK), 158
Phase sensitive extinction ratio, PSER, 141 QPSK regenerator, 146
Phase Sensitive Pulse Measurement, 60 QPSK tributaries, 149
Phase sensitive regenerator, 139 Quantum communication, 356, 359, 406, 407,
Phase Shift Keying, 169 409, 410
Index 511
Quantum dot, 19, 401, 405 Semiconductor optical amplier (SOA), 4,

Quantum frequency conversion (QFC), 357, 1826, 130, 143, 164, 482
398400, 402, 404, 405 Serial-to-parallel converter, 228, 229
Quantum key distribution (QKD), 380, Serial-to-parallel conversion, 263, 265, 269,
406410, 444 270, 273276, 278, 286
Quantum relay, 409411 Shannon capacity, 325327, 329, 330, 332,
Quantum repeater, 409411 333, 335, 336, 340, 342, 343, 346, 347
Quantum well, 19 Shannon-Nyquist sampling theorem (SFWM),
Quasi-phase matching, 165, 166, 178, 374 333, 364, 365, 367, 368, 376, 377,
Quasi-static Equaliser, 299 379384, 389
Quasilossless transmission, 344, 347 Shannon's theorem 12, 329
Qubits, 357, 358, 361 Short-wave infrared, 200
Signal-to-interference ratio, 493
R Signal of interest, 470, 492
Radio frequency (RF) domain, 470 Signal processing functions, 188, 190
Raised cosine, 217, 252 Signal quadratures, 207
Raised cosine Nyquist lter, 240 Signal to noise ratio (SNR), 138
Raised cosine pulse, 220 Silica planar lightwave circuit, 175
Raman amplication, 334, 344 Silicon-on-insulator, 35, 36
Raman response function, 9 Silicon devices, 164, 165, 166
Raman scattering, 107, 110, 118 Silicon nitride, 11, 14, 35, 38, 45
Rayleigh scattering, 194 Silicon waveguide, 7, 12, 13
Receiver DSP, 297 Similariton, 110, 112, 204
Recongurability, 480 Sinc-shape, 240, 241, 248, 251, 254, 255
Recongurable Optical Add Drop Single-mode to multimode (SM-MM)
Multiplexers, 130 combiner, 490
Rectangular frequency comb, 251, 254, 256 Single photon sources, 401, 402
Rectangular-shaped, 255 Size, weight, and power, 472
Reed-Solomon (RS) codes, 302 Slow light, 383, 453
Regeneration, 327, 342, 346, 347 SM-MM combiner, 492
Regenerative channel, 342 SM-MM optical combiner, 492
Regenerator, 316 Smartphones, 470
3R regenerator, 130 Soliton, 82, 83, 106, 107, 112, 115, 116, 118,
Resolution, 295, 296, 298 119, 122, 123
Resonator, 81, 82, 86, 383, 384, 389 Soliton formation, 204
RF beamformers, 485 Soliton self-frequency shift, 118
RF beamforming, 469 Space-division multiplexing, 327
Ring lter, 231 Space-Frequency Domain Retrieval, 195
Rogue waves, 119121 Space-time duality, 266, 270
Roll-off factor, 240, 247 Spatial light modulator, 482, 483
RRC lter, 297 Spectral coding, 442, 443, 446
RZ signal, 286 Spectral compression, 106, 113115
Spectral magnication, 282284, 286
S Spectral Phase Interferometry for Direct
Sampling, 298, 299, 309 Electric-Field Reconstruction, 63
Sampling Rate, 298 Spectral Sculpturing, 108, 112
Sampling theorem, 247 Spectral telescope, 277279, 281, 282, 284,
SBS gain bandwidth, 140 286
Schmidt decomposition, 445 Spectrographic representation, 271, 277
Second Harmonic Generation, 160 Split-and-delay-and-combine multiplexer, 265
Self-phase modulation (SPM), 106, 163, 252, Spontaneous four-wave mixing, 364, 366, 376
338 Spontaneous parametric downconversion
Self-seeded mixing, 201 (SPDC), 364367, 370372, 375, 376,
Self-wavelength cross gain modulation, 482 384, 385, 389, 391
512 Index
Spread spectrum, 444 Tunable laser, 293, 295

Spurious-free dynamic range, 452 Tunable optical OFDM demodulation, 180
Standard single mode bre, 270 Tunable photonic beamformer, 490
Steganography, 494, 499 Twin waves, 328, 345
Stimulated Brillouin scattering, 197, 200, 449, Two-photon absorption (TPA), 35, 36, 42, 43,
452, 453, 464 45, 55, 463
Stokes vector, 74 Two-pump mixer, 209
Subcarrier frequency, 221 Two-pump parametric mixer, 189, 207
Sum-frequency generation, 160, 400, 432 Two-pump parametric mixing, 189, 201, 211
Superchannel, 218, 219, 235, 236
Supercontinuum, 112, 119, 120, 167 U
Supercontinuum Generation, 55 Ultra-long link, 303
Supergaussian spectrum, 268 Ultra-low energy switching, 286
Super large areainverse dispersion ber Ultra-wideband (UWB), 460
(SLA-IDF), 253 Ultrashort pulses, 115, 117, 122
Surface plasmon polariton, 90
Switching window, 307 V
Synthesis, 252 Variable optical attenuators, 175
Vector PSA, 141
T Vernier, 463
Tapped-delay-line, 171, 172, 174, 175, 177, Very-Large-Scale-Integrated (VLSI) circuit,
178, 180, 479 489
TDL lter, 479 Viterbi-Viterbi algorithm, 300
Temporal-to-spectral conversion, 267 Viterbi-Viterbi estimation, 300
Temporal phase coding, 443 Viterbi and Viterbi phase estimation (VV), 298
Temporal resolution, 432
Temporal Talbot effect, 434 W
Thikhonov distribution, 336 Wavefunction, 365, 385
Three-wave mixing, 5 Waveguide synthesis, 198
Thresholder, 174 Waveguides, 368, 373377, 380, 381, 383, 405
Time bin, 397 Wavelength-division multiplexing, 485
Time domain, 299 Wavelength-Preserving, 273
Time-domain OFT, 266, 280 Wavelength-to-space demultiplexing, 133
Time domain butterfly FIR lter, 300 Wavelength conversion, 176
Time-Domain Processing, 284 Wavelength multicasting, 190, 191, 207, 208
Time-frequency domain signal processing, Wavelength selective switch, 283
207, 208 Wave mixing, 159, 162
Time-frequency entanglement, 444 Waveplate, 361
Time interleaving, 133 WDM-to-OTDM conversion, 276
Time lens, 263, 269, 270, 278 WDM correlator, 176, 177, 178
Time lens based OFT, 267 WDM grid manipulation, 277, 279
Time lens based serial-to-parallel converter, Wideband beamforming, 478
281 Wideband photonic beamformer, 491
Time Lens Principle, 266 Wide-band parametric mixing, 193, 195
Time lens system, 285 WIFI network, 473
Time-to-frequency mapping, 269 Wireless base station, 471
Time division multiplexed Nyquist data train, Wireless communication, 469
247 Wireless communication system, 470
Time division multiplex Nyquist pulse, 247
Transfer function, 308 X
Transponder, 291, 292, 302305, 316 XGM, 18, 2023
Triangular pulses, 111 XOR logic operation, 167, 169
True-time delay, 472 XOR operation, 170
TTD beamformer, 486 XPM, 18, 21, 22, 72, 7880, 82, 90

All-Optical Signal Processing

Uploaded by

Copyright:

Available Formats

All-Optical Signal Processing

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

All-Optical Signal Processing

Uploaded by

Copyright:

Available Formats

What is the Springer Series in Optical Sciences?

What is the Springer Series in Optical Sciences?

What topics are covered in the Springer Series in Optical Sciences?

What topics are covered in the Springer Series in Optical Sciences?

Springer Series in Optical Sciences 194

Toshimitsu Asakura Herbert Venghaus

Theodor W. Hnsch Horst Weber

More information about this series at http://www.springer.com/series/624

ISSN 0342-4111 ISSN 1556-1534 (electronic)

Library of Congress Control Number: 2015933154

Springer Cham Heidelberg New York Dordrecht London

Printed on acid-free paper

Springer International Publishing AG Switzerland is part of Springer Science+Business Media

Although it holds the promise for substantial processing speed improvements, in

necessity of electro-optic conversion. Discussed devices include nonlinear mode

Brescia, Italy Stefan Wabnitz

1 Materials and Structures for Nonlinear Photonics . . . . . . . . . . .. 1

2 CMOS Compatible Platforms for Integrated Nonlinear Optics ... 35

3 Optical Guided Wave Switching . . . . . . . . . . . . . . . . . . . . . .... 71

4 Temporal and Spectral Nonlinear Pulse Shaping Methods

5 Optical Regeneration . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..... 129

6 Photonic Signal Processing for Logic and Computation . . . . . . . . 157

7 Wide-Band and Noise-Inhibited Signal Manipulation

8 All-Optical Pulse Shaping for Highest Spectral Efciency . . . . . .. 217

9 Energy-Efcient Optical Signal Processing Using

10 Signal Processing Using Opto-Electronic Devices . . . . . . . . . . ... 291

10.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316

11 Optical Information Capacity Processing . . . . . . . . . . . . . . . . ... 325

12 Nonlinear Optics for Photonic Quantum Networks . . . . . . . . . ... 355

12.3.3 Future Directions with Quantum

13 Biphoton Pulse Shaping . . . . . . . . . . . . . . . . . . . . . . . ........ 423

14 Harnessing Nonlinear Optics for Microwave Signal Processing . .. 449

15 Ultrafast Optical Techniques for Communication

Antonella Bogoni CNIT, National Laboratory of Photonic Networks, Pisa, Italy

Simon Fabbri School of Engineering and Applied Science, Aston Institute of

Mary McCarthy School of Engineering and Applied Science, Aston Institute of

Michael J. Steel MQ Photonics Research Centre and CUDOS, Department of

Xin Gai, Duk-Yong Choi, Steve Madden and Barry Luther-Davies

Abstract In this chapter we summarize progress in materials and structures for

X. Gai D.-Y. Choi S. Madden B. Luther-Davies (&)

Springer International Publishing Switzerland 2015 1

A canonical optical switch consisted of a nonlinear Michelson interferometer in

FOM2PA 1. A similar FOM can be written for linear absorption as reads

1.2 All-Optical Processing Using (2) Nonlinearities

1.3 All-Optical Processing Using (3) Nonlinearities

Fig. 1.3 The group velocity

inequality becomes 4P < k 22 < 0, and Raman scattering does not

1.3.1 Properties of (3) Materials and Devices

1.3.1.1 Crystalline Silicon

Silicon photonic wire waveguides are typically fabricated by plasma etching

To conclude, it is worthwhile noting that in spite of the limitations of silicon, two

1.3.1.2 Hydrogenated Amorphous Silicon (aSi:H)

Fig. 1.7 Minimum pump

1.4 All-Optical Processing in Semiconductor Optical

Semiconductor optical ampliers (SOAs) have a long history in all-optical pro-

1.4.1 Types and Origins of Nonlinear Effects

As electrically pumped devices, SOAs are dominated by carrier density driven

1.4.2 Impairments in SOA Devices

1.4.2.1 Carrier Induced Switching Speed Limitations

Fig. 1.11 Gain and phase

TP jAt j2 =jAi j2 1 4K 1 K cos2 K 0:5cPin L 3:15