Seimetz M. High-Order Modulation For Optical Fiber Transmission (Springer, 2009) (3540937706) PDF

Springer Series in
optical sciences 143

founded by H.K.V. Lotsch
Editor-in-Chief: W. T. Rhodes, Atlanta

Editorial Board: A. Adibi, Atlanta
T. Asakura, Sapporo
T. W. Hänsch, Garching
T. Kamiya, Tokyo
F. Krausz, Garching
B. Monemar, Linköping
H. Venghaus, Berlin
H. Weber, Berlin
H. Weinfurter, München
Springer Series in
optical sciences
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 applications, 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
Georgia Institute of Technology
School of Electrical and Computer Engineering
Atlanta, GA 30332-0250, USA
E-mail: [email protected]
Editorial Board
Ali Adibi Bo Monemar

Georgia Institute of Technology Department of Physics
School of Electrical and Computer Engineering and Measurement Technology
Atlanta, GA 30332-0250, USA Materials Science Division
E-mail: [email protected] Linköping University
Toshimitsu Asakura 58183 Linköping, Sweden
Hokkai-Gakuen University
Faculty of Engineering Herbert Venghaus
1-1, Minami-26, Nishi 11, Chuo-ku Fraunhofer Institut für Nachrichtentechnik
Sapporo, Hokkaido 064-0926, Japan Heinrich-Hertz-Institut
E-mail: [email protected] Einsteinufer 37
Theodor W. Hänsch 10587 Berlin, Germany
Max-Planck-Institut für Quantenoptik E-mail: [email protected]
Hans-Kopfermann-Straße 1
85748 Garching, Germany Horst Weber
E-mail: [email protected] Technische Universität Berlin
Takeshi Kamiya Optisches Institut
Straße des 17. Juni 135
Ministry of Education, Culture, Sports 10623 Berlin, Germany
Science and Technology E-mail: [email protected]
National Institution for Academic Degrees
3-29-1 Otsuka, Bunkyo-ku Harald Weinfurter
Tokyo 112-0012, Japan Ludwig-Maximilians-Universität M
E-mail: [email protected] Sektion Physik
Ferenc Krausz Schellingstraße 4/III
Ludwig-Maximilians-Universität München 80799 München, Germany
Lehrstuhl für Experimentelle Physik E-mail: [email protected]
Am Coulombwall 1
85748 Garching, Germany
and
Max-Planck-Institut für Quantenoptik
Hans-Kopfermann-Straße 1
85748 Garching, Germany
Matthias Seimetz
High-Order Modulation
for Optical Fiber
Transmission
With 132 Figures
123
Dr. Matthias Seimetz
Fraunhofer-Institut für
Nachrichtentechnik
Heinrich-Hertz-Institut
Einsteinufer 37
10587 Berlin
Germany
[email protected]
ISSN 0342-4111 e-ISSN 1556-1534

ISBN 978-3-540-93770-8 e-ISBN 978-3-540-93771-5
DOI 10.1007/978-3-540-93771-5
Springer Dordrecht Heidelberg London New York
Library of Congress Control Number: 2009926839
c Springer-Verlag Berlin Heidelberg 2009

This work is subject to copyright. All rights are reserved, whether the whole or part of the material
is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broad-
casting, reproduction on microfilm or in any other way, and storage in data banks. Duplication of this
publication or parts thereof is permitted only under the provisions of the German Copyright Law of
September 9, 1965, in its current version, and permission for use must always be obtained from Springer.
Violations are liable to prosecution under the German Copyright Law.
The use of general descriptive names, registered names, trademarks, 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.
Cover design: eStudio Calamar S.L.
Printed on acid-free paper
Springer is part of Springer Science+Business Media (www.springer.com)

To my parents
Preface
The deployment of high-order modulation formats in optical fiber transmission sys-

tems is presently seen as a promising way of increasing spectral efficiency and of
making better use of the capacity of currently existing fiber infrastructure. Catering
to this interest, this book presents possible ways of generating and detecting optical
signals with high-order phase and quadrature amplitude modulation and character-
izes their system and transmission properties.
Several implementation options for high-order modulation optical transmitters
are possible. Their optical and electrical parts are described and their individual
signal properties are discussed. Receiver concepts with direct detection, homodyne
differential detection and homodyne synchronous detection are illustrated—starting
with optical frontends and ending with electrical data recovery. The description of
transmitters and receivers provided in the first part of the book does not only help to
demonstrate their functioning, but also allows their complexity and practicability to
be estimated and compared.
To advance understanding of the system and transmission behavior of high-order
modulation formats for optical fiber transmission, various system parameters such
as noise performances, optimal receiver filter bandwidths, required laser linewidths
and the chromatic dispersion and self phase modulation tolerances of a wide range
of modulation formats are highlighted in the second part of the book—considering
different line codes and many transmitter and receiver configurations. Currently, the
determination of attainable transmission distances for multi-span long-haul trans-
mission using high-order modulation formats represents an exciting field of re-
search. Recent results in this area are also covered by this book.
This monograph is intended for researchers in the field of optical communica-
tions, as well as for system designers who would like to learn about the properties
and complexity of optical systems employing high-order modulation.
The author wishes to express his cordial thanks to his colleagues from the Fraun-
hofer Institute for Telecommunications, Heinrich-Hertz-Institut, and to Prof. Peter-
mann for their technical assistance.
Berlin, January 2009 Matthias Seimetz
vii
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Part I Transmitters and Receivers
2 Transmitter Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.1 Transmitter Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.1.1 Lasers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.1.2 External Optical Modulators . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.1.3 Pulse Carvers and Impulse Shapers . . . . . . . . . . . . . . . . . . . . . 21
2.2 Multi-Level Signaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3 ASK Transmitters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.4 DPSK Transmitters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.4.1 Serial DPSK Transmitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.4.2 Parallel DPSK Transmitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.4.3 Differential Encoding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.4.4 Signal Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.5 Star QAM Transmitters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.6 Square QAM Transmitters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.6.1 Differential Quadrant Encoding . . . . . . . . . . . . . . . . . . . . . . . . 37
2.6.2 Serial Square QAM Transmitter . . . . . . . . . . . . . . . . . . . . . . . . 38
2.6.3 Conventional IQ Transmitter . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.6.4 Enhanced IQ Transmitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.6.5 Tandem-QPSK Transmitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.6.6 Multi-Parallel MZM Transmitter . . . . . . . . . . . . . . . . . . . . . . . 52
2.6.7 Signal Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3 Receiver Configurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.1 Receiver Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.1.1 Optical and Electrical Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
ix
x Contents
3.1.2 Delay Line Interferometer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

3.1.3 2 × 4 90◦ Hybrid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.1.4 3 × 3 Coupler . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.2 Direct Detection Receivers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.2.1 Direct Detection Receiver with DLIs . . . . . . . . . . . . . . . . . . . . 67
3.2.2 Direct Detection IQ Receiver with 2 × 4 90◦ Hybrid . . . . . . . 71
3.2.3 Data Recovery for Differential Detection . . . . . . . . . . . . . . . . 73
3.3 Fundamentals of Coherent Detection . . . . . . . . . . . . . . . . . . . . . . . . . . 79
3.3.1 Coherent Detection Principle . . . . . . . . . . . . . . . . . . . . . . . . . . 79
3.3.2 Coherent Detection with Amplifier Noise . . . . . . . . . . . . . . . . 82
3.3.3 Optical Quadrature Frontend . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
3.3.4 Polarization Diversity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
3.4 Homodyne Receivers with Differential Detection . . . . . . . . . . . . . . . . 86
3.4.1 Phase Diversity Receivers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
3.4.2 Digital Differential Demodulation . . . . . . . . . . . . . . . . . . . . . . 91
3.5 Homodyne Receivers with Synchronous Detection . . . . . . . . . . . . . . . 93
3.5.1 Carrier Synchronization Techniques . . . . . . . . . . . . . . . . . . . . 94
3.5.2 Optical Phase Locked Loop . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
3.5.3 Digital Phase Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
3.5.4 Data Recovery for Synchronous Detection . . . . . . . . . . . . . . . 111
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
4 Effort Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119

4.1 Transmitter Complexity and Feasibility . . . . . . . . . . . . . . . . . . . . . . . . 119
4.2 Receiver Complexity and Feasibility . . . . . . . . . . . . . . . . . . . . . . . . . . 122
Part II System Performance
5 System Simulation Aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

5.1 Data Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
5.2 Eye Opening Penalty for Multi-Level Eyes . . . . . . . . . . . . . . . . . . . . . 131
5.3 Monte Carlo Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
5.4 Semi-Analytical BER Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
5.4.1 DBPSK Direct Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
5.4.2 Extension to Higher-Order DPSK and Star QAM . . . . . . . . . 138
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
6 Fiber Propagation Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

6.1 Fiber Attenuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
6.2 Chromatic Dispersion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
6.3 Kerr Nonlinearities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
6.4 Other Propagation Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
6.4.1 Nonlinear Scattering Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
6.4.2 Polarization Mode Dispersion . . . . . . . . . . . . . . . . . . . . . . . . . . 151
6.4.3 Nonlinear Phase Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
Contents xi
7 Back-to-Back and Single-Span Transmission . . . . . . . . . . . . . . . . . . . . . . 155

7.1 Systems with Direct Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
7.1.1 OSNR Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
7.1.2 Optimal Receiver Filter Bandwidths . . . . . . . . . . . . . . . . . . . . 160
7.1.3 Laser Linewidth Requirements . . . . . . . . . . . . . . . . . . . . . . . . . 164
7.1.4 Chromatic Dispersion Tolerances . . . . . . . . . . . . . . . . . . . . . . . 166
7.1.5 Self Phase Modulation Tolerances . . . . . . . . . . . . . . . . . . . . . . 168
7.1.6 Nonlinear Phase Shift Compensation . . . . . . . . . . . . . . . . . . . . 171
7.1.7 Parameter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
7.2 Systems with Homodyne Differential Detection . . . . . . . . . . . . . . . . . 177
7.2.1 Noise Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
7.2.3 Transmission Parameter Tolerances . . . . . . . . . . . . . . . . . . . . . 182
7.3 Systems with Homodyne Synchronous Detection . . . . . . . . . . . . . . . . 185
7.3.1 Noise Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
7.3.3 Chromatic Dispersion Tolerances . . . . . . . . . . . . . . . . . . . . . . . 194
7.3.4 Self Phase Modulation Tolerances . . . . . . . . . . . . . . . . . . . . . . 198
7.3.5 Nonlinear Phase Shift Compensation . . . . . . . . . . . . . . . . . . . . 199
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
8 Multi-Span Long-Haul Transmission . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207

8.1 8PSK Multi-Span Transmission Experiments . . . . . . . . . . . . . . . . . . . 208
8.1.1 Optical Inline CD Compensation . . . . . . . . . . . . . . . . . . . . . . . 208
8.1.2 Electrical CD Compensation at the Receiver . . . . . . . . . . . . . . 212
8.2 Star 16QAM Multi-Span Transmission Experiments . . . . . . . . . . . . . 213
8.2.1 Single-Channel Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
8.2.2 WDM Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
8.3 Comparison of Transmission Distances . . . . . . . . . . . . . . . . . . . . . . . . 219
8.4 Nonlinear Phase Shift Compensation . . . . . . . . . . . . . . . . . . . . . . . . . . 221
8.4.1 Systems with Optical Inline CD Compensation . . . . . . . . . . . 221
8.4.2 Systems with Electrical CD Compensation . . . . . . . . . . . . . . . 224
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
9 Performance Trends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227

9.1 Migration to Higher-Order Formats—Parameter Trends . . . . . . . . . . 227
9.2 Impact of the System Configuration and Pulse Shape . . . . . . . . . . . . . 230
9.3 Reduction of Attainable Transmission Distances . . . . . . . . . . . . . . . . . 232
9.4 Goals of Future Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
A Differential Encoders for 8DPSK / 16DPSK . . . . . . . . . . . . . . . . . . . . . . . 239

xii Contents
B Theoretical BER Formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
Acronyms
Mathematical Symbols
α Fiber attenuation
αd B Fiber attenuation in decibels
αN L Scaling factor for nonlinear phase noise compensation
β0 Propagation constant at ω = ωs
β1 Group delay per length unit at ω = ωs
β2 Chromatic dispersion at ω = ωs
β3 Dispersion slope at ω = ωs
χ1 Linear susceptibility
χ3 Third-order nonlinear susceptibility
δ1 Spreading in the upper level of an eye diagram
δ2 Spreading in the lower level of an eye diagram
1e Eye spreading
1f Reference noise bandwidth (frequency scale)
1 f max Maximum resolvable frequency offset estimate
1λ Reference noise bandwidth (wavelength scale)
1n e f f (t) Change of the effective refractive index
1νe f f Beat-linewidth after coherent detection
1νlo Linewidth of the LO
1νs Linewidth of the signal laser
1ω Angular frequency offset after coherent detection
1ωest Angular frequency offset estimate
1ωestmax Maximum resolvable angular frequency offset estimate
δ P(t) Laser intensity fluctuations due to intensity noise
18i Phase error difference accumulating in D blocks
1φk Differential instantaneous phase samples
1ϕ(t) Modulation phase difference of two consecutive symbols
1ϕk Modulation phase difference samples
1ϕ I (t) Relative phase shift of the MZM-branches in the I -arm
1ϕ M Z M (t) Relative phase shift of the MZM-branches
xiii
xiv Acronyms
1ϕn (t) Random phase change due to overall laser phase noise
1ϕn k Overall laser phase noise samples
1ϕn s (t) Random phase change due to signal laser phase noise
1ϕn k,d Random phase change due to phase noise after d symbols
1ϕ Q (t) Relative phase shift of the MZM-branches in the Q-arm
1t Rise time of the electrical driving signals
1τ P M D Group delay difference of the PSP due to PMD
ǫ0 Vacuum permittivity
ǫre f f Effective relative permittivity
ǫr Relative permittivity
η Quantum efficiency
γ Nonlinear propagation coefficient
λ Wavelength
3 Diagonal matrix with eigenvalues of matrix A
λp Eigenvalues of the matrix A
λs Signal laser wavelength
ν Filter order
Normalized angular frequency ω − ωs
ω Angular frequency
ω0 Band-pass filter center frequency
ωC W Angular frequency of a CW laser
ωc 3 dB bandwidth
ωlo LO angular frequency
ωs Signal laser angular frequency
ωV C O Angular frequency of an electrical VCO
ϕ(t) Modulation phase
ϕ̇(t) Chirp of the optical information signal
8(t) Phase error due to frequency offset and laser phase noise
φ(t) Instantaneous phase
ϕ(z, t) Time and location dependent modulation phase
ϕ0 (t) Initial phase offset after coherent detection
ϕ1 (t) Phase modulation in the upper MZM branch
ϕ2 (t) Phase modulation in the lower MZM branch
ϕbk Phase of a symbol before differential encoding
ϕ DL I Phase shift of a DLI
ϕ DL I I Phase shift of the DLI in the I -branch
ϕ DL I Q Phase shift of the DLI in the Q-branch
8ek Residual phase error of the k-th symbol
ϕ I Q M (t) Phase modulation of an IQM
ϕ I (t) Phase modulation in the I -arm (enhanced IQ TX)
8i Corrected phase error estimate of the i-th block
ϕk Phase of a symbol after differential encoding
8k Phase error samples
φk Instantaneous phase samples
ϕlo LO initial phase
Acronyms xv
ϕn (t) Overall laser phase noise after coherent detection

ϕn k Overall laser phase noise samples
ϕnlo (t) LO phase noise
ϕn s (t) Signal laser phase noise
ϕN L Nonlinear phase shift
ϕ P M (t) Phase modulation of a phase modulator
ϕ Q (t) Phase modulation in the Q-arm (enhanced IQ TX)
ϕs Signal laser initial phase
8 yk (s) Moment generating function of y(tk )
ψ Arbitrary phase shift
ψ̃ Arbitrary phase shift at the hybrid input
9i Phase error estimate of the i-th block
σ2 Noise variance
σϕ2L Phase error variance of the linear noise
σϕ2N L Nonlinear phase shift variance
σϕ2N L ,Comp Nonlinear phase shift variance after compensation
τd Loop delay time
τgr Group delay per length unit
τgr0 Group delay for ω = 0
A Symbol alphabet
An Elements of alphabet A
A Receiver matrix
A(t) Normalized complex modulation envelope
a(t) Normalized modulation amplitude
a(z, t) Time and location dependent modulation amplitude
abk Amplitude of a symbol
Ae f f Effective core area
a I Q M (t) Amplitude modulation of an IQM
a I (t) Amplitude modulation in the I -arm (enhanced IQ TX)
Ak Normalized complex modulation envelope samples
ak Normalized modulation amplitude samples
ak′ Modulation amplitude samples after phase estimation
{an } Data Sequence with elements an , n = (1..b)
a Q (t) Amplitude modulation in the Q-arm (enhanced IQ TX)
b Length of a bit sequence
Bel Electrical receiver filter 3 dB bandwidth
bki In-phase coordinate of a symbol
bk, p Elements of vector bk
bk Complex symbol in the k-th symbol interval
bk Auxiliary vector equal to xk · U
Bn Optical noise bandwidth
bn k n-th bit of a symbol with m bits in the k-th symbol interval
Bopt Optical receiver filter 3 dB bandwidth
q
bk Quadrature coordinate of a symbol
Bs Signal bandwidth
xvi Acronyms
BW D M Electrical bandwidth for WDM channel separation

c Speed of light
cn k n-th Square QAM coder output bit (k-th symbol interval)
d Delay in number of symbol durations
D Processing delay in number of blocks
D Electric flux density vector
D(z, t) Time and location dependent electric flux density
Dλ Dispersion parameter
de Height of an eye diagram
Dmax Maximal accumulated chromatic dispersion
dn k n-th differentially encoded bit of a symbol with m bits
E Electrical field vector
e Electron charge
e Row vector used for Fourier transform
Ecw (t) Normalized electrical field of the signal laser
E I (t) Normalized electrical field in the I -branch
elo LO polarization unit vector
Elo (t) Normalized electrical field of the LO
ep Elements of vector e
E Q (t) Normalized electrical field in the Q-branch
es Signal laser polarization unit vector
E s (t) Normalized electrical signal field
E s (z, jω) Fourier transform of E s (z, t)
E s (z, t) Time and location dependent electrical
p signal field
Ẽ s (z, t) Electrical field normalized to Ae f f
E s,avg Average energy per symbol
ex Polarization unit vector in x-direction
ey Polarization unit vector in y-direction
ek/⊥ Polarization unit vectors of the noise
F Noise figure of an optical amplifier
F Diagonal matrix with Fourier frequency components
f Frequency
Fd B Noise figure of an optical amplifier in decibels
F p, p Diagonal elements of the matrix F
fs Signal laser frequency
G Optical amplifier gain
Gr Resistor’s conductance
h Planck constant
H Diagonal transfer matrix of the optical receiver filter
H1 Receiver transfer matrix (upper DLI output)
H2 Receiver transfer matrix (lower DLI output)
He ( f ) Transfer function of the electrical receiver filter
Ho ( f ) Transfer function of the optical receiver filter
H p, p Diagonal elements of the matrix H
I Identity matrix
Acronyms xvii
I (t) Photocurrent in the I -branch after phase noise cancelation

I Ik Received in-phase electrical signal samples
I I D (t) Photocurrent in the intensity detection branch
I I (t) Photocurrent in the I -branch
Ik In-phase signal samples after phase estimation
ik In-phase coordinate of a symbol, scaled to unity
I M (t) Corrective signal within an OPLL
In n-th order modified Bessel function of the first kind
I Qk Received quadrature electrical signal samples
I Q (t) Photocurrent in the Q-branch
i sh Shot-noise photocurrent
i sh I Shot-noise photocurrent in the I -branch
i sh Q Shot-noise photocurrent in the Q-branch
i th Thermal noise photocurrent
kB Boltzmann constant
L Total number of samples considered for Fourier expansion
l Fiber length per span
le f f Effective length
lel Interaction length
M Number of symbols
m Number of bits per symbol
N Phase estimation block length
n(t) Optical amplifier noise
n(ω, x) Frequency and intensity dependent refractive index
N0 Amplifier noise two-sided PSD in one polarization
n0 Linear refractive index
n2 Nonlinear index coefficient
n k/⊥ (t) Parallel / orthogonally polarized noise components
nk/⊥ Vector with the noise coefficients
nB Binary data sequence order
nc Complex noise
n ck Complex noise samples
n clad Refractive index of the fiber cladding
NClass I Number of Class I symbols in a block of N symbols
n cor e Refractive index of the fiber core
ne f f Effective refractive index
NF S Number of fiber spans
NFW M Number of mixing products generated by FWM
nI In-phase component of complex noise
n Ik In-phase noise samples
ni Phase estimation segment counter
nk Complex noise samples after raising to the M-th power
N L O−AS E Power spectral density of the LO-ASE noise
n L O−AS E (t) LO-ASE noise
n p,k/⊥ Elements of vector nk/⊥
xviii Acronyms
N ph Number of phase states

nQ Quadrature component of complex noise
n Qk Quadrature noise samples
nS Symbol sequence order
n sh k Complex shot-noise samples
n sp Spontaneous emission parameter
NW D M Number of WDM channels
P Electric polarization vector
P(t) Instantaneous optical power
p(t) Pulse shape of the electrical driving signals
P(z, t) Time and location dependent optical power
PL Linear electric polarization vector
Plo Power of the LO
PN L Nonlinear electric polarization vector
PN L (z, t) Time and location dependent nonlinear polarization
Pn Noise power
PR X,avg Average received optical power
PR X (t) Received instantaneous optical power
Ps Power of the signal laser
Ps,avg Average signal power
Psym Symbol error probability
PT X,avg Average fiber input power
PT X (t) Instantaneous fiber input power
p yk (y) Probability density function of y(tk )
Q Transfer matrix of the electrical receiver filter
Q(t) Photocurrent in the Q-branch after phase noise cancelation
Qk Quadrature signal samples after phase estimation
qk Quadrature coordinate of a symbol, scaled to unity
Q p,q Elements of the matrix Q
R Responsivity
r Core radius
r1 Amplitude of the inner circle of a Star QAM constellation
r2 Amplitude of the outer circle of a Star QAM constellation
rB Bit rate
rS Symbol rate
Sλ Dispersion slope parameter
si,n Decision threshold in the in-phase branch
sn Decision threshold for arg-decision
sq,n Decision threshold in the quadrature branch
si,n k Decision result at threshold si,n
sn k Decision result at threshold sn
sq,n k Decision result at threshold sq,n
S pS Number of samples per symbol
T Temperature
t Time
Acronyms xix
t′ Retarded time scale t − τgr · z

T0 Time interval observed for semi-analytical BER estimation
TB Duration of one bit
Tblock Duration of one block for phase estimation
tc Coherence time
Te 1/e pulse duration of the electrical driving signals
tk Sampling instants
TS Duration of one symbol
U Unitary matrix with eigenvectors of matrix A
u(t) Driving voltage
u 1 (t) Driving voltage for the upper MZM branch
u 2 (t) Driving voltage for the lower MZM branch
u I (t) Driving voltage of the MZM in the I -arm
u I M Z M (t) MZM driving voltage for the enhanced IQ TX (I -arm)
u I P M (t) PM driving voltage for the enhanced IQ TX (I -arm)
u I M (t) Driving voltage for intensity modulation
u P M (t) Driving voltage of a PM
u Q (t) Driving voltage of the MZM in the Q-arm
u Q M Z M (t) MZM driving voltage for the enhanced IQ TX (Q-arm)
u Q P M (t) PM driving voltage for the enhanced IQ TX (Q-arm)
Vπ Voltage for π phase shift (MZM power transfer function)
Vπ1 Driving voltage for a phase shift of π, upper MZM branch
Vπ2 Driving voltage for a phase shift of π, lower MZM branch
Wϕ̇ns Power spectral density of the signal laser frequency noise
W E cw (ω) Power spectral density of the emitted CW light
X (t) Received electrical signal phasor
x(t) Complex signal envelope
x(z, t) Time and location dependent complexp signal envelope
x̃(z, t) Complex envelope normalized to Ae f f
x(z, jω) Fourier transform of x(z, t)
xk Vector with Fourier coefficients of x(t)
xk, p Elements of vector xk
Xk Received complex electrical signal samples
X kClass I Received Class I complex electrical signal samples
X k′ Complex electrical signal samples after phase estimation
y(tk ) Photocurrent at the decision gate at sampling instants tk
ỹ(tk ) Photocurrent after the BD at sampling instants tk
zk/⊥ Auxiliary vector equal to nk/⊥ · U
z p,k/⊥ Elements of vector zk/⊥
xx Acronyms
Abbreviations
A/D Analog to Digital
AFC Automatic Frequency Control
ASE Amplified Spontaneous Emission
ASK Amplitude Shift Keying
AWGN Additive White Gaussian Noise
BD Balanced Detector
BER Bit Error Ratio
BPSK Binary Phase Shift Keying
BtB Back-to-Back
CD Chromatic Dispersion
CL Closed Loop
CMA Constant Modulus Algorithm
CSRZ Carrier-Suppressed Return to Zero
CW Continuous Wave
DB Duobinary
dB Decibel
DBBS De Bruijn Bit Sequence
DBPSK Differential Binary Phase Shift Keying
DBSS De Bruijn Symbol Sequence
DC Direct Current
DCF Dispersion Compensating Fiber
DD Direct Detection
DFB Distributed Feedback
DLI Delay Line Interferometer
DMF Dispersion Managed Fiber
DMT Discrete Multi-Tone
DPSK Differential Phase Shift Keying
DQPSK Differential Quadrature Phase Shift Keying
DSF Dispersion Shifted Fiber
DSL Digital Subscriber Line
DSP Digital Signal Processor
ECL External Cavity Laser
EDE Electronic Distortion Equalization
EDFA Erbium Doped Fiber Amplifier
EO Eye Opening
EOP Eye Opening Penalty
FEC Forward Error Correction
FF Feed Forward
FFT Fast Fourier Transform
FIR Finite Impulse Response
FSK Frequency Shift Keying
FWM Four Wave Mixing
HHI Heinrich-Hertz-Institut
Acronyms xxi
I In-phase
IF Intermediate Frequency
IM Intensity Modulation
IP Internet Protocol
IQ In-phase Quadrature
IQM In-phase Quadrature Modulator
IS Impulse Shaper
ISI Inter-Symbol Interference
LAN Local Area Network
LED Light Emitting Diode
LMS Least Mean Square
LO Local Oscillator
LPF Low-pass Filter
MC Monte Carlo
MGF Moment Generating Function
MMA Multiple Moduli Algorithm
MMI Multi-Mode Interference
MPOLSK M-ary Polarization Shift Keying
MZM Mach-Zehnder Modulator
NRZ Non-Return to Zero
OOK On-Off Keying
OPLL Optical Phase Locked Loop
OSNR Optical Signal to Noise Ratio
OTDM Optical Time Division Multiplexing
PBC Polarization Beam Combiner
PBS Polarization Beam Splitter
PD Photodiode
PDF Probability Density Function
PDM Polarization Division Multiplexing
PE Phase Estimation
PLL Phase Locked Loop
PM Phase Modulator
PMD Polarization Mode Dispersion
PRBS Pseudo Random Binary Sequence
PS Power Splitter
PSD Power Spectral Density
PSK Phase Shift Keying
PSP Principle States of Polarization
Q Quadrature
QAM Quadrature Amplitude Modulation
QPSK Quadrature Phase Shift Keying
RIN Relative Intensity Noise
RR Ring Ratio
RX Receiver
RZ Return to Zero
xxii Acronyms
SBS Stimulated Brillouin Scattering

SMF Single Mode Fiber
SNR Signal to Noise Ratio
SPM Self Phase Modulation
SRS Stimulated Raman Scattering
SSMF Standard Single Mode Fiber
TDC Tunable Dispersion Compensation
TX Transmitter
UMTS Universal Mobile Telecommunications System
VCO Voltage Controlled Oscillator
WDM Wavelength Division Multiplexing
WLAN Wireless Local Area Network
XPM Cross Phase Modulation
Chapter 1
Introduction
Abstract After a brief discussion of the evolution of optical networks in the last two
decades, advanced modulation and coherent detection are pointed out as emerging
key technologies for fulfilling the expected bandwidth demands of future optical
fiber networks. Subsequently, an overview of current developments and today’s state
of research in the fields of advanced modulation and coherent detection is given.
Finally, the scope of this book is localized and the chapter’s contents are shortly
outlined.
For millions of people around the world, surfing the Internet day to day to get infor-
mation, check e-mails, dispatch shopping comfortably from their desk and down-
load multimedia and data files has become a natural habit. However, for most of
them, the sophisticated technology they are using to connect to the central office
over a simple twisted copper pair with satisfactory transmission speed remains con-
cealed. Originally laid to transmit voice signals up to 4 kHz bandwidth, twisted pairs
could be utilized to transmit data signals up to 64 kbit/s with cable modems in the
mid nineties. Today, digital subscriber line (DSL) technology has captured the mar-
ket. Several hundred millions of subscribers worldwide have broadband access to
the Internet via DSL, with transmission speeds of up to several Mbit/s in the local
loop. These data rates can only be provided by using advanced transmission tech-
niques like very high-order quadrature amplitude modulation (QAM) and discrete
multi-tone (DMT), where the data band is divided into hundreds of separate chan-
nels, which are adaptively QAM modulated with the aid of intelligent bit loading
algorithms [15].
In contrast to electrical systems, optical fiber communication systems are still
very distant from a commercial practical implementation of high-order modula-
tion formats. The difficult-to-handle optical phase and technological difficulties with
cost-effectively manufacturing more complex high-speed electronic devices restrict
the presently installed optical systems almost exclusively to a simple deployment of
intensity modulation (IM) on the transmitter side and direct detection (DD) at the
receiver end. For optical systems, all system concepts beyond IM-DD can still be
M. Seimetz, High-Order Modulation for Optical Fiber Transmission, 1

Springer Series in Optical Sciences 143, DOI 10.1007/978-3-540-93771-5 1,

2 1 Introduction
qualified as advanced. However, favored by the very low fiber attenuation of about
0.2 dB/km across several THz of bandwidth, optical fiber communication is superior
to other wireline or wireless communication technologies and can support very high
capacities of several Tbit/s over many thousand kilometers, even when advanced
modulation formats are not employed.
Reduction of Costs by Innovative Technologies
With the objective of reducing costs per information bit in optical communication
networks, per fiber capacities and optical transparent transmission lengths have been
stepped up by the introduction of new technology in recent years. A crucial innova-
tion was the Erbium-doped fiber amplifier (EDFA) at the beginning of the nineties
[30]. Using EDFAs, long distances can be bridged without electro–optical conver-
sion. Furthermore, the wavelength division multiplex (WDM) technology which al-
lows a lot of wavelength channels to be simultaneously transmitted over one fiber,
has benefited from the high bandwidth of the EDFA, since several WDM channels
can be amplified using only one EDFA. During the nineties, the capacity-distance
product was further enhanced by employing other optical key technologies such as
optical dispersion compensation, Raman amplification and advanced optical fibers,
as well as through electronic means such as forward error correction (FEC) and the
adaptive compensation of chromatic dispersion (CD) and polarization mode disper-
sion (PMD). Figure 1.1 illustrates the enhancement of capacit y · distance/cost
through innovative technology.
Coherent
Capacity ⋅ Distance / Cost
Adaptive Detection
CD & PMD Kerr
Comp. Comp.
Raman Advanced
Amp. Modulation
FEC
WDM
CD
Fig. 1.1 Enhancement of Comp.
Present
capacit y · distance/cost EDFA
through innovative technol-
ogy in optical communication
networks, based on [12] Time
In WDM systems, cost reduction is not only achieved by increasing the transpar-
ent transmission lengths and the transmitted per fiber capacities, but also by shar-
ing optical components over many channels. An important performance measure
is spectral efficiency, which is defined as the ratio of the data rate per channel to
the WDM channel spacing. For higher spectral efficiency and a fixed per-channel
1 Introduction 3
data rate, more channels can be placed within the limited wavelength window of
the shared components so that the transmitted data volume increases. One possible
approach aimed to increase the capacity of ultra long-haul transoceanic submarine
WDM transmission systems, is making the most of the available bandwidth of stan-
dard C-band optical amplifiers by increasing spectral efficiency.
Two emerging optical key technologies which are seen as a possible further step
towards even more cost effective optical networks, are advanced optical modulation
and coherent detection. Through the adoption of high-order modulation formats,
higher spectral efficiencies can be reached through the reduced symbol rate and the
spectral narrowing therewith aligned. Furthermore, as was theoretically shown in
[13], only coherent detection permits convergence to the ultimate limits of spectral
efficiency. Several bit/s/Hz per polarization can be transmitted with unconstrained
coherent detection, even considering the impact of fiber nonlinearity. On the other
hand, when employing these new technologies, the complexity of the transmitters
and receivers increases, so that cost reduction due to higher spectral efficiency has to
be weighed up against higher hardware costs, as illustrated in Fig. 1.2. Continuing
research and future investigations will help judge the economic potential of high-
order modulation and coherent detection. Thereby, the transmission characteristics
and the reachable transparent transmission lengths, which are largely unknown for
high-order modulation formats, must be also considered.
More complex Advanced Modulation Higher spectral

components efficiency,
Coherent Detection
sharing components
Reach?
Higher costs Cost reduction
Fig. 1.2 Economical issues for the introduction of advanced modulation and coherent detection
Advanced Optical Modulation
The electrical field in single mode fibers (SMF) exhibits three physical parameters
that can be used to carry information. Besides amplitude and phase, polarization can
also be exploited for modulation. Plenty of different modulation formats based on
the modulation of all the quadratures of the optical field were proposed in the early
nineties, primarily in association with coherent detection. However, these investiga-
tions received relatively little attention because the necessary complex high-speed
electronics were rarely available. Moreover, the emergence of the EDFA offered
4 1 Introduction
completely new perspectives for simple IM-DD systems so that there was enough
potential to increase capacity even without high-order modulation. Thus, the inves-
tigation of high-order modulation formats remained mainly confined to the descrip-
tion of some transmitter and receiver structures and the calculation of theoretical bit
error ratio (BER) noise performances. A summary of this early work can be found
for instance in [2].
Having optimized optical systems with binary intensity modulation (also denoted
as on–off keying, OOK) and direct detection over the years using the technologies
described above, a new interest in alternative optical modulation formats emerged
in the late nineties. At first, there was interest in obtaining higher robustness against
fiber propagation effects and extending transmission reach, rather than the pursuit of
higher spectral efficiencies. Differential binary phase shift keying (DBPSK, some-
times simply denoted as DPSK) was shown to feature higher robustness against
nonlinear effects [31, 34]. The DBPSK format seemed to be attractive because di-
rect detection can be further employed by placing a simple optical interferometer in
front of the photodiode, so that the effort growth compared to systems with IM-DD
is relatively small.
Moreover, the influence of the optical pulse shape was examined, and return to
zero (RZ) signals were shown to have different transmission properties than non-
return to zero (NRZ) signals. Data modulation formats with auxiliary phase coding
attracted attention in the optimization of the transmission behavior of binary in-
tensity modulation. For instance, optical duobinary (DB) exhibits higher tolerance
against chromatic dispersion. Modulation formats with auxiliary phase coding are
not covered by this book. Further information can be obtained from [52].
In recent years, the investigation of optical high-order modulation formats has
begun to play an important role in several research projects. By encoding
m = log2 M data bits on M symbols, the symbol rate is reduced by m compared
to the data rate, and higher spectral efficiencies can be obtained due to the spec-
tral narrowing. On the one hand, high-order modulation allows upgrading to higher
channel data rates by using existing lower-speed equipment, and thus exceeding the
limits of present high-speed electronics and digital signal processing. On the other
hand—when keeping the data rate constant—the transmission with lower symbol
rates allows for smaller channel spacings and brings about a higher tolerance against
transmission impairments such as chromatic dispersion and polarization mode dis-
persion. However, these benefits are accompanied by a reduced tolerance to noise
and self phase modulation (SPM), and with a higher complexity of components.
The most simple optical multi-level signaling scheme is M-ary amplitude shift
keying (ASK), where information is encoded in several intensity levels. It was
shown in [49] and [56], that M-ary ASK requires high optical signal to noise ra-
tios (OSNR) for direct detection, especially in optically amplified links due to the
intensity dependence of the signal-spontaneous beat noise. For instance, a 2.5 times
higher dispersion tolerance compared to OOK can be achieved by 4ASK, but only
among acceptance of a 5 dB power penalty due to noise. Thus, the use of M-ary
ASK formats should mainly be considered for short reach applications. M-ary ASK
formats are not further discussed in this book.
1 Introduction 5
As the logical next step after DBPSK, differential quadrature phase shift key-
ing (DQPSK) was the first optical multi-level phase modulation format whose trans-
mission characteristics were intensively examined, for instance in [32, 53]. Since it
features good transmission performance and doubled spectral efficiency, DQPSK is
presently seen as a promising candidate for future networks, in spite of the greater
effort required for practical realization of the transponder hardware. Many of the re-
cently proposed long-haul dense WDM transmission records are based on systems
with RZ-DQPSK modulation.
Encouraged by the current trends and today’s progress in high-speed electronics
and digital signal processing technology, even higher-order modulation formats are
being investigated in present research projects. With direct detection, 8-ary differ-
ential phase shift keying (8DPSK) has been theoretically examined by Ohm [28]
and Yoon et al. [54], and experimentally demonstrated by Serbay et al. [42]. By us-
ing coherent detection, 8-ary PSK has been experimentally reported by Tsukamoto
et al. [47], Seimetz et al. [39], Freund et al. [9], Zhou et al. [57] and Yu et al. [55].
The transmission performance of the 16PSK / 16DPSK formats is presented in detail
within this book.
By combining intensity and phase modulation, the number of phase states can
be reduced for the same number of symbols, leading to larger Euclidean distances
between the symbols. The symbols can be arranged in different circles (Star QAM)
or can be positioned in a square (Square QAM). Star QAM experiments have been
reported with four phase levels in [29] and [41, 43] for 2ASK-DQPSK and 4ASK-
DQPSK, respectively. The 2ASK-8DPSK format, which is denoted in the follow-
ing as Star 16QAM, has been investigated by computer simulations in [35] and
experimentally in [40], and is extensively discussed within this book, just as differ-
ent Square QAM formats. Very recently, Square 16QAM has been experimentally
demonstrated in [33, 51]. Even higher-order Square QAM experiments have been
shown so far only for moderate symbol rates [25].
Coherent Detection
In many wireline and wireless telecommunication systems, receivers are based on

coherent detection because coherent receivers feature a high receiver sensitivity
and can select a channel from a frequency comb by tuning the oscillator to the
desired frequency, as it is known from radio. In optical fiber communication, coher-
ent systems were an important topic of investigation in the late eighties and early
nineties. With the emergence of the EDFA, the former advantage of a higher re-
ceiver sensitivity—compared to direct detection—disappeared. Comparable sensi-
tivities could be achieved by direct detection receivers with optical preamplifiers
[46]. Thus, as it had for the high-order modulation formats, research into this area
ceased, the more so as the components were complex and costly.
Nowadays however, coherent optical systems are reappearing as an area of
interest [19]. The linewidth requirements have relaxed with increasing channel
data rates, and sub-MHz linewidth lasers have recently been developed [6, 16].
6 1 Introduction
More recently, the high-speed digital signal processing available allows for the
implementation of critical operations like phase locking, frequency synchroniza-
tion and polarization control in the electronic domain through digital means. For-
mer concepts for carrier synchronization with optical phase locked loops (OPLL)
[1, 7, 14, 17, 18, 27] can be replaced by subcarrier OPLLs [4] or digital phase esti-
mation [20–23, 26, 36, 37, 45, 48]. Thus, under the new circumstances, the chances
of cost effectively manufacturing stable coherent receivers are increasing.
In addition to the already mentioned potentials of spectral efficiency, coherent
detection provides several advantages. Coherent detection is very beneficial within
the design of optical high-order modulation systems because all the optical field
parameters (amplitude, phase, frequency and polarization) are available in the elec-
trical domain. Therefore the demodulation schemes are not limited to the detec-
tion of phase differences as for direct detection, but arbitrary modulation formats
and modulation constellations can be received. Furthermore, the preservation of the
temporal phase enables more effective methods for the adaptive electronic com-
pensation of transmission impairments like chromatic dispersion and nonlinearities
[11, 44]. When used in WDM systems, coherent receivers can offer tunability and
enable very small channel spacings, since channel separation can be performed by
high-selective electrical filtering.
Two main challenges for the practical fabrication of coherent receivers are the im-
plementation of carrier synchronization and the successful fabrication of key com-
ponents like the optical hybrids [38]. Within this book, both coherent receivers and
direct detection receivers are considered.
Exploding Data Traffic and Bandwidth Demands
In today’s communication society, the need for optical data transmission capacity
is exploding. The Internet is evolving into a multimedia broadband Internet with
services like video on demand, video telephony, online gaming, remote working and
interactive entertainment. The number of broadband subscribers is growing daily,
and future network scenarios plan to provide broadband access with 100 Mbit/s to
the home and several Gbit/s to the office via copper, wireless and all-optic solutions,
as depicted in Fig. 1.3. From the year 2000 on, the global network traffic has been
dominated by internet protocol (IP) data. Studies of traffic in 2005 indicated a global
traffic growth of 115% per year. Assuming a traffic growth of just 50%, the currently
installed Atlantic capacity will be saturated in only a few years. Even when acting
on the rather conservative assumption of only 20% capacity growth, the estimated
ultra long-haul capacity per fiber pair will be 50-100 Tbit/s in 2025 [8].
With the aid of recent technological advancements, optical fiber transmission
capacities have steadily been increased and impressive records were achieved in re-
search laboratories. The highest single wavelength channel data rate was obtained in
the labs of the Heinrich-Hertz-Institut (HHI), where 2.56 Tbit/s could be transmitted
over 160 km dispersion managed fiber (DMF) by combining the techniques of opti-
cal time division multiplexing (OTDM), polarization division multiplexing (PDM)
1 Introduction 7
50 Tbit/s long distance traffic
Core Network WDM, IP
10 Gbit/s
to the
Metro Network company
WDM, IP
Access Network Business

Ring
LAN
WLAN
2.5 Gbit/s
to the office
100 Mbit/s
UMTS to the home
Fig. 1.3 Possible future network scenario (LAN: local area network, WLAN: wireless LAN,
UMTS: universal mobile telecommunications system)
and DQPSK modulation [50]. In [10], the record per fiber capacity of 25.6 Tbit/s
could be achieved by transmitting 160 wavelength channels with 160 Gbit/s per-
channel data rate in the C-band and the L-band over 240 km, employing PDM,
Raman amplification and RZ-DQPSK modulation. A record spectral efficiency of
4.2 bit/s/Hz was obtained in [57] by using a very narrow WDM channel grid of
25 GHz at a channel data rate of 114 Gbit/s. This was enabled by applying RZ-8PSK
modulation and PDM. A transparent optical transmission distance of 18000 km was
bridged by [24] in the laboratory with a capacity of 1.09 Tbit/s, and a distance of
13100 km by [3] with a capacity of 0.96 Tbit/s in a field trial. Moreover, a record
capacity-distance product of 41.82 Pbit/s·km could be achieved in [5] by transmit-
ting 164 PDM-QPSK channels with a channel data rate of 100 Gbit/s over a length
of 2550 km, while using Raman amplification and coherent detection. The current
records are summarized in Table 1.1.
Table 1.1 Current fiber optic transmission records (records are printed in bold face)
Cap. Dist. Cap. · Dist. Sp. Eff. Number of Ch. rate Year Ref.
(Tbit/s) (km) (Pbit/s · km) (bit/s/Hz) channels (Gbit/s)
2.56 160 0.41 - 1 2560 2006 [50]

25.6 240 6.14 3.2 160 160 2008 [10]
0.91 640 0.58 4.2 8 114 2008 [57]
1.09 18000 19.62 0.2 109 10 2003 [24]
0.96 13100 12.58 0.3 96 10 2004 [3]
16.4 2550 41.82 2.0 164 100 2008 [5]
8 1 Introduction
Intention, Scope and Outline of the Book
In spite of the brilliant experimental results and technological progress in optical

networking in the last few years, the capacities reached so far will not suffice to
handle projected future IP traffic. The introduction of optical high-order modulation
formats and coherent detection could be the next important step towards higher net-
work capacities and higher spectral efficiencies. However, today’s network operators
have to handle the steadily growing volume of data traffic in an economic manner.
They must be able to estimate whether these new technologies can be cost-efficiently
integrated in existing optical networks and also in which network segments they can
reasonably be deployed.
Catering to the increasing interest in optical systems with high-order modulation
and coherent detection, this book aims to give an insight into possible system con-
figurations, their complexity and practical feasibility, their system and transmission
properties, and an appropriate choice of parameters. It is intended for researchers in
the field of optical communications, as well as for system designers who would like
to learn about the setup and the properties of optical systems employing high-order
modulation. It helps answering the following questions:
What do the transmitters and receivers look like? In the first part, this book
gives a detailed functional description of transmitter and receiver structures possi-
ble for the generation and detection of optical PSK / DPSK and QAM signals. The
description of different transmitters in Chapter 2 provides all the information rel-
evant to their functioning and includes optical and electrical transmitter parts. It
contains detailed information such as the concrete assembly of the coders and level-
generators. Moreover, the electrical field for the optical transmitter output signals is
analytically derived and differences in signal characteristics are emphasized. Sub-
sequently, Chapter 3 illustrates different receiver concepts which can be employed
to detect optical high-order modulation signals—starting with optical frontends and
ending with electrical data recovery. Receivers are detailed for three different de-
tection schemes: direct detection, homodyne differential detection and homodyne
synchronous detection. The latter scheme gains more and more interest since high-
speed digital signal processing is newly available which enables carrier synchro-
nization to be accomplished through digital means. Therefore, a special focus is
brought to carrier recovery through digital phase estimation.
How great is the additional effort required by a system upgrade? When up-
grading a system to a more sophisticated modulation format, the additional effort
required is determined significantly by the enhanced complexity of the transmit-
ters and receivers. Based on the information provided in Chap. 2 and Chap. 3, the
complexity of different transmitter and receiver configurations is compared and dis-
cussed in Chapter 4.
How does the system react to transmission impairments? The second part of
the book deals with the system and transmission characteristics of high-order mod-
ulation formats. When the interest in alternative modulation formats came up again
1 Introduction 9
a few years ago, the main intention of the research was to obtain a higher robust-
ness against transmission impairments. The optimum modulation format should be
resilient against amplified spontaneous emission (ASE) noise and laser phase noise,
tolerant to CD and PMD, robust to fiber nonlinearities and inaccuracies in dispersion
maps, and unsusceptible to narrow optical filtering, thereby enabling high spectral
efficiency. However, in practice, every modulation format has its own strengths and
weaknesses. After a discussion of system simulation aspects in Chapter 5 and the
degradation effects of optical fiber transmission in Chapter 6, it is the aim of Chap-
ter 7 to show the individual strengths and drawbacks of high-order optical modula-
tion formats in relation to their particular transmission characteristics, namely noise
performance, the optimal receiver filter bandwidths, the linewidth requirements and
the CD and SPM tolerances. Phase modulation formats up to 16PSK / 16DPSK and
different QAM formats (Star 16QAM, Square 16QAM and Square 64QAM) are
characterized for various system configurations, considering nearly all transmitter
and receiver schemes described in the first part of the book.
Which transparent optical transmission lengths can be bridged? Chapter 7 is
restricted to the back-to-back case and to single-span systems without amplifier
noise on the link, whereas Chapter 8 discusses the performance and the distances
attainable for optical systems with high-order modulation for multi-span long-haul
transmission. The achievable transmission distances can not exactly be identified by
an isolated examination of the behavior of a format with respect to particular trans-
mission impairments, because degradation effects interact which each other during
transmission. In fact, the performance of a modulation format can be only evaluated
precisely for the specific system in which it is operated. Currently, the determina-
tion of attainable transmission distances for multi-span long-haul transmission using
high-order modulation formats represents an exciting field of research. Some recent
results of the research group of the author are presented in Chap. 8.
Which spectral efficiencies and network capacities can be obtained? The ap-
plication of high-order modulation formats promises an enhancement of spectral
efficiency and therefore an extension of network capacities. The expected increase
of spectral efficiency depends on the order of the modulation format and may be
about the ratio of the data rate to the symbol rate in systems operated in the linear
regime. However, the spectral efficiencies attainable in practice can only be deter-
mined through investigation of specific WDM systems, with particular attention to
channel filtering, crosstalk, and inter-channel nonlinearities. These issues are sub-
jects for future research and out of the scope of the current edition of this book
which covers mainly single-channel transmission.
Finally, Chapter 9 summarizes the major trends in system performance resulting

from migration to higher-order modulation formats obtained in the second part of
the book and points out goals of future research in this field.
10 1 Introduction
References
1. Barry, J.R., Kahn, J.M.: Carrier synchronization for homodyne and heterodyne detection of
optical quadriphase-shift keying. IEEE Journal of Lightwave Technology 10(12), 1939–1951
(1992)
2. Betti, S., et al.: Coherent Optical Communication Systems. John Wiley & Sons (1995)
3. Cai, J.X., et al.: RZ-DPSK field trial over 13,100 km of installed non slope-matched submarine
fibers. In: Proceedings of Optical Fiber Communication Conference (OFC), PDP34 (2004)
4. Camatel, S., et al.: 2-PSK homodyne receiver based on a decision driven architecture and a
sub-carrier optical PLL. In: Proceedings of Optical Fiber Communication Conference (OFC),
OTuI3 (2006)
5. Charlet, G., et al.: Transmission of 16.4Tbit/s capacity over 2550km using PDM QPSK mod-
ulation format and coherent receiver. In: Proceedings of Optical Fiber Communication Con-
ference (OFC), PDP3 (2008)
6. Chen, X., et al.: Distributed feedback fiber laser with a novel structure. In: Proceedings of
Optical Fiber Communication Conference (OFC), OME10 (2005)
7. Chiou, Y., Wang, L.: Effect of amplifier noise on laser linewidth requirements in long haul
optical fiber communication systems with Costas PLL receivers. IEEE Journal of Lightwave
Technology 14(10), 2126–2134 (1996)
8. Desurvire, E.: Optical communications in 2025. In: Proceedings of European Conference on
Optical Communication (ECOC), Mo2.1.3 (2005)
9. Freund, R., et al.: 30 Gbit/s RZ-8-PSK transmission over 2800 km standard single mode fibre
without inline dispersion compensation. In: Proceedings of Optical Fiber Communication
Conference (OFC), OMI5 (2008)
10. Gnauck, A.H., et al.: 25.6-Tb/s WDM transmission of polarization-multiplexed RZ-DQPSK
signals. IEEE Journal of Lightwave Technology 26(1), 79–84 (2008)
11. Hebebrand, C., et al.: Performance of electronic dispersion compensation for multi-level mod-
ulation formats using homodyne coherent detection. In: Proceedings of European Conference
on Optical Communication (ECOC), We3.P.80 (2006)
12. Kahn, J.M., Ho, K.P.: Ultimate spectral efficiency limits in DWDM systems. In: Proceedings
of OptoElectronics and Communications Conference. Yokohama, Japan (2002)
13. Kahn, J.M., Ho, K.P.: Spectral efficiency limits and modulation/detection techniques for
DWDM systems. IEEE Journal of Selected Topics in Quantum Electronics 10(2), 259–272
(2004)
14. Kahn, J.M., et al.: Heterodyne detection of 310-Mb/s quadriphase-shift keying using fourth-
power optical phase-locked loop. IEEE Photonics Technology Letters 4(12), 1397–1400
(1992)
15. Kaiser, G.: FTTX concepts and applications. John Wiley & Sons, Inc., Hoboken, New Jersey
(2006)
16. Kasai, K., et al.: A13 C2 H2 frequency-stabilized, polarization-maintained erbium fibre ring
laser with no frequency modulation. In: Proceedings of European Conference on Optical
Communication (ECOC), Th1.3.5 (2004)
17. Kazovsky, L.G.: Decision-driven phase-locked loop for optical homodyne receivers: Perfor-
mance analysis and laser linewidth requirements. IEEE Journal of Lightwave Technology
LT-3(6), 1238–1247 (1985)
18. Kazovsky, L.G.: Balanced phase-locked loops for optical homodyne receivers: Performance
analysis, design considerations, and laser linewidth requirements. IEEE Journal of Lightwave
Technology LT-4(2), 182–195 (1986)
19. Kazovsky, L.G.: Homodyne phase-shift-keying systems: Past challenges and future opportu-
nities. In: Proceedings of Optical Fiber Communication Conference (OFC), OTuL3 (2005)
20. Kikuchi, K.: Coherent detection of phase-shift keying signals using digital carrier-phase esti-
mation. In: Proceedings of Optical Fiber Communication Conference (OFC), OTuI4 (2006)
21. Koc, U., et al.: Digital coherent quadrature phase-shift-keying (QPSK). In: Proceedings of
Optical Fiber Communication Conference (OFC), OThI1 (2006)
References 11
22. Louchet, H., et al.: Improved DSP algorithms for coherent 16-QAM transmission. In: Pro-
ceedings of European Conference on Optical Communication (ECOC), Tu.1.E.6 (2008)
23. Ly-Gagnon, D.S., et al.: Unrepeated 210-km transmission with coherent detection and digital
signal processing of 20-Gb/s QPSK signal. In: Proceedings of Optical Fiber Communication
Conference (OFC), OTuL4 (2005)
24. Mollenhauer, L.F., et al.: Demonstration of 109×10G dense WDM over more than 18,000 km
using novel, periodic-group-delay-complemented dispersion compensation and dispersion-
managed solitons. In: Proceedings of European Conference on Optical Communication
(ECOC), Th4.3.4 (2003)
25. Nakazawa, M., et al.: Polarization-multiplexed 1 Gsymbol/s, 64 QAM (12 Gbit/s) coherent
optical transmission over 150 km with an optical bandwidth of 2 GHz. In: Proceedings of
Optical Fiber Communication Conference (OFC), PDP26 (2007)
26. Noe, R.: PLL-free synchronous QPSK polarization multiplex/diversity receiver concept with
digital I&Q baseband processing. IEEE Photonics Technology Letters 17(4), 887–889 (2005)
27. Norimatsu, S., et al.: An 8 Gb/s QPSK optical homodyne detection experiment using external-
cavity laser diodes. IEEE Photonics Technology Letters 4(7), 765–767 (1992)
28. Ohm, M.: Optical 8-DPSK and receiver with direct detection and multilevel electrical signals.
IEEE/LEOS Workshop on Advanced Modulation Formats pp. 45–46 (2004)
29. Ohm, M., Speidel, J.: Receiver sensitivity, chromatic dispersion tolerance and optimal receiver
bandwidths for 40 Gbit/s 8-level optical ASK-DQPSK and optical 8-DPSK. In: Proc. 6th
Conference on Photonic Networks, pp. 211–217. Leipzig, Germany (2005)
30. Olsson, N.A.: Lightwave systems with optical amplifiers. IEEE Journal of Lightwave Tech-
nology 7(7), 1071–1082 (1989)
31. Rohde, M., et al.: Robustness of DPSK direct detection transmission format in standard fiber
WDM systems. Electronics Letters 36, 1483–1484 (1999)
32. Rosenkranz, W.: Robust multi-level phase shift modulation in high-speed WDM transmission.
Proceedings of the SPIE 5625, 241–252 (2005)
33. Sakamoto, T., et al.: 50-km SMF transmission of 50-Gb/s 16 QAM generated by quad-parallel
MZM. In: Proceedings of European Conference on Optical Communication (ECOC), Tu.1.E.3
(2008)
34. Seimetz, M.: Untersuchung der Phasenmodulation als alternatives Übertragungsverfahren für
optische Netze in Wellenlängen-Multiplex-Technik (WDM-Netze). Diplomarbeit, Technische
Fachhochschule Berlin (2000)
35. Seimetz, M.: Optical receiver for reception of M-ary star-shaped quadrature amplitude modu-
lation with differentially encoded phases and its application. German patent, DE 10 2006 030
915.4 (2006)
36. Seimetz, M.: Performance of coherent optical Square-16-QAM-systems based on IQ-
transmitters and homodyne receivers with digital phase estimation. In: Proceedings of
NFOEC, NWA4 (2006)
37. Seimetz, M.: Laser linewidth limitations for optical systems with high-order modulation em-
ploying feed forward digital carrier phase estimation. In: Proceedings of Optical Fiber Com-
munication Conference (OFC), OTuM2 (2008)
38. Seimetz, M., Weinert, C.M.: Options, feasibility and availability of 2 × 4 90◦ -hybrids for
coherent optical systems. IEEE Journal of Lightwave Technology 24(3), 1317–1322 (2006)
39. Seimetz, M., et al.: Coherent RZ-8PSK transmission at 30Gbit/s over 1200km employing
homodyne detection with digital carrier phase estimation. In: Proceedings of European Con-
ference on Optical Communication (ECOC), vol. 3, pp. 265–266 (2007)
40. Seimetz, M., et al.: Transmission reach attainable for single-polarization and PolMux coherent
Star 16QAM systems in comparison to 8PSK and QPSK at 10Gbaud. In: Proceedings of
Optical Fiber Communication Conference (OFC), OTuN2 (2009)
41. Sekine, K., et al.: Proposal and demonstration of 10-Gsymbol/sec 16-ary (40 Gbit/s) opti-
cal modulation / demodulation scheme. In: Proceedings of European Conference on Optical
Communication (ECOC), We3.4.5 (2004)
42. Serbay, M., et al.: Experimental investigation of RZ-8DPSK at 3 × 10.7Gb/s. In: the 18th
Annual Meeting of the IEEE Lasers & Electro-Optics Society, WE3. Sydney, Australia (2005)
12 1 Introduction
43. Serbay, M., et al.: 42.8 Gbit/s, 4 bits per symbol 16-ary Inverse-RZ-QASK-DQPSK transmis-
sion experiment without Polmux. In: Proceedings of Optical Fiber Communication Confer-
ence (OFC), OThL2 (2007)
44. Spinnler, B., et al.: Chromatic dispersion tolerance of coherent optical communication systems
with electrical equalization. In: Proceedings of Optical Fiber Communication Conference
(OFC), OWB2 (2006)
45. Taylor, M.G.: Coherent detection method using DSP for demodulation of signal and subse-
quent equalization of propagation impairments. IEEE Photonics Technology Letters 16(2),
674–676 (2004)
46. Tonguz, O.K., Wagner, R.E.: Equivalence between preamplified direct detection and hetero-
dyne receivers. IEEE Photonics Technology Letters 3(9), 835–837 (1991)
47. Tsukamoto, S., et al.: Coherent demodulation of optical 8-phase shift-keying signals using
homodyne detection and digital signal processing. In: Proceedings of Optical Fiber Commu-
nication Conference (OFC), OThR5 (2006)
48. Tsukamoto, S., et al.: Optical homodyne receiver comprising phase and polarization diver-
sities with digital signal processing. In: Proceedings of European Conference on Optical
Communication (ECOC), Mo4.2.1 (2006)
49. Walklin, S., Conradi, J.: Multilevel signaling for increasing the reach of 10Gb/s lightwave
systems. IEEE Journal of Lightwave Technology 17(11), 2235–2248 (1999)
50. Weber, H.G., et al.: Single channel 1.28Tbit/s and 2.56Tbit/s DQPSK transmission. Electron-
ics Letters 42(3) (2006)
51. Winzer, P., Gnauck, A.H.: 112-Gb/s polarization-multiplexed 16-QAM on a 25-GHz WDM
grid. In: Proceedings of European Conference on Optical Communication (ECOC), Th.3.E.5
(2008)
52. Winzer, P.J., Essiambre, R.J.: Advanced optical modulation formats. Proceedings of the IEEE
94(5), 952–985 (2006)
53. Wree, C., et al.: Differential quadrature phase-shift keying for cost-effective doubling of
the capacity in existing WDM systems. In: Proc. 4th Conference on Photonic Networks,
pp. 161–168. Leipzig, Germany (2003)
54. Yoon, H., et al.: Performance comparison of optical 8-ary differential phase-shift keying sys-
tems with different electrical decision schemes. Optics Express 13(2), 371–376 (2005)
55. Yu, J., et al.: 17 Tb/s (161x114 Gb/s) PolMux-RZ-8PSK transmission over 662km of ultra-low
loss fiber using C-band EDFA amplification and digital coherent detection. In: Proceedings of
European Conference on Optical Communication (ECOC), Th.3.E.2 (2008)
56. Zhao, J., et al.: Analytical investigation of optimization, performance bound, and chromatic
dispersion tolerance of 4-amplitude-shifted-keying format. In: Proceedings of Optical Fiber
Communication Conference (OFC), JThB15 (2006)
57. Zhou, X., et al.: 8x114Gb/s, 25-GHz-spaced, PolMux-RZ-8PSK transmission over 640km of
SSMF employing digital coherent detection and EDFA-only amplification. In: Proceedings of
Chapter 2
Transmitter Design
Abstract This chapter gives a detailed overview of how optical high-order mod-
ulation signals are generated. It describes transmitters for the generation of opti-
cal ASK-signals, DPSK-signals and QAM-signals and considers star-shaped and
square-shaped QAM constellations (Star QAM and Square QAM). Since all the
transmitters are composed of fundamental key components (laser, modulators, pulse
carver), the functionality of these components is discussed at the beginning of this
chapter. The subsequent description of the different transmitters includes optical
transmitter parts as well as electrical parts. It contains much detailed information
such as the concrete assembly of the coders and level-generators. The quality of the
transmitter output signals has a critical influence on the overall system performance.
Therefore the electrical field for the optical output signals is analytically derived for
all transmitters shown, and differences in signal characteristics, symbol transitions
and chirp behavior are emphasized. This helps us understand the system behavior
of the transmitters discussed later on in Chap. 7.
2.1 Transmitter Components
The following subsections briefly describe some fundamental key components of

optical transmitters for high-order modulation.
2.1.1 Lasers
The ideal carrier for optical transmission is a lightwave with constant amplitude,
frequency and phase. In practice, however, this perfect carrier can unfortunately
not be generated. In the last decades, optical light sources have been increasingly
improved. Light emitting diodes (LED) with very high spectral widths of several
tens of nanometers and low output powers and multi-mode Fabry-Perot lasers with


16 2 Transmitter Design
some discrete spectral lines and a total spectral width of several nanometers can now
be replaced with single-mode distributed feedback (DFB) lasers or external cavity
lasers (ECL) with linewidths in the sub-MHz region. When performing direct mod-
ulation (which means that the data is modulated onto the laser drive current), the
width of the emitted optical spectrum is determined by the incidental frequency
modulation of the laser under amplitude modulation, often referred to as chirping of
the laser, when the laser linewidth is small compared with the chirp-induced spec-
tral broadening. To avoid this effect, external modulation can be employed. Then
the laser acts as a continuous wave (CW) light source. Throughout this book, which
deals with advanced modulation and detection schemes, only single-mode lasers
and external modulation are considered.
The normalized electrical field of an ideal optical carrier as emitted from a CW

laser can be expressed in complex notation as
p
Ecw (t) = Ps · e j (ωs t+ϕs ) · es (2.1)
√
In (2.1), Ps represents the field amplitude, ωs /(2π ) the frequency, ϕs the initial
phase and es the polarization of the optical carrier. The character “s” indicates the
signal laser.
In practice, laser phase and amplitude noise, often called “intensity noise”, have
to be taken into account. They have their origin in spontaneous emission photons
which induce intensity fluctuations δ P(t) and phase fluctuations, which are repre-
sented by the signal laser phase noise ϕn s (t).
p
Ecw (t) = Ps + δ P(t) · e j (ωs t+ϕs +ϕns (t)) · es . (2.2)
The laser phase noise is caused by spontaneous emission photons, not generated
in phase with the stimulated emission photons but with random phase [7]. In the
time domain, the evolution of the actual phase can be understood as a random walk.
Within a time interval τ , the phase exhibits a random phase change of
1ϕn s (t) = ϕn s (t) − ϕn s (t − τ ). (2.3)
Since the phase changes 1ϕn s (t) are caused by a high number of independent noise
events—more precisely the generation of spontaneous emission photons—they can
be modeled as Gaussian distributed according to the Central Limit Theorem. Fur-
thermore, when assuming a white power spectral density of frequency noise ϕ̇n s (t),
which represents a realistic practical assumption [8, 13], the variance of the phase
change 1ϕn s (t) can be expressed as
2|τ |
h1ϕn2s (τ )i = Wϕ̇ns · |τ | = , (2.4)
tc
2.1 Lasers 17
where Wϕ̇ns is the constant power spectral density of the frequency noise and tc rep-
resents the coherence time which physically denotes the maximum delay difference
up to which two components of the emitted optical field can stably interfere. When
further neglecting intensity noise, the power spectral density of the optical field can
be shown to exhibit the Lorentzian-shaped spectrum
2tc Ps
W E cw (ω) = . (2.5)
1 + [(ω − ωs )tc ]2
The laser linewidth of the signal laser is defined as the full-width half-maximum
bandwidth of this power spectral density and is specified by
Wϕ̇ns 1
1νs = = . (2.6)
2π πtc
When the laser linewidth given by (2.6) is introduced into (2.4), the variance of the
phase change 1ϕn s (t) can be calculated by
h1ϕn2s (τ )i = 2π 1νs |τ |, (2.7)
showing that the phase uncertainty increases with the laser linewidth and the ob-
served time interval. A more detailed analysis of the mechanisms and statistics of
phase and frequency noise can be found in [13]. As will become clear later on in this
book, laser phase noise can have a limiting effect on system performance, especially
for high-order modulation formats with many phase states and when employing co-
herent synchronous detection.
Intensity noise can also lead to significant degradation in system performance,

in particular for coherent detection with high local oscillator (LO) laser powers
and when not implementing balanced detection [12, 18]. In the data sheets of laser
diodes, the relative intensity noise (RIN) is usually specified. The RIN, integrated
over a reference bandwidth 1 f , relates the variance of the intensity fluctuations to
the squared mean power:
Z 1f hδ P 2 (t)i
RI N( f ) d f = . (2.8)
0 hP(t)i2
The mean optical power hP(t)i is equivalent to the signal laser output power Ps or
the output power of the LO laser Plo , respectively, because hδ P(t)i = 0. In a sim-
ple approach, the intensity fluctuations can be modeled with Gaussian statistics and
a white noise spectrum [2]. In reality, the RIN has more complex spectral charac-
teristics, as can be observed for instance from [13, 20]. RIN values of laser diodes
typically range from -160 dB/Hz to -130 dB/Hz.
2.1.2 External Optical Modulators
The optical part of high-order modulation transmitters is composed of one or more

fundamental external optical modulator structures, which are briefly described in
this subsection: the phase modulator (PM), the Mach-Zehnder modulator (MZM)
and the optical IQ modulator (IQM). The speed attainable as well as the character-
istics of the transmitter output signals depends on the properties of the technology
and materials used for the modulators.
An optical phase modulator can be fabricated as an integrated optical device by

embedding an optical waveguide in an electro-optical substrate, mostly Li N bO3 ,
see Fig. 2.1a. By utilizing the fact that the refractive index of a material, and thus
the effective refractive index n e f f of the waveguide, can be changed by applying an
external voltage via a coated electrode, the electrical field of the incoming optical
carrier can be modulated in phase [20].
a u1 (t )
electro-optic substrate b
u(t )
Ein (t ) Eout (t ) Ein (t ) Eout (t )
waveguide electrode
u2 (t )
Fig. 2.1 a Integrated optical phase modulator. b Integrated optical Mach-Zehnder modulator.
Phase modulation ϕ P M (t) is a function of the wavelength λ, the length of the

electrode lel (interaction length) and the change of the effective refractive index
1n e f f (t). When solely considering the Pockels effect [20], the change of the re-
fractive index can be assumed to be linear to the applied external voltage u(t).
2π
ϕ P M (t) = · 1n e f f (t) · lel ∼ u(t) (2.9)
λ
In the specifications, the necessary driving voltage for achieving a phase shift of π,
denoted as Vπ , is typically given. Thus, the relation of the incoming optical carrier
E in (t) and the outgoing phase modulated optical field E out (t), when neglecting the
constant optical phase shift of the modulator, can be expressed as
u(t)
E out (t) = E in (t) · e jϕ P M (t) = E in (t) · e j Vπ π . (2.10)
2.1 Modulators 19
By utilizing the principle of interference, the process of phase modulation can

also be used to cause an intensity modulation of the optical lightwave, when the in-
terferometric structure shown in Fig. 2.1b is employed. This represents a dual-drive
Mach-Zehnder modulator. In the case of dual-drive MZMs, the phase modulators
in both arms can be driven independently, in contrast to single-drive MZMs. The
incoming light is split into two paths, both equipped with phase modulators. After
acquiring some phase differences relative to each other, the two optical fields are
recombined. The interference varies from constructive to destructive, depending on
the relative phase shift. Without considering the insertion loss, the transfer function
of a MZM is given by
E out (t) 1
= · e jϕ1 (t) + e jϕ2 (t) . (2.11)
E in (t) 2
In (2.11), ϕ1 (t) and ϕ2 (t) represent phase shifts in the upper and lower arms of the
MZM. For a specified driving voltage to obtain a phase shift of π in the upper and
lower arms, Vπ1 and Vπ2 , respectively, and with the driving voltages u 1 (t) and u 2 (t)
as defined in Fig. 2.1b, phase shifts are related to the driving signals with
u 1 (t) u 2 (t)
ϕ1 (t) = π, ϕ2 (t) = π. (2.12)
Vπ1 Vπ2
When operating the MZM in the push-push mode, which means that an identical
phase shift ϕ(t) = ϕ1 (t) = ϕ2 (t) is induced in both arms (for instance with u 1 (t) =
u 2 (t) = u(t) and Vπ1 = Vπ2 = Vπ ), a pure phase modulation is achieved, so
that the relation between the electrical input and output field is given by (2.10)
as for the simple PM. On the other hand, when one arm gets the negative phase
shift of the other arm (ϕ1 (t) = −ϕ2 (t), e.g. with u 1 (t) = −u 2 (t) = u(t)/2 and
Vπ1 = Vπ2 = Vπ ), the MZM is operated in the push-pull mode and a chirp-free
amplitude modulation is obtained. The input and output fields are then related with

1ϕ M Z M (t) u(t)
E out (t) = E in (t) · cos = E in (t) · cos π , (2.13)
2 2Vπ
where 1ϕ M Z M (t) = ϕ1 (t) − ϕ2 (t) = 2ϕ1 (t) is the induced phase difference be-
tween the fields of the upper and lower arm. By squaring (2.13), the power transfer
function of the MZM is obtained:

Pout (t) 1 1 1 1 u(t)
= + · cos (1ϕ M Z M (t)) = + · cos π . (2.14)
Pin (t) 2 2 2 2 Vπ
It should be noted that u(t) was defined in a way that u(t) = Vπ induces a phase
shift of π for the PM as well as a phase shift of π in the power transfer function of
the MZM when it is operated in the push-pull mode.
In Fig. 2.2, two different MZM operation principles are illustrated. For achieving
modulation in intensity, the MZM can be operated at the quadrature point, with a
DC bias of −Vπ /2 and a peak-to-peak modulation of Vπ (see Fig. 2.2a). When the
MZM is operated at the minimum transmission point (see Fig. 2.2b), with a DC
bias of −Vπ and a peak-to-peak modulation of 2Vπ , a phase skip of π occurs when
crossing the minimum transmission point. This becomes apparent from the field
transfer function. This way, the MZM can be used for binary phase modulation and
for modulation of the field amplitude in each branch of an optical IQ modulator.
a b
Operating the MZM at the quadrature point Operating the MZM at the minimum transmission point
1 1
OP
0 0
OP
Vπ 2V π
Field transfer function Field transfer function

Power transfer function Power transfer function
-1 -1
-2Vπ -Vπ 0 Vπ 2Vπ -2Vπ -Vπ 0 Vπ 2Vπ
u(t) u(t)
Fig. 2.2 Operating the MZM in the quadrature point (a) and the minimum transmission point (b)
Mach-Zehnder modulators can be implemented in Lithium Niobate (Li N bO3 ),

Gallium Arsenide (Ga As) and Indium Phosphide (I n P) [23]. Typical Vπ driving
voltages range from approximately 3 V to about 6 V.
A third fundamental modulator structure is the optical IQ modulator. It can be

composed of a PM and two MZMs, and is commercially available in an integrated
form [1]. As illustrated in Fig. 2.3a, the incoming light is equally split into two arms,
the in-phase (I ) and the quadrature (Q) arm. In both paths, a field amplitude modu-
lation is performed by operating the MZMs in the push-pull mode at the minimum
transmission point. Moreover, a relative phase shift of π/2 is adjusted in one arm,
for instance by an additional PM. This way, any constellation point can be reached
in the complex IQ-plane after recombining the light of both branches (see Fig. 2.3b).
Within the IQ modulator pictured in Fig. 2.3a, the induced phase differences of the
MZMs in the upper and lower paths are
u I (t) u Q (t)
1ϕ I (t) = π, 1ϕ Q (t) = π. (2.15)
Vπ Vπ
When neglecting any insertion loss and setting the driving voltage of the PM to
u P M = −Vπ /2, the field transfer function of the IQM can be expressed as

E out (t) 1 1ϕ I (t) 1 1ϕ Q (t)
= cos + j cos . (2.16)
E in (t) 2 2 2 2
2.1 Pulse Carving 21
a Optical IQ modulator b Principle of IQ modulation

uI (t ) Q
Amplitude
Modulation jϕIQM (t)
in the Q (t) ⋅ e
Q-arm
Ein (t ) Eout (t )
I
Amplitude
Reachable
Modulation
Signal
in the
Space
I-arm
uPM = −Vπ / 2 uQ (t )
Fig. 2.3 a Optical IQ modulator. b Principle of IQ modulation.
By using (2.15) and (2.16), the amplitude modulation a I Q M (t) and the phase mod-
ulation ϕ I Q M (t), performed by the IQM, can be calculated by
s
E out (t) 1 u I (t) u Q (t)
a I Q M (t) = = cos2 π + cos2 π , (2.17)
E in (t) 2 2Vπ 2Vπ

u I (t) u Q (t)
ϕ I Q M (t) = arg cos π , cos π . (2.18)
2Vπ 2Vπ
In (2.18), the arg [I, Q] operation denotes the calculation of the angle of a complex
value from the real and imaginary parts in the range between −π and π .
2.1.3 Pulse Carvers and Impulse Shapers
The shape of the transmitted optical pulses significantly affects the overall perfor-
mance of optical fiber transmission systems. The pulse shape used in most commer-
cial systems is NRZ, where a pulse filling the entire bit slot is transmitted for all
symbols with non-zero power. The power does not always go to zero when pass-
ing from one symbol to another. In the case of RZ pulses, the optical power goes
to zero within each symbol period. Therefore, power is smaller during the symbol
transitions and the undesired frequency modulation (chirp) arising during the phase
transitions can not take effect or is at least reduced, depending on the optical pulse
width and the rise time of the electrical driving signals.
Optical signals with RZ pulse shape can be created either by electronically gen-
erating RZ waveforms or by carving RZ pulses in the optical domain, using an extra
optical pulse carver. When employing the latter method, RZ pulses with a duty cy-
cle of 50% can be generated with a MZM, which is operated at the quadrature point
and driven with a sinusoidal electrical signal with a peak-to-peak amplitude of Vπ , a
frequency corresponding to the symbol rate r S = 1/TS and a phase offset of −π/2.
The electrical driving signal is given by u(t) = Vπ /2 · sin(2πt/TS − π/2) − Vπ /2,
where TS denotes the duration of one symbol.
The field transfer function of the optical RZ pulse carver for generating RZ
pulses with a duty cycle of 50% is defined as

E out (t) π t π π
= cos · sin 2π − − . (2.19)
E in (t) 4 TS 2 4
Even when employing optical pulse carving, the final optical pulse form at the
transmitter output depends also on the shape of the electrical driving signals. These
can be formed by electrical impulse shapers (IS) before feeding into the modulator
driving electrodes. In system simulations, electrical pulses without overshoots and
with specified rise times can be generated by filtering a rectangular input time func-
tion with a non-causal linear time invariant filter with the Gaussian shaped impulse
response
2 2
h(t) = √ · e−(2t/Te ) . (2.20)
π Te
The resulting output pulse of the electrical impulse shaper is given by the convolu-
tion of the rectangular signal with the impulse response h(t) and is specified by

1 2 (t − TS ) 2t
p(t) = · er f c − er f c . (2.21)
2 Te Te
In (2.20) and (2.21), Te represents the filter time constant, which can be approxi-
mately related to the electrical rise time 1t as
3
1t ≈ Te , (2.22)
4
as long it is assumed that the symbol time TS is much longer than the filter time
constant Te [2].
Having now discussed some fundamental components used in the various trans-
mitters for high-order modulation, some basics for multi-level signaling are briefly
presented in Sect. 2.2. Afterwards, the transmitters for particular modulation for-
mats are described in detail.
2.3 ASK Transmitters 23
2.2 Multi-Level Signaling
In digital optical transmission

with high-order modulation, m data bits, denoted here
as b1k , b2k , .., bm k , are collected and mapped to a complex symbol bk chosen from
an alphabet A of elements An (n = 1..M, M = 2m ). Each symbol bk can be inter-
q
preted as a complex phasor with the in-phase and quadrature coordinates bki and bk ,
respectively,
q
bk = bki + jbk , (2.23)
and with amplitude and phase states given by

q h i
2 q2 q
abk = bki + bk , ϕbk = arg bki , bk . (2.24)
One of the M = 2m symbols is assigned to each symbol interval (denoted by the

integer k, which has a range of 1 to ∞) of length TS = m · TB , where r B = 1/TB
is the data rate. The assignment of respective combinations of m bits to symbols
with particular amplitude and phase states (bit mapping) is defined in a so called
“constellation diagram”. For the best optical signal to noise ratio (OSNR) perfor-
mance, bit mapping should be arranged so that only one bit per symbol differs from
a neighboring symbol (Gray coding). The symbols are transmitted on the reduced
symbol rate r S = 1/TS = r B /m.
For the theoretical description of the electrical driving signals in Sect. 2.6, the
in-phase and quadrature symbol coordinates are scaled to unity, limiting the max-
imum coordinates of the I-axis and Q-axis to one. These normalized in-phase and
quadrature coordinates are denoted as i k and qk throughout this book. By scaling
q
the symbol coordinates bki and bk to unity, the normalized in-phase and quadrature
coordinates i k and qk can be expressed as
q
bki bk
ik = i
, qk = q , (2.25)
bmax bmax
i q
where bmax and bmax are given by
i q
bmax = max {|Re {An }|} , bmax = max {|I m {An }|} . (2.26)
n n
The relation between i k and qk and the data bits is specified by the bit mapping used
respectively and illustrated more precisely later on.
2.3 ASK Transmitters
The most simple optical multi-level signaling scheme is the M-ary ASK, where in-
formation is encoded into several intensity levels. The binary ASK (2ASK), usually
denoted as OOK, is the standard modulation format in commercially deployed opti-

cal transmission systems. The 2ASK constellation diagram defines only two symbol
points. Just one bit b1k is assigned to each symbol, as it is depicted in Fig. 2.4b. Fig-
ure 2.4a shows a 2ASK transmitter when performing external modulation.
a b q {b1}
Data IS
2ASK signal
MZM 0 1
CW MZM
RZ i
Pulse carving
Fig. 2.4 a 2ASK transmitter with external modulation. b 2ASK constellation diagram.
The optical part consists of a CW laser, an optional MZM for RZ pulse carving
and a MZM for intensity modulation, which is operated at the quadrature point. A
nice side effect of using a MZM for intensity modulation, is the nonlinear compres-
sion of the MZM transfer function at high and low transmission, which can suppress
ripples on electrical driving signals.
The electrical data signal can be formed by an impulse shaper as explained in
Sect. 2.1.3. The optical transmitter output signal for RZ-ASK, when neglecting laser
noises and polarization, can be described by

p u(t) π t π π
E s (t) = Ps · e j (ωs t+ϕs ) · cos π · cos sin 2π − − , (2.27)
2Vπ 4 TS 2 4
whereas the second cosine-term disappears in the case of NRZ. The electrical driv-
ing signal for 2ASK is defined as
X
u(t) = −Vπ + Vπ · b1k · p (t − kTS ) , b1k ∈ {0, 1} . (2.28)
k
High-order ASK formats have been investigated in [21] and [24] where it was
shown that they require high signal to noise ratios for direct detection, especially in
optically amplified links due to the intensity dependence of the signal-ASE noise. In
principle, they can be generated with the same optical transmitter. However, multi-
level electrical signals would have to be produced by an adequate electrical driving
circuit to drive the MZM.
The generation of multi-level electrical driving signals is quite challenging for
high data rates because the eye spreading increases when overlapping different bi-
nary electrical signals to create a multi-level signal, which leads to a degradation of
the system performance. The eye-spreading can be defined as 1e = (δ1 + δ2 )/de ,
2.4 DPSK Transmitters 25
where δ1 and δ2 describe the ripples (or the spreadings) in the upper and lower levels
and de is the height of the eye diagram, as shown in Fig. 2.5. For instance, if two
binary signals are summed to a quaternary signal, the eye spreading is increased by
a factor of three [6].
δ1
Eye spreading
de
Δ e = ( δ1 + δ 2 ) / d e
δ2
Fig. 2.5 Definition of eye spreading for a binary signal, based on [6]
2.4 DPSK Transmitters
Figure 2.6 shows constellation diagrams of different DPSK formats. All the con-
stellation points lie in one circle. Bit mapping can theoretically be chosen arbitrar-
ily. Here it is arranged in Gray code, so that only one bit per symbol differs from a
neighboring symbol, leading to the best noise performance.
q q
{b1} 11 {b1 , b2 }
1 0 10 01
i i
00
DBPSK DQPSK
q q
011 {b1 , b2 , b3 } 0111 0110 {b1 ,.., b4 }
0010
010 001 0101 0011
0100 0001
110 000 1100 0000

i i
1101 1000
111 100 1111 1001

Fig. 2.6 DPSK constellation 1011
101 1110 1010
diagrams with Gray coded bit
mappings 8DPSK 16DPSK
Basically, optical DPSK signals can be constituted by many different transmitter

types. Optical complexity can be reduced through increased electrical complexity
and vice versa. A single PM or MZM in the optical part would be sufficient to
generate arbitrary DPSK signals. However, multi-level electrical driving signals are
required for high-order DPSK formats in that case. Their generation increases the
electrical effort and is problematic due to the eye spreading problem. Another op-
tion is to use an optical IQ modulator alone. In this situation, the necessary number
of states of electrical driving signals corresponds to the number of projections of
the symbols to the I-axis and the Q-axis. From a practical point of view, the IQ
modulator is not the best choice for generating high-order DPSK signals because all
constellation points lie in one circle, and the distances between the signal states of
the in-phase and quadrature driving signals are short.
The discussion of DPSK transmitters within this book is restricted to config-

urations which require solely binary electrical driving signals. In Sect. 2.4.1 and
Sect. 2.4.2, respectively, two different configurations are presented which are de-
noted here as “serial DPSK transmitter” and “parallel DPSK transmitter”.
2.4.1 Serial DPSK Transmitter
One way of generating optical DPSK signals with binary electrical driving signals
is to use m consecutive PMs, where m is the number of bits per symbol. This trans-
mitter is shown in Fig. 2.7, and is called serial transmitter throughout this book.
After the first PM (phase shift π), a DBPSK signal is obtained. After the second PM
(phase shift π/2) a DQPSK signal is obtained, and so on.
d mk IS
bmk
Differential
Encoder
DEMUX
IS
1:m
Data
IS
b1k d1k IS
u PM1 (t ) u PM 2 (t ) u PM 3 (t ) u PM m (t )
CW MZM
PM PM PM PM
RZ
π π/2 π/4 π/2(m-1)

DBPSK DQPSK 8DPSK MDPSK
Fig. 2.7 DPSK transmitter with binary electrical driving signals, serial configuration
2.4 Parallel DPSK Transmitter 27
In the electrical part of the transmitter,the data signal is first parallelized with a
1:m demultiplexer. Parallelized data bits b1k , b2k , .., bm k are then fed into a dif-
ferential DPSK encoder, whose complexity and configuration generally depends on
the order of the DPSK modulation, the structure of the optical transmitter part, as
well as the used bit mapping of the data to the constellation points. The differential
encoding is performed to enable differential detection, or to resolve phase ambiguity
arising from carrier synchronization at the receiver (as described in Sect. 3.5). When
employing synchronous detection techniques without differential decoding, the dif-
ferential encoder can be omitted and only PSK signals generated. The functionality
of the differential encoders is discussed in more detail in Sect. 2.4.3.
At the differential encoder’s outputs, the encoded output data d1k , d2k , .., dm k
is obtained and passed to electrical impulse shapers and subsequently to the optical
modulators as illustrated in Fig. 2.7. The NRZ output signal of the serial DPSK
transmitter for a DPSK format with m bits per symbol is given by
p u P M (t) u P M (t) u P Mm (t)
1 π 2 π
E s (t) = Ps · e j (ωs t+ϕs ) · e j Vπ · ej Vπ · ... · e j Vπ π
, (2.29)
with the binary electrical driving voltages

Vπ X
u P Mn (t) = · dn k · p (t − kTS ) , (2.30)
2n−1
k
where n = {1, 2, .., m}, and dn k ∈ {0, 1} represents the n-th differentially encoded
bit of a symbol consisting of m bits in the k-th symbol interval.
2.4.2 Parallel DPSK Transmitter
A second DPSK transmitter configuration, which also uses binary electrical driving
signals, is composed of a combination of an optical IQ modulator and consecu-
tive phase modulators, in the following called parallel transmitter and depicted in
Fig. 2.8. The optical IQ modulator accomplishes a DQPSK modulation, and higher-
order DPSK signals are generated by the consecutive PMs.
The electrical transmitter part is identical to the one for the serial transmitter,
with the exception of the internal setup of the differential encoder. To accomplish
DQPSK modulation, the Mach-Zehnder modulators in the I-arm and the Q-arm of
the IQM are operated at the minimum transmission point and driven by binary elec-
trical driving signals
X
u I (t) = −2Vπ + 2Vπ · d1k · p (t − kTS ) , (2.31)
k
X
u Q (t) = −2Vπ + 2Vπ · d2k · p (t − kTS ) . (2.32)
k
d mk
bmk IS
Differential
Encoder
DEMUX
IS
1:m
Data
IS
b1k d1k IS
u I (t )
u PM 3 (t ) u PM m (t )
MZM
MZM
CW 3dB 3dB PM PM
RZ
-90° MZM
π/4 π/2(m-1)
DQPSK 8DPSK MDPSK
uQ (t )
Fig. 2.8 Parallel DPSK transmitter with binary electrical driving signals
The driving signals of the consecutive phase modulators are defined by (2.30) for
n = {3..m}. The optical output signal of the parallel DPSK transmitter for NRZ line
coding is specified by
p u P M (t) u P Mm (t)
3 π
E s (t) = Ps · e j (ωs t+ϕs ) · a I Q M (t) · e jϕ I Q M (t) · e j Vπ · ... · e j Vπ π
, (2.33)
where a I Q M (t) and ϕ I Q M (t) describe the amplitude and phase modulation of the
IQM, given by (2.17) and (2.18), respectively, and the parameter Vπ is assumed to
be the same for all the modulators used, for simplicity.
2.4.3 Differential Encoding

In the differential encoder, the data bits b1k , b2k , .., bm k , which are mapped to
symbols as defined by the original bit mapping, for instance the Gray coded bit
mapping in Fig. 2.6, are encoded in a way to represent phase differences. To achieve
this, appropriate absolute phase states ϕk must be adjusted at the encoder output
for given phase differences ϕbk and previously given absolute phase states ϕk−1 ac-
cording to ϕk = ϕk−1 + ϕbk . The symbol assignment at the encoder output, which
describes the mapping of the differentially encoded bits d1k , d2k , .., dm k into sym-
bols with absolute phase states ϕk , must be defined according to a particular optical
transmitter configuration in order to drive the optical modulators adequately to ob-
tain the desired absolute phase states.
Different symbol assignments are appropriate at the encoder output for the serial
and the parallel DPSK transmitter, as shown in Fig. 2.9. Therefore different encoders
are needed for each of the two configurations. For the serial transmitter, the symbol
assignment must be arranged in chronologically increasing order, as illustrated in
2.4 Differential Encoding 29
q q q
01 {d1 , d 2 } 010 {d1 , d 2 , d3 } 0101 0100 {d1 ,.., d 4 }
0011
011 001 0110 0010
0111 0001
10 00 100 000 1000 0000 Serial
i i i transmitter
1001 1111
101 111 1010 1110

11 110 1011 1101
1100
q q q
{d1 , d 2 } 111 {d1 , d 2 , d3 } 1111
1110 {d1 ,.., d 4 }
1101
01 11 010 110 0100 1100
0101 1011
011 101 0110 1010 Parallel

i i i transmitter
0111 1001
00 10 000 100 0000 1000

001 0001 0011
0010
DQPSK 8DPSK 16DPSK
Fig. 2.9 Symbol assignment to absolute phase states at the differential encoder output for the serial
DPSK transmitter (top) and the parallel DPSK transmitter (bottom)
the upper part of Fig. 2.9 for DQPSK, 8DPSK and 16DPSK, respectively. As regards
the parallel transmitter, the symbols must be assigned differently. For instance, when
driving the MZM in the I-arm and the Q-arm of the IQM, each with a logical one,
an optical phase of π/4 is obtained, in contrast to the serial configuration, where
an optical phase of 3/2 · π results for driving both PMs with a logical one. The
symbol assignment for the parallel transmitter used here is shown in the bottom part
of Fig. 2.9 for DQPSK, 8DPSK and 16DPSK, respectively. Because the absolute
phase is not relevant for differential detection, the symbol assignments can also be
arbitrarily rotated for both transmitters.
In any differential encoder, the data bits of the current symbol b1k , b2k , .., bm k ,
representing the current phase difference,
are combined with the previous encoder
output bits d1k−1 , d2k−1 , .., dm k−1 , representing the absolute optical phase of the
previous
symbol, in a logical circuit, in order to specify the next encoder output bits
d1k , d2k , .., dm k which define the current optical phase. The general structure of a
differential encoder is depicted in Fig. 2.10.
The following paragraphs derive logical relations which characterize the logical
circuit of the differential encoders for different DPSK transmitters. They are only
valid for employing the Gray coded bit mapping to phase differences and the symbol
assignment to absolute phases as defined in Fig. 2.6 and Fig. 2.9, respectively. Other
mappings are possible, but would yield other relations.
Fig. 2.10 General structure of

a differential encoder TS
TS
TS
d mk
bmk Logical
circuit d 2k
b2k d1k
b1k
Differential Encoders for DBPSK and DQPSK
In the case of DBPSK, differential encoding can be achieved easily. A logical one
in the current data indicates a phase change of π . By combining the current data bit
with the previous encoder output bit in a simple XOR gate, the next encoder output
bit is obtained.
The encoders required within DQPSK transmitters are yet more complex. They
have to provide two output bits d1k and d2k , which depend on the input data bits b1k
and b2k , as well as on the previous encoder output bits d1k−1 and d2k−1 . Within the
serial transmitter, the encoder output signals are taken to drive the two PMs. The
first PM changes the phase between 0 and π, and the second one changes the phase
between 0 and π/2. In this way, the four absolute phase states 0, π/2, π, and 3/2 · π
can be adjusted with a symbol assignment in chronologically increasing order. The
truth table for the DQPSK encoder of the serial transmitter is given by Table 2.1.
Table 2.1 Truth table for the differential encoder appropriate for the serial DQPSK transmitter
d1k d2k ϕk d1k−1 d2k−1 ϕk−1 b1k b2k ϕbk
0 0 0 0 0 0 0 1 0
0 0 0 0 1 π/2 0 0 3/2 · π
0 0 0 1 0 π 1 0 π
0 0 0 1 1 3/2 · π 1 1 π/2
0 1 π/2 0 0 0 1 1 π/2
0 1 π/2 0 1 π/2 0 1 0
0 1 π/2 1 0 π 0 0 3/2 · π
0 1 π/2 1 1 3/2 · π 1 0 π
1 0 π 0 0 0 1 0 π
1 0 π 0 1 π/2 1 1 π/2
1 0 π 1 0 π 0 1 0
1 0 π 1 1 3/2 · π 0 0 3/2 · π
1 1 3/2 · π 0 0 0 0 0 3/2 · π
1 1 3/2 · π 0 1 π/2 1 0 π
1 1 3/2 · π 1 0 π 1 1 π/2
1 1 3/2 · π 1 1 3/2 · π 0 1 0
2.4 Differential Encoding 31
By using Karnaugh maps, for instance, the logical relations for the encoder out-
put bits d1k and d2k can easily be derived from Table 2.1 for the serial transmitter:
d1k = b2k d1k−1 d2k−1 + b1k d1k−1 d2k−1 + b1k d1k−1 d2k−1 + b2k d1k−1 d2k−1 , (2.34)
d2k = b1k b2k d2k−1 + b1k b2k d2k−1 + b1k b2k d2k−1 + b1k b2k d2k−1 . (2.35)
In (2.34) and (2.35), “+” denotes logical OR, the particular terms are associated by
logical AND, and the “overlines” indicate logical negation. In practice, the encoder
can be implemented using the adequate AND and OR gates.
Due to the different symbol assignment at the encoder output, the differential
encoder of the parallel transmitter differs from the one of the serial transmitter. Its
truth table is shown in Table 2.2. The corresponding logical relations for the two
encoder output bits d1k and d2k for the parallel DQPSK transmitter are
d1k = b1k d1k−1 d2k−1 + b2k d1k−1 d2k−1 + b1k d1k−1 d2k−1 + b2k d1k−1 d2k−1 , (2.36)
d2k = b2k d1k−1 d2k−1 + b1k d1k−1 d2k−1 + b2k d1k−1 d2k−1 + b1k d1k−1 d2k−1 . (2.37)
Table 2.2 Truth table for the differential encoder appropriate for the parallel DQPSK transmitter
d1k d2k ϕk d1k−1 d2k−1 ϕk−1 b1k b2k ϕbk
0 0 5/4 · π 0 0 5/4 · π 0 1 0
0 0 5/4 · π 0 1 3/4 · π 1 1 π/2
0 0 5/4 · π 1 0 7/4 · π 0 0 3/2 · π
0 0 5/4 · π 1 1 π/4 1 0 π
0 1 3/4 · π 0 0 5/4 · π 0 0 3/2 · π
0 1 3/4 · π 0 1 3/4 · π 0 1 0
0 1 3/4 · π 1 0 7/4 · π 1 0 π
0 1 3/4 · π 1 1 π/4 1 1 π/2
1 0 7/4 · π 0 0 5/4 · π 1 1 π/2
1 0 7/4 · π 0 1 3/4 · π 1 0 π
1 0 7/4 · π 1 0 7/4 · π 0 1 0
1 0 7/4 · π 1 1 π/4 0 0 3/2 · π
1 1 π/4 0 0 5/4 · π 1 0 π
1 1 π/4 0 1 3/4 · π 0 0 3/2 · π
1 1 π/4 1 0 7/4 · π 1 1 π/2
1 1 π/4 1 1 π/4 0 1 0
Differential Encoders for Higher-Order DPSK Formats
In the same manner, one can specify differential encoders for the higher-order DPSK
transmitters. Starting from the constellation diagrams with bit mappings and symbol
assignments defined in Fig. 2.6 and Fig. 2.9, respectively, truth tables can be estab-
lished. Subsequently, the corresponding Karnaugh maps can be evaluated in order
to determine the logical relations of the encoders.
Since the equations resulting for 8DPSK and 16DPSK are quite bulky, they are
given in Appendix A. It becomes apparent here that the complexity of the differen-
tial encoder grows significantly with the increasing order of the phase modulation.
For 16DPSK, the differential encoder is very complex. For instance, the relation for
the encoder output bit d1k has 30 OR combined terms, each consisting of 4-7 AND
combined inputs.
Now having provided all the relevant functional information about the setup of
the serial and parallel DPSK transmitters, the following subsection discusses the
properties of their output signals.
2.4.4 Signal Properties
The structure of the transmitter affects the signal characteristics and the transmis-
sion properties of the generated optical DPSK signals, due to the fact that the symbol
transitions (amplitude and phase transitions) of the transmitters are different, espe-
cially in the case of NRZ. An optical high-order modulation signal at the transmitter
output can be generally described by
p
E s (t) = Ps · e j (ωs t+ϕs ) · a(t) · e jϕ(t) , (2.38)
where A(t) = a(t) · e jϕ(t) represents the normalized complex modulation enve-
lope of the optical signal with time dependent amplitude a(t) and phase ϕ(t). The
squared amplitude a 2 (t) times the CW laser power Ps represents the instantaneous
signal power (which is proportional to the signal intensity). Another important pa-
rameter, which has a significant influence on the transmission performance, is the
derivative of the optical phase ϕ̇(t) = dϕ(t)/dt. It is a measure for the undesired
frequency modulation occurring during the symbol transitions, usually denoted as
chirp.
The complex envelope of the DPSK transmitter output signals can be extracted
from (2.29) and (2.33). The normalized intensity eyes, the IQ diagrams (I m {A(t)}
versus Re {A(t)}, where a(t) is scaled here to unity for illustration purposes) and the
chirp characteristics are plotted for both transmitter types discussed above and both
pulse shapes for 8DPSK modulation in Fig. 2.11, assuming a data rate of 40 Gbit/s
and an electrical rise time of 1/4 of the symbol duration.
2.4 DPSK Transmitters - Signal Properties 33
Serial transmitter, NRZ Serial transmitter, RZ Parallel transmitter, NRZ Parallel transmitter, RZ
1 1 1 1
Norm. Intensity
Norm. Intensity
Norm. Intensity
Norm. Intensity
0.8 0.8 0.8 0.8
Intensity
Intensity
Intensity
Intensity
0.6 0.6 0.6 0.6
0.4 0.4 0.4 0.4
0.2 0.2 0.2 0.2
0 0 0 0
0.372 Time 1.122 0.372 1.122 0.372 1.122 0.372 1.122
Time Time Time
1 1 1 1
Im [A(t)]
Im [A(t)]
Im [A(t)]
Im [A(t)]
0 0 0 0
-1 -1 -1 -1
-1 0 1 -1 0 1 -1 0 1 -1 0 1
Re [A(t)] Re [A(t)] Re [A(t)] Re [A(t)]
11 11 11 11
x 10 x 10 x 10 x 10
5 5 5 5
2.5 2.5 2.5 2.5

Chirp (1/s)
Chirp (1/s)
Chirp (1/s)
Chirp (1/s)
0 0 0 0
-2.5 -2.5 -2.5 -2.5
-5 -5 -5 -5
0.372 Time 1.122 0.372 Time 1.122 0.372 Time 1.122 0.372 Time 1.122
11 11 11 11
x 10 x 10 x 10 x 10
5 5 5 5
Chirp ⋅ Norm. Int. (1/s)

2.5 2.5 2.5 2.5
0 0 0 0
-2.5 -2.5 -2.5 -2.5
-5 -5 -5 -5
0.372 Time 1.122 0.372 Time 1.122 0.372 Time 1.122 0.372 Time 1.122
Fig. 2.11 Optical signal properties of different 8DPSK transmitters
When the serial transmitter and NRZ pulse shape are employed, symbol transi-
tions are conducted on circles, and there is constant power during phase changes.
Chirp appears during the symbol transitions, and its magnitude depends on the
steepness of the phase jumps. An intuitive measure for the disturbing effect of the
chirp during transmission is the product of the chirp and the intensity as is shown in
the bottom eye diagrams of Fig. 2.11. It becomes apparent that the chirp has a strong
effect for the serial transmitter and NRZ pulse shape, due to permanent full power.
In the case of RZ, there is almost no optical power during phase changes, and only
a very small residual impact of the chirp can be determined. For the parallel trans-
mitter, the impact of the chirp is reduced even for the NRZ pulse shape. The symbol
transitions are different due to the usage of an IQ modulator, and power is reduced
for some transitions (intensity dips). The chirp shows high peaks when the point
of origin is crossed in the IQ diagram (the phase jumps abruptly by ±π). However
this is not problematic because no optical power exists at this moment. The prod-
uct of the chirp and the intensity is clearly reduced compared with the serial NRZ
transmitter, so a better transmission performance can be expected.
2.5 Star QAM Transmitters
When compared with pure phase modulation, combined phase and amplitude modu-
lation (quadrature amplitude modulation, QAM) exhibits a reduced number of phase
states for the same number of symbols. The constellation points can be arranged in
a square (Square QAM formats) or they can lie on multiple circles (Star QAM for-
mats). The phases are arranged with equal spacing for Star QAM formats, as shown
for Star 16QAM in Fig. 2.13, so the phase difference of any two symbols corre-
sponds to a phase state defined in the constellation diagram and phase information
can be differentially encoded as for DPSK formats. Thus, Star QAM signals with
differentially encoded phases are suitable to be detected by receivers with differen-
tial detection. By contrast, Square QAM signals are conveniently detected by coher-
ent synchronous receivers, but can also be detected by differential detection when
phase pre-integration is employed at the transmitter [9].
To accomplish the generation of Star QAM signals with differentially encoded
phases, the same equipment can be used as for DPSK transmitters just described.
The DPSK transmitters—in serial or parallel configuration—only have to be ex-
tended by an additional MZM for intensity modulation, to be able to place symbols
on different intensity rings.
In principle, arbitrary Star QAM constellations are possible. The Star 8QAM
format (2ASK-DQPSK) was investigated in [11]. A Star 8QAM transmitter can be
composed of a DQPSK transmitter followed by an additional MZM and can use
the same differential encoders as a DQPSK transmitter. The transmitter for Star
16QAM (2ASK-8DPSK) consists of an 8DPSK transmitter, extended by a MZM
for intensity modulation, as shown in Fig. 2.12 for the serial configuration.
b4k IS
d 3k
DEMUX
Differential
IS
Encoder
8DPSK
1:4
Data d 2k
IS
b1k d1k IS
u PM1 (t ) u PM 2 (t ) u PM 3 (t ) u IM (t )
CW MZM
PM PM PM MZM
RZ
π π/2 π/4
Star 16QAM
Fig. 2.12 Optical Star 16QAM transmitter with differential phase encoding, serial configuration
Figure 2.13 illustrates the Star 16QAM constellation diagrams with Gray coded
bit mapping (Fig. 2.13a) and symbol assignments at the encoder output for the serial
and parallel transmitters (Fig. 2.13b and Fig. 2.13c).
2.5 Star QAM Transmitters 35
a q b q c q
0111 {b1 , b2 , b3 , b4 } 0101 {d1 , d 2 , d3 , b4 } 1111 {d1 , d 2 , d 3 , b4 }
0101 0011 0111 0011 0101 1101
0110 0100 1110
0100 0010 0110 0010 0100 1100
r1 r1 r1
1101 1100 0000 0001 1001 1000 0000 0001 0111 0110 1010 1011
r2 i r2 i r2 i
1110 1000 1010 1110 0000 1000
1010 1100 0010
1111 1001 1011 1111 0001 1001
1011 1101 0011
Fig. 2.13 Star 16QAM Gray coded original bit mapping (a) and symbol assignments at the encoder
output for the serial transmitter (b) as well as the parallel transmitter (c)
As for 16DPSK, four bits are mapped to one symbol. However, high spectral
efficiency can be obtained here without using the very complex differential 16DPSK
encoder. Instead, 8DPSK encoders can be employed. When compared with 8DPSK,
the constellation diagram consists of a second circle with eight more symbols. The
fourth bit b4k indicates if a symbol belongs to the inner or the outer circle and is
used to drive the additional MZM.
One degree of freedom which can optimize the OSNR performance for Star
QAM formats with only two amplitude states, is the ring ratio R R = r2 /r1 , where
r1 and r2 are the amplitudes of the inner and outer circle, respectively, as illustrated
in Fig. 2.13. The influence of the ring ratio on the OSNR performance is discussed
later on in the second part of this book.
Another Star QAM constellation with 16 symbols, composed of four amplitude
and four phase states (4ASK-DQPSK), has been investigated in [19]. The definition
of only four phase states has the advantage that data recovery is easier to accomplish
for differential detection. On the other hand, the use of more than two amplitude
states leads to high OSNR requirements for intensity detection and therefore to a
poor overall OSNR performance.
Generally, when using the serial NRZ-Star QAM transmitter, an optical Star
QAM signal with only two amplitude states can be mathematically described by
u P M (t) uPM (t)
p j (ωs t+ϕs ) j Vπ1 π j m−1 π u I M (t)
E s (t) = Ps · e ·e · ... · e V π · cos π . (2.39)
2Vπ
In (2.39), the phase modulator driving signals u P M1 (t)...u P Mm−1 (t) are again de-
fined by (2.30) with n = {1..(m − 1)}. The electrical driving signal for intensity
modulation depends on the desired ring ratio, and is specified by

2 arccos R1R 2 arccos R1R X
u I M (t) = − · Vπ + · Vπ · bm k · p (t − kTS ) , (2.40)
π π
k
where bm k ∈ {0, 1} corresponds to the last data bit of a symbol with m bits in the
k-th symbol interval.
Similarly, the last phase modulator of the parallel DPSK transmitter, whose out-
put signal is given by (2.33), can be replaced with an intensity modulator to obtain
the parallel Star QAM transmitter. Its output signal is then given by
p
E s (t) = Ps · e j (ωs t+ϕs ) · a I Q M (t) · e jϕ I Q M (t)
u P M (t) uPM (t)
j Vπ3 π j m−1 π u I M (t)
·e · ... · e V π · cos π . (2.41)
2Vπ
In Fig. 2.14, the eye diagrams of the normalized intensity, the IQ diagrams, and
the chirp characteristics are depicted for the serial and the parallel Star 16QAM
transmitter, considering NRZ and RZ pulse shapes and assuming a ring ratio of 1.8,
a data rate of 40 Gbit/s and an electrical rise time of 1/4 of the symbol period.
Serial transmitter, NRZ Serial transmitter, RZ Parallel transmitter, NRZ Parallel transmitter, RZ
1 1 1 1
Norm. Intensity
Norm. Intensity
Norm. Intensity
Norm. Intensity
0.8 0.8 0.8 0.8
Intensity
Intensity
Intensity
Intensity
0.6 0.6 0.6 0.6
0.4 0.4 0.4 0.4
0.2 0.2 0.2 0.2
0 0 0 0
0.5 Time 1.5 0.5 1.5 0.5 1.5 0.5 1.5
Time Time Time
1 1 1 1
Im [A(t)]
Im [A(t)]
Im [A(t)]
Im [A(t)]
0 0 0 0
-1 -1 -1 -1
-1 0 1 -1 0 1 -1 0 1 -1 0 1
11 11 11 11
x 10 x 10 x 10 x 10
5 5 5 5
2.5 2.5 2.5 2.5

Chirp (1/s)
Chirp (1/s)
Chirp (1/s)
Chirp (1/s)
0 0 0 0
-2.5 -2.5 -2.5 -2.5
-5 -5 -5 -5
0.5 1.5 0.5 1.5 0.5 1.5 0.5 1.5
Time Time Time Time
11 11 11 11
x 10 x 10 x 10 x 10
5 5 5 5
2.5 2.5 2.5 2.5
0 0 0 0
-2.5 -2.5 -2.5 -2.5
-5 -5 -5 -5
0.5 1.5 0.5 1.5 0.5 1.5 0.5 1.5
Time Time Time Time
Fig. 2.14 Optical signal properties of different Star 16QAM transmitters
The same conclusions can be drawn as for 8DPSK. The impact of the chirp,
which appears at the phase transitions, is reduced when the parallel configuration
is employed, and is almost eliminated when RZ pulses are transmitted. This be-
comes apparent from the product of the chirp and the intensity (bottom diagrams in
Fig. 2.14), introduced as an intuitive performance measure in Sect. 2.4.4.
2.6 Differential Quadrant Encoding 37
2.6 Square QAM Transmitters
In Star QAM constellations described in the last section, first suggested by Cahn in
1960 [3], the same number of symbols is placed on different concentric circles. Star
QAM signals can easily be generated by enhancing a phase modulation transmit-
ter for an additional intensity modulation and can be differentially detected. On the
other hand, these constellations are not optimal as regards noise performance, be-
cause symbols on the inner ring are closer together than symbols on the outer ring.
In order to improve noise performance, Hancock and Lucky suggested placing more
symbols on the outer ring than on the inner ring [5], leading to constellations with
more balanced Euclidean distances. But they came to the conclusion that such sys-
tems are more complicated to implement. In 1962, the Square QAM constellation,
shown in Fig. 2.15 for Square 16QAM, was introduced for the first time by Cam-
popiano and Glazer [4]. Indeed, Square QAM signals are conveniently detected by
coherent synchronous receivers and offer only a small improvement in noise perfor-
mance, but—thinking in terms of two quadrature carriers—relatively simple mod-
ulation and demodulation schemes are possible, due to the regular structure of the
constellation projected on the in-phase and quadrature axis. Today, the Square QAM
is widely used in electrical systems. In optical transmission systems, however, it is
still very distant from a commercial practical implementation.
The next sections illustrate different transmitter options for generating optical
Square QAM signals, which are denoted here as “serial Square QAM transmitter”,
“conventional IQ transmitter”, “enhanced IQ transmitter”, “Tandem-QPSK trans-
mitter” and “multi-parallel MZM transmitter”. Each of these transmitters features
different properties of its output signals and different complexities of its optical and
electrical parts, which can be traded off. Detailed information about the electrical
parts is provided, especially for two particular modulation formats: Square 16QAM,
to which a special focus is brought in this book, and Square 64QAM, which is in-
cluded here as a very ambitious format. Before going into the details of the transmit-
ters, Sect. 2.6.1 outlines the differential quadrant encoding which must be employed
for all transmitter configurations when the quadrant ambiguity arising at the carrier
synchronization at the receiver shall be resolved through differential coding.
2.6.1 Differential Quadrant Encoding
A n times π/2 (n = 0, 1, 2, 3) phase ambiguity (quadrant ambiguity) arises at the

carrier synchronization for synchronous detection of Square QAM signals, which
can be resolved by so-called differential quadrant encoding. This encoding scheme
combines a DQPSK encoding with a specific bit mapping, and was proposed by We-
ber in [22]. Two of the m bits of a Square QAM symbol determine the quadrant and
are differentially encoded using a DQPSK differential encoder. This way, these two
bits can be unambiguously recovered even if the absolute position of the received
constellation diagram is ambiguous with n times π/2. The remaining (m − 2) bits

can also be determined correctly for any n times π/2 rotation when the bit mapping
is arranged as rotation symmetric with respect to these bits. As an undesired con-
sequence, the bit mapping is then not further Gray coded, leading to an OSNR per-
formance degradation compared with Gray coded bit mappings. Gray coded Square
QAM signals can be received for instance when sending training sequences.
a b
{b1 , b2 , b3 , b4 } {d1 , d 2 , b3 , b4 }
q q
1111 1110 0101 0111 0111 0110 1101 1111
1101 1100 0100 0110 0101 0100 1100 1110
i i
1010 1000 0000 0001 0010 0000 1000 1001
1011 1001 0010 0011 0011 0001 1010 1011
Fig. 2.15 a Bit mapping used for Square 16QAM, appropriate for differential quadrant encoding.
b Symbol assignment after the DQPSK differential encoder.
Figure 2.15a shows a non Gray coded Square 16QAM bit mapping which is
appropriate for differential quadrant encoding. It can be observed that any rotation
of n times π/2 (n = 0, 1, 2, 3) causes no difference to the last two bits b3k and
b4k . The first two bits, b1k and b2k , determine the quadrant and are encoded with
a DQPSK differential encoder. When using the differential encoder of the parallel
DQPSK transmitter described in Sect. 2.4.3, the differentially encoded bits d1k and
d2k at the encoder output are assigned to the quadrants as illustrated in Fig. 2.15b.
2.6.2 Serial Square QAM Transmitter
In contrast to Star QAM constellations, the phases are arranged unequally spaced in
Square QAM constellations, so that it is not possible to adjust all the phase states of
the symbols by simply driving consecutive phase modulators with binary electrical
driving signals. In [6], it was shown that any optical QAM signal can be generated by
using a single dual-drive MZM (see Fig. 2.1b). In this case, however, the necessary
number of states of electrical driving signals is quite high (e.g. 16-ary driving signals
are needed for Square 16QAM), and a big electrical effort has to be engaged in to
enable the simplicity of the optical part.
2.6 Serial Square QAM Transmitter 39
Another transmitter with a simple optical part capable of creating any QAM con-
stellation is constituted by only two consecutive optical modulators: a MZM for
adjustment of the amplitude state and a consecutive PM to set the phase. This trans-
mitter is denoted as serial Square QAM transmitter throughout this book and is
shown in Fig. 2.16. One more MZM can be employed for RZ pulse carving. The
simplicity of the optical receiver part necessitates the use of a complex electrical
level generator, since electrical driving signals with a high number of states must be
generated (12-ary electrical driving signals are required for phase modulation in the
case of Square 16QAM, for instance).
Differential
b1k d1k
Encoder
DQPSK
b2k d 2k Generator
DEMUX
IS
Level
1:m
Data b3k
bmk IS
uIM (t ) u PM (t )
MZM
CW MZM PM
RZ
Square QAM
Fig. 2.16 Square QAM transmitter, serial configuration
The optical output signal of the serial Square QAM transmitter for NRZ pulse
shape is given by

p u I M (t) u P M (t)
E s (t) = Ps · e j (ωs t+ϕs )
· cos π · e j Vπ π . (2.42)
2Vπ
In order to adjust the desired amplitude and phase levels, the multi-level electrical
driving signals for the MZM and the PM must be chosen as
 q  
2Vπ X i k2 + qk2
u I M (t) = −Vπ + · arcsin  √  · p (t − kTS ), (2.43)
π 2
k
Vπ X
u P M (t) = · arg i k , qk · p (t − kTS ) . (2.44)
π
k
Equations (2.43) and (2.44) are generally applicable to any QAM constellation, and
i k and qk represent the normalized symbol coordinates. For Square 16QAM, for in-
stance, it holds i k ∈ {−1, −1/3, 1/3, 1} and qk ∈ {−1, −1/3, 1/3, 1}, and the nor-

malized symbol coordinates i k and qk are related to the data bits d1k , d2k , b3k , b4k
as defined by the bit mapping shown in Fig. 2.15b and as specified in Table 2.3.

Table 2.3 Relation between the data bits d1k , d2k , b3k , b4k , the normalized symbol coordinates
i k and qk , the normalized symbol amplitudes and the symbol phases for Square 16QAM after
differential quadrant encoding for the bit mapping defined in Fig. 2.15b
d1k d2k b3k b4k ik qk Normalized amplitude Phase (◦ )
0 0 0 0 -1/3 -1/3 1/3 225

0 0 0 1 -1/3 -1 0.74 251.57
0 0 1 0 -1 -1/3 0.74 198.43
0 0 1 1 -1 -1 1 225
0 1 0 0 -1/3 1/3 1/3 135
0 1 0 1 -1 1/3 0.74 161.57
0 1 1 0 -1/3 1 0.74 108.43
0 1 1 1 -1 1 1 135
1 0 0 0 1/3 -1/3 1/3 315
1 0 0 1 1 -1/3 0.74 341.57
1 0 1 0 1/3 -1 0.74 288.43
1 0 1 1 1 -1 1 315
1 1 0 0 1/3 1/3 1/3 45
1 1 0 1 1/3 1 0.74 71.57
1 1 1 0 1 1/3 0.74 18.43
1 1 1 1 1 1 1 45
The arcsin-function in (2.43) makes sure to approach the appropriate intensity

levels while taking the cosine field transfer function of the MZM into account.
In [16], the electrical driving signals were classified as ideal and non-ideal. The
so-called “ideal” driving signals, which are not very practical to generate, can be
regarded as a theoretical approach to completely compensate for the MZM charac-
teristic, so that the electrical pulse shape is directly transposed to the optical field
amplitude. The “non-ideal” driving signals presented here yield correct intensity and
phase states, but lead to small differences during the symbol transitions because they
do not fully compensate for the cosine MZM characteristic. When observing (2.43),
for instance, the corresponding “ideal” driving signal would be obtained by apply-
ing the arcsin-function to the whole sum. More details about this classification of
the driving signals can be found in [16].
One of the main challenges for the practical implementation of the serial Square
QAM transmitter is the generation of the multi-level electrical driving signals with
an electrical level generator. The level-generator is located behind the differential
encoder, as depicted in Fig. 2.16, and acts as a digital-to-analog converter. Fig-
ure 2.17 shows a possible setup of the level-generator for Square 16QAM, composed
of AND-, NOR-, XOR- and XNOR-gates, an inverter, and attenuators. It illustrates
the complexity of the electrical part in the serial Square QAM transmitter.
2.6 Serial Square QAM Transmitter 41
Level generator
Inverter
d1k -1
+
d 2k PS XOR α
0.5 12-ary signal to PM
XNOR α + IS u PM (t )
0.25
XOR
AND α
b3k PS 0.1
XOR
AND α
0.4
b4k PS
AND
3-ary signal to MZM
NOR α + IS uIM (t )
0.22
XOR α
0.53 Attenuators
Fig. 2.17 Electrical level generator for the serial Square 16QAM transmitter, PS: power splitter
An optical Square 16QAM signal has 3 amplitude levels and 12 different phase
q
2 2
√
states. The normalized amplitudes (given by i k + qk / 2) and phases (arg i k , qk ,
in degrees) assigned to particular symbols after the differential encoder are listed in
Table 2.3. When considering the generation of the 3-ary electrical driving signal for
the MZM, the normalized amplitude is “1” if b3k and b4k are both logical one (ap-
plication of an AND-gate), “1/3” if b3k and b4k are both logical zero (NOR-gate),
and “0.74” if b3k and b4k are different (XOR-gate), respectively (see Table 2.3). This
yields the configuration of the lower part of the level generator depicted in Fig. 2.17.
The amplitude states of the MZM driving signal have to be further adjusted to com-
pensate for the nonlinear MZM characteristic, resulting in the attenuation values
shown in Fig. 2.17. For the generation of the 12-ary driving signal for the phase
modulator, it can be observed from Table 2.3 that the phase is equal to 18.43◦ plus
n times 90◦ if b3k is logical one and b4k is logical zero (realization by an XOR-gate
and an AND-gate), equal to 45◦ plus n times 90◦ if b3k and b4k are both logical zero
or both logical one (XNOR), equal to 71.57◦ plus n times 90◦ if b3k is logical zero
and b4k is logical one (XOR, AND), and that d1k and d2k determine the quadrant,
and thus the value of n (n = 0, 1, 2, 3). To obtain a PM driving signal normalized

to one, the appropriate values of the attenuators are also given in Fig. 2.17.
With the level generator described, the driving signals for the MZM and the PM
are generated with the adequate relative values. It should be noted that the given
attenuation values are only valid if the logical gates are operated with DC coupling
at the outputs to obtain unipolar digital output signals. In practice, both signals must
be amplified by modulator drivers to obtain the appropriate driving voltages for the
modulators.
2.6.3 Conventional IQ Transmitter
It can be concluded from Sect. 2.6.2 that the serial Square QAM transmitter fea-
tures a simple optical part, but requires a complex electrical level-generator which
can not easily be implemented for high data rates. Due to the beneficial projection
of the constellation points on the in-phase and quadrature axis, it is an advantage to
generate square shaped constellations with IQ transmitters. This way, the number
of states of the driving signals and thus the electrical complexity can effectively be
reduced in comparison with the serial Square QAM transmitter. In Fig. 2.18, the
setup of the conventional IQ transmitter for Square QAM is illustrated.
Differential
b1k d1k c1k

Encoder
DQPSK
Level
Gen.
Square QAM
b2k d 2k
DEMUX
IS
Coder
1:m
Data b3k
Level
Gen.
bmk IS
cmk
u I (t )
MZM
MZM
CW 3dB 3dB
RZ
-90° MZM Square QAM Signal
uQ (t )
Fig. 2.18 Conventional IQ transmitter for Square QAM
The optical IQ modulator is also used within the parallel DPSK and Star QAM
transmitter (see Sect. 2.4.2 and Sect. 2.5) to perform a DQPSK modulation. Whereas
the in-phase and quadrature driving signals are binary for DQPSK, multi-level elec-
trical driving signals are required to generate higher-order optical Square QAM con-
stellations. The number of levels of the electrical driving signals is equal to the num-
ber of projections of the symbol points to the I-axis and the Q-axis (e.g. quaternary
driving signals are required for Square 16QAM).
2.6 Conventional IQ Transmitter 43
The optical output signal of the conventional IQ transmitter for Square QAM for
NRZ pulse shape can be simply described by
p
E s (t) = Ps · e j (ωs t+ϕs ) · a I Q M (t) · e jϕ I Q M (t) . (2.45)
In (2.45), a I Q M (t) and ϕ I Q M (t) are the amplitude and phase modulation of the
IQM, which are defined in (2.17) and (2.18), respectively. The in-phase and quadra-
ture driving signals are now multi-level and specified as
2Vπ X
u I (t) = −Vπ + · arcsin (i k ) · p (t − kTS ) , (2.46)
π
k
2Vπ X
u Q (t) = −Vπ + · arcsin (qk ) · p (t − kTS ) . (2.47)
π
k
For the ideal driving case, the arcsin-function would have to be applied to the whole
sum in both equations [16]. In order to generate the appropriate driving signals
with simple level generators, the symbol assignment after the differential DQPSK
encoder (see differential quadrant encoding, Sect. 2.6.1) must be rearranged by an-
other coder, which is denoted as Square QAM coder in the following. In practice,
it is possible to implement this coder together with the DQPSK differential encoder
as one single component, possibly through the use of digital signal processing.
{d1 , d 2 , b3 , b4 } {c1 , c2 , c3 , c4 }
q q
0111 0110 1101 1111 0101 0111 1101 1111
0101 0100 1100 1110 0100 0110 1100 1110
i i
0010 0000 1000 1001 0001 0011 1001 1011
0011 0001 1010 1011 0000 0010 1000 1010
Fig. 2.19 Rearranging of the symbols by the Square 16QAM coder within the conventional IQ
transmitter
Figure 2.19 illustrates how the bits have to be rearranged for Square 16QAM. By
rotating the symbols in the n-th quadrant by n times π/2, the differentially encoded
first bit d1k and the third bit b3k as well as the differentially encoded second bit
d2k and the fourth bit b4k are arranged in chronologically increasing order with
increasing signal levels in the in-phase and quadrature arms, respectively. Table 2.4
shows the truth table for the rearrangement of the symbols. It also illustrates the
relation between the normalized symbol coordinates i k and qk , used in (2.46) and
(2.47), the data bits
d 1k , d 2 k , b3 k , b
4 k and the output bits of the Square 16QAM
coder, denoted as c1k , c2k , c3k , c4k , for the conventional IQ transmitter.
Table 2.4 Truth table of the Square QAM coder and relation of the data bits and the normalized
symbol coordinates i k and qk when using the conventional IQ transmitter for Square 16QAM
d1k d2k b3k b4k c1k c2k c3k c4k ik qk
0 0 0 0 0 0 1 1 -1/3 -1/3
0 0 0 1 0 0 1 0 -1/3 -1
0 0 1 0 0 0 0 1 -1 -1/3
0 0 1 1 0 0 0 0 -1 -1
0 1 0 0 0 1 1 0 -1/3 1/3
0 1 0 1 0 1 0 0 -1 1/3
0 1 1 0 0 1 1 1 -1/3 1
0 1 1 1 0 1 0 1 -1 1
1 0 0 0 1 0 0 1 1/3 -1/3
1 0 0 1 1 0 1 1 1 -1/3
1 0 1 0 1 0 0 0 1/3 -1
1 0 1 1 1 0 1 0 1 -1
1 1 0 0 1 1 0 0 1/3 1/3
1 1 0 1 1 1 0 1 1/3 1
1 1 1 0 1 1 1 0 1 1/3
1 1 1 1 1 1 1 1 1 1
Because the symbol rearrangement is confined to the particular quadrants, the

first two bits are not changed by the Square QAM coder, so that c1k = d1k and
c2k = d2k . By analyzing the truth table given in Table 2.4, the following logical
relations can easily be derived for the third and the fourth output bit of the Square
16QAM coder:
c3k = d1k d2k b3k + d1k d2k b4k + d2k b3k b4k + d1k d2k b4k + d1k b3k b4k , (2.48)
c4k = d1k b3k b4k + d1k d2k b3k + d1k d2k b4k + d1k d2k b3k + d2k b3k b4k . (2.49)
The new symbol assignment allows for the use of a simple level generator to gen-
erate the quaternary electrical driving signals in the in-phase and quadrature arms.
This level generator is shown in Fig. 2.20, and is far less complex than for the serial
Square 16QAM transmitter. The bits c1k and c3k are used as inputs for the level-
generator in the in-phase arm, and the bits c2k and c4k are used as inputs for the one
in the quadrature arm. As can be seen from (2.46) and (2.47), the driving amplitude
has to take the value −Vπ if both input bits are logical zero, − arcsin (1/3) · 2Vπ /π
if the first input bit is logical zero and the second logical one, arcsin (1/3) · 2Vπ /π if
the first input bit is logical one and the second logical zero, and Vπ if both input bits
are logical one. The DC coupled unipolar input signals simply have to be added with
the appropriate weights, and the resulting signal must be passed via a DC blocker
2.6 Conventional IQ Transmitter 45
to the consecutive impulse shaper, and then to a modulator driver to provide for the
required MZM driving voltages.
Level generator
c1k / c2k
+ IS u I (t ) / uQ (t )
c3k / c4k α DC blocker
0.64
Fig. 2.20 Level generator for the conventional Square 16QAM IQ transmitter
The Square 64QAM constellation is composed of 64 symbols in a square array.

In Fig. 2.21, a Square 64QAM constellation diagram with rotation symmetric sym-
bol assignments after differential quadrant encoding is shown.
{b3 , b4 , b5 , b6 } {d1 , d 2 }
01
270°
q 11
1111 1110 1011 1010 0101 0111 1101 1111
1101 1100 1001 1000 0100 0110 1100 1110
0111 0110 0011 0010 0001 0011 1001 1011
0101 0100 0001 0000 0000 0010 1000 1010
180° 90° i
1010 1000 0010 0000 0000 0001 0100 0101
1011 1001 0011 0001 0010 0011 0110 0111
1110 1100 0110 0100 1000 1001 1100 1101
1111 1101 0111 0101 1010 1011 1110 1111

00 10
Fig. 2.21 Rotation symmetric Square 64QAM constellation diagram after differential quadrant
encoding; the bits in the corners are the first two bits of a symbol, being differentially encoded; the
arrows indicate the rearrangement of the symbols by the Square 64QAM coder.
One symbol carries the information of six bits. The first two bits d1k and d2k ,
which are differentially encoded by a DQPSK differential encoder to enable resolv-
ing the quadrant ambiguity of the carrier synchronization at the receiver, are equal
within each quadrant and depicted in the corners of the constellation diagram. The
last four bits b3k ...b6k are arranged as rotation symmetric, so a quadrant ambiguity
at the receiver has also no impact on the information recovery of these data bits. The
symbols have 10 different amplitudes and 52 different phases. Due to the high num-
ber of different phase states, only an IQ transmitter seems to be feasible for signal
generation. For the conventional IQ transmitter, the in-phase and quadrature driving
signals have eight levels each.
Like for Square 16QAM, the symbols should be rearranged by a Square QAM
coder for Square 64QAM to produce a symbol assignment that allows for the usage
of simple level generators to generate the 8-ary electrical in-phase and quadrature
electrical driving signals. When rearranging the bits as indicated by the arrows in
Fig. 2.21, the first, third and fifth bits are sorted in chronologically increasing order
with increasing signal levels in the in-phase arm, and in the same way the second,
fourth and sixth bits with increasing signal levels in the quadrature arm. The re-
sulting constellation diagram with re-assigned symbols behind the Square 64QAM
coder is shown in Fig. 2.22.
{c3 , c4 , c5 , c6 } {c1 , c2 }
01 q 11
0101 0111 1101 1111 0101 0111 1101 1111
0100 0110 1100 1110 0100 0110 1100 1110
0001 0011 1001 1011 0001 0011 1001 1011
0000 0010 1000 1010 0000 0010 1000 1010
i
0101 0111 1101 1111 0101 0111 1101 1111
0100 0110 1100 1110 0100 0110 1100 1110
0001 0011 1001 1011 0001 0011 1001 1011
0000 0010 1000 1010 0000 0010 1000 1010

00 10
Fig. 2.22 Symbol assignment after the Square 64QAM coder, optimized for the Square 64QAM
level generator
2.6 Enhanced IQ Transmitter 47

When denoting the input bits as d1k , d2k , b3k , b4k , b5k , b6k , and the output bits
as c1k , c2k , c3k , c4k , c5k , c6k , the logical circuit of the Square 64QAM coder can
be described as follows:
c3k = d1k b3k b4k + d2k b3k b4k + d2k b3k b4k + d1k d2k b4k + d1k d2k b3k , (2.50)
c4k = d1k d2k b4k + d1k d2k b3k + d1k d2k b3k + d1k b3k b4k + d2k b3k b4k , (2.51)
c5k = d1k d2k b5k + d1k d2k b6k + d2k b5k b6k + d1k d2k b6k + d1k b5k b6k , (2.52)
c6k = d1k b5k b6k + d1k d2k b5k + d1k d2k b6k + d1k d2k b5k + d2k b5k b6k . (2.53)
The first two bits do not change, so it holds true that c1k = d1k and c2k = d2k .
The output bits c1k , c3k and c5k serve as inputs for the Square 64QAM level gen-
erator in the in-phase arm, and the remaining three bits as inputs for the level
generator in the quadrature arm. The Square 64QAM level generators must gen-
erate 8-ary electrical driving signals with the appropriate amplitude levels, which
can be deducted from (2.46) and (2.47). The normalized symbol coordinates in
(2.46) and (2.47) are given as i k ∈ {−1, −5/7, −3/7, −1/7, 1/7, 3/7, 5/7, 1} and
qk ∈ {−1, −5/7, −3/7, −1/7, 1/7, 3/7, 5/7, 1} and can be related to the data bits
using the bit mappings illustrated in Fig. 2.21 and Fig. 2.22.
2.6.4 Enhanced IQ Transmitter
When the conventional IQ transmitter is used, the number of levels of the electrical
driving signals required for a particular modulation format is equal to the number
of states of i k and qk , respectively. With the aim of further reducing the number of
states of the electrical driving signals, a modified IQ transmitter configuration can
be employed, which is denoted here as enhanced IQ transmitter [17]. By replacing
the amplitude modulation in each arm with separate intensity and phase modula-
tions, the necessary number of levels of the driving signals can be reduced to half in
comparison with the conventional IQ configuration. To accomplish intensity modu-
lation, a MZM can be used in each arm which is operated at the quadrature point.
The negative values on the I-axis and the Q-axis are reached by varying the phase
between 0 and π, using phase modulators or MZMs operated at the minimum trans-
mission point. Because the phase has to be varied only between 0 and π , binary
signals are sufficient for phase modulation for any modulation format. This way, the
eye spreading problem, arising with the generation of multi-level electrical driving
signals, can be mitigated in practice.
The application of the enhanced IQ transmitter is of special interest for Square

16QAM. Only binary driving signals are required here for all the modulators. This
results in a simpler electrical transmitter part without level generators, composed of
just a 1 : 4 demultiplexer, the coders and the modulator drivers. In Fig. 2.23, the
enhanced IQ transmitter is illustrated for Square 16QAM, composed of MZMs for
intensity modulation and PMs for phase modulation here.
c1k u I PM (t )
Differential
b1k d1k
Encoder
IS
DQPSK
b2k d 2k c2 k uQPM (t )
DEMUX
16QAM
Square
IS
Coder
1:4
Data b3k c3k u I MZM (t )

b4k IS
c4 k uQMZM (t )
IS
MZM PM
MZM
CW 3dB 3dB
RZ
-90° MZM PM
Square 16QAM
Fig. 2.23 Enhanced IQ transmitter for Square 16QAM
When defining the electrical driving signals of the Mach-Zehnder modulators for
the non-ideal driving case in the in-phase and quadrature branches as
2Vπ X
u I M Z M (t) = −Vπ + · arcsin (|i k |) · p (t − kTS ) , (2.54)
π
k
2Vπ X
u Q M Z M (t) = −Vπ + · arcsin (|qk |) · p (t − kTS ) , (2.55)
π
k
and with the driving signals of the phase modulators in both arms given by
Vπ X
u I P M (t) = · (−sign (i k ) + 1) · p (t − kTS ) , (2.56)
2
k
Vπ X
u Q P M (t) = · (−sign (qk ) + 1) · p (t − kTS ) , (2.57)
2
k
the field transfer function of the enhanced IQ transmitter can be expressed as

2.6 Enhanced IQ Transmitter 49
uI (t)
E out (t) 1 u I M Z M (t) PM
= cos π e j Vπ π
E in (t) 2 2Vπ
u Q (t)
1 u Q M Z M (t) PM
+ j cos π e j Vπ π
2 2Vπ
1 1
= · a I (t) · e jϕ I (t) + j · a Q (t) · e jϕ Q (t) . (2.58)
2 2
The optical Square QAM output signal of the enhanced IQ transmitter for NRZ
pulse shape is
p
E s (t) = Ps · e j (ωs t+ϕs ) · a(t) · e jϕ(t) , (2.59)
where the amplitude and phase of the normalized complex envelope follow from
(2.58) by applying complex arithmetic and are given by
1q 2
a(t) = a I (t) + a 2Q (t) + 2a I (t)a Q (t) sin ϕ I (t) − ϕ Q (t) , (2.60)
2
ϕ(t) = arg {a I (t) cos ϕ I (t) − a Q (t) sin ϕ Q (t),

a I (t) sin ϕ I (t) + a Q (t) cos ϕ Q (t)}. (2.61)
The rearrangement of the bits must be accomplished in a different way than for
the conventional IQ transmitter. Figure 2.24 illustrates how the symbols after the
differential encoder must be rearranged for Square 16QAM within the enhanced IQ
transmitter using a Square QAM coder.
{d1 , d 2 , b3 , b4 } {c1 , c2 , c3 , c4 }
q q
0111 0110 1101 1111 0111 0101 1101 1111
0101 0100 1100 1110 0110 0100 1100 1110
i i
0010 0000 1000 1001 0010 0000 1000 1010
0011 0001 1010 1011 0011 0001 1001 1011
Fig. 2.24 Symbol reassignment within the Square 16QAM enhanced IQ transmitter
Only four symbols in the second and the fourth quadrant have to change their
positions to generate a symbol assignment where the first bit d1k = c1k and the
second bit d2k = c2k (both inverted) define if a phase shift of π is performed by
the phase modulators in the in-phase and the quadrature arms, respectively. The
inverters can be saved, because the resulting rotation of π does not matter at the
receiver when differentially quadrant encoding is employed. By the third bit c3k
and the fourth bit c4k the intensity levels in the in-phase and quadrature arms are
specified as low or high, respectively.
Table 2.5 shows the truth table for symbol rearrangement. It also illustrates
the relation between the normalized symbol coordinates
i k and qk , used in
(2.54)–(2.57),
and
the data bits d1k , d 2k , b 3k , b 4 k as well as the four output bits
c1k , c2k , c3k , c4k of the Square 16QAM coder for the enhanced IQ transmitter.
Table 2.5 Truth table of the Square QAM coder and relation of the data bits and the normalized
symbol coordinates i k and qk when using the enhanced IQ transmitter for Square 16QAM
d1k d2k b3k b4k c1k c2k c3k c4k ik qk
0 0 0 0 0 0 0 0 -1/3 -1/3
0 0 0 1 0 0 0 1 -1/3 -1
0 0 1 0 0 0 1 0 -1 -1/3
0 0 1 1 0 0 1 1 -1 -1
0 1 0 0 0 1 0 0 -1/3 1/3
0 1 0 1 0 1 1 0 -1 1/3
0 1 1 0 0 1 0 1 -1/3 1
0 1 1 1 0 1 1 1 -1 1
1 0 0 0 1 0 0 0 1/3 -1/3
1 0 0 1 1 0 1 0 1 -1/3
1 0 1 0 1 0 0 1 1/3 -1
1 0 1 1 1 0 1 1 1 -1
1 1 0 0 1 1 0 0 1/3 1/3
1 1 0 1 1 1 0 1 1/3 1
1 1 1 0 1 1 1 0 1 1/3
1 1 1 1 1 1 1 1 1 1
The logical circuit of the Square QAM coder for the enhanced Square 16QAM
transmitter is defined by (2.62) and (2.63). These relations can be easily derived
from the truth table given by Table 2.5. Both first bits do not change (so it holds true
that c1k = d1k and c2k = d2k ), and the third and fourth output bits are related to the
input bits as
c3k = d1k d2k b3k + d1k d2k b3k + d1k d2k b4k + d1k d2k b4k , (2.62)
c4k = d1k d2k b3k + d1k d2k b3k + d1k d2k b4k + d1k d2k b4k . (2.63)
The optical part of the enhanced IQ transmitter can be simplified by performing

the separate intensity and phase modulations in each arm with only one component.
For this purpose, a dual-drive MZM can be used, driven simultaneously in the push-
pull mode for intensity modulation and in the push-push mode for phase modulation.
However, the driving signals for intensity and phase modulation must be electrically
combined in that case before being injected into the MZM inputs [16].
2.6 Tandem-QPSK Transmitter 51
2.6.5 Tandem-QPSK Transmitter
Another transmitter also requiring only binary electrical driving signals for Square
16QAM is denoted as Tandem-QPSK transmitter throughout this book and can be
composed of an optical IQ modulator followed by a DQPSK modulator. The latter
can be implemented either with one more IQ modulator, or with two consecutive
phase modulators, as depicted in Fig. 2.25.
d1k u PM1 (t )
Differential
b1k
Encoder
IS
DQPSK
b2k d 2k u PM 2 (t )
DEMUX
IS
1:4
Data b3k u I (t )
IS
b4k uQ (t )
IS
MZM
MZM
CW 3dB 3dB PM PM
RZ
-90° MZM π π/2
Square 16QAM
Fig. 2.25 Optical Tandem-QPSK transmitter for Square 16QAM
Like for the enhanced IQ transmitter, the MZMs within the IQ modulator achieve
modulation in intensity. This way, only positive values on the in-phase and quadra-
ture axis are addressed. As regards Square 16QAM, the MZMs are driven by binary
electrical signals, and a constellation composed of four symbols in the first quadrant
is created. In a similar way to the enhanced IQ transmitter, the electrical driving
signals of the MZMs are defined here as
2Vπ X h i
u I (t) = −Vπ + · arcsin i k1 · p (t − kTS ) , (2.64)
π
k
2Vπ X h i
u Q (t) = −Vπ + · arcsin qk1 · p (t − kTS ) , (2.65)
π
k
where i k1 and qk1 represent the normalized symbol coordinates in the first quadrant
of the constellation diagram and are directly related to the data bits b3k and b4k ,
respectively. For the bit mapping depicted in Fig. 2.15a, i k1 = 1(1/3) for b3k = 1(0)
and qk1 = 1(1/3) for b4k = 1(0). Operating the IQ modulator in this way has the
same effect as interfering a DQPSK signal with a CW wave, as proposed in [10].
With two consecutive phase modulators which perform phase shifts of π and
π/2, respectively, the three other quadrants can be approached, thus creating a
complete Square QAM constellation. It is a beneficial side-effect of this transmitter

type that—initiated through signal generation—the resulting constellation is inher-
ently symmetric in rotation as regards the data bits b3k and b4k , so that no additional
Square QAM coder for symbol rearrangement is needed to constitute a rotation sym-
metric symbol assignment which is required to handle the quadrant ambiguity at the
receiver. It should be noted that the differential encoder for the serial DQPSK trans-
mitter must be employed when performing the quadrant shift with two consecutive
phase modulators as shown in Fig. 2.25.
With the amplitude and phase modulation of the IQ modulator, a I Q M (t) and
ϕ I Q M (t), defined by (2.17) and (2.18), respectively, and with the phase modulator
driving signals u P M1 (t) and u P M2 (t) given by (2.30) as for the serial DPSK trans-
mitter (n = 1, 2), the optical output signal of the Tandem-QPSK transmitter for
NRZ pulse shape can be described by
p u P M (t) u P M (t)
1 π 2 π
E s (t) = Ps · e j (ωs t+ϕs ) · a I Q M (t) · e jϕ I Q M (t) · e j Vπ · ej Vπ . (2.66)
The Tandem-QPSK transmitter is a promising option for practically implement-

ing Square 16QAM systems, since the driving signals are binary and the signal
generation is well suited for creating rotation symmetric constellations.
2.6.6 Multi-Parallel MZM Transmitter
Another option for generating Square QAM signals, which has been recently pro-
posed in [14, 15], is to use a multi-parallel MZM transmitter. Its setup for Square
16QAM is illustrated in Fig. 2.26.
By arranging two IQ modulators in parallel, a Square 16QAM signal can be syn-
thesized from two QPSK signals. The so-called “large-amplitude QPSK” is created
by the upper IQM and determines the quadrant to which the symbol is mapped.
The “small-amplitude QPSK” is obtained after the attenuator in the lower branch
(attenuation 6 dB) and fixes the position of a symbol within the quadrant. The
Square 16QAM constellation is finally assembled by the combination of both QPSK
signals. All MZMs are operated at the minimum transmission point. The Square
16QAM signal is generated by driving them only with binary electrical signals, so
that the transmitter is free from handling multi-level electrical driving signals—just
like the enhanced IQ transmitter and the Tandem-QPSK transmitter. The Square
16QAM transmitter shown in Fig. 2.26 is denoted as quad-parallel MZM transmit-
ter because four MZMs are used in parallel.
Generally, a M-ary Square QAM signal can be created by a multi-parallel MZM
transmitter being composed of m/2 optical IQMs and accordingly m MZMs. The
power attenuation in the n-th IQM branch must then be chosen as (n − 1) · 6 dB.
To generate the modulator driving signals, the same coders as those appropri-
ate for the conventional IQ transmitter can be employed in the electrical part of
the multi-parallel MZM transmitter. In Fig. 2.19, it is shown how the bit mapping
2.6 Multi-Parallel Transmitter 53
c1k
Differential
b1k d1k
Encoder
IS
DQPSK
b2k d 2k c2 k
DEMUX
16QAM
Square
IS
Coder
1:4
Data b3k c3k
b4k IS
c4 k
IS
u I L (t )
IQ-Modulator 1
MZM
3dB 3dB
-90° MZM
uQL (t )
MZM
CW 3dB 3dB
RZ u I S (t )
IQ-Modulator 2
MZM Square 16QAM
3dB 3dB α
-90° MZM 6dB
uQS (t )
Fig. 2.26 Quad-parallel MZM transmitter for generation of Square 16QAM signals
after differential encoding must be rearranged by a Square QAM coder for Square
16QAM to create a symbol assignment suitable for driving the four MZMs of the
quad-parallel MZM transmitter. To create the large-amplitude QPSK, the MZMs in
the in-phase and quadrature branches in the upper IQM branch are driven by
X
u I /Q L (t) = −2Vπ + 2Vπ · c1/2k · p (t − kTS ) , (2.67)
k
where c1k = d1k and c2k = d2k represent the first and second output bit of the
Square 16QAM coder, respectively. The driving signals for the generation of the
small-amplitude QPSK depend on the third and fourth Square 16QAM coder output
bits and are specified by
X
u I /Q S (t) = −2Vπ + 2Vπ · c3/4k · p (t − kTS ) . (2.68)
k
After combining the QPSK signals of both branches in a 3 dB coupler at the

output of the quad-parallel MZM transmitter, the optical Square 16QAM output
signal at the lower output port of the 3 dB coupler—when neglecting constant phase
shifts and assuming NRZ line coding—is given by
1p h
E s (t) = Ps · e j (ωs t+ϕs ) · a I Q M L (t) · e jϕ I Q M L (t)
2
1 jϕ I Q M S (t)
+ a I Q M S (t) · e . (2.69)
2
In (2.69), a I Q M L/S (t) and ϕ I Q M L/S (t) describe the amplitude and phase modulation
of the IQM in the upper / lower branch. These can be calculated with (2.17) and
(2.18) when applying the electrical driving signals defined by (2.67) and (2.68).
2.6.7 Signal Properties
In the following paragraphs, the signal properties at the transmitter outputs are dis-
cussed for selected Square QAM transmitters. The amplitude and phase character-
istics of the optical transmitter output signals for NRZ pulse shape are defined for
the serial Square QAM transmitter by (2.42), for the conventional IQ transmitter
by (2.45)–(2.47) using (2.17) and (2.18), for the enhanced IQ transmitter by (2.60)
and (2.61), and for the Tandem-QPSK transmitter by (2.66). To obtain the optical
output signals for RZ pulse shape, the transfer function of the pulse carver given by
(2.19) must additionally be considered. In Fig. 2.27, the intensity eye diagrams, the
IQ diagrams (with a(t) scaled to unity) and the chirp characteristics are depicted for
NRZ pulse shape, assuming a data rate of 40 Gbit/s and an electrical rise time of 1/4
of the symbol duration.
a Serial TX b Conventional IQ TX c Enhanced IQ TX d Tandem QPSK TX

2
1 1 1
Norm. Intensity
Norm. Intensity
Norm. Intensity
Norm. Intensity
0.8 0.8 1.5 0.8

Intensity
Intensity
Intensity
Intensity
0.6 0.6 0.6

1
0.4 0.4 0.4
0.5
0.2 0.2 0.2
0 0 0 0
0.5 1.5 0.5 1.5 0.5 1.5 0.5 1.5
Time Time Time Time
1 1 1 1
Im [A(t)]
Im [A(t)]
Im [A(t)]
Im [A(t)]
0 0 0 0
-1 -1 -1 -1
-1 0 1 -1 0 1 -1 0 1 -1 0 1
11 11 11 11
x 10 x 10 x 10 x 10
5 5 5 5
2.5 2.5 2.5 2.5

Chirp (1/s)
Chirp (1/s)
Chirp (1/s)
Chirp (1/s)
0 0 0 0
-2.5 -2.5 -2.5 -2.5
-5 -5 -5 -5
0.5 1.5 0.5 1.5 0.5 1.5 0.5 1.5
Time Time Time Time
11 11 11 11
x 10 x 10 x 10 x 10
5 5 5 5
2.5 2.5 2.5 2.5
0 0 0 0
-2.5 -2.5 -2.5 -2.5
-5 -5 -5 -5
0.5 1.5 0.5 1.5 0.5 1.5 0.5 1.5
Time Time Time Time
Fig. 2.27 Optical signal properties of different Square 16QAM transmitters (NRZ)
2.6 Square QAM Signal Properties 55
From the intensity eye diagrams in Fig. 2.27, it can be observed that there are
undershoots for the optical IQ transmitters (the intensity goes to zero). When the en-
hanced IQ structure is employed, where the amplitude modulation is replaced with
separate intensity and phase modulations, additional overshoots arise at the sym-
bol transitions. As a result of the separate phase modulation, the optical power can
be temporarily higher than it is for the constellation points with the highest power
level. This becomes equally apparent in the corresponding constellation diagram,
where phasors with an amplitude greater than one are possible during the symbol
transitions. A comparison of the constellation diagrams shows that there are sym-
bol transitions through zero for the IQ transmitters but not for the serial transmitter
and the Tandem-QPSK transmitter, where optical power is always greater than zero.
When considering chirp, the best transmission performance can be expected from
the conventional IQ transmitter, since the product of chirp and normalized intensity
is comparatively small. The chirp characteristic of the other transmitters is more
disadvantageous because chirp appears simultaneously with high power levels. Fig-
ure 2.28 illustrates the signal characteristics for some further selected cases (“ideal”
driving, shorter electrical rise time, RZ pulse shape).
a Conventional IQ TX, b Enhanced IQ TX, c Tandem QPSK TX,

ideal driving, NRZ reduced rise time, NRZ RZ
2
1 1
Norm. Intensity
Norm. Intensity
Norm. Intensity
0.8 1.5 0.8

0.6 0.6
1
0.4 0.4
0.5
0.2 0.2
0 0 0
0.5 1.5 0.5 1.5 0.5 1.5
Time Time Time
1 1 1
Im [A(t)]
Im [A(t)]
Im [A(t)]
0 0 0
-1 -1 -1
-1 0 1 -1 0 1 -1 0 1
Re [A(t)] Re [A(t)] Re [A(t)]
11 11 11
x 10 x 10 x 10
5 5 5
2.5 2.5 2.5

Chirp (1/s)
Chirp (1/s)
Chirp (1/s)
0 0 0
-2.5 -2.5 -2.5
-5 -5 -5
0.5 1.5 0.5 1.5 0.5 1.5
Time Time Time
11 11 11
x 10 x 10 x 10
5 5 5
2.5 2.5 2.5
0 0 0
-2.5 -2.5 -2.5
-5 -5 -5
0.5 1.5 0.5 1.5 0.5 1.5
Time Time Time
Fig. 2.28 Optical signal properties of Square 16QAM transmitters for “ideal” driving (a), shorter
rise time (b) and RZ pulse shape (c), selected cases
The use of the “ideal” driving signals (which were discussed in Sect. 2.6.2 and
can be regarded as a theoretical approach to compensate for the cosine character-
istic of the MZM) leads to only small differences, as is exemplarily shown for the
conventional IQ transmitter in Fig. 2.28a. A comparison with Fig. 2.27b shows that
the eyes are not as steep as with the “non-ideal” driving case, and that the symbol
transitions are only slightly different (when the conventional IQ transmitter is used,
symbol transitions take the shortest way between the constellation points for “ideal”
driving). To cope with the overshoots of the enhanced IQ transmitter for NRZ, either
steeper driving signals for phase modulation must be applied (resulting in shorter
overshoots, but higher chirp, as depicted in Fig. 2.28b for a rise time of 1/16 of the
symbol duration), or RZ pulse shape must be used. RZ pulse carving leads to similar
eyes and constellation diagrams in any configuration, and eliminates unwanted in-
tensity ripples at symbol transitions for the enhanced IQ transmitter. Moreover, the
influence of chirp is reduced when using RZ, as exemplarily shown for the Tandem-
QPSK transmitter in Fig. 2.28c.
References
1. Covega corporation, http://www.covega.com

2. VPIsystems, Photonic Modules Reference Manual, 2005
3. Cahn, C.R.: Combined digital phase and amplitude modulation communication system. IRE
Transactions on Communications CS-8, 150–155 (1960)
4. Campopiano, C.N., Glazer, B.G.: A coherent digital amplitude and phase modulation system.
IRE Transactions on Communications CS-10, 90–95 (1962)
5. Hancock, J.C., Lucky, R.W.: Performance of combined amplitude and phase modulated com-
munication system. IRE Transactions on Communications CS-8, 232–237 (1960)
6. Ho, K.P., Cuei, H.W.: Generation of arbitrary quadrature signals using one dual-drive modu-
lator. IEEE Journal of Lightwave Technology 23(2), 764–770 (2005)
7. Hooijmans, P.W.: Coherent Optical System Design. John Wiley & Sons (1994)
8. Kazovsky, L.G.: Optical Fiber Communication Systems. Artech House (1996)
9. Kikuchi, N., Sasaki, S.: Optical dispersion-compensation free incoherent multilevel signal
transmission over single-mode fiber with digital pre-distortion and phase pre-integration
techniques. In: Proceedings of European Conference on Optical Communication (ECOC),
Tu.1.E.2 (2008)
10. Kikuchi, N., et al.: Proposal of inter-symbol interference (ISI) supression technique for optical
multilevel signal generation. In: Proceedings of European Conference on Optical Communi-
cation (ECOC), Tu4.2.1 (2006)
12. Patzak, E., Langenhorst, R.: Sensitivity degradation of conventional and balanced 3 × 3 port
phase diversity receivers due to thermal and local oscillator intensity noise. Electronics Letters
25(8), 545–547 (1989)
13. Petermann, K.: Laser diode modulation and noise. Kluwer Academic, Dor-
drecht / Boston / London (1988)
14. Sakamoto, T., et al.: 50-Gb/s 16 QAM by a quad-parallel Mach-Zehnder modulator. In: Pro-
ceedings of European Conference on Optical Communication (ECOC), PDP2.8 (2007)
References 57
(2008)
16. Seimetz, M.: Multi-format transmitters for coherent optical M-PSK and M-QAM transmis-
sion. In: Proceedings of ICTON, Th.B1.5. Barcelona (2005)
17. Seimetz, M.: Optical IQ-transmitter with a serial IQ-modulator. German utility patent, DE 20
2006 000 197.2 (2006)
18. Seimetz, M.: Phase diversity receivers for homodyne detection of optical DQPSK signals.
IEEE Journal of Lightwave Technology 24(9), 3384–3391 (2006)
20. Voges, E., Petermann, K.: Optische Kommunikationstechnik. Springer Verlag,
Berlin / Heidelberg (2002)
21. Walklin, S., Conradi, J.: Multilevel signaling for increasing the reach of 10Gb/s lightwave
systems. IEEE Journal of Lightwave Technology 17(11), 2235–2248 (1999)
22. Weber, W.J.: Differential encoding for multiple amplitude and phase shift keying systems.
IEEE Transactions on Communications COM-26(3), 385–391 (1978)
23. Winzer, P.J., Essiambre, R.J.: Advanced optical modulation formats. Proceedings of the IEEE
94(5), 952–985 (2006)
24. Zhao, J., et al.: Analytical investigation of optimization, performance bound, and chromatic
dispersion tolerance of 4-amplitude-shifted-keying format. In: Proceedings of Optical Fiber
Communication Conference (OFC), JThB15 (2006)
Chapter 3
Receiver Configurations
Abstract This chapter describes receiver configurations for the detection of optical
signals with high-order modulation. At the beginning of the chapter, receiver key
components (filters, delay line interferometer, optical hybrids) are discussed. After-
wards, receivers are detailed for three different detection schemes: direct detection,
homodyne differential detection and homodyne synchronous detection. A complete
analytical description of the optical and electrical receiver parts is thereby provided,
starting with optical frontends and ending with electrical data recovery. Direct de-
tection receivers are shown to be capable of detecting arbitrary modulation formats
with differentially encoded phases such as DPSK and Star QAM formats. Subse-
quently, homodyne receivers with differential detection are presented and shown
to be suitable for detecting the same signals as is possible with direct detection.
Finally, homodyne synchronous detection which is appropriate for arbitrary high-
order modulation formats is highlighted. A particular stress is laid on carrier syn-
chronization, which was traditionally performed using an optical phase locked loop.
Its current state is briefly reviewed. Digital phase estimation schemes are discussed
in more detail, because digital coherent receivers gain more and more interest due
to high-speed digital signal processing newly available.
3.1 Receiver Components
The following subsections describe some key components used within optical re-
ceivers for high-order modulation.
3.1.1 Optical and Electrical Filters
Filters act as crucial and multi-purpose components in optical receivers. In direct

detection receivers with optical pre-amplification, an optical band-pass filter should


60 3 Receiver Configurations
be located behind the pre-amplifier to limit the noise bandwidth and thus reduce
noise components falling and beating into the detection bandwidth, resulting in a
higher receiver sensitivity. Furthermore, optical band-pass filters are used for WDM
channel separation. In the electrical domain, low-pass filters are employed to further
reduce amplifier noise, as well as the receiver’s shot-noise and thermal noise.
In Signal and System theory, mostly ideal filters are observed—like the ideal
low-pass filter, which transmits all frequencies inside the pass-band without any dis-
tortion and completely rejects frequencies inside the stop-band [16]; or the matched
filter, whose impulse response matches the received signal and which maximizes
the output signal to noise ratio [54].
In system simulations, optical band-pass filters are often modeled as non-causal

Gaussian filters with Gaussian amplitude and real-valued transfer function, provid-
ing a first-step approximation of a real filter. The transfer function of a Gaussian
optical band-pass filter is given by [2]
2ν
− 0.5·ln (2)·2
2ν ·(ω−ω0 )2ν
Ho (ω) = e ωc . (3.1)
In (3.1), ω0 represents the center frequency, ωc is the 3 dB bandwidth, and ν is the

filter order.
Electrical filters are often modeled as Bessel filters. They can be physically im-
plemented and exhibit a linear phase response and an excellent step response with
minimal overshoot and ringing [2]. The normalized transfer function of an electrical
Bessel low-pass filter can be described as
Bν (S = 0)
He (S) = , S = jωτgr0 , (3.2)
Bν (S)
where ν again denotes the filter order, and τgr0 the group delay for ω = 0. The
polynomials Bν (S) for ν = 1 and ν = 2 are specified by B1 (S) = 1 + S and
B2 (S) = 3 + 3S + S 2 , respectively. The polynomials for higher orders can be
calculated recursively by the relation [68]
Bν (S) = (2ν − 1) · Bν−1 (S) + S 2 · Bν−2 (S). (3.3)
The 3 dB bandwidth of the electrical Bessel low-pass filter can be related to the
group delay τgr0 by ωc = 1.3616/τgr0 for ν = 2, and can be approximately calcu-
lated for higher filter orders with
√
(2ν − 1) · ln 2
ωc ≈ , ν ≥ 3. (3.4)
τgr0
More detailed information about Bessel filters and other practical filter realiza-
tions can be found in [68].
3.1 Delay Line Interferometer 61
3.1.2 Delay Line Interferometer
In order to evaluate the phase information of an optical signal without using coherent
detection techniques, phase information must be converted to intensity information
before the detection process, since intensity alone can be detected by a photodiode.
The optical component, which is commonly used for this purpose, is a delay line
interferometer (DLI). As illustrated in Fig. 3.1, the received light is split by a 3 dB
coupler into two branches. In one branch, the optical signal is delayed by one symbol
duration TS . Additionally, an arbitrary phase shift ϕ DL I can be accomplished in one
of the branches before the light of both branches is recombined in a second 3 dB
coupler.
E# o u t1
Ein1 TS Eout1
Ein2 ϕDLI Eout2

Fig. 3.1 Structure of a delay
line interferometer (DLI)
E# o u t 2
With the nomenclature used in Fig. 3.1 for the 3 dB coupler at the DLI input, the
input and output relationship of a single lossless 3 dB coupler can be expressed in
matrix notation as

Ẽ out1 1 1 j E in 1
=√ · · . (3.5)
Ẽ out2 2 j 1 E in 2
One property of the 3 dB coupler is that the relative phase shift experienced by both
added input fields is 90◦ at the upper and -90◦ at the lower coupler output. Thus, the
difference of these relative phase shifts is 180◦ . For this reason, the 3 dB coupler is
often denoted as 180◦ hybrid.
When feeding a signal E in 1 into the upper port of the input coupler of the DLI,
and when neglecting the common phase shift within the interferometer length, the
two fields obtained at the DLI output ports are
1 1
E out1 (t) = E in 1 (t − TS ) − E in 1 (t)e jϕ DL I , (3.6)
2 2
1 1
E out2 (t) = j E in 1 (t − TS ) + j E in 1 (t)e jϕ DL I . (3.7)
2 2
For a DLI input signal with time dependent modulation amplitude a(t) and modu-
lation phase ϕ(t), which is corrupted by laser phase noise ϕn s (t),
p
E in 1 (t) = Ps · e j (ωs t+ϕs ) · a(t) · e jϕ(t) · e jϕns (t) , (3.8)
the optical power at the DLI outputs—when further neglecting frequency drift ef-
fects (so that ωs remains constant)—can be specified by
∗ 1 1
Pout1 (t) = E out1 (t) · E out1
Ps · a 2 (t) + Ps · a 2 (t − TS )
(t) =
4 4
1
− Ps · a(t)a(t − TS ) cos 1ϕ(t) + 1ϕn s (t) + ϕ DL I , (3.9)
2
∗ 1 1
Pout2 (t) = E out2 (t) · E out2
Ps · a 2 (t) + Ps · a 2 (t − TS )
(t) =
4 4
1
+ Ps · a(t)a(t − TS ) cos 1ϕ(t) + 1ϕn s (t) + ϕ DL I , (3.10)
2
where the modulation phase difference between two consecutive symbols, 1ϕ(t),
and the phase change due to laser phase noise at the DLI outputs, 1ϕn s (t), are
defined as
1ϕ(t) = ϕ(t) − ϕ(t − TS ), 1ϕn s (t) = ϕn s (t) − ϕn s (t − TS ). (3.11)
For the derivation of (3.9) and (3.10), it is further assumed that an integer number
of wavelengths fits into the symbol time (ωs · TS = 2mπ, m ∈ N0 ) because an extra
relative phase shift would otherwise occur. It can be concluded from (3.9) and (3.10)
that the optical power at the DLI outputs depends on the current modulation ampli-
tude a(t), the modulation amplitude delayed by one symbol duration a(t − TS ), the
modulation phase difference 1ϕ(t), the laser phase noise 1ϕn s (t) with a variance
of h1ϕn2s i = 2π 1νs TS and the DLI phase shift ϕ DL I . For an ideal DPSK with con-
stant a(t) = a(t − TS ) and for negligible laser phase noise, the DLI output signals
can be used for evaluation of 1ϕ(t) when an appropriate DLI phase shift ϕ DL I is
adjusted. In principle, one of the DLI output ports is sufficient for signal detection
(single-ended detection). To eliminate the first two terms in (3.9) and (3.10) and not
waste half the power, however, a balanced detector should be employed.
In practice, the symbol delay TS is implemented by using different path lengths
within the DLI. Its accuracy is not critical in terms of hitting the symbol time, but
rather in terms of fine-tuning the length difference with sub-wavelength accuracy to
control the interference conditions.
3.1.3 2 × 4 90◦ Hybrid
The 2 × 4 90◦ hybrid is a key component in optical receivers for high-order modula-
tion because it enables the detection of the in-phase and quadrature components of
an optical signal. It can be used within coherent receivers to superposition the signal
light and the LO light, as well as for conversion of the phase difference information
3.1 2 × 4 90◦ Hybrid 63
into intensity information in direct detection receivers, provided that an optical delay
of one symbol time is foreseen at one of the hybrid inputs. When generally defining
the two input signals of the 2 × 4 90◦ hybrid as

E in 1 (t) = E in 1 (t) · e jφ1 (t) , E in 2 (t) = E in 2 (t) · e jφ2 (t) , (3.12)
the following output powers are desired at the four hybrid outputs (n = 0, 1, 2, 3)
to enable the detection of the in-phase and quadrature components
∗ 1 2 1 2
Poutn (t) = E outn (t) · E outn
(t) = E in 1 (t) + E in 2 (t)
4 4
1
+ E in 1 (t) E in 2 (t) · cos φ1 (t) − φ2 (t) − n · 90◦ + ψ ,(3.13)
2
where the phase shift ψ is allowed to be arbitrary, since the initial phases of the input
signals are arbitrary as well. It can be observed that the beating terms of two adjacent
output signals are in quadrature, respectively, and the remaining output signals can
be used to employ balanced detection. The quadrature property can be exploited to
detect the in-phase and quadrature components of high-order modulation signals,
as well as for the implementation of special coherent receivers such as the phase
diversity receiver [30], the image-rejection receiver [23] and receivers with optical
phase locked loops [12, 27].
To provide the desired powers specified by (3.13) at the four outputs, the 2 × 4
90◦ hybrid (when assuming a lossless device with ideal uniformity) must exhibit the
field transfer function
   jψ 
E out1 (t) e 11 e jψ12
 E out (t) 1 e jψ21 j · e jψ22  E in (t)
2 = · · 1
 E out3 (t) 2 e jψ31 −e jψ32  E in 2 (t) ,
 (3.14)
E out4 (t) e jψ41 − j · e jψ42
where the phase coefficients ψ11 ..ψ42 must satisfy the phase condition
ψ11 − ψ12 = ψ21 − ψ22 = ψ31 − ψ32 = ψ41 − ψ42 = ψ. (3.15)
Since the difference of the relative phase shifts of the added input fields at the four
outputs is n times 90◦ , this component is denoted as a 90◦ hybrid.
Different implementation options for the 2 × 4 90◦ hybrid and their practical
feasibility are discussed in detail in [61]. Figure 3.2 illustrates three different options
for implementation of the hybrid.
One possibility, shown in Fig. 3.2a, is to construct the hybrid with four 3 dB cou-
plers and an additional phase shifter in one branch. When setting the phase shift in
the lower branch to 90◦ and using the inputs and outputs as defined in Fig. 3.2a, the
transfer function of the whole component is equal to (3.14) with ψ = 0 [61]. The
configuration should be implemented in an integrated form to achieve sufficient IQ
balance. A version fabricated on Li N bO3 is analyzed and discussed in [19]. The de-
vice is commercially available [1] and can be adjusted with six different electrodes.
a 3dB couplers + phase shifter
Ein1 Eout3
3dB 3dB
Eout1
Eout4
3dB 3dB
Ein2 90° Eout2
b 4x4 MMI coupler c 3dB coupler + PBS
Eout1 Eout1
Ein1 PBS
4x4 Eout4 Ein1 Eout2
Ein2 3dB
MMI Eout2 Ein2 Eout3
Eout3 PBS
Eout4
Fig. 3.2 Options for implementation of the 2 × 4 90◦ hybrid
Four electrodes control the uniformity of the 3 dB couplers (which is a measure of

the differences in the output port powers). With the remaining two electrodes, the
phase shifts in the upper and the lower branches can be set [26]. This allows the
adjustment of phase shifts of ψ 6= 0, which can be a beneficial feature for rotating
received constellation diagrams into the appropriate position. To ensure orthogonal-
ity, the relative phase shift between two branches has to be tuned to 90◦ . Imprecise
relative phase shifts lead to a degradation of the IQ balance, whereas the asymme-
tries of the 3 dB couplers affect the power symmetry of the hybrid output signals and
thus the symmetry of the subsequent balanced detection processes. For commercial
application, even for the integrated implementation of this component on Li N bO3 ,
it is necessary to establish an active control loop to stabilize the 90◦ phase relation.
Even though the hybrid on Li N bO3 is commercially available, the active control
loop is not, and would have to be implemented with ones own resources. A possible
design for the active control loop was proposed in [61].
The most promising option for obtaining a stable commercial 2 × 4 90◦ hybrid
component without the need of an additional active control, involves exploiting the
properties of a 4 × 4 multi-mode interference (MMI) coupler. Using the appropriate
inputs, as shown in Fig. 3.2b for instance, and for accurate waveguide dimension-
ing, this component inherently exhibits the desired phase relations [50, 61]. Unfor-
tunately, 4 × 4 MMI couplers are commercially not available at present. However,
they are a very interesting alternative because they promise stable phase relations.
Furthermore, MMI couplers are broadband. This makes them suitable for multi-
band WDM applications. In addition, the balanced detectors of the receiver can be
integrated on the chip, so that the whole optical receiver frontend can be integrated,
possibly with polarization diversity. The device has to be carefully designed in order
3.1 3 × 3 Coupler 65
to achieve equal splitting ratios together with the appropriate phase relations, as it
was shown using simulations in [61].
A third option of the 2 × 4 90◦ hybrid which has been realized with discrete
components [36], as well as in an integrated form [35], is a configuration with a 3 dB
coupler in combination with two polarization beam splitters (PBS), as depicted in
Fig. 3.2c. This arrangement, however, requires specific polarization states from the
signals feeding into the hybrid inputs. One input signal must be linearly polarized at
45◦ with respect to the PBS reference directions, and the other one must be circularly
polarized:
1
Ein 1 (t) = √ · ex + e y · E in 1 (t) · e jφ1 (t) , (3.16)
2
1 ◦

Ein 2 (t) = √ · ex + e y e j90 · E in 2 (t) · e jφ2 (t) . (3.17)
2
When using these input polarizations, the outputs fields of the two polarization beam
splitters are obtained as
 1 jφ (t) 
  2 ex · E in 1(t) · e
1 + E in 2(t) · e jφ2 (t)
Eout1 1

jφ1 (t) + E in (t) · e jφ2 (t) · e j90◦ 
Eout   2 e y · E in 1 (t) · e

2 
2
Eout3  =  21 ex · E in 1 (t) · e jφ1 (t) − E in 2 (t) · e jφ2 (t)
   , (3.18)

 jφ (t) jφ (t) j90◦ 
Eout4 1
e · E (t) · e 1 − E (t) · e 2 · e
2 y in 1 in 2
and result again in the optical powers given by (3.13) with ψ = 0. So far, solutions
were implemented using free space optics [38]. A fiber-based approach could be
even more difficult to implement. For lab experiments, it is advantageous that this
setup can be configured with discrete components.
3.1.4 3 × 3 Coupler
One way to avoid the deployment of a 2 × 4 90◦ hybrid and to be able to detect
in-phase and quadrature components anyway, is to use a 3 × 3 fiber coupler. An
ideal lossless 3 × 3 coupler with perfect uniformity provides the functionality of a
120◦ hybrid with an input and output relationship [53]
     
E out1 1 1 1 E in 1
1 ◦ ◦
 E out2  = √ · 1 e j120 e− j120  ·  E in 2 . (3.19)
E out3 3 ◦
1 e− j120 e j120
◦
E in 3
In practice, non-ideal uniformity and lossy couplers result in deviations from the
ideal phase relations [52, 53]. When feeding the two signals defined by (3.12) into
the first and second input of the 3 × 3 coupler, the optical power at the output n
(n = 1, 2, 3) is given by
∗ 1 2 1 2
Poutn (t) = E outn (t) · E outn
(t) = E in 1 (t) + E in 2 (t)
3 3
2
+ E in 1 (t) E in 2 (t) cos φ1 (t) − φ2 (t)+(n − 1) 120◦ . (3.20)
3
It becomes apparent that the cosine terms of the three output powers are 120◦ out of
phase with each other. From these signals, two signals 90◦ out of phase and with the
same amplitude can be easily constructed by performing the following operations
[35, 43, 53]:
PI (t) = Pout1 (t) − Pout2 (t)

2
= √ E in 1 (t) E in 2 (t) · sin φ1 (t) − φ2 (t) + 60◦ , (3.21)
3
1
PQ (t) = √ · Pout1 (t) + Pout2 (t) − 2Pout3 (t)
3
2
= √ E in 1 (t) E in 2 (t) · cos φ1 (t) − φ2 (t) + 60◦ . (3.22)
3
In practice, these operations can be performed electronically or by digital means
after photo-detection.
3.2 Direct Detection Receivers
Receiver schemes for the detection of optical high-order modulation signals can be
roughly divided into two basic groups: Direct detection and coherent detection. Di-
rect detection receivers, whose functionality is shown in this section, are convinc-
ingly simple. No phase, frequency or polarization control is necessary. Although
only the intensity of the optical field can be detected by a simple photodiode, the
information encoded in the optical phase can also be obtained when employing ad-
ditional optics. By using an optical interferometer, the phase difference information
of two consecutive symbols can be converted into intensity information, which then
can be detected by a photodiode. This allows for the detection of arbitrary DPSK
signals. With a separate intensity detection branch, arbitrary Star QAM signals with
differentially encoded phases can also be received when appropriate data recovery
methods are employed.
In Sect. 3.2.1 and Sect. 3.2.2, different receiver configurations with direct detec-
tion are described. One option is to use DLIs for phase detection, another option to
employ a 2 × 4 90◦ hybrid combined with a delay of one symbol duration in front
3.2 Direct Detection Receiver with Delay Line Interferometers 67
of one of the hybrid inputs. Subsequently, Sect. 3.2.3 illustrates for a wide range
of modulation formats how data recovery can be accomplished in receivers with
differential detection.
3.2.1 Direct Detection Receiver with DLIs
The usual way for constructing direct detection receivers is employing DLIs to
convert differential phase modulation into intensity modulation before photodiode
square-law detection.
Multiple DLI Receiver
One receiver option—whose optical part is shown in Fig. 3.3—is to use N ph /2

DLIs with appropriate phase shifts, where N ph represents the number of phase
states (N ph = M for MDPSK signals). For the detection of DPSK signals, only the
branch with the DLIs (phase detection branch) is needed. Another branch (intensity
detection branch) must be provided for a separate evaluation of the intensity when
detecting Star QAM signals. Phase information can be demodulated by performing
binary decisions on the resulting N ph /2 electrical photocurrents.
Intensity detection branch Intensity
Phase detection branch
DLI
1
DLI
2
3dB To data
1:Nph /2 recovery
DLI
Nph /2-1
DLI
Nph /2
BD
Fig. 3.3 Direct detection receiver composed of N ph /2 DLIs; for the detection of DPSK signals the
phase detection branch is sufficient. A separate intensity detection branch must be provided in the
case of the detection of Star QAM signals. BD: balanced detector
The receiver concept with multiple DLIs was investigated for 8DPSK in [71].
Unfortunately, the optical effort becomes quite high for modulation formats with a
high number of phase states. Four DLIs are needed for 8DPSK, and as many as eight
DLIs for 16DPSK. For this reason, the following description of direct detection
receivers composed of DLIs is restricted to a configuration with only two DLIs.
IQ Receiver with DLIs
Generally, two DLIs are sufficient to obtain the phase difference information of
arbitrary DPSK and Star QAM signals by detecting their in-phase and quadrature
components (IQ receiver). However, decisions on multi-level electrical signals with
multiple thresholds become necessary for modulation formats with N ph > 4 in that
case. The setup of the optical receiver part for Star QAM is shown in Fig. 3.4.

I ID (t )
EI ,1 (t )
TS In-phase
E s (t ) EI (t ) ϕ DLI I
EI ,2 (t ) I I (t )
3dB
3dB EQ ,1 (t )
TS Quadrature
EQ (t ) ϕ DLI Q EQ ,2 (t ) IQ (t )
DLI BD
Fig. 3.4 Direct detection IQ receiver composed of two DLIs; in the case of the detection of DPSK
signals, the phase detection branch is sufficient. A separate intensity detection branch must be
provided for the detection of Star QAM signals.
The received signal E s (t), which is specified by (3.8) when neglecting all trans-
mission impairments, is split by a coupler into the intensity detection branch and
the phase detection branch. Here, the coupler is assumed to be a 3 dB coupler. The
output current of the photodiode in the intensity detection branch is
E s (t) E s∗ (t) R
I I D (t) = R · √ · √ + i sh + i th = · a 2 (t) · Ps + i sh + i th . (3.23)
2 2 2
This photocurrent is proportional to the CW power Ps and the square of the mod-
ulation amplitude a 2 (t). Thus, the power of a symbol is detected in the intensity
detection branch. In (3.23), R represents the responsivity of the photodiode, which
is equal to
2π e
R=η , (3.24)
hωs
3.2 Direct Detection Receiver with Delay Line Interferometers 69
where e = 1.6 · 10−19 C is the charge per electron, hωs /2π is the energy per photon
with h = 6.63 · 10−34 J s being the Planck constant, and η is the quantum efficiency
of the photodiode that corresponds to the average number of electrons generated
per photon. The two variables i sh and i th in (3.23) describe the photocurrent shot-
noise

2 and
the thermal noise of the
2 receiver, respectively. Their variances are given
by i sh = 2 · e · hI I D i · Bel and i th = 4 · k B · T · G r · Bel , where Bel represents the
electrical receiver bandwidth, k B is the Boltzmann constant, T is the temperature,
and G r is the resistor’s conductance of the electrical circuit [66].
To enhance the sensitivity, an optical pre-amplifier, commonly followed by an
optical filter, is typically placed in front of the DD receiver (not shown in Fig. 3.4).
The complex noise of the optical amplifier, whose baseband representation is de-
fined here as n(t) = n k (t)ek + n ⊥ (t)e⊥ , can be partitioned into noise components
polarized in parallel and orthogonally with respect to the signal, denoted as n k (t)
and n ⊥ (t), and can be assumed to be additive white Gaussian noise (AWGN) with
a two-sided power spectral density (PSD) per polarization of
N0 = n sp (G − 1) h · f s , (3.25)
where n sp represents the spontaneous emission parameter and G is the amplifier

gain [51, 66]. The variance of the in-phase and quadrature components of n k (t) and
n ⊥ (t) is given by σ 2 = σ I2 = σ Q2 = (N0 /2)· Bn , where Bn is the optical noise band-
width. When taking the amplifier noise into account but neglecting optical filtering,
the photocurrent in the intensity detection branch becomes
R h ih i∗
I I D (t) = Es (t) + n(t)e jωs t Es (t) + n(t)e jωs t + i sh + i th
2
R h i
= |Es (t)|2 + Es (t)n∗ (t)e− jωs t + E∗s (t)n(t)e jωs t + |n(t)|2 + i sh + i th
2
R Rh i
= |E s (t)|2 + E s (t)n ∗k (t)e− jωs t + E s∗ (t)n k (t)e jωs t
2 | {z } 2 | {z }
Signal power
Signal-ASE noise
R h i
+ n k (t)2 + |n ⊥ (t)|2 +i sh + i th , (3.26)
2 | {z }
ASE-ASE noise
In comparison with (3.23), two new noise components arise which are the signal-
ASE noise and the ASE-ASE noise. For the detection of symbols with sufficiently
high signal power and for small optical noise bandwidth, the signal-ASE noise be-
comes dominant compared with the ASE-ASE noise. The shot-noise and the thermal
noise can be usually neglected in systems with optical amplification.
The following description intends to demonstrate the general functionality of the
direct detection receiver for high-order modulation, so all the noise sources men-
tioned are not further regarded. A more detailed discussion of the impact of ampli-
fier noise on the receiver performance, where optical and electrical filtering and the
whole receiver structure are considered, is carried out in Sect. 5.4.
As illustrated in Fig. 3.4, the second output of the coupler at the receiver input
meets the phase detection branch. In the phase detection branch, the signal is split
by another coupler for detection of the in-phase and quadrature components. When
assuming 3 dB couplers and taking the phase shifts of the couplers into account, two
fields with a phase difference of 90◦ are obtained at the upper and the lower output
of the second coupler
1 p ◦
E I (t) = · a(t) · Ps · e j(ωs t+ϕs ) · e jϕ(t) · e jϕns (t) · e j90 , (3.27)
2
1 p ◦
E Q (t) = · a(t) · Ps · e j(ωs t+ϕs ) · e jϕ(t) · e jϕns (t) · e j180 . (3.28)
2
When looking at the in-phase branch and considering the transfer characteristic of
the DLI defined by (3.6) and (3.7), the electrical fields obtained at the two DLI
outputs are
1 p ◦
E I,1 (t) = · a(t − TS ) · Ps · e j(ωs t+ϕs ) · e jϕ(t−TS ) · e jϕns (t−TS ) · e j90
4
1 p ◦
+ a(t) Ps · e j(ωs t+ϕs ) · e jϕ(t) · e jϕns (t) · e− j90 · e jϕ DL I I , (3.29)
4
1 p ◦
E I,2 (t) = · a(t − TS ) · Ps · e j(ωs t+ϕs ) · e jϕ(t−TS ) · e jϕns (t−TS ) · e j180
4
1 p ◦
+ a(t) Ps · e j(ωs t+ϕs ) · e jϕ(t) · e jϕns (t) · e j180 · e jϕ DL I I , (3.30)
4
with ϕ DL I I representing the phase shift of the DLI in the in-phase branch. When
detecting the fields given by (3.29) and (3.30), the two photocurrents within the
balanced detectors (BD) are given by
R R
I I,1 (t) = R · E I,1 (t) · E ∗I,1 (t) =
Ps · a 2 (t) + Ps · a 2 (t − TS )
16 16
R
− · Ps · a(t)a(t − TS ) · cos 1ϕ(t) + 1ϕn s (t) + ϕ DL I I , (3.31)
8
R R
I I,2 (t) = R · E I,2 (t) · E ∗I,2 (t) =
Ps · a 2 (t) + Ps · a 2 (t − TS )
16 16
R
+ · Ps · a(t)a(t − TS ) · cos 1ϕ(t) + 1ϕn s (t) + ϕ DL I I , (3.32)
8
where the difference of the modulation phase of two consecutive symbols, 1ϕ(t),
and the phase error due to laser phase noise, 1ϕn s (t), were defined in (3.11). The
variance of the phase error is given by h1ϕn2s i = 2π 1νs TS .
3.2 Direct Detection IQ Receiver with 2 × 4 90◦ Hybrid 71
Within the balanced detector, both photocurrents are subtracted. The first two
terms are the same in (3.31) and (3.32), and disappear after balanced detection. The
photocurrent after balanced detection becomes
I I (t) = I I,2 (t) − I I,1 (t)

R
= · Ps · a(t)a(t − TS ) · cos 1ϕ(t) + 1ϕn s (t) + ϕ DL I I . (3.33)
4
In the same manner, when performing the same calculations for the quadrature
branch, the quadrature photocurrent is obtained as
I Q (t) = I Q,2 (t) − I Q,1 (t)

R
= · Ps · a(t)a(t − TS ) · cos 1ϕ(t) + 1ϕn s (t) + ϕ DL I Q . (3.34)
4
In (3.34), ϕ DL I Q denotes the phase shift of the DLI in the quadrature branch. When
detecting DQPSK signals, for instance, ϕ DL I I should be set to −45◦ and ϕ DL I Q
to −135◦ to enable that decisions on binary electrical signals in the in-phase and
quadrature arms can be performed. Generally, by choosing the phase shifts of the
DLI as ϕ DL I I = 0◦ and ϕ DL I Q = −90◦ , the in-phase and quadrature components
of arbitrary DPSK constellations can be obtained. When further neglecting the laser
phase noise, the in-phase and quadrature photocurrents are
R
I I (t) = · Ps · a(t)a(t − TS ) · cos [1ϕ(t)] , (3.35)
4
R
I Q (t) = · Ps · a(t)a(t − TS ) · sin [1ϕ(t)] . (3.36)
4
For DPSK formats, a(t) and a(t −TS ) are not modulated, so that phase difference
information 1ϕ(t) can easily be recovered from the photocurrents given by (3.35)
and (3.36). By contrast, the dependence of (3.35) and (3.36) on the non-delayed
and delayed modulation amplitudes can complicate the recovery of phase difference
information for Star QAM. Nevertheless, data recovery can be accomplished for
arbitrary Star QAM modulation formats with differentially encoded phases, as will
be shown in Sect. 3.2.3.
3.2.2 Direct Detection IQ Receiver with 2 × 4 90◦ Hybrid
Another IQ receiver with direct detection can be composed of a 2 × 4 90◦ hybrid

with an additional delay of one symbol duration in front of one of the hybrid inputs.
This alternative receiver configuration is functionally equivalent to the direct detec-
tion receiver with two DLIs and is illustrated in Fig. 3.5.

I ID (t )
In-phase
Es (t ) Ein 1 (t ) I I (t )
3dB 2x4
ψ# 90°
3dB
TS Hybrid Quadrature
Ein 2 (t ) IQ (t )
BD
Fig. 3.5 Direct detection IQ receiver whose setup is based on a 2 × 4 90◦ hybrid; for the detection
of DPSK signals the phase detection branch is sufficient. A separate intensity detection branch
must be provided for the detection of Star QAM signals.
A part of the receiver, which consists of the intensity detection branch and two
couplers, is identical to the DLI configuration. When observing the phase detection
branch and assuming the use of 3 dB couplers, the input fields of the 2×4 90◦ hybrid
are given by
1 p ◦
E in 1 (t) = · a(t) · Ps · e j(ωs t+ϕs ) · e jϕ(t) · e jϕns (t) · e j90 · e j ψ̃ , (3.37)
2
1 p ◦
E in 2 (t) = · a(t − TS ) · Ps · e j(ωs t+ϕs ) · e jϕ(t−TS ) · e jϕns (t−TS ) · e j180 , (3.38)
2
where the angle ψ̃ corresponds to an arbitrary phase shift in front of the upper
hybrid input. The fields described by (3.37) and (3.38) combine in the optical 2 × 4
90◦ hybrid, whose field transfer function is defined in (3.14) and (3.15). The phase
shift ψ introduced there is set to zero for all following calculations. Altering the
arbitrary phase shift ψ̃ in front of the hybrid input has the same effect on the hybrid
output signals as altering the phase shift ψ within the hybrid would have. The four
optical signals at the hybrid outputs are detected by two balanced detectors. When
connecting the appropriate hybrid outputs to the proper balanced detector inputs
and neglecting amplifier noise, shot-noise, thermal noise and laser phase noise, the
following photocurrents can easily be derived after the balanced detectors
R h i
I I (t) = Iout4 (t) − Iout2 (t) = · Ps · a(t)a(t − TS ) · cos 1ϕ(t) + ψ̃ , (3.39)
4
R h i
I Q (t) = Iout1 (t) − Iout3 (t) = · Ps · a(t)a(t − TS ) · sin 1ϕ(t) + ψ̃ . (3.40)
4
It becomes apparent that the equations (3.39) and (3.40) are identical to (3.35) and
(3.36) if the phase shift ψ̃, which can also be interpreted as a relative phase shift
3.2 Data Recovery for Differential Detection 73
between the upper and the lower input of the hybrid, is equal to zero. An advantage
of the hybrid configuration, compared with the configuration with two DLIs, is that
the 2 × 4 90◦ hybrid inherently features the orthogonal relation between the in-
phase and quadrature axes. By adjusting the phase shift ψ̃, the whole received con-
stellation diagram can be rotated (the same effect can be achieved by setting the
phase shift ψ when simultaneously adjusting the phase shifts in the upper and lower
branches of the hybrid composed of 3 dB couplers and phase shifters, as described in
Sect. 3.1.3). In the DLI configuration, however, the phase shifts in both DLIs have
to be adjusted correctly to ensure appropriate orthogonality as well as the correct
absolute phase. Finally, choosing a configuration will depend on practical imple-
mentation considerations.
3.2.3 Data Recovery for Differential Detection
This section describes the data recovery for modulation formats with differentially
encoded phases whose differential phase information is determined by IQ receivers
with differential detection. Differential detection can be accomplished in the optical
domain as shown for direct detection receivers in Sect. 3.2.1 and Sect. 3.2.2, or in the
electrical domain after coherent detection as described in Sect. 3.4. Because phase
differences are identified, no differential decoder is needed at the receiver when
the phase was differentially encoded at the transmitter. The datarecovery circuits
derived in this section are appropriate for recovering the data bits b1k , b2k , .., bm k
of systems employing the bit mappings defined in Chap. 2. Figure 3.6 illustrates two
different data recovery techniques which can be employed in optical IQ receivers
with differential detection.
a ARG-DECISION
In-phase b1k
LPF s1k
Multiplexer
Operation
Recovery
Data
Logic
Data
ARG
….
….
Quadrature
LPF s Nphk
bmk
b IQ-DECISION
si , nk
In-phase b1k
Multiplexer
LPF
….
Data
Recovery
Logic
Data
….
Quadrature
….
LPF
s q , nk bmk
Fig. 3.6 Data recovery techniques applicable within the phase detection branch of DPSK receivers.
a Arg-decision. b IQ-decision. LPF: low-pass filter
Usually, the electrical in-phase and quadrature photocurrents in the phase detec-
tion branch are low-pass filtered beforehand to achieve a better BER performance.
Then they can be processed by the techniques shown in Fig. 3.6 to recover the data
bits corresponding to the phase information. These techniques differ in the position-
ing of the thresholds, their practical implementation and their performance.
For maximum distances between symbols and thresholds and for optimal noise
performance, the thresholds should be positioned radially between the phase states
as shown for 8DPSK and 16DPSK in Fig. 3.7c and Fig. 3.7d, respectively. This
decision scheme is denoted here as “arg-decision”. The electrical receiver config-
uration for arg-decision is illustrated in Fig. 3.6a. By performing an arg-operation
(calculation of the angle of a complex value) on the electrical in-phase and quadra-
ture signals, a multi-level electrical signal whose states represent the received phase
differences arises. A decision is performed on these multi-level signals, by apply-
ing the radially positioned thresholds. The outcome of the decision is processed by
appropriate data recovery logic to obtain the data bits, which, in the end, can be
multiplexed to obtain the original data stream. In practice, the arg-operation could
be implemented by digital means. The performance then depends on the resolution
of the analog-to-digital (A/D) converter used.
Another option is performing the decisions directly on the in-phase and quadra-
ture photocurrents as depicted in Fig. 3.6b. This technique is denoted here with
“IQ-decision”. For DBPSK and DQPSK, decisions are performed on binary sig-
nals with an optimal threshold at zero. For higher-order DPSK formats, the in-phase
and quadrature signals are multi-level and the decision circuits have multiple thresh-
olds, as shown for 8DPSK in Fig. 3.7b and discussed in [49]. The thresholds must
be adjusted according to the signal power received. The binary data at the output of
the decision modules must be processed by appropriate data recovery logic which is
different from the one for arg-decision. For IQ-decision, the thresholds are not op-
timally placed between the symbols, so a worse noise performance can be expected
than for ideal arg-decision.
To better understand the reconstruction of the information, the data recovery pro-
cess is illustrated in greater detail in the following paragraphs. In the decision cir-
cuits, incoming signals are sampled once per symbol at sample times tk . In the case
of IQ-decision, this provides the electrical in-phase and quadrature signal samples
I Ik and I Q k . The decision results si,n k and sq,n k (logical one or zero) for IQ-decision
in the k-th symbol interval at the outputs of the decision circuit in the in-phase and
quadrature arms are specified by

1 I Ik > si,n 1 I Q k > sq,n
si,n k = , sq,n k = , (3.41)
0 I Ik < si,n 0 I Q k < sq,n
where si,n and sq,n represent the thresholds in the in-phase and quadrature branches,
respectively.
a DQPSK, IQ-DECISION b 8DPSK, IQ-DECISION
{b1 , b2 } 011 {b1 , b2 , b3 }

10 11 010 001
Sq,1
Sq,1 110 000
Sq,2
00 01 111 100
101
Si,1 Si,2 Si,1
c 8DPSK, ARG-DECISION d 16DPSK, ARG-DECISION
S5 S4
S3 S2 S6 S3
011
{b1 , b2 , b3 } {b1 , b2 , b3 , b4 }
0111 0110 0010
S7
010 0101 0011 S2
001
S4 S1
0100 0001
S8 S1
110 000 1100 0000

S9 S16
1101 1000
S5 S8
111 100 S10 1111 1001 S15
101 1110 1010 1011

S11 S14
S6 S7 S12 S13
Fig. 3.7 Positioning of the thresholds for DPSK data recovery in the case of IQ-decision (a,b) and
arg-decision (c,d)
When looking specifically at the data recovery for DQPSK and using the bit
mapping shown in Fig. 3.7a, the first bit b1k is obtained from the decision result sq,1k
at threshold sq,1 (decision circuit in the Q-branch) and the second bit b2k is equal
to the decision result si,1k at threshold si,1 (decision circuit in the I-branch), placing
both thresholds optimally at zero. IQ-decision for 8DPSK can be accomplished by
placing two thresholds in the in-phase and the quadrature arms respectively. The
positioning of these thresholds is illustrated in Fig. 3.7b. The data in the k-th symbol
interval can be obtained from a data recovery logic defined by
b1k = sq,2k + si,2k sq,1k , (3.42)
b2k = si,2k + si,1k sq,1k , (3.43)
b3k = si,1k sq,2k + si,2k sq,1k . (3.44)

In the case of arg-decision, the phase difference information is determined using

the electrical in-phase and quadrature signal samples by calculating

1ϕk = arg I Ik , I Q k . (3.45)
The signal samples after the arg-operation can be evaluated by a decision module
with N ph thresholds, where N ph denotes the number of phase states. The thresholds
sn should be optimally placed at sn = (2n − 1)π/N ph for n = (1, .., N ph /2) and
sn = (2n − 1)π/N ph − 2π for n = (N ph /2 + 1, .., N ph ). The decision results sn k
for arg-decision at the N ph outputs of the decision circuit in the symbol interval k
are

1 1ϕk > sn
sn k = . (3.46)
0 1ϕk < sn

The data recovery logic required to obtain the data bits b1k , b2k , .., bm k is different
from the one required for IQ-decision. For 8DPSK and the bit mapping shown in
Fig. 3.7c, the data recovery circuit is specified by
b1k = s4k + s8k , (3.47)
b2k = s2k + s6k , (3.48)
b3k = s1k s3k + s5k s7k . (3.49)
When looking at 16DPSK, IQ-decision can be expected to exhibit very poor

performance due to small distances between the signal levels and the thresholds in
the in-phase and quadrature arms. Therefore, only arg-decision is considered here.
With the thresholds and bit mappings illustrated in Fig. 3.7d, the four information
bits b1k , b2k , b3k and b4k can be recovered with the data recovery logic
b1k = s8k + s16k , (3.50)
b2k = s4k + s12k , (3.51)
b3k = s2k s6k + s10k s14k , (3.52)
b4k = s1k s3k + s5k s7k + s9k s11k + s13k s15k . (3.53)
The remainder of this section deals with the data recovery for Star QAM formats
[58]. As can be observed from the equations (3.35) and (3.36) or (3.39) and (3.40)
respectively, the in-phase and quadrature photocurrents are not only dependent on
the phase difference between consecutive symbols but also on the non-delayed and
delayed modulation amplitudes a(t) and a(t − TS ). These amplitudes are ideally
equal for DPSK formats and then do not affect the evaluation of the in-phase and
quadrature signals for designation of the phase difference information. However, for
Star QAM formats, the product a(t) · a(t − TS ) can take different values which de-
pend on the amplitudes of two consecutive symbols. To illustrate this, the constella-
tion diagrams received in the phase detection branch, constructed from the in-phase
and quadrature photocurrents after differential detection, are shown in Fig. 3.8 for
Star 8QAM (ASK-DQPSK) and Star 16QAM (ASK-8DPSK). It can be observed
that what were originally two amplitude states have changed to three possible am-
plitude states as a result of the differential detection process.
IQ IQ
II II
Fig. 3.8 Star QAM constella-
tion diagrams received in the
phase detection branch after Star 8QAM Star 16QAM
differential detection (ASK-DQPSK) (ASK-8DPSK)
With the arg-decision scheme it is still possible to determine the phase difference
information by performing an arg-operation as described in (3.45). To recover the
bits encoded in the phase differences, the same data recovery logic can be used as
for the DPSK format with the same number of phase states. 8DPSK data recovery
is appropriate for Star 16QAM, for instance. The bit encoded in the amplitude can
simply be recovered by a binary decision in the intensity detection branch.
Data recovery using IQ-decision, where the signals in the in-phase and quadra-
ture branches are evaluated separately, is only easy to accomplish for Star QAM
formats with a maximum of four phase states. When the phases are positioned at
π/4 + n · π/2 (n = 0, 1, 2, 3), phase difference information can be obtained by
performing binary decisions at zero on the signals in the in-phase and quadrature
branches. But this works only if the thresholds can be set to zero and just one phase
information bit must be recovered in each branch. Thus, this method is limited to
Star QAM formats with only four phase states. It was used for ASK-DQPSK in [49]
and for 16APSK (4ASK-DQPSK) in [62].
Additional effort must be undertaken to be able to perform IQ-decision for Star
QAM signals with more than four phase states. In that case, the information from the
intensity detection branch can be used to normalize the constellation and get rid of
the unwanted amplitude information within the in-phase and quadrature photocur-
rents [58]. This process is denoted as “normalization” throughout this book and is
illustrated in Fig. 3.9 for two different implementation options.
a b
Normalization to original QAM constellation Normalization to one DPSK circle
In-phase x In-phase x x
x/y x/y x/y
y y y
Intensity Intensity
TS TS
y y y
Quadrature x Quadrature x x
x/y x/y x/y
Fig. 3.9 ”Normalization”, necessary when performing IQ-decision for Star QAM formats with
more than four phase states
For the implementation option depicted in Fig. 3.9a, the received constellation is
normalized to the original QAM constellation by dividing the in-phase and quadra-
ture photocurrents given by (3.35) and (3.36) by the square root of the signal from
the intensity branch defined by (3.23), which is delayed by one symbol duration.
After this process, the in-phase and quadrature photocurrents are given by
r r
R · Ps R · Ps
I I (t) = · a(t) · cos [1ϕ(t)] , I Q (t) = · a(t) · sin [1ϕ(t)] . (3.54)
8 8
These signals represent the original Star QAM constellation and can be evaluated
using any conventionally employed decision scheme for Star QAM formats. An-
other option is normalizing the received constellation to one circle by dividing the
in-phase and quadrature signals by the square root of the non-delayed and the de-
layed signal from the intensity branch, as illustrated in Fig. 3.9b. This way, the
signals after normalization become independent of the modulation amplitudes.
1 1
I I (t) = · cos [1ϕ(t)], I Q (t) = · sin [1ϕ(t)]. (3.55)
2 2
Using the in-phase and quadrature signals given by (3.55), the data bits correspond-
ing to the phase information can be determined in the same way as for DPSK formats
with the same number of phase states.
Arbitrary DPSK and Star QAM formats with differentially encoded phases can
be demodulated by applying the methods described in this section. The practical
implementation of critical operations such as normalization and arg-calculation is
a challenge for high data rates and can prospectively be performed by using high-
speed digital signal processing technology.
3.3 Coherent Detection Principle 79
3.3 Fundamentals of Coherent Detection
When employing coherent detection, all the information of the optical signal wave
(amplitude, frequency, phase and polarization) is transferred to the electrical do-
main. In the case of the detection of optical signals with high-order modulation, this
has the advantage that demodulation can be performed completely electrically, and
optical complexity—in terms of interferometric optical demodulation structures for
the conversion of phase to intensity modulation—can be reduced. Moreover, coher-
ent receivers exhibit enhanced possibilities for electronic compensation of transmis-
sion impairments and can be used as flexible tunable WDM receivers with highly
selective channel separation. On the other hand, coherent detection does not only
provide for the availability of the desired field parameters in the electrical domain
but also necessitates a controlled state of the remaining field parameters in order to
be able to evaluate the information in demand.
3.3.1 Coherent Detection Principle
To better understand the principles of coherent detection and see how the field pa-
rameters of the optical information signal are made available in the electrical do-
main, let us first observe the simple structure shown in Fig. 3.10.
I I ,1 (t )
EI,1 (t)
Es (t)
3dB I I (t )
Elo (t)
EI,2 (t) BD
Fig. 3.10 Principle of coher- I I ,2 (t )
LO
ent detection
It becomes apparent that the optical information signal is interfered with the light
of a LO laser before photo-detection, e.g. in a 3 dB coupler. With the normalized
electrical field of the received optical information signal defined like before as
p
Es (t) = Ps · e j (ωs t+ϕs ) · a(t) · e jϕ(t) · e jϕns (t) · es , (3.56)
and the CW light of the LO given by

p
Elo (t) = Plo · e j (ωlo t+ϕlo ) · e jϕnlo (t) · elo , (3.57)
where Ps and Plo represent the CW powers, ωs and ωlo are the angular frequencies,
ϕs and ϕlo are the initial phases, ϕn s (t) and ϕnlo (t) represent the laser phase noise,
and es and elo are the polarization unit vectors of the signal and the LO light, the
resulting photocurrents within the balanced detector are
1 1 1
I I,1 (t) = R · (Es + jElo ) · (Es + jElo )∗ + i sh 1 = R Ps · a 2 (t) + R Plo + i sh 1
2 p 2 2
+R Ps Plo · a(t) · es elo · sin [1ωt + ϕn (t) + ϕ0 + ϕ(t)], (3.58)
1 1 1
I I,2 (t) = R · ( jEs + Elo ) · ( jEs + Elo )∗ + i sh 2 = R Ps · a 2 (t) + R Plo + i sh 2
2 p 2 2
−R Ps Plo · a(t) · es elo · sin [1ωt + ϕn (t) + ϕ0 + ϕ(t)], (3.59)
where i sh 1 and i sh 2 represent the shot-noise photocurrents of the two photodiodes,

and the angular frequency offset 1ω, the overall laser phase noise ϕn (t) and the
initial phase offset ϕ0 are defined as
1ω = ωs − ωlo , ϕn (t) = ϕn s (t) − ϕnlo (t), ϕ0 = ϕs − ϕlo . (3.60)
The variance of the random phase change due to the overall laser phase noise within
a time interval τ , 1ϕn (t) = ϕn (t) − ϕn (t − τ ), can be calculated by
h1ϕn2 (τ )i = 2π 1νe f f |τ |, (3.61)
where 1νe f f is the beat-linewidth and given by the sum of the linewidth of the
signal laser, 1νs , and the linewidth of the LO, 1νlo , [17]
1νe f f = 1νs + 1νlo . (3.62)
The first and second terms are equal in (3.58) and (3.59) and represent the directly
detected signal power and LO power, respectively. They are not interesting for eval-
uating signal information because it is usually true that Ps << Plo . During the
balanced detection process (assuming an ideal symmetric balanced detection) they
disappear anyway and only the beating term which contains all the field parameters
of the superimposed optical fields remains
I I (t) = I I,1 (t) − I I,2 (t)

p
= 2R Ps Plo · a(t) · es elo · sin [1ωt + ϕn (t) + ϕ0 + ϕ(t)] + i sh .(3.63)
In (3.63), i sh = i sh 1 − i sh 2 is the overall

2 shot-noise photocurrent after balanced
detection, whose variance is given by i sh 2 i + hi 2 i when assuming uncor-
= hi sh 1 sh 2
related noise events of the two photodiodes.
Two fundamental coherent detection principles can be distinguished: homodyne
and heterodyne detection. In the case of homodyne detection, the carrier frequencies
of the signal laser and the LO laser aspire to be identical and the optical spectrum is
directly converted to the electrical baseband. One of the main challenges for homo-
dyne receivers, especially for homodyne synchronous (non-differential) detection,
is the implementation of the carrier synchronization which synchronizes the carrier
frequencies and phases of the signal laser and the LO.
3.3 Coherent Detection Principle 81
In the case of heterodyne detection, the frequencies of the signal laser and the
LO are chosen to be different, so that the field information of the optical signal
wave is transferred to an electrical carrier at an intermediate frequency (IF) which
corresponds to the frequency difference of the signal laser and the LO (1ω). The
IF must be at least as high as the baseband bandwidth of the information signal. On
the one hand, heterodyne detection permits simple demodulation schemes and en-
ables carrier synchronization with an electrical phase locked loop. On the other, the
occupied electrical bandwidth for heterodyne detection is more than twice as high
as for homodyne detection, and image-rejection techniques are required to allow
for acceptable spectral efficiencies for WDM. Therefore, when considering spectral
efficiency and the practical feasibility at high data rates, homodyne receivers are
superior to their heterodyne counterparts and seem to be a better choice for future
optical networks. For this reason, coherent receiver configurations described in this
book are restricted to homodyne detection. Actual research about heterodyne de-
tection of optical signals with high-order modulation can be found in [20, 41, 42],
where very high-order Square QAM transmission has been experimentally demon-
strated at moderate symbol rates.
From (3.63) it can be observed that for evaluation of the signal information in
demand (for phase and quadrature amplitude modulation: the modulation phase ϕ(t)
and / or the normalized modulation amplitude a(t)) all the remaining parameters
must be controlled to take defined states. In homodyne receivers, the frequencies of
the signal laser and the LO aspire to be equal, whereas a stable IF at 1ω = ωs − ωlo
is desired in heterodyne receivers. An uncompensated difference of the initial phases
ϕ0 causes a fixed phase offset. The overall phase noise ϕn (t) leads to a random walk
of the phase which corresponds to a permanent rotation of the received constellation
diagram in arbitrary directions for quadrature detection of high-order modulation
signals. Furthermore, the polarizations of the signal laser and the LO have to be
aligned properly to obtain a maximal photocurrent.
The availability of polarization information provides an auxiliary degree of free-
dom for optical transmission systems. Pure polarization modulation schemes such
as M-ary polarization shift keying (MPOLSK), where information is encoded into
different states of polarization [4], and combined QAM and polarization modula-
tion (M-4Q-QAM), where all the four quadratures of the optical field are exploited
[9], can be realized. A more simple way of making use of polarization is given by
polarization division multiplexing (PDM). Here, the spectral efficiency of any mod-
ulation format can be approximately doubled.
The derivation of the photocurrent given by (3.63) ignores the influence of the
thermal noise of the receiver. For sufficiently high LO powers, the thermal noise can
be neglected [13] and the shot-noise becomes the dominant noise source (shot-noise
limited detection). The determination of the shot-noise limited receiver performance
is the traditional way of characterizing coherent receivers. However, optical trans-
mission systems with coherent detection and optical amplifiers on the link or in
front of the receiver are no longer shot-noise limited but limited by the amplifier
noise (amplifier noise limited detection). The influence of amplifier noise on the
coherent detection process is briefly examined in Sect. 3.3.2.
3.3.2 Coherent Detection with Amplifier Noise
When the noise of an optical amplifier, whose low-pass representation was defined
in Sect. 3.2.1 as n(t) = n k (t)ek + n ⊥ (t)e⊥ , is incorporated into the coherent de-
tection process, the electrical field at the upper input of the balanced detector of the
receiver shown in Fig. 3.10 is given by
1 h i
E I,1 (t) = √ · Es (t) + n(t)e jωs t + jElo (t) , (3.64)
2
where Es (t) and Elo (t) were defined by (3.56) and (3.57), respectively. The resulting
photocurrent at the upper photodiode within the balanced detector—when the same
polarizations of the received signal and the LO laser (es = elo ) are assumed, for
simplicity—is specified by
I I,1 (t) = R · E I,1 (t) · E∗I,1 (t) + i sh 1

Rh ∗
i
= |Es (t)|2 + |Elo (t)|2 − jEs (t)Elo (t) + jE∗s (t)Elo (t)
2
Rh i
+ Es (t)n∗ (t)e− jωs t + E∗s (t)n(t)e jωs t + |n(t)|2
2
Rh i
+ jElo (t)n∗ (t)e− jωs t − jElo ∗
(t)n(t)e jωs t + i sh 1
2
R R R ∗

= |E s (t)|2 + |Elo (t)|2 + − j E s (t)Elo (t) + j E s∗ (t)Elo (t)
2 | {z } 2 | {z } 2 | {z }
Signal power LO power Signal-LO beating
Rh i R h
n k (t)2 + |n ⊥ (t)|2
i
+ E s (t)n ∗k (t)e− jωs t + E s∗ (t)n k (t)e jωs t +
2 | {z } 2 | {z }
Signal-ASE noise ASE-ASE noise
Rh i
+ j Elo (t)n ∗k (t)e− jωs t − j Elo
∗
(t)n k (t)e jωs t +i sh 1 . (3.65)
2 | {z }
LO-ASE noise
The first three terms (signal and LO power, signal-LO beating) describe the pho-
tocurrent without amplifier noise given by (3.58). The signal-ASE noise and the
ASE-ASE noise are known from the direct detection receiver discussed in Sect. 3.2.1.
In coherent receivers, the additional LO-ASE noise is usually dominant compared
with all the other noise components [11, 17].
In the same manner, the photocurrent of the lower photodiode within the balanced
detector I I,2 (t) can be derived. The electrical field at the lower input of the balanced
detector is
1 h i
E I,2 (t) = √ · jEs (t) + jn(t)e jωs t + Elo (t) , (3.66)
2
3.3 Coherent Detection with Amplifier Noise 83
and the photocurrent of the lower photodiode within the balanced detector can be
described by
I I,2 (t) = R · E I,2 (t) · E∗I,2 (t) + i sh 2

R R R ∗

= |E s (t)|2 + |Elo (t)|2 + j E s (t)Elo (t) − j E s∗ (t)Elo (t)
2 2 2
Rh i R h
n k (t)2 + |n ⊥ (t)|2
i
+ E s (t)n ∗k (t)e− jωs t + E s∗ (t)n k (t)e jωs t +
2 2
Rh i
+ − j Elo (t)n ∗k (t)e− jωs t + j Elo ∗
(t)n k (t)e jωs t + i sh 2 . (3.67)
2
When neglecting all noise components except for the LO-ASE noise and assuming
full correlation of the noise events of both branches, the photocurrent after balanced
detection is given by
I I (t) = I I,1 (t) − I I,2 (t)

∗

= R − j E s (t)Elo (t) + j E s∗ (t)Elo (t)
h i
+R j Elo (t)n ∗k (t)e− jωs t − j Elo ∗
(t)n k (t)e jωs t
p
= 2R Ps Plo · a(t) · sin [1ωt + ϕn (t) + ϕ0 + ϕ(t)]
| {z }
Signal-LO beating
n o
∗
+ 2R · Re − j Elo (t)n k (t)e jωs t . (3.68)
| {z }
LO-ASE noise
The signal and LO powers have disappeared after balanced detection. The signal
information is contained in the signal-LO beating term which is the same as derived
before in (3.63). The LO-ASE noise can be written as
n o
∗
n L O−AS E (t) = 2R · Re − j Elo (t)n k (t)e jωs t
n o
∗
= 2R · Re − j Elo (t)e jωs t · Re n k (t) + j I m n k (t)
p
= 2R Plo · Re n k (t) · sin 1ωt − ϕnlo (t) − ϕlo
p
+2R Plo · I m n k (t) · cos 1ωt − ϕnlo (t) − ϕlo . (3.69)

The variances of Re n k (t) and I m n k (t) are given by σ I2 = σ Q2 = (N0 /2) · Bn .
When neglecting the phase noise of the LO laser ϕnlo (t), the LO-ASE noise can be
interpreted as Gaussian noise with a power spectral density of
N L O−AS E = 2R 2 Plo N0 . (3.70)
Compared with direct detection, where the power of the signal-ASE noise depends
on the power of the detected symbols, the noise is symbol-power independent for
coherent detection limited by LO-ASE noise.
3.3.3 Optical Quadrature Frontend
To be able to detect the in-phase and quadrature components of high-order optical

modulation signals, an optical quadrature frontend must be employed, where the
signal and LO light are superposed in a 2 × 4 90◦ hybrid whose output signals are
detected by two balanced detectors. The setup of the optical quadrature frontend is
shown in Fig. 3.11.
Eout1 (t )
In-phase
I I (t )
Es (t) 2x4 Eout3 (t )

Elo (t) 90°
Hybrid Eout2 (t )
Fig. 3.11 Optical quadra- Quadrature
ture frontend for coherent
IQ (t )
detection of the in-phase and
quadrature components of Eout4 (t )
high-order modulation signals BD
The field transfer function of the hybrid is defined by (3.14) and (3.15). When
setting the phase shift ψ to zero and neglecting phase shifts commonly experienced
by both input fields, the hybrid output fields (without amplifier noise) are
   
Eout1 (t) Es (t) + Elo (t)
Eout (t) 1 Es (t) + jElo (t)
2
Eout3 (t) = 2 ·  Es (t) − Elo (t) . (3.71)
   
Eout4 (t) Es (t) − jElo (t)
After detecting the output fields Eout1 (t) and Eout3 (t) with the upper and Eout2 (t)
and Eout4 (t) with the lower balanced detector, the in-phase and quadrature photocur-
rents are obtained as
I I (t) = R · Eout1 (t) · E∗out1 (t) − R · Eout3 (t) · E∗out3 (t)

p
= R Ps Plo a(t) es elo cos [1ωt + ϕn (t) + ϕ0 + ϕ(t)] + i sh I , (3.72)
I Q (t) = R · Eout2 (t) · E∗out2 (t) − R · Eout4 (t) · E∗out4 (t)

p
= R Ps Plo a(t) es elo sin [1ωt + ϕn (t) + ϕ0 + ϕ(t)] + i sh Q , (3.73)
where i sh I and i sh Q are the overall shot-noise photocurrents in the in-phase and
quadrature arms, respectively.
3.3 Polarization Diversity 85
3.3.4 Polarization Diversity
As can be observed from (3.63) and (3.72)–(3.73), the photocurrents depend on the
polarization of the signal and LO light. Parallel polarizations are necessary to obtain
a maximal photocurrent. For slow polarization variations, adjusting the polarization
manually in lab experiments suffices. For commercial application, polarization di-
versity which increases the effort should be implemented.
Polarization diversity can be implemented by using two polarization beam split-
ters and doubling the optical frontend as shown in Fig. 3.12 for the quadrature re-
ceiver. The optical information signal is allowed to exhibit an arbitrary polarization
state and is split proportionately to the power corresponding to the PBS reference
directions. The LO must be polarized at 45◦ with respect to the PBS reference di-
rections for even LO power splitting. Signal light then interferes with the LO light
in both quadrature optical frontends with defined parallel polarization.
The illustration chosen here has to be understood schematically. In practice, both
separated polarization components of the information signal at the PBS outputs ex-
hibit the same linear polarization state, and it suffices when the LO light—whose
polarization must then be aligned to the polarization of the signal at the two PBS
outputs—is equally split with a 3 dB coupler.
Es (t) Optical I I (t )
PBS Quadrature
Frontend
Combination
Network
Optical I Q (t )
Elo (t) Quadrature
PBS Frontend
Fig. 3.12 Configuration of a polarization diversity receiver; PBS: polarization beam splitter
After photo-detection, the photocurrents are combined electrically. If the signal’s

polarization components have a relative phase delay, then the photocurrents of both
frontends exhibit this delay as well. This must be considered during the combina-
tion. Several combination schemes are discussed in [17]. Simply adding the in-phase
signals of both frontends, as well as the quadrature signals, leads to a penalty com-
pared with ideal polarization control, even if phase matching is ensured.
Currently, digital coherent receivers for detection of high-order modulation sig-
nals with polarization division multiplexing are extensively investigated [57, 72].
These receivers use the same optical frontend as illustrated in Fig. 3.12 for polariza-
tion diversity. In the electrical domain, the four photocurrents at the outputs of the
two optical quadrature frontends are analog-to-digital converted. Afterwards, polar-
ization de-multiplexing and compensation of degradation effects is accomplished
by adaptive digital equalization. More details can be found in Sect. 3.5.3.
3.4 Homodyne Receivers with Differential Detection
In contrast to direct detection receivers, homodyne receivers with differential detec-

tion accomplish the differential demodulation in the electrical domain after homo-
dyne detection. Homodyne differential detection can be performed either by analog
means (“phase diversity receivers”) as illustrated in Sect. 3.4.1, or by digital dif-
ferential demodulation as described in Sect. 3.4.2. In the description of homodyne
receivers with differential detection presented in these sections, the shot-noise and
the amplifier noise are neglected and the in-phase and quadrature photocurrents at
the outputs of the optical quadrature frontend are assumed to be
p
I I (t) = R Ps Plo · a(t) · cos [1ωt + ϕn (t) + ϕ0 + ϕ(t)], (3.74)
p
I Q (t) = R Ps Plo · a(t) · sin [1ωt + ϕn (t) + ϕ0 + ϕ(t)]. (3.75)
Starting from these photocurrents, it is shown in the next two sections how ampli-
tude information a(t) and phase difference information 1ϕ(t) = ϕ(t) − ϕ(t − TS )
can be recovered for ASK, DPSK and Star QAM formats by means of homodyne
differential detection.
3.4.1 Phase Diversity Receivers
In the late eighties, optical homodyne receivers using a multi-branch structure and
appropriate analog electrical processing were extensively investigated for the binary
modulation formats 2ASK, binary frequency shift keying (2FSK), and DBPSK, for
instance in [10, 29, 30]. These receivers are traditionally called “phase diversity
receivers”, although the term “phase diversity” is not necessarily confined to ho-
modyne differential detection and can also be used more generally for all receivers
employing multiple branches with different relative phase relations.
After homodyne detection with the optical quadrature frontend, the squared am-
plitude information of optical ASK and QAM signals can be obtained by squaring
and adding the in-phase and quadrature signals given by (3.74) and (3.75) as shown
in Fig. 3.13a.
I (t) = I I2 (t) + I Q2 (t) = R 2 Ps Plo a 2 (t) (3.76)
It can be observed that all the phase information, including the phase noise, is ef-
fectively disposed because (3.76) is not dependent on ϕn (t), ϕ(t) and ϕ0 . Moreover,
arbitrary frequency offsets 1ω are allowed in principle, and no automatic frequency
control is needed for the ASK phase diversity receiver. In practice, the electrical
receiver bandwidth limits the tolerable frequency offset.
3.4 Phase Diversity Receivers 87
a 2-branch ASK phase diversity receiver
I I (t )
( )2
2x4 I (t )
90° +
Hybrid IQ (t )
( )2
LO
BD
b 3-branch ASK phase diversity receiver
( )2
3x3 I (t )
Coupler ( )2 +
LO ( )2
DC blocker
Fig. 3.13 Homodyne phase diversity receivers for ASK employing a two-branch structure (a) and
a three-branch structure (b)
Phase diversity receivers can also be based on a three-branch structure using

a 3 × 3 coupler as depicted in Fig. 3.13b, where a 3-branch ASK phase diversity
receiver is shown without balanced detection so that additional DC blocks are nec-
essary to cancel the DC current emerging from the directly detected LO light. As
illustrated in Sect. 3.1.4, the signals after the 3 × 3 coupler in the three branches
are 120◦ out of phase with each other. The cancelation of phase information for the
three-branch structure becomes apparent from the simple trigonometric relation
3
cos2 [φ] + cos2 φ + 120◦ + cos2 φ + 240◦ = . (3.77)
2
Thus, after squaring and adding the signals of the three branches, the resulting pho-
tocurrent is no longer dependent on phase information.
The amplitude is not the only signal parameter that can be evaluated with phase
diversity receivers. Phase modulated signals with differentially encoded phases can
be demodulated just as well. When a DBPSK signal is detected by an optical quadra-
ture frontend, phase difference information can be obtained by implementing the
electronic demodulation network shown in Fig. 3.14. Starting from the in-phase and
quadrature photocurrents given by (3.74) and (3.75), the signal after the electronic
network can be calculated as
Q
I I (t )
×
2x4 TS I (t )
90° +
Hybrid IQ (t ) I
TS
×
LO
BD DBPSK
Fig. 3.14 Homodyne phase diversity receiver for DBPSK composed of a quadrature optical fron-
tend and an electronic demodulation network
I (t) = I I (t) · I I (t − TS ) + I Q (t) · I Q (t − TS )

= R 2 Ps Plo a(t)a(t − TS ) · cos [1ωt + ϕn (t) + ϕ0 + ϕ(t)]
· cos [1ω · (t − TS ) + ϕn (t − TS ) + ϕ0 + ϕ(t − TS )]
+R 2 Ps Plo a(t)a(t − TS ) · sin [1ωt + ϕn (t) + ϕ0 + ϕ(t)]
· sin [1ω · (t − TS ) + ϕn (t − TS ) + ϕ0 + ϕ(t − TS )]. (3.78)
Defining once again the difference of the modulation phase of two consecutive sym-
bols as 1ϕ(t) = ϕ(t) − ϕ(t − TS ), and the phase change due to laser phase noise
within the symbol duration as 1ϕn (t) = ϕn (t)−ϕn (t − TS ), (3.78) can be simplified
to the following equation by applying simple trigonometric calculations:
I (t) = R 2 Ps Plo a(t)a(t − TS ) · cos [1ϕ(t) + 1ϕn (t) + 1ωTS ]. (3.79)
In the case of an idealized DBPSK, where 1ϕ(t) takes the values 0 and π and where
it holds true that a(t) = a(t − TS ) = 1, and when neglecting the phase noise and the
frequency offset whose influence is discussed later on, a bipolar binary photocurrent
results and the data can be obtained by binary decisions at zero.
The phase diversity concept can be extended to arbitrary DPSK formats and
Star QAM modulation with differentially encoded phases. In [40], phase diversity
is investigated for higher-order DPSK formats for optical systems with coherence
multiplexing. Recently, a phase diversity receiver for DQPSK was analyzed in [59].
Whereas the electronic network shown in Fig. 3.14 yields the in-phase component
of arbitrary DPSK signals given by (3.79), the quadrature component can be ob-
tained by multiplying the in-phase and quadrature photocurrents with the delayed
photocurrents crossover and then subtracting the results
Q(t) = I Q (t) · I I (t − TS ) − I I (t) · I Q (t − TS )

= R 2 Ps Plo a(t)a(t − TS ) · sin [1ωt + ϕn (t) + ϕ0 + ϕ(t)]
· cos [1ω · (t − TS ) + ϕn (t − TS ) + ϕ0 + ϕ(t − TS )]
−R 2 Ps Plo a(t)a(t − TS ) · cos [1ωt + ϕn (t) + ϕ0 + ϕ(t)]
· sin [1ω · (t − TS ) + ϕn (t − TS ) + ϕ0 + ϕ(t − TS )]. (3.80)
3.4 Phase Diversity Receivers 89
By applying trigonometric calculations, (3.80) can easily be simplified and the

quadrature photocurrent becomes
Q(t) = R 2 Ps Plo a(t)a(t − TS ) · sin [1ϕ(t) + 1ϕn (t) + 1ωTS ]. (3.81)
The resulting configuration of a phase diversity receiver which is qualified to de-

tect the in-phase and quadrature components of arbitrary DPSK signals is shown in
Fig. 3.15. The part of the receiver for detecting the in-phase component I (t) is equal
to the phase diversity receiver for DBPSK depicted in Fig. 3.14.
×
TS
I I (t ) I (t )
× +
TS
2x4
90°
Hybrid IQ (t )
TS Q(t )
× -
-
LO
BD
TS
×
Fig. 3.15 Homodyne phase diversity receiver for detection of the in-phase and quadrature compo-
nents of arbitrary DPSK signals
It should be noted that the phase noise is not completely canceled for the DPSK
phase diversity receiver (in contrast to the ASK phase diversity receiver). The de-
modulation is based on differential phase detection and phase noise becomes criti-
cal if the phase change between two consecutive symbols takes considerable values.
Phase noise requirements are slightly more stringent than for direct detection, but
they are in a similar range (see Sect. 7.2.2).
Moreover, frequency offsets are a critical task for phase detection in phase diver-
sity receivers. Fixed frequency offsets lead to corresponding fixed phase rotations of
1ω TS and frequency offset drifts lead to slow varying rotations of the constellation
diagram. Thus, automatic frequency control (AFC, see Sect. 3.5.2) must be imple-
mented in practice which is not as critical as an optical phase locked loop because
the frequency offset drift δ1ω/δt is relatively slow.
The differentially encoded phase information contained in 1ϕ(t) in (3.79) and
(3.81) can be evaluated by the DPSK data recovery methods described in Sect. 3.2.3,
in the same way as for the direct detection receivers.
By performing a separate detection of intensity information, the DPSK phase

diversity receiver concept can be extended to detect Star QAM signals with differ-
entially encoded phases. The designation of intensity information can be based on

coherent detection, for instance by applying the ASK phase diversity concept as
depicted in Fig. 3.16.
Electronic
I I (t ) Intensity
Network
ASK
Phase Diversity
2x4
90°
In-phase
Hybrid IQ (t ) Electronic
Network
DPSK Quadrature
LO Phase Diversity
BD
Fig. 3.16 Homodyne phase diversity receiver for detection of Star QAM signals with differentially
encoded phases, intensity detection based on ASK phase diversity
Alternatively, intensity information can be determined by direct detection—by

employing an additional intensity detection branch as for the Star QAM direct de-
tection receiver. This receiver concept is shown in Fig. 3.17 and has the drawback
that the typical benefits of homodyne detection (the possibility of selecting WDM
channels by just tuning the LO and the enhanced possibilities for electronic miti-
gation of transmission impairments) can not be exploited for the intensity detection
branch. Furthermore, an optical amplifier must be used in the intensity detection
branch to obtain a similar sensitivity to the phase detection branch.
Intensity detection branch

Intensity
I I (t )
In-phase
3dB 2x4 Electronic
90° Network
DPSK
Hybrid IQ (t ) Quadrature
Phase Diversity
LO
BD
Fig. 3.17 Homodyne phase diversity receiver for detection of Star QAM signals with differentially
encoded phases, intensity detection based on direct detection
Data recovery for the Star QAM phase diversity receivers can be accomplished
with the same methods as used for the direct detection receivers. These methods
are described in Sect. 3.2.3. The data bits encoded in the amplitude can be directly
obtained from the intensity signals. The phase difference information contained in
3.4 Digital Differential Demodulation 91
the in-phase and quadrature signals after the electronic network can either be evalu-
ated by arg-decision or IQ-decision. In the latter case, a normalization (see Fig. 3.9)
becomes necessary for detection of Star QAM signals with more than four phase
states because the signals after the electronic network, defined by (3.79) and (3.81),
are dependent on the current amplitude a(t) and the amplitude a(t − TS ) delayed by
one symbol duration.
When being compared with direct detection receivers which enable detection of
the same modulation formats, homodyne receivers with differential detection offer
enhanced possibilities for electronic distortion equalization (EDE) and allow the se-
lection of WDM channels with high selectivity using a tunable LO. On the other
hand, direct detection receivers do not require any frequency synchronization and
polarization control. In [59], the potential of the phase diversity receiver for elec-
tronic distortion equalization and its suitability for WDM is highlighted. In phase
diversity receivers for WDM, no optical filter is needed to detect the desired wave-
length channel. In fact, unwanted interference terms (direct detection interference,
channel cross-channel interference) are eliminated by balanced detection. Finally,
the desired channel can be separated from other channels using an electrical low-
pass filter which must be placed directly behind the optical quadrature frontend
(in front of the electronic network). This enables highly selective channel separa-
tion and spectrally efficient WDM transmission. Moreover, in [59] the impact of
particular receiver impairments such as RIN, phase noise, frequency offset, receiver
asymmetries (e.g. gain and delay imbalance) and WDM channel crosstalk on a phase
diversity receiver for DQPSK are analyzed.
In principle, the electrical part of the phase diversity receivers discussed in this
section can also be implemented by digital means. However, when digital signal
processing is employed, the receiver illustrated in Sect. 3.4.2 is a simpler imple-
mentation option.
3.4.2 Digital Differential Demodulation
An alternative receiver concept for homodyne differential detection becomes feasi-

ble with recently available high-speed digital signal processing technology and is
shown in Fig. 3.18. The in-phase and quadrature signals at the outputs of the optical
quadrature frontend, defined by (3.74) and (3.75), are sampled by an A/D-converter,
for instance once per symbol. This provides the in-phase sample I Ik and the quadra-
ture sample I Q k at the k-th symbol interval
p
I Ik = R Ps Plo · ak · cos 1ωtk + ϕn k + ϕ0 + ϕk , (3.82)
p
I Q k = R Ps Plo · ak · sin 1ωtk + ϕn k + ϕ0 + ϕk , (3.83)
Digital differential demodulation
I I (t )
I Ik
ARG-Operation
ARG-Operation
A/D φk
2x4 Δφk
EDE
90° -
-
Hybrid IQ (t ) I Qk TS
φk −1
A/D
LO
BD
Fig. 3.18 Homodyne receiver with digital differential demodulation
where ak , ϕk and ϕn k represent the samples of the modulation amplitude, the mod-
ulation phase and the overall laser phase noise respectively. Optionally, electronic
distortion equalization can be applied first. By performing an arg-operation on the
in-phase and quadrature samples, the instantaneous phase of the current symbol can
then be calculated as

φk = arg I Ik , I Q k = 1ωtk + ϕn k + ϕ0 + ϕk . (3.84)
By subtracting the phase sample delayed by one symbol time φk−1 from the current
phase sample φk , the current phase difference 1φk can be determined as
1φk = φk − φk−1 = 1ϕk + 1ϕn k + 1ωTS , (3.85)
with the samples of the modulation phase difference and phase change due to laser
phase noise being defined as 1ϕk = ϕk −ϕk−1 and 1ϕn k = ϕn k −ϕn k−1 , respectively.
In practice, these steps necessitate only a table-lookup for phase determination and
a subtraction operation for phase differentiation. Since differential demodulation is
employed, the absolute phase is not important and laser phase noise becomes not
critical until the phase noise induced phase change takes considerable values within
the symbol duration TS . In [32], a homodyne receiver with digital differential de-
modulation is investigated for DQPSK modulation. Additionally to the impact of
phase noise, the influence of frequency offset is studied in this publication through
simulations and experiments. The algorithm for the digital frequency offset esti-
mation employed in [32] is easy to implement and capable of dealing with large
frequency offsets (0.94 GHz at 10 Gbaud) without any significant penalty.
Of course, Star QAM signals with differentially encoded phases can be demod-
ulated using the homodyne receiver with digital differential demodulation just as
well. The amplitude information can be obtained from the in-phase and quadrature
samples by simply calculating
ak2 = I I2k + I Q2 k . (3.86)
Finally, data recovery can be accomplished by employing the arg-decision scheme

described in Sect. 3.2.3.
3.5 Homodyne Receivers with Synchronous Detection 93
3.5 Homodyne Receivers with Synchronous Detection
In homodyne receivers with differential detection, the phase information of the op-
tical information signal is determined by differential phase detection. As a result,
the LO phase does not have to be synchronized with the carrier phase of the sig-
nal (asynchronous detection). By contrast, the absolute phases of the information
signal are designated with homodyne synchronous detection after mixing the signal
with the LO reference carrier. This way, higher receiver sensitivities can be ob-
tained than for differential detection, where the delayed modulation signal—which
is also noisy—acts as a reference signal during differential demodulation. Moreover,
synchronous detection has the advantage that arbitrary modulation formats can be
conveniently received because demodulation is not based on the determination of
phase differences but on absolute phases. On the other hand, a carrier synchroniza-
tion becomes necessary—a challenging task for practical implementation. However,
its chances for a successful commercial implementation are growing because laser
linewidth and frequency offset requirements become more relaxed with increasing
data rates. Furthermore, recent progress in high-speed digital signal processing tech-
nology permits the application of digital carrier synchronization techniques to opti-
cal receivers.
Before going into the details of carrier synchronization, let us first have a look
at its need for homodyne synchronous detection. The in-phase and quadrature pho-
tocurrents emerging at the outputs of the optical quadrature frontend for coherent
detection are specified by
p
I I (t) = R Ps Plo · a(t) · cos [1ωt + ϕn (t) + ϕ0 + ϕ(t)] + n I , (3.87)
p
I Q (t) = R Ps Plo · a(t) · sin [1ωt + ϕn (t) + ϕ0 + ϕ(t)] + n Q , (3.88)
where 1ω represents the frequency offset, ϕn (t) is the overall laser phase noise
and ϕ0 is the initial phase offset which are all defined in (3.60). The parameters n I
and n Q correspond to additive noise in the in-phase and quadrature branches re-
spectively. In the case of shot-noise limited detection, n I and n Q can be substituted
by the shot-noise photocurrents in the in-phase and quadrature branches, i sh I and
i sh Q , respectively (see Sect. 3.3.1 and Sect. 3.3.3). In systems with optical ampli-
fiers limited by the LO-ASE noise, n I and n Q can be interpreted as the LO-ASE
noise n L O−AS E I (t) and n L O−AS E Q (t) in the in-phase and quadrature branches, re-
spectively. The LO-ASE noise is discussed in Sect. 3.3.2.
The goal is to recover the modulation information contained in a(t) and ϕ(t). The
complex phasor which describes the received amplitude and phase in the IQ-plane
is given by
p
X (t) = I I (t) + j I Q (t) = R Ps Plo · A(t) · e j8(t) + n c , (3.89)
where A(t) represents the normalized complex envelope containing the modulation
information, 8(t) is the phase error due to the frequency offset, the laser phase noise
and the initial phase offset, and n c is the complex shot-noise / amplifier noise. These
parameters are defined by
A(t) = a(t) · e jϕ(t) , 8(t) = 1ωt + ϕn (t) + ϕ0 , n c = n I + jn Q . (3.90)
As can be observed from the received in-phase and quadrature signals given by
(3.87) and (3.88) or from the complex phasor defined in (3.89), the recovery of
information can fail due to the shot-noise / amplifier noise, as well as for an uncom-
pensated phase error which causes an arbitrary rotation of the constellation diagram.
For cancelation of the phase error, carrier synchronization which compensates for
laser phase noise, frequency offset and initial phase offset is needed.
3.5.1 Carrier Synchronization Techniques
The aim of carrier synchronization is to estimate the phase error 8(t) from the
received signal and then to rotate the constellation diagram to the appropriate posi-
tion by correcting the received phase with the determined phase error. The individ-
ual phase error contributors (the laser phase noise and the frequency offset) can be
treated separately or using a joint frequency and phase estimation technique.
Frequency and phase recovery techniques can be classified into three main cat-
egories: data aided, decision directed and non data aided. When trying to extract
the phase error from the received signal, this is impaired by the complex noise n c ,
which also contributes to the received phase, as well as by data modulation. The
latter impairment can be disposed by employing data aided carrier recovery. Here, a
part of the data sequence is known. This is usually achieved by applying burst-mode
transmission and sending a preamble with a known pattern at the beginning of each
burst. If the data sequence is not known a priori, decision directed or non data aided
carrier recovery techniques must be employed. For decision directed carrier syn-
chronization, the data is assumed to be estimated properly by the data recovery and
substitutes the true data. A similar performance as for data aided carrier recovery
can be reached at the steady state for a high signal to noise ratio and thus few de-
cision errors. As a third option, non data aided techniques neither make use of the
knowledge of the current data values nor their estimates—the received information
signal alone is used to determine the phase error.
Carrier recovery techniques can be further classified according to the receiver
configuration used. In closed loop receiver concepts, a corrective signal proportional
to the phase error is generated and fed back to synchronize the phases of the signal
and the LO. This way, the constellation diagram can be held locked in a fixed po-
sition by permanently compensating for newly appearing phase offsets. In contrast,
feedback of a control signal is not required in an open loop concept, which is de-
noted as feed forward (FF) carrier recovery. Here, desired parameters are directly
extracted from the received signal and a correction is performed afterwards.
3.5 Optical Phase Locked Loop 95
Homodyne optical receivers with an optical phase locked loop (OPLL, closed
loop scheme) were extensively studied at the end of the eighties and at the beginning
of the nineties [3, 8, 25, 27, 28, 47]. This work is briefly reviewed in Sect. 3.5.2. The
OPLL is suitable for high-speed analog implementation.
Recently, high-speed digital signal processing technology has emerged and there
has been renewed interest in coherent detection. In “digital coherent receivers”,
as they have often been referred to recently, digital equalization is performed very
efficiently by digital means, and carrier recovery is accomplished using digital phase
estimation. The problems associated with the OPLL—its stringent laser linewidth
requirements and difficult practical implementation—can be avoided. In the current
edition, this book does not cover equalization in detail, but digital phase estimation
is precisely described for some selected schemes in Sect. 3.5.3. Phase estimation
schemes based on feed forward and closed loop concepts and various modulation
formats such as PSK, Star QAM and Square QAM are thereby considered.
3.5.2 Optical Phase Locked Loop
Homodyne receivers with optical phase locked loop have been investigated in dif-
ferent variants, particularly for binary phase shift keying (BPSK) and quadrature
phase shift keying systems.
Optical Phase Locked Loops for BPSK
To understand the principles of the OPLL, one simple implementation option which
is denoted as Costas loop is depicted in Fig. 3.19.
I I (t ) to decision
BPSK signal LPF
2x4
90° ×
Hybrid IQ (t )
LPF
LO
BD
Loop
I M (t )
Filter
Fig. 3.19 Optical Costas loop receiver for BPSK

In combination with the electrical multiplier behind, the optical quadrature fron-
tend acts as a phase detector which generates a corrective signal proportional to the
phase error. Assuming (3.87) and (3.88) and neglecting the complex noise n c , the
corrective signal behind the multiplier can be derived as
1 2
I M (t) = I I (t) · I Q (t) = R Ps Plo · a 2 (t) · sin [28(t) + 2ϕ(t)], (3.91)
2
with the phase error 8(t) as defined in (3.90). For BPSK modulation, it is ideally
true that a 2 (t) = 1 and ϕ(t) = n · 180◦ (n ∈ {0, 1}), so that the modulation is
eliminated and I M (t) ∼ sin [28(t)]. For small phase errors, I M (t) is approximately
linearly dependent on the phase error, so that I M (t) ∼ 28(t). The corrective signal
is filtered by a loop filter which regulates the bandwidth of the OPLL and has to be
dimensioned properly to obtain stable operation of the loop. The filtered corrective
signal is fed into a tunable LO which adjusts the frequency of the LO lightwave
subject to the phase error.
The optical Costas loop discussed above is based on a non data aided carrier
recovery scheme and is analyzed in [8, 12, 18, 67] with respect to laser linewidth
requirements and optimal splitting ratio, the latter of which defines the fraction of
power provided in the in-phase and quadrature arms. An alternative OPLL structure
for BPSK is described in [28] and denoted as “balanced OPLL”. This OPLL fea-
tures a simple configuration, functioning with a 180◦ hybrid instead of a 90◦ hybrid.
However, a residual carrier must be provided by the transmitted signal (which can be
achieved using an incomplete phase modulation and phase shifts of less than 180◦ ),
and the modulation impact is not completely canceled, which leads to a perfor-
mance degradation due to data to phase-lock crosstalk [28]. The data to phase-lock
crosstalk can be considerably reduced by applying a decision directed configuration
[24, 63], where the decision circuit’s output signal is subtracted from the phase error
signal before it is fed into the loop filter. A third OPLL structure for BPSK, based on
a decision directed carrier recovery scheme, is analyzed in [27] and denoted there as
“decision driven OPLL”. The linewidth requirements here are more relaxed than for
the balanced OPLL. As for the improved balanced OPLL, decision results are used
to eliminate the modulation information. However, the deployment of a 90◦ hybrid
is as necessary as for the optical Costas loop.
Optical Phase Locked Loops for QPSK
Optical phase locked loops were also shown for QPSK, for instance in [3, 46, 47].
To illustrate the increasing complexity of the phase detector for higher-order mod-
ulation formats, the decision directed OPLL structure investigated in [46] is shown
in Fig. 3.20. The principle operation of the loop can be understood by looking at
the output signals of the three electrical multipliers. When not taking the decisions
into account, it is true for the output signal of the first multiplier that I1 (t) ∼
sin [28(t) + 2ϕ(t)]. The signal after the second multiplier can be calculated as
I2 (t) ∼ cos [28(t) + 2ϕ(t)]. Thus, a corrective signal I M (t) ∼ sin [48(t) + 4ϕ(t)]
3.5 Optical Phase Locked Loop 97
I I (t ) Data I
QPSK signal
LPF
2x4
90° ×
Hybrid IQ (t ) Data Q
LPF
LO
BD + -
×
I 2 (t )
I M (t ) I1 (t )
Loop
×
Filter
Fig. 3.20 Decision directed OPLL for QPSK modulation, investigated in [46]
results after the third multiplier. In the case of QPSK, it can be ideally assumed
ϕ(t) = 45◦ + n · 90◦ (n ∈ {0, 1, 2, 3}), so I M (t) becomes independent of the
transmitted data stream. Because a corrective signal proportional to the four-fold
phase error is generated, a quadrant ambiguity results, which can be solved by dif-
ferentially encoding the transmitted data and employing differential decoding at the
receiver.
The decision directed OPLL for QPSK was experimentally demonstrated in [47]
at a data rate of 8 Gbit/s. Instead of the multipliers, digital XOR gates were em-
ployed. The beat-linewidth of the external cavity lasers used in the experiment was
60 kHz. But even a phase noise this small resulted in an error floor caused by a
loop propagation delay of 12 ns. It is shown in [46] that the loop delay of an OPLL
becomes non-negligible if the loop delay time τd is greater than the bit duration
1/r B and that this loop delay becomes the dominant restriction for the linewidth
requirements at high data rates. The required beat linewidths to keep the receiver
sensitivity penalty below 1 dB at BER=10−10 are derived for the decision directed
OPLL in [46] and are summarized for BPSK and QPSK in Table 3.1.
Table 3.1 Required beat linewidths for a receiver sensitivity penalty of 1 dB at BER=10−10 for ho-
modyne receivers with decision driven OPLL, for zero loop delay time τd and under consideration
of τd , derived in [46]
BPSK QPSK
1ν [H z] (τd ≫ 1/r B ) 2.04 · 10−3 /τ d 2.86 · 10−4 /τd

1ν [H z] (τd = 0) 5.99 · 10−4 · rB 9.56 · 10−5 · r B
For a data rate of r B = 40 Gbit/s, for instance, the required beat linewidths are
24 MHz and 3.86 MHz for BPSK and QPSK, respectively, when the loop delay can
be neglected. However, when the loop delay is 12 ns as in the experiment performed
in [47], the beat linewidth requirements change to 170 kHz and 24 kHz for BPSK
and QPSK, respectively. This is beyond the specifications of laser diodes currently
available for commercial applications. Even when neglecting the delay of the com-
ponents, a cable length of about 2.4 m alone corresponds to a delay of 12 ns. Thus,
very compact, preferably integrated structures have to be aspired to.
Until now, homodyne detection experiments with OPLL were not reported for
modulation formats beyond QPSK, although phase locked loop (PLL) schemes for
higher-order PSK and QAM formats are known from electrical systems—for in-
stance the decision directed carrier recovery with selective gated PLL for Square
16QAM described in [21, 69, 70]. In principle, these phase locked loop schemes
could be adapted to optical receivers. However, the laser linewidth requirements
become even more stringent for higher-order modulation formats, so that the imple-
mentation of OPLL receivers for these formats only seems to be realistic when loop
delay is kept very small.
Subcarrier Based Optical Phase Locked Loop
Another challenge for an OPLL is the practical implementation of the optical

voltage-controlled oscillator (VCO). The optical counterpart to an electrical VCO is
a tunable local laser. However, the requirements for fast frequency tuning make this
component a complex and expensive optical device. Recently, a new solution for the
implementation of the optical VCO was proposed in [6, 7], where the tunable LO
is replaced by a simple continuous wave laser whose light is intensity modulated
externally (for instance using a MZM) by the output signal of an electrical VCO as
shown in Fig. 3.21.
Elo (t )
Loop
I M (t )
MZM Electrical
VCO Filter
Fig. 3.21 Optical VCO com-
posed of a CW laser, an
intensity modulator and an
electrical VCO, alternative CW
solution to a tunable LO, Optical VCO
proposed in [6]
The spectrum of the signal arising at the optical VCO output contains two main
subcarriers at the frequencies (ωC W ± ωV C O ) /2π , where ωC W and ωV C O are the
angular frequencies of the CW laser and the electrical VCO, respectively. The se-
lection of one of the subcarriers provides an optical carrier, which can be fine-tuned
with the speed and stability of the electrical VCO.
3.5 Digital Phase Estimation 99
Automatic Frequency Control
In the homodyne receivers with differential detection discussed in Sect. 3.4, phase
differences are detected, so no optical phase locked loop must be employed. Nev-
ertheless, frequency offsets 1ω which cause phase offsets of 1ωTS , as well as fre-
quency offset drifts δ1ω/δt must be compensated for. This can be accomplished
with an automatic frequency control (AFC) loop. Frequency offsets can be detected
by a delay line frequency discriminator which has been employed in [45] and is
shown in Fig. 3.22.
I I (t )
×
Fig. 3.22 Delay line fre- τ I M (t )
quency discriminator for - - Q(t ) Integrator
generation of the corrective
signal within an automatic IQ (t ) τ
frequency control loop, ×
see [45]
The setup of the delay line frequency discriminator is similar to the quadrature
part of the electronic network of the phase diversity receiver for DPSK formats
shown in Fig. 3.15. The signal arising at the integrator input is
Q(t) ∼ sin [1ωτ + ϕn (t) − ϕn (t − τ ) + 1ϕ(t)], (3.92)
where τ in this case represents the delay of the electrical delay lines. Because the fre-
quency drift is slow compared with the phase noise and the modulation, and when
furthermore the expectation values of the phase noise and the modulation are as-
sumed to be zero, a corrective signal I M (t) proportional to the integral of sin (1ωτ )
results after the integrator which represents a phase error signal and can be taken to
drive the optical voltage controlled oscillator.
3.5.3 Digital Phase Estimation
Homodyne optical receivers based on OPLLs suffer from implementation difficul-

ties and stringent laser linewidth requirements, especially when migration to higher-
order optical modulation formats is desired. Fortunately, recent advances in high-
speed digital signal processing technology open the door to handle carrier synchro-
nization in homodyne optical receivers with digital techniques.
With high-speed digital signal processors (DSPs)—which can be successfully
applied to optical transmission systems—tasks like electronic pre-distortion can be
performed on the transmitter side. At the receiver end, coherent detection mech-
anisms in particular benefit from digital technology. In digital coherent receivers,
transmission impairments like chromatic dispersion and nonlinearities can be digi-

tally compensated for with electronic distortion equalization, and digital carrier syn-
chronization allows for a free running LO which does not have to be phase locked
by an OPLL. Figure 3.23 shows the basic setup of a digital coherent receiver with
homodyne synchronous detection when polarization diversity and polarization divi-
sion de-multiplexing are not employed.
Digital signal processing
I I (t ) I Ik Ik
Phase Estimation
Timing Recovery
LPF A/D
Adaptive EDE
2x4
to data
90°
recovery
Hybrid I Qk
IQ (t ) Qk
LPF A/D
LO
BD
Fig. 3.23 Digital coherent receiver with homodyne synchronous detection employing adaptive
electronic distortion equalization (EDE) and digital phase estimation; single-polarization scheme
without polarization diversity and polarization de-multiplexing
After being low-pass filtered, the in-phase and quadrature signals I I (t) and I Q (t)
are sampled by A/D-converters. The first functional block in the digital signal pro-
cessing part is often a non-adaptive time or frequency domain equalizer (not shown
in Fig. 3.23) which compensates for the main part of chromatic dispersion having
accumulated along the fiber link [33, 56]. Afterwards, a timing recovery is accom-
plished in order to synchronize the sample rate with the signal‘s symbol rate. Widely
used algorithms are the Gardner [14] and the square timing recovery [48] here. Tim-
ing recovery is typically followed by a time domain equalizer, which is usually im-
plemented as a FIR filter whose coefficients are adapted using the constant modulus
algorithm (CMA) or the decision-directed least mean square (LMS) algorithm. In
order to ensure a proper operation of the equalizers, a sample rate of at least twice
the symbol rate is mostly chosen (fractionally spaced equalizer). For digital phase
estimation—the functional block behind the adaptive EDE—just one sample per
symbol is required which must be properly selected for the case that more than one
sample per symbol is utilized for equalization.
Digital coherent receivers are often used for detection of signals with polariza-
tion division multiplexing [57, 72]. A digital coherent receiver with polarization
de-multiplexing is shown in Fig. 3.24. It uses the same optical frontend as described
in Sect. 3.3.4 for polarization diversity. In the electrical domain, the four photocur-
rents at the outputs of the balanced detectors are analog-to-digital converted and
then further processed by the digital signal processing. Compensation for degrada-
tion effects as well as polarization de-multiplexing are accomplished by the adaptive
digital equalization. The time domain equalizer is often implemented as a FIR but-
terfly equalizer here [56, 57]. Phase estimation can be performed by treating both
polarizations independently or by using a joint-polarization approach [34].
Digital signal processing

X Ik
A/D
Es (t) 2x4
PBS 90°
Hybrid
Digital Phase Estimation

X Qk
Timing Recovery
A/D
Adaptive EDE
YI k
A/D
Elo (t) 2x4
PBS 90°
Hybrid
YQk
LO A/D
BD
Fig. 3.24 Digital coherent receiver with homodyne synchronous detection employing adaptive
electronic distortion equalization (EDE), digital phase estimation and polarization de-multiplexing
The following paragraphs skip timing recovery and equalization, but in detail de-
scribe carrier synchronization by digital phase estimation for the single-polarization
receiver depicted in Fig. 3.23.
When the in-phase and quadrature signals I I (t) and I Q (t) are sampled by A/D-
converters once per symbol at sampling instants tk (in the middle of the symbols,
assuming an ideal clock recovery), the sampled signals I Ik and I Q k —when neglect-
ing the electrical filtering—are defined by (3.82) and (3.83). The complex phasor
obtained after sampling is then given by
p
X k = I Ik + j I Q k = R Ps Plo · Ak · e j8k + n ck , (3.93)
where the samples of the normalized complex envelope Ak , the phase error 8k , and
the complex noise n ck are defined as
Ak = ak · e jϕk , 8k = 1ωtk + ϕn k + ϕ0 , n ck = n Ik + jn Q k . (3.94)
In principle, the algorithms for phase estimation can be based on data aided,
decision directed or non data aided techniques, as well as on feed forward or closed
loop concepts. The phase estimation algorithms applied so far to optical receivers
are mainly based on the non-data aided feed forward M-th power scheme which
allows for abandoning the provision of pilot sequences and the feedback of decision
results [15, 31, 38, 39, 44, 60, 64, 65].
In the next paragraph, the so-called “feed forward M-th power block scheme”,
similar to the one investigated for QPSK in [15, 39], is described for MPSK, Star
QAM and Square QAM formats. It is a candidate for commercial employment in
the near future due to its relatively simple implementation.
Feed Forward M-th Power Block Scheme
When the phase error of an individual symbol is calculated, a fraction of this phase
error is caused by phase drifting effects, but the remainder results from the shot-
noise / amplifier noise. To be able to get an accurate estimate of the phase error 8k ,
it has to be isolated from the phase error caused by the shot-noise / amplifier noise.
This can be attained by performing an averaging over a block of N symbols.
In a M-th power phase estimation scheme with symbol-to-symbol correction,
a phase error estimate is calculated individually for every symbol while averaging
over a block of N symbols. This corresponds to a sliding window technique. A
different approach, which can be denoted as “M-th power block scheme”, is to cal-
culate a common phase error estimate for all N symbols within a block. On the one
hand, the accuracy of the carrier phase estimation is reduced when applying this
technique (the phases of each symbol within a block are corrected with the same
phase error estimate). On the other hand, the M-th power block scheme leads to
a reduced implementation complexity. A block diagram of the feed forward M-th
power block scheme for MPSK is illustrated in Fig. 3.25.
i-th block of N symbols i-th corrected block of N symbols
X k = I I k + jI Qk X k ' = I k + jQk
1:N DEMUX 1:N MUX

X 1+ ( i −1) N X 2+ (i −1) N ... X N + (i −1) N X '1+ ( i −1) N X '2 + ( i −1) N ... X 'N + ( i −1) N
-jΦ i -jΦ i -jΦ i

e e ... e
M M M
() () ... ()
∑
Ψi
Ψi-1 1 M ⋅ arg ( ) Φi = Ψi + ni ⋅ 2π/M
Ψi ni
SEGMENT CHANGE DETECTOR

Ψi - Ψi-1 < -π/M ni = ni −1 + 1
Ψi - Ψi-1 > π/M ni = ni −1 − 1
else ni = ni −1
Fig. 3.25 Feed forward M-th power block scheme with field averaging for digital carrier phase
estimation on differentially encoded MPSK signals
The incoming in-phase and quadrature samples are first combined into complex
samples X k = I Ik + j I Q k . The stream of complex samples is then parallelized by
a 1 : N demultiplexer, where N denotes the number of samples per block (which
corresponds to the number of symbols per block, because the sampling rate is equal

to the symbol rate). Thus, the samples X 1+(i−1)N , X 2+(i−1)N , ..., X N +(i−1)N are
available in parallel in the i-th block. Parallelization is very useful with regard to
practical implementation because the clock speeds of currently available DSPs are
limited to about 1GHz, so that it is necessary to process samples in parallel to han-
dling higher symbol rates. In the architecture shown in Fig. 3.25, the clock rate is
reduced by a factor of N with respect to the symbol rate. An even higher degree of
parallelization is possible when the incoming sample stream is demultiplexed into
multiple blocks of N samples as done in [15, 39].
After demultiplexing, the N complex samples of the i-th block are raised to the
M-th power to eliminate the phase modulation
p M
X kM = R Ps Plo · Ak · e j8k + n ck
p
= (R Ps Plo ) M · akM · e j Mϕk · e j M8k + n k
p
= (R Ps Plo ) M · akM · e j M8k + n k , (3.95)
where n k contains the shot-noise / amplifier noise n ck raised to the M-th power and
the multiple mixing terms of the shot-noise / amplifier noise and the signal, and with
k = p + (i − 1)N , p ∈ {1, 2, ..., N }, in the i-th block. The phase modulation is
canceled because it holds e j Mϕk = 1 for MPSK signals, whose modulation angles
are given by ϕk = n · 2π/M with n ∈ {0, 1..., M − 1}.
As mentioned above, averaging reduces the influence of the shot-noise / amplifier
noise on the determination of the phase error estimate. In [64, 65], it is shown that
the optimal phase estimate is attained by applying a Wiener filter. Another option—
leading to a lower phase noise tolerance, but featuring more simple practical
implementation—is to use a rectangular filter which calculates the arithmetic mean
of the input samples while summing the input samples with equal weights. One ap-
proach with rectangular filtering, whichnis denoted here as field averaging, is to sum o
4
the N parallel available raised samples X 1+(i−1)N 4
, X 2+(i−1)N , ..., X 4N +(i−1)N as
shown in Fig. 3.25. The angle of the resulting phasor divided by M then yields an
estimate 9i for the average phase error of the i-th block
 
N
1 X
9i = · arg  XM
p+(i−1)N .
 (3.96)
M
p=1
On the one hand, averaging lowers the influence of the shot-noise / amplifier
noise on the phase error estimate. On the other, an inherent error is introduced since
an average phase error estimate is calculated, commonly used for the phase cor-
rection of all symbols in the block. An optimal block length N can be found as a
trade-off between the shot-noise / amplifier noise and the phase noise effects.
Since the angle values calculated by the arg-operation are limited to the inter-
val ]−π, π], the phase error estimates 9i take values between −π/M and π/M and
a M-fold phase ambiguity of n ·2π/M, n ∈ {0, 1, ..., M − 1}, is induced. This prob-
lem can be overcome by periodically sending synchronization sequences, or better
still by the use of differential encoding. More precisely, DPSK signals are gener-
ated at the transmitter side, so that the absolute position of the constellation diagram
at the receiver side is allowed to be ambiguous with n · 2π/M. It should be noted
that a synchronous detection is performed nevertheless, because the demodulation
is not based on determination of the phase differences of two consecutive symbols
but on the designation of absolute signal phases relative to the LO reference carrier.
The differential decoding is performed here on the logical plane by using a differ-
ential decoder after phase estimation and data recovery, as described in Sect. 3.5.4.
Therefore, an improved OSNR performance can be expected in comparison with
differential detection, although the performance limit of coherent synchronous de-
tection can not be completely achieved due to error propagation effects during the
differential decoding [5]. To indicate that synchronous detection is performed with
digital phase estimation receivers, the phase modulation formats detected by ho-
modyne synchronous receivers will be just denoted here as MPSK, even though
MDPSK signals are sent by the transmitter.
Although the use of differential coding can solve the problem of permanent false
rotations of n · 2π/M with constant n, symbol errors arise if the phase ambiguity
error of two consecutive blocks is different (change of n). This occurs if the random
walk of the phase noise is passing one of the boundaries between two segments at
n ·2π/M. In this case, the phase error estimate performs phase jumps (“cycle slips”)
and does not follow the trajectory of the physical phase [39, 44]. The phase jumps
must be corrected by performing a phase unwrapping and can be detected by a so-
called “segment change detector” which is also shown in Fig. 3.25. A variable n i ,
representing the actual segment, is updated in each block, depending on whether a
segment change has occurred or not. For positive phase jumps which can be detected
by 9i − 9i−1 > π/M, the segment number n i is decremented by one, for negative
phase jumps (9i − 9i−1 < −π/M), n i is incremented by one, and n i is left equal if
no segment change has appeared. After that, the phase error estimate 9i in the i-th
block is corrected by
2π
8i = 9i + n i · , (3.97)
M
where 8i represents the corrected phase error estimate of the i-th block which is
taken to correct the phase of the received symbols by calculating
p
X k′ = R Ps Plo · Ak · e j8k + n ck · e− j8i
p
= R Ps Plo · Ak · e j (8k −8i ) + n ck · e− j8i , (3.98)
again with k = p + (i − 1)N , p ∈ {1, 2, ..., N }, in the i-th block. In (3.98), the
quantity 8ek = 8k − 8i describes the residual phase error of the k-th symbol.
In the scheme described here, the phase correction for segment changes defined
by (3.97) is applied to all symbols of the block. In principle, it is sufficient to correct
only the last symbol of each block as done in [15, 39]. In this case, however, correc-
tion has to be performed at the logical level during differential decoding, whereas
the phase estimation and the differential decoding can be completely decoupled for
the scheme presented here.
As a result of phase noise, symbol errors can occur because phase errors 8k can
not be distinguished from phase errors of 8k + n · 2π/M, n ∈ {0, 1, ..., M − 1} due
to the M-fold phase ambiguity. This happens in spite of differential encoding and
the appropriate detection of segment changes. The phase error estimate 8i becomes
imprecise
and symbol errors can arise if the phase error walk exceeds a phase inter-
val of −π/M, π/M within the block duration Tblock = N · TS .
The deployment of field averaging defined by (3.96) yields phase error

estimate
√ inaccuracies when the summed phasors are not of the same length
(R Ps Plo ) M · akM . Thus, field averaging might not be appropriate for phase es-
timation of highly distorted MPSK signals and can not be used for QAM formats
without further modification. Field averaging can be improved and made usable for
carrier phase estimation of Star QAM signals by normalizing the phasors to an am-
plitude of one before being summed. This scheme is denoted as normalized field
averaging throughout this book. After normalizing the phasors of the raised com-
plex fields, the average phase error estimate 9i is again obtained by calculating the
angle of the resulting sum phasor divided by M
 
1 XN XM
p+(i−1)N
9i = · arg  . (3.99)
M M
p=1 X p+(i−1)N
In the case of Star QAM, the number of symbols M in (3.99) must be replaced with
the number of phase states N ph .
In contrast to MPSK and Star QAM where the modulation phases are equally
spaced and the modulation can be eliminated by raising the received complex sig-
nal samples to the N ph -th power, the phase states are not arranged equidistantly for
Square QAM formats. However, the M-th power block scheme can still be applied
when the constellation points are partitioned into two groups. This is illustrated for
Square 16QAM in Fig. 3.26a and for Square 64QAM in Fig. 3.26b, the latter show-
ing only one quadrant. The Class I symbols (solid points in Fig. 3.26) all exhibit
modulation angles of π/4 + n · π/2, n ∈ {0, 1, 2, 3}, so that modulation informa-
tion can be eliminated in the same way as for QPSK by raising to the fourth power
when selecting only these symbols for determination of the phase error estimate.
The selection of the Class I and Class II symbols can be accomplished by perform-
ing amplitude decisions on the received signal samples.
a Square 16QAM b Square 64QAM (one quadrant)
q q
Class I symbols Class II symbols
Fig. 3.26 Class partitioning for Square 16QAM (a) and Square 64QAM (b)
In the case of Square 16QAM, the symbols on the inner and outer circle belong to
Class I and the symbols on the middle circle to Class II (open points in Fig. 3.26). A
symbol X k can be identified as a Class I symbol X kClass I when the decision s1k + s2k
yields a logical one. The decision results s1k and s2k are specified by

1 |X k | > sn
sn k = , n = (1, 2), (3.100)
0 |X k | < sn
where s1 and s2 denote the thresholds in-between the inner and middle circle and
the middle and outer circle, respectively, shown as dashed lines in Fig. 3.26a. On
average, only half of the symbols are Class I symbols for Square 16QAM, so larger
block lengths N are necessary to ensure that a sufficient number of Class I symbols
is available in each block for appropriate averaging.
The ratio of Class I and Class II symbols is even smaller for Square 64QAM,
as shown in Fig. 3.26b. A quarter of the constellation points exhibit modulation
angles of π/4 + n · π/2, but only three points in each quadrant can effectively be
selected by amplitude decisions. Because the constellation points on the third circle
are close to the thresholds, phase estimation probably yields better results when only
the symbols on the inner and outer circle are selected for the calculation of the phase
error estimate. The selection of just an eighth of the total number of constellation
points leads to high necessary block lengths as illustrated later on in Sect. 7.3.2.
The phase estimation procedure for Square QAM formats is summarized in
Fig. 3.27. After demultiplexing the complex signal samples into blocks of N sym-
bols, the class partitioning just described is performed. NClass I symbols of the N
symbols of a block belong to Class I and are used to determine the phase error esti-
mate 8i which is used to correct the phase of all the symbols in the block.
1:N DEMUX 1:N MUX

-jΦ i -jΦ i -jΦ i

e e ... e
CLASS PARTITIONING
Select the Class I points
.......... ..........
X kClassI
CALCULATE PHASE ERROR ESTIMATE Φi

as for QPSK (M = 4)
Fig. 3.27 Phase estimation procedure for Square QAM formats
The phase error estimate for Square QAM formats can be obtained in the same
way as for QPSK by applying the feed forward M-th power block scheme, for in-
stance. When symbols of the inner and the outer circle with different amplitudes
are selected by the class partitioning, normalized field averaging must be employed.
With M = 4, replacing N with NClass I and averaging only over the Class I symbols
X kClass I , (3.99) changes to
 
NX
Class I 4
1 X
9i = · arg  kClass I . (3.101)
4 4
k=1 X kClass I
The corrected phase error estimate 8i can then be calculated after segment change
detection by (3.97).
Closed Loop M-th Power Block Scheme
Alternatively, digital phase estimation can be based on closed loop concepts with
feedback of a control signal. When the feed forward M-th power block scheme
is employed, a phase error estimate is calculated for each block from a received
constellation diagram with an arbitrary phase offset corresponding to the current
phase error. In contrast, the position of the constellation diagram is permanently
tracked when the closed loop M-th power block scheme is applied.
1:N DEMUX 1:N MUX

Φi-D
-jΦ i-D -jΦ i-D -jΦ i-D -j∆Φi -j∆Φi -j∆Φi

e e ... e e e ... e
M M M
() () ... ()
arg( ) arg( ) ... arg( )
∑
M ⋅N
∆Φi
Φi-D +
Φi
Fig. 3.28 Closed loop M-th power block scheme with argument averaging for digital carrier phase
estimation of differentially encoded MPSK signals
The block diagram of the closed loop M-th power block scheme for MPSK is
depicted in Fig. 3.28. Before calculating a new phase error estimate, the received
signal samples are rotated by −8i−D which represents the negative of the phase
error accumulated in the past. Afterwards, the symbols are in the appropriate posi-
tion with the exception of the phase error difference 18i accumulated in the last D
blocks. To minimize 18i and for relaxed linewidth requirements, the accumulated
phase error 8i−1 of the last block should ideally be available for rotation of the cur-
rent block (D = 1). However, this is not attainable in practice when the processing
delay for the calculation of 8i is greater than one block length. For this reason, only
the accumulated phase error 8i−D is available in the current block, with a process-
ing delay of D block lengths. The processing delay, which has a comparable effect
to the loop delay in OPLLs, leads to more stringent laser linewidth requirements
for the closed loop than for the feed forward M-th power block scheme, which can
be implemented without the need for using distant past results. For this reason, the
closed loop scheme is only of practical interest in systems where laser phase noise
is not a critical parameter.
For determination of the phase error difference 18i , the same mechanisms can
be adopted as for the feed forward M-th power block scheme. After rotating the
signal samples by 8i−D and raising the rotated samples to the M-th power, field
averaging or normalized field averaging can be employed. With field averaging, the
current phase error difference is obtained analogous to (3.96) as
 
N M
1 X
18i = · arg  X p+(i−1)N · e− j8i−D . (3.102)
M
p=1
Another option for averaging, denoted here as argument averaging and illustrated
in Fig. 3.28, is to calculate the arguments of the raised complex fields first, and then
to average the arguments to obtain an estimate of the current phase error difference
which is given by
N M
1 X
18i = · arg X p+(i−1)N · e− j8i−D . (3.103)
NM
p=1
Argument averaging is applicable to the closed loop concept where the constel-
lation diagram is held constantly in a fixed position and the raised samples are cen-
tered around an average phase of near zero due to the permanent phase tracking.
However, it can not be applied to the feed forward scheme where the samples are
centered around an arbitrary phase. To give an example, average phase errors of
zero and average phase errors of π/4 would both yield phase error estimates of zero
for QPSK, because the arguments are calculated modulo 2π with values in the in-
terval ]−π, π]. Since argument averaging is independent of phasor amplitudes, it
is applicable to Star QAM and Square QAM formats. In the case of Square QAM,
class partitioning must be performed before raising to the fourth power.
Finally, the accumulated phase error of the current block, obtained after a pro-
cessing delay of D block lengths, can be calculated by
8i = 8i−D + 18i , (3.104)
and the phase of all the symbols in the block is corrected by −8i . This has the same
effect as correcting the symbols by −18i after rotation with 8i−D as illustrated in
Fig. 3.28. After the correction, the corrected samples are obtained as
p
X k′ = R Ps Plo · Ak · e j (8k −8i ) + n ck · e− j8i , (3.105)
with k = p + (i − 1)N , p ∈ {1, 2, ..., N }, in the i-th block and a residual phase
error of 8ek = 8k − 8i .
The last paragraphs described two digital phase estimation schemes which fea-
ture a relatively simple implementation. Phase estimation can be improved by using
more sophisticated techniques. For instance, block-by-block phase correction can be
replaced with symbol-to-symbol phase correction. Averaging can be optimized by
employing hyperbolic or Wiener filtering. Non data aided carrier recovery schemes
can be replaced with decision directed techniques [22]. Moreover, enhanced phase
estimation algorithms for Square QAM allow to incorporate all constellation points
into the calculation of the phase error estimate [37, 55]. In the next paragraph, it is
shown how a separate frequency offset estimation can be accomplished.
Digital Frequency Offset Estimation
Generally, the phase error estimate 8i provides an estimate of the phase error 8k
which contains the phase error due to the frequency offset 1ω, the initial phase off-
set ϕ0 and the phase noise ϕn k , as can be observed from (3.94). Because the walk
of the phase must remain within a certain interval −π/M, π/M within the block
length Tblock to avoid symbol errors, the frequency offset should be estimated sepa-
rately before phase estimation, since permanent constant phase offsets of 1ω · Tblock
are induced within each block duration. For instance, at a symbol rate of r S =
10 Gbaud and a block length of N = 8, a frequency offset of 1ω = 2π · 100 MHz
shifts the phase by 1ω · Tblock = 2π · 100 MHz · 8 · 100 ps= 0.5 rad. This would sig-
nificantly restrain the phase noise tolerance for QPSK, and lets the phase estimation
fail for modulation formats with even more phase levels.
One algorithm possible for frequency offset estimation is the phase differential
algorithm. As for the M-th power phase estimation, the received samples are raised
to the M-th power to eliminate phase modulation
p
X kM = (R Ps Plo ) M · akM · e j M 1ωtk +ϕ0 +ϕnk + n k . (3.106)
Next, the current raised sample is multiplied by the complex conjugate of the raised
sample delayed by d symbols. In the case of MPSK (akM = ak−d M ), and when ne-
glecting the noise n k , for simplicity, one obtains

h i
M ∗
p j M 1ω(tk −tk−d )+ ϕn k −ϕn k−d
X kM · X k−d = (R Ps Plo ) · ak2M
2M
·e
p
1ω·τ +1ϕn k
= (R Ps Plo )2M · ak2M · ejM , (3.107)
where τ = tk − tk−d = d · TS denotes the duration of d symbols, 1ω · τ is

the phase offset accumulating during d symbols due to the frequency offset and
1ϕn k = ϕn k − ϕn k−d represents the phase noise induced random phase change
within the time interval τ which can be assumed to be Gaussian distributed as il-
lustrated in Sect. 2.1.1. Similarly to the isolation of the phase error induced by laser
phase noise from the phase error induced by shot-noise / amplifier noise during dig-
ital phase estimation, the frequency offset can be isolated from the phase noise by
averaging over a longer observation interval with N −d single estimates. This yields
the frequency offset estimate
 
N
1 X ∗
1ωest = · arg  X kM · X k−d
M . (3.108)
Mτ
k=1+d
3.5 Data Recovery for Synchronous Detection 111
Since the angle values of the arg-operation are limited to the interval ]−π, π], the
maximum resolvable frequency offset is given by
1ωestmax 1 1 rS
1 f max = = · (±π) = ± =± . (3.109)
2π 2π Mτ 2Mτ 2Md
The frequency offset tolerance is reduced with increasing M and increasing d. On
the other hand, the frequency estimation accuracy improves with increasing d, so
there is a trade-off between estimation range and estimation accuracy [55].
3.5.4 Data Recovery for Synchronous Detection
After carrier synchronization, the constellation diagrams are appropriately aligned

and data can be recovered. The position of the constellation diagram is still am-
biguous due to the phase ambiguity of the carrier recovery. The phase ambiguity
is 2π/M for MPSK formats, 2π/N ph for Star QAM formats and π/2 for Square
QAM. However, this phase ambiguity induces no difficulties when data is differen-
tially decoded after the data recovery.
The recovery of information can be based on arg-decision or IQ-decision (see
Sect. 3.2.3). Data recovery for synchronous detection of PSK signals with arg-
decision is illustrated in the block diagram in Fig. 3.29.
Ik d#1k
b1k
d#
Multiplexer
Differential
s1k
Operation
from
Recovery
Data
Decoder
1k −1
TS
Logic
Data
ARG
carrier d#mk
….
….
….
recovery Qk
sM k d#mk −1
TS bmk
Fig. 3.29 Data recovery for synchronous detection of arbitrary PSK signals using arg-decision
After performing the arg-operation and decision as described by (3.46), a data re-
covery logic assigns symbols with respective bits {d̃1k , d̃2k , .., d̃m k } to the absolute
phase states. In principle, this symbol assignment to absolute phases can be chosen
arbitrarily. To reverse the differential encoding of the transmitter and to resolve the
phase ambiguity, a differential decoder is used after the data recovery logic which
evaluates the differences of the absolute phases on a logical level. In the differen-
tial decoder for any PSK format, the bits of the current output symbol of the data
recovery logic, denoted here as {d̃1k , d̃2k , .., d̃m k }, are combined with the previous
output bits of the data recovery logic, defined by {d̃1k−1 , d̃2k−1 , .., d̃m k−1 }, to spec-
ify the current decoded symbol representing the original data bits {b1k , b2k , .., bm k }.
The symbol assignment to phase differences at the output of the decoder is forced
by the original bit mapping defined at the transmitter side.
Since the symbol assignment to absolute phases can be chosen arbitrarily, the
data recovery circuits for the differential detection of DPSK signals, which are de-
scribed in Sect. 3.2.3, can be re-used for the synchronous detection of PSK signals.
The decoder must then provide for the appropriate final conversion of the symbol
mapping to absolute phase states defined by {d̃1k , d̃2k , .., d̃m k } to the original data
information {b1k , b2k , .., bm k } representing the phase differences.
When the same data recovery logic is used as for differential detection, the map-
pings of the input and output constellations of the differential decoder are identical.
In the case of QPSK (m = 2), the truth table is then given by Table 3.2 and the
following logical relations can easily be derived for the differential decoder
b1k = d̃1k d̃2k d̃1k−1 + d̃1k d̃2k d̃2k−1 + d̃1k d̃2k d̃1k−1 + d̃1k d̃2k d̃2k−1 , (3.110)
b2k = d̃1k d̃1k−1 d̃2k−1 + d̃2k d̃1k−1 d̃2k−1 + d̃1k d̃1k−1 d̃2k−1 + d̃2k d̃1k−1 d̃2k−1 . (3.111)
Table 3.2 Truth table of the differential decoder for QPSK when using the data recovery logic
described in Sect. 3.2.3 for the recovery of the data defined by the Gray coded bit mapping depicted
in Fig. 2.6 (special case of equal input and output mappings)
d̃1k d̃2k ϕk d̃1k−1 d̃2k−1 ϕk−1 b1k b2k ϕbk
0 0 3/2 · π 0 0 3/2 · π 0 1 0
0 0 3/2 · π 0 1 0 0 0 3/2 · π
0 0 3/2 · π 1 0 π 1 1 π/2
0 0 3/2 · π 1 1 π/2 1 0 π
0 1 0 0 0 3/2 · π 1 1 π/2
0 1 0 0 1 0 0 1 0
0 1 0 1 0 π 1 0 π
0 1 0 1 1 π/2 0 0 3/2 · π
1 0 π 0 0 3/2 · π 0 0 3/2 · π
1 0 π 0 1 0 1 0 π
1 0 π 1 0 π 0 1 0
1 0 π 1 1 π/2 1 1 π/2
1 1 π/2 0 0 3/2 · π 1 0 π
1 1 π/2 0 1 0 1 1 π/2
1 1 π/2 1 0 π 0 0 3/2 · π
1 1 π/2 1 1 π/2 0 1 0
Alternatively to re-using the data recovery circuit employed for differential de-
tection, it can be advantageous to design another data recovery logic which allows
for the re-use of the differential encoder as a differential decoder. As can be observed
from the comparison of (3.110)–(3.111) with the logical relations for the differen-
tial encoders given by (2.34)–(2.35) and (2.36)–(2.37), the QPSK decoder described
above is not identical with both DQPSK encoders. However, it can be shown that
the differential encoder for the parallel transmitter can be also used as a differential
decoder when the data recovery is based on the symbol assignment of the constel-
lation diagram shown in Fig. 2.9, bottom left. The data recovery logic must then be
defined as d̃1k = si,1k and d̃2k = sq,1k , in contrast to Sect. 3.2.3. In this special case,
the DQPSK encoder and the QPSK decoder feature the same configuration.
In the same manner, the logical circuits of the differential decoders for higher-
order PSK formats can be derived. Knowing the symbol assignment to absolute
phase states determined by the data recovery logic, and considering the original
mapping of the data bits to symbols, the truth tables can be established for the par-
ticular decoders and the logical relations can easily be obtained by evaluating these
truth tables.
When looking at the data recovery for Star QAM formats, the bits containing
the phase information can be obtained by arg-decision in the same way as for the
PSK formats. In the case of Star 16QAM, the bit b4k —which encodes the amplitude
information—can be recovered by performing a single decision
q on the magnitude
of the detected QAM symbol which is given by ak′ = Ik2 + Q 2k . The ideal po-
sition of the threshold depends on the ring ratio adjusted at the transmitter. When
the Star QAM signals have more than two amplitude levels, several decisions on a
multi-level signal become necessary in order to obtain the different data bits which
contain the amplitude information.
In the case of Square QAM, the position of the constellation diagram after carrier
recovery is ambiguous with n times π/2 (n = 0, 1, 2, 3). This is denoted as quad-
rant ambiguity. For this reason, a differential quadrant encoding of two of the m bits
of a symbol, b1k and b2k , is performed at the transmitter, as illustrated in Sect. 2.6.1,
and must be reversed at the receiver by a QPSK differential decoder. The differen-
tial decoder is placed behind the data recovery logic and finally recovers the data
bits b1k and b2k from the differentially encoded quadrant bits d̃1k and d̃2k and the
delayed bits d̃1k−1 and d̃2k−1 , as shown in Fig. 3.30.
si ,1k d#1k
b1k
d#
Differential
Ik
….
1k −1
Decoder
TS
QPSK
s( i , d#2k
Multiplexer
from M −1) k b2k

Recovery
d#
Data
Logic
carrier
Data
2k −1
sq ,1k TS
recovery
Qk b3k
….
s( q ,
….
M −1) k bmk
Fig. 3.30 Data recovery for synchronous detection of Square QAM signals
The remaining m − 2 data bits b3k , .., bm k are arranged so as to be symmetric in

rotation, so that the quadrant ambiguity does not have any influence on them. No
differential decoding is necessary to recover these bits. In Fig. 3.31, the constella-
tion diagram of Square 16QAM is shown with the symbol assignment defined in
Sect. 2.6, also illustrating the placement of the thresholds.
{b3 , b4 } {d# , d# }
1 2
01 11
11 10 01 11
Sq,1
01 00 00 10
Sq,2
10 00 00 01
Sq,3
Fig. 3.31 Symbol assignment
and placement of the thresh- 11 01 10 11
olds for Square 16QAM data 00 10
recovery Si,3 Si,2 Si,1
The bits on the corners determine the quadrant and are easily obtained from a data
recovery logic defined by d̃1k = si,2k and d̃2k = sq,2k , where the decision results of
the decision circuits in the in-phase and quadrature branches are specified by

1 Ik > si,n 1 Q k > sq,n
si,n k = , sq,n k = , (3.112)
0 Ik < si,n 0 Q k < sq,n
√
with n = (1, .., M −1) and si,n and sq,n representing the thresholds in the in-phase
and quadrature branches, respectively. The data bits b1k and b2k are then obtained at
the two outputs of the QPSK differential decoder. The remaining two data bits b3k
and b4k can be recovered by a data recovery logic specified by
b3k = si,1k sq,2k + si,2k sq,1k + si,3k sq,2k + si,2k sq,3k , (3.113)
b4k = si,2k sq,1k + si,3k sq,2k + si,2k sq,3k + si,1k sq,2k . (3.114)
The constellation diagram of Square 64QAM together with the thresholds is de-
picted in Fig. 3.32—assuming the symbol assignment defined in Sect. 2.6. The two
bits which define the quadrant are obtained in the same way as for Square 16QAM.
After performing decisions at the thresholds si,4 and sq,4 , the differentially encoded
bits d̃1k and d̃2k are decoded by a QPSK differential decoder. This yields the data
bits b1k and b2k . By the way, the differential decoder has an identical configuration
for all Square QAM formats.
{b3 , b4 , b5 , b6 } {d# , d# }
1 2
01 11
1111 1110 1011 1010 0101 0111 1101 1111
Sq,1
1101 1100 1001 1000 0100 0110 1100 1110
0001 Sq,2
0111 0110 0011 0010 0011 1001 1011
0001 0000 Sq,3

0101 0100 0000 0010 1000 1010
Sq,4
1010 1000 0010 0000 0000 0001 0100 0101

Sq,5
1011 1001 0011 0001 0010 0011 0110 0111

Sq,6
1110 1100 0110 0100 1000 1001 1100 1101

Sq,7
1111 1101 0111 0101 1010 1011 1110 1111

00 10
Si,7 Si,6 Si,5 Si,4 Si,3 Si,2 Si,1
Fig. 3.32 Symbol assignment and thresholds for Square 64QAM data recovery
The remaining four bits b3k , .., b6k are again arranged as rotation symmetric as
it is depicted in Fig. 3.32. The logical circuit for the data recovery of these bits is
defined by
b5k = si,1k sq,4k + si,2k si,3k sq,4k + si,4k sq,1k + si,4k sq,2k sq,3k
+si,7k sq,4k + si,5k si,6k sq,4k + si,4k sq,7k + si,4k sq,5k sq,6k , (3.117)
b6k = si,4k sq,1k + si,4k sq,2k sq,3k + si,7k sq,4k + si,5k si,6k sq,4k
+si,4k sq,7k + si,4k sq,5k sq,6k + si,1k sq,4k + si,2k si,3k sq,4k . (3.118)
References
1. CeLight, http://www.celight.com
2. VPIsystems, Photonic Modules Reference Manual, 2005
3. Barry, J.R., Kahn, J.M.: Carrier synchronization for homodyne and heterodyne detection of
optical quadriphase-shift keying. IEEE Journal of Lightwave Technology 10(12), 1939–1951
(1992)
4. Benedetto, S., Poggiolini, P.: Theory of polarization shift keying modulation. IEEE Transac-
tions on Communications 40(4), 708–721 (1992)
5. Cai, Y., Pilipetskii, A.N.: Comparison of two carrier phase estimation schemes in optical co-
herent detection systems. In: Proceedings of Optical Fiber Communication Conference (OFC),
OMP5 (2007)
6. Camatel, S., et al.: Optical phase-locked loop for coherent detection optical receiver. Elec-
tronics Letters 40(6), 384–385 (2004)
7. Camatel, S., et al.: 2-PSK homodyne receiver based on a decision driven architecture and a
sub-carrier optical PLL. In: Proceedings of Optical Fiber Communication Conference (OFC),
OTuI3 (2006)
8. Chiou, Y., Wang, L.: Effect of amplifier noise on laser linewidth requirements in long haul
optical fiber communication systems with Costas PLL receivers. IEEE Journal of Lightwave
Technology 14(10), 2126–2134 (1996)
9. Cusani, R., et al.: An efficient multilevel coherent optical system: M-4Q-QAM. IEEE Journal
of Lightwave Technology 10(6), 777–786 (1992)
10. Davis, A.W., et al.: Phase diversity techniques for coherent optical receivers. IEEE Journal of
Lightwave Technology LT-5(4), 561–572 (1987)
11. Desurvire, E.: Erbium-Doped Fiber Amplifiers. John Wiley & Sons, Inc. (1994)
12. Djordjevic, I.B., et al.: Performance optimization and exact laser linewidth requirements eval-
uation for optical PSK homodyne communication systems with Costas loop or DDL. Journal
of Optical Communications 20(5), 178–182 (1999)
13. Franz, J.: Optische Übertragungssysteme mit Überlagerungsempfang. Springer Verlag,
Berlin/Heidelberg (1988)
14. Gardner, F.M.: A BPSK/QPSK timing-error detector for sampled receivers. IEEE Transactions
on Communications COM-34(5), 423–429 (1986)
15. Goldfarb, G., Li, G.: BER estimation of QPSK homodyne detection with carrier phase esti-
mation using digital signal processing. Optics Express 14(18), 8043–8053 (2006)
16. Haykin, S.: Communication Systems. John Wiley & Sons, Inc. (1994)
17. Ho, K.P.: Phase-Modulated Optical Communication Systems. Springer (2005)
18. Hodgkinson, T.G.: Costas loop analysis for coherent optical receivers. Electronics Letters
22(7), 394–396 (1986)
19. Hoffmann, D., et al.: Integrated optics eight-port 90◦ -hybrid on Li N bO3 . IEEE Journal of
Lightwave Technology 7(5), 794–798 (1989)
20. Hongou, J., et al.: 1 Gsymbol/s, 64 QAM coherent optical transmission over 150 km with a
spectral efficiency of 3 bit/s/Hz. In: Proceedings of Optical Fiber Communication Conference
(OFC), OMP3 (2007)
21. Horikawa, I., et al.: Design and performance of a 200Mbit/s 16QAM digital radio system.
IEEE Transactions on Communications COM-27(12), 1953–1958 (1979)
22. Ip, E., Kahn, J.M.: Feedforward carrier recovery for coherent optical communications. IEEE
Journal of Lightwave Technology 25(9), 2675–2692 (2007)
23. Jorgensen, B.F., et al.: Analysis of amplifier noise in coherent optical communication systems
with optical image rejection receivers. IEEE Journal of Lightwave Technology 10(5), 660–671
(1992)
24. Kahn, J.M.: BPSK homodyne detection experiment using balanced optical phase-locked loop
with quantized feedback. IEEE Photonics Technology Letters 2(11), 840–843 (1990)
25. Kahn, J.M., et al.: Heterodyne detection of 310-Mb/s quadriphase-shift keying using fourth-
power optical phase-locked loop. IEEE Photonics Technology Letters 4(12), 1397–1400
(1992)
References 117
26. Kaplan, A., Achiam, K.: Li N bO3 integrated optical QPSK modulator and coherent receiver.
In: Proceedings of ECIO, WeA3.2, pp. 79–82 (2003)
27. Kazovsky, L.G.: Decision-driven phase-locked loop for optical homodyne receivers: Perfor-
mance analysis and laser linewidth requirements. IEEE Journal of Lightwave Technology
LT-3(6), 1238–1247 (1985)
28. Kazovsky, L.G.: Balanced phase-locked loops for optical homodyne receivers: Performance
analysis, design considerations, and laser linewidth requirements. IEEE Journal of Lightwave
Technology LT-4(2), 182–195 (1986)
29. Kazovsky, L.G.: Phase- and polarization-diversity coherent optical techniques. IEEE Journal
of Lightwave Technology 7(2), 279–292 (1989)
30. Kazovsky, L.G., et al.: Wide-linewidth phase diversity homodyne receivers. IEEE Journal of
Lightwave Technology 6(10), 1527–1536 (1988)
31. Kazovsky, L.G., et al.: Homodyne phase-shift-keying systems: Past challenges and future op-
portunities. IEEE Journal of Lightwave Technology 24(12), 4876–4884 (2006)
32. Koc, U., et al.: Digital coherent quadrature phase-shift-keying (QPSK). In: Proceedings of
Optical Fiber Communication Conference (OFC), OThI1 (2006)
33. Kuschnerov, M., et al.: Joint equalization and timing recovery for coherent fiber optic re-
ceivers. In: Proceedings of European Conference on Optical Communication (ECOC),
Mo.3.D.3 (2008)
34. Kuschnerov, M., et al.: Joint-polarization carrier phase estimation for XPM-limited coherent
polarization-multiplexed QPSK transmission with OOK-neighbors. In: Proceedings of Euro-
pean Conference on Optical Communication (ECOC), Mo.4.D.2 (2008)
35. Langenhorst, R.: Optische Koppelelemente für den kohärent optischen Mehrtorempfänger.
Ph.D. thesis, Technische Universität Berlin (1992)
36. Leeb, W.R.: Optical 90◦ hybrid for Costas-type receivers. Electronics Letters 26, 1431–1432
(1990)
38. Ly-Gagnon, D.S., et al.: Unrepeated 210-km transmission with coherent detection and digital
signal processing of 20-Gb/s QPSK signal. In: Proceedings of Optical Fiber Communication
Conference (OFC), OTuL4 (2005)
39. Ly-Gagnon, D.S., et al.: Coherent detection of optical quadrature phase-shift keying signals
with carrier phase estimation. IEEE Journal of Lightwave Technology 24(1), 12–21 (2006)
40. Meijerink, A., et al.: Balanced optical phase diversity receivers for coherence multiplexing.
41. Nakazawa, M., et al.: 20 Msymbol/s, 128 QAM coherent optical transmission over 500 km
using heterodyne detection with frequency-stabilized laser. In: Proceedings of European Con-
ference on Optical Communication (ECOC), Mo4.2.2 (2006)
42. Nakazawa, M., et al.: Polarization-multiplexed 1 Gsymbol/s, 64 QAM (12 Gbit/s) coherent
optical transmission over 150 km with an optical bandwidth of 2 GHz. In: Proceedings of
43. Nicholson, G., Stephens, T.D.: Performance analysis of coherent optical phase-diversity re-
ceivers with DPSK modulation. IEEE Journal of Lightwave Technology 7(2), 393–399 (1989)
44. Noe, R.: PLL-free synchronous QPSK polarization multiplex/diversity receiver concept with
digital I&Q baseband processing. IEEE Photonics Technology Letters 17(4), 887–889 (2005)
45. Noe, R., et al.: New FSK phase-diversity receiver in a 150 Mbit/s coherent optical transmission
system. Electronics Letters 24(9), 567–568 (1988)
46. Norimatsu, S., Iwashita, K.: Linewidth requirements for optical synchronous detection sys-
temes with nonnegligible loop delay time. IEEE Journal of Lightwave Technology 10(3),
341–349 (1992)
47. Norimatsu, S., et al.: An 8 Gb/s QPSK optical homodyne detection experiment using external-
cavity laser diodes. IEEE Photonics Technology Letters 4(7), 765–767 (1992)
48. Oerder, M., Meyr, H.: Digital filter and square timing recovery. IEEE Transactions on Com-
munications 36(5), 605–612 (1988)
50. Paiam, M.R., MacDonald, R.I.: Design of phased-array wavelength division multiplexers us-
ing multimode interference couplers. Applied Optics 36(21), 5097–5108 (1997)
51. Petermann, K.: Einführung in die optische Nachrichtentechnik. Vorlesungsskript, Technische
Universität Berlin (2003)
52. Pietzsch, J.: Scattering matrix analysis of 3 × 3 fiber couplers. IEEE Journal of Lightwave
Technology 7(2), 303–307 (1989)
53. Priest, R.G.: Analyis of fiber interferometer utilizing 3 × 3 fiber coupler. IEEE Journal of
Quantum Electronics QE-18(10), 1601–1603 (1982)
54. Proakis, J.G.: Digital Communications. McGraw-Hill (2001)
55. Rice, F.: Bounds and algorithms for carrier frequency and phase estimation. Ph.D. thesis,
University of South Australia (2002)
56. Savory, S.J.: Compensation of fibre impairments in digital coherent systems. In: Proceedings
of European Conference on Optical Communication (ECOC), Mo.3.D.1 (2008)
57. Savory, S.J., et al.: Transmission of 42.8Gbit/s polarization multiplexed NRZ-QPSK over
6400km of standard fiber with no optical dispersion compensation. In: Proceedings of Op-
tical Fiber Communication Conference (OFC), OTuA1 (2007)
58. Seimetz, M.: Optical receiver for reception of M-ary star-shaped quadrature amplitude modu-
lation with differentially encoded phases and its application. German patent, DE 10 2006 030
915.4 (2006)
59. Seimetz, M.: Phase diversity receivers for homodyne detection of optical DQPSK signals.
60. Seimetz, M.: Laser linewidth limitations for optical systems with high-order modulation em-
ploying feed forward digital carrier phase estimation. In: Proceedings of Optical Fiber Com-
munication Conference (OFC), OTuM2 (2008)
61. Seimetz, M., Weinert, C.M.: Options, feasibility and availability of 2 × 4 90◦ -hybrids for
coherent optical systems. IEEE Journal of Lightwave Technology 24(3), 1317–1322 (2006)
63. Sun, L., Ye, P.: Optical homodyne receiver based on an improved balance phase-locked loop
with the data-to-phaselock crosstalk suppression. IEEE Photonics Technology Letters 2(9),
678–680 (1990)
64. Taylor, M.G.: Accurate digital phase estimation process for coherent detection using a par-
allel digital processor. In: Proceedings of European Conference on Optical Communication
(ECOC), Tu4.2.6 (2005)
65. Taylor, M.G.: Coherent detection for optical communications using digital signal processing.
In: Proceedings of Optical Fiber Communication Conference (OFC), OMP1 (2007)
67. Wang, Y., Leeb, W.R.: Sensitivity analysis and performance optimization of an optical Costas
phase-locked loop. Journal of Optical Communications 8(1), 29–31 (1987)
68. von Wangenheim, L.: Aktive Filter in RC- und SC-Technik. Hüthig Buch Verlag Heidelberg
(1991)
69. Webb, W., Hanzo, L.: Modern Quadrature Amplitude Modulation. IEEE Press and Pentech
Press (1994)
70. Xiong, F.: Digital Modulation Techniques. Artech House, Inc. (2000)
71. Yoon, H., et al.: Performance comparison of optical 8-ary differential phase-shift keying sys-
tems with different electrical decision schemes. Optics Express 13(2), 371–376 (2005)
Chapter 4
Effort Comparison
Abstract When a system is upgraded to a more sophisticated modulation format, the

additional effort required is determined, in large part, by the enhanced complexity of
the transmitters and receivers. In this chapter, the complexity of different transmitter
and receiver configurations is briefly discussed and compared.
It was illustrated at the beginning of this book that the deployment of high-order
modulation formats and coherent detection is presently seen as a promising way
of increasing the spectral efficiency of optical fiber transmission and of better ex-
ploiting the capacity of the currently installed fiber infrastructure. Some questions
were posed in the introduction which system designers would like to have answered
in order to rank the potential of this new technology. Two of these are: What do
the transmitters and receivers look like? How great is the additional effort required
by a system upgrade? The last two chapters have provided a lot of material which
assists in answering these questions and have described the composition of many
optical transmitters and receivers. Based on the information given in these chapters,
an explicit discussion and comparison of the complexity of different transmitter and
receiver configurations takes place in the following two sections.
4.1 Transmitter Complexity and Feasibility
The suitability and feasibility of a particular transmitter concept depends on the

modulation format it is applied to. For circular constellations, transmitter config-
urations with a single modulator or several consecutive modulators are favorable,
whereas IQ-structure based transmitters are more beneficial for generating square-
shaped constellations.
Generally, the migration to higher-order formats brings about an increase in
transmitter complexity. The upgrade can be performed by adding optical modu-
lators and accordingly creating more elaborate optical modulator structures or by


120 4 Effort Comparison
providing more complex electrical level generators for the generation of multi-level
electrical driving signals. As illustrated in Fig. 4.1, the overall complexity of the
transmitters can be traded-off between the optical and electrical parts. Since eye
spreading complicates the generation of high-quality multi-level electrical driving
signals for high data rates through analog means, transmitters with binary electrical
driving signals may be preferred.
Optical
Electrical complexity
complexity
Fig. 4.1 Transmitter com- Generation of More complex

plexity: trade-off between the multi-level optical modulator
optical and electrical parts driving signals Trade-Off structures
The description of DPSK and Star QAM transmitters in Sect. 2.4 and Sect. 2.5
focusses on configurations which require only binary electrical driving signals. Ta-
ble 4.1 compares the complexity of these transmitters for various modulation for-
mats. 2ASK acts as reference (“0”), and all the other formats are characterized with
one or more “+” to evaluate their particular complexity.
Table 4.1 Comparison of the complexity of DPSK and Star QAM transmitters with binary elec-
trical driving signals for various modulation formats. Se.: Serial, Pa.: Parallel
Mod. Format Tx Type PM/MZM/IQM Opt. Complexity Encoder Type El. Complexity
2ASK - 0/1/0 0 - 0
DBPSK - 1/0/0 0 DBPSK +
DQPSK Se. 2/0/0 + DQPSK (Se.) ++
Pa. 0/0/1 ++ DQPSK (Pa.) ++
8DPSK Se. 3/0/0 ++ 8DPSK (Se.) +++
Pa. 1/0/1 +++ 8DPSK (Pa.) +++
16DPSK Se. 4/0/0 +++ 16DPSK (Se.) +++++
Pa. 2/0/1 ++++ 16DPSK (Pa.) +++++
Star 16QAM Se. 3/1/0 +++ 8DPSK (Se.) +++
Pa. 1/1/1 ++++ 8DPSK (Pa.) +++
The complexity of the electrical part of the transmitters depends mainly on the
complexity of the differential encoder. This complexity increases disproportionately
to the number of differentially encoded phase states, so that the differential encoder
becomes very complex for high-order DPSK formats such as 16DPSK. When com-
paring the optical part of the transmitters, the effort required in creating an IQ mod-
ulator within the parallel transmitter is higher than for the creation of a pure serial
4.1 Transmitter Complexity and Feasibility 121
configuration consisting of consecutive phase modulators. Furthermore, the optical

effort required increases with the order of the modulation format and the corre-
sponding increasing number of modulators. The number of necessary modulators
given in Table 4.1 does not include the MZM used for RZ pulse carving.
In principle, the high complexity of the optical transmitter part for higher-order
DPSK and Star QAM formats can be reduced by using a single conventional IQ
modulator, but this necessitates a further increase in electrical complexity because
multi-level electrical driving signals must be generated. Moreover, it should be noted
that the in-phase and quadrature driving signals for circular constellations with many
phase states have very small distances between signal levels, which is detrimental
for system implementation.
As regards transmitters for Square QAM, five different transmitter structures are
described in Sect. 2.6. The complexity of these transmitters for Square 16QAM is
compared in Table 4.2.
Table 4.2 Comparison of the complexity of different transmitters for Square 16QAM
Transmitter PM/MZM/IQM Opt. Compl. Enc. Type QAM Cod./Lev. Gen. El. Compl.
Serial QAM 1/1/0 + DQPSK No / Yes (+) +++++

Convent. IQ 0/0/1 ++ DQPSK Yes / Yes ++++
Enhanced IQ 2/0/1 +++ DQPSK Yes / No +++
Tandem-QPSK 2/0/1 ++++ DQPSK No / No ++
Quad-pa. MZM 0/0/2 +++++ DQPSK Yes / No +++
The serial Square QAM transmitter features a simple optical part, but the elec-
trical level generator must generate electrical driving signals with a high number
of levels and is very complex. To generate square-shaped constellations, the con-
ventional IQ transmitter is a good choice, featuring beneficial chirp characteristics
and a moderate optical complexity, but still requiring multi-level electrical driving
signals. However, a restriction to binary driving signals—and thus a decrease of
electrical complexity—is possible even for Square QAM. The enhanced IQ trans-
mitter, the Tandem-QPSK transmitter and the quad-parallel MZM transmitter re-
quire only binary driving signals for Square 16QAM (an electrical level generator
is not needed) and are therefore three promising options for implementation of a
practical system—in spite of their more complex optical parts.
In any Square QAM transmitter and for any Square QAM order, the same differ-
ential encoder—a DQPSK encoder with relatively moderate complexity—is appro-
priate when the quadrant ambiguity arising at the carrier synchronization in the re-
ceiver shall be resolved by differential decoding. However, a further coder (denoted
within this book as Square QAM coder) must then be provided in the conventional
IQ transmitter, the enhanced IQ transmitter and the quad-parallel MZM transmitter
to create a bit mapping which is symmetric in rotation.
The spectral efficiency of any modulation format can be doubled using polariza-
tion division multiplexing. In this case, the transmitter complexity is approximately
double for any configuration.
The respective transmitters can be rated on the basis of the effort spent in de-
veloping their optical and electrical transmitter parts, but also by considering the
influence of individual signal properties such as intensity shape, symbol transitions
and chirp characteristic on the overall system performance. This issue, among many
others, is discussed in the second part of this book.
4.2 Receiver Complexity and Feasibility
Various receiver configurations can be used to detect optical high-order modulation

signals. These can be based on direct detection or coherent detection, as shown
in Chap. 3. The receiver scheme suitable for a particular system must be chosen
according to the modulation format received, as well as through consideration of the
given system specifications (e.g. sensitivity, laser linewidth) and desired accessory
functionalities, such as electronic equalization. Furthermore, the complexity and
feasibility of the receivers represents an important criterion for a decision.
Direct detection receivers are suitable for detecting DPSK and Star QAM sig-
nals. Their optical frontend can be constructed by delay line interferometers or a
2 × 4 90◦ hybrid. Table 4.3 summarizes the complexity of the optical and electrical
parts of two different direct detection receivers described in Sect. 3.2.1: the multiple
DLI receiver and the IQ receiver composed of DLIs.
Table 4.3 Complexity comparison of direct detection receivers (multiple DLI receiver and direct
detection IQ receiver) for various modulation formats; BD: balanced detector, PD: photodiode
Mod. Format RX Type DLI and BD Opt. Complexity El. Complexity
2ASK - - 0 0
DBPSK - 1 + 0
DQPSK Mult. DLI 2 ++ +
IQ 2 ++ +
8DPSK Mult. DLI 4 +++ ++
IQ 2 ++ +++
16DPSK Mult. DLI 8 +++++ +++
IQ 2 ++ ++++
Star 16QAM Mult. DLI 4 (+1 PD) ++++ +++
IQ 2 (+1 PD) +++ ++++
4.2 Receiver Complexity and Feasibility 123
Similarly to the transmitters, the complexity of the electrical receiver parts can, in
principle, be reduced through the introduction of more complex optical structures,
for instance a higher amount of delay line interferometers for direct detection. The
multiple DLI receiver exhibits a high optical complexity for high-order formats,
but electrical decisions can be performed very easily with fixed thresholds at zero.
When an IQ receiver is employed and the number of delay line interferometers is
reduced to two for any modulation format, decisions on multi-level electrical signals
with multiple thresholds become necessary for modulation formats with more than
two bits per symbol. Alternatively, the two DLIs can be replaced by a 2 × 4 90◦
hybrid (see Sect. 3.2.2). This IQ receiver then exhibits a similar optical complexity
and the same electrical complexity as the direct detection IQ receiver with DLIs.
The optical part of the IQ receivers is identical for all high-order phase modulation
formats. As regards Star QAM, the additional intensity detection branch slightly
increases the complexity in comparison to DPSK receivers. The complexity of the
electrical part of the receivers is determined primarily by the number of decision
circuits necessary and the complexity of the data recovery logic, and thus increases
with the modulation format order.
Table 4.4 compares the complexity of receivers adopting different detection
schemes, including the direct detection IQ receiver, the homodyne receiver with
differential detection and the digital coherent receiver with homodyne synchronous
detection for single-polarization and polarization division multiplexing.
Table 4.4 Complexity comparison of different detection schemes: direct detection, homodyne dif-
ferential detection and digital homodyne synchronous detection for single-polarization and polar-
ization division multiplexing (PDM)
Receiver Scheme 90◦ Hy./BD Opt. Compl. Pol.-/Fre.-/Ph.-Syn. Decoder El. Compl.
Direct Det. IQ 1/2 0 No / No / No No 0

Hom. Diff. Det. 1/2 + Yes / Yes / No No +
Hom. Syn. Det. 1/2 + Yes / Yes / Yes Yes ++
Hom. Syn. PDM 2/4 ++ No / Yes / Yes Yes ++++
The optical frontend of homodyne receivers for high-order modulation is usually

composed of a 2 × 4 90◦ hybrid followed by two balanced detectors. Although the
complexity of the optical frontend of the direct detection IQ receiver and the ho-
modyne receivers is similar (the optical complexity is rated here slightly higher for
the homodyne receivers due to the LO required additionally), the direct detection
IQ receiver is simpler, since a carrier synchronization is not needed. When per-
forming homodyne differential detection, differential demodulation must be imple-
mented in the electrical domain. Thereby, the laser phase noise is not such a critical
task, but frequency synchronization and polarization alignment of the signal and
the LO wave must be ensured. In the case of homodyne synchronous detection, the
carrier phase synchronization becomes a practical challenge, especially for high-
order modulation formats. Fortunately, high-speed digital signal processing offers
new alternatives for the implementation of carrier synchronization. Receivers with

homodyne synchronous detection in particular are reliant on digital technology to
replace the analog optical phase locked loop with digital phase estimation.
When comparing direct detection receivers and homodyne receivers, it should
be borne in mind that, in homodyne receivers, the polarization of the signal and LO
waves must be aligned by a polarization control or better the optical quadrature fron-
tend should be doubled to implement polarization diversity. In the latter case, it is a
nice side-effect that signals with polarization division multiplexing can be received
without further modifications of the optical frontend. This, however, necessitates a
significantly more complex electrical receiver part.
The data recovery logic of direct detection IQ receivers is similar to that of homo-
dyne receivers. Multi-level in-phase and quadrature photocurrents arise for higher-
order formats, and the effort required for electrical data recovery increases according
to the order of the modulation format. When the phase ambiguity of the carrier syn-
chronization is resolved by differential coding, a differential decoder must be placed
behind the carrier recovery in homodyne synchronous receivers. As for carrier syn-
chronization, digital signal processing can greatly facilitate data recovery for direct
detection and coherent detection and will play an important role in future receiver
realizations.
Despite their higher complexity, coherent receivers offer their own benefits such
as the high receiver sensitivity (even under abdication of optical pre-amplification),
an enhanced potential for the electronic mitigation of transmission impairments, a
possible application as tunable WDM receivers, as well as suitability for detection
of arbitrary modulation formats with homodyne synchronous detection.
When looking ahead, it is a future challenge for practical system implementation

to develop integrated novel optical modulator structures and high-speed digital-to-
analog converters for the generation of high-quality multi-level electrical driving
signals on the transmitter side. At the receiver end, practical solutions which inte-
grate the whole optical frontend (the 2 × 4 90◦ hybrid together with the balanced
detectors) in a single chip are being developed. Moreover, further technological
progress and improved algorithms in the field of digital signal processing are in-
dispensable for the future commercial implementation of optical transmitters and
receivers for high-order modulation formats.
The first part of this book illustrated the generation and detection of optical high-
order modulation signals and provided a detailed insight into the setup, complexity
and properties of various transmitter and receiver configurations. The second part of
the book now to come deals with the system and transmission characteristics of op-
tical high-order modulation signals and discusses the influence of many transmitter
and receiver configurations described in the first part of the book on system perfor-
mance. This allows the different transmitters and receivers to be rated, not only on
the basis of their complexity and practical feasibility as performed in this chapter,
but also by considering aspects of their system performance.
Chapter 5
System Simulation Aspects
Abstract Computer simulations offer a cost-effective way of estimating the physical

properties and the performance of communication systems before practical system
implementation. In this chapter, some aspects relevant for the simulation of optical
fiber transmission systems with high-order modulation are briefly discussed. The
discussion covers data sequences and performance measures commonly used for
system characterization. Moreover, a detailed illustration of semi-analytical BER
estimation provides a deeper insight into the noise characteristics within direct de-
tection receivers.
5.1 Data Sequences
An appropriate choice of data sequences is of crucial importance for reliable results

in system simulations and experiments, especially when investigating high-order
modulation formats. In the simulation of optical transmission systems, pseudo ran-
dom binary sequences (PRBS) are widely used for representation of the data. In
this section, three pseudo random binary sequences and the proper choice of data
sequences according to the system memory length are discussed.
Maximum Length Sequences
By using a linear feedback shift register with n B binary storage elements and feed-
backs at the appropriate locations, a so-called maximum length sequence with a
period of 2n B − 1 which contains all possible n B -bit patterns except for the one
with n B zeros can be produced. Each pattern appears only once within each period.
Several options exist for generating maximum length sequences for given n B [6]. It
can be observed from the polynomials listed in [6] where the feedbacks have to be
placed. Initially, the shift register is loaded with an initial value, often denoted as


128 5 System Simulation Aspects
seed, which can be any number between 1 and 2n B − 1. A change of seed causes a
shift of the periodic sequence.
De Bruijn Bit Sequences
In order to be able to simulate the effects of inter-symbol interference (ISI) in a

system with binary modulation and a memory length of n B bits correctly, one has
to ensure that all possible n B -bit patterns are incorporated in the used bit sequence.
When enhancing a maximum length sequence by adding a zero digit to the run with
n B − 1 zeros, a pseudo random binary sequence results which contains all n B -bit
patterns. Such a sequence is called “de Bruijn bit sequence”. In a de Bruijn bit
sequence, the ones and zeros occur each with a probability of 1/2.
De Bruijn Symbol Sequences
For modulation formats with more than one bit per symbol, the requirement of con-
taining all n B -bit patterns changes to the requirement of containing all n S -symbol
patterns, where n S must be chosen according to the memory length of the system in
symbols. A de Bruijn symbol sequence with b = m·M n S bits contains all n S -symbol
patterns of M symbols, each consisting of m bits. When looking at Star 16QAM, for
instance, the de Bruijn symbol sequence has 163 = 4096 symbols when considering
all 3-symbol patterns. Four bits are assigned to each symbol, so that the resulting
binary data sequence has a length of b = 4096 · 4 = 16384 bits. For higher memory
lengths and high-order modulation formats, de Bruijn symbol sequences become
quite long and can exceed the limits of simulative feasibility. When using shorter
sequences, ISI-induced signal distortions are not further accounted for ideally. Like
de Bruijn bit sequences, de Bruijn symbol sequences can be generated by using lin-
ear feedback shift registers. Instead of the modulo-2 arithmetic, arithmetic in Galois
fields must then be used [6].
Choice of Data Sequences According to the System Memory Length
The required data sequence length is determined by the number of symbols which
can interact in the investigated system due to ISI. Inter-symbol interference is caused
by optical and electrical filtering, as well as by linear and nonlinear fiber degradation
effects. For a nonlinear dispersive channel, all the ISI caused by dispersion and intra-
channel Kerr nonlinearities can be captured by considering the number of symbols
in a time interval τ corresponding to the relative time shift experienced by two
spectral components separated by the width of the signal spectrum Bs [16]. Thus,
the approximate required order of the data sequence n S can be estimated by
nS c
≥ τ = Dmax 2 Bs + 1. (5.1)
rS fs
5.1 Data Sequences 129
When considering only CD and observing a simple fiber link without CD compen-
sation, Dmax simply represents the maximum dispersion which has accumulated on
the link. For instance, a maximum accumulated dispersion of Dmax = 320 ps/nm
corresponds to a length of 20 km for a single mode fiber with a dispersion co-
efficient of Dλ = 16 ps/(nm·km). With f s = 193.1 THz, r B = 40 Gbit/s and
r S = r B /m = Bs , the de Bruijn symbol sequence order must be chosen as
n S ≥ (6, 3, 2, 2, 2) for m = (1, 2, 3, 4, 6) which leads to relaxed requirements
on the data sequence length of b = M n S · m = (26 , 43 · 2, 82 · 3, 162 · 4, 642 · 6).
In practical systems with chromatic dispersion, intra-channel Kerr nonlinearities
and CD compensation, Dmax stands for the maximum magnitude reached by the
accumulated dispersion at locations in the dispersion map where the signal power
is sufficiently large to generate nonlinear effects. When looking at a fiber link of
length l = 80 km with 100% CD post-compensation, for instance, the effective
length le f f = 1 − e−α·l /α, corresponding to the length beyond which the addi-
tionally accumulating nonlinear phase shift becomes negligible [1], is about 21km
for an attenuation of αdb = 0.2 dB/km. In that case, Dmax ≈ 320 ps/nm can also
be assumed when implying that the dispersion beyond the effective length can be
ideally compensated for.
To better illustrate the influence of the data sequence type and length on system
characterization, some examples are given in the following paragraphs. Figure 5.1a
shows the dependence of the eye opening penalty (EOP, defined in Sect. 5.2) on
different de Bruijn bit and de Bruijn symbol sequences for linear fiber transmission,
an accumulated dispersion of 320 ps/nm and a data rate of 40 Gbit/s for some se-
lected modulation formats and direct detection. Inter-symbol interference induced
by optical and electrical receiver filters with 3 dB bandwidths of Bopt = 2.5 · r S and
Bel = 0.75 · r S is also included.
5 1E-2
Eye Opening Penalty [dB]
a b
4 12
De Bruijn bit 2
1E-3
3 DQPSK NRZ
De Bruijn bit 2
14
BER
8DPSK RZ
2
1E-4
De Bruijn bit 216
1 Star 16QAM RZ
Square 64QAM NRZ

0 1E-5
2 4 6 8 10 12 14 16 18 20 -37 -36 -35 -34 -33 -32
Length of data sequence log2 b Received Power [dBm]
Fig. 5.1 a EOP dependence on the data sequence length b for 320 ps/nm accumulated dispersion at
40 Gbits/s for different modulation formats; Solid lines: de Bruijn bit sequences of length b = 2n B ;
Symbols: de Bruijn symbol sequences with a period of b = m · M n S bits. b BER vs. received power
for different de Bruijn bit sequences and back-to-back MC simulations for Square 64QAM.
A de Bruijn bit sequence with the same data sequence length b as the correspond-
ing de Bruijn symbol sequence, for instance b = 212 = 4 · 163 , does not contain all
symbol patterns of the de Bruijn symbol sequence. However, as becomes apparent

from Fig. 5.1a, the choice of a 212 de Bruijn bit sequence is sufficiently accurate
for the modulation formats with m = (2, 3, 4) and the single-span system scenario
with Dmax = 320 ps/nm observed here. When using de Bruijn symbol sequences,
the order should be at least n S = 5 for m = 2 (211 = 2 · 45 ), n S = 4 for m = 3
(211 < 3 · 84 ), and n S = 3 for m = 4 (214 = 4 · 163 ).
Whereas a simulation of deterministic distortions with data sequences not con-
taining all the ISI-relevant symbol transitions can only lead to an underestimation
of the EOP, the bit error ratio (BER) can be overestimated by the use of too short
de Bruijn bit sequences with too many poorly performing symbol patterns when
simulating noise. This becomes apparent from Fig. 5.1b which shows the back-
to-back BER performance for NRZ-Square 64QAM for different de Bruijn bit se-
quence lengths, obtained by Monte Carlo (MC) simulations (see Sect. 5.3). The 212
de Bruijn bit sequence overestimates the BER because too many bad symbol pat-
terns are contained. A 216 de Bruijn bit sequence leads to more accurate results. An
even better choice would be a 643 de Bruijn symbol sequence. However, this yields
a data sequence length of b = 643 · 6 = 1572864 which can hardly be simulated
with currently available memory sizes.
Figure 5.2 illustrates the back-to-back performance of 16DPSK determined by
MC simulations (a) and semi-analytical BER estimation (b) described in Sect. 5.4,
again for 3 dB receiver filter bandwidths of Bopt = 2.5 · r S and Bel = 0.75 · r S and a
data rate of 40 Gbit/s. For MC simulations, the use of 212 de Bruijn bit sequences and
adequate de Bruijn symbol sequences leads to very similar results. The variations
here are rather caused by MC estimate inaccuracies than by the particular choice of
the data sequences. When lower BER values in the range of 10−9 are determined
with semi-analytical BER estimation, de Bruijn symbol sequences with appropriate
average distributions of the symbol patterns should be used. An OSNR difference of
1 dB can be observed in Fig. 5.2b between the back-to-back performances for NRZ,
caused by a different consideration of the ISI induced by the receiver filters.
1E-2 1E-3
a b 16DPSK NRZ
16DPSK NRZ
De Bruijn
1E-5 symbol 16
3
De Bruijn
1E-3 symbol 16
3
16DPSK RZ
BER
BER
1E-7 De Bruijn
3
16DPSK RZ symbol 16 +
12
de Bruijn bit 2
1E-4
1E-9
De Bruijn
De Bruijn bit 212
symbol 163 De Bruijn bit 212
1E-5 1E-11
18 20 22 24 26 28 22 24 26 28 30 32 34 36
OSNR [dB] OSNR [dB]
Fig. 5.2 16DPSK back-to-back performance determined by MC simulations (a) and semi-
analytical BER estimation (b) for 40 Gbit/s when using 212 de Bruijn bit sequences and 163 de
Bruijn symbol sequences for NRZ (solid lines) and RZ (dashed lines)
5.2 Eye Opening Penalty 131
When looking at multi-span long-haul transmission, the data sequence length re-
quirements can hardly be fulfilled when high magnitudes of accumulated chromatic
dispersion lead to inter-symbol interference across many symbols. This is particu-
larly problematic for systems without optical in-line CD compensation.
5.2 Eye Opening Penalty for Multi-Level Eyes
A signal passing through a transmission system is degraded by stochastic and de-

terministic degradation effects. An easy way to characterize the quality of digital
signals degraded by deterministic distortions is the determination of the eye open-
ing penalty (EOP). When sampling a binary signal once per bit period at a specified
sample time, the eye opening is defined as the difference of the minimum value of
the “one” state and the maximum value of the “zero” state. The eye opening penalty
for binary signals is defined as the ratio of the eye opening of the non-distorted ref-
erence eye, E Or e f , and the eye opening of the distorted eye, E O, and usually given
in decibels
E Or e f
E O P = 10 lg [d B]. (5.2)
EO
For high-order modulation, the electrical signals have multiple states and data
recovery is accomplished by performing several decisions at different thresholds. In
this case, the ratio of the eye openings of the reference eye to the distorted eye is
calculated for all the eye openings of the multi-level eye relevant for decision. The
maximum value defines the eye opening penalty of the multi-level eye

E On,r e f
E O P = max 10 lg [d B], (5.3)
n E On
where n corresponds here to the number of eye openings relevant for decision. For
arg-decision, the EOP can be calculated from the multi-level eyes arising after the
arg-calculation, whereas the EOP must be identified individually for the signals in
the in-phase and quadrature branches for IQ-decision. In the latter case, the larger
value finally determines the overall system performance.
Figure 5.3 schematically shows how the reference eye (left) and the distorted
eye (right) are received in the in-phase and quadrature branches for 8PSK. When
employing IQ-decision, only the two inner eye openings are relevant for the decision
process (as observable from Fig. 3.7b) and the EOP of the multi-level signal in each
branch can be calculated using (5.3) with n = {1, 2}.
Reference eye Distorted eye
EO1
EO1,ref
EO2,ref EO2
Fig. 5.3 Determination of the EOP for multi-level eyes; The reference eye and the distorted eye
(how they are received in the in-phase and quadrature branch) and the relevant eye openings are
schematically shown here for 8PSK and IQ-decision.
5.3 Monte Carlo Simulations
The most significant criterion for performance characterization in digital transmis-

sion systems is the BER, which is simply defined as the ratio between the bits re-
ceived with errors and the number of bits that have been transmitted. An intuitive
approach to the simulative determination of the BER is sending a preferably long
input bit sequence through the investigated system and then simply counting the bit
errors of the output bit sequence and relating them to the total number of transmitted
bits. This method is denoted as Monte Carlo (MC) simulation.
An advantage of the MC method is its general applicability to the simulation of
arbitrary systems and the ability to consider all deterministic and stochastic degrada-
tion effects. On the other hand, the accuracy of the calculated BER estimate strongly
depends on the total number of simulated bits b. A fairly large number of bits must
be simulated in order to obtain accurate estimates. As a rule of thumb, the simu-
lation of b = 10/BER bits is required for a confidence interval of 95% with an
uncertainty factor of two on the error rate scale, corresponding to a few tenths of a
dB on the OSNR scale [6]. Thus, the determination of BER values lower than 10−6
necessitates the consideration of more than ten million bits and turns out to be too
complex even for the strength of modern computers. √ With increasing b, the confi-
dence interval narrows relatively slowly with 1/ b, so a trade-off must be found
between computational effort and statistical accuracy.
Because the computing time increases disproportionally with increasing b (the
computational effort for the fast fourier transform scales with b · lg b), simulations
should be split into multiple runs. A reduced number of bits is then simulated in each
run, whereat the random number seeds of the noise generators must be readapted in
each run for an appropriate simulation of noise.
5.4 Semi-Analytical BER Estimation for DBPSK Direct Detection 133
5.4 Semi-Analytical BER Estimation
A method which allows for the determination of low BER values without very long
run-times is the semi-analytical BER estimation. “Semi-analytical” means that parts
of this method are based on analytical derivations but other parts on numerical cal-
culations. With knowledge of the exact statistics of the photocurrents appearing at
the decision gates it becomes possible to estimate the BER by integration over the
probability density function (PDF), without needing to simulate a very large number
of bits as necessary in MC simulations. Unfortunately, in many cases the determina-
tion of the exact PDF turns out to be difficult. As a first approach, the photocurrent
statistics are often approximated by a Gaussian PDF. However, it has been shown
that this yields inaccurate results for pre-amplified direct detection systems limited
by signal-ASE noise, especially for the detection of DPSK signals [5, 11].
Several closely related methods used to exactly evaluate the bit error probability
of optically pre-amplified direct detection systems have been proposed. They are
mainly directed towards OOK, for instance in [4, 7, 8, 11]. By incorporating the
altered receiver structure with delay line interferometers and balanced detection into
the analysis, the semi-analytical BER estimation is extended to binary and high-
order DPSK formats in [3, 5, 9, 14], as well as to Star 16QAM and different receiver
configurations in [9].
5.4.1 DBPSK Direct Detection
Next, the fundamental theory of semi-analytical BER estimation is illustrated on the

basis of the DBPSK direct detection receiver. The system model of the receiver is
shown in Fig. 5.4.
Ho ( f ) He ( f )
x (t ) x(t ) + n(t ) TS
y# (t ) y (t k )
BPF ϕDLI LPF
DLI BD
Fig. 5.4 System model used for semi-analytical BER estimation for DBPSK direct detection
If it is assumed that only signal fractions within a time interval (tk − T0 , tk ) are
relevant for the decision
√ of each symbol at the sampling instants tk , the complex
envelope x(t) = Ps · a(t) · e jϕ(t) of the electrical field at the receiver input
p
E s (t) = Ps · a(t) · e jϕ(t) · e j(ωs t+ϕs ) = x(t) · e j(ωs t+ϕs ) (5.4)
can be interpreted as periodic with T0 at each sampling instant tk , and can be ex-
panded in a Fourier series
L
X
x(t) ≃ xk, p · e j2π( p−L/2)(t−tk )/T0 = xk · e T ∗ , (5.5)
p=1
with the fourier coefficients defined as

L
1X qT0 − j2qπ( p−L/2)/L
xk, p = x tk − T0 + e , (5.6)
L L
q=1
which are represented by a vector with L rows xk . The elements of the row vector e
in (5.5) are given by
− j 2π( p−L/2) (t−tk )

ep = e T
0 , (5.7)
where “T ” and “*” denote transposition and complex conjugation, respectively.

The total number of samples considered for the Fourier expansion is given by
L = S pS T0r S , with S pS denoting the number of samples per symbol. The duration
T0 chosen should be large enough to take all the relevant inter-symbol interference
caused by filtering into account.
As for the signal, the noise n(t) = n k (t)ek + n ⊥ (t)e⊥ emerging at the optical
pre-amplifier can also be expanded into a fourier series (the expansion of a centered
stochastic process into uncorrelated random variables is known as Karhunen-Loève
expansion in the literature). This way, the noise components, polarized in parallel
and orthogonally with respect to the signal, and denoted here as n k (t) and n ⊥ (t),
respectively, can be expressed as
L
X
n k/⊥ (t) ≃ n p,k/⊥ · e j2π( p−L/2)(t−tk )/T0 = nk/⊥ · e T ∗ , (5.8)
p=1
with the fourier coefficients

L
1X qT0 − j2qπ( p−L/2)/L
n p,k/⊥ = n k/⊥ tk − T0 + e . (5.9)
L L
q=1
As will become clear later, the method described here for semi-analytical BER
estimation implies a restriction to stationary Gaussian noise with zero mean in front
of the optical receiver filter. This is valid, for instance, in single-span systems with-
out noise on the transmission link where the overall noise in front of the optical
filter input is solely confined to the ASE-noise of the optical pre-amplifier and
can be assumed to be additive white Gaussian noise (AWGN) with a two-sided
power spectral density (PSD) per polarization given by (3.25). The variance of the
in-phase and quadrature components of each of the elements of nk/⊥ is given as
σ p2 = σ 2 = (N0 /2) · Bn = (N0 /2) · L/T0 · 1/L = N0 /(2T0 ).
Usually, noise performance of optical receivers is characterized with the OSNR

in decibels, which is defined here by the ratio of the average signal power Ps,avg and
the noise power in both polarizations within the reference bandwidth 1 f

Ps,avg
O S N Rd B = 10 lg . (5.10)
2 · N0 · 1 f

In this context, the average signal power is given as Ps,avg = |x(t)|2 , and 1 f =
1λ · f s2 /c, where c is the speed of light. Normally, the reference bandwidth 1λ is
set to 0.1nm, and the OSNR is specified in front of the optical receiver filter.
After having described the signal and noise in front of the optical receiver filter,
the exact PDF can be derived at the decision gate. The optical filter can be described
with a diagonal matrix H, whose elements result from the low-pass equivalent trans-
fer function Ho ( f )

p − L/2
H p, p = Ho . (5.11)
T0
The transfer characteristic of the DLI with respect to the upper output is incorpo-
rated in the analysis by defining the diagonal matrix
1
H1 = · H · F − I · e jϕ DL I , (5.12)
2
where F is a diagonal matrix with diagonal elements F p, p = e j2π( p−L/2)·TS /T0
and I represents a identity matrix with the same size as F. The matrix H1 accounts
for the overall transfer characteristic of the receiver up to the upper DLI output.
Analogously, the overall transfer function with respect to the lower DLI output can
be described by the diagonal matrix
1
H2 = · H · jF + jI · e jϕ DL I . (5.13)
2
Because the spontaneous noise from the amplifier usually dominates receiver shot
and thermal noises, it is assumed here that the photodiodes are ideal square-law
photo-detectors. In this case, the signal after the balanced detector at the sample
instant tk can be calculated as
2 h 2 i
ỹ(tk ) = xk + nk H1 + |n⊥ H1 |2 − xk + nk H2 + |n⊥ H2 |2
T ∗
= xk + nk H1 H1 T ∗ xk + nk + n⊥ H1 H1 T ∗ n⊥T ∗
h T ∗ i
− xk + nk H2 H2 T ∗ xk + nk + n⊥ H2 H2 T ∗ n⊥T ∗ , (5.14)
where the vectors e and e T ∗ are left out since their elements are equal to one for t =
tk . Furthermore, if the electrical filter with the transfer function He ( f ) is included
by defining the elements of the transfer matrix Q as

q−p
Q p,q = He , (5.15)
T0
the signal arising at the decision gate after electrical filtering and sampling can be
expressed as
T ∗
y(tk ) = xk + nk H1 QH1 T ∗ xk + nk + n⊥ H1 QH1 T ∗ n⊥T ∗
T ∗
− xk + nk H2 QH2 T ∗ xk + nk − n⊥ H2 QH2 T ∗ n⊥T ∗
T ∗
= xk + nk H1 QH1 T ∗ − H2 QH2 T ∗ xk + nk

+n⊥ H1 QH1 T ∗ − H2 QH2 T ∗ n⊥T ∗ . (5.16)
The receiver matrix A = H1 QH1 T ∗ − H2 QH2 T ∗ reflects the special properties of a

DPSK receiver composed of DLIs. For particular DPSK receivers and their different
branches, different DLI phase shifts are appropriate and determine the values of the
matrices H1 , H2 and A. As regards DBPSK, the DLI phase shift is set to ϕ DL I = 0◦ .
The procedure of semi-analytical BER estimation for higher-order DPSK formats
and Star QAM is briefly discussed in Sect. 5.4.2. In the case of 2ASK, the receiver
matrix is simply given by A = HQHT ∗ [11].
With the intention of writing the photocurrent y(tk ) as a sum of weighted squared
random variables, an eigendecomposition of the matrix A can be performed. Due to
the definition of the electrical filter matrix Q, the matrix A is a hermitian matrix,
whose eigenvalues are real. That way, the matrix A can be expressed as
A = U3UT ∗ , (5.17)
where U is a unitary matrix whose columns contain the eigenvectors of the matrix
A, and 3 is a diagonal matrix whose diagonal elements are given by the real-valued
eigenvalues λ p of the matrix A. When furthermore imposing the vectors bk = xk U
and zk/⊥ = nk/⊥ U with their elements bk, p and z p,k/⊥ , respectively, the photocur-
rent given by (5.16) can be rewritten as
T ∗
y(tk ) = xk + nk U3UT ∗ xk + nk + n⊥ U3UT ∗ n⊥T ∗
T ∗
= bk + zk 3 bk + zk + z⊥ 3z⊥T ∗
L
X L
2 X 2
= λ p bk, p + z p,k +
λ p z p,⊥
p=1 p=1
L
X L
2 X 2
= λ p Re bk, p + z p,k + λ p I m bk, p + z p,k
p=1
| {z } p=1 | {z }
non-central χ 2 non-central χ 2
L
X L
2 X 2
+ λ p Re z p,⊥ + λ p I m z p,⊥ . (5.18)
p=1
| {z } p=1 | {z }
central χ 2 central χ 2
The photocurrent described by (5.18) can be interpreted as a weighted sum of

squared Gaussian random variables. Half of the summands (the terms containing
signal and noise) are given by the square of Gaussian random variables with non-
zero mean. The statistics of the sum of n squared Gaussian random variables are
known as a non-central χ 2 -distribution with n degrees of freedom [15]. The re-
maining summands are given by the square of zero mean Gaussian random variables
and exhibit a central χ 2 -distribution. The PDFs of the non-central and central χ 2 -
distribution with n degrees of freedom can be found in [10, 15]. The quested overall
PDF of y(tk ) can be obtained by the convolution of the PDFs of the particular sum-
mands. However, it is easier to determine the moment generating functions (MGF)
of the particular PDFs (the MGF is defined as the conjugate complex Laplace trans-
form of the PDF [10]), multiply the particular MGFs and calculate the PDF of y(tk )
from the overall MGF by the conjugate complex inverse Laplace transform.
The noise in front of the optical receiver filter is assumed here to be AWGN,
so that the elements of nk/⊥ are independent Gaussian random variables with zero
mean and equal variances σ 2 , and the elements of zk/⊥ are Gaussian and of equal
variance σ 2 as well because the matrix U consists of a set of orthonormal eigen-
vectors. In this case, by employing the known MGFs for χ 2 -distributed random
variables [15], the overall MGF of y(tk ) can be derived as [4, 9, 11]
2

sλ p |bk, p |
YL exp 1−2sλ p σ 2
8 yk (s) = (5.19)
(1 − 2sλ p σ 2 )2
p=1
with s = jω. When considering colored Gaussian noise at the receiver input (which
is not further regarded here), the elements of nk/⊥ have different variances σ p2 and
the elements of zk/⊥ have different variances σ̂ p2 . These can be substituted for the
constant σ 2 in (5.19), provided that they are known.
The photocurrent PDF can be obtained from the MGF (5.19) by the conjugate
complex inverse Laplace transform
Z u 0 + j∞
∗ 1
p yk (y) = L−1 8 yk (s) = 8 yk (s)e−sy ds, (5.20)
2π j u 0 − j∞
where u 0 must be in the range −1/ max |2λ− 2 + 2

p σ | < u 0 < 1/ max |2λ p σ | to ensure
−/+
the convergence of the integral [2, 4] and λ p denote the negative / positive eigen-
values of the receiver matrix A. The PDF given by (5.20) reflects the exact statistics
of the photocurrent appearing at the decision gate.
The probability of an error at the sample time tk conditional upon the b-bit data
sequence {an }, with n = (1..b), can be calculated for the DBPSK receiver, when
setting the decision threshold to yth , by
R u th
P {y(tk ) < yth } = R−∞ p yk (y)dy, an = 1
P (ek |{an }) = ∞ . (5.21)
P {y(tk ) > yth } = u th p yk (y)dy, an = 0
The integrals of (5.21) can be approximately solved by the so-called “saddlepoint

approximation”. Details of this integration method can be found in the literature, for
instance in [4, 9]. The saddlepoint approximation for the calculation of the condi-
tional probabilities in (5.21) is given as
±
e9 yk (u 0 )
P {y(tk ) ≷ yth } ≃ ± q , (5.22)
2π 9 y′′k (u ±0 )
′′
where 9 yk (s) = ln 8 yk (s)−ln(s)−s · yth , 9 yk (s) is the second derivative of 9 yk (s),
and u + −
0 > 0 and u 0 < 0 are the saddle points on the real s-axis which must be used
for the “>” or “<” sign, respectively. With (5.21) and (5.22), the BER can be finally
estimated as
b
1X
BER = P (ek |{an }) . (5.23)
b
n=1
5.4.2 Extension to Higher-Order DPSK and Star QAM
A DQPSK direct detection receiver composed of DLIs consists of two independent

branches (the in-phase and quadrature branch), both having the same configuration
as a DBPSK receiver (see Sect. 3.2.1). By setting the phase shifts of the DLIs in
these branches to −45◦ and −135◦ , respectively, the in-phase and quadrature com-
ponents of DQPSK signals can be obtained by performing binary decisions at zero.
The error probability must then be estimated separately in the in-phase and quadra-
ture branches. Individual receiver matrices A result for the two branches due to the
different DLI phase shifts. Because the BER has to be calculated twice—once for
every component—the simulation time enlarges in comparison to the DBPSK re-
ceiver. The overall BER is finally obtained by averaging the BER values calculated
in the in-phase and quadrature branches.
Semi-analytical BER estimation can also be applied to high-order DPSK and
Star QAM formats. As shown in Sect. 3.2.1, arbitrary DPSK signals can be detected
by using a multiple DLI receiver composed of N ph /2 DLIs with N ph /2 individ-
ual phase shifts, where N ph is equal to M for MDPSK formats and represents the
number of phase states. The multiple DLI receiver exhibits nearly the same the-
oretical BER performance as the direct detection IQ receiver with only two DLIs
and arg-decision. The direct application of semi-analytical BER estimation to the
IQ receiver with arg-decision is problematic because a nonlinear operation (arg) is
performed before decision. To calculate the BER of the multiple DLI receiver, an
individual receiver matrix A must be adopted to every DLI branch. N ph /2 differ-
ent receiver matrices have to be evaluated and N ph /2 individual error probabilities
arise. Every error probability is defined as the probability of an erroneous semicircle
decision. The overall BER is finally obtained through an appropriate aggregation of
5.4 Semi-Analytical BER Estimation for Higher-Order DPSK and Star QAM 139
the individual error probabilities [9]. For the 8DPSK receiver, for instance, four er-
ror probabilities are calculated. A symbol error rate can then be derived based on
these four values. When defining the probability of an erroneous decision as P(yk ),
k ∈ {1, 2, 3, 4}, the probability of a symbol error is given by
Psym = P(y1 ) + P(y2 ) + P(y3 ) + P(y4 )

+P(y1 )P(y2 ) + P(y1 )P(y4 ) + P(y2 )P(y3 ) + P(y3 )P(y4 )
+P(y1 )P(y2 )P(y3 ) + P(y1 )P(y3 )P(y4 )
+P(y1 )P(y2 )P(y4 ) + P(y2 )P(y3 )P(y4 )
+P(y1 )P(y2 )P(y3 )P(y4 ) (5.24)
The first line represents the dominant fraction of the symbol error probability for
reasonably small error probabilities, because all other terms contain products of
the error probabilities. Neglecting these product terms (taking only one erroneous
decision per symbol into account) is equivalent to considering only symbol errors
corresponding to false decisions of neighboring symbols. When the bit mapping is
chosen as Gray coded, one symbol error corresponds to one bit error. Therefore, the
BER for the multiple DLI receiver can finally be approximated by
N ph /2
1 X
BER ≈ P(yk ). (5.25)
m
k=1
As discussed in Sect. 3.2.1, a direct detection IQ receiver with just two DLIs,
whose phase shifts in the in-phase and quadrature arms are adjusted to ϕ DL I I = 0◦
and ϕ DL I Q = −90◦ , is sufficient for detecting the in-phase and quadrature compo-
nents of arbitrary high-order DPSK signals. For this receiver, the error probability
must be estimated separately in the in-phase and quadrature branches. Individual
receiver matrices A result for the two branches due to the different DLI phase shifts.
Since the levels of the in-phase and quadrature photocurrents represent the projec-
tion of the symbol states to the in-phase and quadrature axis, multiple decisions on
multi-level electrical signals become necessary for DPSK formats with more than
four phase states such as 8DPSK or 16DPSK, as described in Sect. 3.2.3. Figure 3.7
illustrates in this section how thresholds must be positioned within the in-phase
and quadrature eyes for 8DPSK. It can be observed that just one fixed threshold at
zero is not sufficient, but that instead two thresholds per branch are necessary. For
every threshold, an individual error probability has to be evaluated. Because the ap-
propriate threshold depends on the particular received symbol, the received signal
power and the filter transfer functions, the thresholds at each sample time tk must be
adapted. More precisely, the integration interval must be properly chosen when de-
termining the error probability by integration over the PDF. Finally, the overall BER
is obtained through an appropriate aggregation of the individual error probabilities,
which can be done in a similar way as illustrated for the multiple DLI receiver in
(5.24) and (5.25), where P(yk ) then represents the probability of an erroneous de-
cision at the different thresholds in the in-phase and quadrature arms.
In the case of Star QAM, the error probability corresponding to the bits which
encode the phase information can be determined in the same way as for the DPSK
format with the same number of phase states. When the Star QAM signals have
only two amplitude levels, just one additional bit for which the error probability has
to be calculated from the signal in the intensity detection branch is encoded in the
additional amplitude modulation. The same BER estimation procedure can then be
employed as for 2ASK systems. The receiver matrix A for 2ASK receivers was de-
rived in [11] and simplifies to A = HQHT ∗ . Of course, the threshold in the intensity
detection branch must be optimized due to the dependence of the signal-ASE noise
on the received signal power.
Using the semi-analytical BER estimation described in the last sections, the BER
of direct detection receivers can be estimated very accurately and quickly without a
large number of bits needing to be simulated. The method presented here takes any
deterministic signal distortions and arbitrary pulse shaping into account (this infor-
mation is contained in the received signal samples), as well as the exact structure
and transfer characteristic of the receiver with its filters and interferometers. How-
ever, application possibilities are limited to the appearance of stationary Gaussian
noise with zero mean in front of the receiver—only in this case the photocurrent can
be described as a sum of χ 2 -distributed random variables with known MGF. This
prohibits an exact characterization of multi-span systems taking into consideration
all relevant transmission effects, as for instance the nonlinear phase noise.
The exact noise statistics at the decision gates of homodyne receivers with dif-
ferential and synchronous detection are not derived here. As shown in [13], optical
pre-amplified direct detection receivers are mathematically equivalent with hetero-
dyne receivers with delay demodulation under certain conditions. Moreover, homo-
dyne phase diversity DPSK receivers are shown to have the same performance than
heterodyne DPSK receivers in [12]. Therefore, the semi-analytical BER estimation
for direct detection can probably also be used for BER evaluation in systems with
homodyne differential detection. In digital coherent receivers with homodyne syn-
chronous detection, the noise statistics are altered by equalization and digital phase
estimation. Further research is necessary to enable semi-analytical BER estimation
for those complex receiver structures.
References
1. Agrawal, G.P.: Nonlinear Fiber Optics. Academic Press (3. Edition, 2001)
2. Bronstein, I., et al.: Taschenbuch der Mathematik. Harry Deutsch Verlag, Frankfurt a. M.
(2001)
3. Coelho, L.D.: Numerical optimization of fiber optic communication systems with advanced
modulation formats at 40Gbit/s channel data rate. Master´s thesis, Munic University of Tech-
nology (2005)
References 141
4. Forestieri, E.: Evaluating the error probability in lightwave systems with chromatic disper-
sion, arbitrary pulse shape and pre- and postdetection filtering. IEEE Journal of Lightwave
Technology 18(11), 1493–1503 (2000)
5. Gnauck, A.H., Winzer, P.J.: Optical phase-shift-keyed transmission. IEEE Journal of Light-
wave Technology 23(1), 115–130 (2005)
6. Jeruchim, M.C., et al.: Simulation of Communication Systems - Modeling, Methodology and
Techniques. Kluwer Academic/Plenum Publishers (2000)
7. Kac, M., Siegert, A.J.F.: On the theory of noise in radio receivers with square law detectors.
J. Appl. Phys. 18, 383–397 (1947)
8. Lee, J.S., Shim, C.S.: Bit error rate analysis of optically preamplified receivers using an eigen-
function expansion method in optical frequency domain. IEEE Journal of Lightwave Tech-
nology 12, 1224–1229 (1994)
9. Nölle, M.: Übertragungseigenschaften von höherwertigen optischen DPSK und QAM
Modulationsformaten mit Direktempfang auf Basis eines semi-analytischen Bitfehlerraten-
Schätzverfahrens. Diplomarbeit, Technische Universität Berlin (2007)
10. Papoulis, A.: Probability, Random Variables, and Stochastic Processes. McGraw-Hill, Inc.
(1984)
11. Randel, S.: Analysis of fibre-optic transmission systems with wavelength-division multiplex
at 160 Gb/s data rate per channel. Ph.D. thesis, Technische Universität Berlin (2005)
12. Siuzdak, J., van Etten, W.: BER evaluation for phase and polarization diversity optical homo-
dyne receivers using noncoherent ASK and DPSK demodulation. IEEE Journal of Lightwave
Technology 7(4), 584–599 (1989)
14. Wang, J., Kahn, J.M.: Impact of chromatic and polarization-mode dispersions on DPSK sys-
tems using interferometric demodulation and direct detection. IEEE Journal of Lightwave
Technology 22(2), 362–371 (2004)
15. Whalen, A.D.: Detection of Signals in Noise. Academic Press (1971)
16. Wickham, L.K., et al.: Bit pattern length dependence of intrachannel nonlinearities in pseudo-
linear transmission. IEEE Photonics Technology Letters 16(6), 1591–1593 (2004)
Chapter 6
Fiber Propagation Effects
Abstract During fiber transmission, optical signals are distorted by several linear
and nonlinear degradation effects. This chapter provides a brief description of the
most important fiber degradation effects such as fiber attenuation, chromatic disper-
sion, Kerr nonlinearities, nonlinear scattering effects, polarization mode dispersion
and nonlinear phase noise, laying the foundations for understanding the specific
behavior of high-order modulation formats in relation to various transmission im-
pairments, discussed in detail later on in Chap. 7 and Chap. 8.
In the previous chapters, only the time dependency of the electrical field was of
interest. In the case of fiber propagation, the time and location dependent complex
electrical field must be examined. For propagation in z-direction and when neglect-
ing the initial phase and the phase noise, the electrical field can be described by
E s (z, t) = x(z, t) · e− jβ0 z+ jωs t , (6.1)

√
where x(z, t) = Ps · a(z, t) · e jϕ(z,t) represents the slowly varying time and
location dependent complex envelope of the monochromatic electrical field with
the propagation constant β0 at the angular optical frequency ωs . The evolution of
x(z, t) along an optical fiber can be described by the so-called generalized nonlin-
ear Schroedinger equation [11], which is given by
∂ x(z, t ′ ) α β2 ∂ 2 x(z, t ′ ) β3 ∂ 3 x(z, t ′ )

= − x(z, t ′ ) + j +
∂z | 2 {z } | 2 ∂t ′2 {z 6 ∂t ′3 }
attenuation chromatic dispersion
2
− jγ x(z, t ′ ) x(z, t ′ ), (6.2)
| {z }
Kerr nonlinearities
with the retarded time scale t ′ = t − τgr · z, where τgr · z represents the accumulated
group delay. The generalized nonlinear Schroedinger equation takes into account the
fiber attenuation described by the attenuation coefficient α, the effect of chromatic


144 6 Fiber Propagation Effects
dispersion characterized by β2 and β3 , as well as the fiber nonlinearities caused

by the intensity dependence of the refractive index (Kerr effect), whose strength is
defined by the nonlinear propagation coefficient γ . Equation (6.2) can be solved nu-
merically by applying the split-step Fourier method [1]. With the split-step Fourier
method, the fiber is divided into multiple segments with step size 1z. Knowing
the complex envelope x(z, t ′ ) at location z, it is possible to calculate x(z + 1z, t ′ )
at location z + 1z. The linear and nonlinear parts of (6.2) are treated separately.
Whereas the Kerr nonlinearities are handled in the time domain, attenuation and
chromatic dispersion are considered in the frequency domain employing the Fast
Fourier Transform (FFT). Long computation times can arise for high optical pow-
ers, which necessitate small step sizes.
In the following sections, fiber attenuation, chromatic dispersion and Kerr non-
linearities are more closely examined. Afterwards, additional effects such as non-
linear scattering effects, polarization mode dispersion and nonlinear phase noise are
briefly discussed. Because the latter three effects are not further detailed in Chap. 7
and Chap. 8, their principal impact on the transmission of high-order modulation
formats is already presented here.
6.1 Fiber Attenuation
When neglecting the chromatic dispersion and the Kerr nonlinearities in (6.2), the
complex envelope of the electrical field at the output of an optical fiber of length l
is given by
α
x(z = l, t ′ ) = x(z = 0, t ′ ) · e− 2 ·l , (6.3)
where x(z = 0, t ′ ) denotes the complex electrical field envelope at the fiber input.
For the optical power it holds
P(z = l, t ′ ) = x(z = l, t ′ ) · x(z = l, t ′ )∗ = P(z = 0, t ′ ) · e−α·l . (6.4)
The attenuation coefficient α has the unit 1/km. Commonly, the fiber loss is ex-
pressed in units of dB/km with
P(z = 0, t ′ )
αd B · L = 10 · lg , (6.5)
P(z = l, t ′ )
where the attenuation coefficient in decibels αd B is related to α approximately as
αd B ≈ 4.343 · α · dB.
The two main sources of loss in silica optical fibers are material absorption
and Rayleigh scattering. When observing the wavelength band between 800 nm-
1600 nm used for optical communication systems, the Rayleigh scattering, caused
by the orderless molecular structure of glass and the resulting random fluctuations
6.2 Chromatic Dispersion 145
of the refractive index, is the dominant loss factor in the low-wavelength range,
whereas the molecular infrared absorption limits the utilizable wavelength window
to wavelengths smaller than 1600 nm. Apart from the negligible waveguide loss of
pure silica, further loss factors are the ultraviolet absorption (caused by lifting elec-
trons to higher energy levels) and absorption due to material impurities (for instance
the OH absorption).
In spite of all these effects, a minimum loss of αd B < 0.2 dB/km can be achieved
for silica fibers at λ = 1550 nm when employing appropriate manufacturing pro-
cesses. Although the loss depends on the wavelength of light (the exact wavelength
dependency of αd B can be looked up for instance in [11]), αd B can be assumed to
be constant within the modulation bandwidth of a single channel for current data
rates such as 40 Gbit/s.
The attenuation of fiber links can be compensated for by the use of optical am-
plifiers, for instance EDFAs. Because EDFAs are very broadband, many channels
of a WDM signal can be amplified simultaneously by a single EDFA. To be able to
bridge longer distances, several sections can be cascaded, each consisting of a fiber
link and an optical amplifier (multi-span transmission). Unfortunately, each optical
amplifier adds noise with a power spectral density given by (3.25) and degrades the
OSNR. This noise accumulation is one of the major reasons for the limitation of the
reachable transmission lengths in transparent optical networks.
6.2 Chromatic Dispersion
The frequency dependence of the group delay in optical fibers is denoted as chro-
matic dispersion and caused by the frequency dependence of the refractive index
(material dispersion) and the frequency dependent wave guidance in the fiber core
and fiber cladding (waveguide dispersion). Mathematically, the effect of chromatic
dispersion can be described by expanding the frequency dependent propagation con-
stant β(ω) into a Taylor series at ωs = 2πc/λs
1
β(ω) = ω
c · n(ω) = β0 + β1 (ω − ωs ) + β2 (ω − ωs )2
2
1 3
+ β3 (ω − ωs ) + ... (6.6)
6

In (6.6), β0 is the propagation constant at ω = ωs and β1 = dβ(ω)/dω ω=ωs is
equal to the group delay per length unit, τgr (ω), at ω = ωs . The parameters β2
and β3 qualify the chromatic dispersion as well as the chromatic dispersion slope at
ω = ωs , respectively, and are given by

d 2 β(ω) dτgr (ω)
β2 = = , (6.7)
dω2 ω=ωs dω ω=ωs

d 3 β(ω) d 2 τgr (ω)
β3 = = . (6.8)
dω3 ω=ωs dω2
ω=ωs
Distortion free transmission can only be achieved if τgr (ω) is constant within the
modulation bandwidth, which is the case for βi = 0 for i ≥ 2. Mostly, the dispersion
and the dispersion slope are specified as the first and second derivatives of the group
delay on the wavelength, denoted here as Dλ and Sλ , respectively. By using the
relation dλ = −dω · λ2 /(2πc), Dλ and Sλ can be calculated from β2 and β3 at
λ = λs as

dτgr (λ) 2π · c
Dλ = = − 2 · β2 , (6.9)
dλ λ=λs λs
2
d 2 τgr (λ) 2π · c 4π · c
Sλ = = · β3 + · β2 . (6.10)
dλ2 λ2s λ3s
λ=λs
To examine the impact of the chromatic dispersion on the signal propagation

along the fiber, the fiber input electrical field E s (z = 0, t) can be Fourier trans-
formed and multiplied with the transfer function of the fiber in the frequency domain
to obtain the Fourier transform of the signal at location z
E s (z, jω) = E s (z = 0, jω) · e− jβ(ω)z , (6.11)
where β(ω) is given by (6.6). To further work out the propagation of the complex
envelope x(z, t ′ ) with the retarded time scale t ′ = t − τgr · z, the Fourier transform
of x(z, t ′ ) can be calculated by using the time shift theorem as
Z +∞
j·τgr ·z
x(z, j) = e x(z, t)e− jt dt, (6.12)
−∞
with = ω − ωs . Resolving (6.1) for x(z, t), and then inserting the result in (6.12),
leads to
x(z, j) = e j·τgr ·z · e jβ0 ·z · E s (z, jω). (6.13)
With E s (z, jω) given by (6.11) and E(z = 0, jω) = x(z = 0, j), it is finally
obtained
1 2 1 3
x(z, j) = x(z = 0, j) · e− j 2 β2 z · e− j 6 β3 z , (6.14)
which can be rewritten as differential equation as

∂ x(z, j) 1 1
= x(z, j) · − j β2 2 − j β3 3 . (6.15)
∂z 2 6
6.3 Kerr Nonlinearities 147
Transforming (6.15) back to the time domain with ∂/∂t ′ = j yields the differential
equation
∂ x(z, t ′ ) β2 ∂ 2 x(z, t ′ ) β3 ∂ 3 x(z, t ′ )

= j + . (6.16)
∂z 2 ∂t ′2 6 ∂t ′3
The terms on the right side of (6.16) are identically to the dispersion terms in the
generalized nonlinear Schroedinger equation given by (6.2).
For standard single mode fibers (SSMF), the dispersion coefficient Dλ is typ-
ically zero for λ = 1310 nm and increases up to about 16 ps/(nm·km) at λ =
1550 nm. This means that two spectral components in the λ = 1550 nm range,
which are separated by 1 nm, experience a delay difference of 16 ps per kilometer.
This causes a broadening of the transmitted pulses and leads to inter-symbol inter-
ference.
The possible influencing of the waveguide dispersion can be used to develop
special fibers, for instance the dispersion shifted fiber (DSF) with Dλ = 0 at
λ = 1550 nm or the dispersion compensating fiber (DCF). The latter can be used to
fully compensate for the chromatic dispersion of fiber links for single channel trans-
mission or even for broadband dispersion compensation across the whole WDM
band. In the latter case, an appropriate compensation of the dispersion slope addi-
tionally becomes important.
Whereas the chromatic dispersion can in principle be ideally compensated for
for point-to-point links, an adaptive compensation becomes necessary in flexible
routed optical networks. For high symbol rates exceeding 40 Gbaud, length fluctu-
ations caused by temperature can necessitate an adaptive dispersion compensator
to compensate for the steadily changing amount of accumulated dispersion [7]. As
shown in Sect. 3.5.3, digital coherent receivers provide the opportunity for perform-
ing adaptive dispersion compensation efficiently in the electrical domain.
6.3 Kerr Nonlinearities
The Kerr nonlinearities of an optical fiber imply all the nonlinear effects originating
from the intensity dependence of the refractive index. For a better understanding
of the reason for this intensity dependence, it is useful to look at the electric flux
density D, which is related to the electrical field as
D = ǫ0 E + P, (6.17)
where ǫ0 is the vacuum permittivity and P represents the electric polarization which
reflects the material properties of the medium and can be divided into a linear and
nonlinear part as P = P L + P N L .
The linear part of the electric polarization is given by P L = ǫ0 χ1 E, where χ1
denotes the linear susceptibility. Without consideration of the nonlinear electrical
polarization it holds D = ǫ0 ǫr E, with the relative permittivity of the medium de-
fined as ǫr = 1 + χ1 . When considering only third-order nonlinear effects which
is generally sufficient for non-doped silica fibers [1, 2, 8], the nonlinear electric
polarization is given by
P N L = ǫ0 χ3 E3 , (6.18)
The parameter χ3 represents the third-order nonlinear susceptibility. An optical

wave with an electrical field described by (6.1), propagating in an optical fiber with
an effective core area Ae f f [1], exhibits a field intensity within the fiber core of
|x̃(z, t)|2 = |x(z, t)|2 /Ae f f (in units of W/m2 ). The electrical field given by (6.1)
p p
normalized to Ae f f is denoted here as Ẽ s (z, t) = E s (z, t)/ Ae f f and has the unit
√
W /m. It can be written as a real-valued signal by adding the conjugate complex
(c.c.) component (in nonlinear optics, one must calculate with real-valued signals
because Re[z 1 · z 2 ] 6= Re[z 1 ] · Re[z 2 ], with z 1 and z 2 being complex numbers).
Using these definitions, the nonlinear polarization can be calculated by
3
1 1
PN L (z, t) = ǫ0 χ3 · Ẽ s (z, t) + Ẽ s∗ (z, t)
2 2
1 h 3 i
= ǫ0 χ3 · · x̃ (z, t) · e− j3β0 z+ j3ωs t + c.c.
8
3 h i
+ ǫ0 χ3 · · |x̃(z, t)|2 · x̃(z, t) · e− jβ0 z+ jωs t + c.c. . (6.19)
8
When neglecting the component at 3 · ωs and returning to complex notation, the
electric flux density under consideration of the linear and nonlinear electrical polar-
ization becomes
3
D(z, t) = ǫ0 Ẽ s (z, t) + ǫ0 χ1 Ẽ s (z, t) + ǫ0 χ3 · |x̃(z, t)|2 · Ẽ s (z, t)
4
3
= ǫ0 · 1 + χ1 + χ3 · |x̃(z, t)|2 · Ẽ s (z, t), (6.20)
4
where the term in brackets can be interpreted as an effective relative permittivity

ǫr,e f f (x̃). The refractive index is then given by
q r
3
n(x̃) = ǫr,e f f (x̃) = 1 + χ1 + χ3 · |x̃(z, t)|2 = n 0 + n 2 · |x̃(z, t)|2 .(6.21)
4
It can be observed that√the refractive index becomes intensity-dependent for χ3 6= 0.
The parameter n 0 = 1 + χ1 denotes the linear refractive index and n 2 is the so-
called “nonlinear index coefficient” in units of m2 /W. It can be shown that it is true
that n 2 = 3χ3 /8n 0 by expanding the square root in (6.21) in a power series.
In the same way as for the refractive index, the propagation constant becomes
dependent on the intensity of the light. For the final derivation of the term describing
the Kerr nonlinearities in the generalized nonlinear Schroedinger equation (6.2) it is
referred to in [1].
6.3 Kerr Nonlinearities 149
The relation between the nonlinear propagation coefficient γ in W−1 m−1 used
in (6.2) and the nonlinear index coefficient n 2 is derived in [1] and given by
n 2 · ωs
γ = . (6.22)
c · Ae f f
Having understood the origin of the Kerr nonlinearities, let us now look at their
effect on signal propagation. When neglecting attenuation and chromatic dispersion,
the simple solution of (6.2) is
′ 2
x(z, t ′ ) = x(z = 0, t ′ ) · e− jγ ·|x(z=0,t )| z . (6.23)
Obviously, the intensity dependence of the refractive index leads to a phase modu-
lation of the information signal with the intensity-dependent nonlinear phase shift
2
ϕ N L (z, x, t ′ ) = −γ · x(z = 0, t ′ ) z. The phase modulation causes a broadening
of the signal spectrum but does not directly influence the amplitude of the complex
envelope. Only the interaction with the chromatic dispersion induces a conversion
of the phase modulation into an amplitude modulation and thus an indirect impact
2
of the Kerr effect on the signal power x(z, t ′ ) detected in IM-DD systems. In the
case of phase and quadrature amplitude modulation, however, the nonlinear phase
shift has a direct influence on the signal information as illustrated in more detail
later on in Sect. 7.1.6 and Sect. 7.3.5.
When neglecting chromatic dispersion but including fiber attenuation and ob-
serving a multi-channel transmission system composed of two channels with the
electrical fields
E 1 (z, t ′ ) = x1 (z, t ′ ) · e− jβ0,ω1 z+ jω1 t , E 2 (z, t ′ ) = x2 (z, t ′ ) · e− jβ0,ω2 z+ jω2 t , (6.24)
the totally induced nonlinear phase shifts of the complex field envelopes of the two
channels at ω1 and ω2 are given by [1, 2]
1 − e−αz h 2 2 i
ϕ N L (z, x1 , x2 , t ′ )ω = −γω1 x1 (z = 0, t ′ ) + 2 x2 (z = 0, t ′ ) , (6.25)
1 α
1 − e−αz h 2 2 i
ϕ N L (z, x1 , x2 , t ′ )ω = −γω2 x2 (z = 0, t ′ ) + 2 x1 (z = 0, t ′ ) , (6.26)
2 α
with ω1 and ω2 representing the angular frequencies, γω1 and γω2 the nonlinear
propagation coefficients, and x1 (z = 0, t ′ ) and x2 (z = 0, t ′ ) the complex en-
velopes at the fiber input of the two channels, respectively, and assuming same
attenuation coefficients
α for both channels. In (6.25) and (6.26), the quantity
le f f (z) = 1 − e−α·z /α, which was introduced in Sect. 5.1, can be interpreted as
an effective length and indicates that an attenuation free fiber of length le f f would
induce the same nonlinear phase shift.
Two important propagation effects can be explained from (6.25) and (6.26).
Firstly, the propagating channels are phase modulated as a result of the intensity
change of their own channel which is denoted as self phase modulation (SPM).
Secondly, the phases of the channels are shifted due to the intensity fluctuations of
the neighboring channels. The latter effect is referred to as cross phase modulation
(XPM) and is twice as effective as SPM. However, the XPM efficiency is reduced
for non-parallel polarizations of the channels and due to chromatic dispersion be-
cause the channels propagate with different group velocities and XPM occurs only
in the time intervals in which the propagating pulses are superimposed.
A third effect, which can not be deduced from (6.25) and (6.26), is four wave
mixing (FWM). It describes the generation of new waves at other frequencies by
raising the total electrical field of a WDM signal to the third power, according to
(6.18). When considering the interaction of three waves with the frequencies f i ,
f j and f k to compose a fourth waveat f i jk = f i + f j − f k , a total number
of N F W M = 1/2 · N W 3 2
D M − N W D M new mixing products is generated, where
N W D M corresponds to the number of WDM channels. In standard WDM systems
with equidistantly arranged channel spacings, the mixing products fall directly into
the spectral bands of the WDM channels. As for XPM, the efficiency of FWM is
decreased due to chromatic dispersion. Nevertheless, FWM can be a limiting effect
when using fibers with low dispersion and for narrow channel spacings.
In single-channel systems, the SPM is the only Kerr nonlinearity which must be
considered. The SPM is often further partitioned into different intra-channel effects,
depending on whether the effects are caused by the own pulse or the neighboring
pulses. These effects are then denoted as intra-channel XPM (I-XPM) and intra-
channel FWM (I-FWM). The exact impact of SPM on the signal quality has to
be determined by examining the SPM as a combined effect with attenuation and
chromatic dispersion. Whereas chromatic dispersion can be compensated for very
efficiently, SPM-induced signal distortions are one of the main limiting factors for
single-channel multi-span fiber-optic transmission systems for high fiber input pow-
ers, just as the amplifier noise limits the transmission reach attainable with low fiber
input powers. As shown in Chap. 8, an optimal fiber input power can be found as a
trade-off of noise and SPM effect. The robustness against SPM of various high-order
modulation formats is discussed in Sect. 7.1.5, Sect. 7.2.3 and Sect. 7.3.4.
6.4 Other Propagation Effects
The next subsections show how some other effects, namely nonlinear scattering
effects, polarization mode dispersion and nonlinear phase noise, influence the sig-
nal propagation in optical fibers for high-order modulation formats. This is briefly
discussed here rather than later because these effects are not part of the system per-
formance characterization in Chap. 7 and Chap. 8.
6.4 Polarization Mode Dispersion 151
6.4.1 Nonlinear Scattering Effects
Nonlinear scattering effects such as stimulated Brillouin scattering (SBS) and stim-
ulated Raman scattering (SRS), which manifest themselves as an intensity depen-
dent attenuation of the signal wave, are not considered in the generalized nonlinear
Schroedinger equation given by (6.2). SBS induces a reflected optical wave, whose
frequency is reduced by about 10 GHz with respect to the frequency of the signal
wave, and becomes relevant if the optical power exceeds some milliwatts within
a bandwidth of 100 MHz [8]. Therefore, modulation formats with a more constant
power spectral density allow for higher fiber input powers—without being espe-
cially impaired by SBS—than modulation formats with unsuppressed carrier, espe-
cially when low-linewidth lasers [2] are used.
SRS takes effect only for high optical powers greater than 500 mW [8] and causes
an energy transfer to other frequencies. The main part of the scattered power is con-
tained in the peak of the Raman gain spectrum which is located at an about 13.2 THz
lower frequency. For this reason, SRS-induced crosstalk must be considered partic-
ularly in WDM systems with a high number of channels.
6.4.2 Polarization Mode Dispersion
A fiber mode can be interpreted as a solution for the wave equations which satisfies
the boundary conditions at theq
core-cladding interface [9]. For proper dimensioning
of the fiber with (2π/λs ) · r · n 2cor e − n 2clad < 2.405, where r represents the core
radius and n cor e and n clad are the refractive indices of the core and the cladding,
respectively, only a single mode is supported, denoted as the fundamental mode. A
fiber in which only the fundamental mode can propagate is termed as a single mode
fiber. However, this notation is basically imprecise, since two linearly independent
solutions for the wave equation, which would have the same propagation constant in
an ideal perfectly circularly symmetric fiber, correspond to the fundamental mode.
When assuming the fundamental mode as a transverse field, the two solutions are
orthogonally polarized in the x-y-plane (for propagation in the z-direction) and de-
noted as the principle states of polarization (PSP).
Because fibers are not perfectly circularly symmetric in practice (this property
is denominated as “birefringence”), both PSPs have slightly different propagation
constants which leads to signal distortions when the energy of the propagating signal
is split between the two PSPs. This effect is denoted as polarization mode dispersion
(PMD). The statistic for the group delay difference of both PSPs, caused by random
changes in birefringence occurring along the fiber, follows a Maxwell distribution.
The mean value of the group delay difference 1τ P M D for a fiber of length l with
the PMD-parameter P M D is given by [1, 10, 11]
√
h1τ P M D i = P M D · l. (6.27)
√
For good fibers, the PMD-parameter
√ has values of about P M D = 0.1 ps/ km.
Since h1τ P M D i scales with l, the PMD effect is very weak for those fibers for
current data rates such as 40 Gbits/s, even for large distances. In commercially de-
ployed fiber networks, however, particular fibers can show high birefringence, so
that the PMD can be a limiting propagation effect, especially for data rates equal to
or higher than 40 Gbit/s.
The application of high-order modulation formats offers a way to relax the re-
quirements on the fiber birefringence, because a certain group delay difference has
a smaller impact on neighboring pulses for reduced symbol rates implicating longer
pulse durations.
6.4.3 Nonlinear Phase Noise
Another effect with influence on the transmission behavior of optical high-order

modulation signals for multi-span long-haul transmission is the nonlinear phase
noise. When looking at the complex plane, the complex noise of the optical am-
plifiers n k/⊥ (t) leads to amplitude and phase fluctuations of the transmitted signals.
These phase fluctuations are induced directly and can be referred to as “linear phase
noise”. However, the amplitude fluctuations also have influence on the optical phase,
since they are converted to phase fluctuations during transmission through the Kerr
effect. This additional phase noise is denoted as “nonlinear phase noise” and can be
induced by SPM or XPM depending on whether it is caused by the amplitude noise
of the own channel or other channels.
The corruption of the optical phase due to the nonlinear phase noise was inves-
tigated in [4] for the first time. In [6] the variance of the nonlinear phase shift for
multi-span transmission at high SNR is derived. It is given by
2 · hϕ N L i2
σϕ2N L ≈ , (6.28)
3 · SN R
with the mean nonlinear phase shift
hϕ N L i ≈ −γ · |x(z = 0, t)|2 · N F S · le f f , (6.29)
where N F S corresponds to the number of fiber spans, having an effective length of

le f f each, and |x(z = 0, t)|2 represents the power launched into the fiber in each
span. Thereby, the effect of chromatic dispersion is neglected. The SNR in (6.28) is
defined over a bandwidth matched to the signal’s symbol rate r S and related to the
OSNR of (5.10) by
1f E s,avg Ps,avg
SN R = 2 · OSN R · = = , (6.30)
rS No 2 · NF S · σ 2
6.4 Nonlinear Phase Noise 153
where E s,avg is the average energy per symbol and σ 2 represents the variance of
the linear amplifier noise n k/⊥ (t) per fiber span per dimension. Furthermore, it has
been shown in [6] that the variance of the nonlinear phase noise can significantly be
reduced by using a compensator which rotates the received phase proportionally to
the received optical power scaled with the factor α N L . This factor is optimized for
α N L ≈ −γ · le f f (N F S + 1) /2. This way, the variance of the nonlinear phase shift
can be reduced to
hϕ N L i2
σϕ2N L ,Comp ≈ , (6.31)
6 · SN R
which permits the doubling of the transmission distance in systems limited by the
nonlinear phase noise and makes it less likely that the nonlinear phase noise be-
comes the dominant transmission impairment. As illustrated in [5, 12], compensa-
tion of the nonlinear phase noise can be performed using a simple optical phase
modulator in front of the receiver or by electronic means.
For an initial estimation of the impact of the nonlinear phase noise on optical
systems with high-order phase modulation, the ratio between the variance of the
nonlinear phase noise σϕ2N L and the variance of the phase error due to the linear
noise σϕ2L can be examined. As derived in [5], σϕ2L is given for high SNR by
1
σϕ2L ≈ . (6.32)
2 · SN R
The ratio between σϕ2N L and σϕ2L can be easily calculated by using (6.28), (6.29) and
(6.32). It can be interpreted as a measure for whether the nonlinear phase noise gets
dominant over the linear noise. This happens for
σϕ2N L 4 4
= · hϕ N L i2 = · γ 2 · |x(z = 0, t)|4 · N F2 S · le2f f > 1. (6.33)
σϕ2L 3 3
As can be observed from (6.33), the nonlinear phase noise becomes more dominant
for higher fiber input powers and longer transmission lengths. So far, the optimum
fiber input powers for high-order modulation formats associated with multi-span
transmission are largely unknown. Since the attainable transmission lengths can be
expected to be lower for high-order modulation formats, it can be presumed that the
nonlinear phase noise may become a less dominant effect.
In the case of quadrature amplitude modulation, similar conclusions concerning
the nonlinear phase noise are valid as for high-order phase modulation. However,
an additional problem occurs here due to SPM. Symbols with different power levels
experience different mean nonlinear phase shifts, so that QAM signal constellations
are additionally distorted. Therefore, a compensation of the mean nonlinear phase
shift becomes essential to avoid a severe performance degradation due to SPM. This
effect, as well as its compensation, is discussed in Sect. 7.1.6 and Sect. 7.3.5. More-
over, (6.28) and (6.29) show that the nonlinear phase shift variance is smaller for
symbols with less power, so that the nonlinear phase noise is more critical for sym-
bols on outer intensity rings.
An exact investigation of the impact of the nonlinear phase noise on optical sys-
tems with high-order modulation is beyond the scope of this book, but an important
issue for future research. Recent work [3] tries to integrate the nonlinear phase noise
into tools for semi-analytical BER estimation. This should help us to learn more
about the exact influence and the critical nature of this effect.
References
1. Agrawal, G.P.: Nonlinear Fiber Optics. Academic Press (3. Edition, 2001)
2. Coelho, L.D.: Numerical optimization of fiber optic communication systems with advanced
modulation formats at 40Gbit/s channel data rate. Master´s thesis, Munic University of Tech-
nology (2005)
3. Coelho, L.D., et al.: Numerical and experimental investigation of the effect of dispersion on
nonlinear phase noise in RZ-DPSK systems. In: Proceedings of European Conference on
Optical Communication (ECOC), vol. 5, pp. 195–196 (2007)
4. Gordon, J.P., Mollenauer, L.F.: Phase noise in photonic communication systems using linear
amplifiers. Optics Letters 15(23), 1351–1353 (1990)
6. Ho, K.P., Kahn, J.M.: Electronic compensation technique to mitigate nonlinear phase noise.
7. Kato, T., et al.: Temperature dependence of chromatic dispersion in various types of optical
fiber. Optics Letters 25(16), 1156–1158 (2000)
9. Ramaswami, R., Sivarajan, K.: Optical Networks: A Practical Perspective. Morgan Kaufmann
(2. Edition, 2001)
10. Rohde, M., et al.: Robustness of DPSK direct detection transmission format in standard fiber
WDM systems. Electronics Letters 36, 1483–1484 (1999)
12. Xu, C., Liu, X.: Postnonlinearity compensation with data-driven phase modulators in phase-
shift keying transmission. Optics Letters 27(18), 1619–1621 (2002)
Chapter 7
Back-to-Back and Single-Span Transmission
Abstract This chapter discusses the system performance of high-order modulation

formats for back-to back and single-span transmission. The individual strengths and
drawbacks of phase and quadrature amplitude modulation formats in relation to im-
portant system parameters such as noise performance, optimal receiver filter band-
widths, laser linewidth requirements, chromatic dispersion tolerances and self phase
modulation tolerances are identified—for systems with direct detection, homodyne
differential detection and homodyne synchronous detection. Moreover, the influ-
ence of the signal characteristics of different transmitter configurations on system
performance is illustrated for NRZ and RZ pulse shapes. Clear performance tenden-
cies are ascertained and help the reader understand the consequences resulting from
migration to high-order modulation formats, as well as the impact of the choice of
particular system configurations.
Each modulation format shows a specific performance in relation to system impair-

ments due to its individual properties. Euclidean distances and signal characteristics
already determine the particular behavior for back-to-back transmission. When op-
tical high-order modulation signals are propagated along the fiber, their individual
characteristics—for instance the signal shape in the time domain, the shape of the
spectrum and the spectral width—result in a particular robustness or vulnerability
in relation to selected fiber degradation effects.
In this chapter, the system performance of high-order modulation formats for
the back-to-back case and for single-channel single-span transmission without opti-
cal amplifier noise on the link is detailed. This allows for tendencies concerning the
influence of single parameters on the system performance to be highlighted. Consid-
ered parameters are noise performance, optimum optical and electrical receiver filter
bandwidths, laser linewidth requirements, chromatic dispersion tolerances and self
phase modulation tolerances. This isolated examination of single parameters gives
a first indication of the transmission behavior for multi-span transmission over long
distances which is discussed later on in Chap. 8.


156 7 Back-to-Back and Single-Span Transmission
The illustration of system performance provided in this chapter is based on re-

sults from computer simulations at a fixed data rate of 40 Gbit/s (the symbol rates
are then reduced to 40/m Gbaud). In many cases, the results can be scaled also to
other data rates. A wide range of modulation formats is considered. The discussion
of systems with direct detection and homodyne differential detection involves bi-
nary amplitude modulation (2ASK), various phase modulation formats up to 16-ary
phase modulation and the Star 16QAM. With homodyne synchronous detection, ar-
bitrary modulation formats can be conveniently received, and the Square 16QAM as
well as the Square 64QAM are additionally included. Throughout the whole chap-
ter, NRZ and RZ pulse shapes are regarded, where a duty cycle of 50% is assumed
for the RZ pulses. A rise time of 1/4 of the symbol duration is fixed for the electrical
driving signals of all transmitters examined here.
As quality measures for system performance, the eye opening penalty (defined in
Sect. 5.2), the BER—determined by Monte Carlo simulations and semi-analytical
BER estimation as described in Sect. 5.3 and Sect. 5.4—and the receiver sensi-
tivity penalty / OSNR penalty are used. This enables a system characterization at
BER=10−9 for direct detection via semi-analytical BER estimation, whereas a BER
reference of 10−4 is used for systems with homodyne detection which are explored
by Monte Carlo simulations. For BER determination, the decision thresholds are
assumed to be optimally centered between the signal states but not to be optimized
for non-uniformly distributed distortions of different symbols which may appear.
Solely for ASK and the intensity detection branch for Star 16QAM, the threshold
level is optimized with respect to the optimum BER for direct detection to take the
power dependence of the signal-ASE noise into account. The sample time is al-
ways assumed to be located in the middle of the symbols. The window length T0 for
semi-analytical BER estimation is properly chosen according to the system mem-
ory length. The de Bruijn bit sequences and de Bruijn symbol sequences used as
data sequences are appropriate to take the inter-symbol interference caused by the
receiver filters and the examined transmission effects into account.
7.1 Systems with Direct Detection
In this section, the system properties of high-order modulation formats for direct
detection are addressed. Firstly, back-to-back OSNR requirements are shown for
3 dB filter bandwidths of Bopt = 2.5 · r S (optical receiver filter) and Bel = 0.75 · r S
(electrical receiver filter) in Sect. 7.1.1. Afterwards, optimal filter bandwidths are
illustrated for all examined modulation formats in Sect. 7.1.2. The laser linewidth
requirements for interferometric direct detection are considered as an important as-
pect in Sect. 7.1.3. Afterwards, the robustness and sensitivity of the different formats
against fiber transmission impairments is detailed by discussing chromatic disper-
sion tolerances and SPM tolerances in Sect. 7.1.4 and Sect. 7.1.5, respectively. Since
QAM formats exhibit a very poor SPM tolerance, a compensation of the mean non-
linear phase shift can become necessary. Two simple compensation schemes and
7.1 Back-to-Back OSNR Requirements for Direct Detection 157
their performance gains are illustrated in Sect. 7.1.6. Finally, the system charac-
teristics of high-order modulation formats for direct detection are summarized in
Sect. 7.1.7.
7.1.1 OSNR Requirements
The back-to-back optical signal-to-noise ratio (OSNR) requirements can be deter-

mined using the system setup depicted in Fig. 7.1.
DD receiver with optical pre-amplification
TX BPF RX
PRX,avg OSNR (noise bandwidth 0.1nm)
Fig. 7.1 Setup for determination of the back-to-back OSNR requirements
The OSNR is specified here between the optical pre-amplifier and the optical
receiver filter and given by the average signal power Ps,avg divided by the noise
power in both polarizations Pn = 2· N0 ·1 f = 2· N0 ·1λ· f s2 /c as defined in (5.10).
The noise bandwidth is set here to 1λ = 0.1 nm. According to (3.25), the two-sided
power spectral density of the optical pre-amplifier noise per polarization is given by
N0 = n sp (G − 1) · h · f s ≈ F/2 · (G − 1) · h · f s , where the approximation is true
for high amplifier gain G [10] and F represents the noise figure of the optical pre-
amplifier. The amplifier noise (Signal-ASE noise and ASE-ASE noise) is assumed to
be the dominant noise source, so the shot-noise and the thermal noise of the receiver
are neglected. The amplifier gain is given by G = Ps,avg /PR X,avg , where PR X,avg
represents the average optical power received in front of the optical pre-amplifier.
The OSNR behind the optical pre-amplifier can be calculated as
Ps,avg PR X,avg
OSN R = ≈ , (7.1)
Pn F · h · fs · 1 f
where the approximation in (7.1) is valid for high gain. By taking the logarithm of
both sides in (7.1), the OSNR in decibels can be related to the received power for
1λ = 0.1 nm and f s = 193.1 THz by O S N Rd B = 58 dBm + PR X,avg,d Bm − Fd B ,
with PR X,avg,d Bm = 10 lg(PR X,avg /1 mW) and Fd B = 10 lg F. This way, the re-
ceiver sensitivities can easily be calculated from the OSNR requirements presented
in the following paragraphs.
Figure 7.2 shows the OSNR requirements of all examined modulation formats for
NRZ (a) and RZ (b) at 40 Gbit/s when the serial transmitter structure (transmitters
shown in Fig. 2.7 and Fig. 2.12) and direct detection receivers (depicted in Fig. 3.4
and Fig. 3.5) with arg-decision (Fig. 3.6a) are employed. As receiver filters, second-
order Gaussian optical bandpass filters and fifth-order electrical Bessel filters are
adopted, with 3 dB bandwidths of Bopt = 2.5 · r S and Bel = 0.75 · r S , respectively.
1E-2
a
1E-3
1E-4 16DPSK
1E-5
BER
2ASK
Star 16QAM
(ASK-8DPSK)
1E-6 DBPSK RR 1.6
1E-7 DQPSK
1E-8 8DPSK
NRZ
1E-9
10 12 14 16 18 20 22 24 26 28
OSNR [dB]
1E-2
b
1E-3
1E-4
2ASK
1E-5 16DPSK
BER
1E-6 DBPSK
Star 16QAM
1E-7 (ASK-8DPSK)
DQPSK
RR 1.6
1E-8 8DPSK
RZ
1E-9
10 12 14 16 18 20 22 24 26 28
OSNR [dB]
Fig. 7.2 Back-to-back OSNR requirements of direct detection receivers for NRZ (a) and RZ (b) at
r B = 40 Gbit/s for receiver filter 3 dB bandwidths of Bopt = 2.5 · r S and Bel = 0.75 · r S when the
serial transmitter structure and arg-decision are employed, lines: semi-analytical BER estimation,
symbols: results from MC simulations, RR: ring ratio
7.1 Back-to-Back OSNR Requirements for Direct Detection 159
The lines were calculated by semi-analytical BER estimation, whereas the sym-
bols represent results of MC simulations. Both are in excellent agreement. As re-
gards performances, RZ outperforms NRZ for all formats because the distances be-
tween the symbols are greater for the same average power (higher peak-to-average
power ratio). Generally, the OSNR requirements are roughly determined by the Eu-
clidean distances. Therefore, phase modulation formats with an increasing number
of bits per symbol exhibit increasing OSNR requirements and Star QAM formats
feature a better noise performance than DPSK formats with the same number of
symbols. The best OSNR performance of all the formats examined here features
the DBPSK. When comparing the DPSK formats with 2ASK, DQPSK requires a
lower OSNR than 2ASK, but 8DPSK is significantly worse. The comparison of the
two 16-ary formats Star 16QAM (ASK-8DPSK) and 16DPSK shows that the Star
16QAM requires a significantly lower OSNR than the 16DPSK and its performance
is almost as good as for 8DPSK in a comparison at the same data rate. However, the
very poor performance of NRZ-16DPSK can be somewhat improved when employ-
ing a broader optical receiver filter as shown later on in Sect. 7.1.2.
Optimum Ring Ratio for Star 16QAM
The curves for Star 16QAM in Fig. 7.2 are illustrated for the optimum ring ratio,
which defines the ratio of the amplitudes of the outer and inner circle. The optimum
ring ratio arises if same error rate performances are obtained in the intensity and
phase detection branches—yielding a minimum overall BER. Figure 7.3a depicts
the BER for different ring ratios for a fixed OSNR (20.46dB for NRZ, 18.46dB
for RZ) when the serial transmitter structure and arg-decision are used. The BER
is minimized for a ring ratio of about 1.6. When the parallel transmitter and IQ-
decision are employed, the optimum ring ratios are 1.65 for NRZ and 1.6 for RZ as
can be observed from Fig. 7.3b.
Serial transmitter, arg-decision Parallel transmitter, IQ-decision

1E-2 1E-2
a b
1E-3 Star 16QAM NRZ 1E-3 Star 16QAM RZ
@ OSNR = 20.46dB @ OSNR = 18.46dB
BER
BER
1E-4 1E-4
1E-5 1E-5
Star 16QAM RZ Star 16QAM NRZ
@ OSNR = 18.46dB @ OSNR = 20.46dB
1E-6 1E-6
1,2 1,4 1,6 1,8 2,0 1,2 1,4 1,6 1,8 2,0
Ring Ratio Ring Ratio
Fig. 7.3 Ring ratio optimization for Star 16QAM when employing the serial transmitter and arg-
decision (a) / the parallel transmitter and IQ-decision (b)
Influence of the Transmitter Configuration and Decision Technique
In Sect. 2.4.4 and Sect. 2.5, the output signals of the serial and parallel transmitters
for DPSK and Star QAM were shown to exhibit different properties with respect to
their signal shapes, symbol transitions and chirp characteristics. Furthermore, two
decision techniques (arg-decision and IQ-decision) were described in Sect. 3.2.3
and expected to show a different noise performance due to the different positioning
of the thresholds. To illustrate the influence of the transmitter structure and the de-
cision method on the back-to-back OSNR requirements, a performance comparison
is shown in Fig. 7.4, exemplarily for 8DPSK.
TX / RX comparison 8DPSK
1E-2
Se. / IQ NRZ
1E-3
Se. / arg NRZ
1E-4
Pa. / IQ NRZ
1E-5
BER
Pa. / arg NRZ

1E-6
Both TX,
Fig. 7.4 Impact of the trans- 1E-7 arg-dec., RZ
mitter configuration and de- 1E-8 Both TX,

IQ-dec., RZ
cision method on the OSNR 1E-9
performance of 8DPSK for 16 17 18 19 20 21 22 23 24 25
direct detection. OSNR [dB]
Looking at RZ pulse shape, it makes no difference whether the serial or the par-
allel transmitter is used because power is decreased during the symbol transitions
and the signal shapes are nearly identical. For NRZ, however, the use of the serial
transmitter leads to a worse OSNR performance. This is caused by the smaller peak-
to-average power ratio and the higher chirp which results in larger distortions at the
optical receiver filter. When comparing the decision techniques, arg-decision shows
a better performance than IQ-decision, since the thresholds are optimally placed be-
tween the symbols for arg-decision, whereas the distances between the thresholds
and the symbols are smaller for IQ-decision. The tendencies shown here for 8DPSK
are also valid for other formats.
7.1.2 Optimal Receiver Filter Bandwidths
In Sect. 7.1.1, the OSNR requirements were illustrated for fixed 3 dB receiver fil-
ter bandwidths of Bopt = 2.5 · r S and Bel = 0.75 · r S , conjointly defined for all
formats. In this section, the optimal receiver filter bandwidths are discussed. These
were calculated using semi-analytical BER estimation and are shown in the contour
diagrams of Fig. 7.5—all together in a single figure to simplify the comparison of
the different formats and configurations, and to recognize the tendencies.
7.1 Optimal Receiver Filter Bandwidths for Direct Detection 161
ASK NRZ DBPSK Se. TX, NRZ ASK RZ DBPSK Se. TX, RZ
3.5 3.5 3.5 3.5
1 2 0.4 0.6
2.5 3 0.8
3 2 3 3 0.2 3
0.4
1 1.5 1.5 0.2
2 2.5
2.5 2.5 2.5 2.5

Bopt / rS
18.4 0.2
0.5
1.5 15.6
1 1
2 2 2 2
1
0.5 0.5 0.2 0.2
0.5 0.4 0.6 0.4
1.5 1.5 1.5 1.5 0.6
19.0 0.5
16.5 0.8 0.8
1 0.5 1
1 1 1 1
0.6 0.7 0.8 0.9 1 1.1 0.6 0.8 1 1.2 1.4 1.6 1.8 0.6 0.7 0.8 0.9 1 1.1 0.6 0.8 1 1.2 1.4 1.6 1.8
DQPSK Se. TX, NRZ DQPSK Pa. TX, NRZ DQPSK Se. TX, RZ DQPSK Pa. TX, RZ
3.5 3.5 3.5 3.5
2.5 3 3.5 0.8 0.8
0.6
3 0.2
3 1.5 2 3 3 0.2 0.4 0.6 3 0.4
0.5 1
1 1.5 2 2.5 0.2
2.5 2.5 2.5 2.5
Bopt / rS
19.1 0.2 17.2

0.5 0.2
17.1
2 2 1.5 2 2
1 0.2
0.5 1 0.4 0.2
1.5 0.6
2.5 2 0.8 0.4 0.6
3 1
1.5 1.5 1.5 1.5 0.8
17.4 0.5
5
0.5
1
1 1 1 1
0.6 0.8 1 1.2 1.4 1.6 1.8 0.6 0.8 1 1.2 1.4 1.6 1.8 0.6 0.8 1 1.2 1.4 1.6 1.8 0.6 0.8 1 1.2 1.4 1.6 1.8
8DPSK Se. / arg, NRZ 8DPSK Pa. / arg, NRZ 8DPSK Se. / arg, RZ 8DPSK Pa. / arg, RZ
3.2 3.2 3.2 3.2
3.5 0.6
4 0.6
23.6 2.5
2.8 0.5 1 3 2.8 1.5 2.8 0.2 0.4 2.8 0.4
1.5 2 21.1 0.2
2.4 0.5 2.4 0.5 1 2.4 21.2 2.4

Bopt / rS
1
22.3 2.5 0.2
1.5 0.2
2 0.4
2 2 0.5 2 2 0.6 2
0.8 0.4
1 0.8 0.6
5
1
1.6 1.6 1.6 1.6
1.5
1
5 1
1.2 1.2 1.2 1.2
0.7 0.9 1.1 1.3 1.5 1.7 1.9 0.5 0.7 0.9 1.1 1.3 1.5 1.7 1.9 0.7 0.9 1.1 1.3 1.5 1.7 1.9 0.7 0.9 1.1 1.3 1.5 1.7 1.9
8DPSK Se. / IQ, NRZ 8DPSK Pa. / IQ, NRZ 8DPSK Se. / IQ, RZ 8DPSK Pa. / IQ, RZ
3.2 3.2 3.2 3.2
4 0.6
3 3.5 2.5 0.6
24.3 0.4
2.8 1 2 2.8 2.8 0.4 2.8
2.5 1 2 0.4
2.4 0.5 0.5 2.4 0.5 1.5 2.4 22.0 0.2 2.4 21.8
Bopt / rS
1.5 0.2
1
2 1.5 2 2 0.2 2
2 0.2
22.8 0.5 0.6
0.4 0.4
1 0.8 0.6
1.6 5 1 0.8
1.6 1.6 1.6 1
5 1
1.2 1.2 1.2 1.2
0.7 0.9 1.1 1.3 1.5 1.7 1.9 0.5 0.7 0.9 1.1 1.3 1.5 1.7 1.9 0.9 1.1 1.3 1.5 1.7 1.9 0.7 0.9 1.1 1.3 1.5 1.7 1.9
16DPSK Se. / arg, NRZ Star 16QAM Se. / arg, NRZ 16DPSK Se. / arg, RZ Star 16QAM Se. / arg, RZ
3.6 3.4 3.2 3.4
2 1
28.6 0.5 24.5 0.6 0.8
3.2 0.5 0.4
3 1 1.5 2.8 0.2 3 0.6
2.5
0.5 0.2 26.0 0.4
2.8 1 3
2.6 2.4 0.4 2.6 0.2
Bopt / rS
1.5 2 3.5 4.5 5

2.4 4 0.6
1 0.8
2.2 2 2 1 2.2 22.0
2 3 0.2
0.4
5 1
1.8 1.6 1.8
1.6
1.2 1.4 1.2 1.4

0.7 0.9 1.1 1.3 1.5 1.7 1.9 0.7 0.9 1.1 1.3 0.7 0.9 1.1 1.3 1.5 1.7 1.9 1.1 1.3 1.5 1.7 1.9
Bel / rS Bel / rS Bel / rS Bel / rS
Fig. 7.5 Optimal receiver filter bandwidths for direct detection at r B = 40 Gbit/s for second-order
Gaussian optical bandpass filters and fifth-order electrical Bessel filters; The stars indicate the
points with minimum required OSNR and are labeled with the minimum OSNR value at a BER of
10−9 . The contour lines are labeled with the OSNR penalty at a BER of 10−9 with respect to the
minimum OSNR. RZ duty cycle: 50%
The stars in Fig. 7.5 indicate points with optimized filter bandwidths and are
labeled with the minimum OSNR value at a BER of 10−9 . The contour lines are
labeled with the OSNR penalty at a BER of 10−9 with respect to the minimum
OSNR value for optimized filter bandwidths. It can be concluded that NRZ signals
are much more sensitive to filter bandwidth variations than signals with RZ pulse
shape. In the case of RZ, the contour lines are labeled in steps of 0.2 (compared
to steps of 0.5 for NRZ) and high deviations from the optimum electrical receiver
bandwidths can be tolerated without significant penalties, as well as deviations of
the optical receiver bandwidth down to a certain minimum value which depends on
each modulation format and increases with the modulation format order. Moreover,
there are only minor differences between systems with serial and parallel transmit-
ters for RZ. For NRZ, the optimal receiver filter bandwidths are strongly dependent
on the transmitter configuration used and the parallel transmitter proves to be much
more tolerant than the serial transmitter as regards bandwidth deviations of the op-
tical filter—especially for the higher-order formats, as becomes apparent from the
results for 8DPSK. When the serial transmitter is used in combination with NRZ, an
optical bandwidth should be chosen that is not too small, since the signals become
very vulnerable to narrow optical filtering with increasing order of the phase modu-
lation. This can be observed by comparing the contour plots of NRZ-DQPSK, NRZ-
8DPSK and NRZ-16DPSK for the serial transmitter configuration. In Table 7.1, the
optimum optical and electrical receiver filter bandwidths are summarized for all
modulation formats and system configurations examined here.
Table 7.1 Optimum receiver filter bandwidths for back-to-back in systems with direct detection
Modulation Format (Configuration) Bel /r S , Bopt /r S for NRZ Bel /r S , Bopt /r S for RZ
2ASK 0.7, 1.4 0.7, 2.4

DPBSK 1.3, 1.4 1.3, 2.2
DQPSK (Se./arg) 0.9, 2.4 1.1, 2.2
DQPSK (Pa./arg) 0.7, 1.4 0.7, 2.4
8DPSK (Se./arg) 0.9, 3.2 1.1, 2.4
8DPSK (Se./IQ) 0.9, 3.0 1.1, 2.4
8DPSK (Pa./arg) 0.7, 2.4 0.7, 2.6
8DPSK (Pa./IQ) 0.7, 2.0 0.9, 2.4
16DPSK (Se./arg) 0.9, 3.4 1.3, 2.6
Star 16QAM (Se./arg) 0.9, 3.2 1.5, 2.2
Star 16QAM (Pa./arg) 0.7, 2.4 0.7, 2.6
In Fig. 7.6, the BER is illustrated over the OSNR for optimized filter bandwidths,
again assuming for all formats the use of the serial transmitter structure and arg-
decision. When these curves are compared with the curves for fixed optical and
electrical receiver filter bandwidths of Bopt = 2.5 · r S and Bel = 0.75 · r S depicted
in Fig. 7.2, it becomes apparent that the differences are nearly negligible for RZ
7.1 Optimal Receiver Filter Bandwidths for Direct Detection 163
pulse shape, whereas significant improvements are obtained for NRZ-ASK / NRZ-
DBPSK and especially for the NRZ-16DPSK format which is very sensitive to too
narrow optical filtering.
1E-2 1E-2
1E-3
a 1E-3 b
16DPSK 16DPSK
1E-4 1E-4
2ASK
Star 16QAM
1E-5 1E-5 2ASK
BER
BER
RR 1.6
DQPSK
Star 16QAM
1E-6 RR 1.6 1E-6
DQPSK
1E-7 DBPSK 1E-7
DBPSK
1E-8 8DPSK 1E-8 8DPSK
NRZ RZ
1E-9 1E-9
10 12 14 16 18 20 22 24 26 28 10 12 14 16 18 20 22 24 26 28
OSNR [dB] OSNR [dB]
Fig. 7.6 OSNR requirements for direct detection receivers for NRZ (a) and RZ (b) at a data rate
of r B = 40 Gbit/s for optimal receiver filter bandwidths when the serial transmitter structure and
arg-decision are employed, calculated by semi-analytical BER estimation
The conclusions concerning the comparison of different transmitter and decision

schemes drawn in Sect. 7.1.1 are just as valid for optimized filter bandwidths, as
can be observed from the 8DPSK curves in Fig. 7.7. These can be compared to the
curves for non-optimized bandwidths shown in Fig. 7.4.
TX / RX comparison 8DPSK, opt. BW

1E-2
Se. / IQ NRZ
1E-3
Se. / arg NRZ
1E-4
Pa. / IQ NRZ
1E-5
BER
Pa. / arg NRZ

1E-6
Fig. 7.7 Performance com- Both TX,
parison of different transmit- 1E-7 arg-dec., RZ
ter structures and decision 1E-8 Both TX,

IQ-dec., RZ
schemes for 8DPSK when 1E-9
optimal receiver filter band- 16 17 18 19 20 21 22 23 24 25
widths are applied OSNR [dB]
Although the back-to-back OSNR requirements are shown here for a fixed data
rate of 40 Gbit/s, they can also be scaled to other data rates. When quadrupling the
data rate, for instance, the fourfold amount of noise power falls into the detection
bandwidth and the fourfold signal power PR X,avg is required at the receiver input
to obtain the same BER. With the OSNR definition of (7.1), this corresponds to a
required OSNR which is 6 dB higher. Alternatively, the system noise performance
can be expressed in a form that is independent of the data rate by illustrating the
BER with respect to the SNR defined in (6.30). In that case, the relative distances
between the curves of different modulation formats change.
7.1.3 Laser Linewidth Requirements
This section gives an insight into the laser linewidth requirements of direct detection
systems with high-order modulation. In the case of negligible influence of the optical
receiver filter, an analytical formula for calculation of the BER of DBPSK systems
under consideration of the laser linewidth 1νs is presented in [4] as
∞
1 S N R · e−S N R X (−1)n
BER = − ·
2 2 2n + 1
n=0

SN R S N R 2 −(2n+1)2 π 1νs TS
· In + In+1 ·e , (7.2)
2 2
where In represents the n-th order modified Bessel function of the first kind. How-
ever, the phase-noise statistics of light are modified by optical filtering. In [15], an
analytical method is described for taking the statistics of filtered DBPSK signals
with phase noise into account. The influence of different optical receiver filters on
the phase noise statistics is illustrated in [1], where a semi-analytical BER estima-
tion method for DQPSK systems was proposed under consideration of the laser
phase noise.
The laser linewidth requirements presented here were calculated by MC simula-
tions, where the laser phase noise is modeled as a random walk process as described
in Sect. 2.1.1, assuming a white power spectral density of the frequency noise and
a Lorentzian line-shape. Within a symbol duration of TS , the optical signal phase
exhibits a Gaussian distributed random phase change of 1ϕn s (t), whose variance is
given by h1ϕn2s (TS )i = 2π 1νs TS .
1E-3 1E-3
∆ν / rB = 3.75e-3
a ∆ν / rB = 25e-3 b
∆ν / rB = 20e-3
BER
BER
∆ν / rB = 2.5e-3
1E-4 1E-4
∆ν / rB = 15e-3
∆ν / rB = 0 ∆ν / rB = 0
∆ν / rB = 1.25e-3
∆ν / rB = 5e-3
∆ν / rB = 0.625e-3
RZ-DBPSK NRZ-DQPSK
1E-5 1E-5
11 12 13 14 15 16 17 14 15 16 17 18 19 20
OSNR [dB] OSNR [dB]
Fig. 7.8 BER performance of RZ-DBPSK (a) and NRZ-DQPSK (b) considering laser phase noise
7.1 Laser Linewidth Requirements for Direct Detection 165
In Fig. 7.8, the BER is shown against the OSNR for back-to-back, considering
different linewidth-to-data-rate ratios, exemplarily for RZ-DBPSK (a) and NRZ-
DQPSK (b). The receiver filters are assumed to be second-order Gaussian optical
bandpass filters and third-order electrical Bessel filters with 3 dB bandwidths of
Bopt = 2.5 · r S and Bel = 0.75 · r S . The laser phase noise leads to BER floors which
are a function of the ratio of the laser linewidth to the data rate. It should be noted
that the laser linewidth requirements are much more stringent when calling for a
BER of 10−9 instead of 10−4 due to the appearance of the BER floors.
From the BER curves, the OSNR penalties at a BER of 10−4 can be derived.
These are defined by the differences of the required OSNR at a BER of 10−4 for
the inspected curves and the reference curve for zero linewidth, and illustrated with
respect to the linewidth-to-data-rate ratio for all modulation formats examined here
in Fig. 7.9 for NRZ (a) and RZ (b).
4 4
a Penalty @ BER=10 [dB]
Penalty @ BER=10 [dB]
b
Star
3 16QAM 3
-4
-4
Star DBPSK
16QAM
2 DBPSK 2 DQPSK
16DPSK DQPSK 16DPSK
1 1
8DPSK NRZ 8DPSK RZ

0 -5 -4 -3 -2 -1
0 -5 -4 -3 -2 -1
10 10 10 10 10 10 10 10 10 10
Linewidth / data rate Linewidth / data rate
Fig. 7.9 OSNR penalties at a BER of 10−4 with respect to the linewidth-to-data-rate ratio for
various modulation formats for NRZ (a) and RZ (b) pulse shapes.
When comparing the phase modulation formats, the linewidth requirements in-
crease with an increasing number of phase states. This is reasonable, since a cer-
tain level of laser phase noise becomes more critical for smaller phase distances.
Additionally, the reduction of the symbol rate makes the laser phase noise more
critical for modulation formats with a higher number of bits per symbol. This way,
the slightly worse performance of Star 16QAM compared with 8DPSK can be ex-
plained. With regard to the pulse shapes, differences between NRZ and RZ are rel-
atively small.
It can be concluded that the linewidth requirements for direct detection are quite
relaxed. Even for NRZ-16DPSK, which shows the most stringent requirements on
the laser phase noise of the formats discussed here, the required linewidth-to-data-
rate ratio is 2.6 · 10−5 . This corresponds to 1νs =1 MHz at r B =40 Gbit/s.
7.1.4 Chromatic Dispersion Tolerances
The last sections have discussed the influence of some performance parameters for
the back-to-back case. In the following sections, system parameters for single-span
optical fiber transmission are highlighted. Next, the chromatic dispersion tolerances
are presented. These can be determined by sending the optical signals over a disper-
sive single mode fiber, without using any dispersion compensation. Other disturbing
effects such as laser phase noise and Kerr nonlinearities are neglected.
Figure 7.10 shows the chromatic dispersion tolerances at r B = 40 Gbit/s of all
modulation formats discussed here for direct detection. In Fig. 7.10 is assumed that
serial transmitters and arg-decision are employed, as well as optical and electrical
filters with 3 dB bandwidths of Bopt = 2.5 · r S and Bel = 0.75 · r S , respectively. As
performance measures, two different quality criteria are used. The OSNR penalties
at a BER of 10−9 were calculated by semi-analytical BER estimation (the OSNR
values for zero dispersion serve as a reference). Additionally, eye opening penal-
ties are illustrated which are shown separately for the intensity detection and phase
detection branches for Star 16QAM.
4 4
a
2ASK b 2ASK,
DBPSK
3 3
-9
-9
DBPSK
DQPSK 8DPSK, DQPSK

16DPSK,
2 Star 2 8DPSK
16QAM
16DPSK
1 1
Star
NRZ 16QAM RZ
0 0
-320 -160 0 160 320 -320 -160 0 160 320
Dispersion [ps/nm] Dispersion [ps/nm]
4 4
c 2ASK d 2ASK,
DBPSK
3 DBPSK 3
DQPSK
DQPSK, 8DPSK
8DPSK,
2 16DPSK
2
Star Star Star
16DPSK
16QAM 16QAM 16QAM
Phase Int. Star Int.
1 1 16QAM
Phase
NRZ RZ
0 0
-320 -160 0 160 320 -320 -160 0 160 320
Fig. 7.10 Chromatic dispersion tolerances of various modulation formats for direct detection at
40 Gbit/s when using the serial transmitters and arg-decision; The two upper diagrams show the
OSNR penalties at a BER of 10−9 against the chromatic dispersion for NRZ (a) and RZ (b). The
bottom diagrams (c, d) illustrate the eye opening penalties.
7.1 Chromatic Dispersion Tolerances for Direct Detection 167
The principal tendencies for the chromatic dispersion tolerances for the migra-
tion to high-order modulation formats become clear when looking at the diagrams
for RZ pulse shape (Fig. 7.10b/d). For a comparison at a fixed data rate, the chro-
matic dispersion tolerances increase with the order of the modulation format due to
the reduced symbol rates and the narrowed spectral widths. However, it becomes
apparent from Fig. 7.10a/c that this is not valid without exception for NRZ when
employing the serial transmitter structure, since the advantage of the smaller band-
width for the high-order formats is neutralized by detrimental chirp characteristics
of the transmitter output signals.
Generally, RZ outperforms NRZ for all the non-binary modulation formats. In the
case of binary modulation, the smaller bandwidth of the NRZ signals compared with
RZ is the reason for superior NRZ performance, whereas for high-order modulation
formats the reduced chirp of the RZ signals becomes the dominant influencing factor
and yields significantly higher chromatic dispersion tolerances for RZ.
Both 16-ary formats, RZ-16DPSK and RZ-Star 16QAM, show by far the highest
dispersion tolerances of all the formats observed here. When accepting an OSNR
penalty of 2 dB at BER=10−9 , a dispersion of about 290 ps/nm can be tolerated and
the EOP for 320 ps/nm is about 1dB. For Star 16QAM, the EOP in the intensity de-
tection branch is slightly worse than in the phase detection branch for—with respect
to OSNR—the optimum ring ratio of 1.6 (see Fig. 7.10c/d). For other ring ratios
this relation can change as discussed in [13]. It can be concluded from a comparison
of the OSNR penalties and the eye opening penalties that the EOP overestimates
the acceptable residual dispersion compared to the OSNR penalties at BER=10−9 .
To better illustrate the disturbing effect of chromatic dispersion, the IQ diagrams
and the phase and intensity eyes for Star 16QAM are depicted in Fig. 7.11 for zero
dispersion (a) and 320 ps/nm dispersion (b).
a Dispersion 0ps/nm b Dispersion 320 ps/nm
NRZ
RZ
IQ-Plot Phase eye Intensity eye IQ-Plot Phase eye Intensity eye
Fig. 7.11 Effect of chromatic dispersion on Star 16QAM signals

Impact of the Transmitter Configuration and the Decision Scheme
The signal shapes, symbol transitions and chirp characteristics are different for the
output signals of the serial and parallel transmitters (see Sect. 2.4.4 and Sect. 2.5)
and have a significant influence on the obtainable chromatic dispersion tolerance.
In Fig. 7.12, this influence is illustrated for 8DPSK (a) and Star 16QAM (b), just as
the impact of the decision method (see Sect. 3.2.3) is.
TX / RX comparison 8DPSK TX / RX comparison Star 16QAM

4 4
a

Se. / arg NRZ b Se. / arg NRZ
3 3
-9
Pa. / arg NRZ
-9
Pa. / IQ NRZ
Se. / arg RZ Se. / arg RZ

2 2
Pa. / IQ RZ Pa. / arg RZ
1 1
0 0
-320 -160 0 160 320 -320 -160 0 160 320
Fig. 7.12 Impact of the transmitter configuration and decision method on the chromatic dispersion
tolerances at 40 Gbit/s for direct detection. a 8DPSK. b Star 16QAM.
Unlike the decision technique, which does not significantly affect the attainable
dispersion tolerance, the transmitter configuration has a substantial effect. The se-
rial transmitters show higher penalties than the parallel transmitters due to the dis-
advantageous chirp characteristics. These tendencies are consistent with the signal
characteristics discussed in Sect. 2.4.4 and Sect. 2.5 and illustrated in Fig. 2.11 and
Fig. 2.14, respectively. The product of the chirp and the intensity was introduced
in these sections as an intuitive measure for the chirp influence on the transmission
performance and the parallel transmitter was already expected to exhibit a better
transmission behavior than the serial transmitter.
7.1.5 Self Phase Modulation Tolerances
Self phase modulation (SPM) tolerances for single-span transmission can be deter-
mined using the system setup shown in Fig. 7.13. The optical signals are transmitted
over a single dispersive and nonlinear fiber link with a length of l = 80 km. The
dispersion parameter and the nonlinear index coefficient of the SMF at the optical
carrier frequency f s = 193.1 THz are assumed here to be Dλ = 16 ps/(nm·km) and
n 2 = 2.6 · 10−20 m2 /W (Ae f f = 80 · 10−12 m2 ), respectively, and the dispersion
slope is assumed to be zero (Sλ = 0). The chromatic dispersion is compensated for
7.1 Self Phase Modulation Tolerances for Direct Detection 169
behind the link by a dispersion compensating fiber (degree of compensation: 100%)

which is assumed to be linear. The average fiber input power PT X,avg is varied.
SMF DCF
TX BPF RX
80km 100%
CD-comp.
PTX OSNR (noise bandwidth 0.1nm)
Fig. 7.13 Single-span system setup for determination of self phase modulation tolerances
In Fig. 7.14, the SPM tolerances of all examined formats are depicted for a data
rate of 40 Gbit/s, where the use of serial configurations at the transmitter side is
assumed, as well as the application of second-order Gaussian optical bandpass filters
and third-order electrical Bessel filters—with 3 dB bandwidths of Bopt = 2.5 · r S
and Bel = 0.75 · r S —and arg-decision at the receiver end.
4 4
NRZ RZ
a
8DPSK 16DPSK b
3 3
-9
2ASK
-9
2ASK 8DPSK
Star
2 16QAM 2 DQPSK
DQPSK
16DPSK DBPSK
1 1 Star
16QAM
DBPSK
0 0
0 3 6 9 12 15 0 3 6 9 12 15
Fiber input power [dBm] Fiber input power [dBm]
4 4
NRZ
Star
8DPSK c RZ 2ASK,
Star
d
16QAM
16QAM
3 Int. 3 Int.
2ASK
16DPSK
2 16DPSK 2
8DPSK
Star DQPSK
16QAM Star DBPSK,
1 Phase 1 16QAM DQPSK
DBPSK Phase
0 0
0 3 6 9 12 15 0 3 6 9 12 15
Fig. 7.14 SPM tolerances of various modulation formats for direct detection at 40 Gbit/s when
serial transmitters and arg-decision are used; The two upper diagrams show the OSNR penalties at
a BER of 10−9 against the fiber input power for NRZ (a) and RZ (b). The bottom diagrams (c, d)
illustrate the eye opening penalties.
Figure 7.14a/b illustrates the OSNR penalties (with respect to the linear case),
and Fig. 7.14c/d illustrates the eye opening penalties. It can be seen that RZ-DPSK
signals exhibit higher tolerable fiber input powers than NRZ-DPSK signals due to
their better chirp characteristics. Higher chirp leads to a greater influence of the
chromatic dispersion and thus to a stronger interaction between chromatic disper-
sion and SPM. A comparison of the different DPSK formats shows that the SPM
tolerance decreases with an increasing number of symbols which is equivalent to a
decrease in the phase distances. Moreover, DPSK formats exhibit higher SPM tol-
erances than ASK formats. For instance, even RZ-16DPSK shows a similar SPM
performance to RZ-2ASK.
Whereas the Star 16QAM gives a considerably better performance than the
16DPSK with respect to noise and a similar chromatic dispersion tolerance, the
SPM tolerance of Star 16QAM is very poor for NRZ and RZ. For RZ-16DPSK,
an OSNR penalty of less than 2 dB is observed for 11 dBm fiber input power. In
contrast, a fiber input power of about 5.5 dBm yields an OSNR penalty of 2 dB for
RZ-Star 16QAM. In the case of NRZ, this performance difference can not be de-
tected because the NRZ-16DPSK is highly sensitive against chirp when the serial
transmitter is employed. As becomes apparent from the EOP diagrams shown in
Fig. 7.14c/d, the remarkably bad SPM tolerance of Star QAM formats is caused by
a poor performance in the phase detection branch. The reasons for this effect, as
well as solutions for a possible performance improvement using appropriate SPM
compensators, are illustrated in detail in Sect. 7.1.6. Another possibility for enhanc-
ing the poor SPM tolerance of Star 16QAM in the phase detection branch which is
simultaneously associated with a performance degradation in the intensity detection
branch, is to modify the ring ratio as shown in Fig. 7.15. However, this is not an
ideal solution because the OSNR performance becomes worse in this case [13].
RZ-Star 16QAM, phase detection branch RZ-Star 16QAM, intensity detection branch
4 4
a RR 4
b RR 1.19
3 3 RR 1.41
RR 2
RR 2
2 RR 1.41 2
RR 1.19
1 1
RR 4
0 0
0 3 6 9 12 15 0 3 6 9 12 15
Fig. 7.15 Influence of the ring ratio (RR) on the SPM tolerance in the phase detection branch (a)
and the intensity detection branch (b) for RZ-Star 16QAM
7.1 Nonlinear Phase Shift Compensation for Direct Detection 171
Influence of the Transmitter Configuration and the Decision Scheme
A comparison of the SPM tolerances of 8DPSK and Star 16QAM for different
transmitter configurations and decision techniques is depicted in Fig. 7.16a and
Fig. 7.16b, respectively.
TX / RX comparison 8DPSK TX / RX comparison Star 16QAM

4 4

a
b
Se. / arg NRZ
3 3
-9
Se. / arg NRZ
-9
Se. / IQ NRZ Se. TX,

arg + IQ
RZ
Pa. / arg NRZ
2 2
Pa. / IQ NRZ Pa. TX, Pa. / arg NRZ
arg + IQ
RZ
1 1 Se. / arg RZ
Se. / IQ RZ
0 0
0 3 6 9 12 15 0 3 6 9 12 15
Fig. 7.16 Impact of the transmitter configuration and decision method on the SPM tolerances at
40 Gbit/s for direct detection for 8DPSK (a) and Star 16QAM (b)
As can be seen from Fig. 7.16a, DPSK systems with serial transmitters exhibit a
worse SPM tolerance than DPSK systems with parallel transmitters. This difference
appears more distinctly for NRZ. A comparison of the decision methods shows
no strong differences. The SPM performance is very poor throughout all system
configurations for Star 16QAM, as shown in Fig. 7.16b. The reasons for this effect
and its compensation are discussed in Sect. 7.1.6.
7.1.6 Nonlinear Phase Shift Compensation
Section 6.3 showed that an optical signal experiences a SPM-induced nonlinear

2
phase shift of ϕ N L (z, x, t ′ ) = −γ · le f f (z) · x(z = 0, t ′ ) = −γ · le f f (z) · PT X (t ′ )
when it propagates in a fiber (under negligence of the chromatic dispersion). Ob-
viously, the nonlinear phase shift depends on the instantaneous fiber input power
PT X (t ′ ). For DPSK formats, the power levels of the different symbols are equal
(in the idealized case) and every symbol gets the same nonlinear phase shift during
transmission. Since phase differences are evaluated for differential demodulation,
these nonlinear phase shifts have no influence on information recovery and the de-
termined modulation phase differences are still correct. In contrast, symbols with
different power levels experience different nonlinear phase shifts for QAM formats.
Therefore, the modulation phase differences are incorrect after differential detection
when two consecutive symbols with different power levels are detected. This effect
reflects an inherent problem of optical QAM transmission that leads to performance

degradation even when not taking the interaction between SPM and chromatic dis-
persion into account.
The distortion of a RZ-Star 16QAM signal after differential detection is shown
in the electrical IQ diagram on the right side of Fig. 7.17b for an average fiber input
power of PT X,avg = 8 dBm after transmission over a non-dispersive fiber of length
l = 80 km (le f f ≈ 21.17 km for αd B = 0.2 dB/km) with n 2 = 2.6 · 10−20 m2 /W and
Ae f f = 80 · 10−12 m2, which corresponds to a nonlinear propagation coefficient of
γ = 2π f s n 2 / c Ae f f = 1.31 W−1 km−1 for f s = 193.1 THz.
a 16DPSK DD b Star 16QAM DD
w/o SPM w. SPM w/o SPM w. SPM
Fig. 7.17 Effect of the SPM-induced nonlinear phase shift on differentially detected 16DPSK (a)
and Star 16QAM (b) signals
The SPM-induced nonlinear phase shift has no influence on the differential de-
tection of (idealized) DPSK signals, as can be observed from Fig. 7.17a. In the case
of Star 16QAM, however, the detected symbols are positioned on three different cir-
cles after differential detection as explained in Sect. 3.2.3. It becomes apparent from
the right part of Fig. 7.17b that the middle circle (corresponding to the case of de-
tecting consecutive symbols with different power levels) is significantly corrupted
by SPM-induced phase errors which are equal to the differences of the nonlinear
phase shifts experienced by two consecutive symbols with different power levels.
The phase distortions caused by non-zero nonlinear phase shift differences can
be significantly reduced by using relatively simple compensation schemes—at least
in the case of single-span transmission. When taking only SPM and attenuation
into account and neglecting chromatic dispersion and any other effects disturbing
the signal intensity, the intensity shape of the fiber output signal can be assumed
to be equal to the one of the fiber input signal. The nonlinear phase shift can then
be calculated from the instantaneous fiber output power. When the same parameter
values are assumed as specified above, it holds
ϕ N L (t ′ ) = −γ · le f f (z) · PT X (t ′ )
αd B ·l
= −γ · le f f (z) · PR X (t ′ ) · 10 10 ≈ 1104 · PR X (t ′ ). (7.3)
Under these simplified conditions, the phase distortions can be ideally compensated
for using the fiber output signal if the fiber and system parameters are known. In
practice, the compensation is not ideal because the intensity shape of the propagat-
ing signal changes along the fiber, for instance due to chromatic dispersion.
7.1 Nonlinear Phase Shift Compensation for Direct Detection 173
In Fig. 7.18, two different compensation schemes are illustrated, both of which
may compensate for nonlinear phase shift. Both schemes can also be used to reduce
the variance of the nonlinear phase noise.
a Optical compensation of NL phase shift b Electrical compensation of NL phase shift
Intensity
PM -
TS
3dB
In-phase
ROTATE
Quadrature
Fig. 7.18 Schemes for compensation of the nonlinear phase shift for QAM formats. a Compen-
sation with a phase modulator (PM) in the optical domain in front of the QAM receiver before
differential detection. b Compensation in the electrical domain after differential detection.
Figure 7.18a shows a simple optical compensator, proposed in [4, 13, 18] and
typically placed in front of the QAM receiver. The nonlinear phase shift is rotated
back by an optical phase modulator before differential detection. The electrical driv-
ing current of the phase modulator is proportional to the fiber output power and must
be adjusted according to the system parameters to ensure appropriate compensation
of the accumulated nonlinear phase shift. Care has to be taken that the delays of the
upper and the lower paths are equal. SPM compensation can also be accomplished
in the electrical domain after differential detection as depicted in Fig. 7.18b. In this
case, a corrective electrical signal proportional to the difference of the nonlinear
phase shifts of two consecutive symbols is generated from the received photocur-
rent in the intensity detection branch and is used to rotate back the phase of the
symbols received in the phase detection branch.
The efficiency of the optical compensator for Star 16QAM is illustrated in
Fig. 7.19. The electrical compensator exhibits nearly the same performance.
4
RZ
uncompensated
RZ
3
-9
compensated
NRZ
2 uncompensated
Fig. 7.19 Enhancement of the NRZ

SPM tolerance by nonlinear 1 compensated
phase shift compensation

for Star 16QAM systems 0
with serial transmitter and 0 3 6 9 12 15
arg-decision Fiber input power [dBm]
It can be observed that the SPM tolerance can be significantly improved by com-
pensating for the nonlinear phase shift. For NRZ, the enhancement of the SPM tol-
erance is smaller than for RZ, simply caused by the fact that the SPM tolerance
for NRZ is worse than for RZ anyway. A comparison of Fig. 7.19 with Fig. 7.14a
shows that the SPM tolerance for NRZ-Star 16QAM with compensation is even
in the same range as the SPM tolerance for NRZ-8DPSK. In the case of RZ-Star
16QAM, a fiber input power which is more than 5 dB higher can be tolerated thanks
to the nonlinear phase shift compensation and the same performance can be reached
as for RZ-16DPSK.
To better illustrate the effect of compensation, the received NRZ-Star 16QAM
IQ diagrams, phase eyes and intensity eyes are depicted in Fig. 7.20 for a fiber input
power of PT X,avg = 8 dBm and three different cases: without SPM (a), with SPM
without compensation (b) and with SPM when using the compensation (c). Without
compensation, the phase eye is completely closed due to SPM, whereas the intensity
eye is only slightly distorted. By using the nonlinear phase shift compensation, the
phase eye can be clearly opened and the SPM tolerance improves significantly.
a
w/o SPM
b
w. SPM, w/o compensation
c
w. SPM, w. compensation
IQ-Plot Phase eye Intensity eye
Fig. 7.20 Effect of nonlinear phase shift compensation for NRZ-Star 16QAM
The nonlinear phase shift compensation schemes shown in this section are quite
effective for single-span transmission. But even if the chromatic dispersion is com-
pletely compensated for before nonlinear phase shift compensation, the interaction
between chromatic dispersion and SPM prevents a complete compensation of the
nonlinear phase shift. So there is the potential to obtain even higher performance
gains with compensation techniques which better take the mutation of the signal in-
tensity during transmission into account and more accurately estimate the nonlinear
phase shift accumulated for each symbol. Potentially, compensation efficiency can
be enhanced using digital equalization. The compensation of the nonlinear phase
shift for multi-span long-haul transmission is discussed later on in Chap. 8.
7.1 Parameter Summary Direct Detection 175
7.1.7 Parameter Summary
Table 7.2 summarizes the performance parameters of 40 Gbit/s direct detection sys-
tems, considering all high-order modulation discussed in the previous sections and
NRZ / RZ pulse shapes. The OSNR requirements given in Table 7.2 are valid for
the optimized receiver filter bandwidths specified in Table 7.1. The shown receiver
sensitivities imply a pre-amplifier noise figure of Fd B = 3 dB and are calculated
from the OSNR requirements by PR X,avg,d Bm = O S N Rd B − 58 d Bm + Fd B as
described in Sect. 7.1.1. Moreover, Table 7.2 lists the laser linewidth requirements,
the chromatic dispersion tolerances and the SPM tolerances. The last two parame-
ters are specified by eye opening penalties as well as by OSNR penalties. All given
parameter values are valid for direct detection systems with serial transmitter con-
figurations and arg-decision.
Table 7.2 Performance parameters of various modulation formats for direct detection at 40 Gbit/s
for NRZ / RZ. The given parameter values are valid for direct detection systems with serial trans-
mitters and arg-decision. Ph: Phase, Int: Intensity, Comp: Nonlinear phase shift compensation
Modulation Format 2ASK DBPSK DQPSK 8DPSK 16DPSK Star 16QAM
O S N Rd B [dB] 15.2 / 12.9 / 14.8 / 19.1 / 23.8 / 19.6 /

@ BER=10−4 14.7 12.1 13.1 16.8 21.5 17.3
PR X,avg,d Bm [dBm] -39.8 / -42.1 / -40.2 / -35.9 / -31.2 / -35.4 /
@ BER=10−4 -40.3 -42.9 -41.9 -38.2 -33.5 -37.7
O S N Rd B [dB] 19.0 / 16.5 / 19.1 / 23.6 / 28.6 / 24.4 /
@ BER=10−9 18.4 15.6 17.1 21.2 26.0 22.0
PR X,avg,d Bm [dBm] -36.0 / -38.5 / -35.9 / -31.4 / -26.4 / -30.6 /
@ BER=10−9 -36.6 -39.4 -37.9 -33.8 -29.0 -33.0
1ν/r B -/ 1.8e-2 / 1.7e-3 / 2.6e-4 / 2.6e-5 / 2e-4 /
@ Pen. 2 dB (10−4 ) - 1.6e-2 1.6e-3 3.2e-4 4.5e-5 2.3e-4
1ν [MHz], 40 Gbit/s, -/ 720 / 68 / 10.4 / 1.0 / 8.0 /
@ Pen. 2 dB (10−4 ) - 640 64 12.8 1.8 9.2
Disp. Tol. [ps/nm] 48 / 64 / 77 / 111 / 115 / 114 /
@ Pen 2 dB (10−9 ) 40 35 123 205 280 290
Disp. Tol. [ps/nm] 86 / 84 / 118 / 118 / 118 / Ph 193 / >320
@ EOP 2 dB 37 42 151 302 >320 Int 160 / >320
SPM Tol. [dBm] 8.9 / >15 / 10.7 / 6.2 / 2.9 / 4.4 / 5.5
@ Pen. 2 dB (10−9 ) 11.9 >15 >15 13.8 11.3 Comp 6.5 / 11.0
SPM Tol. [dBm] 10.3 / >15 / 12.6 / 9.1 / 7.4 / Ph 6.1 / 6.0
@ EOP 2 dB 13.2 >15 >15 15.0 13.8 Comp 7.8 / 11.1
Int 11.8 / 12.9
The migration to high-order modulation formats with more bits per symbol re-
sults in higher spectral efficiencies, higher dispersion tolerances and higher PMD
tolerances (as discussed in Sect. 6.4.2). But they come along with higher OSNR
requirements and worse SPM tolerances. Both are critical parameters when attain-
able distances of many thousand kilometers for multi-span long-haul transmission
are considered. This point is further detailed in Chap. 8. The laser linewidth re-
quirements increase with a growing number of phase states and when reducing the
symbol rate. They remain above one megahertz for all shown modulation formats at
40 Gbit/s and are therefore relatively uncritical for direct detection.
As regards the single-channel single-span system configurations described in this
chapter, RZ features a significantly better system performance than NRZ for all sys-
tem parameters discussed here (except for the linewidth requirements which are
comparatively independent of the pulse shape). However, this may no longer be
valid for systems where optical signals are narrowly filtered and the signal character-
istics undergo a significant change, for instance in WDM systems with small channel
spacings. In particular, the performance differences between NRZ and RZ become
significant in systems where the serial transmitter configuration is employed. Per-
formance differences are smaller for the parallel transmitter structure, as can be
observed from the comparison of different system configurations for 8DPSK sum-
marized in Table 7.3.
Table 7.3 Comparison of different 8DPSK system configurations for direct detection at 40 Gbit/s
for NRZ / RZ pulse shapes
Configuration 8DPSK Se./arg 8DPSK Se./IQ 8DPSK Pa./arg 8DPSK Pa./IQ
O S N Rd B [dB] 19.1 / 19.7 / 17.8 / 18.3 /

@ BER=10−4 16.8 17.5 16.8 17.5
O S N Rd B [dB] 23.4 / 24.2 / 22.3 / 22.8 /
@ BER=10−9 21.2 21.9 21.1 21.8
1ν/r B 2.6e-4 / -/ -/ 2.9e-4 /

@ Pen. 2 dB (10−4 ) 3.2e-4 - - 2.6e-4
1ν [MHz], 40 Gbit/s, 10.4 / -/ -/ 11.4 /
@ Pen. 2 dB (10−4 ) 12.8 - - 10.5
Disp. Tol. [ps/nm] 111 / -/ -/ 160 /

@ Pen. 2 dB (10−9 ) 205 - - 274
Disp. Tol. [ps/nm] 118 / 138 / 212 / 230 /
@ EOP 2 dB 302 269 >320 302
SPM Tol. [dBm] 6.2 / 6.1 / 8.3 / 9.3 /

@ Pen. 2 dB (10−9 ) 13.8 13.8 14.3 14.6
SPM Tol. [dBm] 9.1 / 9.0 / 10.0 / 10.1 /
@ EOP 2 dB 15.0 14.5 >15 15.0
7.2 Systems with Homodyne Differential Detection 177
Each modulation format has its own benefits and drawbacks. When evaluating
the benefits of particular formats, it can be said that the application of the DBPSK
format instead of 2ASK leads to an improved OSNR performance and to a signif-
icant enhancement of the SPM tolerance at almost the same chromatic dispersion
tolerance. The DQPSK format seems to be a very attractive candidate for future net-
works. The OSNR requirements for DQPSK are still more relaxed than for 2ASK
and only slightly higher than for DBPSK. However, the chromatic dispersion tol-
erance can be enhanced by almost a factor of four for RZ pulse shape without re-
ducing the SPM tolerance to a great extent. When migrating to even higher-order
modulation formats, the spectral efficiency and the dispersion tolerance can be fur-
ther enhanced. However, this results in stricter OSNR requirements. Furthermore,
SPM tolerances decrease. A comparison of the two 16-ary modulation formats dis-
cussed here—16DPSK and Star 16QAM—shows clear advantages for Star 16QAM
as regards its OSNR requirements. On the other hand, the SPM tolerance of QAM
formats in the phase detection branch is very poor, and a compensation of the mean
nonlinear phase shift becomes necessary to avoid a significant performance degra-
dation due to SPM.
7.2 Systems with Homodyne Differential Detection
This section discusses the performance of systems with homodyne differential de-
tection. As for direct detection, the recovery of phase information is accomplished
by evaluating the phase differences of two consecutive symbols. Thus, the same
modulation formats can be received and the same data recovery schemes can be
employed at the receiver as for direct detection.
Under certain conditions, systems with homodyne differential detection exhibit a
similar or even the same performance as systems with direct detection. This can be
analytically explained based on the theory presented in [16] and [14]. The coming
sections present results from computer simulations. A similar behavior of systems
with direct detection and homodyne differential detection is thereby identified, re-
garding a wide range of modulation formats. Exemplarily, systems with homodyne
phase diversity receivers (see Sect. 3.4.1) are discussed. However, receivers with
digital differential demodulation described in Sect. 3.4.2 show a nearly the same per-
formance. Next, noise performances are detailed in Sect. 7.2.1—for shot-noise lim-
ited detection and amplifier noise limited detection. Afterwards, the laser linewidth
requirements for homodyne differential detection are illustrated in Sect. 7.2.2. Chro-
matic dispersion and SPM tolerances are briefly presented by eye opening penalties
in Sect. 7.2.3. In Sect. 7.2.4, the properties of systems with homodyne differential
detection are summarized.
7.2.1 Noise Performance
The traditional way of characterizing the noise performance of coherent receivers is

to determine the shot-noise limits. The transmitters and receivers are directly con-
nected as depicted in Fig. 7.21a. Amplifier noise is not present. The shot-noise is
assumed to be the dominant noise source, whereas the thermal noise is neglected
(which can be done for sufficiently large LO powers).
However, optical transmission systems with coherent detection and optical am-
plification on the link or in front of the receiver are no longer shot-noise limited, but
limited by the amplifier noise. Then, the dominant noise component usually is the
LO-ASE noise, as discussed in Sect. 3.3.2 and shown in [3]. In the case of amplifier
noise limited detection, the noise performance of coherent receivers can be charac-
terized by the OSNR requirements in the same way as for direct detection receivers.
This can be done using the setup depicted in Fig. 7.21b.
a Shot-noise limited detection (quantum limit η = 1)
TX Coherent
RX
η=1
PRX,avg
b Amplifier-noise limited detection (with FdB = 3dB)
Coherent
TX BPF RX
FdB = 3dB Shot-noise negligible

PRX,avg OSNR (0.1nm, both polarizations)
Fig. 7.21 Characterization of the receiver noise performance for shot-noise limited homodyne
detection (a) and amplifier noise limited homodyne detection (b) in the quantum limit for η = 1
and for Fd B = 3 dB; The required receiver input powers PR X,avg for a given BER are the same for
both configurations when using a quadrature receiver.
For η = 1/n sp = 2/F, both configurations illustrated in Fig. 7.21 exhibit the
same receiver performance and require the same power PR X,avg at the indicated
points to obtain a certain BER—when using a homodyne quadrature receiver and
neglecting the influence of the optical filtering in Fig. 7.21b. At first view, this is a
surprise because shot-noise limited homodyne detection is known to be 3 dB su-
perior in noise performance in comparison with homodyne detection limited by
amplifier noise [3, 4]. But this is no longer valid when using a quadrature re-
ceiver, where the optical power is equally split between the in-phase and quadra-
ture branches and the SNR degrades by 3 dB in each branch for shot-noise limited
detection, leading to a performance equivalent to that of amplifier noise limited
7.2 Noise Performances for Homodyne Differential Detection 179
detection. Due to this equivalence, the OSNR requirements at the reference point
behind the optical amplifier for amplifier noise limited detection can be calculated
from the necessary receiver input powers PR X,avg for shot-noise limited detection
by O S N Rd B = 58 dBm + PR X,avg,d Bm − Fd B with η = 2/F. The receiver sensi-
tivities for direct detection systems are specified for Fd B = 3 dB in Table 7.2. For
a fair comparison with these results, the receiver sensitivities of the homodyne re-
ceivers for amplifier noise limited detection are also specified here for Fd B = 3 dB.
This corresponds to an equivalent shot-noise limited system in the quantum limit
with η = 1 and a responsivity of R = 1.25 at f s = 193.1 THz.
In the following paragraphs, the back-to-back receiver sensitivities at BER=10−4 ,
which are defined as the necessary receiver input powers PR X,avg to obtain a BER
of 10−4 , are discussed by looking at the shot-noise limited case. These receiver
sensitivities can easily be scaled to the OSNR requirements for the amplifier noise
limited case as described in the previous paragraph. The BER versus the receiver
input power for shot-noise limited detection in the quantum limit for η = 1 at
40 Gbit/s is shown in Fig. 7.22 for all considered formats, regarding NRZ (a) and
RZ (b) pulse shapes. Thereby, it is assumed that third-order electrical Bessel low-
pass filters are used at two different locations of the receiver depicted in Fig. 3.15.
Two filters with an electrical 3 dB bandwidth of BW D M = 1.25·r S are placed behind
the two outputs of the quadrature optical frontend (in front of the electronic network)
in the in-phase and quadrature arms. These filters are typically used to select desired
channels of WDM signals. Two more electrical filters with Bel = 0.75·r S are located
in the in-phase and quadrature arms behind the electronic circuit. In the case of Star
16QAM, the receiver configuration depicted in Fig. 3.17 with amplitude detection
based on direct detection is observed here. Noise performances shown in Fig. 7.22
are valid for system configurations with serial transmitters and arg-decision.
1E-2 1E-2
a b
Star 16QAM
1E-3 RR 1.65 1E-3 16DPSK
DBPSK 16DPSK
BER
BER
2ASK
2ASK Star 16QAM
8DPSK RR 1.65
1E-4 8DPSK 1E-4
DQPSK
DBPSK
DQPSK
NRZ RZ
1E-5 1E-5
-45 -42 -39 -36 -33 -30 -27 -45 -42 -39 -36 -33 -30 -27
Received power [dBm] Received power [dBm]
Fig. 7.22 BER against receiver input power for homodyne phase diversity differential detection for
NRZ (a) and RZ (b) at r B = 40 Gbit/s for Bel = 0.75 · r S when employing the serial transmitter
structures and arg-decision
The comparison of the receiver sensitivities for homodyne phase diversity differ-
ential detection illustrated in Fig. 7.22 with the receiver sensitivities of direct detec-
tion receivers listed in Table 7.2 shows sizable similarities. As already mentioned
above, this can be analytically explained based on the theory presented in [16] and
[14]. Optical pre-amplified direct detection receivers are shown in [16] to be math-
ematically equivalent to shot-noise limited heterodyne receivers with delay demod-
ulation for η = 1/n sp = 2/F when ignoring the amplifier noise from the orthog-
onal polarization. When the noise from the orthogonal polarization is included, an
additional penalty of about 0.4 dB arises for direct detection [4]. Moreover, homo-
dyne phase diversity DBPSK receivers are shown to have the same performance
as heterodyne DBPSK receivers in [14]. Homodyne two-branch ASK phase diver-
sity receivers with squarers are 0.25 dB worse than ideal ASK heterodyne detectors
[14]. The results from Monte Carlo simulations presented here show that the system
equivalence also holds true for the high-order formats.
Small differences in the noise performances of both receiver schemes arise here
due to the different filters which have been employed. Almost identical perfor-
mance values can be observed for RZ, whereas for NRZ, the receiver sensitivities
of the direct detection receivers are partly better because the electrical filter band-
widths have not been optimized for homodyne phase diversity differential detection.
Furthermore—for shot-noise limited detection—a sensitivity degradation of about
3 dB can be detected for Star 16QAM homodyne differential detection compared
with direct detection, caused by the power splitting of the signal into the phase and
intensity detection branches and the smaller optical powers at the balanced detector
inputs. However, this disadvantage disappears for amplifier noise limited detection
and for the alternative receiver structure shown in Fig. 3.16.
The optimum ring ratios for the homodyne phase diversity Star 16QAM re-
ceivers are in the same range as for the Star 16QAM direct detection receivers
and correspond to a value of about 1.6. As regards the comparison of systems with
different transmitter configurations and decision schemes shown in Fig. 7.23 for
Star 16QAM, the same conclusions as for the direct detection receivers can be
drawn. Arg-decision leads to an improved noise performance in comparison with
IQ-decision due to better placed thresholds. This can be observed from the curves
for RZ pulse shape in Fig. 7.23. In the case of NRZ, the worse chirp behavior of the
serial transmitter neutralizes the advantage of the arg-decision scheme.
TX / RX comparison Star 16QAM

1E-2
Se. / arg NRZ
1E-3
Pa. / IQ NRZ
BER
Se. / arg RZ
Fig. 7.23 Impact of the trans- 1E-4
mitter configuration and de-
cision scheme on the noise Pa. / IQ RZ
performance of homodyne 1E-5

phase diversity receivers in -37 -36 -35 -34 -33 -32 -31
the case of Star 16QAM Received power [dBm]
7.2 Linewidth Requirements for Homodyne Differential Detection 181
In the case of direct detection as well as homodyne differential detection, the dif-
ferential demodulation is impaired by the laser phase noise. Whereas the linewidth
of the signal laser is relevant for the differential demodulation in the optical domain
for direct detection, for homodyne differential detection the beat-linewidth, which
is given by the sum of the linewidths of the signal laser and the LO, is the determin-
ing parameter for the electrical differential demodulation. Thus, when assuming the
same linewidths for the signal laser and the LO, the linewidth requirements on each
laser are doubled.
The receiver sensitivity penalties at BER=10−4 (with respect to the phase noise
free system) versus the ratio of the linewidth per laser and the data rate are illustrated
for all formats discussed here in Fig. 7.24. They were calculated by Monte Carlo
simulations, where the 3 dB bandwidths of the electrical third-order Bessel receiver
filters are assumed to be BW D M = 1.25 · r S and Bel = 0.75 · r S . For simplicity, the
same linewidths are assumed for the signal laser and the LO.
4 4
a b
Star
3 3
-4
-4
16QAM
Star
16QAM DBPSK DBPSK
2 2
DQPSK 16DPSK DQPSK
16DPSK
1 1
8DPSK
NRZ 8DPSK RZ
0 -6 -5 -4 -3 -2 -1
0 -6 -5 -4 -3 -2 -1
10 10 10 10 10 10 10 10 10 10 10 10
Linewidth per laser / data rate Linewidth per laser / data rate
Fig. 7.24 Receiver sensitivity penalties at BER=10−4 versus the linewidth-to-data-rate ratio for
homodyne phase diversity differential detection for NRZ (a) and RZ (b) pulse shapes
The laser phase noise becomes more critical with an increasing number of phase
states and with a reduction of the symbol rate, just like for direct detection. The
absolute values show that the requirements on the linewidth of each laser are twice
as high as for the signal laser for direct detection. Nevertheless, they are still rel-
atively relaxed, even for 8DPSK and Star 16QAM, where the linewidth per laser
must be in the range of some megahertz at 40 Gbit/s. Solely for NRZ-16DPSK, the
laser linewidth is required to be in the sub-MHz range at 40 Gbit/s (400 kHz for a
receiver sensitivity penalty smaller than 2 dB). Again it should be borne in mind
that the laser linewidth requirements become much more stringent when calling for
a BER of 10−9 instead of 10−4 due to the appearance of the BER floors.
7.2.3 Transmission Parameter Tolerances
Direct detection and homodyne phase diversity detection are both based on dif-
ferential demodulation. Thus, the same signals can be received and the same data
recovery techniques can be employed. When the same signals are propagated along
the fiber, the received signals are equally distorted by inter-symbol interference
caused by chromatic dispersion and SPM. In principle, differences in the receiver
structure can lead to a different impact of these deterministic signal distortions on
the system performance. As discussed in Sect. 7.2.1, based on the analysis per-
formed in [16] and [14], direct detection receivers with optical pre-amplification
have been demonstrated to be mathematically equivalent to homodyne phase diver-
sity receivers with differential detection under certain conditions. This equivalence
includes the sensitivity against deterministic signal distortions caused by chromatic
dispersion and SPM. The identical sensitivity against deterministic signal distortions
can be also understood from the description of the receivers given in Chap. 3. The in-
phase / quadrature photocurrents of the direct detection receiver and the homodyne
phase diversity receiver are given by (3.35) / (3.36) and (3.79) / (3.81), respectively,
the only difference between them being a constant factor.
Figure 7.25 shows some simulation results for systems with homodyne phase
diversity differential detection at 40 Gbit/s. These demonstrate the identical perfor-
mance of systems with direct detection and homodyne differential detection in rela-
tion to chromatic dispersion (a) and SPM (b), and can be compared with Fig. 7.10d
concerning the chromatic dispersion tolerances and with Fig. 7.14d with respect to
the SPM tolerances. Only minor differences arise caused by the different filters used
in each of the two receiver schemes.
Phase Diversity CD Tolerance RZ Phase Diversity SPM Tolerance RZ

4 4
a 2ASK, b 2ASK,
Star
DBPSK 16QAM
3 3 Int.
DQPSK
8DPSK 16DPSK
2 2
Star 8DPSK
16DPSK 16QAM
Int. Star DBPSK,
Star
1 16QAM 1 16QAM DQPSK
Phase Phase
RZ RZ
0 0
-320 -160 0 160 320 0 3 6 9 12 15
Dispersion [ps/nm] Fiber input power [dBm]
Fig. 7.25 Chromatic dispersion tolerances (a) and SPM tolerances (b) of various modulation for-
mats at 40 Gbit/s for homodyne phase diversity differential detection and RZ pulse shape.
By the way, the same compensators as for direct detection (see Sect. 7.1.6) can
be used to compensate for the mean nonlinear phase shift and to improve the poor
SPM tolerance of QAM formats.
7.2 Parameter Summary Homodyne Differential Detection 183
In Table 7.4, some performance parameters of 40 Gbit/s systems with homodyne

phase diversity differential detection are summarized, regarding all high-order mod-
ulation formats discussed in the previous sections and NRZ / RZ pulse shapes. The
given parameter values imply the application of the serial transmitter configurations,
as well as arg-decision at the receiver.
Table 7.4 Performance parameters of various modulation formats for homodyne phase diversity
differential detection at 40 Gbit/s for NRZ / RZ. The given parameter values are valid for systems
with serial transmitters, arg-decision and Bel = 0.75·r S . Ph: Phase, Int: Intensity, Comp: Nonlinear
phase shift compensation
Modulation Format 2ASK DBPSK DQPSK 8DPSK 16DPSK Star 16QAM
PR X,avg,d Bm [dBm] -38.7 / -41.5 / -39.8 / -35.4 / -27.2 / -31.4 /

@ BER=10−4 -40.5 -43.1 -41.8 -38.2 -33.5 -34.4
O S N Rd B [dB] 16.3 / 13.5 / 15.2 / 19.6 / 27.8 / 20.6 /
@ BER=10−4 14.5 11.9 13.2 16.8 21.5 17.6
1ν/r B -/ 8e-3 / 8.4e-4 / 1.2e-4 / 1e-5 / 7.6e-5 /
@ Pen. 2 dB (10−4 ) - 7.8e-3 8.6e-4 1.5e-4 2.9e-5 1.3e-4
1ν [MHz], 40 Gbit/s, -/ 320 / 33.6 / 5.0 / 0.4 / 3.0 /
@ Pen. 2 dB (10−4 ) - 312 34.4 6.0 1.2 5.2
Disp. Tol. [ps/nm] 87 / 86 / 124 / 124 / 150 / Ph 222 / >320
@ EOP 2 dB 39 43 156 316 >320 Int 168 / >320
SPM Tol. [dBm] 10.4 / >15 / 13.0 / 10.0 / 8.6 / Ph 6.4 / 5.8
@ EOP 2 dB 13.1 >15 >15 15.0 13.8 Comp 8.5 / 11.1
Int 11.8 / 12.8
Systems with homodyne differential detection are very similar in terms of per-
formance to systems with direct detection. This similarity becomes apparent after a
comparison of Table 7.4 with Table 7.2.
The values for PR X,avg,d Bm in Table 7.4 specify the receiver sensitivities at
BER=10−4 for shot-noise limited detection in the quantum limit with η = 1. These
are equivalent to the required receiver input powers for amplifier noise limited de-
tection with Fd B = 3 dB when a quadrature receiver is used. Therefore, the OSNR
requirements for obtaining a BER of 10−4 for amplifier noise limited detection,
given in the third row of Table 7.4, can be calculated from the receiver sensitivi-
ties for shot-noise limited detection by O S N Rd B = 55 dBm + PR X,avg,d Bm . In the
case of shot-noise limited detection, a 3 dB poorer receiver sensitivity compared
with direct detection can be observed for Star 16QAM caused by the optical power
splitting before photo-detection within the receiver illustrated in Fig. 3.17. This dis-
advantage disappears in the case of amplifier noise limited detection or when the
receiver structure shown in Fig. 3.16 is used instead.
The linewidth requirements listed in Table 7.4 correspond to a receiver sensitivity

penalty of 2 dB at BER=10−4 . They would be more stringent at BER=10−9 . The
linewidth requirements on each laser are doubled compared with direct detection
when the same linewidths are assumed for the signal laser and the LO, since the
effective phase noise taking effect on the electrical differential demodulation process
is determined by the beat-linewidth.
The transmission parameter tolerances are specified here by the eye opening
penalties and are almost identical with direct detection. Minor differences to sys-
tems with direct detection only arise here due to the use of different filters in each
of the two receiver schemes.
A performance comparison of two different 8DPSK system configurations with
homodyne phase diversity differential detection is presented in Table 7.5. The same
conclusions can be drawn as for direct detection. Arg-decision leads to a receiver
sensitivity improvement compared with IQ-decision. Systems with a parallel trans-
mitter outperform systems with a serial transmitter in relation to the noise perfor-
mances and the dispersion and SPM tolerances for NRZ.
Table 7.5 Comparison of different 8DPSK system configurations for homodyne phase diversity
differential detection at 40 Gbit/s for NRZ / RZ pulse shapes
Configuration 8DPSK Se./arg 8DPSK Pa./IQ
PR X,avg,d Bm [dBm] @ BER=10−4 -35.4 / -38.2 -35.9 / -37.4

O S N Rd B [dB] @ BER=10−4 19.6 / 16.8 19.1 / 17.6
1ν/r B @ Pen. 2 dB (10−4 ) 1.2e-4 / 1.5e-4 1.3e-4 / 1.5e-4

1ν [MHz], 40 Gbit/s, @ Pen. 2 dB (10−4 ) 5.0 / 6.0 5.2 / 6.0
Disp. Tol. [ps/nm] @ EOP 2 dB 124 / 316 246 / 307
SPM Tol. [dBm] @ EOP 2 dB 10.0 / 15.0 10.6 / 15.0
The benefits and drawbacks of particular modulation formats were already dis-
cussed for the direct detection receivers in Sect. 7.1.7 and are also true for homodyne
differential detection. Since systems with direct detection and homodyne differential
detection exhibit very similar performances, the question may arise of which system
should be used. At this point, general considerations comparing systems with direct
and coherent detection come into play. Systems with direct detection do not require
a LO and frequency synchronization. On the other hand, coherent detection offers
enhanced possibilities for electronic equalization and allows the selection of WDM
channels with high selectivity using a tunable LO.
7.3 Noise Performances for Homodyne Synchronous Detection 185
7.3 Systems with Homodyne Synchronous Detection
This section describes the performance of systems with homodyne synchronous

detection in the case that receivers with digital phase estimation according to
Sect. 3.5.3 are used. A wide range of modulation formats is regarded. Since arbi-
trary modulation formats can be received with those receivers, the Square 16QAM
and the Square 64QAM are additionally included in the discussion.
The receiver sensitivities at BER=10−4 for the shot-noise limited case and the
OSNR requirements for the amplifier noise limited case are presented in Sect. 7.3.1,
assuming an ideal carrier recovery. In the same section, a performance compari-
son is carried out for the different Square 16QAM transmitters which are described
in Sect. 2.6. A special focus is laid on the laser linewidth requirements for homo-
dyne phase estimation receivers. These are characterized by the receiver sensitivity
penalties at BER=10−4 and illustrated in Sect. 7.3.2, considering the phase estima-
tion schemes described in Sect. 3.5.3. Besides, the impact of the block length on
the phase noise requirements is shown. Subsequently, the influence of chromatic
dispersion and SPM on the performance of systems with homodyne synchronous
detection is detailed in Sect. 7.3.3 and Sect. 7.3.4, respectively. Section 7.3.5 demon-
strates that the SPM-induced nonlinear phase shift leads to drastic distortions of the
QAM signal constellations and highlights improvement of the SPM performance by
nonlinear phase shift compensation for Star 16QAM, Square 16QAM and Square
64QAM. In Sect. 7.3.6, systems with homodyne synchronous detection are finally
discussed and the most important performance parameters are summarized.
7.3.1 Noise Performance
The noise performance of homodyne receivers with digital phase estimation at a

data rate of r B = 40 Gbit/s is illustrated for a wide range of formats in Fig. 7.26.
1E-2 1E-2
a NRZ
b RZ
Star
1E-3 16QAM 1E-3
RR 1.8
BER
BER
8PSK Star
16PSK
QPSK 16QAM 16PSK
Square RR 1.8
1E-4 8PSK 1E-4 16QAM
Square Square
Square 64QAM QPSK 64QAM
16QAM
1E-5 1E-5
-46 -44 -42 -40 -38 -36 -34 -32 -46 -44 -42 -40 -38 -36 -34 -32
Received power [dBm] Received power [dBm]
Fig. 7.26 BER against received power for homodyne receivers with digital phase estimation for
NRZ (a) and RZ (b) at r B = 40 Gbit/s for Bel = 0.75 · r S and shot-noise limited detection in the
quantum limit with η = 1, when assuming serial transmitters and ideal carrier synchronization
The curves in Fig. 7.26 assume shot-noise limited detection in the quantum limit
with η = 1 and an ideal carrier synchronization, as well as the use of serial trans-
mitter configurations, arg-decision for MPSK / Star QAM formats and IQ-decision
for Square QAM formats.
One of the advantages of homodyne synchronous detection compared with di-
rect detection and homodyne differential detection is its superior noise performance.
From the theory, a performance gain of 0.5 dB / 2.3 dB / 3 dB / 3 dB can be expected
for synchronously detected BPSK / QPSK / 8PSK / 16PSK in comparison with dif-
ferentially detected DBPSK / DQPSK / 8DPSK / 16DPSK, respectively [6, 11, 17].
The comparison of Fig. 7.26 with the receiver sensitivities for direct detection and
homodyne differential detection listed in Table 7.2 and Table 7.4, respectively,
shows a performance difference of a little less than usually obtained when com-
paring synchronous and differential detection. The reason for this is the differential
coding used to resolve the phase ambiguity of the digital phase estimation. The “dif-
ferentially encoded MPSK” (DEMPSK)—as it is often referred to—can not achieve
the performance limits of the ideal synchronous MPSK and leads to a penalty of
about 0.5 dB [2, 17].
A further sensitivity degradation, which is not considered in Fig. 7.26, must be
borne in mind for including the phase estimation in the receiver performance. As
shown in Sect. 7.3.2, this penalty is caused if the laser phase noise necessitates an
averaging over small block lengths. In that case, the noise performance improvement
of homodyne synchronous phase estimation receivers compared with differential
detection receivers is relatively small. On the other hand, Square QAM formats
exhibit the best noise performance for a given number of bits per symbol and can
only be detected conveniently by synchronous detection. The noise performance
of RZ-Square 16QAM is in the same range as for RZ-8PSK. RZ-Square 64QAM
shows nearly the same performance as RZ-16PSK.
Moreover, it should be noted that Fig. 7.26 implies the use of electrical third-
order Bessel filters with a 3 dB bandwidth of Bel = 0.75 · r S , conjointly defined
for all formats. These are located in the in-phase and quadrature arms behind the
outputs of the balanced detectors. Especially for NRZ and some high-order formats,
performance can be improved by optimizing the receiver filter bandwidths.
The noise performances presented here are the results of numerical calculations
and characterize specific system configurations under consideration of the transmit-
ter setups, filtering, data recovery and signal shape. In Appendix B, they are com-
pared with results calculated using theoretical BER formulas, well known from the
literature [4, 11, 17] and reflecting the matched filtering case.
As described in Sect. 7.2.1, receiver sensitivities for the shot-noise limited case
can be easily scaled to the OSNR requirements for the amplifier noise limited
case by O S N Rd B = PR X,avg,d Bm + 55 dBm when a pre-amplifier noise figure of
Fd B = 3 dB and the use of a quadrature receiver are assumed. However, this rela-
tion becomes imprecise when an optical filter is included into the system, located
behind the optical pre-amplifier in front of the optical quadrature frontend. Figure
7.27 illustrates the OSNR requirements for amplifier noise limited detection if an
optical receiver filter with Bopt = 2.5 · r S is additionally considered.
7.3 Noise Performances for Homodyne Synchronous Detection 187
1E-2 1E-2
a NRZ
b RZ
1E-3 16PSK 1E-3

8PSK Star 8PSK Star
Square
BER
BER
16QAM 16QAM
RR 1.8 64QAM
RR 1.8
Square
16QAM Square
1E-4 1E-4 16QAM
16PSK
Square
QPSK 64QAM
QPSK
1E-5 1E-5
10 12 14 16 18 20 22 24 10 12 14 16 18 20 22 24
OSNR [dB] OSNR [dB]
Fig. 7.27 OSNR requirements at r B =40 Gbit/s for homodyne phase estimation receivers, am-
plifier noise limited detection, Bopt = 2.5 · r S and Bel = 0.75 · r S , assuming the use of serial
transmitters and an ideal carrier synchronization. a NRZ. b RZ.
From Fig. 7.26 and 7.27, a performance improvement can be observed for
most formats when employing an optical receiver filter with a 3 dB bandwidth of
Bopt = 2.5 · r S . The required OSNR for BER=10−4 for NRZ-8PSK without optical
filtering can be calculated by O S N Rd B = −38.1 dBm + 55 dBm=16.9 dB and is
decreased to 16.1 dB by the optical filtering, for instance. However, the OSNR per-
formance degrades when the optical filter bandwidth is not appropriately adapted to
the modulation format used, as shown here for Square 64QAM (20.0 dB / 18.6 dB
without optical filtering in comparison with 21.7 dB / 20.7 dB with optical filtering
for NRZ / RZ). Ideally, the optical and electrical filter bandwidths should be opti-
mized for each format individually to obtain the best OSNR performances, in the
same way as discussed for the direct detection receivers in Sect. 7.1.2.
Optimum Star 16QAM Ring Ratio / Square 16QAM Transmitter Comparison
The optimum ring ratio for Star 16QAM in the case of homodyne synchronous de-
tection corresponds to a value of about 1.8. This becomes apparent from Fig. 7.28a.
In contrast to direct detection, where the Star 16QAM constellation comprises three
different rings after differential detection and the noise power depends on the sym-
bol amplitude, the optimum ring ratio can be easily understood for coherent syn-
chronous detection. If the amplitude of the outer circle is normalized to one, the
amplitude of the inner circle is given by 1/R R, so the Euclidean distance between
the two rings—relevant for the error probability of amplitude decision—is given by
d1 = 1 − 1/R R. By adopting simple geometry, the Euclidean distance between two
p √
symbols on the inner ring can be shown to be d2 = 2 − 2/R R. Best overall
BER performance is obtained for equal error probabilities for phase and amplitude
decision, occurring for d1 = d2 . From this condition, an optimum ring ratio of
R R ≈ 1.77 can easily be derived.
Figure 7.28b illustrates the impact of the transmitter configuration on the noise
performance for Square 16QAM systems. All transmitters exhibit nearly the same
Optimal Star 16QAM Ring Ratio TX Comparison Square 16QAM

1E-2 1E-2
a b Serial +
Enhanced IQ +
Tandem QPSK,
Star 16QAM NRZ
1E-3 @ PRX = -37dBm 1E-3 NRZ
BER
BER
Conv. IQ, NRZ
1E-4 1E-4
All TX, RZ
Star 16QAM RZ
@ PRX = -39dBm
Se. / arg
1E-5 1E-5
1,5 1,6 1,7 1,8 1,9 2,0 2,1 2,2 -43 -42 -41 -40 -39 -38 -37 -36 -35
Ring ratio Received Power [dBm]
Fig. 7.28 a Optimal Star 16QAM ring ratio for homodyne synchronous detection. b Impact of the
transmitter configuration on the noise performance for Square 16QAM.
performance for RZ. The conventional IQ transmitter shows a performance im-

proved by 1 dB compared with the serial / enhanced IQ / Tandem-QPSK transmitters
for NRZ. This is attributable primarily to the higher peak-to-average power ratio.
This section discusses the laser linewidth requirements of systems with homo-
dyne phase estimation receivers for shot-noise limited back-to-back transmission.
Thereby, the feed forward (FF) M-th power block scheme and the closed loop (CL)
M-th power block scheme (both described in Sect. 3.5.3) are considered, as well as
different averaging schemes denoted as field averaging, normalized field averaging
and argument averaging and defined by (3.96), (3.99) and (3.103), respectively. As
a quality criterion, the receiver sensitivity penalty at BER=10−4 is used, where the
receiver input powers required for the phase noise free systems without phase esti-
mation are taken as power references. The linewidth requirements are specified by
the linewidth per laser and assume the same linewidths for the signal laser and the
LO, so that the required beat-linewidths 1νe f f = 1νs + 1νlo can be easily ob-
tained by doubling the given values. Bessel low-pass filters with a 3 dB bandwidth
of Bel = 0.75 · r S are located in the in-phase and quadrature arms of the receiver.
Comparison of Different Phase Estimation Schemes for NRZ-QPSK
In Fig. 7.29, different phase estimation schemes are compared for NRZ-QPSK and
a block length of N = 8. The curves for the closed loop scheme are shown for
the limit of zero processing delay with D = 1. It can be observed that normalized
field averaging, where the amplitudes of the phasors are normalized to one before
being summed, shows a slightly better performance than field averaging, since the
7.3 Linewidth Requirements for Homodyne Synchronous Detection 189
Fig. 7.29 Comparison of 3

different phase estimation CL no proc. delay
concepts for NRZ-QPSK at Arg averaging
-4
r B = 40 Gbit/s, assuming 2
a block length of N = 8 FF + CL no proc. delay
Field averaging
and zero processing delay
(D = 1) 1
FF + CL no proc. delay
Norm. field averaging
0
0 2 4 6 8 10
Linewidth per laser [MHz]
phase error is estimated more accurately. It can be expected that this performance
difference is even higher when the signals are additionally distorted by inter-symbol
interference caused by fiber propagation effects. Compared with field averaging,
argument averaging exhibits an improved performance for smaller linewidths. On
the other hand, the receiver performance degrades faster for higher linewidths. For
zero processing delay, the CL scheme performs identically to the FF scheme when
the same averaging mechanism is employed.
When considering practical implementation, however, the accumulated phase er-
ror (which should be available from the last block for the calculation of the phase
error of the current block in the ideal case) is only available after a processing delay
of D block lengths for the CL scheme. The processing delay has an effect compa-
rable to the loop delay in optical phase locked loops and leads to a drastic reduction
of the phase noise tolerance. For NRZ-QPSK this is shown for field averaging in
Fig. 7.30a and for argument averaging in Fig. 7.30b.
3 3
a CL, field averaging
b CL, arg averaging
Proc. delay
8ns
-4
-4
Proc. delay
2 8ns
2
Proc. delay
Proc. delay 4ns
Proc. delay 2ns
4ns
1 1 Proc. delay
2ns
0 0
0 0,2 0,4 0,6 0,8 1,0 1,2 0 0,2 0,4 0,6 0,8 1,0 1,2
Linewidth per laser [MHz] Linewidth per laser [MHz]
Fig. 7.30 Degradation of the phase noise tolerance due to the processing delay for NRZ-QPSK at
r B = 40 Gbit/s for the M-th power CL block scheme, a block length of N = 8 and processing
delays of 2 ns (D = 5), 4 ns (D = 10) and 8 ns (D = 20). a Field averaging. b Argument averaging.
At 40 Gbit/s, the duration of a QPSK symbol is TS = 50 ps and the block dura-

tion is 400 ps for N = 8, so that processing delays of 2 ns / 4 ns / 8 ns correspond to
D = 5 / 10 / 20 blocks, respectively. The crucial influence of the processing de-

lay on the performance of the CL scheme is clarified by comparing the linewidth
requirements for a receiver sensitivity penalty of 2 dB at BER=10−4 for zero pro-
cessing delay, shown in Fig. 7.29, with the diagrams of Fig. 7.30. For zero process-
ing delay, the linewidths required for NRZ-QPSK are about 6 MHz and 10 MHz
for argument averaging and field averaging, respectively. When taking loop delay
into account, the linewidth requirements increase drastically to 1070 kHz / 470 kHz /
270 kHz for field averaging and to 1320 kHz / 550 kHz / 260 kHz for argument av-
eraging for processing delays of 2 ns / 4 ns / 8 ns, respectively. For comparison, the
linewidths required for a receiver sensitivity penalty of 1 dB at BER=10−10 of
QPSK receivers with OPLL, derived in [9] and listed in Table 3.1, are 3.86 MHz
for zero loop delay and 143 kHz / 72 kHz / 36 kHz for loop delays of 2 ns / 4 ns / 8 ns,
respectively. When considering the higher BER reference of 10−10 and the lower
penalty reference of 1 dB, the requirements are in a comparable range. Therefore,
higher phase noise tolerances—compared with receivers with OPLL—can only be
achieved using digital phase estimation receivers when feed forward schemes are
used for phase estimation or processing delay can be kept small.
Linewidth Requirements for the Feed Forward M-th Power Block Scheme
The contour plots in Fig. 7.31 show the receiver sensitivity penalties at BER=10−4
with respect to the linewidth per laser and the block length N at 40 Gbit/s for the
feed forward M-th power block scheme, regarding NRZ and RZ pulse shapes and a
wide range of modulation formats. Field averaging according to (3.96) is assumed
to be used for the MPSK formats, whereas normalized field averaging according
to (3.99) is applied to Star 16QAM because phasors with different amplitudes are
averaged. As described in Sect. 3.5.3, a class partitioning is performed for the Square
QAM formats before the phase error estimate is calculated by (3.101). All the Class
I symbols are used for phase error estimation in the case of Square 16QAM, but
only the eight symbols on the inner and outer rings are used for the calculation of
the phase error estimate in the case of Square 64QAM. The block lengths are varied
in steps of 2n (n = 0, 1, 2, ..).
Obviously, a certain minimum block length is necessary to obtain a satisfac-
tory performance. For the PSK formats, block lengths of at least four are required
for receiver sensitivity penalties smaller than 2 dB, even for zero linewidth. When
implying the phase estimation scheme with class partitioning considered here, sig-
nificantly higher block lengths are needed for Square QAM formats, since only a
part of the symbols of each block are used for determination of the phase error es-
timate. For block lengths of 32 and 512, the penalty is still greater than 2 dB for
Square 16QAM and Square 64QAM, respectively. For zero linewidth, the receiver
performance can be improved almost up to the limit of ideal carrier recovery when
increasing the block length towards infinity. With increasing linewidth, the optimum
block lengths are getting smaller because they are determined by a trade-off between
the shot-noise (or amplifier noise) and the phase noise.
QPSK Se. TX, NRZ 8PSK Se. / arg, NRZ 16PSK Se. / arg, NRZ
32 32 32
28 28 4 28
3.5 5
2.5
24
Block Length N
Block Length N
24 24
Block Length N
3 4.5
20 2.5 4
20 20 2
1.5 2 3.5
16 0.5 3
1 16 0.5 1 1.5 16 1 1.5
3
2.5
12 2 12 12
1.5 2.5
8 1.5 8 1 1.5 8 2
1 1.5 2
1.5 2.5
2 3
4 2 4 2.5 4
2.5 3 5
1 1 1
0 1 2 3 4 5 6 7 8 9 10 0 0.2 0.4 0.6 0.8 1 0 50 100 150 200
Linewidth per laser [MHz] Linewidth per laser [MHz] Linewidth per laser [kHz]
QPSK Se. TX, RZ 8PSK Se. / arg, RZ 16PSK Se. / arg, RZ

32 32 32
4
4.5
28 5 28 3.5 28
4.5 4
3
24 24 24 3.5
Block Length N
Block Length N
Block Length N
3.5 2.5 3
4
20 1 2.5 20 1 20
3 2 2.5
16 0.5 1.5 16 0.5 16 1 1.5 2
1.5
12 12 12
8 1.5 2 8 1 8 1.5
1 1.5 1.5
1.5
4 2 4 2 4 2
2.5 2.5 3
1 1 1
0 1 2 3 4 5 6 7 8 9 10 0 0.2 0.4 0.6 0.8 1 0 50 100 150 200
Linewidth per laser [MHz] Linewidth per laser [MHz] Linewidth per laser [kHz]
Star 16QAM Se. / arg, NRZ Square 16QAM Conv. IQ, NRZ Square 64QAM Conv. IQ, NRZ
32 256 4096
4.5
0.5
28 4
5
3.5
192 3072 1
Block Length N
Block Length N
24
Block Length N
3 4.5
5
2.5 4
20
2 2.5 4.5 3.5
128 1 1.5 3 2048
16 1 1.5 3.5 4 3
12 2.5
64 1.5 2 2.5 3 1024 1.5 2

8 0.5 2
1 2.5 3
3 3.5 4
4 1.5
2 5 5
1 1 1
0 0.2 0.4 0.6 0.8 1 0 50 100 150 200 0 500 1000 1500 2000
Linewidth per laser [MHz] Linewidth per laser [kHz] Linewidth per laser [Hz]
Star 16QAM Se. / arg, RZ Square 16QAM Conv. IQ, RZ Square 64QAM Conv. IQ, RZ
32 256 4096
4.5 0.5 5
28 4 0.5
3.5 4.5
24 192 3072 1
Block Length N
Block Length N
3
Block Length N
4
2.5 4 5
20 3
1 2 3.5
128 1 2.5 3.5 4.5 2048
16 0.5 1.5
1.5 3
12
64 1.5 2 2.5 1024 2 2.5 3
8 0.5 1 3 2
2.5 3
1 3 3.5
4 1.5 4
2 5 5
1 1 1
0 0.2 0.4 0.6 0.8 1 0 50 100 150 200 0 500 1000 1500 2000
Linewidth per laser [MHz] Linewidth per laser [kHz] Linewidth per laser [Hz]
Fig. 7.31 Receiver sensitivity penalties at BER=10−4 versus the linewidth per laser and the block
length N for homodyne phase estimation receivers employing the feed forward M-th power block
scheme at r B = 40 Gbit/s, for various modulation formats and NRZ and RZ pulse shapes
For accepting a receiver sensitivity penalty of 2 dB at BER=10−4 and searching

for the block length with the greatest linewidth tolerable, the optimal block lengths
in steps of 2n (n = 0, 1, 2, ..) are N = 8 for the PSK formats and Star 16QAM,
N = 64 for Square 16QAM and N = 1024 for Square 64QAM. For a clearer
illustration, the receiver sensitivity penalties at BER=10−4 versus the linewidth per
laser are illustrated for selected block lengths in Fig. 7.32, exemplarily for NRZ-
8PSK (a) and RZ-Square 16QAM (b).
NRZ-8PSK linewidth requirements RZ-Square 16QAM linewidth requirements

4 4

N=16 b N=128
3 3
-4
-4
N=32
2 N=4 2 N=64
N=8
1 1
FF, field averaging FF, norm. field averaging

0 0
0,0 0,2 0,4 0,6 0,8 1,0 1,2 1,4 1,6 1,8 0 50 100 150 200 250
Linewidth per laser [MHz] Linewidth per laser [kHz]
Fig. 7.32 Receiver sensitivity penalties at BER=10−4 versus the linewidth per laser for selected
block lengths N at r B = 40 Gbit/s, when employing the FF M-th power block scheme, illustrated
for NRZ-8PSK (a) and RZ-Square 16QAM (b)
Higher block lengths should be used when the linewidth requirements for a par-
ticular modulation format can easily be fulfilled. However, smaller block lengths
are necessary to achieve a better system performance when the laser phase noise
becomes a critical parameter. For RZ-Square 16QAM, for instance, a receiver sen-
sitivity penalty of about 1 dB can be obtained with lasers possessing a linewidth in
the 50 kHz range by choosing N = 128. However, if the laser linewidths are around
100 kHz, a block length of N = 64 is the appropriate choice to obtain the best
performance. As already discussed in Sect. 7.3.1, the penalty caused by the phase
estimation degrades the overall receiver performance and reduces the performance
gain of synchronous detection compared with differential detection.
As mentioned above, the optimal block lengths are specified here in steps of just
2n (n = 0, 1, 2, ..) and the term “optimal block length” is defined here by the block
length allowing for the greatest tolerable linewidth at a receiver sensitivity penalty
of 2 dB for the phase estimation. In this context, the optimal block lengths are N = 8
for the MPSK formats and for Star 16QAM, N = 64 for Square 16QAM and N =
1024 for Square 64QAM. The receiver sensitivity penalties versus the linewidth
per laser to data rate ratio for optimal block lengths—calculated with Monte Carlo
simulations— are summarized for all modulation formats discussed here in Fig. 7.33
for NRZ (a) and RZ (b) pulse shapes.
The comparison of the modulation formats shows that the linewidth require-
ments increase with an increasing number of phase states. In spite of having a
3 3
a

QPSK b QPSK
Square Square
64QAM 16QAM
-4
-4
2 2
16PSK
Square
64QAM
Square
16QAM
1 1 16PSK
8PSK
8PSK
Star Star
16QAM NRZ, FF field averaging 16QAM RZ, FF field averaging
0 -8 -7 -6 -5 -4 -3
0 -8 -7 -6 -5 -4 -3
10 10 10 10 10 10 10 10 10 10 10 10
Linewidth per laser / data rate Linewidth per laser / data rate
Fig. 7.33 Receiver sensitivity penalties at BER=10−4 versus the linewidth-to-data-rate ratio for
homodyne receivers with feed forward M-th power digital phase estimation and field averaging
(PSK) / normalized field averaging (Star 16QAM, Square QAM), for optimal block lengths and
NRZ (a) and RZ (b) pulse shapes
smaller symbol rate, the Star 16QAM exhibits the same performance as the 8PSK
in Fig. 7.33 because normalized field averaging is assumed to be used here only for
QAM formats, leading to a slight performance improvement compared with field
averaging as illustrated in Fig. 7.29 for QPSK. When the feed forward phase estima-
tion algorithm based on class partitioning described in Sect. 3.5.3 is implemented,
the Square QAM formats require lower linewidths for a certain number of bits per
symbol due to the higher block lengths necessary. For instance, the phase distances
for 16PSK (22.5◦ ) are smaller than the smallest phase distance for Square 16QAM
(26.6◦ ), but the RZ-Square 16QAM is nevertheless more phase noise sensitive than
the RZ-16PSK, since larger block lengths are needed.
The previous paragraphs have shown that—for current data rates—the laser
phase noise becomes a critical parameter for 16PSK, Square 16QAM and Square
64QAM, even when employing the feed forward M-th power block scheme which
is not impaired by processing delay. Square QAM formats in particular require low
linewidths when only some of the symbols are used for calculation of the phase error
estimate. The laser linewidths must be in the range of 240 kHz / 120 kHz / 1 kHz for
RZ-16PSK / RZ-Square 16QAM / RZ-Square 64QAM at 40 Gbit/s. These require-
ments can be fulfilled with high-spec lasers for the lab, but not with lasers currently
available for commercial use. The Star 16QAM features the most relaxed phase
noise requirements of the 16-level formats. These are around 1 MHz at 40 Gbit/s.
To enable a commercial application of modulation formats such as the Square
16QAM or the 16PSK, the development of commercial lasers with linewidths in the
range of 100 kHz is necessary or else more sophisticated phase estimation schemes
must be adopted. The linewidth requirements for Square QAM formats can possi-
bly be relaxed by the application of decision directed techniques [5], or by the use
of enhanced feed forward algorithms which allow to incorporate all constellation
points into the calculation of the phase error estimate [8, 12].
Moreover, it should be noted that the optimal block lengths shown here are valid
for the back-to-back case, considering only the trade-off between shot-noise and
phase noise effects. In practical transmission systems, fiber degradation effects such
as XPM necessitate a further optimization of phase estimation block lengths.
7.3.3 Chromatic Dispersion Tolerances
Next, the chromatic dispersion tolerances of optical fiber transmission systems with
homodyne phase estimation receivers are discussed, provided that no electrical dis-
persion compensation is implemented at the receiver. They are specified here by the
receiver sensitivity penalties at BER=10−4 , assuming digital phase estimation with
the optimum block lengths for back-to-back highlighted in Sect. 7.3.2, as well as by
the eye opening penalties for ideal carrier synchronization. In Fig. 7.34, chromatic
dispersion tolerances at r B = 40 Gbit/s are shown for a wide range of formats. The
diagrams imply the use of serial structures at the transmitter and the use of electrical
Bessel filters with 3 dB bandwidths Bel = 0.75 · r S at the receiver.
4 4
a b
-4
3 3
-4
QPSK, QPSK
8PSK,
16PSK, 16PSK,
2 Square 2 Square
16QAM Star 16QAM,
16QAM Star 8PSK
Square 16QAM
1 1 64QAM
Square
64QAM
0 NRZ 0 RZ
-320 -160 0 160 320 -320 -160 0 160 320

4 4
c Square d
16QAM
3 3
QPSK 8PSK, QPSK
16PSK 16PSK,
2 2 Square 8PSK
Star 16QAM,
Square Star
16QAM,
64QAM 16QAM
Int.+Ph.
1 Square
1
64QAM
NRZ RZ
0 0
-320 -160 0 160 320 -320 -160 0 160 320
Fig. 7.34 Chromatic dispersion tolerances for systems with homodyne detection and digital phase
estimation at r B = 40 Gbit/s when using the serial transmitter structures; The two upper diagrams
show the receiver sensitivity penalties at BER=10−4 against the chromatic dispersion for NRZ (a)
and RZ (b) when employing digital phase estimation with the optimal block lengths. The bottom
diagrams (c, d) illustrate the eye opening penalties for ideal carrier synchronization.
7.3 Dispersion Tolerances for Homodyne Synchronous Detection 195
The receiver sensitivity penalties at BER=10−4 and the eye opening penalties
show very similar results, so the EOP can be used as a good first performance crite-
rion when MC simulations with high computational effort must be avoided. A com-
parison of the eye opening penalties discovered here with the eye opening penalties
for differential detection listed in Table 7.2 and Table 7.4 indicates slightly higher
dispersion tolerances for homodyne synchronous detection.
In the same way as for differential detection, the dispersion tolerances for RZ
increase with the order of the modulation format due to the smaller symbol rates
and reduced spectral widths. In the case of NRZ, the detrimental chirp characteris-
tics neutralize the advantage of the reduced spectral widths for the high-order for-
mats, especially when serial transmitter configurations are employed. RZ signals are
less distorted by chromatic dispersion than NRZ signals. This difference becomes
significant for high-order formats in particular. For Square 64QAM, the curves in
Fig. 7.34 are shown for the conventional IQ transmitter and the performance would
be worse for NRZ if a serial transmitter structure was used.
Influence of the Transmitter Configuration for Square 16QAM
The impact of the configuration of the transmitter on the chromatic dispersion tol-
erance is illustrated in Fig. 7.35 for some of the Square 16QAM transmitters intro-
duced in Sect. 2.6.
TX comparison Square 16QAM

4
Serial +
3
-4
Enhanced IQ +
Tandem QPSK,
NRZ
2
Conv. IQ, NRZ
1
Fig. 7.35 Impact of the trans- All TX, RZ
mitter configuration on the
0
chromatic dispersion toler-
ance for Square 16QAM at -320 -160 0 160 320
r B = 40 Gbit/s Dispersion [ps/nm]
All transmitter types show nearly the same performance for RZ. For NRZ, the
usage of the conventional IQ transmitter significantly improves the dispersion toler-
ance in comparison with the serial transmitter, the enhanced IQ transmitter and the
Tandem-QPSK transmitter. This could be expected as a result of the discussion of
signal properties in Sect. 2.6.7. The product of the chirp and the normalized inten-
sity was shown to be small there for the conventional IQ transmitter. For the other
configurations, chirp appears simultaneously with high power levels and leads to a
higher influence of chromatic dispersion.
Vivid Illustration of the Chromatic Dispersion on Signal Shape
A vivid illustration of the impact of the chromatic dispersion on the signal quality
of optical high-order modulation signals is provided in Fig. 7.36 and Fig. 7.37.
Dispersion 0ps/nm Dispersion 320ps/nm
Original Ideal IQ-Plot IQ-Plot In-phase eye

IQ-Plot in-phase eye before PE after PE+LPF after PE+LPF
QPSK
Pa./IQ NRZ
QPSK
Se./IQ NRZ
QPSK
Se./IQ RZ
8PSK
Se./arg NRZ
8PSK
Se./arg RZ
16PSK
Se./arg NRZ
16PSK
Se./arg RZ
Fig. 7.36 Impact of chromatic dispersion on PSK signals at r B = 40 Gbit/s, representation with
time scales adapted to the particular symbol rates; PE: phase estimation, LPF: low-pass filter
7.3 Dispersion Tolerances for Homodyne Synchronous Detection 197
Dispersion 0ps/nm Dispersion 320ps/nm
Original Ideal IQ-Plot IQ-Plot In-phase eye

IQ-Plot in-phase eye before PE after PE+LPF after PE+LPF
Star 16QAM
Se./arg NRZ
Star 16QAM
Se./arg RZ
Square 16QAM
Se. TX NRZ
Square 16QAM
Conv. IQ TX NRZ
Square 16QAM
Enh. IQ TX NRZ
Square 16QAM
Enh. IQ TX RZ
Square 64QAM
Conv. IQ TX NRZ
Square 64QAM
Conv. IQ TX RZ
Fig. 7.37 Impact of chromatic dispersion on QAM signals at r B = 40 Gbit/s, representation with
time scales adapted to the particular symbol rates; PE: phase estimation, LPF: low-pass filter
The undistorted electrical eyes and IQ diagrams of various PSK and QAM signals
for r B = 40 Gbit/s are depicted on the left side in Fig. 7.36 and 7.37, respectively,
and compared to the distorted eyes and IQ diagrams after an accumulated dispersion
of 320 ps (three diagrams on the right in each figure). Thereby, the time scales of the
eye diagrams are adapted to the particular symbol rates. The distorted IQ diagrams
are shown before phase estimation where they are partly rotated by a constant phase
offset, as well as after phase estimation and electrical low-pass filtering with a third-
order Bessel filter with a 3 dB bandwidth of Bel = 0.75 · r S . The averaging of
the phase estimation is performed over the entire number of symbols, so that the
constellation diagrams are rotated back only by the constant phase offset.
The distorted eye diagrams after 320 ps accumulated dispersion are more clearly
opened for the high-order formats and for RZ pulse shape. Moreover, it can be ob-
served from the IQ diagrams for QPSK and Square 16QAM that the specific chirp
characteristics of different transmitters lead to individually different signal distor-
tions. The parallel transmitter, the conventional IQ transmitter and all RZ transmit-
ters tend to result in equally distributed distortions of the symbols. In contrast, the
intensity of the distortions is symbol dependent for systems employing the serial
transmitter structure and NRZ pulse shape.
7.3.4 Self Phase Modulation Tolerances
The SPM tolerances of high-order modulation formats in systems with homodyne

synchronous detection and digital phase estimation receivers can be identified using
the same system setup as for direct detection. As shown in Fig. 7.13, the optical
signals are transmitted over a dispersive and nonlinear fiber link with a length of
l = 80 km. The data rate is r B = 40 Gbit/s and the 3 dB bandwidth of the electrical
receiver filters is assumed to be Bel = 0.75 · r S . The SMF parameters are specified
by Dλ = 16 ps/(nm·km), Sλ = 0 and n 2 = 2.6 · 10−20 m2 /W. Chromatic dispersion
is compensated for by ideal dispersion compensation behind the link. The average
fiber input power is varied and the receiver sensitivity penalties at BER=10−4 and
the eye opening penalties are determined. In Fig. 7.38, the receiver sensitivity penal-
ties (a, b) and the eye opening penalties (c, d) versus fiber input power are shown for
the case of using the serial configurations on the transmitter side, except for Square
64QAM where the use of the conventional IQ transmitter is assumed.
The same principal tendencies can be observed as for differential detection. The
SPM tolerance of PSK formats decreases with increasing modulation format order
and thus smaller phase distances. Signals with RZ pulse shape tolerate higher fiber
input powers than NRZ signals due to the more beneficial chirp characteristics. All
QAM formats show very poor SPM performance, since symbols on different circles
experience different nonlinear phase shifts. This leads to drastic deformations of the
signal constellations as further illustrated in Sect. 7.3.5.
7.3 Nonlinear Phase Shift Compensation 199
4 4

a
b Square
Square
3 16QAM 3 16QAM
-4
16PSK
-4
Square
64QAM 8PSK
Star
2 2 16QAM 8PSK
Star Square
QPSK 64QAM
16QAM
1 1 QPSK
0 0
16PSK
NRZ RZ
-1 -1
-6 -3 0 3 6 9 12 15 -6 -3 0 3 6 9 12 15
4 4

c Square Square 16PSK d Star

16QAM, 64QAM Square 16QAM,
3 Star 3 16QAM Int.
16QAM,
8PSK
Phase Star
16QAM, 16PSK
2 Star 2 Phase
16QAM,
Int. Square 8PSK
64QAM
1 QPSK 1 QPSK
NRZ RZ
0 0
-6 -3 0 3 6 9 12 15 -6 -3 0 3 6 9 12 15
Fig. 7.38 SPM tolerances of various modulation formats for homodyne detection with digital
phase estimation at 40 Gbit/s when using the serial transmitter structures; the two upper diagrams
show the receiver sensitivity penalties at BER=10−4 against the fiber input power for NRZ (a) and
RZ (b). The bottom diagrams (c, d) illustrate the eye opening penalties.
7.3.5 Nonlinear Phase Shift Compensation
During fiber propagation, an optical signal experiences an intensity dependent SPM-

induced nonlinear phase shift as described in Sect. 6.3 and discussed for direct de-
tection receivers in Sect. 7.1.6. QAM symbols have different intensity levels. There-
fore, the intensity dependent nonlinear phase shift causes drastic deformations of
the received QAM signal constellations. This effect is shown in Fig. 7.39 for Star
16QAM and Square 16QAM after transmission over a dispersive nonlinear fiber
with a length of l = 80 km and Dλ = 16 ps/(nm·km) and γ = 1.31 W−1 km−1 ,
where it is assumed that the chromatic dispersion is compensated for ideally behind
the link and the average input power launched into the fiber is PT X,avg = 8 dBm.
It can be observed that symbols with different power levels undergo different
degrees of phase rotation. The resulting distortions of the signal constellations can
not be compensated for by phase estimation solely, which just rotates back the en-
tire constellation (or more precisely: all symbols within a block) by the phase error
as shown in Fig. 7.39b, but must be compensated for by an additional nonlinear
phase shift compensator to enable further use of simple decision techniques. For
Square 16QAM, for instance, the optimal decision boundaries are spiral-like when
Star 16QAM Star 16QAM Square 16QAM Square 16QAM

NRZ RZ NRZ RZ
a
w/o compensation
w/o PE
b
w/o compensation
w. PE
c
w. compensation
w. PE
Fig. 7.39 Deformation of the signal constellations of Star 16QAM and Square 16QAM caused by
the SPM-induced nonlinear phase shift, assuming a fiber length of l = 80 km and an average fiber
input power of PT X,avg = 8 dBm. a Without (w/o) phase estimation (PE) and w/o nonlinear phase
shift compensation. b With (w.) PE and w/o compensation. c With PE and compensation.
not employing compensation, whereas the usual straight-line decision boundaries

can be used after nonlinear phase shift compensation. Further investigation con-
cerning QAM signal detection impaired by SPM—also under consideration of the
nonlinear phase noise—is presented in [7].
The distortions caused by the SPM-induced nonlinear phase shift can be com-
pensated for using the optical compensator depicted in Fig. 7.18a, which can be
placed in front of the coherent receiver. The obtainable improvements of the SPM
tolerance for Star 16QAM and Square 64QAM are illustrated in Fig. 7.40.
Star 16QAM, serial TX, arg-decision Square 64QAM, conventional IQ TX

4 4
NRZ
a b
uncompen- RZ
NRZ
3 3
-4
uncompen-
-4
sated uncompen-
sated
RZ sated NRZ
compensated compensated
RZ
2 uncompen- 2
sated NRZ
RZ
compen-
compensated
sated
1 1
0 0
0 3 6 9 12 15 -6 -3 0 3 6 9 12 15
Fig. 7.40 Enhancement of the SPM tolerance by nonlinear phase shift compensation for Star
16QAM (a) and Square 64QAM (b), l = 80 km, r B = 40 Gbit/s
7.3 Nonlinear Phase Shift Compensation 201
As regards the single-span system configuration discussed here, it can be seen

in Fig. 7.40 that SPM tolerance can be greatly enhanced for both formats shown.
With compensation, the performance for NRZ-Star 16QAM is nearly in the same
range as for NRZ-8PSK and the SPM tolerance can be increased by more than 5 dB
for RZ-Star 16QAM. In the case of Square 64QAM, the performance gain is even
higher—almost 10 dB for NRZ and RZ pulse shapes.
Discussion of the Impact of the Transmitter Configuration for Square 16QAM
A comparison of the SPM tolerances of Square 16QAM systems comprising differ-

ent transmitter structures is shown in Fig. 7.41.
TX comparison Square 16QAM

4
All TX, NRZ All TX
uncompen- RZ
3
-4
sated compensated
All TX, RZ
2 uncompen-
Conv. IQ
sated
Serial + NRZ
Fig. 7.41 SPM tolerances for enhanced IQ compensated
Square 16QAM when using 1 NRZ
compensated
different transmitter types,
with and without nonlinear 0
phase shift compensation, 0 3 6 9 12 15
l = 80 km, r B = 40 Gbit/s Fiber input power [dBm]
The SPM tolerance of all Square 16QAM system configurations is very poor
without the compensation. When using the compensation, a performance gain of
more than 6 dB can be achieved for RZ and the conventional IQ transmitter shows a
higher robustness against SPM than the other transmitters for NRZ.
The nonlinear phase shift compensation scheme discussed here turns out to be
quite effective for the single-span system configuration examined in this chapter.
Later on in Chap. 8, its application to multi-span transmission systems is briefly
shown. It should also be borne in mind that the nonlinear phase noise (see Sect. 6.4)
must be considered for multi-span transmission.
Alternatively or additionally to the compensation scheme described here, signal
distortions through the SPM-induced mean nonlinear phase shift can potentially be
reduced by means of digital equalization in the electrical part of the receiver, for
instance using decision directed adaptive equalization schemes, or by applying pre-
distortion techniques on the transmitter side.
Some performance parameters of 40 Gbit/s systems with homodyne synchronous

detection and digital phase estimation receivers are summarized in Table 7.6, re-
garding all high-order modulation discussed before and NRZ / RZ pulse shapes. The
given parameter values are valid for systems using serial transmitter configurations
(except for Square 64QAM where the use of the conventional IQ transmitter is as-
sumed), arg-decision for the PSK formats and Star 16QAM, IQ-decision for Square
QAM and electrical Bessel filters with Bel = 0.75·r S in the in-phase and quadrature
branches of the receiver.
Table 7.6 Performance parameters of 40 Gbit/s systems with homodyne synchronous detection
and digital phase estimation receivers, regarding various modulation formats and NRZ / RZ pulse
shapes. The given parameter values are valid for systems with serial transmitter structures (except
for Square 64QAM where the use of the conventional IQ transmitter is assumed), arg-decision for
the PSK formats and Star 16QAM, IQ-decision for Square QAM and Bel = 0.75 · r S .
Modulation Format QPSK 8PSK 16PSK St.16QAM Sq.16QAM Sq.64QAM
PR X,avg,d Bm [dBm] -42.0 / -38.1 / -32.6 / -36.6 / -37.8 / -34.9 /
@ BER=10−4 -43.8 -40.4 -36.1 -38.9 -40.0 -36.3
O S N Rd B [dB], BER=10−4 13.0 / 16.9 / 22.4 / 18.4 / 17.2 / 20.1 /
w/o optical filter 11.2 14.6 18.9 16.1 15.0 18.7
O S N Rd B [dB], BER=10−4 12.5 / 16.1 / 22.4 / 17.8 / 16.6 / 21.7 /
opt. filter, Bopt = 2.5 · r S 11.0 14.2 19.0 15.8 14.8 20.7
1ν/r B , FF 2.5e-4 / 2.8e-5 / 2.7e-6 / 3.5e-5 / 2.9e-6 / 3.7e-8 /

@ Pen. 2 dB (10−4 ) 2.5e-4 3.7e-5 6.2e-6 3.8e-5 2.9e-6 2.8e-8
1ν [kHz], 40 Gbit/s, FF 10000 / 1120 / 108 / 1400 / 116 / 1.5 /
@ Pen. 2 dB (10−4 ) 10000 1480 248 1520 116 1.1
Optimal N 8/ 8/ 8/ 8/ 64 / 1024 /
(steps 2n ) 8 8 8 8 64 1024
Disp. Tol. [ps/nm] 184 / 172 / 190 / 248 / 176 / >320 /

@ Pen. 2 dB (10−4 ) 150 294 >320 >320 >320 >320
Disp. Tol. [ps/nm] 143 / 179 / 188 / 233 / 284 / >320 /
@ EOP 2 dB 179 >320 >320 >320 >320 >320
SPM Tol. [dBm] 13.8 / 10.0 / 7.9 / 7.6 / 4.9 / -0.7 /

@ Pen. 2 dB (10−4 ) >15 15.0 13.4 6.1 4.1 -2.6
SPM Tol. w. comp. [dBm] -/ -/ -/ 9.2 / 7.3 / 8.7 /
@ Pen. 2 dB (10−4 ) - - - 11.5 10.5 7.0
SPM Tol. [dBm] 13.0 / 9.6 / 8.0 / 6.3 / 6.0 / 2.5 /
@ EOP 2 dB >15 >15 14.1 7.1 6.4 0.7
7.3 Parameter Summary Homodyne Synchronous Detection 203
The receiver sensitivities at BER=10−4 for shot-noise limited detection with

η = 1 , as well as the OSNR requirements for a BER of 10−4 for amplifier noise lim-
ited detection (with and without consideration of the influence of an optical receiver
filter with Bopt = 2.5·r S ), are summarized in the first three rows of Table 7.6. When
these values are compared with the parameters of systems with differential detection
listed in Table 7.2 and Table 7.4, it can be seen that systems with homodyne syn-
chronous detection have the principal advantage of superior noise performance in
comparison with systems with direct and homodyne differential detection. The per-
formance gain depends on the modulation format used. It is about 2 dB on average
and slightly smaller than usual since differential coding is employed to resolve the
phase ambiguity arising at the carrier synchronization. It should be borne in mind
that a further penalty, caused by the phase estimation for averaging over small block
lengths, must be considered when the linewidth requirements are in a critical range
for a particular modulation format. In this case, similar noise performance can be
obtained in systems with differential detection which feature simpler implementa-
tion and more relaxed linewidth requirements.
Even when employing feed forward phase estimation which can be practically
implemented without impairment from processing delay, the laser linewidth require-
ments become critical at r B = 40 Gbit/s for 16PSK, Square 16QAM and Square
64QAM. The particular requirements on laser linewidth for systems which estimate
the phase using the feed forward M-th power block scheme (with class partitioning
for Square QAM formats, as described in Sect. 3.5.3) are listed in Table 7.6. This
scheme is a candidate for practical employment in the near future due to its rela-
tively simple implementation. Potentially, the laser phase noise requirements can
be further relaxed by using symbol-to-symbol phase correction, Wiener filtering
and—for Square QAM, in particular—decision-directed phase estimation schemes
or enhanced phase estimation algorithms which allow to incorporate all constella-
tion points into the calculation of the phase error estimate [8].
The chromatic dispersion and self phase modulation tolerances of the different
modulation formats can be also found in Table 7.6. The same tendencies arise as
for systems with differential detection. The migration to higher-order modulation
formats with more bits per symbol results in higher dispersion tolerances and worse
SPM tolerances. SPM tolerance is very poor for all QAM formats, but can be greatly
improved by nonlinear phase shift compensation. RZ signals show a more beneficial
system behavior than NRZ signals in relation to all criteria shown (except for the
linewidth requirements which are almost identical)—at least for the single-channel
single-span system configurations without narrow optical filtering at the transmitter
or the transmission link discussed in this chapter.
A comparison of system performance when employing different Square 16QAM
transmitter configurations is presented in Table 7.7. Nearly the same performances
values can be observed for the different transmitters in the case of RZ pulse shape.
For NRZ, the conventional IQ transmitter features a superior system performance
with respect to its robustness against noise, chromatic dispersion and SPM (with
nonlinear phase shift compensation) in comparison with the other transmitters. The
improved performance for RZ pulse shape in comparison with NRZ pulse shape
Table 7.7 Properties of Square 16QAM systems for homodyne detection with digital phase esti-
mation at 40 Gbit/s for NRZ / RZ when using different transmitter configurations
Transmitter Type Serial Conv. IQ Enh. IQ Tand.-QPSK
PR X,avg,d Bm [dBm], BER=10−4 -37.8 / -40.0 -38.8 / -40.1 -37.9 / -40.0 -37.7 / -39.9
Disp. Tol. [ps/nm], Pen. 2 dB (10−4 ) 176 / >320 >320 / >320 184 / >320 198 / >320
SPM Tol. [dBm], Pen. 2 dB (10−4 ) 4.9 / 4.1 5.1 / 4.2 5.1 / 4.2 -/-
SPM Tol. w. comp., Pen. 2 dB (10−4 ) 7.3 / 10.5 9.6 / 10.5 7.2 / 10.4 -/-
becomes particularly noticeable for the serial transmitter, the enhanced IQ transmit-
ter and the Tandem-QPSK transmitter, but also for the conventional IQ transmitter,
to a reduced extent.
In this chapter, tendencies concerning the influence of single parameters on sys-

tem performance were highlighted looking at the back-to-back case and single-span
transmission and considering a wide range of high-order modulation formats. The
tendencies observed give a first indication of the transmission behavior of partic-
ular formats for multi-span transmission over long distances. Independently from
the detection scheme used, the migration to higher-order modulation formats with
more bits per symbol has the advantage of higher chromatic dispersion and PMD
tolerances. On the other hand, two main parameters limiting the distances attainable
in multi-span long-haul transmission systems are becoming more critical: noise per-
formance and Kerr nonlinearities. Therefore, it can be expected that the achievable
transmission lengths are reduced when migrating to higher-order formats. This issue
is discussed further in Chap. 8.
References
1. Avlonitis, N.S., Yeatman, E.M.: Performance evaluation of optical DQPSK using saddle point
approximation. IEEE Journal of Lightwave Technology 24(3), 1176–1185 (2006)
2. Cai, Y., Pilipetskii, A.N.: Comparison of two carrier phase estimation schemes in optical co-
herent detection systems. In: Proceedings of Optical Fiber Communication Conference (OFC),
OMP5 (2007)
3. Desurvire, E.: Erbium-Doped Fiber Amplifiers. John Wiley & Sons, Inc. (1994)
5. Ip, E., Kahn, J.M.: Feedforward carrier recovery for coherent optical communications. IEEE
Journal of Lightwave Technology 25(9), 2675–2692 (2007)
6. Kahn, J.M., Ho, K.P.: Spectral efficiency limits and modulation/detection techniques for
DWDM systems. IEEE Journal of Selected Topics in Quantum Electronics 10(2), 259–272
(2004)
7. Lau, A.P.T., Kahn, J.M.: Signal design and detection in presence of nonlinear phase noise.
References 205
9. Norimatsu, S., Iwashita, K.: Linewidth requirements for optical synchronous detection sys-
temes with nonnegligible loop delay time. IEEE Journal of Lightwave Technology 10(3),
341–349 (1992)
12. Rice, F.: Bounds and algorithms for carrier frequency and phase estimation. Ph.D. thesis,
University of South Australia (2002)
13. Seimetz, M., et al.: Optical systems with high-order DPSK and Star QAM modulation based
on interferometric direct detection. IEEE Journal of Lightwave Technology 25(6), 1515–1530
(2007)
14. Siuzdak, J., van Etten, W.: BER evaluation for phase and polarization diversity optical homo-
dyne receivers using noncoherent ASK and DPSK demodulation. IEEE Journal of Lightwave
Technology 7(4), 584–599 (1989)
15. Smith, P., et al.: Optical heterodyne binary-DPSK systems: A review of analysis and perfor-
mance. IEEE Journal on Selected Areas in Communications 13(3), 557–568 (1995)
18. Xu, C., Liu, X.: Postnonlinearity compensation with data-driven phase modulators in phase-
shift keying transmission. Optics Letters 27(18), 1619–1621 (2002)
Chapter 8
Multi-Span Long-Haul Transmission
Abstract This chapter outlines some current activities performed in the author’s re-
search group and aimed at identifying the performance and the distances attainable
in optical multi-span long-haul transmission systems with high-order modulation.
Firstly, some system experiments are presented which investigate achievable dis-
tances with 8PSK and Star 16QAM when using homodyne synchronous detection.
In these experiments, system configurations with optical inline CD compensation
as well as electrical CD compensation at the receiver are considered. Subsequently,
the transmission lengths which have been achieved with different modulation for-
mats are compared, showing that they are significantly reduced when migrating to
higher-order formats. Finally, system degradation by the SPM-induced mean non-
linear phase shift in multi-span QAM transmission is highlighted and the efficiency
of possible compensation schemes is shown.
Chapter 7 illustrated the behavior of a wide range of modulation formats in re-

lation to particular impairments for single-span transmission. This allows tenden-
cies concerning the properties for multi-span long-haul transmission to be pre-
estimated. However, achievable transmission distances and long-haul transmission
performance of a particular modulation format can be only evaluated precisely for
the specific system in which it is operated, since several degradation effects interact
with each other within the whole system.
Optical multi-span long-haul transmission systems, which are typically com-
posed of multiple transmission sections each containing a fiber—usually with a
length of about 80 km—and optical amplifiers compensating for fiber attenuation,
are mainly limited by amplifier noise and fiber nonlinearities. Chromatic dispersion
can be compensated for within each span (inline CD compensation) or electrically
at the receiver. In Chap. 7, it was shown for single-span transmission that noise
and self phase modulation become more critical in the case of high-order modula-
tion. Thus, the attainable transmission lengths for multi-span transmission can be
expected to be reduced when migrating to higher-order formats.


208 8 Multi-Span Long-Haul Transmission
Already installed long-haul fiber transmission systems are mainly based on on-
off keying and differential binary phase shift keying. Currently, quadrature phase
keying is on the way towards a commercial deployment. Even higher-order formats
are not yet adopted in commercially deployed systems. But the imminent need for
optical data transmission capacity feeds the interest in system concepts allowing
for high spectrally efficient transmission by the use of higher-order modulation for-
mats and motivates the current research activities in this field. This chapter presents
some initial research results of the research group of the author dealing with the
multi-span transmission behavior of some high-order modulation formats and the
achievable transmission lengths.
Section 8.1 presents re-circulating fiber loop experiments which identify the
transmission distances attainable with RZ-8PSK. In these experiments, system con-
figurations with optical inline CD compensation and electrical CD compensation
at the receiver are considered. In Sect. 8.2, multi-span transmission experiments
with RZ-Star 16QAM are described. They highlight the limitation of transmission
lengths due to noise, implementation imperfections of the system practically in-
vestigated and signal distortions caused by the SPM-induced nonlinear phase shift.
Subsequently, transmission distances achieved with RZ-QPSK, RZ-8PSK and RZ-
Star 16QAM are compared in Sect. 8.3. Finally, system degradation by the SPM-
induced mean nonlinear phase shift in multi-span QAM transmission is illustrated
on the basis of RZ-Star 16QAM in Sect. 8.4, and the efficiency of some possible
compensation schemes is shown.
8.1 8PSK Multi-Span Transmission Experiments
The next two subsections present some experimental results which have been pub-
lished in [5] and [1]. Section 8.1.1 describes a RZ-8PSK multi-span transmission
experiment with optical inline CD compensation and homodyne synchronous detec-
tion. In Sect. 8.1.2, distances experimentally achieved with RZ-8PSK are illustrated
in the case of employing electrical CD compensation at the receiver.
8.1.1 Optical Inline CD Compensation
In Fig. 8.1, the schematic of the experimental system setup with optical inline CD
compensation used in [5] is shown.
For RZ-8PSK signal generation, the parallel transmitter structure (see Sect. 2.4.2)
is employed. The CW light is emitted by an external cavity laser with a linewidth
specified as 100 kHz. A Mach-Zehnder modulator is used for RZ pulse carving.
Afterwards, the optical IQ modulator generates an optical QPSK signal and the
consecutive phase modulator accomplishes a π/4 phase modulation for constitution
of the 8PSK constellation. The transmitted data signal is a 211 de Bruijn sequence
8.1 8PSK Experiments with Optical Inline CD Compensation 209
Fig. 8.1 Experimental system

setup for the 30 Gbit/s co- RZ 90° MZM
herent RZ-8PSK multi-span ECL MZM MZM
PM
Transmitter
long-haul transmission exper- MZM
LO
iment with inline chromatic
dispersion compensation per- t
SSMF
DATA
formed in [5] 4 km Pulse DATA
Pat.Gen.
8PSK
Transmission Link
3 dB
EDFA
SSMF DCF
80 km Loop 13 km EDFA
EDFA
Receiver
Bal.
Processing
EDFA Data Digital
Det.
Offline
Pol.-ctrl. 90°- Storage
Hybrid Oscillosc.
Bal.
LO 40 GSa/s
Det.
which is given to the modulator inputs with different delays. The system symbol
rate is r S = 10 Gbaud, corresponding to a data rate of r B = 30 Gbit/s.
The transmission link is based on a re-circulating loop with an adjustable number
of sections. Each section consists of 80 km SSMF and about 13 km DCF which fully
compensates for the SSMF dispersion. The fiber parameters are measured for the
SSMF as αd B = 0.2 dB/km, Dλ = 16.8 ps/(nm·km), Sλ = 0.0585 ps/(nm2 · km)
and γ = 0.99 W−1 km−1 , as well as αd B = 0.56 dB/km, Dλ = −104.3 ps/(nm·km),
Sλ = −0.217 ps/(nm2 ·km) and γ = 5.1 W−1 km−1 for the DCF. Erbium doped fiber
amplifiers are used to compensate for the fiber loss and for controlling the launch
powers into the SSMF and DCF. The noise power of the optical amplifiers is reduced
by optical filters. The signal can be sent to the receiver after being transmitted over
a desired number of cascaded sections by the use of acousto-optical switches.
At the receiver end, the RZ-8PSK signal is interfered with the light of a local
oscillator (Plo =10 dBm) in a Li N bO3 2×4 90◦ hybrid. For experimental simplicity,
the LO light is taken from the transmitter laser to avoid an automatic frequency
control loop. In the back-to-back case where the transmitter is directly connected to
the receiver, the received information signal and the LO signal are de-correlated by a
4 km long SSMF. The polarization is controlled manually in front of the inputs of the
2 × 4 90◦ hybrid. The hybrid output signals are detected by two balanced detectors
and the in-phase and quadrature photocurrents are digitized using a 40 GSa/s digital
storage oscilloscope. Finally, data is recovered off-line by applying an 8-th power
digital phase estimation algorithm (N =8) and an appropriate data recovery circuit,
including a differential decoder to remove the 8-fold phase ambiguity.
For system characterization, the noise loaded back-to-back performance was

measured in [5] and compared with Monte Carlo simulations. As can be observed
from Fig. 8.2a, an OSNR of about 13 dB was required in the experiment for a BER
of 10−3 . The Monte Carlo simulation for RZ-8PSK at r B = 30 Gbit/s for the sys-
tem configuration with the parallel transmitter and arg-decision indicates a required
OSNR of about 11.5 dB. The difference of about 1.5 dB must be considered an im-
plementation penalty when using off-the-shelf components. Furthermore, the mea-
sured optical signal spectrum at the transmitter output is depicted in Fig. 8.2b. The
20 dB bandwidth of the optical 30 Gbit/s RZ-8PSK signal can be seen to be approx-
imately 0.2 nm (25 GHz).
Norm. power in 0.01nm [dBm]

1E-2 0
a b
-10
1E-3
Experiment -20
BER
Simulation
-30
1E-4
-40
RZ-8PSK, 30Gbit/s RZ-8PSK, 30Gbit/s
1E-5
8 10 12 14 16 18 1550,4 1550,6 1550,8 1551,0 1551,2
OSNR [dB] Wavelength [nm]
Fig. 8.2 Measured back-to-back OSNR performance (a) and optical signal spectrum at the trans-
mitter output (b) in the 30 Gbit/s RZ-8PSK multi-span transmission experiment performed in [5]
To determine the possible transmission lengths of 30 Gbit/s RZ-8PSK on widely

used SSMF and DCF spans, the optical RZ-8PSK signal was sent over the re-
circulating fiber loop. The launch powers into the SSMF and DCF (PS M F and
PDC F , respectively) were varied. In Fig. 8.3a, the measured BER is shown versus
the transmission length for different values of PS M F and PDC F .
Reach with inline CD compensation SSMF input power optimization at 960km

1E-2 1E-2
PSMF = -6.0dBm
PDCF = -7.6dBm
PDCF ≈ PSMF - 5dB
1E-3 PSMF = -1.0dBm 1E-3
PDCF = -6.6dBm
BER
BER
PSMF = -0.7dBm
1E-4 PDCF = -2.2dBm 1E-4
a RZ-8PSK, 30Gbit/s b RZ-8PSK, 30Gbit/s

1E-5 1E-5
0 480 960 1440 1920 2400 -8 -6 -4 -2 0 2 4
Transmission length [km] SSMF input power [dBm]
Fig. 8.3 BER vs. transmission distance (a) and power optimization at 960 km (b) in the RZ-8PSK
experiments with inline chromatic dispersion (CD) compensation performed in [5]
8.1 8PSK Experiments with Optical Inline CD Compensation 211
First, PDC F was chosen about 1.5 dB below PS M F and best BER performance
was obtained for PS M F = −0.7 dBm. When decreasing PDC F , the bridgeable trans-
mission distance can be further increased, but only slightly. Figure 8.3b shows the
influence of SSMF launch power variation on the BER at a fixed transmission length
of 960 km when PDC F is chosen about 5 dB below PS M F . Optimal performance was
obtained for PS M F = −1 dBm and PDC F = −6.6 dBm. In this case, the trade-off
between noise and nonlinearities allows for the largest transmission distances.
For a BER of about 10−3 which allows for error free transmission when using
forward error correction coding, a transmission distance of 1360 km (17 cascaded
sections) could be achieved, choosing the optimum launching condition. The plotted
BER values were calculated for evaluating 2.304.000 bits and averaging over 6 re-
alizations. In Fig. 8.4, the IQ diagrams for back-to-back and after 960 km are shown
as displayed by the digital storage oscilloscope when employing synchronous sam-
pling, taking four samples per symbol and showing 2048 samples.
NRZ-8PSK, BtB NRZ-8PSK, 960km RZ-8PSK, BtB RZ-8PSK, 960km
Fig. 8.4 Received 8PSK IQ diagrams for back-to-back and after 960 km for NRZ and RZ pulse
shapes, as displayed by the digital storage oscilloscope when employing synchronous sampling,
taking four samples per symbol and showing 2048 samples
The IQ diagrams for NRZ in Fig. 8.4 were obtained by bypassing the pulse carver
of the transmitter. They show the typical symbol transitions of the parallel trans-
mitter structure. The received RZ-8PSK constellation diagrams from selecting one
sample per symbol in the middle of the symbols are depicted in Fig. 8.5 for the
back-to-back case before (a) and after (b) phase estimation and at 1200km after
phase estimation (c) by displaying 65.536 received symbols.
a b c
BtB, before PE BtB, after PE 1200km, after PE
Fig. 8.5 Received RZ-8PSK
constellation diagrams for
back-to-back before (a) and
after (b) digital phase estima-
tion (PE) and at 1200 km after
phase estimation (c)
8.1.2 Electrical CD Compensation at the Receiver
A great advantage of the homodyne IQ receiver in comparison with the direct detec-
tion receiver is its capability to efficiently compensate for transmission impairments
in the electrical domain. All information parameters of the optical signal are ac-
cessible after detection. The accumulated chromatic dispersion can be compensated
for by convolution of the received complex signal with the inverse fiber impulse
response using a complex finite impulse response (FIR) filter. The transfer function
of this filter—when neglecting the dispersion slope—is given by
2 ·N
He (ω) = e j0.5·β2 ·ω F S ·l , (8.1)
where β2 represents the fiber dispersion, N F S the number of cascaded transmission

sections and l the fiber length in each section.
To discover the attainable transmission lengths without optical inline compen-
sation of chromatic dispersion, the DCF and the EDFA in front of the DCF were
removed from the transmission section shown in Fig. 8.1 in the experiments de-
scribed in [1], and the raw data was electrically equalized off-line by an ideal FIR
filter before digital phase estimation within the receiver. For practical filter imple-
mentation, the equalization performance will be limited, for instance by the number
of taps. In Fig. 8.6b, the BER against the launch power into the SSMF (PS M F ) after
transmission through 1680 km is shown. An optimum value of PS M F ≈ −3.5 dBm
was measured which is about 2.5 dB below the optimum value for inline chromatic
dispersion compensation.
Reach with electrical CD compensation at RX SSMF input power optimization at 1680km

1E-2 1E-2
a b
PSMF = -1.0dBm
1E-3 1E-3
BER
BER
1E-4 PSMF = -3.5dBm 1E-4
RZ-8PSK, 30Gbit/s RZ-8PSK, 30Gbit/s

1E-5 1E-5
960 1440 1920 2400 2880 3360 -8 -6 -4 -2 0 2
Fig. 8.6 BER vs. distance (a) and power optimization at 1680 km (b) in the RZ-8PSK experiments
with electrical chromatic dispersion (CD) compensation at the receiver conducted in [1]
In Fig. 8.6a, the BER against the transmission length is shown for two differ-
ent values of PS M F . When choosing the optimum launching condition, a maximum
transmission distance of 2800 km (35 cascaded sections) for a target BER of 10−3
could be achieved. This is approximately twice the maximum transmission distance
compared with optical inline dispersion compensation and can be explained by an
8.2 Single-Channel Star 16QAM Experiments with Inline CD Compensation 213
improvement of OSNR, caused by the noise reduction through the removal of the
EDFA-amplified DCF and measured to have an increase of about 1 dB at a transmis-
sion distance of 1440 km. Moreover, the accumulated fiber nonlinearities decrease
due to the removal of the DCF and lower optimum launch power into the SSMF.
From another point of view, the amplifier spacing can be enlarged to achieve the
same transmission performance as for inline dispersion compensation at a certain
distance. This can result in cost savings.
It can be concluded from the outlined 8PSK experiments that a distance of

1360 km could be bridged with RZ-8PSK transmission and homodyne detection at
30 Gbit/s for a target BER of 10−3 when employing optical inline dispersion com-
pensation with DCF modules. The maximum transmission distance could be dou-
bled to approximately 2800 km by replacing optical inline CD compensation with
electrical CD compensation at the receiver.
8.2 Star 16QAM Multi-Span Transmission Experiments
Next, in Sect. 8.2.1, a multi-span transmission experiment with Star 16QAM and
optical inline compensation of chromatic dispersion is described which has been
published in [6]. It investigates the transmission performance of Star 16QAM
single-channel systems for single-polarization and polarization division multiplex-
ing (PDM) and highlights the limitation of attainable transmission lengths due to
amplifier noise and SPM-induced signal distortions. Subsequently, a Star 16QAM
WDM experiment is presented in Sect. 8.2.2. Five WDM channels are transmit-
ted over more than 1000 km and demodulated with the aid of enhanced electronic
equalization at the receiver.
8.2.1 Single-Channel Experiments
Figure 8.7 illustrates the schematic of the experimental setup for investigation of
Star 16QAM multi-span transmission systems with optical inline CD compensation
and polarization division multiplexing used in [6].
The main part of the transmitter is equal to the RZ-8PSK transmitter employed
in the 8PSK experiments described in Sect. 8.1. Another Mach-Zehnder modulator
is used for Star 16QAM signal generation. By changing the driving and bias volt-
ages of this modulator, different ring ratios can be adjusted. The system symbol rate
is r S = 10 Gbaud, resulting in a data rate of 40 Gbit/s for single-polarization and
80 Gbit/s for polarization division multiplexing. Polarization division multiplexed
transmission is investigated by splitting the signal with a polarization beam split-
ter, delaying one polarization component, and afterwards adding both polarization
components in a polarization beam combiner (PBC).
QPSK 8PSK Star 16QAM

RZ 90° MZM
Transmitter
LO ECL MZM PM MZM 3 dB
PBS 3 dB
PBC
t
MZM
EDFA
t 1 t 2 t3
Pulse DATA
Pat.Gen.
SSMF DATA
4 km
Transmission Link
3 dB EDFA
SSMF Loop DCF

80 km 13 km
EDFA EDFA
EDFA
Bal.
Data
90°- Det.
Receiver
LO Hybrid
Processing
PBS Bal. Digital
Offline
Det.
Storage
Bal. Oscillosc.
3 dB Data Det. 50 GSa/s
90°-
LO Hybrid Bal.
Det.
Fig. 8.7 Experimental system setup for the 80 Gbit/s coherent RZ-Star 16QAM multi-span trans-
mission experiment with inline chromatic dispersion compensation and polarization division mul-
tiplexing performed in [6]
The transmission link is based on a re-circulating fiber loop with an adjustable

number of sections and optical inline CD compensation in each section. It has the
same configuration as in the experiment described in Sect. 8.1.1. In the case of PDM
experiments, the signal launched into the receiver is split by a PBS first. Afterwards,
both polarization components are interfered with the LO light (which is again taken
from the transmitter laser to avoid an automatic frequency control loop) in two 2 × 4
90◦ hybrids. The hybrid output signals are detected by a pair of balanced detectors
and the photocurrents are digitized using a 50 GSa/s digital storage oscilloscope.
Data is then recovered off-line by applying digital phase estimation (using the feed-
forward block scheme with rectangular time domain filtering and averaging over 8
symbols) and appropriate data recovery. Further electrical equalization of transmis-
sion impairments was not performed in this experiment.
Figure 8.8a depicts the transmission lengths achieved with 10 Gbaud RZ-Star
16QAM in the experiments, showing the measured BER versus the transmission
length for launched powers into the SSMF (PSS M F ) as specified there. In all cases,
the power launched into the DCF (PDC F ) was chosen as 5 dB below PS M F . The in-
fluence of SSMF launch power variation on the BER at fixed transmission lengths of
800 km (single-polarization) and 320 km (PDM) is shown in Fig. 8.8b and indicates
8.2 Single-Channel Star 16QAM Experiments with Inline CD Compensation 215
optimal SSMF input powers of -4 dBm for single-polarization and 0 dBm for PDM.
Even when applying the optimal launch powers, the transmission distance was lim-
ited to about 600 km for single-polarization and 400 km for PDM case.
Reach with inline CD compensation SSMF input power optimization

1E-2 1E-1
PDM, RR 1.8
a b
PSMF = 0dBm Single Pol., RR 1.8
at 800km
1E-3 1E-2
BER
BER
Single Pol., RR 1.9
PSMF = -1dBm
1E-4 1E-3
Single Pol., RR 1.9
PSMF = -4dBm
PDM, RR 1.8
at 320km RZ-Star 16QAM
RZ-Star 16QAM
1E-5 1E-4
0 200 400 600 800 1000 -8 -6 -4 -2 0 2 4
Fig. 8.8 BER vs. transmission distance (a) and power optimization (b) in the RZ-Star 16QAM
experiments with inline chromatic dispersion compensation conducted in [6], r S = 10 Gbaud.
Compared with RZ-8PSK, transmission distances are considerably reduced for

RZ-Star 16QAM. The reasons for that are the following: Firstly, the Euclidean dis-
tances between the symbols for Star 16QAM are smaller (the required OSNR at
BER=10−3 for the single-polarization system in the experiment was 16.5 dB for RZ-
Star 16QAM, but only about 13.5 dB for RZ-8PSK) and ISI effects appearing within
the transmitter impaired system performance. In the practical transmitter setup, ev-
ery new modulation stage can lead to higher ISI caused by pattern effects and thus
to higher implementation penalties. Secondly, symbols with different power levels
experience different SPM-induced mean nonlinear phase shifts as observable for
the single-polarization case from the constellation diagram received after 560 km,
shown on the right side of Fig. 8.9. Symbols on different rings have obtained dif-
ferent phase rotations. When these phase shifts are not being compensated for, they
limit attainable transmission lengths even for ISI- and noise-free systems.
Received constellation diagrams for single-polarization
Back-to-back 400km 560km
Fig. 8.9 Received Star 16QAM constellation diagrams for back-to-back, after 400 km and after
560 km in the experiments conducted in [6] for the single-polarization case
Figure 8.10b illustrates how BER performance can be improved when the rela-
tive nonlinearity-induced phase difference of both rings is compensated for. After
720 km, optimum BER performance is obtained when symbols on the inner ring
are rotated by about -0.19 rad. Using the mean nonlinear phase shift compensa-
tion, achievable distances can probably be increased to about 1000 km for single-
polarization (see Fig. 8.10a), whereas the performance gain was only small for PDM
where other impairments such as polarization crosstalk limited the performance of
the system investigated in [6].
Star 16QAM reach with /without compensation NL phase shift compensation at 720km
1E-2 1E-2
PDM, with comp.
a b
PDM, w/o comp. Single Pol., w/o comp.
1E-3
BER
BER 1E-3
1E-4
Single Pol., with comp.
RZ-Star 16QAM Single-polarization RZ-Star 16QAM

1E-5 1E-4
0 200 400 600 800 1000 -0,4 -0,3 -0,2 -0,1 0,0
Transmission length [km] Phase shift inner ring [rad]
Fig. 8.10 a BER versus distance for single-polarization and PDM RZ-Star 16QAM systems
with / without nonlinear phase shift compensation. b BER improvement through mean nonlinear
phase shift compensation at 720 km for single-polarization RZ-Star 16QAM. r S = 10 Gbaud.
An important performance parameter which must be optimized in Star QAM sys-

tems is the ring ratio. Figure 8.11 illustrates the BER dependence on the ring ratio
for single-polarization RZ-Star 16QAM, measured in [6]. After 400 km, the opti-
mum ring ratio is about 1.77, which is consistent with the theory (see Sect. 7.3.1).
In the case of back-to-back, a higher ring ratio was optimal in the experiment. This
can be explained by the ISI-induced radial distortions of the constellation points due
to pattern effects which can be recognized in the left diagram of Fig. 8.9.
Star 16QAM ring ratio optimization

1E-2
At 400km
1E-3
1E-4
BER
Fig. 8.11 Experimental opti- 1E-5 Back-to-back
mization of the Star 16QAM

1E-6 RZ-Star 16QAM
ring ratio for back-to-back
Single-polarization
transmission and at 400 km 1E-7
for the single-polarization 1,6 1,7 1,8 1,9 2,0 2,1 2,2 2,3 2,4
case, performed in [6] Ring ratio
8.2 Star 16QAM WDM Experiments with Electrical CD Compensation 217
8.2.2 WDM Experiments
Another RZ-Star 16QAM transmission experiment which uses electrical dispersion

compensation at the receiver has been performed in the WDM systems group of
the Fraunhofer Institute for Telecommunications, Heinrich-Hertz-Institut, in collab-
oration with VPIsystems GmbH. Five 10 Gbaud RZ-Star 16QAM WDM channels
were transmitted over 720 km / 1200 km SSMF with / without PDM on a 50 GHz
frequency grid centered at 1550.92 nm. The central channel was demodulated with
the aid of enhanced electronic equalization at the receiver.
The system setup employed in the experiment is very similar to the single-
channel experimental setup shown in Fig. 8.7, but upgraded to WDM at some lo-
cations. The transmitter consists of five external cavity lasers which are coupled by
a set of 3 dB couplers. The transmission section within the re-circulating fiber loop
is composed of three 80 km SSMF spans without inline dispersion compensating
modules. Erbium doped fiber amplifiers are used to compensate for the loop loss
and control the launch powers into the fiber spans. The noise power of the ampli-
fiers is reduced by optical bandpass filters and a gain equalizer controls the gain of
each WDM channel. At the receiver, the central channel is selected by an optical
bandpass filter and then split by a polarization beam splitter in the case of PDM.
Afterwards, both polarization components are interfered with the LO light in two
2 × 4 90◦ hybrids which are followed by four balanced detectors.
In the electrical domain, the photocurrents are digitized and processed by dig-
ital equalization. Chromatic dispersion is compensated for by using a filter that
is implemented in the frequency domain. Thereafter, an enhanced constant mod-
ulus algorithm—denoted as multiple moduli algorithm (MMA) and published in
[3] where it was applied to Square 16QAM—is used to compensate for residual
transmission impairments such as nonlinearities and polarization crosstalk. After
equalization, feed forward M-th power phase estimation is used to compensate for
laser phase noise. Finally, decoding and error counting are performed.
Figure 8.12 shows the measured WDM spectra at the output of the transmitter
and after transmission over 1200 km for a fiber launch power of -1 dBm / channel.
0
-5
Power in 0.1nm [dBm]
-10
-15 At 1200km
-20
Fig. 8.12 Measured five- -25
Transmitter
channel WDM RZ-Star -30 output
16QAM power spectra within -35
0.1 nm at the transmitter -40 Single-polarization WDM RZ-Star 16QAM
output and after transmission -45
over 1200 km for the 40 Gbit/s 1548 1549 1550 1551 1552 1553 1554
single-polarization system Wavelength [nm]
BER values of the central channel were measured after different numbers of
loop round trips. Figure 8.13a shows BER versus transmission distance when us-
ing only the non-adaptive frequency domain equalizer to compensate for chromatic
dispersion. Applying this approach, transmission distances of 480 km / 950 km were
achieved for 80 Gbit/s / 40 Gbit/s RZ-Star-16QAM with / without PDM (assuming a
BER reference of 10−3 ). In Fig. 8.13b, the BER is shown for the case that an addi-
tional MMA equalizer with 9 taps per complex filter structure is employed. Using
this additional equalizer, transmission distance could be extended to about 800 km
and 1400 km for 80 Gbit/s and 40 Gbit/s RZ-Star 16QAM with and without PDM,
respectively.
CD compensation in frequency domain Additional adaptive MMA equalization

1E-2 1E-2
a PDM b
PDM
1E-3 1E-3
BER
BER
Single-polarization
1E-4 1E-4
Single-polarization
WDM RZ-Star 16QAM WDM RZ-Star 16QAM

1E-5 1E-5
0 400 800 1200 1600 0 400 800 1200 1600
Transmission length [km] Transmission length [km]
Fig. 8.13 BER versus transmission distance for 80 Gbit/s and 40 Gbit/s WDM RZ-Star-16QAM
with and without PDM, using only frequency domain CD compensation (a) and MMA equalization
additionally (b).
It can be concluded from the RZ-Star 16QAM experiments described in the last
two sections that distances of 600 km and 400 km over a link with optical inline CD
compensation could be bridged at 10 Gbaud without and with PDM, respectively,
achieving a BER lower than 10−3 . By employing compensation of the nonlinear
phase shift, distances could be increased to about 1000 km for single-polarization.
It can be expected that even higher transmission distances can be achieved in sys-
tems with inline CD compensation when transmitter performance is further im-
proved and additional electronic distortion equalization at the receiver is used. In
the five-channel WDM experiment with electrical CD compensation at the receiver
described in the last paragraphs, 80 Gbit/s and 40 Gbit/s WDM RZ-Star 16QAM
transmission has been demonstrated over 800 km and 1400 km for PDM and single-
polarization, respectively. This was possible by applying enhanced electronic dis-
tortion equalization techniques.
8.3 Comparison of Transmission Distances 219
8.3 Comparison of Transmission Distances
Based on the results obtained in the experiments described in Sect. 8.1 and Sect. 8.2,
transmission distances at 10 Gbaud achieved experimentally with RZ-QPSK, RZ-
8PSK and RZ-Star 16QAM are compared in this section. Results for RZ-QPSK,
which have not been shown in the last sections, are attained using the same exper-
imental setup as illustrated for RZ-8PSK in Fig. 8.1. At the transmitter, simply the
π/4 phase modulation of the phase modulator must be switched off. At the receiver
end, the quadrature optical frontend remains unchanged and just the phase estima-
tion and the data recovery must be adapted.
A first indicator for the transmission length achievable with a particular modu-
lation format is the back-to-back noise performance. In Fig. 8.14, the back-to-back
OSNR requirements measured in [6] for RZ-QPSK, RZ-8PSK and RZ-Star 16QAM
are compared for single-polarization (Fig. 8.14a) and PDM (Fig. 8.14b).
OSNR requirements single-polarization OSNR requirements for PDM

1E-2 1E-2
a b
RZ-Star 16QAM RZ-Star 16QAM
1E-3 1E-3
BER
BER
RZ-8PSK
RZ-8PSK
1E-4 1E-4
RZ-QPSK RZ-QPSK
1E-5 1E-5
6 8 10 12 14 16 18 20 22 24 10 12 14 16 18 20 22 24 26 28
OSNR [dB] OSNR [dB]
Fig. 8.14 Back-to-back OSNR requirements of RZ-QPSK, RZ-8PSK and RZ-Star 16QAM mea-
sured in [6] for single-polarization (a) and polarization division multiplexing (b), assuming a com-
mon symbol rate of r S = 10 Gbaud.
To obtain a BER of 10−3 , an OSNR of about 16.5 dB and 20.0 dB was required
for RZ-Star 16QAM in the case of single-polarization and PDM, respectively. The
measured OSNR penalty at BER=10−3 is 2-3 dB and 9 dB compared with RZ-8PSK
and RZ-QPSK, respectively. The difference in the required OSNR between RZ-Star
16QAM and RZ-QPSK is larger than expected from numerical simulation (7.5 dB,
as shown in Sect. 7.3.1), since in the practical transmitter setup the modulation
stage performing the 45◦ phase modulation led to an implementation penalty caused
by pattern effects of the electrical driving signals. OSNR requirements increase by
about 3 dB when upgrading from single-polarization to PDM.
Transmission distances achieved in [6] with RZ-QPSK, RZ-8PSK and RZ-Star
16QAM in multi-span transmission systems with optical inline CD compensation
are compared in Fig. 8.15 for single-polarization (a) and PDM (b), assuming a com-
mon symbol rate of r S = 10 Gbaud for all formats.
Reach comparison single-polarization Reach comparison for PDM

1E-2 1E-2
a RZ-Star 16QAM b
w/o NL PS comp. RZ-Star 16QAM
w/o NL PS comp.
1E-3 1E-3
RZ-QPSK
BER
BER
RZ-8PSK RZ-8PSK
RZ-QPSK
1E-4 1E-4
Opt. Inline CD comp. Opt. Inline CD comp.

1E-5 1E-5
0 1000 2000 3000 4000 0 1000 2000 3000 4000
Transmission length [km] Transmission length [km]
Fig. 8.15 Transmission distances achieved in [6] with RZ-QPSK, RZ-8PSK and RZ-Star 16QAM
for multi-span transmission with optical inline CD compensation for single-polarization (a) and
PDM (b), assuming a common symbol rate of r S = 10 Gbaud.
The experimental results presented in Fig. 8.15 assume optimized launch powers
into the SSMF and DCF and demonstrate that the attainable transmission distances
are considerably reduced when migrating from QPSK to 8PSK, and even more when
applying Star 16QAM. This is primarily caused by the more stringent OSNR re-
quirements of the higher-order formats, as well as by their reduced tolerance against
nonlinear effects. However, it should be noted that the curves for RZ-Star 16QAM
in Fig. 8.15 are shown without compensation of the mean nonlinear phase shift.
Distances for Star 16QAM can be increased when the relative nonlinearity-induced
phase difference of both rings is compensated for, as described in Sect. 8.2.1 and
further discussed in Sect. 8.4. Moreover, distances for all formats can potentially be
increased by means of adaptive electrical equalization within the coherent receiver,
which has not been employed in the experiments conducted in [6].
As illustrated for RZ-8PSK and RZ-Star 16QAM in Sect. 8.1.2 and Sect. 8.2.2,
longer transmission distances are possible when optical inline CD compensation is
replaced with electrical CD compensation within the coherent receiver. Transmis-
sion distances experimentally achieved with optical inline CD compensation and
electrical CD compensation at r S = 10 Gbaud are compared in Fig. 8.16a for the
single-polarization case, exemplarily for RZ-QPSK and RZ-8PSK.
1E-2
RZ-8PSK,
inline comp.
Fig. 8.16 Reach compari- 1E-3 RZ-QPSK,
son of RZ-QPSK and RZ- inline comp.
BER
8PSK at r S = 10 Gbaud for RZ-8PSK,

el. comp.
single-polarization multi-span 1E-4 RZ-QPSK,
transmission with optical in- el. comp.
line CD compensation and

electronic CD compensation, 1E-5
based on the experiments 0 1000 2000 3000 4000 5000 6000
performed in [1, 5]. Transmission distance [km]
8.4 Nonlinear Phase Shift Compensation in Systems with Inline CD Compensation 221
8.4 Nonlinear Phase Shift Compensation
As explained in Sect. 7.1.6 and Sect. 7.3.5 for single-span systems and briefly dis-
cussed in Sect. 8.2.1 for RZ-Star 16QAM multi-span systems, optical QAM trans-
mission inherently suffers from performance degradation caused by different SPM-
induced mean nonlinear phase shifts experienced by symbols with different power
levels. This section clarifies the influence of this effect in multi-span transmission
systems and points out differences between system configurations with optical in-
line CD compensation (Sect. 8.4.1) and electrical CD compensation at the receiver
(Sect. 8.4.2), regarding the single-polarization and PDM case. Possible compensa-
tion schemes and their efficiency for Star 16QAM are discussed on the basis of
results from numerical simulations obtained in [4].
8.4.1 Systems with Optical Inline CD Compensation
Figure 8.17 shows the RZ-Star 16QAM multi-span system setup with optical in-
line CD compensation employed in [4] for simulative investigation of the single-
polarization case. The RZ-Star 16QAM signal is generated by using the parallel RZ-
Star 16QAM transmitter structure described in Sect. 2.5. The transmission link con-
sists of N F S sections, each being composed of 80 km SSMF, 13 km DCF (fully com-
pensating for the chromatic dispersion of the SSMF) and optical amplifiers (OA)
with a noise figure of 5.6 dB. An additional attenuation of 10 dB was used in each
section to better emulate the behavior of an experimental re-circulating fiber loop
test bed. At the receiver side, the signal is detected by a homodyne phase estimation
receiver which is described in detail in Sect. 3.5.3.
RZ-Star 16QAM Transmitter Star 16QAM Homodyne Receiver

MZM
Phase Estimation
Electr. CD Comp.
Data Recovery
MZM A/D
CW 3dB 3dB PM MZM
RZ 2x4
-90° MZM 90°
Hybrid
A/D
LO
× NFS
Transmission Link
Inline CD compensation
PSMF SSMF PDCF DCF
80 km 13 km 10dB 10dB
OA OA OA α
Comp. Comp.
Case A Case B
Fig. 8.17 Single-polarization RZ-Star 16QAM multi-span system setup used in [4] to investigate
different schemes for compensation of the SPM-induced mean nonlinear phase shift.
In the case of PDM, the transmitter is doubled and both polarizations are
multiplexed in a polarization beam combiner before the PDM signal is launched
into the fiber. Moreover, the receiver frontend is enhanced as shown in Fig. 8.7.
SPM-induced signal distortions are different for single-polarization and PDM sys-
tems, as illustrated in Fig. 8.18 for RZ-Star 16QAM transmission over a single
non-dispersive noise-free transmission section with nonlinear propagation coeffi-
cients of the SSMF and DCF given by γ S M F = 1.43 W−1 km−1 and γ DC F =
5.84 W−1 km−1 , respectively, and for fiber input powers of PS M F = 6 dBm and
PDC F = 1 dBm.
a b
Single-polarization PDM
Fig. 8.18 Effect of the SPM-

induced mean nonlinear phase
shift on RZ-Star 16QAM
signals in single-polarization
systems (a) and for PDM (b).
It can be observed from the single-polarization case (Fig. 8.18a) that symbols
with different power levels undergo different degrees of phase rotation. In the case
of PDM, distortions are different due to nonlinear cross-polarization effects (see
Fig. 8.18b). As already discussed in Chap. 7 for single-span transmission systems,
the resulting distortions of the signal constellations must be compensated for by a
nonlinear phase shift compensator. Without compensation, attainable transmission
lengths for multi-span QAM transmission are strongly limited. This was already
demonstrated in the experiments described in Sect. 8.2.1. For comparison, some
results for PDM systems at 10 Gbaud determined by computer simulations in [4]
are illustrated in Fig. 8.19. These are valid for optimized fiber input powers and
indicate that the transmission distances achieved experimentally for 8PSK in [5]
and Star 16QAM in [6] can potentially be increased by additional practical system
optimization. Nevertheless, attainable transmission distances for RZ-Star 16QAM
are limited to about 800 km at BER=10−3 due to the SPM-induced mean nonlinear
phase shift and significantly reduced in comparison to RZ-8PSK.
1E-1
1E-2 RZ-Star 16QAM

Fig. 8.19 Transmission dis-
tances attainable for PDM
BER
RZ-Star 16QAM in com- 1E-3

RZ-QPSK
parison to RZ-8PSK and RZ-8PSK
RZ-QPSK at 10 Gbaud when 1E-4

nonlinear phase shift com-
PDM
pensation is not employed. 1E-5
Results were determined by 0 1000 2000 3000 4000 5000 6000
computer simulations in [4]. Transmission distance [km]
8.4 Nonlinear Phase Shift Compensation in Systems with Inline CD Compensation 223
The distortions caused by the SPM-induced mean nonlinear phase shift can be
partly compensated for using the simple optical compensator depicted in Fig. 7.18a.
The optical phase is rotated back proportionally to the instantaneous power at the
compensator input. The proportionality factor depends on the link parameters and
the location where the compensator is placed within the system. In multi-span trans-
mission systems with inline CD compensation, the compensator could principally
be placed behind each fiber in each span (denoted here as “Case A”), but this is not a
very practical approach. Another, more practical option is to place the compensator
only in front of the coherent receiver (denoted here as “Case B”). Both compensa-
tion schemes are indicated in Fig. 8.17. It should be noted, that in both cases com-
pensation is not ideal, since the intensity shape of the propagating signal changes
along the fiber and interaction between chromatic dispersion and SPM prevents a
complete compensation of the mean nonlinear phase shift. Moreover, the compen-
sator depicted in Fig. 7.18a does not work ideally for PDM where distortions due
to cross-polarization effects necessitate a more complex compensator for achieving
best performance. Furthermore, the nonlinear phase noise (see Sect. 6.4.3) should be
considered additionally in practical systems and an appropriate scaling factor α N L
should be found to reduce the variance of the nonlinear phase shift [2].
Nevertheless, both compensation schemes presented here, which use only the
simple compensator shown in Fig. 7.18a, lead to a significant transmission reach
enhancement. This is illustrated in the case of RZ-Star 16QAM transmission for
single-polarization in Fig. 8.20a and for PDM in Fig. 8.20b, assuming optimized
launched powers into the SSMF and DCF.
Reach enhancement single-polarization Reach enhancement for PDM

1E-2 1E-2
a Case B, αN L= 1 b
1E-3 1E-3 Case A, αN L= 0.85

BER
BER
w/o comp. w/o comp. Case B, αN L= 0.85
1E-4 Case A, αN L= 1
1E-4 Case B, αN L= 1
RZ-Star 16QAM RZ-Star 16QAM

1E-5 1E-5
0 500 1000 1500 2000 2500 0 500 1000 1500 2000 2500
Transmission distance [km] Transmission distance [km]
Fig. 8.20 Enhancement of transmission reach for RZ-Star 16QAM at 10 Gbaud for single polar-
ization (a) and PDM (b) using different schemes of nonlinear phase shift compensation based on
the optical compensator depicted in Fig. 7.18a.
In single-polarization systems, the transmission lengths attainable with RZ-Star

16QAM at 10 Gbaud can be increased from 900 km to about 1500 km when plac-
ing the compensator only at the receiver (Case B) and almost doubled to 1750 km
when using a compensator behind each fiber (Case A). However, compensation with
this simple optical compensator does not work equally effective for PDM where
transmission distances are increased to 1100 km and 1200 km for Case B with scal-
ing factors of α N L = 1 and α N L = 0.85, respectively, and to 1400 km for Case A
(with α N L = 0.85). It can be observed from Fig. 8.20b that scaling factors not equal
to one are optimal for PDM due to nonlinear cross-polarization effects. Nonlinear
phase noise was neglected in these investigations.
8.4.2 Systems with Electrical CD Compensation
When chromatic dispersion is not compensated for periodically in each transmission

section but solely by an electrical CD compensation module within the receiver (see
Fig. 8.17; the DCF and the optical amplifier in front of the DCF are then removed
from the transmission link), the mean nonlinear phase shift difference between sym-
bols with different power levels is smaller because the symbol power levels become
indistinguishable after certain transmission distances due to chromatic dispersion.
In Fig. 8.21a and Fig. 8.21b, the received constellation diagrams before digital phase
estimation within the receiver in systems with inline CD compensation after 960 km
are shown for SSMF input powers of -5 dBm (optimal) and -1 dBm, respectively.
The mean nonlinear phase shift difference between symbols of the different inten-
sity rings can be clearly seen as the limiting degradation effect. On the contrary, the
relative nonlinearity-induced phase difference of both rings is smaller in systems
without optical inline CD compensation. This becomes apparent from the constel-
lation diagram depicted in Fig. 8.21c which is received at 1600 km after electrical
CD compensation when an optimal SSMF input power of -1 dBm is chosen.
Optical inline CD compensation Electrical CD comp. within the RX
a b c
960km, PSMF = -5dBm 960km, PSMF = -1dBm 1600km, PSMF = -1dBm
Fig. 8.21 RZ-Star 16QAM constellation diagrams received in systems with optical inline CD com-
pensation (a, b) and electrical CD compensation at the receiver (c) for selected transmission dis-
tances and fiber input powers.
Even without nonlinear phase shift compensation, transmission distances of

1700 km (single-polarization) and 1500 km (PDM) can be bridged in systems with
electrical CD compensation at the receiver, as illustrated in Fig. 8.22. These dis-
tances are similar to or even greater than in systems with optical inline CD compen-
sation which additionally use nonlinear phase shift compensation.
8.4 Nonlinear Phase Shift Compensation in Systems with Electrical CD Compensation 225
1E-2
Fig. 8.22 Distances attainable Inline CD
for RZ-Star 16QAM in sys- compensation
tems with optical inline CD 1E-3 PDM

Single-pol.
compensation and electrical
BER
Single-pol.
CD compensation at the re- PDM
ceiver for single-polarization 1E-4 Electrical CD

compensation
and PDM at 10 Gbaud without
nonlinear phase shift compen- RZ-Star 16QAM
sation, determined in [4]. 1E-5
0 500 1000 1500 2000 2500
Transmission distance [km]
Transmission distances in systems with electrical CD compensation at the re-

ceiver can be further increased by compensating for the small relative phase dif-
ference of both rings observable in Fig. 8.21c. Alternatively or additionally, signal
distortions through the SPM-induced mean nonlinear phase shift can be reduced
by means of adaptive digital equalization within the receiver or by applying pre-
distortion techniques on the transmitter side. Both techniques are also applicable to
systems with optical inline CD compensation.
In summary, the SPM-induced mean nonlinear phase shift limits achievable dis-
tances for optical Star 16QAM transmission in multi-span systems with optical
inline CD compensation when it is not being compensated for. However, trans-
mission distances can be increased from 900 km / 800 km (without compensation)
to 1750 km / 1400 km for single-polarization / PDM using the nonlinear phase shift
compensation described above. The SPM-induced mean nonlinear phase shift is
less critical in systems with electrical CD compensation at the receiver, in which
distances of 1700 km / 1500 km can be attained even without nonlinear phase shift
compensation.
The research results which have been described in this chapter are just a first
small step towards the characterization of high-order modulation formats with re-
spect to their properties for optical multi-span transmission. On the one hand, spec-
tral efficiencies can be increased by the application of high-order modulation for-
mats, but on the other hand, transmission distances are decreased—to evaluate to
what exact extent is an important issue of future research. Currently, more and more
research groups become interested in this field. In their investigations, the trans-
mission lengths and spectral efficiencies attainable in WDM systems are a matter of
particular interest. Recently, impressive experimental results have been shown in [9]
and [8] where a record spectral efficiency of 4.2 bit/s/Hz and a new record C-band
capacity of 17 Tbit/s could be obtained with RZ-8PSK. Moreover, a transmission
distance of 315 km has been demonstrated with Square 16QAM in a 10 × 112 Gbit/s
WDM environment in [7]. Goals of future research in this field are discussed in de-
tail later on in Sect. 9.4.
References
1. Freund, R., et al.: 30 Gbit/s RZ-8-PSK transmission over 2800 km standard single mode fi-
bre without inline dispersion compensation. In: Proceedings of Optical Fiber Communication
Conference (OFC), OMI5 (2008)
3. Louchet, H., et al.: Improved DSP algorithms for coherent 16-QAM transmission. In: Proceed-
ings of European Conference on Optical Communication (ECOC), Tu.1.E.6 (2008)
4. Seimetz, M.: System degradation by the SPM-induced mean nonlinear phase shift in opti-
cal QAM transmission. In: Proceedings of Optical Fiber Communication Conference (OFC),
JWA38 (2009)
5. Seimetz, M., et al.: Coherent RZ-8PSK transmission at 30Gbit/s over 1200km employing ho-
modyne detection with digital carrier phase estimation. In: Proceedings of European Confer-
ence on Optical Communication (ECOC), vol. 3, pp. 265–266 (2007)
Star 16QAM systems in comparison to 8PSK and QPSK at 10Gbaud. In: Proceedings of Optical
Fiber Communication Conference (OFC), OTuN2 (2009)
(2008)
Chapter 9
Performance Trends
Abstract This chapter summarizes the major trends in system performance which
result from migration to higher-order modulation formats. Tendencies regarding rel-
evant system parameters such as noise performance, laser linewidth requirements,
chromatic dispersion tolerances and self phase modulation tolerances are identi-
fied. The principal impact of the choice of particular transmitter configurations and
receiver schemes on system performance is outlined. Furthermore, the reduction
of transmission distances achievable with high-order modulation formats is briefly
discussed. Finally, goals for future research in this field are pointed out.
In order to assess the potential application of highly spectral efficient modulation

formats in their networks, system designers would like to know how high-order
modulation formats typically behave with respect to relevant performance degra-
dation effects and attainable transmission distances. Thus, in addition to system
complexity issues which have been detailed in the first part of this book, two ma-
jor questions which have importance for system design are: How does the system
react to transmission impairments? Which transparent optical transmission lengths
can be bridged? The second part of the book has provided insight in the system
and transmission characteristics of a wide range of phase and quadrature amplitude
modulation formats for single-span and multi-span optical fiber transmission and
has addressed these questions.
9.1 Migration to Higher-Order Formats—Parameter Trends
In Chap. 7, several system parameters of interest such as noise performance, opti-

mal receiver filter bandwidths, laser linewidth requirements, chromatic dispersion
tolerances and self phase modulation tolerances have been discussed by looking
at back-to-back and single-span system configurations. Concrete parameter values
are summarized in Table 7.2, Table 7.4 and Table 7.6 for systems which use direct


228 9 Performance Trends
detection, homodyne differential detection and homodyne synchronous detection

with digital phase estimation, respectively, assuming the use of serial transmitter
configurations (described in Chap. 2). By using this material, the individual system
properties and parameter tolerances of particular modulation formats can be rated in
comparison with other formats. Clear performance tendencies valuable for system
design can be ascertained and these are briefly summarized in the following para-
graphs.
The migration from traditionally used modulation formats to formats with more
bits per symbol leads to a reduction of the symbol rate and narrowed spectral widths.
Therefore, higher spectral efficiencies and per fiber capacities can be realized. This
is the main motivation for a system upgrade to higher-order modulation. At the same
time, migration to higher-order modulation formats strongly influences system per-
formance. Positive effects are the improvement of chromatic dispersion tolerances
and an increased robustness against polarization mode dispersion at reduced symbol
rates. However, multi-span long-haul optical fiber transmission systems are typically
limited by noise and Kerr nonlinearities. These impairments become more critical
for higher-order modulation formats, resulting in a reduction of transmission reach.
The noise performance degrades significantly for high-order PSK / DPSK formats
in particular, whereas self phase modulation has a critical impact on QAM signals
in particular, due to the unequal nonlinear phase shifts obtained by symbols with
different power levels. The latter effect can partly be compensated for by nonlinear
phase shift compensation, but is an inherent problem of optical QAM transmission.
Moreover, the laser linewidth requirements increase with an increasing number of
phase states and decreasing symbol rates. Currently, they are hardly able to be ful-
filled at 40 Gbit/s by lasers for commercial use in case of 16PSK, Square 16QAM
and Square 64QAM in systems with homodyne synchronous detection, even when
feed forward digital phase estimation is used. Next, the system performance param-
eters just mentioned are discussed individually in some more detail.
Noise performance: Noise performance degrades with an increasing number of
bits per symbol as the Euclidean distances between the symbols become smaller.
Assuming a fixed data rate, quaternary phase modulation shows a noise perfor-
mance similar (synchronous detection) or only about 1 dB worse (differential detec-
tion) than binary phase modulation. But high performance penalties arise for higher-
order phase modulation formats (8PSK requires an OSNR about 3.5 dB higher than
QPSK, 16PSK an OSNR about 4.5 dB higher than 8PSK). High-order QAM formats
exhibit a significantly better noise performance than high-order phase modulation
formats for a certain number of bits per symbol, in particular Square QAM for-
mats. In comparison with 16PSK, Square 16QAM has an OSNR performance gain
of about 4 dB, for instance.
Laser linewidth requirements: Laser linewidth requirements increase with an in-
creasing number of phase states, since a certain level of laser phase noise is more
critical for closer phase distances. In addition, the reduction of the symbol rate
makes the laser phase noise more critical for modulation formats with a higher num-
ber of bits per symbol. In systems with direct detection, the linewidth requirements
9.1 Migration to Higher-Order Modulation Formats—System Parameter Trends 229
are relatively relaxed. Even 16DPSK can tolerate a linewidth of about 1 MHz at
r B = 40 Gbit/s. In the case of homodyne differential detection, the effective phase
noise taking effect on the electrical differential demodulation process is determined
by the beat-linewidth. The linewidth requirements on each laser are approximately
doubled in comparison with direct detection when the same linewidths are assumed
for the signal laser and the LO. In systems with homodyne synchronous detection,
the laser phase noise becomes a very critical parameter for high-order formats such
as 16PSK, Square 16QAM and Square 64QAM, even when using feed forward dig-
ital phase estimation which is not impaired by processing delay. When the feed
forward M-th power block scheme described in Sect. 3.5.3 is employed for phase
estimation, the required linewidths at 40 Gbit/s are in the range of 240 kHz, 120 kHz
and 1 kHz for 16PSK, Square 16QAM and Square 64QAM, respectively. These re-
quirements are not able to be fulfilled with lasers currently available for commercial
use. A commercial application of those modulation formats in systems with syn-
chronous detection necessitates the development of low-cost lasers with very low
linewidth specifications. Moreover, the application of improved phase estimation
schemes offers a way of further relaxing the requirements on laser linewidth.
Chromatic dispersion tolerances: The principal tendencies for the chromatic dis-
persion tolerances of high-order modulation formats become clear when looking at
signals with RZ pulse shape. At a fixed data rate, the dispersion tolerances increase
with the order of the modulation format due to the reduced symbol rates and the re-
lated narrower spectral widths. As regards high-order modulation signals with NRZ
pulse shape, the chromatic dispersion tolerances are smaller in comparison to RZ
and depend strongly on the chirp behavior of the transmitter. When a chromatic dis-
persion of 320 ps/nm has been accumulated along the transmission link, the receiver
sensitivity penalty at BER=10−4 is still below 1 dB for modulation formats of at
least fourth order in systems using homodyne synchronous detection and RZ pulse
shape, even when no electrical chromatic dispersion compensation is implemented
within the receiver.
Self phase modulation tolerances: Generally, the modulation formats tend to re-
sult in SPM tolerances which become worse with an increasing number of phase
states. Each symbol of an idealized phase modulated signal with constant power
would be effected by the same nonlinear phase shift during fiber propagation if
there was no other effect than SPM. In this case, the received constellation would be
rotated, but not distorted. However, chromatic dispersion and SPM interact during
propagation. Power fluctuations induced by chromatic dispersion cause the nonlin-
ear phase shifts obtained by the symbols to become different and the received con-
stellation diagrams to become distorted in amplitude and phase. When QAM signals
have been propagated through the fiber, the constellation diagrams are deformed
even when the chromatic dispersion is not taken into account, since symbols with
different power levels are effected by different mean nonlinear phase shifts. This
additional effect on them points to an inherent problem of optical QAM transmis-
sion and is the reason for the poor SPM performance of all QAM formats. However,
relatively simple techniques can be employed to compensate for this effect.
Polarization mode dispersion and nonlinear phase noise: The impact of polar-
ization mode dispersion and nonlinear phase noise on the transmission of optical sig-
nals with high-order modulation was only briefly discussed in Sect. 6.4. In the same
manner as for chromatic dispersion, high-order modulation formats offer a way of
relaxing the requirements on polarization mode dispersion since a certain group de-
lay difference has a smaller impact on neighboring pulses for reduced symbol rates
and longer pulse durations. Moreover, it was presumed in Sect. 6.4 that nonlinear
phase noise may become a less dominant effect for higher-order modulation formats
when compared to the linear phase noise because the transmission lengths achiev-
able with higher-order formats are shorter. However, this must be proved by future
research.
9.2 Impact of the System Configuration and Pulse Shape
The second part of the book has also discussed the impact of the choice of particu-
lar system configurations on system performance. Nearly all transmitter and receiver
structures described in the first part of this book have been examined in this discus-
sion. This allows the transmitters and receivers to be rated not only on the basis
of their complexity and practical feasibility as done in Chap. 4, but also according
to aspects of their system performance. Moreover, differences between high-order
modulation signals with NRZ and RZ (50% duty cycle) pulse shapes have been ex-
posed. Parameter values of systems with direct detection and homodyne differential
detection are summarized in Table 7.3 and Table 7.5, respectively, where various
8PSK / 8DPSK system configurations which comprise different transmitter config-
urations with individual signal characteristics and different decision schemes are
taken into consideration. Parameter values of systems with homodyne synchronous
detection considering several Square 16QAM transmitter configurations are given
in Table 7.7. In what follows, the major performance tendencies are outlined.
Impact of the transmitter configuration and pulse shape: For any modulation
format, the transmitter can be composed of different structures. The optical output
signals of the respective transmitters exhibit individual properties in signal shape,
spectrum, symbol transitions and chirp characteristics. These signal properties have
a crucial influence on the overall system performance. Generally, transmitter con-
figurations where optical power is reduced during symbol transitions show a bene-
ficial system behavior. Firstly, noise performance is improved due to higher peak-
to-average power ratios. Secondly, chromatic dispersion tolerances are increased
because chirp does not appear simultaneously with high power levels. Thirdly, the
self phase modulation tolerances are enhanced, since power fluctuations induced
by chromatic dispersion are less distinct when the impact of chirp is insignificant.
Moreover, higher deviations from the optimal bandwidths of the optical and elec-
trical receiver filters can be tolerated. For these reasons, the parallel DPSK / Star
QAM transmitter outperforms the serial DPSK / Star QAM transmitter for NRZ
9.2 Impact of the System Configuration and Pulse Shape on System Performance 231
pulse shape. In the case of Square QAM and NRZ, the conventional IQ transmitter is
superior to other Square QAM transmitters where symbol transitions are conducted
at higher power levels. Furthermore, RZ signals show a predominantly superior sys-
tem performance in comparison with NRZ signals and their transmission charac-
teristics are almost independent of the transmitter structure used. All tendencies
mentioned here tend to become more significant with increasing modulation format
order, because the percentage of unfavorable symbol transitions with high chirp then
increases. To sum up, transmitter setups where high chirp appears simultaneously
with high power levels should be avoided. However, it should be noted that the con-
clusions drawn here may no longer be valid in systems where optical signals are
narrowly filtered at the transmitter output or the transmission link and where sig-
nal characteristics undergo a significant change, for instance in WDM systems with
small channel spacings.
Impact of the employed receiver concept: Apart from heterodyne detection,
which was not dealt with within this book due to high receiver bandwidth require-
ments, three detection schemes can be employed for the reception of optical high-
order modulation signals: direct detection, homodyne differential detection and ho-
modyne synchronous detection. Homodyne synchronous detection with carrier syn-
chronization through digital phase estimation features the principal advantage of
an improved noise performance compared with the other two detection schemes.
The performance gain depends on the modulation format used and is about 2 dB
on average—as long as laser phase noise does not degrade receiver sensitivity. But
laser linewidth requirements for homodyne synchronous detection are critical for
many of the high-order formats, so differential detection schemes which show more
relaxed linewidth requirements can probably achieve similar receiver sensitivities in
practice and should then be preferred, since they feature simpler implementation.
However, it should be borne in mind that Square QAM formats are conveniently
detected only by synchronous techniques and feature the best noise performance for
a given number of bits per symbol. The general benefits of coherent detection com-
pared to direct detection can be exploited using homodyne synchronous as well as
homodyne differential detection. Transmission impairments such as chromatic dis-
persion, polarization mode dispersion, and possibly even Kerr nonlinearities, can be
compensated for efficiently in the electrical domain. WDM channels can be selected
with high selectivity using a tunable local oscillator. The great potential of coherent
detection combined with the electronic compensation of chromatic dispersion for
transmission reach and cost savings was demonstrated in the multi-span transmis-
sion experiments without inline CD compensation described in Chap. 8. However,
direct detection has its benefits as well, since it features the most relaxed linewidth
requirements and does not require any frequency or phase synchronization or polar-
ization control. Moreover, the positioning of the thresholds for data recovery influ-
ences the system noise performance. In systems limited by shot-noise or amplifier
noise, the thresholds in circular constellations should optimally be arranged radially
to maximize distances between symbol states and thresholds.
Tolerance to deviations from the optimal receiver filter bandwidths: Optimal

receiver filter bandwidths for systems with direct detection in back-to-back con-
figuration have been discussed in Sect. 7.1.2. It has been shown that RZ signals
tolerate much higher deviations from the optimum optical and electrical receiver
bandwidths than NRZ signals. In the case of NRZ, the parallel transmitter features
reduced power during symbol transitions and proves to be more tolerant than the se-
rial transmitter—as far as deviations from the optimal bandwidths of the optical and
electrical receiver filters are concerned. When the serial transmitter configuration is
used in combination with NRZ pulse carving, the signals become very vulnerable
to narrow optical filtering with an increasing order of phase modulation. For in-
stance, an optical bandwidth of at least 3 · r S is then necessary for 16DPSK to avoid
high performance penalties. Generally, in more complex systems such as multi-span
transmission systems comprising many fiber spans and optical filtering on the link,
the optimal receiver filter bandwidths depend on the setup of the whole system and
must be optimized for every system configuration individually.
9.3 Reduction of Attainable Transmission Distances
In Sect. 9.1, it was discussed that the migration to modulation formats with more bits
per symbol leads to higher spectral efficiencies and higher chromatic dispersion and
polarization mode dispersion tolerances. At the same time, these are accompanied
by inferior noise performance and smaller self phase modulation tolerances. The
two latter effects considerably determine the distances achievable for single-channel
multi-span long-haul transmission. Thus, systems applying high-order modulation
formats show a reduced transmission reach.
The evaluation of transmission distances attainable using high-order modula-
tion formats represents an important field of current and future research. Some
recent experimental results obtained in the research group of the author have
been presented in Chap. 8. These confirm the expected reduction of transmis-
sion reach when migrating to higher-order formats. In multi-span transmission sys-
tems with optical inline CD compensation, distances of 3700 km / 1360 km / 600 km
and 3200 km / 960 km / 400 km could be bridged with RZ-QPSK / RZ-8PSK / RZ-
Star 16QAM for single-polarization and polarization division multiplexing, respec-
tively, assuming a common symbol rate of 10 Gbaud, single-channel transmission
and a BER reference of 10−3 . In the case of single-polarization, the transmis-
sion distance for RZ-Star 16QAM could be increased to about 1000 km by ap-
plying compensation of the mean nonlinear phase shift. Moreover, the transmis-
sion lengths can be increased by replacing optical inline CD compensation with
an electrical CD compensation at the receiver. This way, transmission distances
of > 6000 km / 2800 km / 950 km could be attained for single-polarization RZ-
QPSK / RZ-8PSK / RZ-Star 16QAM. The transmission distance of 950 km for RZ-
Star 16QAM was achieved in a five-channel WDM environment and could even
be extended to 1400 km by using adaptive electronic equalization. Results from
9.4 Goals of Future Research in the Field of High-Order Modulation 233
computer simulations indicate that a further extension of transmission reach is pos-

sible for the higher-order formats by optimizing the practical system setups. Gen-
erally, as regards optical QAM transmission, a compensation of the SPM-induced
mean nonlinear phase shift is essential for achieving longer distances.
It should be noted that the comparison of transmission distances in this book
assumes a common symbol rate for all the modulation formats. The differences
detected between transmission distances would have been smaller for comparisons
made at the same data rate.
9.4 Goals of Future Research
In the future, an extension of network capacities is desirable while maintaining an

attractive system reach. Network capacities can be increased by applying high-order
modulation formats which provide a higher spectral efficiency. The reduction of
transmission distances aligned with this can be mitigated by attention to the prac-
tical optimization and high-quality fabrication of the system components required
for generating and detecting optical signals with high-order modulation, but also
by diminishing transmission impairments such as noise and fiber nonlinearities us-
ing low-noise optical amplification and Kerr effect compensation. Future research
should cover the following areas:
Transmission distances achievable with high-order modulation formats: Anal-
ysis of multi-span fiber transmission systems with high-order modulation is still at
an early stage. Some initial results, exploring the transmission distances for single-
channel transmission, have been presented in this book. In order to enable trans-
mission of signals with high-order modulation over long distances, more detailed
investigations are indispensable. Link configurations must be optimized for optimal
fiber input powers and dispersion maps in systems with optical inline chromatic dis-
persion compensation—individually for any format and for a wide range of formats.
A key issue is the development of techniques which will efficiently compensate for
fiber nonlinearities. It is necessary to pursue the question of how the inherent phase
distortions of QAM signals through the unequal nonlinear phase shifts obtained by
symbols with different power levels can best be compensated for in multi-span trans-
mission systems, when single-channel and WDM systems are taken into consider-
ation. Furthermore, the influence of the nonlinear phase noise must be estimated
to be able to finally evaluate the question: Which transparent optical transmission
lengths can be bridged?
Behavior of high-order modulation formats in WDM systems: The transmis-
sion lengths and channel spacings achievable with high-order modulation formats
in WDM systems are a matter of particular interest. Attention must paid to some
aspects which have not been considered in detail within this book: channel filter-
ing, crosstalk and inter-channel nonlinearities. The channel spacing attainable de-
pends on the signal bandwidth and on how narrowly optical signals can be filtered.
Narrower channel spacing induces higher penalties due to cross phase modulation
and four wave mixing. Thus, the system penalty induced by narrow optical filtering
and the impact of linear and nonlinear inter-channel crosstalk must be determined
for the various modulation formats. Moreover, narrow filtering has an influence on
the signal’s transmission properties, so the conclusions drawn within this book con-
cerning the transmission behavior of NRZ and RZ signals may require amendment.
Another issue stems from limitation of the maximal power per fiber which is regu-
lated in WDM systems and limited practically by the maximum output power of the
optical amplifiers. In the case of higher spectral efficiencies and thus a higher num-
ber of channels within a certain bandwidth, the power available for a single channel
is reduced—perhaps by too much to provide each channel with the optimal fiber
input power determined by a trade-off between noise and fiber nonlinearities. From
this point of view, nonlinear fiber effects may play a less important role in highly
spectral efficient future networks with high-order modulation formats. Future re-
search should address the question: Which transmission lengths, channel spacings
and channel numbers can be obtained in WDM systems?
Capacity and spectral efficiency attainable in WDM systems: Theoretically, if
the fiber were linear and there were no system degradation through fiber nonlin-
earities, spectral efficiency could be increased to infinity by applying modulation
formats of higher and higher order. Thereby, the expected increase of spectral effi-
ciency would be about the ratio of the data rate to the symbol rate. Increasing noise
requirements of the higher-order formats could then be fulfilled by simply launch-
ing more and more power into the fiber. However, in practical transmission systems,
performance degrades due to fiber nonlinearities when the fiber input power is in-
creased. Theoretical calculations show that spectral efficiencies attainable in WDM
fiber transmission systems are upper-bounded. Even if low-noise distributed Raman
amplification is assumed and intra-channel nonlinear effects are compensated for
by applying reverse propagation, spectral efficiency in one polarization is estimated
to be limited to 5 bit/s/Hz and 7 bit/s/Hz for WDM transmission over 2000 km and
1000 km, respectively [3, 4], caused by inter-channel nonlinear effects. A lot of re-
search investigating the properties of practical WDM systems with high-order mod-
ulation must be undertaken in order to finally answer the question: Which spectral
efficiencies and network capacities can be obtained practically, while maintaining
attractive transmission distances?
Increase of the capacity-distance product: Whereas it is evident that the spectral
efficiency and network capacity can be increased by the application of high-order
modulation formats, it has not been clarified yet whether the capacity-distance prod-
uct can be improved, due to the reduced transmission distances. In [2], experiments
with the greatest capacity-distance products have been reviewed. These experiments
use binary and quaternary phase modulation and show the trend that the highest sys-
tem capacities are achieved over the most moderate distances. The stars in Fig. 9.1
indicate capacities and distances attained in the experiments performed in [1, 2, 8].
Moreover, two reference lines are shown in Fig. 9.1, corresponding to constant
capacity-distance products of 10 Pbit/s·km and 40 Pbit/s·km. In [1], the second-best
9.4 Goals of Future Research in the Field of High-Order Modulation 235
Fig. 9.1 Can the capacity- 12000

distance product be increased 41.03 Pbit/s⋅km
10000 DBPSK, Cai et al., 2003
by the application of high-
Distance [km]
order modulation formats? 8000 41.82 Pbit/s⋅km
QPSK, Charlet et al., 2008
The stars indicate results
6000
experimentally achieved in 40 Pbit/s⋅km 11.25 Pbit/s⋅km
[1, 2, 8]. The solid lines are 4000 8PSK, Yu et al., 2008
reference lines for constant

2000 10 Pbit/s⋅km
?
capacity-distance products.
0
0 5 10 15 20 25 30
Capacity [Tbit/s]
capacity-distance product so far, 41.03 Pbit/s·km, has been demonstrated by trans-

mitting 373 RZ-DPSK channels in the C-band and the L-band at 10 Gbit/s over a
transoceanic distance of 11000 km while using hybrid EDFA / Raman amplification
and FEC. In [2], the record capacity-distance product of 41.82 Pbit/s·km has been
attained, sending 164 PDM-QPSK channels at a channel data rate of 100 Gbit/s over
a length of 2550 km while using Raman amplification and coherent detection. Very
recently, initial impressive results using RZ-8PSK have been obtained. A record
capacity of 17 Tbit/s in the C-band was demonstrated by sending 161 polarization
division multiplexed RZ-8PSK WDM channels over 662 km ultra-low loss fiber,
with a record spectral efficiency of 4.2 bit/s/Hz [8]. The capacity-distance product
thereby achieved is 11.25 Pbit/s·km and four times smaller than the record prod-
uct of 41.82 Pbit/s·km obtained in [2], due to the comparatively small transmission
length. However, latent potential for further practical system optimization and for
increasing experience with systems applying high-order modulation leaves the ques-
tion open: Can the capacity-distance product be improved by applying high-order
modulation formats?
Utilization of polarization: Polarization information provides an additional de-
gree of freedom in optical fiber transmission systems. Utilizing polarization, the
spectral efficiency of any modulation format can be doubled by means of polariza-
tion division multiplexing. Here, how far crosstalk between the multiplexed chan-
nels degrades the performance of systems applying high-order modulation remains
to be clarified in future research. Moreover, the compensation of the SPM-induced
nonlinear phase shift for QAM transmission becomes more critical in the case of po-
larization division multiplexing. In addition to the phase and quadrature amplitude
modulation formats discussed within this book, formats exploiting all the parame-
ters of the electrical field and encoding information additionally into the polarization
are available for optical transmission, so the question remains: How efficiently can
polarization information be exploited in future optical networks?
Practical system optimization: For many system experiments with high-order
modulation currently performed, the main challenge is still a practical optimiza-
tion of the system components. Signal distortions caused by pattern effects and ac-
cumulating in multiple modulator stages have led to implementation penalties in
the Star 16QAM experiment described in Sect. 8.2.1 [6]. In recent Square 16QAM
experiments limited to distances of some hundred kilometers, an enhanced gener-

ation of the multi-level electrical driving signals for the conventional IQ transmit-
ter [7] and an improved practical fabrication of the integrated quad-parallel MZM
transmitter [5] are important steps in achieving improved system performance and
reach. Thus, in order to compose transmitters performing closely to the theoretical
performance limits, it is aimed to develop high-speed integrated optical modulator
structures and electrical level generators of high quality. At the receiver end, it is an
aim to design practical solutions which integrate the whole optical receiver frontend
in a single chip, and to exploit digital signal processing technology to compensate
for performance degradation effects and facilitate the recovery of information. So
an important question is: What are the performance limits of practically optimized
systems employing high-order modulation?
Simulative system evaluation: Performance limits of optical fiber transmission
systems with high-order modulation can also be explored via computer simulations.
But this is not easily accomplished. In Monte Carlo simulations with a high number
of bits, the computational effort can become unmanageable when multi-span WDM
systems with long transmission distances, high per channel powers and a high num-
ber of channels are investigated. Currently available tools for semi-analytical BER
estimation can not consider all transmission effects and can not easily be imple-
mented for complex receiver configurations with complicated noise statistics at the
decision gates, for instance the homodyne receiver with digital equalization and
phase estimation. Furthermore, the data sequence length requirements can not be
fulfilled when high magnitudes of accumulated chromatic dispersion lead to inter-
symbol interference across many symbols—being a particular problem for systems
without optical inline CD compensation. This all raises the question: How can the
performance of optical multi-span transmission systems with high-order modulation
be evaluated exactly via computer simulations?
Having finished all this research and answered all the questions posed above, it
is possible that the final question may arise: Which is the most favorable modulation
format? However, the ultimate modulation format will not be able to be identified.
Each modulation format has its own benefits and drawbacks with respect to partic-
ular system parameters, so its suitability depends on the respective specifications
of the system which it operates in and on the field of application, for instance the
network segment which it is intended for. The future will show for which system ap-
plications and in which network segments particular high-order modulation formats
can most profitably be deployed.
References
1. Cai, J.X., et al.: A DWDM demonstration of 3.73 Tb/s over 11,000 km using 373 RZ-DPSK
channels at 10 Gb/s. In: Proceedings of Optical Fiber Communication Conference (OFC),
PD22 (2003)
References 237
2. Charlet, G., et al.: Transmission of 16.4Tbit/s capacity over 2550km using PDM QPSK modula-
tion format and coherent receiver. In: Proceedings of Optical Fiber Communication Conference
(OFC), PDP3 (2008)
3. Essiambre, R.J., et al.: The capacity of fiber-optic communication systems. In: Proceedings of
Optical Fiber Communication Conference (OFC), OTuE1 (2008)
4. Essiambre, R.J., et al.: Exploring capacity limits of fibre-optic communication systems. In:
Proceedings of European Conference on Optical Communication (ECOC), We.1.E.1 (2008)
(2008)
Star 16QAM systems in comparison to 8PSK and QPSK at 10Gbaud. In: Proceedings of Optical
Fiber Communication Conference (OFC), OTuN2 (2009)
(2008)
Appendix A
Differential Encoders for 8DPSK / 16DPSK
In this appendix, the logical circuits of the differential encoders for 8DPSK and
16DPSK are specified. In Sect. 2.4.3, differential encoding for DPSK formats is
described and the logical relations for the differential encoders of the serial and par-
allel DQPSK transmitter are derived. In the same manner, the differential encoders
for higher-order DPSK transmitters can be obtained.
Starting from the constellationdiagrams with the original bit mappings given,
which assign the original data bits b1k , b2k , .., bm k to particular symbols, and with
symbol assignments
specified
at the encoder output mapping the differentially en-
coded bits d1k , d2k , .., dm k into particular symbols with absolute phase states, the
truth tables can be established. Subsequently, the corresponding Karnaugh maps can
be evaluated to determine the logical relations of the encoders.
Differential Encoders for 8DPSK Transmitters
In the case of 8DPSK, the truth tables of the encoders have 64 rows. When using
the bit mappings and symbol assignments as defined in Fig. 2.6 and Fig. 2.9 respec-
tively, the logical circuit
of the serial
8DPSK transmitter, which relates the current
encoder output bits d1k , d2k, d3k to the current data bits b1k , b2k , b3k and the
previous encoder output bits d1k−1 , d2k−1 , d3k−1 , is specified by
d1k = b1k d1k−1 d2k−1 d3k−1 + b1k b3k d1k−1 d2k−1 + b2k b3k d1k−1 d3k−1
+ b2k d1k−1 d2k−1 d3k−1 + b1k b2k d1k−1 d2k−1 + b2k b3k d1k−1 d3k−1
+ b1k b3k d1k−1 d2k−1 + b1k d1k−1 d2k−1 d3k−1 + b1k b2k d1k−1 d3k−1
+ b1k b3k d1k−1 d2k−1 d3k−1 + b1k b3k d1k−1 d2k−1 d3k−1 , (A.1)
d2k = b3k d2k−1 d3k−1 + b3k d2k−1 d3k−1 + b1k b2k d2k−1 d3k−1
+ b1k b2k d2k−1 d3k−1 + b1k b2k d2k−1 d3k−1 + b1k b2k d2k−1 d3k−1 , (A.2)
239
240 A Differential Encoders for 8DPSK / 16DPSK
d3k = b1k b2k b3k d3k−1 + b1k b2k b3k d3k−1 + b1k b2k b3k d3k−1
+ b1k b2k b3k d3k−1 + b1k b2k b3k d3k−1 + b1k b2k b3k d3k−1
+ b1k b2k b3k d3k−1 + b1k b2k b3k d3k−1 . (A.3)
Section 2.4.3 has demonstrated that the setup of the differential encoder changes
when the parallel transmitter configuration is employed instead of the serial one.
When the bit mappings and symbol assignments shown in Fig. 2.6 and Fig. 2.9 are
assumed, respectively, the logical relations describing the differential encoder of the
parallel 8DPSK transmitter can be derived as
d1k = b1k d1k−1 d2k−1 d3k−1 + b2k b3k d1k−1 d3k−1 + b1k b2k d1k−1 d3k−1
+ b1k b2k d2k−1 d3k−1 + b1k b3k d2k−1 d3k−1 + b2k b3k d1k−1 d3k−1
+ b1k b3k d2k−1 d3k−1 + b1k b2k d1k−1 d3k−1 + b2k d1k−1 d2k−1 d3k−1 , (A.4)
d2k = b2k d1k−1 d2k−1 d3k−1 + b1k b3k d1k−1 d3k−1 + b1k d1k−1 d2k−1 d3k−1
+ b1k b2k d2k−1 d3k−1 + b2k b3k d2k−1 d3k−1 + b1k b3k d1k−1 d3k−1
+ b2k b3k d2k−1 d3k−1 + b1k b2k d1k−1 d3k−1 + b1k d1k−1 d2k−1 d3k−1 , (A.5)
d3k = b1k b2k b3k d3k−1 + b1k b2k b3k d3k−1 + b1k b2k b3k d3k−1
+ b1k b2k b3k d3k−1 + b1k b2k b3k d3k−1 + b1k b2k b3k d3k−1
+ b1k b2k b3k d3k−1 + b1k b2k b3k d3k−1 . (A.6)
The relations for the third encoder output bit, given by (A.3) and (A.6), can be
seen to be identical.
Differential Encoder for the Serial 16DPSK Transmitter
In the case of 16DPSK, the differential encoders are quite complex. Here, only the
differential encoder for the serial 16DPSK transmitter is described. The derivation
of its logical relations necessitates the evaluation of a truth table with 256 rows. The
relation for the encoder output bit d1k , for instance, has 30 OR combined terms, each
consisting of 4-7 AND combined inputs.
Based on the bit mapping defined in Fig. 2.6 and the symbol assignment shown
in Fig. 2.9, the logical circuit of the differential encoder of the
serial 16DPSK trans-

mitter can be derived and the current encoder output bits d1k , d2k , d3k , d4k can
A Differential Encoders for 8DPSK / 16DPSK 241

be shown to be related
to the current data bits b1k , b2k , b3k , b4k and the previous
encoder output bits d1k−1 , d2k−1 , d3k−1 , d4k−1 with
d1k = b1k b2k d1k−1 d2k−1 + b1k b2k d1k−1 d2k−1 + b1k b2k d1k−1 d2k−1
+ b1k b2k d1k−1 d2k−1 + b1k b3k b4k d1k−1 d2k−1 + b2k b3k b4k d1k−1 d2k−1
+ b2k b3k d1k−1 d3k−1 d4k−1 + b2k b4k d1k−1 d2k−1 d3k−1 + b2k d1k−1 d2k−1 d3k−1 d4k−1
+ b1k b3k b4k d1k−1 d2k−1 + b2k b4k d1k−1 d2k−1 d3k−1 + b2k b3k d1k−1 d3k−1 d4k−1
+ b2k b3k b4k d1k−1 d2k−1 + b1k b2k d1k−1 d3k−1 d4k−1
+ b1k b2k b4k d1k−1 d2k−1 d3k−1 + b1k b2k b3k d1k−1 d2k−1 d4k−1
+ b1k b4k d1k−1 d2k−1 d3k−1 d4k−1 + b2k b3k b4k d1k−1 d2k−1 d4k−1
+ b1k b2k b4k d1k−1 d2k−1 d3k−1 + b2k b4k d1k−1 d2k−1 d3k−1 d4k−1
+ b1k b2k b3k d1k−1 d2k−1 d3k−1 + b2k b4k d1k−1 d2k−1 d3k−1 d4k−1
+ b1k b4k d1k−1 d2k−1 d3k−1 d4k−1 + b2k b3k b4k d1k−1 d2k−1 d4k−1
+ b1k b2k b4k d1k−1 d2k−1 d3k−1 d4k−1 + b1k b2k b4k d1k−1 d2k−1 d3k−1 d4k−1 , (A.7)
d2k = b3k b4k d2k−1 d4k−1 + b3k b4k d2k−1 d4k−1 + b1k b2k b4k d2k−1 d3k−1
+ b1k b2k d2k−1 d3k−1 d4k−1 + b1k b2k b4k d2k−1 d3k−1 + b1k b2k d2k−1 d3k−1 d4k−1
+ b2k b3k d2k−1 d3k−1 d4k−1 + b1k b2k b4k d2k−1 d3k−1 + b2k b3k d2k−1 d3k−1 d4k−1
+ b1k b2k d2k−1 d3k−1 d4k−1 + b2k b3k d2k−1 d3k−1 d4k−1 + b1k b2k b4k d2k−1 d3k−1
+ b1k b2k d2k−1 d3k−1 d4k−1 + b2k b3k d2k−1 d3k−1 d4k−1
+ b1k b2k b4k d2k−1 d3k−1 d4k−1 + b1k b2k b4k d2k−1 d3k−1 d4k−1 , (A.8)
d3k = b4k d3k−1 d4k−1 + b1k b2k b3k d3k−1 d4k−1 + b1k b2k b3k d3k−1 d4k−1
+ b1k b2k b3k d3k−1 d4k−1 + b1k b2k b3k d3k−1 d4k−1 + b1k b2k b3k d3k−1 d4k−1
+ b2k b4k d3k−1 d4k−1 + b1k b2k b3k d3k−1 d4k−1
+ b1k b2k b3k d3k−1 d4k−1 + b2k b4k d3k−1 d4k−1
+ b1k b2k b3k d3k−1 d4k−1 , (A.9)
242 A Differential Encoders for 8DPSK / 16DPSK
d4k = b1k b2k b3k b4k d4k−1 + b1k b2k b3k b4k d4k−1 + b1k b2k b3k b4k d4k−1
+ b1k b2k b3k b4k d4k−1 + b1k b2k b3k b4k d4k−1 + b1k b2k b3k b4k d4k−1
+ b1k b2k b3k b4k d4k−1 + b1k b2k b3k b4k d4k−1
+ b1k b2k b3k b4k d4k−1 + b1k b2k b3k b4k d4k−1 . (A.10)
It can be concluded that the complexity of the differential encoders increases

significantly with an increasing order of the phase modulation.
Appendix B
Theoretical BER Formulas
This appendix compiles theoretical BER formulas for systems with direct detec-
tion and coherent synchronous detection which are well known from the literature
[1–4]. These formulas describe the theoretical SNR requirements of systems with
PSK / DPSK and QAM for idealized receivers with matched filtering. With matched
filtering, the SNR of the decision samples is maximized, becomes independent of
the signal shape and is solely a function of the average symbol energy and the power
spectral density of the noise
E s,avg 1f
SN R = = 2 · OSN R · . (B.1)
N0 rS
In contrast to the system characterization conducted in Chap. 7, the impact of
specific setups used at the transmitter, the pulse shape of the signal, the optical and
electrical receiver filters and the data recovery techniques employed at the receiver
are not considered here. For some cases, the formulas represent approximations.
More details can be found in [1–4].
SNR Requirements of Systems with Direct Detection
Equations (B.2)–(B.5) describe the theoretical SNR requirements of direct detection

systems with OOK, DBPSK, DQPSK and M-ary DPSK, respectively, assuming the
matched filtering case.

1 −S N R/2 SN R
BER OOK = ·e · 1+ , yth = R · Ps,avg /2 (B.2)
2 2

1 −S N R SN R
BER DB PSK = ·e · 1+ (B.3)
2 4
243
244 B Theoretical BER Formulas
∞
3 S N R · e−S N R X sin(nπ/4)
BER DQ PSK = − ·
8 4 n
n=1

SN R SN R 2
· I n−1 + I n+1 (B.4)
2 2 2 2
r !
2 SN R π
BER M DPSK ≈ · er f c · sin (B.5)
log2 M 2 M
In (B.4), In is the n-th order modified Bessel function of the first kind. The BER
of Star MQAM systems can be estimated from the BER formulas for PSK / DPSK
formats by calculating the BER for the inner ring, while assuming the same average
power as for the corresponding PSK / DPSK constellation with an equal number of
phase states. Moreover, an appropriate pre-factor of 2/log2 M must be assigned. In
this calculation, it is not taken into account that the noise levels on the rings are
different for direct detection.
SNR Requirements of Systems with Coherent Synchronous Detection
In (B.6)–(B.9), the theoretical SNR requirements of systems with coherent syn-

chronous detection for 2ASK, BPSK, MPSK and Square MQAM is specified, as-
suming amplifier noise limited detection and matched filtering.
r !
1 SN R
B E R 2AS K = · er f c (B.6)
2 2
1 √
BER B PSK = · er f c SN R (B.7)
2
2 √ π
BER M PSK ≈ · er f c S N R · sin (B.8)
log2 M M
s !
2 1 3S N R
B E R Squar e M Q AM ≈ · 1− √ · er f c (B.9)
log2 M M 2(M − 1)
References 245
In Table B.1, OSNR requirements for B E R = 10−4 calculated with (B.8), (B.9)
and (B.1) are compared with values obtained by computer simulations for the spe-
cific system configurations discussed in Sect. 7.3.1.
Table B.1 Comparison of OSNR requirements calculated with (B.8), (B.9) and (B.1) with values
obtained by computer simulations (sim.) for the system configurations discussed in Sect. 7.3.1,
assuming a data rate of r B = 40 Gbit/s and a reference noise bandwidth of 1λ = 0.1 nm.
Format QPSK 8PSK 16PSK St.16QAM Sq.16QAM Sq.64QAM
SNR [dB], theory 11.8 16.9 22.6 20.0 18.2 24.3

OSNR [dB], theory 10.8 14.2 18.6 16.0 14.2 18.5
OSNR [dB], RZ, sim. 11.2 14.6 18.9 16.1 15.0 18.7
OSNR [dB], NRZ, sim. 13.0 16.9 22.4 18.4 17.2 20.1
The values calculated with the theoretical BER formulas, reflecting the matched
filtering case, show the same relative performance tendencies as observed for the
specific system configurations discussed in Sect. 7.3.1. Moreover, it can be seen
that the theoretical performance limits of matched filtering are approached more
closely in the case of RZ.
References

3. Schwartz, M.: Information, Transmission, Modulation, and Noise. McGraw-Hill (1990)
Index
Symbols Bessel filter 60

Bessel function, modified 164
2 × 4 90◦ hybrid 62, 71, 84 Birefringence 151
3 × 3 coupler 65, 87 Bit error ratio (BER) 132
χ 2 -distribution 137 theoretical performance 243
120◦ hybrid 65 Bit mapping 23, 25, 34
180◦ hybrid 61 rotation symmetric 38, 45, 114
3 dB coupler 61 Block length 103, 106, 190
Block-by-block phase correction 102
A Boltzmann constant 69
Burst-mode transmission 94
Absorption 144 Butterfly equalizer 100
Access networks 6
Active control loop 64 C
Additive white Gaussian noise 69, 134
Amplifier noise 69, 82, 134, 145, 157, 207 Capacity 2, 234
Amplifier noise limited detection 178, 186 Capacity-distance product 2, 7, 234
Amplitude shift keying (ASK) 4, 23 Carrier recovery
Analog-to-digital converter 91, 100 data aided 94
Arg-decision 74, 111 decision directed 94, 110
Arg-operation 21 joint-polarization 100
Argument averaging 109, 188 non data aided 94
ASE-ASE noise 69, 82 Carrier synchronization 6, 94
Asynchronous detection 93 Central Limit Theorem 16
Attenuation 143, 144 Channel cross-channel interference 91
Attenuation coefficient 144 Channel separation 91
Automatic frequency control (AFC) 89, 99 Channel spacing 2, 233
Auxiliary phase coding 4 Chirp 16, 32, 167, 170, 195
Chirp-intensity product 33, 36, 55, 168, 195
B Chromatic dispersion (CD) 145, 167, 196
Chromatic dispersion compensation
Back-to-back transmission 155 electrical 207, 212, 217, 224
Balanced detection 62, 70, 80, 84 optical, inline 207, 208, 213, 221
Balanced OPLL 96 Chromatic dispersion tolerances
Band-pass filter 60 direct detection 166, 175
Bandwidth demands 6 homodyne detection
Beat-linewidth 80, 97, 181 differential detection 182, 183
248 Index
synchronous detection 194, 202 Digital storage oscilloscope 209

summary 229 Digital subscriber line (DSL) 1
Class partitioning 105, 190 Digital-to-analog converter 40
Closed loop M-th power block scheme 107 Direct detection 66
performance 188 complexity 122
Closed loop carrier recovery 94 performance 156, 175
Coherence time 17 Direct detection interference 91
Coherent detection 3, 5, 79 Direct modulation 16
complexity 122 Dispersion compensating fiber (DCF) 147
Combination 85 Dispersion managed fiber 6
Complex envelope 32, 93, 101 Dispersion map 233
Complexity 3 Dispersion shifted fiber (DSF) 147
receiver 122 Dispersion slope 145
transmitter 119 Distributed feedback (DFB) laser 16
Constant modulus algorithm (CMA) 100 Driving signals 24
Constellation diagram 23 Duobinary 4
Continuous wave 16 Duty cycle 22
Conventional IQ transmitter 42, 121, 204
Costas loop 95 E
Cross phase modulation (XPM) 150, 234
intra-channel 150 Effective core area 148
Crosstalk 233 Effective length 149, 152
Cycle slips 104 Eigendecomposition 136
Eigenvalue 136
D Eigenvector 136
Electric flux density 147
Data rate 23 Electric polarization 147
Data recovery 73, 111 Electrical field 16, 79, 143
Data recovery logic 74, 111, 124 Electrical field parameters 3, 79, 235
Data sequences 127, 236 Electronic distortion equalization (EDE) 91,
Data to phase-lock crosstalk 96 92, 100
De Bruijn bit sequence 128 Enhanced IQ transmitter 47, 121, 204
De Bruijn symbol sequence 128 Equalization 6, 95, 100, 174, 225
Decision boundaries 200 Equalizer
Decision circuit 74, 106, 114 frequency domain 100
Decision driven OPLL 96 time domain 100
Degree of compensation 169 Erbium doped fiber amplifier (EDFA) 2, 145
Delay line frequency discriminator 99 Euclidean distance 37, 159
Delay line interferometer (DLI) 61, 135 External cavity laser (ECL) 16
Demultiplexer 27, 103 External modulation 16
Differential decoder 73, 97, 104, 111, 121 Eye opening 131
Differential detection 66, 73, 86 Eye opening penalty (EOP) 131, 167
performance 156, 177 Eye spreading 24, 47
Differential encoder 27, 28, 97, 104, 239
Differential phase shift keying (DPSK) 25 F
Differential quadrant encoding 37, 113
Differentially encoded PSK (DEPSK) 186 Fabry-Perot laser 15
Digital coherent receiver 85, 95, 100, 140 Fast Fourier transform (FFT) 132, 144
Digital differential demodulation 91, 177 Feed forward M-th power block scheme 102
Digital frequency offset estimation 92, 110 performance 188, 190
Digital phase estimation 6, 95, 99 Feed forward carrier recovery 94
performance 186, 188, 202 Field averaging 103, 109, 188
Digital signal processing 6, 91, 99 Filters 59
Digital signal processor (DSP) 99, 103 Finite impulse response (FIR) filter 212
Index 249
Forward error correction (FEC) 2, 235 Karnaugh maps 31

Four wave mixing (FWM) 150, 234 Kerr effect 128, 144, 147
intra-channel 150
Fourier series 134 L
Fractionally spaced equalizer 100
Laplace transform 137
Free space optics 65
Large-amplitude QPSK 52
Frequency noise 16
Laser 15
Frequency offset 86, 89, 92, 93, 99, 110
Laser linewidth 16
Frequency shift keying (FSK) 86
Laser linewidth requirements 5, 98
Frequency synchronization 6, 91, 123
direct detection 164, 175
Fundamental mode 151
homodyne detection
differential detection 181, 183
G
synchronous detection 188, 202
Galois fields 128 summary 228
Gaussian filter 60 Laser phase noise 16, 70, 80, 86, 89, 93
Gray coding 23, 25, 34 Least mean square (LMS) algorithm 100
Group delay 145 Level generator 40, 44, 121
Light emitting diode (LED) 15
H Linear phase noise 152
Linewidth-to-data-rate ratio 165, 192
Heterodyne detection 81, 140, 180 LO-ASE noise 82, 83, 178
Homodyne detection 80 Local loop 1
differential detection 86 Local oscillator 79
complexity 123 tunable 96, 98
performance 177, 183 Loop delay 97, 108, 189
synchronous detection 93, 104, 111 Loop filter 96
complexity 123 Low-pass filter 60, 91
performance 185, 202
M
Hyperbolic filter 110
Mach-Zehnder modulator (MZM) 19
I Matched filter 60, 186, 243
Material dispersion 145
Ideal driving 40, 55 Maximum length sequence 127
Image-rejection receiver 63, 81 Minimum transmission point 20
Impulse shaper 22 Modulator driver 42
Initial phase offset 94, 110 Modulators 18
Intensity 148 Moment generating function (MGF) 137
Intensity detection branch 67, 68, 140 Monte Carlo (MC) simulation 132, 159
Intensity dips 33 Multi-level signaling 23
Intensity modulation 4, 19 Multi-mode interference (MMI) coupler 64
Intensity noise 16 Multi-parallel MZM transmitter 52
Intensity ripples 56 Multi-span transmission 207
Inter-symbol interference (ISI) 128, 147 Multiple DLI receiver 67, 122, 138
Intermediate frequency (IF) 81 Multiple moduli algorithm (MMA) 217
Internet 1, 6
IQ modulator (IQM) 20, 42 N
IQ receiver
with 2 × 4 90◦ hybrid 71 Network scenarios 6
with DLIs 68, 122, 139 Noise figure 157, 221
IQ-decision 74, 111 Non-ideal driving 40, 56
Non-return to zero (NRZ) 4, 21, 176
K Nonlinear index coefficient 148
Nonlinear phase noise 140, 152, 173, 230
Karhunen-Loève expansion 134 Nonlinear phase shift 149, 171, 199, 215
250 Index
mean 152 Phase modulation 5, 18, 25

variance 152 Phase modulator 18
Nonlinear phase shift compensation Phase offset 81, 99
compensators 173 Phase pre-integration 34
direct detection 171 Phase shifter 63, 72
future research 233 Phase tracking 107, 109
homodyne detection Phase unwrapping 104
synchronous detection 199, 202 Photodiode 68, 135
multi-span transmission 216, 221 Planck constant 69
Nonlinear propagation coefficient 144, 149 PM-IM conversion 61, 149
Nonlinear scattering effects 151 PMD-parameter 151
Nonlinear Schroedinger equation 143, 147 Pockels effect 18
Nonlinearities 128, 144, 147, 207, 234 Polarization beam combiner (PBC) 213
Normalization 77 Polarization beam splitter (PBS) 65, 85, 213
Normalized field averaging 105, 107, 188 Polarization control 6, 91, 124
Normalized symbol coordinates 23 Polarization diversity 85, 124
Polarization division multiplexing (PDM) 6,
O 81, 85, 100, 124, 213, 219, 235
Polarization information 81, 235
On-off keying (OOK) 4, 24 Polarization mode dispersion 151, 230
Optical phase locked loop (OPLL) 6, 95, 190 Polarization modulation 81, 235
Optical quadrature frontend 84 Polarization shift keying (POLSK) 81
Optical signal to noise ratio (OSNR) 135 Polarization states 65
Optical time division multiplexing 6 Power spectral density 69, 157
Optical voltage controlled oscillator 98 Pre-amplifier 69, 134
Optimal receiver filter bandwidths 160, 231 Pre-distortion 225
OSNR penalty 165, 167 Preamble 94
OSNR requirements 160 Principle states of polarization (PSP) 151
direct detection 157, 175 Probability density function (PDF) 133, 137
experiments 219 Processing delay 108, 188
homodyne detection Propagation constant 143, 145, 148
differential detection 178, 183 Pseudo random binary sequence 127
synchronous detection 185, 202 Pulse carver 22
summary 228 Pulse shape 4, 21, 176, 230
Push-pull mode 19
P Push-push mode 19
Parallel DPSK transmitter 27, 33, 120, 176 Q

Parallel Star QAM transmitter 36, 120, 221
Parallelization 103 Quad-parallel MZM transmitter 52, 121
Pattern effects 215 Quadrant ambiguity 37, 97, 113
Peak-to-average power ratio 160, 188 Quadrature amplitude modulation 5, 34, 37
Permittivity 147, 148 Quadrature point 19
Phase ambiguity 37, 104, 111 Quantum efficiency 69
Phase detection branch 67, 70 Quantum limit 179
Phase detector 96
Phase differential algorithm 110 R
Phase differentiation 92
Phase diversity receiver 63, 86, 140 Raman amplification 2, 7
performance 177, 180 Raman gain spectrum 151
Phase error 93, 94, 101, 110 Random number seed 128, 132
Phase locked loop (PLL) 81, 98 Random walk 16, 81, 104, 164
Phase locking 6 Rayleigh scattering 144
Phase matching 85 Re-circulating fiber loop 209, 214
Index 251
Receiver configuration, impact 231 Spectral efficiency 2, 7, 81, 234

Receiver matrix 136, 138 Split-step Fourier 144
Receiver sensitivities Spontaneous emission 16
direct detection 157, 175 Spontaneous emission parameter 69
homodyne detection Square QAM 5, 37
differential detection 178, 183 Square QAM coder 43, 46, 49, 121
synchronous detection 185, 202 Star QAM 5, 34
summary 228 Stimulated Brillouin scattering (SBS) 151
Receiver sensitivity 5, 93 Stimulated Raman scattering (SRS) 151
Receiver sensitivity penalty 181 Subcarrier based OPLL 98
Records 6 Susceptibility 147
Rectangular filter 103 Symbol error probability 139
Reference bandwidth 135 Symbol rate 23
Refractive index 18, 145, 147 Symbol-to-symbol phase correction 102
Relative intensity noise (RIN) 17, 91 System memory length 128
Responsivity 68
Return to zero (RZ) 4, 21, 176 T
Reverse propagation 234
Ring ratio 35, 159, 170, 187, 216 Tandem-QPSK transmitter 51, 121, 204
Rise time 22, 55 Thermal noise 69, 135
Timing recovery
S Gardner 100
square 100
Saddlepoint approximation 138 Training sequence 38, 101, 104
Scaling factor 153, 223 Transmission distances, attainable
Segment change detector 104 8PSK 210, 212
Selective gated PLL 98 comparison 219, 222
Self phase modulation (SPM) 150, 174 future research 233
Self phase modulation tolerances records 7
direct detection 168, 175 Star 16QAM 215, 218
homodyne detection summary 232
differential detection 182, 183
Transmitter configuration, impact 230
synchronous detection 198, 202
Truth table 30
summary 229
Twisted pair 1
Semi-analytical BER estimation 133, 140,
159, 236
U
Serial DPSK transmitter 26, 33, 120, 176
Serial Square QAM transmitter 38, 121, 204
Serial Star QAM transmitter 35, 120 Uniformity 63, 65
Shot-noise 69, 80, 94, 135
Shot-noise limited detection 178, 186 V
Signal to noise ratio (SNR) 152
Signal-ASE noise 24, 69, 82 Voltage controlled oscillator (VCO) 98
Signal-LO beating 80, 83
Single mode fiber (SMF) 147, 151 W
Single-ended detection 62
Single-span transmission 155 Waveguide dispersion 145, 147
Sliding window 102 Wavelength division multiplexing (WDM)
Small-amplitude QPSK 52 2, 217, 233
SNR requirements, matched filtering 243 Wiener filter 103, 110

Seimetz M. High-Order Modulation For Optical Fiber Transmission (Springer, 2009) (3540937706) PDF

Uploaded by

Copyright:

Available Formats

Seimetz M. High-Order Modulation For Optical Fiber Transmission (Springer, 2009) (3540937706) PDF

Uploaded by

Document Information

Original Title

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

Seimetz M. High-Order Modulation For Optical Fiber Transmission (Springer, 2009) (3540937706) PDF

Uploaded by

Copyright:

Available Formats

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

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

What are the main roles and responsibilities of the Editorial Board?

What are the main roles and responsibilities of the Editorial Board?

Springer Series in

optical sciences 143

Editor-in-Chief: W. T. Rhodes, Atlanta

Ali Adibi Bo Monemar

With 132 Figures

ISSN 0342-4111 e-ISSN 1556-1534

Library of Congress Control Number: 2009926839

c Springer-Verlag Berlin Heidelberg 2009

Cover design: eStudio Calamar S.L.

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

The deployment of high-order modulation formats in optical fiber transmission sys-

Berlin, January 2009 Matthias Seimetz

Part I Transmitters and Receivers

3.1.2 Delay Line Interferometer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

4 Effort Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119

Part II System Performance

5 System Simulation Aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

6 Fiber Propagation Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

7 Back-to-Back and Single-Span Transmission . . . . . . . . . . . . . . . . . . . . . . 155

8 Multi-Span Long-Haul Transmission . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207

9 Performance Trends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227

A Differential Encoders for 8DPSK / 16DPSK . . . . . . . . . . . . . . . . . . . . . . . 239

B Theoretical BER Formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243

ϕn (t) Overall laser phase noise after coherent detection

BW D M Electrical bandwidth for WDM channel separation

I (t) Photocurrent in the I -branch after phase noise cancelation

N ph Number of phase states

t′ Retarded time scale t − τgr · z

SBS Stimulated Brillouin Scattering

M. Seimetz, High-Order Modulation for Optical Fiber Transmission, 1

Reduction of Costs by Innovative Technologies

More complex Advanced Modulation Higher spectral

Advanced Optical Modulation

In many wireline and wireless telecommunication systems, receivers are based on

Exploding Data Traffic and Bandwidth Demands

50 Tbit/s long distance traffic

Core Network WDM, IP

Access Network Business

2.56 160 0.41 - 1 2560 2006 [50]

Intention, Scope and Outline of the Book

In spite of the brilliant experimental results and technological progress in optical

Finally, Chapter 9 summarizes the major trends in system performance resulting

2.1 Transmitter Components

The following subsections briefly describe some fundamental key components of

M. Seimetz, High-Order Modulation for Optical Fiber Transmission, 15

The normalized electrical field of an ideal optical carrier as emitted from a CW

1ϕn s (t) = ϕn s (t) − ϕn s (t − τ ). (2.3)

h1ϕn2s (τ )i = 2π 1νs |τ |, (2.7)

Intensity noise can also lead to significant degradation in system performance,

2.1.2 External Optical Modulators

The optical part of high-order modulation transmitters is composed of one or more

An optical phase modulator can be fabricated as an integrated optical device by

Phase modulation ϕ P M (t) is a function of the wavelength λ, the length of the

By utilizing the principle of interference, the process of phase modulation can

In digital optical transmission