Chaotic Communication

Kehui Sun
Chaotic Secure Communication

Also of interest
Coding and Cryptography

Natas̆a Z̆ivić, 2013
ISBN 978-3-486-75212-0, e-ISBN 978-3-486-78126-7,
Set-ISBN 978-3-486-80581-9
Modern Communications Technology

Natas̆a Z̆ivić, 2016
ISBN 978-3-11-041337-3, e-ISBN 978-3-11-041338-0,
e-ISBN (EPUB) 978-3-11-042390-7
A Course in Mathematical Cryptography

G. Baumslag, B.Fine, M. Kreuzer, G. Rosenberger
ISBN 978-3-11-037276-2, e-ISBN 978-3-11-037277-9,
e-ISBN (EPUB) 978-3-11-038616-5
Computational Methods in Applied Mathematics

Carsten Carstensen (Editor-in-Chief), 4 issues per year
ISSN 1609-4840, e-ISSN 1609-9389
Kehui Sun
Chaotic Secure
Communication
Principles and Technologies

This work is co-published by Tsinghua University Press and Walter de Gruyter GmbH.
Author
Prof. Kehui Sun
School of Physics and Electronics
Central South University
Hunan, China
ISBN 978-3-11-042688-5
e-ISBN (PDF) 978-3-11-043406-4
e-ISBN (EPUB) 978-3-11-043326-5
Set-ISBN 978-3-11-043407-1
Library of Congress Cataloging-in-Publication Data

A CIP catalog record for this book has been applied for at the Library of Congress.
Bibliographic information published by the Deutsche Nationalbibliothek

The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed
bibliographic data are available on the Internet at http://dnb.dnb.de.
© 2016 Walter de Gruyter GmbH, Berlin/Boston

Cover image: scanrail/iStock/thinkstock
Typesetting: Integra Software Services Pvt. Ltd.
Printing and binding: CPI books GmbH, Leck
Printed on acid-free paper
8
Printed in Germany
www.degruyter.com
Preface
Since meteorologist Lorenz discovered chaotic phenomenon in 1963, it has been
widely and deeply studied. Chaos theory is known as the third scientific theoretical
revolution following the theory of relativity and quantum theory in the 20th century.
Especially, Pecora and Carroll reported chaos synchronization in 1990, which aroused
an upsurge of study on chaotic secure communications. After more than 20 years of
research and development, chaotic secure communication technology is becoming a
new growth point in the field of information technology. On one hand, the application
of chaos will continue to promote people’s profound understanding of the nature of
chaos. On the other hand, the new problems in the application of chaos will further
lead to the further research of chaos. Although chaotic secure communication techno-
logy still has a long way to go, the application of chaos in secure communication field
shows us a promising prospect.
This book includes nine chapters. In the first chapter, the history of chaos research
is reviewed, and the important characters and stories in the development of chaos are
introduced, and the advances in research on chaotic synchronization control techno-
logy and chaotic secure communication method are reviewed. In the second chapter,
the main analysis methods of the chaotic characteristic of the nonlinear system are
presented. The typical chaotic dynamic system models are introduced briefly in the
third chapter. In the fourth chapter, the principle and technology of the synchron-
ization control between chaotic systems are investigated. In the fifth chapter, the
technology and scheme of secure communication based on chaotic synchronization
are discussed. In the sixth chapter, the data encryption technology based on chaotic
sequences is discussed. In the seventh chapter, the audio and video real-time en-
cryption communication technology based on chaos is studied. In the eighth chapter,
the latest research progress, dynamic characteristic analysis, and simulation methods
of fractional-order chaotic system are studied. In the ninth chapter, the design and
implementation technology of chaotic circuit is studied.
In order to facilitate readers to grasp the important knowledge points in the book,
the author lists a few questions at the end of each chapter. In order to facilitate the
search of relevant literature and to publish the research results on time, journals about
chaos research are listed as Appendix A of this book. In addition, in order to facilit-
ate readers to quickly grasp the chaos of research methods, the author collected the
relevant analysis programs as shown in Appendix B.
Based on the research results in the field and a large number of related literatures,
this book is written through repeated modification.
In the field of research of chaos, the author has got the guidance and inspiration
of many professors including Professor Sprott J C from the University of Wisconsin
Madison in the United States, Professor Qiu S S from South China University of Tech-
nology, Professor Yu S M from Guangdong University of Technology, Professor Wang
VI Preface
G Y from Hangzhou University of Electronic Science and Technology, Professor Zhang

T S, Professor Sheng L Y, and Professor Zhu C X from Central South University. The au-
thor expresses sincere gratitude to all the professors. In addition, in the preparation
of this book, previous graduate students gave a lot of support. They are Mou J, Zhou J
L, Tan G Q, Ren J, Shang F, Wang X, Bao S Q, Yang J L, Cheng W, Liu X, He S B, Wang
H H, Wang Y, Zuo T, Ai X X, Liu W H, Ruan J Y, Zhang L M, Xu Y X, Yu M Y, Yin Y H.
Thanks for their contributions!
I would sincerely like to appreciate and acknowledge the comments and sugges-
tions from the readers for this book.
Kehui Sun
January 2015, Changsha
Contents
1 Introduction 1
1.1 Linear, Nonlinear, and Chaos 1
1.2 Development of Chaos 2
1.3 Famous Scientists and Important Events 4
1.3.1 Lorenz and Butterfly Effect 4
1.3.2 Li Tien-Yien and the Concept of Chaos 5
1.3.3 Feigenbaum and Feigenbaum Constant 6
1.3.4 Leon Ong Chua and Chua’s Circuit 7
1.3.5 Guanrong Chen and Chen’s Attractor 8
1.4 Definition and Characteristics of Chaos 9
1.5 Overview of Chaos Synchronization Method 10
1.5.1 Chaos Synchronization Methods and Characteristics 11
1.5.2 Other Synchronization Methods and Problems 14
1.6 Summary of Chaos Secure Communication 15
1.6.1 Chaotic Analog Communication 16
1.6.2 Chaotic Digital Communication 16
1.6.3 Encryption Communication based on Chaotic
Sequence 17
1.7 Chaos Research Methods and Main Research Contents 18
Questions 20
2 Characteristic Analysis Methods for Nonlinear System 21

2.1 Phase Diagram Analysis Method 21
2.2 Power Spectral Analysis Method 22
2.3 Poincaré Section Method 24
2.4 Lyapunov Characteristic Exponent Method 25
2.4.1 Definition of Lyapunov Characteristic Exponent 25
2.4.2 Lyapunov Exponent Spectrum 27
2.4.3 Physical Meaning of Lyapunov Exponent 27
2.4.4 Condition Lyapunov Exponent 29
2.5 Fractal Dimension Analysis Method 30
2.6 0–1 Test Method 30
2.6.1 0–1 Test Algorithm 30
2.6.2 Improved 0–1 Test Algorithm 32
2.6.3 Application of 0–1 Test Algorithm 33
2.7 Dividing Frequency Sampling Method 33
2.8 Pseudo-Phase Space Method 34
2.9 Complexity Measure Algorithm 35
2.9.1 Spectral Entropy Complexity Algorithm 36
2.9.2 C0 Complexity Algorithm 38
Questions 40
VIII Contents
3 Typical Chaotic Systems 41

3.1 Discrete-Time Chaotic Map 41
3.1.1 Logistic Map 41
3.1.2 Tent Map 44
3.1.3 Hénon Map 44
3.1.4 Tangent Delay-Ellipse Reflecting Cavity System Map 46
3.2 Continuous-Time Chaotic System 47
3.2.1 Duffing Oscillator 48
3.2.2 van der Pol Oscillator 49
3.2.3 Lorenz System 49
3.2.4 Rössler System 50
3.2.5 Chua’s Circuit 51
3.2.6 Chen System 55
3.2.7 Lü System 55
3.2.8 Unified Chaotic System 57
3.2.9 Simplified Lorenz System 59
3.2.10 Standard for a New Chaotic System 62
3.3 Hyperchaotic System 62
3.3.1 Rössler Hyperchaotic System 62
3.3.2 Chen Hyperchaotic System 63
3.3.3 Folded Towel Hyperchaotic Map 64
Questions 65
4 Chaotic Synchronization Principle and Method 66

4.1 Definition of Chaos Synchronization 66
4.2 Performance Index of Chaotic Synchronization System 67
4.3 Principle and Performance of Feedback Control Synchronization 69
4.3.1 Multivariable drive feedback synchronization for chaotic
system 69
4.3.2 Linear and Nonlinear Feedback Synchronization for
Discrete Chaotic System 73
4.4 Parameter Adaptive Synchronization Based on Pecora–Carroll
Synchronization Criterion 84
4.4.1 Principle for Adaptive Synchronization 84
4.4.2 Choose Control Law and Control Parameters 85
4.4.3 Determine the Boundary and Range of the Control
Constant 86
4.4.4 Simulation Results and Discussion 87
4.5 Adaptive Synchronization Control Based on Lyapunov Stable Theory 88
4.5.1 Adaptive Synchronization Control with Certain
Parameter 88
4.5.2 Adaptive Synchronization Control with Uncertain
Parameters 91
Contents IX
4.6 Intermittent Feedback Synchronization Control 94

4.6.1 Intermittent Synchronization Control between Different
Systems 94
4.6.2 Synchronization Simulation and Performance
Analysis 97
4.7 Synchronization Control Based on State Observer Method 99
4.7.1 Design Principle of State Observer 99
4.7.2 Design of State Observer for Unified Chaotic
System 101
4.7.3 Simulation and Discussion 101
4.8 Chaos Synchronization Based on Chaos Observer 104
4.8.1 Design Principle of Chaos Observer 104
4.8.2 Chaos Observer of the Unified Chaotic System 105
4.8.3 Synchronization Simulation and Performance
Analysis 106
4.9 Projective Synchronization 108
4.9.1 Principle and Simulation of Proportional Projection
Synchronization 109
4.9.2 Principle and Simulation of Function Projective
Synchronization 113
4.9.3 Principle and Simulation of the Adaptive Function
Projective Synchronization 116
4.10 Problems and Development of Chaos Synchronization 123
Questions 124
5 Secure Communication Technology

Based on Chaos Synchronization 125
5.1 Chaotic Masking Communication 125
5.2 Chaos Shift Keying Communication 130
5.2.1 COOK Communication 131
5.2.2 CSK Communication 132
5.2.3 DCSK Communication 137
5.2.4 FM-DCSK Communication 139
5.2.5 QCSK Communication 140
5.3 Chaos Parameter Modulation 145
5.4 Chaotic Spread Spectrum Communication 148
5.4.1 Principle of Chaotic DS/SS Communication 148
5.4.2 Generation of Chaotic Spread Spectrum Code 150
5.4.3 Simulation Module Design of Multiuser Chaotic Spread
Spectrum System 152
5.4.4 Design and Simulation of Multiuser Chaotic Spread
Spectrum System Based on Rake Receiver 158
X Contents
5.4.5 Performance Analysis of Multiuser Chaotic Spread

Spectrum System Based on Rake Receiver 159
Questions 163
6 Data Encryption Based on Chaotic Sequence 164
6.1 Chaos Cryptography 164
6.1.1 Traditional Cryptography 164
6.1.2 Relationship between Chaos Cryptography and Traditional
Cryptography 166
6.1.3 Principle of Chaos Sequence Encryption 167
6.2 Data Encryption Algorithm Based on Chaotic Sequence 168
6.2.1 Basic Requirements of Chaotic Encryption
Algorithm 168
6.2.2 Logistic Map and Its Statistics 170
6.2.3 Chaotic Encryption Model and Algorithm 170
6.2.4 Implement of Encryption Algorithm Based on Chaotic
Sequence 172
6.2.5 System Performance Analysis 175
6.3 Binary Watermark Image Encryption Algorithm Based on Chaos 176
6.3.1 Improvement of Chaotic Pseudo-Random Sequence
Generator 178
6.3.2 Binary Watermark Image Encryption Algorithm Based on
TD-ERCS Map 178
6.3.3 Performance Analysis of Encryption Algorithm 179
6.4 Color Image Watermarking Algorithm Based on Chaotic Map and Wavelet
Transform 182
6.4.1 Wavelet Transformation Algorithm and Its
Application 183
6.4.2 Design and Implementation of Watermark Embedding and
Extraction Algorithm 186
6.4.3 Simulation and Performance Analysis 190
6.5 Chaotic Image Encryption Algorithm 196
6.5.1 Chaos-Based Image Encryption Technology 196
6.5.2 A New Chaotic Image Encryption Scheme 199
Questions 205
7 Audio and Video Chaotic Encryption

and Communication Technology 206
7.1 Principle of Real-Time Voice Encryption Communication 206
7.1.1 Principle of Digital Encryption Communication
System 206
7.1.2 Real-Time Voice Encryption and Communication
System 208
Contents XI
7.1.3 Cipher Synchronization Technology 209

7.2 Real-Time Data Stream Encryption Algorithm 210
7.2.1 3DES Encryption Algorithm 210
7.2.2 Encryption Algorithm Based on 3DES and Chaotic
Map 211
7.2.3 Network Synchronization Method 213
7.2.4 Security Analysis of Real-Time Voice Encryption
System 215
7.3 Realization of Real-Time Voice Encryption System 216
7.3.1 Structure of Real-Time Voice Encryption System 216
7.3.2 Real-Time Transmission Protocol and Its Data
Structure 218
7.3.3 Implementation of RTP Based on Adaptive Communication
Environment 220
7.3.4 System Debugging and Performance Testing 222
7.4 MPEG Video Encryption Algorithm Based on Chaotic Sequence 224
7.4.1 Encryption Principle and Procedure of MPEG Video
Data 225
7.4.2 Design of Stream Cipher Based on Chaos 226
7.4.3 Design of Scrambling Algorithm Based on Chaos 229
7.4.4 Encryption Experiment and Encryption Effect
Analysis 229
7.4.5 Performance Analysis of MPEG Video Encryption
System 230
7.5 H.264 Video Encryption Algorithm Based on Chaos 232
7.5.1 Chaotic Stream Cipher Based on Tangent Delay-Ellipse
Reflecting Cavity System 233
7.5.2 H.264 Video Encryption Scheme 233
7.5.3 Performance Analysis of Video Encryption 236
Questions 237
8 Analysis and Simulation of Fractional-Order Chaotic System 238

8.1 Development of Fractional-Order Chaotic System 238
8.2 Definition and Physical Meaning of Fractional Calculus 241
8.3 Solution Methods of Fractional-Order Chaotic System 243
8.3.1 Frequency Domain Method for Solving Fractional-Order
Chaotic System 243
8.3.2 Time Domain Method for Solving Fractional-Order Chaotic
System 247
8.4 Simulation Method for Fractional-Order Chaotic System 251
8.4.1 Dynamic Simulation Method for Fractional-Order Chaotic
System 251
XII Contents
8.4.2 Circuit Simulation Method of Fractional-Order Chaotic

System 253
8.4.3 Numerical Simulation Method of Fractional-Order Chaotic
System 256
8.4.4 Comparison of Simulation Methods 258
8.5 Dynamics of Fractional-Order Unified System Based on Frequency
Domain Method 259
8.5.1 Fractional-Order Unified Chaotic System Mode 259
8.5.2 Dynamic Simulation of Fractional-Order Unified Chaotic
System 260
8.5.3 Dynamics of Fractional-Order Unified System with Fixed
Parameter 261
8.5.4 Dynamics of Fractional-Order Unified System with Different
Parameter 264
8.6 Dynamics of Fractional-Order Diffusionless Lorenz System Based on
Time Domain Method 266
8.6.1 Fractional-Order Diffusionless Lorenz System
Model 266
8.6.2 Chaotic Characteristic of Fractional-Order Diffusionless
Lorenz System 267
8.6.3 Bifurcation Analysis with Varying System
Parameter R 268
8.6.4 Bifurcation Analysis with Varying Fractional
Order q 269
8.6.5 Bifurcations with Different Fractional Order 270
Questions 272
9 Simulation and Hardware Implementation of Chaotic System 274

9.1 Dynamic Simulation of Chaotic System 274
9.1.1 Dynamic Simulation Steps on Simulink 274
9.1.2 Simulation Result Analysis and Performance
Improvement 276
9.2 Circuit Simulation of Chaotic System 276
9.3 Design of Chaotic Analog Circuit 277
9.3.1 Procedure of Chaotic Circuit Design 277
9.3.2 Modular Design of Chaotic Circuit 279
9.3.3 Basic Components of Chaotic Circuit 279
9.4 Improved Modular Design of Chaotic Circuit 282
9.4.1 Design of Multiscroll Jerk Circuit Based on OA 283
9.4.2 Design of Multiscroll Jerk Circuit Based on CC 283
9.4.3 Multiscroll Jerk Circuit Implementation 286
9.5 Digital Signal Processor Design and Implementation of Chaotic
Systems 287
Contents XIII
9.5.1 DSP Experimental Platform of Chaotic System 287

9.5.2 DSP Implementation of Integer-Order Chaotic
System 290
9.5.3 DSP Implementation of Fractional-Order Chaotic
System 292
Questions 294
Appendix A: Important Academic Journals
in the Field of Chaos 295
Appendix B: Matlab Source Programs
of Chaos Characteristic Analysis 296
B.1 Mode of the Simplified Lorenz System 296
B.2 Programs about Characteristic Analysis of Integer-Order Simplified
Lorenz System 297
B.2.1 Simulation Codes for Matlab Toolbox with Runge–Kutta
Algorithm 297
B.2.2 Solution Based on Euler and Runge–Kutta
Algorithm 297
B.2.3 Program of Calculating LE 300
B.2.4 Program of Calculating the Maximum LE 303
B.2.5 Program for Calculating Bifurcation Diagram 304
B.2.6 Program for Plotting Poincaré Section 305
B.2.7 Program for 0–1 Test 307
B.3 Solution Program for the Fraction-Order Simplified Lorenz System 308
B.3.1 Solution Program Based on Predictor–Corrector
Approach 308
B.3.2 Solution Program Based on Adomian Decomposition
Approach 309
B.3.3 Analysis Program of Complexity and Chaos
Diagram 312
B.4 Simulation of Chaotic Synchronization 316
B.4.1 Drive–Response Synchronization 316
B.4.2 Coupling Synchronization 318
B.4.3 Tracking Synchronization 320
References 323
Index 331
1 Introduction
After decades of development, people are no longer strange to chaos, but how to
correctly understand and study chaos and its application is still the main task of aca-
demic circles. As we all know, chaos is nonlinear, while what we know well is linear.
What is the relationship between them? The discussion begins here.
1.1 Linear, Nonlinear, and Chaos

Linear and nonlinear are used to explain the relationship between function y = f (x)
and an independent variable x. To be specific, linear refers to the proportional re-
lationship between two independent variables. It represents regular and smooth
movement in space and time. However, nonlinear refers to the disproportionate re-
lationship between two dependent variables. It represents irregular and mutation
movement in space and time. For example, how many times the visual acuity of two
eyes is that of an eye? It is very easy to think that it is two times, but actually, it is 6–10
times. This is the nonlinear relationship. Calculation result of one plus one is not two.
Although nonlinear relationship is ever changing and complicated, there is some-
thing in common that is different from linear relationship. Linear relationship is un-
related independence, while nonlinear relationship is interaction. It is the interaction
that makes the entirety, which is not simply equal to the sum of the parts. On the con-
trary, unlike linear superposition, gain or loss may occur. For example, the generation
of laser is nonlinear. Laser scatters in all directions like lamp when the voltage is small.
When the applied voltage reaches a certain value, a new phenomenon applied will
appear suddenly. The excited atoms scatter a kind of homogeneous light whose emit
phase and direction are consistent as if they received an order to look right. This kind
of homogeneous light is laser. For another example, the frequency components of lin-
ear signal remain the same, but nonlinear relationship leads to changes of frequency
configuration. As long as there is nonlinearity, even if it is a small nonlinearity, it will
produce sum frequency, difference frequency, frequency doubling, and other compon-
ents. Behavioral mutations are caused by nonlinear relationships, but small deviation
for the linear relationship doesn’t lead to behavioral mutations in general. And accord-
ing to the original linear conditions, the small deviation can be described and com-
prehended by modified linear theory. System behavior may change suddenly when
nonlinear relationship reaches a certain degree. Mutations often occur at a series of
parameter thresholds of the nonlinear system, and each mutation is accompanied by
a new frequency component. The system will eventually enter the chaotic state.
According to the characteristics of the nonlinear relationship, it can be concluded
that if there exists chaos, the system must be a nonlinear system. There is a muta-
tion behavior when a nonlinear system turns into the state of chaos. How can we
2 1 Introduction
know whether a system has turned into the state of chaos? How to judge a system
that presents phenomenon for a long period of time? How to distinguish whether a
system is subject to external random disturbances? All those questions are important
issues in chaos research.
Chaos theory belongs to nonlinear science. Nonlinear scientific research al-
ways seems to make people’s understanding of “normal” things and “normal”
phenomenon to explore the “abnormal” things and “abnormal” phenomenon. For
example, solitary wave is not a regular propagation with periodic oscillation. Un-
conventional edit methods will be taken when a large number of unconventional
phenomena occur when using multimedia technology for information storage, com-
pression, transmission, conversion, and control process encountered. Chaos broke the
convention that the future movement of a deterministic equation is determined strictly
by the initial conditions and led to a so-called strange attractor phenomenon.
In nonlinear science, chaos is not exactly the same as what it means. Chaos
phenomenon refers to a determined but unpredictable motion in nonlinear science.
The external manifestations of chaos are very similar to purely random motion.
That is, they are all unpredictable. Comparing with random motion, chaos motion is
determinate. And its unpredictability is derived from the instability of the system of
internal movement. Chaos system is sensitive to initial conditions and little change,
no matter how small they are. Therefore, after a period of time evolution, the system
will deviate from its original direction of evolution. Chaos is a common phenomenon
in nature, such as the change of weather. An important feature of chaos is butterfly
effect. So chaos means a nonperiodic motion generated by a deterministic system.
Chaos, fractals, and soliton are the three most important nonlinear science concepts.
Chaos theory is a part of nonlinear science, and only nonlinear system can produce
chaotic motion.
In 1977, the first international chaos conference was held in Italy. It marks the of-
ficial birth of the science of chaos. The physicist Ford who was one of the assembly
moderators considered that chaos theory was the third revolution of physics following
relativity theory and quantum mechanics in the twentieth century. He said, relativity
theory eliminated illusions about absolute space and time, and quantum mechanics
broke Newton’s dream about controllable measurement process; and chaos burst the
bubble of determinism of predictability by Laplace [1].
1.2 Development of Chaos

In the latter half of the twentieth century, nonlinear science has been developing
rapidly, which accounted for a great proportion of the study on chaos. Chaos, as a
remarkable frontier project and academic hot topic, reveals the unity of simplicity and
complexity, order and disorder, and certainty and randomness in nature and human
society, and broadens people’s view greatly, and deepens people’s understanding of
1.2 Development of Chaos 3
the objective world. In the field of natural science and social science, chaos is striking
and changing the field of science and technology, and it presents a huge challenge to
people.
In the early 1960s, scientists began to explore some elusive phenomenon in
nature. In 1963, Lorenz proposed deterministic nonperiodic flow model [2], and later
he proposed the butterfly effect. In 1975, Li and Yorke published “Period three implies
chaos” [3] in American Mathematical Monthly, which revealed the evolution process
from order to chaos. This famous article also defined “chaos” as a new scientific term
that officially appeared in the literature. In 1976, American biologist May published
“Simple mathematical models with very complicated dynamics” [4] in Nature. This
article pointed out that a very simple one-dimensional iterative map can produce
complex period-doubling and chaotic motion. He revealed to the people that some
simple deterministic mathematical model in ecology can generate chaotic behavior.
In 1978 and 1979, Feigenbaum independently discovered scaling property and univer-
sal constant [5, 6] in the phenomenon of period-doubling bifurcation, which lays a
solid theoretical foundation for chaos in modern science. Lorenz defined that chaos
is a science that uses fractal geometry to analyze and study the nonlinear dynamics
problem from butterfly effect, aperiodicity, and so on.
In the 1980s, chaos science has been further developed. Many researchers have fo-
cused on how an order system turns into the chaotic state and on the characteristics of
chaos. In 1981, Takens proposed experimental method of determining strange attract-
ors [7]. In 1983, Glass published an article to set off a wave of computing time series
dimension [8]. In 1986, Grassber proposed dynamical systems theory and methods
of reconstruction [9]. By calculating chaotic characteristics, such as fractal dimen-
sion, Lyapunov exponent from the time series, chaos theory entered into the stage of
practical application.
In the 1990s, chaos science got mutual infiltration, promotion, and wide applic-
ation with other subjects, such as synchronization, secure communication, chaos
cryptography, chaotic neural networks, and chaotic economics research areas. Ex-
ploring the unique nature, role, and function of chaotic systems to benefit humanity
becomes a very great task.
Research on deterministic chaotic system has undergone three stages. The first
stage is to study how an order system turns into chaos, which focuses on the condi-
tions, mechanisms, and generated pathways. The second stage is to study what is the
order of chaotic system, which focuses on universality, statistical characteristics, and
the fractal structure in chaos. The third stage is to study how a chaotic system turns
into an order system, which focuses on controlling chaos to achieve order. The current
study of chaos focuses on how to control and utilize chaos.
Due to being sensitive on initial conditions, even two identical chaotic systems
evolve from almost the same initial conditions, and their orbits also quickly be-
come uncorrelated over time, which makes the chaotic signal become unpredictable
and with long-term anti-interception capability. At the same time, chaotic system is
4 1 Introduction
deterministic, which is completely determined by the equations, nonlinear system

parameters, and initial conditions, and therefore chaotic signal is easy to be gener-
ated and replicated. Due to these features, chaos has excellent prospects in secure
communications and becomes a hot research topic in chaotic application.
Internationally, the study of secure communications originated in the early 1990s,
and it becomes one of the central issues in the field of information science. In 1990,
Pecora and Carroll proposed a chaos synchronization method and first observed the
phenomenon of chaotic synchronization [10, 11] on electronic circuits. In the same
year, Ott, Grebogi, and Yorke realized the control to the chaotic attractor unstable
periodic orbits [12]. These works have greatly aroused people’s interest in chaos mech-
anism and applications, and quickly pushed the theoretical and experimental study
on chaos synchronization and control of chaos, and opened the prelude to apply chaos
in engineering field. In fact, chaotic encrypted communication will play an important
role not only in the field of military, national security, and communications counter-
measure, but also in the business world and people’s lives, which is the new trend of
future information security technology.
1.3 Famous Scientists and Important Events
1.3.1 Lorenz and Butterfly Effect
Edward Norton Lorenz (1917–2008) was born in Hartford, Connecticut, USA. He loved
science from a young age and graduated from Harvard University in 1940. During
World War II, Lorenz served as a forecaster in the United States Army Air Corps. After
the war, he received his master’s and doctorate in meteorology. Then he taught at the
Massachusetts Institute of Technology. In 1972, he put forward the famous “butterfly
effect” theory. In 1975, he became an academician in the US National Academy of Sci-
ences. He won the “Crafoord Prize,” which is called Ecology “Nobel Prize” in 1983 and
won the “Kyoto Prize” in 1991. The judges thought his chaos theory “brought the most
dramatic change to the human nature view after Newton’s theory.”
The butterfly effect is one of his important scientific contributions, and its dis-
covery is interesting and enlightening. In 1961 winter, Lorenz used computer for
numerical prediction calculation. In order to save time, he selected a row data (equi-
valent to one day’s weather conditions) from the original calculation results as an
initial value to calculate again, and then left his office to drink a cup of coffee. After
an hour, when he returned to the laboratory, the computer had already calculated the
predicted results of the next two months. He was shocked by the new results, which
were quite different from the original results. Minor deviation doubled every four days
until the similarity between the old and the new data was completely lost. The differ-
ence of initial value was less than 1/1,000, but it causes the second calculation result
to be completely different from that of the first one. This is the “butterfly effect,” which
shows the initial value sensitivity.
In 1979, Lorenz delivered a speech entitled “Butterfly Effect” on the 139th meet-
ing of the American Academy of Science Development and proposed a seemingly
preposterous assertion: a butterfly flapping its wings in Brazil, there may be a tornado
in Texas, United States, a month later. Thus, he pointed out that it is very difficult to
forecast weather accurately. Today, people still talk about this assertion. More import-
antly, it stimulated people’s interest in chaos. With the rapid development of computer
technology, chaos has become a frontier field with far-reaching influence and rapid
development. “Butterfly Effect” is fascinating, exciting, and thought provoking, so
it not only is a bold imagination and charming color aesthetics, but also lies in the
profound scientific connotation and inherent philosophical charm. Lorenz’s discovery
laid the foundation for chaos theory. The chaos theory not only affects meteorology,
but also profoundly affects every branch of science.
1.3.2 Li Tien-Yien and the Concept of Chaos
Li Tien-Yien (1945–) was born in Shaxian, Fujian Province, China. He and his fam-
ily moved to Taiwan when he was three years old. He received education there until
he graduated from university. He is the 68th graduate of mathematics from Tsinghua
University, Hsinchu, Taiwan province, China. In 1969, he went to the Department of
Mathematics, University of Maryland, United States, for doctoral degree, under Pro-
fessor York. He was a lecturer in the University of Utah in the United States from 1974
to 1976. Since 1976, he taught at Michigan State University in the United States. Li Tien-
Yien has made several important pioneering works and extraordinary achievements in
the field of applied mathematics and computational mathematics. The paper “Period
Three Implies Chaos” was written by him and York, which first proposed the concept
of chaos in science and opened a new era in the scientific community for research in
chaotic dynamical systems.
In the past few decades, Professor Li Tien-Yien is suffering from a long-term ill-
ness. So far, he has undergone major surgeries under general anesthesia more than
ten times and countless surgeries under local anesthesia. Innumerable wounds are
on his body. However, he struggled in the face of adversity and overcame the disease
with an indomitable spirit of optimism again and again. He sought breakthrough in
adversity and fought the disease optimistically. He overcame all the difficulties with a
surge of strong determination and ultimately made a first-class research work under
difficult environments.
Li Tien-Yien has had rigorous scholarship for decades [13]. He believes that the
only way to success is insistence. He said to his students that he was not smart and
that it is not really important to be smart. The most important issue is to get to the
6 1 Introduction
bottom. He often emphasized to think on the problem just a minute more than others.
He believed the precious one minute paves the road for success. He often exhorts his
students that with tremendous effort, one must persist till the end of everything and
should never give up. He also said that being engaged in researches they must have a
thorough understanding, especially on mathematical logic.
Li Tien-Yien witnessed interesting research experience in the birth of chaos.
The famous paper, “Period Three Implies Chaos,” also underwent topic selection,
research, submission, rejection, and modification, and finally was published in Amer-
ican Mathematical Monthly. It focused on the meaning of section iterative, definitions,
and main results of some basic concepts. And some courses are provided as an ap-
pendix. A mathematician, who most values clarity and precision, had even talked
about mess, which has attracted many readers. Physicists and biologists have also
quickly accepted the mathematical articles. Since then, “chaos” which was put for-
ward first by Li and York, has become a technical term in chaos theory and has even
become synonymous with chaos theory [14].
1.3.3 Feigenbaum and Feigenbaum Constant
Feigenbaum M. J. (1944–), a professor of physics at Cornell University, entered the

Graduate School of MIT after graduating from New York College in 1964. He received
his PhD in basic particle physics in 1970 and went to Cornell University to engage in
postdoctoral research in the same year, but there was no progress for two years. Then
he went to Virginia Polytechnic Institute for two years without success. Carruthers P.,
a former professor in Cornell University, recognized his excellence and selected him as
his assistant in Los Alamos (affiliated to California campuses) for free research. Then,
after two years, Feigenbaum found a wonderful constant 4.6692. . . [5]. He published
the important result in the Journal of Statistical Physics in 1979 [6] and became famous
overnight.
Feigenbaum studied about spacing ratio changes in period-doubling bifurca-
tions, which are shown in Table 1.1. He found that bifurcation point of two adja-
cent spacing Bk forms a geometric sequence with the increase in the number k of
bifurcation
kn – kn–1
$ = lim = 4.6692016091 . . . . (1.1)
n→∞ kn+1 – kn
Bifurcation width wi also forms a geometric sequence

wn
! = lim = 2.5029078750 . . . . (1.2)
n→∞ wn+1
Calculation of Feigenbaum constant is shown in Figure 1.1.

The discovery of Feigenbaum constant was one of the twentieth century’s greatest
discoveries. It found a deeper regularity in the periodic-doubling bifurcations and
Table 1.1: Spacing ratio changes of period-doubling bifurcation ( f (x) = +x(1 – x)).
k Bifurcation Bifurcation value +k Spacing ratio (+k – +k–1 )/(+k+1 – +k )
1 1 to 2 = 21 3
2 2 to 4 = 22 3.449489743 4.751466
3 4 to 8 = 23 3.544090359 4.656251
4 8 to 16 = 24 3.564407266 4.668242
5 16 to 32 = 25 3.468759420 4.66874
6 32 to 64 = 26 3.569691610 4.6691
.. .. .. ..
. . . .
∞ Periodic solution → chaos 3.569945972 4.669201609
w1
x = 0.5 w3
w2
k2 – k1 k3 – k2
k
Figure 1.1: Calculation of Feigenbaum constant.
revealed the secret of nonlinear systems from order to chaos. The discovery of Feigen-
baum constant greatly changed humans’ understanding of the universe and promoted
chaos research from qualitative analysis to quantitative calculation. It became an im-
portant milestone in the study of chaos. As we all know, the birth of a new theory
is often accompanied with the emergence of new constant in the development of
physical theory, such as gravitational constant G in Newton mechanics, Planck’s con-
stant h in quantum mechanics, and the speed of light c in relativity. So we can easily
understand the importance of Feigenbaum constant in chaos theory research and
development.
1.3.4 Leon Ong Chua and Chua’s Circuit
Leon Ong Chua (1936–) is a professor at Department of Electrical Engineering and

Computer Science, University of California, Berkeley. He received his bachelor’s de-
gree in electrical engineering in the Philippines Mapúa Institute of Technology in 1959.
8 1 Introduction
L NR
C2 C1
Figure 1.2: Chua’s circuit.
He received his master’s degree from MIT in 1961 and his doctoral degree from the Uni-
versity of Illinois, Urbana – Champaign in 1964. He taught at the Purdue University in
1964–1970 and became a professor and an IEEE fellow at the University of California,
Berkeley, in 1971. He proposed the memristor, Chua’s circuit, and cell-type neural net-
work theory. He is considered the father of the theory of nonlinear circuit analysis. He
predicted the existence of the memristor. Thirty-seven years later, the research team
of Williams R. S. in Hewlett-Packard developed a solid memristor.
Chua’s circuit provides a hardware foundation for chaos theory and applications
research. When Professor Chua was a visiting scholar at Waseda University in 1983, he
designed a simple nonlinear electronic circuit, which is shown in Figure 1.2 [15]. The
circuit consists of an energy-storage component (capacitors C1 and C2 ), an inductor L1 ,
an active resistance R, and a nonlinear resistor NR . Nonlinear resistor consists of three
linear resistors and an operational amplifier. The circuit can exhibit standard chaotic
behavior. Its simple manufacturability makes it an example of chaotic system in real
world and becomes a typical model of chaotic systems.
1.3.5 Guanrong Chen and Chen’s Attractor
Guanrong Chen (1948–) is a professor in the Department of Electronic Engineering in

the City University of Hong Kong. During the Cultural Revolution (for about 10 years),
he studied on his own and was directly admitted to the Department of Mathematics
of Sun Yat-Sen University at the end of the Cultural Revolution. He graduated in 1981
and received his master’s degree in computational mathematics. Later, he studied ap-
plied mathematics for his doctoral degree in Texas A & M University in 1987. He worked
as a visiting assistant professor at the Rice University in the United States from 1987
to 1990, and then as an assistant professor, associate professor, and professor at the
Houston University from 1990 to 2000. In 1996, Professor Chen became an IEEE Fellow
1.4 Definition and Characteristics of Chaos 9
40
z
20
0
–20 0 20
x
Figure 1.3: Chen attractor (x–z plane).
because of his contributions in chaos control and bifurcation theory and application.
Since 2000, he has been serving as a professor of electronic engineering in the City
University of Hong Kong and the director of the chaotic and complex network of aca-
demic research center. His research focused on related applications areas of control
theory of nonlinear systems and dynamics analysis and complex network. In 1999, he
proposed a new chaotic attractor [16], known as Chen’s attractor, which is shown in
Figure 1.3. It has become an important model in chaos theory and applied research.
1.4 Definition and Characteristics of Chaos

Due to the singularity and complexity of chaotic systems, there is no unified defini-
tion for chaos. Different definitions reflect the nature of chaotic motion from different
angles. The larger impact is the definition of Li–Yorke chaos [3], which is defined from
the interval mapping, and it is described as follows.
Li–Yorke theorem: Let f (x) be a continuous map on [a, b]. If there are three cycle
points, then for any positive integer n, there are n periodic points for f (x).
Li–Yorke chaos definition: For continuous map f (x) on closed interval I, if the follow-
ing conditions are met, it is confirmed that there is chaos phenomenon: (1) period of
periodic points of f (x) is unbounded and (2) the uncountable set S, on closed interval
I, meets the following conditions:

1. ∀x, y ∈ S, lim supf n (x) – f n (y) > 0 when x ≠ y,
n→∞
2. ∀x, y ∈ S, lim inf f n (x) – f n (y) = 0,
n→∞
3. ∀x ∈ S and any periodic point y of f , lim supf n (x) – f n ( y) > 0.
n→∞
10 1 Introduction
Chaos has aperiodic bounded dynamic behaviors in deterministic nonlinear dynam-

ical systems, which is sensitive to initial conditions. Chaotic motion is random-like
phenomenon that occurs in deterministic systems. Chaos has the following main
features:
1. Boundedness. Chaos is bounded, and its track of movement has always been con-
fined to a certain area that is called the chaotic attracting basin. No matter how
instable the chaotic system is, its trajectory will not be out of the chaotic attractor
basin. So chaotic systems are stable in general.
2. Ergodicity. Chaotic motion is ergodic in their attractor area, which means chaotic
orbits experience every status point within a limited time in the chaotic region.
3. Randomness. Chaos is uncertain behavior generated by a certain system, and it
has the intrinsic randomness, regardless of external factors. Although the equa-
tions of a system are deterministic, its dynamic behavior is difficult to determine.
The probability density function of any area is nonzero in its attractor. In fact,
unpredictability and sensitivity to initial conditions of chaos lead to internal
randomness, but it is also proved that chaos is locally instable.
4. Fractal dimension. Fractal dimension character refers to the behavior charac-
teristics of chaos trajectories in phase space. The dimension is a quantitative
description for the geometry complexity of an attractor. It shows chaotic motion
features with multi-leaf and a multilayer structure, and each leaf and layer is fine,
performing infinite hierarchy self-similar structure.
5. Scaling property. Scaling property refers that the chaotic motion is ordered in the
disordered state. It can be considered that as long as the precision of value or
laboratory equipment is high enough, you can always see the orderly movement
in small-scale chaotic region.
6. Universality. It refers to the different systems that show some common features in
the route to chaos, and it does not change with systems equations or parameters.
Specific performance is universal constant in chaotic systems, such as the fam-
ous Feigenbaum constant. Universality is a reflection of the inherent regularity in
chaos.
7. Positive Lyapunov exponent. Lyapunov exponent is the quantitative characteriza-
tion of approaching or separating trajectories generated by nonlinear system from
each other. Positive Lyapunov exponents indicate that tracks are unstable in each
locality and adjacent tracks separated exponentially. Meanwhile, the positive Lya-
punov exponent also indicates the loss of the adjacent point information – the
greater the value, the more serious the loss of information, the higher the degree
of chaos.
1.5 Overview of Chaos Synchronization Method

Chaos is so sensitive to initial conditions that people once thought that chaos syn-
chronization is almost impossible. Until 1990, Pecora and Carroll proposed chaos
self-synchronization method [10]. Synchronization between two chaotic systems was

achieved by using the drive-response method for the first time. This breakthrough
research progress broke the traditional idea that movement pattern of chaos is
dangerous and uncontrollable. The research results show that chaos not only can
achieve control and synchronization, but also can serve as the dynamic basis of
information transmission and processing. It is possible to apply chaos to secure
communication.
Generally speaking, synchronization belongs to the category of chaos control.
Synchronization and control have been combined into one issue by Chua et al. [17].
That is to say, the problem of chaotic synchronization can be considered as a kind
of control problem of the chaotic orbit of the controlled system moving according to
the orbit of the target system. Traditional chaos control is generally to control a sys-
tem on unstable periodic orbit, while chaos synchronization is to realize complete
reconstruction of two chaotic systems.
1.5.1 Chaos Synchronization Methods and Characteristics
Based on chaos synchronization theory and chaos control methods, researchers have
put forward many chaos synchronization schemes [18] as shown in Table 1.2.
Drive-Response Synchronization: In 1990, Pacora-Carroll first realized synchroniz-
ation between two chaotic systems by using drive-response synchronization method.
Later, Cuomo and Oppenheim also successfully simulated the synchronization of
Lorenz system by using electronic circuit [19]. Soon, Carroll and Pecora promoted
drive-response chaos synchronization method to higher-order cascaded chaotic sys-
tem [20]. The biggest feature of drive-response synchronization is that there is a
relationship of drive and response between two nonlinear dynamic systems. Beha-
vior of the response system depends on the drive system, while the behavior of the
drive systems has nothing to do with the response system. However, for some actual
Table 1.2: Chaos synchronization methods.
Time chaotic synchronization Spatiotemporal chaos Hyperchaos synchronization

synchronization
Drive-response synchronization Feedback technology Drive-response synchronization

Active-passive synchronization Driven variable feedback Coupled synchronization
Mutual coupling synchronization Other Variable feedback synchronization
Continuous variable feedback Hybrid synchronization
Adaptive synchronization Drive and variable feedback
Pulse synchronization Active-passive feedback
Projective synchronization
D-B synchronization
12 1 Introduction
nonlinear systems, they cannot be decomposed due to physical nature or natural

characteristics or other reasons. Then, the drive-response synchronization method
will fail.
Active-Passive Synchronization: In 1995, Kocarev and Parlitoz proposed active-

passive synchronization method [21]. The advantage of this method is that it can
choose the function of drive signal without limitation. Therefore, it is very flexible
and has greater universality and practicality, and it includes the drive-response syn-
chronization method as a special case. Active-passive synchronization method is
particularly suitable for applications in secure communications. The method can also
be used to synchronize control between hyperchaotic systems.
Coupling Synchronization: In 1990, Winful and Rahman theoretically studied the

possibility of laser chaos synchronization. In 1994, American scholars Roy and
Thornburg and Japan scholars Suga-Wara and Tachikawa independently observed
synchronization between two laser chaotic systems in experiments. Study results
show that mutually coupled chaotic system can achieve chaotic synchronization un-
der certain conditions. Kapitaniak and Chua achieved synchronization between two
Chua’s chaotic circuits with mutual coupling method [22]. Dynamical behaviors of mu-
tually coupled nonlinear systems are very complex. So far there is no universal theory.
However, mutually coupled nonlinear systems widely existed in nature, and therefore,
it is very important to study this synchronization method.
Continuous Variable Feedback Synchronization: In 1993, German scholars Pyragas

and Tamasevicius proposed a method to control continuous nonlinear chaotic sys-
tems [23]. Later, this idea was used to study the synchronization between two chaotic
systems, and it is called continuous variable feedback synchronization method. Vari-
able feedback synchronization method is simple and easy to implement. According to
different systems, the single variable, multivariables, or even all variables of the sys-
tem can be flexibly used to feedback control. Research results show that the minimum
number of feedback variables should be equal to the number of positive Lyapunov
exponent of the system without perturbation. Meanwhile, multivariable feedback is
more effective than single-variable feedback.
Adaptive Synchronization: In 1990, Huberman and Lumer employed adaptive

method to control chaos [24]. John and Amritker applied this principle to chaos syn-
chronization, and the phase space trajectory of the chaotic system is synchronized
with the desired unstable orbit [25]. Specifically, the adaptive synchronization method
is to achieve chaos synchronization by automatically adjusting certain parameters
of system with adaptive control technology. There are two prerequisites for the ap-
plication of this method. (1) At least one or more parameters of the system can be
obtained. (2) For the desired orbit, the values of these parameters are known. Ad-
justment of system parameters depends on two factors: (1) the difference between
output variable of the system and corresponding variable of the desired orbit and (2)
the difference between the controlled parameter value and the corresponding variable
value of desired system.
Pulse Synchronization: In 1997, Yang and Chua put forward pulse synchronization
control method [26, 27]. For pulse synchronization, the response system is driven by
single pulse transferred from drive signal. The transmitted signal is incomplete chaotic
signal, so it is more secure. Pulse synchronization has strong noise immunity and ro-
bustness, so it is a promising method of chaos synchronization. According to the basic
theory of impulsive differential equations, the stability of synchronization error sys-
tem depends on its comparative systems. When the comparative system approaches
stable state asymptotically, pulses synchronization becomes stable, and the pulse
intervals can be obtained.
Projective Synchronization: In 1999, Mainieri and Rehacek observed a new phe-

nomenon – projective synchronization in the research of some linear chaotic systems
[28]. By employing this synchronization, for some linear chaotic systems, when an
appropriate controller is selected, the output phase of the drive system and re-
sponse system will be locked, and the amplitude of each corresponding output is
also evolved according to a certain proportional relation. This new synchronization
method caught researchers’ attention, and a lot of projective synchronization schemes
are proposed. Recently, a function projective synchronization method is proposed
[29]. Compared with the general projective synchronization, function projective syn-
chronization means that the drive system and response system can be synchronized
according to a certain function proportion, which has important significance for the
realization of chaotic secure communication. In engineering practice, the parameters
of a system may be unknown. In order to realize parameter identification of chaotic
systems with unknown parameters, people began to apply adaptive synchronization
method to the synchronization research of chaotic systems with unknown parameters.
Obviously, combining adaptive control and function projective synchronization to re-
search the synchronization of chaotic systems with unknown parameters has more
theoretical value and practical significance [30].
Dead-Beat Synchronization: In 1995, Angeli et al. proposed a synchronization
method for discrete chaotic systems, called dead-beat synchronization method [31].
The most important feature of this method is that as long as there are several steps of
iteration, chaos synchronization can be achieved accurately. Although several syn-
chronization methods described above can be extended to a discrete system, they
are different from dead-beat synchronization method. First, dead-beat chaotic syn-
chronization is accurate synchronization, and other synchronization is asymptotic
synchronization in a certain sense. Second, the synchronization time for dead-beat
synchronization method is the time for iteration N (N is the dimension of the state
space). The synchronization time for other synchronization methods is not only
closely related with system parameters, but also depends on the given accuracy range.
14 1 Introduction
Performance of different synchronization methods is different. Based on Lorenz

system, researchers have conducted simulation experiments about synchronization
time and sensitivity to parameter’s change with the drive-response synchronization,
mutual coupling synchronization, feedback perturbation synchronization, adaptive
control synchronization, and pulse synchronization. Results show that for feedback
synchronization between Lorenz systems, the robustness to parameter changes is
the best, and the establishment of synchronization is the fastest. Setup time for
adaptive synchronization and pulse synchronization is longer. Weak robustness to
parameter change means relatively complex, but its security is better. Therefore, from
the synchronization setup time and robustness to parameter changes, the feedback
synchronization method is the optimal synchronization method.
1.5.2 Other Synchronization Methods and Problems
In the twenty-first century, chaos synchronization study shows a new trend. First, new
synchronization methods have been put forward. In addition to improving the existing
synchronization methods, the main work is to introduce advanced control theory and
technology into chaos synchronization research, such as fuzzy control [32], genetic
algorithm [33], and state observer method [34]. All this methods have achieved good
results. Second, the object of study has turned to discrete chaotic systems from con-
tinuous chaotic system, from low-dimensional chaotic system to high-dimensional
hyperchaotic system. Third, how to improve the performance of chaotic synchroniza-
tion began to draw more attention [35], such as to improve the synchronization system
performance with system identification and neural network technology.
Synchronization is the key to achieve chaotic security communication. The most
recent and the most competitive studies are to achieve chaotic secure communication
based on chaos synchronization. As the foundation of chaos theory and technology,
chaos synchronization also is an important aspect of a chaotic mechanism. Many
chaos synchronization methods perhaps will not soon get practical value, but each
new proposed method will give a new inspiration to open up a new avenue of research.
Although many chaos synchronization methods have been proposed, they have not
reached the mature stage. There are still many theoretical and technical issues to be
resolved in practical applications.
1. Establishing chaos synchronization theory. At present, some new chaotic syn-
chronization methods, such as genetic algorithm, neural network, and fuzzy
control, still lack universal theory. The synchronization theory needs further
establishment and improvement.
2. Improving synchronization performance. Research on the performance of syn-
chronization system is directly related to the engineering application of chaos.
Carrying out this research is of great significance. In the case of mismatch between
noise and parameter, chaos synchronization may be lost. It is the key technology
of chaotic secure communication to improve the stability of synchronization and

to increase the anti-disturbance ability of the synchronization system by using the
robust control method and the stability of dynamic system. We will discuss this in
Chapter 4 in detail.
3. Developing chaotic network synchronization method. With the development of
computer technology and network technology, the Internet has become the main
means to transmit information for people. It is convenient, and meanwhile it
has brought huge information security risks. How to achieve information-secure
transmission becomes a hot spot. It is an effective way to realize the information
security by exploring the real-time information encryption transmission based on
chaotic network synchronization and network transmission protocol. It has broad
application prospects.
4. Combining other chaos communication methods to promote the development
of chaotic communications. As described above, besides chaos synchronous
communication, there is chaos-coded communication. The two methods can be
combined together. If an information signal is encoded first, and it is transmit-
ted based on chaotic synchronization, then it effectively increases the difficulty of
deciphering. The fifth, sixth, and seventh chapters of the book will do in-depth
research on real-time chaotic secure communication based on chaotic encryption
and chaotic synchronization.
1.6 Summary of Chaos Secure Communication

Since 1990, chaotic secure communication and chaotic encryption technology have
become a hot topic in the field of international electronic communications [36–39].
So far, the chaos secure communication applications are roughly divided into three
categories [18]. The first is to achieve secure communication using chaos directly. The
second is to achieve secure communication based on chaotic synchronization signal.
The third is digital encrypted communication based on chaos sequence. At present,
the second class, chaos synchronous communication, is an international research fo-
cus. It has become a new high-tech field. With the development of network technology
and computer technology, the third class of chaotic encrypted communication has
captured more attention.
So far, four chaos synchronous secure communication technologies have been
proposed and developed, including chaotic masking [40], chaos shift keying [41],
chaotic modulation [42], and chaotic spread spectrum [43, 44]. The first category
belongs to analog communications, and the other three belong to digital communic-
ations. At present, the four secure communication schemes are the most competitive
technologies. There are three main ways to realize chaos secure communication, such
as electrical systems, laser systems, and computer networks. The most mature tech-
nology is electrical systems. Here, the overview will focus on the basic principles and
16 1 Introduction
study progress of chaotic synchronization secure communication technology based

on circuit systems and chaotic encryption technology based on computer network.
1.6.1 Chaotic Analog Communication
The typical technique of chaotic analog communication is chaotic masking. Chaotic

masking is to superpose the message signal into the chaotic signal directly, and
the signal is masked by the randomness of chaotic signal. Cuomo and Oppenheium
constructed chaotic masking secure communication system based on Lorenz system
[19, 45], combining two response subsystems into one complete response system,
which the structure is identical with drive system. The receiver can replicate all states
of the sender and achieve synchronization. At the sender, the message signal over-
laps with larger amplitude chaotic signal to form noise-like signal. At the receiver, the
chaotic signal is obtained from the response system, and the message signal is re-
covered by subtracting the chaotic signal generated by the response system from the
received signal.
Chaotic masking requires that parameters of chaos circuit of the sender and
the receiver are accurately matched. Minor differences of parameters may cause
synchronization fail. It is very difficult for an attacker to obtain the key parameters,
but the strict matching of parameters also puts forward a very high request to the
circuit design.
1.6.2 Chaotic Digital Communication
Chaos digital communications include chaos shift keying, chaotic parameter modu-
lation, and chaotic spread-spectrum communication. Chaos shift keying is a chaos
communication method that is suitable for digital communications. It can code a
binary signal using different attractors produced by different parameters. For ex-
ample, code 1 indicates parameter ,1 , and its corresponding chaotic attractor is A1 .
Code 0 indicates parameter ,2 , and its corresponding chaotic attractor is A2 . The
behaviors of chaos switch between A1 and A2 . Response time of the system is con-
trolled by changing the parameter. A parameter at the transmitter is modulated by
using a binary signal, and then the chaotic signal is sent to the receiver. The mod-
ulated signal can be detected at the receiver by using the synchronization error.
Thereby the real signal is recovered. Chaos shift keying has better robustness than
that of chaotic masking method, and the anti-interference performance is better, but
it has lower information transmission rate. The original signal may be recovered by
using the short-time zero crossing rate method, but the security will be reduced.
So researchers have proposed an improved chaos shift keying digital communica-
tion schemes, including chaotic on-off keying (COOK), differential chaos shift keying
(DCSK), frequency modulation differential chaos shift keying (FM-DCSK) modulation,
etc. [46–48].
Chaos parameter modulation is to modulate certain parameters of chaotic sys-

tem by using the original signal in chaos oscillator in the sender. The changes of this
parameter reflect changes of the original signal. There is a synchronous chaotic sys-
tem in the receiver. Because the changes of chaotic state also include the changes
in the original signal, the original signal can be detected and recovered by a nonlin-
ear filter. Yang and Chua proposed a chaotic parameter modulation scheme in 1996,
which is suitable for general signal modulation [42] and comprehensively analyzed
several methods for parameter modulation based on Chua circuit. The simulation
results show that, when modulating the different parameters of Chua circuit, the sig-
nal recovery accuracy is different. If we modulate capacitor and resistor, the recovery
accuracy is higher.
Chaotic spread-spectrum communication is to achieve spread-spectrum commu-
nication by using chaotic sequence instead of spreading codes. The traditional code
division multiple access (CDMA) technology is mainly limited by the periodicity of the
PN code and the available number of orthogonal PN code address. Because chaotic
systems are sensitive to initial conditions, a large number of spreading sequences
with good correlation properties can be generated through the evolution of chaotic
system. Based on the statistical characteristics of the chaotic sequence, it is very
effective for the realization of the chaotic spread-spectrum code division multiple ac-
cess communication. Chaotic sequence can be generated by an initial value and a
mapping formula, without having to store values of the sequence. Therefore, it is
more suitable for spreading code in spread-spectrum communication, and it has broad
application prospects.
In all, chaotic communication has the following advantages [18]:
1. High security. The security of chaotic communication mainly comes from complex
dynamic behavior of chaos system, sensitivity to initial conditions, and long-term
unpredictability of dynamic behavior.
2. High capacity of dynamic storage.
3. Low power and low accessibility.
4. Low cost.
Based on the advantages above, applications of chaos in secure communication are

still in the initial stage of the study; it has captured a lot of attention and interest from
physics, information science, and other interdisciplinary fields. It is foreseeable that
chaotic secure communication will play an important role in people’s lives, especially
in the military and national information security.
1.6.3 Encryption Communication based on Chaotic Sequence
In 1989, British mathematician Matthews proposed a new encryption method –

chaotic encryption method [49], and thus the study of chaos cryptography began.
18 1 Introduction
In the past decades, along with the deepening theoretical study of chaos, the ap-
plication scope of chaos theory is also expanding. Chaos cryptography application
has become a hot topic, and a number of chaotic encryption algorithms have been
proposed [50, 51]. Chaotic systems can provide pseudorandom sequence with good
randomness, correlation, and complexity. The key sequences generated by a chaotic
system not only exhibit excellent cryptographic performances, but also have a rich
source. In addition, the chaotic encryption method greatly simplifies the design pro-
cess of traditional sequence cipher. Those attractive features make it possible to be a
stream cryptosystem. The unique advantages of chaos in cryptography attracted many
scholars and academic communities around the world to do research.
Unlike chaos analog communication, the anti-crack ability of chaotic encrypted
communication is strong. Chaos analog communication using small signal modula-
tion cannot resist the attacks of neural network and regression map. While for chaos
encrypted communications, the attacker cannot get the complete original data, and
the chaotic signals generated by original equation or regression map reconstruction
cannot be obtained according to neural network or regression method. It is diffi-
cult to decipher the encrypted signals and it can be applied to highly confidential
communication systems.
With the development of computer technology, digital technology, and network
technology, chaos encryption software has been emerging persistently [52]. Chaos-
based engineering applications have also appeared [53], and the hardware achieve-
ments based on Digital Signal Processing (DSP) and Field Programmable Gate Array
(FPGA) technologies have also been reported [54]. Chaotic encryption with digital
technology has many advantages, such as low requirements for accuracy of circuit
elements, easy hardware implementation, easy computer processing, less loss of in-
formation transmission, versatility, and so on. Especially when it is used in real-time
signal processing, the encrypted information is difficult to be deciphered. Therefore,
chaotic encryption shows strong vitality in secure communications.
Compared with conventional packet encryption system, chaos encryption system
has the following characteristics:
1. Complex key signal generated by simple circuit or iterative equation.
2. Key signal is unpredictable. There is no exclusive corresponding plaintext–
ciphertext pair, and the same plaintext corresponds to different and irrelevant
ciphertext. So it is similar to one-time pad.
3. Design method and implementation technology of key space need to be further
studied.
4. Security of systems can only be tested by limited numerical algorithm.
1.7 Chaos Research Methods and Main Research Contents

Because of the complexity of chaos, the research on the chaotic motion should employ
many methods, including theoretical analysis, numerical simulation, experimental
1.7 Chaos Research Methods and Main Research Contents 19
study, and so on. At present, the methods on the chaos theory mainly include nu-
merical calculation, symbolic dynamics, Melnikov and Shilikov method, phase space
reconstruction method, and so on. The numerical methods for judging the chaotic
state are the Lyapunov exponents, fractal dimension, power spectrum, Poincaré
section, direct observation, and so on. Among them, the Lyapunov exponents and
fractal dimension are quantitative methods, and the others are qualitative descrip-
tions. Power spectral density shows the signal changes with the frequency by Fourier
transform. The power spectrum of a chaotic signal is changed continuously. Poincaré
section is a section in the phase space. The phase trajectory leaves a point at the cross
section when it passes the cross section so that the flow in n-dimension can be re-
duced to a map with n – 1 dimension, and many properties of the original system are
preserved. For a chaotic system, these points have a fractal structure and cannot fill
up the entire section. To observe whether there exists butterfly effect is a typical dir-
ect observation method. Comparing the time series of chaotic system, two very similar
initial values after a short-time evolution will be very different. It reflects the sensit-
ivity of chaotic system to initial value. For the chaotic systems with known dynamics
equations, the symbolic dynamics method is used to study the dynamics equation.
In the second chapter of this book, we will focus on the analysis methods of chaotic
characteristics.
Numerical simulation is an important method for the study of chaos theory and
application. First of all, numerical simulation method provides the application issues
associated with the real system for chaos theory analysis, and it broadens the field
of theory research. Simulation system provides ideal plasticity experimental model
for theoretical research, and it opens up the new field of chaotic application study.
Second, the simulation method can provide the basis for the feasibility of construct-
ing the real and practical system. The third and fourth chapters of this book are based
on the Matlab/Simulink numerical simulation method. The third, fourth, and fifth
chapters mainly use the numerical simulation method based on Matlab/Simulink to
study.
Real experimental study is an important method for the practical application
of chaos. Since the 1990s, the results of a large number of theoretical studies are
based on real system experiments and applications. Real experiments include circuit
experiments and computer experiments. With the development of microelectronics
technology, it is possible to design various kinds of high precision chaotic compon-
ents and systems. The development of network technology and computer technology
provides a platform for constructing and testing various chaotic encryption softwares.
The fifth and sixth chapters of this book will be mainly based on computer system and
network.
Three kinds of chaos analysis methods discussed above are mutual depend-
ence and mutual promotion, and constantly push forward the development of chaos
research. In this book, the theoretical analysis, numerical simulation, and real exper-
iments are combined. This book will mainly use the Lyapunov exponents, conditional
20 1 Introduction
Lyapunov exponents, and fractal dimension for the theoretical study. Based on Mat-
lab/Simulink simulation platform, we analyze chaotic system synchronization control
mechanism and the chaotic synchronization performances. Through computer ad-
vanced language and computer network, we carry out real experimental researches,
designing practical chaotic encryption software, testing it, and analyzing the security.
In the study of chaos synchronization control, both the continuous system
synchronization and discrete system synchronization are studied. The relationship
between the performance and the control parameters is particularly studied. In the
application of chaotic systems, we will focus on the integration of chaotic dynam-
ics, advanced control technology, modern cryptography, and modern communication
technology. Combining the physical synchronization of chaotic system with com-
puter programming technology, static data file encryption softwares are designed. The
network-based self-synchronization method is investigated. Chaotic encryption soft-
ware to achieve real-time voice signal transmission is designed by using the advanced
computer language. Those above are the main achievements of the research group in
the field of chaotic application.
Questions
1. What is chaos? What are the main characteristics of chaos?
2. What is butterfly effect? What is its philosophical significance?
3. What is the definition and significance of the Feigenbaum constant?
4. What are the main techniques of chaotic secure communication? What are the
characteristics?
5. Why chaos synchronization is the key technology of secure communication?
2 Characteristic Analysis Methods for Nonlinear
System
Because of the complexity of the nonlinear system, how to judge the chaotic behavior
of a nonlinear system is an important question in the study of chaos theory and ap-
plication. Generally, methods to determine the chaotic characteristics of a nonlinear
system include phase diagram analysis, power spectral analysis, Lyapunov exponent
spectrum analysis, fractal dimension analysis, dividing frequency sampling method,
pseudo-phase space method, Poincaré section, 0–1 test, and so on.
2.1 Phase Diagram Analysis Method

Phase diagram analysis is one of the direct observation methods. Characteristics of
the aperiodic motion of a chaotic motion can be displayed by the method of phase
plane graph. Periodic motion repeats the previous movement at every cycle, so the
trajectory of its movement is a closed curve. Chaotic motion is a nonperiodic motion,
so its trajectory curve is a never-closed curve. The reciprocating movement reflected
in the trajectory curve is confined to a bounded area. It does not diverge to infinity,
and it does not converge to a stable point; thus it forms a strange attractor. When we
simulate a nonlinear system, its dynamic behavior can be determined by the phase
diagrams of the system directly, including limit circle, period movement, and chaos.
The disadvantage of phase trajectories diagram is that when the periodic motion of
the cycle is very long, it is difficult to accurately distinguish periodic motion or chaotic
motion only based on the phase plane diagram.
A chaotic attractor has a complex structure by stretching and folding, which leads
to the system being kept in a limited space. It is the result of the global stability and
local instability of the dynamic system. The global stability makes all the movements
outside the attractor get closer to the attractor, and the trajectory converges to the at-
tractor. While the local instability makes all the arrived inside the attractor trajectory
mutually reject, and the trajectory spreads in a certain direction. A small perturba-
tion is stable in the chaotic attractor, and it will eventually reach the attractor. But in
the interior of the chaotic attractor, the motion state of the system is very sensitive to
the initial condition. In other words, if the location entering the chaotic attractor is
slightly different, then the difference will grow exponentially and will eventually lead
to different chaotic orbits. A chaotic attractor has other properties, such as infinite
embedded self-similarity structure and fractal dimension.
For example, the periodic forced nonlinear system equation is [55]
ẍ + aẋ – x + bsgn(x) = F sin(9t), (2.1)

22 2 Characteristic Analysis Methods for Nonlinear System
0
y
–1
–2
–2 –1 0 1 2
x
Figure 2.1: Strange attractor of system (2.1) on x–y plane.
where a, b, F, and 9 are constants; sgn(x) is a symbol function. When a = 1.05, b = 1,

F = 1.1, and 9 = 1, the ‘Onion’ chaotic attractor of the system in the x–y plane is shown
in Figure 2.1.
2.2 Power Spectral Analysis Method

Power spectral density function is often used to characterize the randomness of
chaotic motion. Power spectrum indicates the statistical characteristics of the fre-
quency components in the random motion process, so it is the basic method to
analyzes chaotic motion in nonlinear systems. According to the randomness of chaotic
motion, we often use the spectrum analysis method to identify chaos in a nonlinear
system. Chaotic motion is a nonperiodic motion that can be regarded as the superpos-
ition of periodic motion with infinite number of different frequency, and its power
spectrum has the characteristics of random signal. Obviously, the power spectrum
of chaotic motion is continuous spectrum. Power spectrum analysis can be used
to determine whether the motion is random, but it cannot be determined whether
this random motion is due to the external random disturbance or the internal ran-
domness nature of the deterministic nonlinear system. Therefore, it is difficult for
power spectrum analysis to distinguish between chaotic motion and real random
motion.
The power spectrum analysis can provide the signal frequency domain inform-
ation, and it can be used to observe whether a nonlinear system has chaotic char-
acteristics. For periodic motion, the power spectrum is discrete spectrum, and for
chaotic motion, the power spectrum is continuous spectrum. So we can judge whether
a system is chaotic through the power spectrum diagram of the system.
2.2 Power Spectral Analysis Method 23
According to Fourier transformation, for a function x(t), –∞ < t < ∞, if x(t) satisfies
the Dirichlet condition
+∞
|x(t)|dt < ∞, (2.2)
–∞
then we have
+∞
x(t) = F(f )ei20 ft df , (2.3)
–∞
∞
where F(f ) = –∞ x(t)e–i20 ft dt. We call F(f ) the Fourier transformation or the spectrum
of x(t). The following is the relationship between F(f ) and x(t):
+∞ ∞
|x(t)|2 dt = |F(f )|2 df , (2.4)
–∞ –∞
where the left-side term represents the total energy of x(t) over the range (–∞, ∞),
while |F(f )|2 in the right side represents the spectral density of x(t), which means the
energy or power per unit interval is on the frequency scale. Equation (2.4) is the spec-
tral representation of the total energy of time function. The average power per unit
interval on the frequency scale is
1
G(f ) = lim |F(f )|2 . (2.5)
T→∞ 2T
It is also defined as the average power spectral density of x(t).

According to Fourier analysis, a periodic signal x(t) can be decomposed into
Fourier series
∞

x(t) = cn ejn90 t , (2.6)
n=–∞
where
T/2
1
cn = x(t)e–jn90 t . (2.7)
T T/2
A quasi-periodic signal also can be decomposed into a series of frequencies of the

superposition of the sine vibration. Both of them are discrete spectrum.
If an aperiodic signal x(t) satisfies the condition of being absolutely integrable,
then it can be represented by Fourier integral
∞
1
x(t) = X(9)ej9t d9, (2.8)
20 –∞
∞
X(9) = x(t)e–j9t dt. (2.9)
–∞
Obviously, the frequency spectrum of nonperiodic motion signal is continuous spec-

trum.
To represent the frequency domain characteristics of chaotic signals, we can cal-
culate the Fourier transform of its autocorrelation function Rxx (4) and analyze its
frequency domain characteristics by auto-power spectral density function Sxx (f )
∞
Rxx (4) = Sxx (f )e–j20f 4 df . (2.10)
–∞
For periodic motion, the power spectrum appears at the fundamental frequency and
its frequency doubling. For the quasi-periodic motion, the power spectrum is located
in irreducible fundamental frequency and their superimposed frequency spikes. The
power spectral can display bandwidth characteristics of the noise signal. When the
period-doubling bifurcation occurs, frequency division and frequency doubling occur
in power spectrum, and peaks will display at this frequency points of the power spec-
trum. The chaotic motion is characterized by a continuous spectrum with the noise
background in the power spectrum, which contains the peak corresponding to the
periodic motion. According to these characteristics, the motion of a system can be
easily identified as the periodic, quasi-periodic, random, or chaotic.
In order to obtain the reliable power spectrum, the average of several sequential
sampling sequence spectra is required. In addition, the transition process must be
ended before the start of the sampling. If the original data contains a large amount
of noise and external interference, the appropriate filtering or smoothing should be
considered.
2.3 Poincaré Section Method

Poincaré section was proposed by Poincaré at the end of the nineteenth century,
which was used to analyze the motion characteristics of the multivariable autonomous
system.
The phase diagram of the Poincaré map well characterizes the reciprocating non-
periodic characteristics of chaos. If the Poincaré map is neither a finite set nor a closed
curve, then the corresponding system motion is a chaotic motion state. More specific-
ally, if the system does not have an external noise disturbance, and there is a certain
damping, the Poincaré map will be a point set with a detailed structure. If the system
disturbed by external noise or damping is small, the Poincaré map will be a set of fuzzy
points. If the Poincaré is a finite set of points, then the corresponding system motion
is periodic motion state. Since the resolution of Poincaré section is higher than that
of the phase plane graph, the Poincaré section is a popular method for the analysis of
chaotic dynamics.
A section is appropriately selected in phase space. On this cross section, if a pair
of conjugate variables is fixed, it is called Poincaré section. The intersections of the
Poincaré section with the continuous trajectory in the phase space are called the cutoff
points. By observing the distribution of the cutoff points of the Poincaré section, we
can judge whether the chaotic phenomenon occurs. When there is only one fixed point
or a small number of discrete points on the Poincaré section, the motion is periodic.
When there is a closed curve on the Poincaré section, its movement is quasi-periodic.
When there are pieces of dense points with a fractal structure on the Poincaré section,
the motion is chaotic.
To observe the route to chaos of the system (2.1), the phase diagram and the Poin-
caré section of the system are shown in Figure 2.2 with a = b = 1 and F change. The
Poincaré section is z mod 20 = 1. With the increase of F, the limit cycle phase diagram
is observed as shown in Figure 2.2(a1 ), and its corresponding Poincaré section is a
point as shown in Figure 2.2(a2 ). To increase the parameter F, the limit cycle phase dia-
gram by period doubling is obtained as shown in Figure 2.2(c1 ) and its corresponding
Poincaré section is the two points as shown in Figure 2.2(c2 ). To increase the paramet-
ers F further, the limit cycle phase diagrams by period doubling again is obtained as
shown in Figure 2.2(d1 ) and the corresponding Poincaré section is the four points as
shown in Figure 2.2(d2 ). Finally, the chaotic attractor phase diagram is shown in Fig-
ure 2.2(e1 ) and the infinite points on the corresponding Poincaré section are shown
in Figure 2.2(e2 ). Thus the process of chaos of the system (2.1) by the period-doubling
bifurcation is demonstrated by the phase diagram and the Poincaré section.
At the same time, we can observe fork bifurcation with parameter F increasing
from 0.65 to 1.045 as shown in Figure 2.2(b), and then the two coexisting attractors
expand as shown in Figure 2.2(c) and (d) and finally merge by an attractor-merging
crisis bifurcation at the onset of chaos as shown in Figure 2.2(e).
2.4 Lyapunov Characteristic Exponent Method

To quantitatively characterize the sensitivity to initial value of chaotic system, the
Lyapunov characteristic exponent is needed.
2.4.1 Definition of Lyapunov Characteristic Exponent
Lyapunov exponent refers to the average change rate of two trajectory separations or
aggregates to each other with the time going on in the phase space. In the practical
application, the maximum Lyapunov exponent +1 is of great significance, and it can
be calculated by
M

1 D(tk )
+1 = ln , (2.11)
tM – t0 D(tk–1 )
k=1
where D(tk ) is the distance of most adjacent two points at tk ; M is the iterative number;
+1 is not only the quantitative index of the chaotic attractor, but also the quantitative
2 2
(a1) (a2)
1 1
0 0
–1 –1
–2 –2
–2 –1 –0 –1 –2 –2 –1 –0 –1 –2
2 2
(b1) (b2)
1 1
0 0
–1 –1
–2 –2
–2 –1 –0 –1 –2 –2 –1 –0 –1 –2
2 2
(c1) (c2)
1 1
0 0
–1 –1
–2 –2
–2 –1 –0 –1 –2 –2 –1 –0 –1 –2
2 2
(d1) (d2)
1 1
0 0
–1 –1
–2 –2
–2 –1 –0 –1 –2 –2 –1 –0 –1 –2
2 2
(e1) (e2)
1 1
0 0
–1 –1
–2 –2
–2 –1 –0 –1 –2 –2 –1 –0 –1 –2
Figure 2.2: Phase diagram (left) and its corresponding Poincaré section (right) of system (2.1):
(a) F = 0.65, 1T period; (b) F = 0.95, 1T period; (c) F = 1, 2T periods; (d) F = 1.01, 4T periods; and
(e) F = 1.045, chaotic motion.
description of the initial deterministic amplification ratio of chaotic system. In addi-

tion, its reciprocal is generally considered to be the maximum predictable time scale
of the metric system state.
2.4.2 Lyapunov Exponent Spectrum
For systems with dimensions greater than 1, there is a set of Lyapunov exponents, of-
ten referred to as the Lyapunov exponent spectrum. Each of the Lyapunov exponents
characterizes the convergence of the trajectory in a certain direction.
Now, a definition of Lyapunov exponent spectrum is given. For a continuous dis-
sipative dynamical system with n-dimensional phase space, it is assumed that the
long-term evolution of an infinitesimal n-dimensional sphere among the initial condi-
tions of the system can be controlled. And because of the evolution of the natural de-
formation, the ball will become ellipsoid. If all the spindles in the sphere center on the
ellipsoid attractor are ranked by the order from the fastest to the slowest speed, then
the ith Lyapunov exponent is defined according to the increase rate pi (t) of the ith axis

1 pi (t)
+i = lim ln . (2.12)
t→∞ t p0 (t)
Note that the linear range of the ellipse increases with e+1 t , and the area defined by
the first two principal axes increases with e(+1 ++2 )t , and the volume defined by the
first three principal axes increases with e(+1 ++2 ++3 )t , and so on. In fact, this feature is a
definition of the Lyapunov exponent spectrum, i.e. the sum of the first j exponents is
determined by the long-term average rate of the volume defined by the first j principal
axes. If the evolution of all the points on the sphere is considered, the behavior of the
phase space can be visualized.
For example, the simplified Lorenz system equation is [56]
⎧
⎪
⎨ẋ = 10(y – x)
⎪
ẏ = –xz + (24 – 4c)x + cy , (2.13)
⎪
⎪
⎩ż = xy – 8z/3
where c is the system parameter, and when c ∈ [–1.59, 7.75], the system is chaotic. Its
two Lyapuov exponents are shown in Figure 2.3. When +1 > 0, +2 = 0 and when +3 < 0,
the system is chaotic state. When +1 = 0, +2 < 0, and +3 < 0, the system is periodic
state.
2.4.3 Physical Meaning of Lyapunov Exponent
Sensitivity of a system to the initial value can be effectively characterized by the

Lyapunov exponent spectrum of the system. If the Lyapunov exponent is negative,
0.5
LE sprectrum
–0.5
–2 0 2 4 6 8
c
Figure 2.3: Lyapunov exponents of the simplified Lorenz system with different parameter c.
it indicates that the phase volume of the system shrinks in this direction, and the
movement of the system is stable. For a nonlinear dissipative system, the sum of the
Lyapunov exponents must be negative, and it indicates that the overall motion of the
system in the phase space is stable and contraction. While a positive Lyapunov ex-
ponent indicates that the system phase volume expands and folds in one direction
so that the originally adjacent trajectories in the attractor become increasingly irrel-
evant, and the long-term behavior of the system becomes unpredictable, namely the
so-called initial value sensitivity. At this time, the motion is chaotic state, and it has
a strange attractor. So, for a chaotic system, one Lyapunov exponent is greater than
zero at least.
Lyapunov exponent can depict the strange attractors. The strange attractor is a
phase diagram with fractal structure in the phase space. Since the chaotic attractor
is often of noninteger dimension, the confirmation of strange attractors is signific-
ant for studying the chaotic motion. If a system’s Lyapunov exponent spectrum is +1 ,
+2 , . . . , +n (from large to small), and the system has a chaotic attractor, it must satisfy
the following conditions:
1. At least one positive Lyapunov exponent.
2. At least one of Lyapunov exponents +i = 0, 1 < i < n.

3. The sum of the Lyapunov exponents is negative, that is, ni=1 +i < 0.
The first condition above indicates that the adjacent track in the phase space is separ-
ated with the exponential rate, and the system is sensitive to the initial value, which
is the prominent feature of the chaos. So the Lyapunov exponent spectrum must have
positive component, which is an important symbol to distinguish the strange attractor.
Sum of all positive Lyapunov exponents is defined as Kolmogorov entropy, which is a
measure of the degree of disorder of the system, and it is a characterization of chaos.

The second condition is the periodic performance of the chaotic attractor, and the
period is infinite. The third condition shows that the phase volume of the system is
contracted, thus ensuring the stability of the system. For a three-dimensional chaotic
system, the symbols of Lyapunov exponent spectrum have only one case, namely
(+, 0, –), and +1 < –+3 .
2.4.4 Condition Lyapunov Exponent
Condition Lyapunov exponent (CLE) is a generalization of the Lyapunov exponent in

the chaotic synchronization system. In the research of chaos synchronization, the CLE
is an important quantitative index, and it is important to guarantee synchronization.
Here, the conditional Lyapunov exponent is illustrated by employing Pecora–Carroll
(PC) chaos synchronization method.
According to PC chaotic synchronization method, an n-dimensional autonomous
dynamical system
u̇ = f (u) (2.14)
is decomposed into two subsystems:
v̇ = g(v, w), ẇ = h(v, w), (2.15)
where the first subsystem variable v = (u1 , u2 , . . . , um ), and the second subsys-
tem variables v = (um+1 , um+2 , . . . , un ), g = (f1 (u), f2 (u), . . . , fm (u)), h = (fm+1 (u),
fm+2 (u), . . . , fn (u)). Copy a new system
ẇ′ = h(v, w′ ). (2.16)
It is exactly the same as the system w. Obviously, the synchronization condition

between w′ and w is the error w′ – w = Bw → 0 as t → ∞.
If Dw h is the Jacobian matrix of the subsystem, its corresponding Lyapunov ex-
ponent is called the conditional Lyapunov exponent of the subsystem w, and it is
defined as
1
+j = lim ln (Dw ht )x0 v . (2.17)
t→∞ t
where j = m + 1, . . . , n, and v is the unit matrix, and (Dw ht )x0 = (Dw h)xt
(Dw h)xt–1 , . . . , (Dw h)x0 .
Only when all of the CLEs are negative, system w′ can synchronize with w.
2.5 Fractal Dimension Analysis Method

Dimension is an important feature of the nonlinear system. Divergence of orbits of
chaotic system makes the whole phase space of the system become impossible to
be filled by phase trajectories, so chaotic attractor has a noninteger dimension. In
theory, the chaotic attractor has infinite-level self-similar structure. Therefore, fractal
dimension is an effective parameter to distinguish chaotic motion.
There are many definitions for dimension, for example, capacity dimension, cor-
relation dimension, information dimension, similarity dimension, and so on. The
fractional dimension of the chaotic system is generally discussed in its capacity di-
mension. The definition of capacity dimension is the number of cubes L(%) when one
uses an N-dimensional cube epsilon with the side length % to measure a given space.
It is defined by
ln L(%)
Dc = lim . (2.18)
%→0 ln(1/%)
The relationship between the capacity and the Lyapunov exponent spectrum is
j
+i
Dc = j + , (2.19)
–+i+1 + 1
i=1
where j satisfies +1 ≥ +2 ≥ ⋯ > +j ≥ 0 ≥ +j+1 ≥ ⋯ ≥ +N . If the above formula contains

+ = 0, then the number of + should be added to j.
The fractional dimension of the simplified Lorenz system with different parameter
c is shown in Figure 2.4. Obviously, when Dc = 0, the system converges to equilibrium
point. When Dc = 1, the system is periodic (limit cycle). When Dc > 2, the system is
chaotic. So the fractal dimension like the Lyapunov exponent can be used to describe
the dynamics of a nonlinear system.
2.6 0–1 Test Method

Recently, Gottwald and Melbourne [57] proposed a reliable and effective binary test
method for testing whether a nonlinear system is chaotic, which is called “0–1 test.”
The basic idea is to create a random dynamic process for the data and then study
how the scale of the stochastic process changes with time. Next, we discuss this
method and apply to analyze the characteristics of the fractional-order simplified
Lorenz system.
2.6.1 0–1 Test Algorithm
Consider a set of continuous time series x(t) to characterize the dynamic system, and
6(x) represents the observable data set. This test method is completely independent
2.6 0–1 Test Method 31
2.5
2
Fractal dimension
1.5
0.5
–0.5
–2 0 2 4 6 8
c
Figure 2.4: Fractional dimension of the simplified Lorenz system with different c.
of the specific form 6(x), that is, the condition can be satisfied for any form of 6. For
example, if x = (x1 , x2 , . . . , xn ), then 6(x) = x1 satisfies the condition. Similarly 6(x) =
x2 or 6(x) = x1 + x2 also satisfies the condition. Choose an arbitrary constant c > 0 and
define the following equations:
t
((t) = ct + 6(x(s))ds, (2.20)
0
t
p(t) = 6(x(s)) cos(((s))ds. (2.21)
0
Thus, p(t) in eq. (2.21) is only related to the observation data set 6(x), which demon-
strates the general property of the testing method. This test algorithm has no relation-
ship with the original data of the system and the system dimension. From the above
definition, we can get two conclusions as follows:
1. If the system is not chaotic, then the movement of p(t) is bounded.
2. If the system is chaotic, then the movement of p(t) is similar to that of the Brown
movement.
To characterize the growth of p(t), the root mean square displacement of p(t) is
defined as
T
1
M(t) = lim [p(t + 4) – p(4)]2 d4. (2.22)
T→∞ T 0
If the motion of p(t) is the Brown movement, M(t) linearly increases with time. If the
motion of p(t) is bounded, then M(t) is bounded. Define the asynchronous growth
rate of M(t) as
K = lim log M(t) / log t. (2.23)

t→∞
When the integral time t is long enough, then K = 0 or K = 1, which characterizes the
nonchaotic motion or chaotic motion, respectively.
Since the integral is only suitable for continuous systems, the solution of vector
integrals is not very convenient. An improved 0–1 test algorithm is introduced, which
is used to replace integrating with summing so that it has wider applications.
2.6.2 Improved 0–1 Test Algorithm
Consider a set of discrete data 6(n) sampled at times n = 1, 2, 3, . . ., which represents a

one-dimensional observable data set. Choose a constant c at random and define [58]
n

p(n) = 6(j) cos(((j)), n = 1, 2, 3, ..., (2.24)
j=1
n

s(n) = 6(j) sin(((j)), n = 1, 2, 3, ..., (2.25)
j=1
where
j

((j) = jc + 6(i), j = 1, 2, 3, ..., n. (2.26)
i=1
Then, on the basis of function p(n), or s(n), define the root mean square displace-
ment as
N
1 2
M(n) = lim p(j + n) – p(j) , n = 1, 2, 3, ... (2.27)
N→∞ N
j=1
Obviously, if the behavior of p(n) (s(n) behaves in the same way) is Brownian, then
M(n) grows linearly in time. If the behavior of p(n) is bounded, then M(n) is bounded.
The asynchronous growth rate is defined by
log M(n)
K = lim . (2.28)
n→∞ log n
Just as the same for continuous time system, K = 0 means that the system is regular
(i.e. periodic or quasi-periodic), and K = 1 means the system is chaotic. For a finite
data set 6(n), 1 ≤ n ≤ N, we define
N–n
1 2
M(n) = p(j + n) – p(j) . (2.29)
N–n
j=1
2.7 Dividing Frequency Sampling Method 33
If 1 ≤ n ≤ N, the scales of M(n) is similar to that described by eq. (2.27). Therefore, by

plotting log (M(n) + 1) against log n for 1 ≤ n ≤ N/10, the test still used to check whether
K is close to 0 (regular motion) or close to 1 (chaotic motion).
Note that another interesting feature of the 0–1 test is shown that the inspection of
the dynamics of the (p, s) trajectories provides a simple visual test about whether the
underlying dynamics is chaotic or nonchaotic, namely bounded trajectories in the (p,
s) plane imply regular dynamics, while Brownian-like (unbounded) trajectories imply
chaotic dynamics.
2.6.3 Application of 0–1 Test Algorithm
The simplified Lorenz system is described by eq. (2.13). Now, we carry out the 0–1
test with c = 7.2 and c = –1, respectively, and we obtain K = 0.0736 at c = 7.2, and
K = 0.7630 at c = –1, which corresponds to periodic state and chaotic state. The state
space plots are shown in Figure 2.5 and the corresponding translation components
(p, s) is shown in Figure 2.6, which corresponds to periodic motion and Brown motion.
Obviously, the results of 0–1 test are the same with that of Lypunov exponents and
phase diagram analysis [59].
2.7 Dividing Frequency Sampling Method

To avoid the confusion of complex motion in phase space, we can observe the rep-
resentative points (i.e. sampling points) in the phase space at every time interval (i.e.
the sampling period). Thus, the continuous trajectory in phase space is represented
by a series of discrete points. Dividing frequency sampling method is a generalization
of the scintillation sampling method in the experimental physics. It is the most
effective method for identifying the long period of chaos. It is mainly used to study
the period-doubling bifurcation and chaos phenomenon of nonlinear oscillator with
the periodic force.
(a) (b)
5 50
0
y
0
y
–5 –50
5
20 50
5
0 0
x –5 0 z x –20 0 z
Figure 2.5: State space diagrams of the simplified Lorenz system: (a) c = 7.2 and (b) c = –1
5 6000
(a) (b)
4000
–5
2000
s(n)
s(n)
–15
0
–25
–2000
–35 –4000
–25 –20 –15 –10 –5 0 5 10 15 –10000 –6000 –2000 2000
p(n)
p(n)
Figure 2.6: (p, s) plot of the simplified Lorenz system: (a) c = 7.2 and (b) c = –1.
For the forced vibration, the sampling period is often set to the period of the external
control force. When the sampling result is a point, the system motion is a periodic
motion (special case for the stable state). When the sampling results are n discrete
points, the system movement is also a periodic, and its motion period is n times of
the external control period. When the sampling results are infinite points, the system
movement is random. When the sampling results are concentrated in a certain region
and have a hierarchical structure, the system movement is chaotic. Continuously in-
creasing resolution, we can get the fine structure of the original distribution. It is a
kind of infinite-level self-similarity, and the self-similarity of the infinite level is the
scaling invariance of the nonlinear system.
Dividing frequency sampling method is suitable for all the nonlinear systems
driven by a periodic force for which its resolution is much higher than that of other
methods. The resolution is limited by a computer word “length.” But this method
also has some shortcomings. One is that it is not unique for the result interpretation,
and the other shortcoming is that the higher frequency over the sampling frequency
cannot be distinguished.
2.8 Pseudo-Phase Space Method

When we analyze the chaotic characteristics of the unknown dynamical systems, the
dividing frequency sampling method and the Poincaré section method are not ap-
plicable. Moreover, sometimes, it is easy to measure the time series of a variable in
the experiments, and we can use the time series to reconstruct the phase space. The
embedding theorem is used to solve the problem of establishing and describing the fi-
nite dimensional strange attractor and to reconstruct the dynamic system from a single
time series. Reconstructed phase space, i.e. the dimension of the pseudo-phase space
method (the embedding dimension), should satisfy m ≥ 2n + 1, where n is the real
dimension of phase space. Assuming the time series is {x(k), k = 1, 2, . . . , N}, appro-
priately select the time delay 4, which is an integer multiple of the sampling period.
Choosing x(k), x(k + 4), x(k + 24), . . . , x(k + (m – 1)4) as the coordinate axis, we can draw
a pseudo-phase space trajectory.
The process of reconstructing the attractor is equivalent to mapping the time
series {x(k)} to the m-dimensional Euclidean space Rm , and the topological properties
of the original unknown attractors may be maintained in the Rm space. Although the
pseudo-phase space method is used to construct the phase space of a variable at dif-
ferent time, the change of one variable of the dynamical system is related to the other
variables of the system. The dynamics of the whole system is implied by the change
of the variable with time. So the trajectory of phase space reconstruction also reflects
the evolution law of the system state. For the steady state, the results obtained by this
method are still a certain point. For periodic motion, the results are limited points,
and for chaotic system, the results are some discrete points with certain distribution
patterns or structures.
2.9 Complexity Measure Algorithm

Complexity is a quantitative description of the complexity by employing a certain
method. Obviously, different aspects of complexity are characterized by different
complexity algorithms. So far, there is no unified definition about complexity. Each
research field has its own understanding and interpretation about complexity. Horgan
pointed out that there are at least 45 definitions about complexity at present, such as
time complexity, space complexity, semantic complexity, Kolmogorov complexity, and
so on [60].
Complexity of chaotic systems means the degree of the chaotic sequence close
to the random sequence by use of correlation algorithm. The greater the complexity,
the more the sequence is close to the random sequence, and the security of the corres-
ponding application system is higher. In essence, the complexity of the chaotic system
is a kind of chaos dynamics.
Complexity of chaotic sequences is divided into behavior complexity and struc-
tural complexity. Behavior complexity is used to measure the size of the new pattern
probability of the short-time window. The larger the probability of generating a new
pattern is, the more complex the sequence is. At present, there are many compu-
tational complexity algorithms for chaotic pseudo-random sequence, and most of
them are based on Kolmogorov method and Shannon entropy, such as Lempl-Ziv al-
gorithm [61], approximate entropy (ApEn) algorithm [62, 63], fuzzy entropy (FuzzyEn)
algorithm [64, 65], intensity statistics complexity (ISC) algorithm [66, 67], and symbol
entropy (SymEn) algorithm [68]. The calculation speed of these algorithms is fast and
the result is accurate, but if the dimension is too high or the symbol space of the
pseudo-random sequence is too large, the result will overflow. Structure complexity
is used to measure the complexity of the sequence by the frequency characteristic and
energy spectrum in the transformation domain. The more uniform distribution of the
energy spectrum is, the original sequence is more close to the random signal, and
the original sequence is more complex. The structural complexity focuses on analyz-
ing the energy characteristic in the transform domain, which is based on all but not
the local sequence, and thus compared with the behavior complexity, the result has a
global statistical significance.
At present, structural complexity algorithms include spectral entropy [69] and
wavelet entropy algorithm [70], which are based on the Fourier transform and wavelet
transform and are applied to complexity analysis of discrete chaotic system and Elec-
troencephalogram (EEG) signals. In addition, the computing speed of C0 algorithm [71]
based on fast Fourier transform (FFT) is fast, and it is used in the field of biomedicine,
the brain electrical signal, brain optical images successfully for complexity analysis.
The important properties of C0 algorithm are proved in Ref. [72], which shows that
C0 algorithm can describe the random degree of the time series and can describe the
complexity. The algorithm is improved in Ref. [73], and the improved C0 algorithm is
more excellent.
In fact, the complexity of chaotic systems is one of the methods to characterize
the dynamics of chaotic systems. The effects of the method are similar to that of the
Lyapunov exponent, the bifurcation diagram, and the phase diagram. Here, spectral
entropy (SE) algorithm and C0 algorithm are introduced here.
2.9.1 Spectral Entropy Complexity Algorithm
Spectral entropy complexity algorithm can produce energy distribution by Four-

ier transform and calculates the corresponding spectrum entropy according to the
Shannon entropy; the algorithm is described as follows [74]:
Step 1: Remove the DC. We can use the next formula to remove the DC part of the
chaotic pseudo-random sequence {xN (n), n = 0, 1, 2, . . . , N – 1}, so that the spectrum
can reflect the signal energy more effectively.
x(n) = x(n) – x̄, (2.30)

N–1
where x̄ = 1
N n=0 x(n).
Step 2: Do Fourier transformation. Do discrete Fourier transformation for the se-

quence xN
N–1
N–1

20
X(k) = x(n)e–j N nk = x(n)WNnk , (2.31)
n=0 n=0
where k = 0, 1, 2, . . . , N – 1.
Step 3: Calculate the relative power spectrum. The first half of the sequence X(k) is
used to calculate. According to the Parseval theorem, the power spectrum of a certain
frequency point is calculated by
1
p(k) = |X (K )|2 , (2.32)
N
where k = 0, 1, 2, . . . , N/2–1. The total power of the sequence is defined as

N/2–1
1
ptot = |X(k)|2 (2.33)
N
k=0
Then, the relative power spectrum of the sequence is
N |X(k)|
1 2
p(k) |X(k)|2
Pk = = = . (2.34)
ptot 1
N/2–1
N/2–1
N |X(k)|2 |X(k)|2
k=0 k=0

N/2–1
We know that Pk = 1.
k=0
Step 4: Calculate spectral entropy. By using the relative power spectral density Pk and
the Shannon entropy, the spectral entropy se is obtained by
N/2–1

se = – Pk ln Pk , (2.35)
k=0
where, if Pk = 0, define Pk ln Pk = 0. It can be proved that the value of the spec-

tral entropy converges to ln(N/2). For comparative analysis, the spectral entropy SE is
normalized as
se
SE(N) = . (2.36)
ln(N/2)
Obviously, when the distribution of the sequence power spectrum is not uniform, the
spectrum structure of the sequence is simpler and the signal has a distinct oscillation
pattern. The SE measure is smaller in this case, and it indicates that the complexity is
smaller. Otherwise, the complexity is larger.
The SE calculation flow chart is shown in Figure 2.7. First, the average value of the
sequence is subtracted; that is, the DC part of the signal is removed. Next do FFT. The
energy distribution in the transformed domain is calculated by the Shannon entropy,
and the SE complexity of the sequence is obtained by the normalization.
When we calculate the SE complexity, we can do the Fourier transformation by
using the MATLAB function Y = fft(X). The calculation speed of this function is very
fast. The obtained spectrum is symmetric, so we only choose the first N/2 points to
calculate.
Start
Data from x N = x N – x
FFT for X(k)
N / 2–1
se = – Σ Pk In Pk
k=0
k=0
Normalization
N / 2–1
Pk = | X(k) | 2 ∑ | X(k) |2
k=0
End
k=k+1
Y
k ≤ N / 2–1
N
Figure 2.7: Flow chart of SE algorithm.
2.9.2 C0 Complexity Algorithm
The main idea of C0 complexity is to decompose the sequence into the regular and
the irregular components, and the measure value is the proportion of the irregular
components in the sequence. The algorithm calculation steps are as follows [75]:
Step 1: Discrete Fourier transformation. Do discrete Fourier transformation for time

series {xN (n), n = 0, 1, 2, . . . , N – 1}
N–1
N–1

20
X(k) = x(n)e–j N nk = x(n)WNnk , (2.37)
n=0 n=0
where k = 0, 1, . . . , N – 1.
Step 2: Remove irregular parts. The mean square value of {X(k), k = 0, 1, 2, . . . , N – 1}
is defined by
N–1
1
GN = |X(k)|2 . (2.38)
N
k=0
To introduce a parameter r, and retain the spectrum that is more than the mean square
value of r times, let the rest of the part be zero; that is

X(k) if |X(k)|2 > rGN
X̃(k) = . (2.39)
0 if |X(k)|2 ≤ rGN
Step 3: Fourier inverse transformation. Do inverse Fourier transformation for the

sequence X̃(k)
N–1 N–1
1 20 1
x̃(n) = X̃(k)ej N nk = X̃(k)WN–nk , (2.40)
N N
k=0 k=0
where n = 0, 1, . . . , N – 1.
Step 4: Define C0 complexity. C0 complexity is defined as
N–1
N–1

C0 (r, N) = |x(n) – x̃(n)|2 / |x(n)|2 . (2.41)
n=0 n=0
C0 algorithm can also be implemented in MATLAB, and the flow chart is shown in Fig-
ure 2.8. FFT and Inverse Fast Fourier Transform (IFFT) are, respectively, implemented
by function Y = fft(X) and Y = ifft(X).
In addition, there are some other chaotic dynamical characteristics analysis
methods, such as measure entropy method. It is often combined with qualitative
analysis methods (phase diagram, dividing frequency sampling, or Poincaré section)
and quantitative analysis method (Lypunov exponents, power spectral density, or
0–1 test).
Start
IFFT for the regular
spectrum
Data preparation x N
Remove the regular
X(k) by FFT
Calculation C0 complexity
Remove irregular spectrum
End
Figure 2.8: Flow chart of the C0 complexity algorithm.

Questions
1. What is the relationship between the Lyapunov exponent spectrum and the
capacity dimension?
2. What is the conditional Lyapunov exponent?
3. What are the characteristics and deficiencies of the 0–1 test method for analyzing
chaotic dynamics?
4. What is the complexity of the chaotic system? How to measure it?
3 Typical Chaotic Systems
Chaos exists widely in nature. During the research and development of chaos theory,
researchers found a lot of chaotic dynamics models, and some of them are typical
for the chaos theory and application research, including discrete-time chaotic maps,
continuous-time chaotic systems, and hyperchaotic systems. In this chapter, we will
discuss the mathematical model and its main characteristics of some typical chaotic
systems.
3.1 Discrete-Time Chaotic Map

First, the definition of discrete-time chaotic map is presented. Consider a nonlinear
system
x(k + 1) = Gu(k) ○ F(x(k)), x(0) = x0 , (3.1)
where “○” represents a compound operation between two functions. x(k) ∈ M ⊂ R,

u(k) ∈ RP , {Gu }u∈RP is the diffeomorphism group with p parameters on M. F: M → M is a
diffeomorphism. If x(k) is chaotic, and x0 ∈ M, then we call x(k), which all pass through
x0 on smooth n-dimensional manifold M, is discrete chaos. The physical meaning
is that chaos is generated by a discrete map described by a nonlinear difference
equation, which can usually be achieved by a software program or sampler.
3.1.1 Logistic Map
Logistic map is the most widely used discrete chaotic map model, and its equa-
tion is [4]
xn+1 = ,xn (1 – xn ), n = 1, 2, 3, . . . , (3.2)
where the parameter , ∈ (0, 4] and xn ∈ (0, 1). When 3.5699 . . . < , ≤ 4, it is chaotic.
The dynamical state of the system changes with the system parameter, which includes
the following aspects:
1. When , ∈ [0, 1], the system has a fixed point xn → 0.
2. When , ∈ [1, 3], the system has fixed points xn → 0 and xn → 1 – 1/,. This is called
period 1 solution.
3. When , ∈ [3, ,′ ), ,′ = 3.569945672 . . ., where orbital period changes such as 1 →
2 → 4 → 8 . . . occur. This is called period-doubling bifurcation phenomenon.
4. When , ∈ [,′ , 4), the system shows a chaotic phenomenon.
42 3 Typical Chaotic Systems
0.9
0.8
0.7
0.6
0.5
xn
0.4
0.3
0.2
0.1
0
2 2.2 2.4 2.6 2.8 3 3.2 3.4 3.6 3.8 4
μ
Figure 3.1: Bifurcation diagram of Logistic map.
1
Lyapunov exponent
–1
–2
–3
2 2.5 3 3.5 4
μ
Figure 3.2: Lyapunov exponent of Logistic map.
The bifurcation diagram of the system is shown in Figure 3.1. The iterative initial value
of x0 is 0.6, and the iteration number N is 2,000. The step size of , is 0.01. As shown
in Figure 3.1, when time series experiences four stages of evolution, from stable fixed
point, unstable fixed points, period to chaos, complexity of the system increases. If , =
4.0, the complexity is largest, which is shown in Figure 3.2 by the Lyapunov exponent.
When , = 4.0, Lyapunov exponent is equal to 0.69, which is the maximum.
Next, we will discuss the statistical properties of Logistic map.
The probability density function of Logistic mapping system is

⎧
⎨ √1 , 0<x<1
1(x) = 0 x(1–x) . (3.3)
⎩0, others
The average value is
N
1
x̄ = lim xi = . (3.4)
N→∞ 2
i=1
The autocorrelation function is

N–1

1 0.125, 4=0
Raa (4) = lim an an+4 = . (3.5)
N→∞ N 0, 4≠0
n=0
The cross-correlation function is

N–1
1
Rab (4) = lim an bn+4 = 0. (3.6)
N→∞ N
n=0
The advanced Logistic map, which is also called improved logistic map, is
xn+1 = 1 – 2(xn )2 , xn ∈ (–1, 1). (3.7)
The probability density function is

⎧
⎨ √ 1 , x ∈ (–1, 1)
1(x) = 0 1–x2 . (3.8)
⎩0, others
N

x̄ = lim xi = 0. (3.9)
N→∞
i=1

N–1

1 0.5, 4 = 0
Raa (4) = lim an an+4 = . (3.10)
N→∞ N 0, 4≠0
n=0

N–1
1
Rab (4) = lim an bn+4 = 0. (3.11)
N→∞ N
n=0
3.1.2 Tent Map
Tent map is another commonly used discrete chaotic map. The chaotic sequence gen-
erated by the tent map has been widely applied in the field of chaotic spread spectrum
communication, chaotic encryption system, chaotic optimum algorithm, and so on. Its
equation is

xn /q, 0 < xn ≤ q
xn+1 = , (3.12)
(1 – xn )/(1 – q), q < xn < 1
where the system parameter q ∈ (0, 1) and the variable xn ∈ (0, 1). Tent map and lo-
gistic map are topological conjugates. When q varies as (0, 1), the system is chaotic.
In particular, when q = 0.5, the system is in a short-cycle state, such as (0.2, 0.4, 0.8,
0.4). In this case, the system is simple and its complexity is small. Therefore, for the
application of tent map system, we should choose reasonable system parameter and
initial value. Tent map system has the following statistical properties.
N
1
x̄ = lim xi = . (3.13)
N→∞ 2
i=1

N–1

1 1/12, 4=0
Raa (4) = lim an an+4 = . (3.14)
N→∞ N 0, 4≠0
n=0

N–1
1
Rab (4) = lim an bn+4 = 0. (3.15)
N→∞ N
n=0
3.1.3 Hénon Map
Hénon map is a two-dimensional discrete chaotic system, which was proposed by

Hénon in 1976. The map equation is [76]

xn+1 = 1 – axn2 + yn
. (3.16)
yn+1 = bxn
When 1.07 ≤ a ≤ 1.4 and b = 0.3, it is chaotic. The variable xn ∈ (–1.5, 1.5).
When a = 1.4, the complexity of the system is largest. Generally, in the study of
the chaotic synchronization and encryption, we choose a = 1.4 and b = 0.3. The
0.4
0.2
0
yn
–0.2
–0.4
–2 –1 0 1 2
xn
Figure 3.3: Strange attractor of Hénon map.
0.5
0.4
0.3
0.2
0.1
yn
–0.1
–0.2
–0.3
–0.4
0 0.5 1 1.5
a
Figure 3.4: Bifurcation diagram of Hénon map.
strange attractor of Hénon map system is shown in Figure 3.3, which looks like the
galaxy in the sky, where the initial value xn (0) = yn (0) = 0.5, and the iteration
times are 1,000. When b = 0.3, and a ∈ [0, 1.5], the bifurcation diagram of Hénon
map system is shown in Figure 3.4. Obviously, it enters into chaos by period-doubling
bifurcation.
3.1.4 Tangent Delay-Ellipse Reflecting Cavity System Map
In 2004, Sheng et al. proposed a discrete chaotic system based on tangent delay (TD),
which is called as TD-ellipse reflecting cavity system (TD-ERCS) map [77]. The map
equation is
⎧
⎪
⎪ 2kn–1 yn–1 + xn–1 (,2 – kn–1
2 )
⎪
⎪ xn = –
⎪
⎪ , – kn–1
2 2
⎪
⎪
⎪
⎪ ′ ′2
⎨k = 2kn – kn–1 + kn–1 kn–1 , n = 1, 2, 3, . . . ,
⎪
n ′ ′2
1 + 2kn–1 kn – kn (3.17)
⎪
⎪
⎪
⎪ ′ x n–m
⎪
⎪ kn–m =– ,2 (m ≤ n)
⎪
⎪ yn–m
⎪
⎪
⎪
⎩y = k (x – x ) + y
n n–1 n n–1 n–1
where the system parameter , ∈ (0, 1], and the variables |xn | ≤ 1 and |yn | ≤ 1. m is
an integer, which represents the number of TD. kn–m′ is the ellipse tangent slope after
delaying m. k0 is determined by incident angle !, and the method is as follows.
First, we calculate y0 and the slope of k0′ through the initial value x0 and system
parameter ,

y0 = , 1 – x02 , (3.18)
x0
k0′ = – ,2 . (3.19)
y0
Then according to trigonometric function, we can obtain:
tan ! + k0′
k0 = . (3.20)
1 – k0′ tan !
So, if we know the system parameters , and m, and the initial values x0 and !, we can
get the chaotic sequence {xn , kn }. When the TD m= 0, the system becomes ERCS map.
When m ≥ 1, the TD-ERCS is chaotic.
With the change in system parameter ,, the largest Lyapunov exponent of the
system is shown in Figure 3.5. Because the curve chart looks like a flying goose, it is
called “Goose.” The iteration number n is 400,000. Obviously, when , = 1, the max-
imum Lyapunov exponent LE = 0.92. When , = 0.5, the maximum Lyapunov exponent
LE = 0.33. Thus, the maximum Lyapunov exponent is greater than zero in the range of
,, which shows that TD-ERCS is chaotic in the whole range.
When m = 1, there exists a square attractor for TD-ERCS as shown in Figure 3.6,
where we choose , = 0.8123, x0 = 0.7654, ! = 0.9876, and the iteration number N is
1,500. At the beginning, the phase points are not on the attractor, but with the increase
of iteration, phase points will fall on the attractor. The experiment also shows that if
0 < , < 1, the position of square attractor is unchangeable and has no relation with
0.8
0.6
LE
0.4
0.2
0
0 0.5 1
μ
Figure 3.5: Maximum Lyapunov exponent of TD-ERCS system (x0 = 0.7654, ! = 0.9876).
0.5
Start point
arctan(K)/π
–0.5
0 0.5 1
arccos(X)/π
Figure 3.6: Square attractor of TD-ERCS with m = 1.
the initial value. When , = 1, the position of attractor is related to the initial values,
but all phase points are on the attractor.
3.2 Continuous-Time Chaotic System

First, we present the definition of continuous-time chaotic system. Consider the
nonlinear system
ẋ = F(x, ,, t), x(t0 ) = x0 , (3.21)
where x(t) ∈ M ⊂ Rn ; it is an integral curve through x0 in the vector field of F. x0 ∈ M

and , ⊂ RP are parameter vectors. If x(t) is chaotic, all x(t) going through x0 on the
smooth n-dimensional manifold M are continuous chaos. From physical meaning, a
continuous chaotic system is a nonlinear dynamical system and can produce chaotic
phenomenon, which is described by the differential equation ẏ = F(x) and usually can
be realized by analog circuit.
There are two types of continuous chaotic systems. One is autonomous sys-
tem, such as the Lorenz system, Rössler system, and Chua system. The other is
nonautonomous system, such as Duffing oscillator and van der Pol oscillator.
3.2.1 Duffing Oscillator
Duffing oscillator is a typical forced vibration nonlinear dynamic system, and it

belongs to nonlinear vibration problem. The Duffing system equation is [78]
ẍ + kẋ – f (x) = e(t), (3.22)
where k is damping coefficient, f (x) is a nonlinear function, and e(t) is the drive term
with period T. When f (x) and e(t) take different forms, we will get Duffing equation
with different forms. Here, consider the following Duffing equation:
ẍ + kẋ – x + x3 = F cos 9t, (3.23)
where k is the dissipative coefficient, and F and 9 are amplitude and frequency of
the periodic force, respectively. It is an important type that can generate chaos in
nonlinear vibration.
Let ẋ = y, then eq. (3.23) is translated into

ẋ = y
. (3.24)
ẏ = –ky + x – x3 + F cos 9t
When k = 0, F = 1.5, and 9 = 1, the attractor of the system is shown in Figure 3.7.
0
y
–2
–4
–2 0 2
x
Figure 3.7: Chaotic attractor of Duffing system in x–y plane.

0
y
–5
–2 0 2
x
Figure 3.8: Chaotic attractor of van der Pol oscillator in x–y plane.
3.2.2 van der Pol Oscillator
In 1926, van der Pol proposed a nonlinear nonautonomous system that is driven by
external force, and the equation is [79]

ẋ = y
. (3.25)
ẏ = –x + b(1 – x2 )y + A sin Kt
When parameters b = 3, A = 5, and K = 1.788, the chaotic attractor of the system is

shown in Figure 3.8.
3.2.3 Lorenz System
When American meteorologist Lorenz investigated atmospheric convection model, he

summarized a three-dimensional equation and found “butterfly effect,” which lays a
foundation for chaos research. For Lorenz system, the mathematical model is a three-
dimensional one-order nonlinear differential equation with three parameters, and it
becomes one of the most typical examples to study chaos theory and its applications.
Lorenz system has abundant bifurcation and chaotic phenomenon.
Lorenz system equation is [2]
⎧
⎪
⎨ẋ = 3(y – x)
⎪
ẏ = rx – y – xz , (3.26)
⎪
⎪
⎩ż = xy – bz
where 3, r, and b are the system parameters. The third-order ordinary differential
equation is based on the simplified model of heat fluid convection movement between
the infinite flats. Its variables do not contain time, so it is called autonomous equation.
x represents convection strength; y represents the temperature difference between
50
40
30
z
20
10
0
–20 –10 0 10 20
x
Figure 3.9: Chaotic attractor of Lorenz system in x–z plane.
upflow and downflow between the units; z represents the nonlinear intensity in the
vertical temperature distribution; r represents the ratio of driving factors and inhibited
factor that causes convection and turbulent, which are the main control parameters of
the system. 3 = v/k is Prandtl value (where v and k are molecular viscosity coeffi-
cient and heat transfer coefficient, respectively); b represents the ratio of shape that is
related to convection in aspect ratio; and 3 and b are dimensionless constant values.
Numerical results show that when 3 = 10, b = 8/3, and 28 ≤ r < ∞, except some
values near b (such as b = 100, 160, 260, and so on), the Lorenz system is in a chaotic
state. Let 3 = 10, r = 28, and b = 8/3 and take numerical simulation for the Lorenz
system, we get the famous attractor – “butterfly,” as shown in Figure 3.9. The butterfly
attractor has two fixed points, and the orbit line transfers randomly around the two
fixed points and forms the spiral shape, just like two wings of a butterfly.
3.2.4 Rössler System
Rössler equation looks like the Lorenz system equation, but only a nonlinear term,
which is also one of the typical systems in chaotic application research.
Rössler system model is described as [80]
⎧
⎪
⎨ẋ = –y – z
⎪
ẏ = x + ay , (3.27)
⎪
⎪
⎩ż = b + z(x – c)
where a, b, and c are system parameters. When a = b = 0.2, c = 5.7, the system is in a
chaotic state. The Lyapunov exponent spectra are (0.0714, 0, –5.3943). The movement
of the attractor describes the complexity of the system. The attractor of Rössler system
is shown in Figure 3.10.
10
0
y
–5
–10
–15
–10 0 10
x
Figure 3.10: Chaotic attractor of Rössler system in x–y plane.
3.2.5 Chua’s Circuit
In 1983, Professor Chua designed the famous nonlinear chaotic circuit at the first
time, which is called as Chua’s circuit. Not only it can generate the rich and com-
plex dynamic behaviors, but also the structure of the circuit is relatively simple.
It has become a classical chaotic system for application of chaos control and
chaos secure communication. It can produce period-doubling bifurcation, single-
scroll, period 3, double-scroll attractor, and other complex nonlinear dynamic be-
haviors by adjusting the circuit parameters. On the basis of Chua’s circuit, the
study of multi-scroll Chua’s chaotic attractors has become a research hotspot at
present.
The normalized state equation of Chua’s circuit is [15]
⎧
⎪
⎨ẋ = ![y – h(x)]
⎪
ẏ = x – y + z , (3.28)
⎪
⎪
⎩ż = –"y
h(x)
xeq1 xeq3
m0 m1 m2 m3 m4
c2 c3 c4
x
c1
xeq0 xeq2 xeq4
Figure 3.11: Schematic diagram of piecewise linear function h(x).

0.5 4
(a) (b)
0 0
z
y
–2
–0.5 –4
–2 0 2 –2 0 2
x x
Figure 3.12: Chaotic attractors of the double scroll: (a) x–y plane and (b) x–z plane.
where h(x) is a piecewise linear function, and its waveform is shown in Figure 3.11.
When h(x) is a different piecewise linear function, the system produces double-scroll
and multi-scroll chaotic attractors. For chaos secure communication, the more the
scroll is, the more secure the chaotic secure communication system is. Here only
the double-scroll, three-scroll, five-scroll, and seven-scroll chaotic systems are sim-
ulated.
3.2.5.1 Double-Scroll Chaotic Attractor

For double-scroll attractor, h(x) and other parameters are defined by
⎧
1
2q–1
⎪
⎪
⎪
⎪ h(x) = m 2q–1 x + (mi–1 – mi ) ⋅ (|x + ci | – |x – ci |)
⎪
⎪
2
i=1
⎨
q = 1, ! = 9, " = 100/7 . (3.29)
⎪
⎪
⎪m = [m0 ; m1 ] = [–1/7; 2/7]
⎪
⎪
⎪
⎩
c = [c1 ] = [1]
By Matlab simulation, we obtain the chaotic attractor as shown in Figure 3.12.
3.2.5.2 Three-Scroll Chaotic Attractor

For three-scroll chaotic attractor, h(x) and other parameters are defined by
⎧
1
2q–1
⎪
⎪
⎪
⎪ h(x) = m 2q–1 x + (mi–1 – mi ) ⋅ (|x + ci | – |x – ci |)
⎪
⎪
2
i=1
⎨
q = 2, ! = 9, " = 100/7 (3.30a)
⎪
⎪
⎪
⎪ m = [m0 , m1 , m2 , m3 ] = [0.9/7, –3/7, 3.5/7, –2.4/7]
⎪
⎪
⎩
c = [c1 , c2 , c3 ] = [1, 2.15, 4]

m = [m0 , m1 , m2 , m3 ] = [–1/7, 2/7, –4/7, 2/7]
Or (3.30b)
c = [1, 2.15, 3.6]
1 1
(a) (b)
0.5 0.5
h(x)
h(x)
0 0
–0.5 –0.5
–1 –1
–5 0 5 –5 0 5
x x
1 2
(c) (d)
0.5 1
0 0
y
–0.5 –1
–1 –2
–5 0 5 –10 –5 0 5 10
x x
Figure 3.13: Nonlinear curve and the three-scroll chaotic attractor (a), (b) nonlinear curve; (c),
(d) The three-scroll chaotic attractor in x-y plane.
Through Matlab simulation, the nonlinear function and three-scroll chaotic attractor
are shown in Figure 3.13.
3.2.5.3 Five-Scroll Chaotic Attractor

For five-scroll chaotic attractor, h(x) and other parameters are given by
⎧
⎪
⎪
2q–1
⎪
⎪ h (x) = m2q–1 x + 21 (mi–1 – mi ) ⋅ (|x + ci | – |x – ci |)
⎪
⎪ i=1
⎪
⎪
⎪
⎨q = 3, ! = 9, " = 100/7
(3.31a)
⎪
⎪ m = [m0 , m1 , m2 , m3 , m4 , m5 ]
⎪
⎪

⎪
⎪ = 0.9/7, –3/7, 3.5/7, –2.7/7, 4/7, –2.4/7
⎪
⎪
⎪
⎩
c = [c1 , c2 , c3 , c4 , c5 ] = [1, 2.15, 3.6, 6.2, 9]

m = [m0 , m1 , m2 , m3 , m4 , m5 ] = –1/7, 2/7, –4/7, 2/7, –4/7, 2/7
Or . (3.31b)
c = [c1 , c2 , c3 , c4 , c5 ] = [1, 2.15, 3.6, 6.2, 9]
By Matlab simulation, the five-scroll chaotic attractor is shown in Figure 3.14

5
(a) (c)
1
0 0
y
y
–1
–5
–10 –5 0 5 10 –20 0 20
x x
Figure 3.14: Five-scroll chaotic attractor: (a) Parameter 1 (x–y plane) and (b) parameter 2 (x–y plane).
0
y
–1
–2
–20 –10 0 10 20
x
Figure 3.15: Seven-scroll chaotic attractor (x–y plane).
3.2.5.4 Seven-Scroll Chaotic Attractor

For seven-scroll chaotic attractor, h(x) and other parameters are described as
⎧
1
2q–1
⎪
⎪
⎪
⎪ h ( x ) = m2q–1 x + (mi–1 – mi ) ⋅ (|x + ci | – |x – ci |)
⎪
⎪
2
i=1
⎪
⎪
⎪
⎨q = 4, ! = 9, " = 100/7
. (3.32)
⎪
⎪m = [m0 , m1 , m2 , m3 , m4 , m5 ]
⎪
⎪

⎪
⎪ = 0.9/7, –3 7, 3.5/7,–2.4/7, 2.52/7, –1.68 7, 2.52/7, –1.68/7
⎪
⎪
⎪
⎩
c = [c1 , c2 , c3 , c4 , c5 ] = [1, 2.15, 3.6, 6.2, 9, 14, 25]
Through Matlab simulation, the seven-scroll chaotic attractor is shown in Figure 3.15.
3.2.6 Chen System
In 1999, in the research of anti-control of chaos, Professor Chen discovered a new

chaotic attractor. From the phase diagram of the system, compared with the “butter-
fly” attractor, it is more complicated on the phase space.
The mathematical model of Chen attractor is [16]
⎧
⎪
⎨ẋ = a (y – x)
⎪
ẏ = (c – a) x – xz + cy , (3.33)
⎪
⎪
⎩ż = xy – bz
where X = (x, y, z)T ∈ R3 is the state of the system; a > 0, b > 0, and c > 0 are the
system parameters. Although Chen system is similar to the other two famous chaotic
systems – Lorenz system and Rössler system in the form, Professor Chen proved that
it isn’t topological equivalence with the two chaotic systems. The attractor of this sys-
tem is different from any previous chaotic attractors, and qualitatively analysis the
features of the attractor, such as bifurcation, the route to chaos. If a = 35, b = 3, and
c = 28, the strange attractor of Chen chaotic system is shown in Figure 3.16.
Hardware circuit of Chen system is designed and tested in Ref. [81], and the cir-
cuit diagram and experimental results are shown in Figures 3.17 and 3.18, respectively.
Obviously, the experimental results are consistent with the simulation results.
3.2.7 Lü System
In 2002, Lü and Chen discovered a new chaotic attractor, which connects the famous
Lorenz attractor and Chen attractor. The mathematical model of Lü system is [82]
50
40
30
z
20
10
0
–20 0 20
x
Figure 3.16: Phase diagram of Chen’s chaotic attractor in x–z plane.

R4 C1
R7
R1
y – R5
– R6
x + –
R2 + x –x
+
R3
R12 R14
x C2
Y
R9
z X R13
– R15
–
–x + –
R10 + y
+
y
R8
R11
R19 C3
x Y R16
X R20
y –
–
+ z
R17 +
R18
Figure 3.17: Circuit diagram of Chen system.
Figure 3.18: Chaotic attractor of Chen system from oscilloscope.
⎧
⎪
⎨ẋ = a(y – x)
⎪
ẏ = –xz + cy , (3.34)
⎪
⎪
⎩ż = xy – bz
where a, b, and c are system parameters. When a = 36, b = 3, and c = 20, the system
is in a chaotic state. The chaotic attractor of the system is shown in Figure 3.19.
40
30
20
z
10
0
–20 –10 0 10 20
x
Figure 3.19: Attractor of Lü system in x–z plane.
3.2.8 Unified Chaotic System
In 2002, Lü and Chen proposed a new chaotic system, and the system connects
the Lorenz attractor and Chen attractor. Lü system is just a special one of the new
chaotic system, so it is called the unified chaotic system. According to the definition
of Vanĕček and Čelikovský [83], Lorenz system, Chen system, and Lü system belong to
different topology types. Mathematical model of the unified chaotic system is [84]
⎧
⎪
⎨ẋ = (25! + 10)(y – x)
⎪
ẏ = (28 – 35!)x – xz + (29! – 1)y , (3.35)
⎪
⎪
⎩ż = xy – (8 + !)z/3
where the system parameter ! varies in the interval [0, 1]. At this range, the unified
chaotic system has full chaotic characteristics. According to the definition of Vanĕček
and Čelikovský [83], when ! varies in the interval [0, 0.8), the unified chaotic system
belongs to the generalized Lorenz system. When ! varies in the interval (0.8, 1], the
unified chaotic system belongs to the generalized Chen system. When ! = 0.8, the
unified chaotic system belongs to Lü system. So the unified chaotic system plays
an important role in connecting Lorenz system with Chen system, and it is a one-
parameter continuous chaotic system. Namely, using only one parameter !, we can
control the entire system. When ! gradually increases from 0 to 1, the system also
gradually transits from the generalized Lorenz system to the generalized Chen system
[85]. Under the different parameters of unified chaotic system, the chaotic attractors
in the x–z plane are shown in Figure 3.20. The unified chaotic system parameter !
has the largest Lyapunov exponent in the range of [0, 1] as shown in Figure 3.21, ex-
cept the three obvious periodic windows, W1 = [0.369, 0.371], W2 = [0.468, 0.470],
W3 = [0.575, 0.597]. The largest Lyapunov exponent is greater than zero in the range
50 50
(a) (b)
40 40
30 30
z
z
20 20
10 10
0 0
–20 0 20 –20 0 20
x x
50 50
(c) (d)
40 40
30 30
z
z
20 20
10 10
0 0
–20 0 20 –20 0 20
x x
Figure 3.20: Attractors of unified chaotic system with different parameters in x–z plane. (a) ! = 0;
(b) ! = 0.5; (c) ! = 0.8; and (d) ! = 1.
2.5
2.0
Maximum LE
1.5
1.0
0.5
0
0 0.2 0.4 0.6 0.8 1
α
Figure 3.21: Maximum Lyapunov exponent of unified system with different !.
of [0, 1], so the unified chaotic system has full chaos characteristics. Due to its excel-
lent performances, the control and synchronization of the system become a hot topic
for the chaotic research [86], and it has extensive application prospect in the field of
secure communications [87].
0.5
Maximum LE
–0.5
–2 0 2 4 6 8
c
Figure 3.22: Maximum Lyapunov exponent of simplified Lorenz system with different c.
3.2.9 Simplified Lorenz System
Dynamic equation of simplified Lorenz system is [56]
⎧
⎪
⎨ẋ = 10(y – x)
⎪
ẏ = –xz + (24 – 4c)x + cy , (3.36)
⎪
⎪
⎩ż = xy – 8z/3
where c is a bifurcation parameter of the system. The maximum Lyapunov exponent of

simplified Lorenz system is shown in Figure 3.22. When c ∈ [–1.59, 7.75], the system is in
a chaotic state obviously. But there exists nine period windows in this range at least,
namely, W1 = [3.507, 3.509], W2 = [4.581, 6.612], W3 = [4.6911, 4.722], W4 = [5.122, 5.127],
W5 = [5.167, 5.169], W6 = [5.599, 5.600], W7 = [5.7601, 5.770], W8 = [5.820, 5.830], and
W9 = [6.415, 6.425].
With the change of the system parameter c, the bifurcation diagram of the sim-
plified Lorenz system is shown in Figure 3.23. The chaotic attractor of the simplified
Lorenz system is shown in Figure 3.24.
Simplified Lorenz system has the following properties:
1. It is chaotic over most of the range c ∈ [–1.59, 7.75].
2. For c = –1, it is the usual Lorenz system with the standard parameters.
3. For c = 0, the variable y is removed from the second equation.
4. For c = 6, the variable x is removed from the second equation.
5. There is a rich set of bifurcation as c is varied over the range.
6. According to the topological definition by Vanĕček and Čelikovský [83], the lin-
earization of system (3.36) about the origin produces a 3 × 3 constant matrix of
partial derivatives, A = [aij ]3×3 , in which the sign of a12 a21 distinguishes nonequi-
valent topologies. According to this criterion, when c < 6, a12 a21 > 0; when c = 6,
a12 a21 = 0; and when c > 6, a12 a21 < 0.
20
18
16
14
12
χmax
10
8
6
4
2
0
–2 –1 0 1 2 3 4 5 6 7 8
c
Figure 3.23: Bifurcation diagram of the simplified Lorenz system.
20
0
x
–20
20 40
0
20
y
–20 0 z
Figure 3.24: Chaotic attractor of the simplified Lorenz system (c = 2).
Simplified Lorenz system can be implemented by analog circuit. The nonlinear

part uses operational amplifier LM741 and multiplier AD633 to implement. Accord-
ing to the simulation, the variable amplitude exceeds the voltage range that op-
erational amplifier and multiplier provide, so the system state variables cannot
be directly as the voltage variables. To realize the circuit systems, the state vari-
ables must be scaled for transformation appropriately. Simplified Lorenz system
circuit is shown in Figure 3.25 [88], and c is the system parameter of simplified
Lorenz system; thus, the change of R14 is equal to the change of system para-
meter c value. Figure 3.26 is an attractor phase diagram captured by an oscillo-
scope.
R3 C1
R1
– R5
–
R2 + + V1
R4
Y R6 R7 R8
X R16 C2
– R9
– R17
+ –
+ V2
R15 +
R13
R14
R11
R10
– R12
–
+ +
R20 C3
Y
R18 V3
X – R22
–
+
R19 +
R21
Figure 3.25: Circuit diagram of the simplified Lorenz system.
Figure 3.26: Attractor of simplified Lorenz system in x–z plane from oscilloscope.
3.2.10 Standard for a New Chaotic System
At present, many mathematical models of low-dimensional discrete time and continu-

ous time have been put forward, and their solutions are chaotic, which are proved
by a positive Lyapunov exponent. Researchers are continuously finding new chaotic
systems and publishing them in a pile of journals about nonlinear dynamics. These
papers just reported some characteristics about nonperiodic orbit in the state space
and bifurcation diagram, Lyapunov exponents, and other dynamics and topological
properties. Obviously, these papers almost are not wrong, but they usually only
provide another example that is known and understood fully to all of us. So they can’t
be called as a new chaotic system.
For this phenomenon, Professor Sprott thought if you want to say the proposed
system is a new chaotic system, the system must satisfy at least one of the following
criteria [89]:
1. The system should credibly model some important unsolved problems in nature
and shed insight on that problem.
2. The system should exhibit some behaviors previously unobserved.
3. The system should be simpler than all other known examples exhibiting the
observed behaviors.
The above standards provide a sufficient condition for the publication of new chaotic
system. The famous Lorenz system meets the three standards when first published
since it deals with atmospheric turbulence, exhibited sensitive dependence on initial
conditions, and is the simplest system in its time known to have these proper-
ties. Lorenz devoted considerable effort to simplifying the system from the original
seven-dimensional system with 13 quadratic nonlinearities that was suggested by
Saltzman [90].
3.3 Hyperchaotic System

For the chaotic movement of high-dimensional partial differential equations, in many
cases, there is more than one direction of instability, which is called as hyperchaotic
system. Hyperchaotic system has two or more positive Lyapunov characteristic index,
which has more complex dynamical behaviors than general chaotic system.
3.3.1 Rössler Hyperchaotic System
Rössler hyperchaotic system is a simple four-dimensional oscillator model that was

found by Rössler in 1979. It can generate hyperchaotic attractor in more than two
directions of hyperbolic unstable. Rössler equation is [91]
3.3 Hyperchaotic System 63
⎧
⎪
⎪ẋ1 = –x2 – x3
⎪
⎪
⎪
⎨ẋ = x + ax + x
2 1 2 4
. (3.37)
⎪ẋ3 = b + x1 x3
⎪
⎪
⎪
⎪
⎩ẋ = –cx + dx
4 3 4
When the system parameters a = 0.25, b = 3.0, c = 0.5, and d = 0.05, the system is in
a chaotic state, and the Lyapunov exponent spectrum is (0.112, 0.019, 0, –25.188). Its
attractor is shown in Figure 3.27.
3.3.2 Chen Hyperchaotic System
In 2005, Li et al. designed the hyperchaotic Chen system by employing state feedback
control, and the equation is [92]
⎧
⎪
⎪ ẋ = a(y – x) + w
⎪
⎪
⎪
⎨ẏ = dx – xz + cy
, (3.38)
⎪ż = xy – bz
⎪
⎪
⎪
⎪
⎩ẇ = yz – rw
where x, y, z, and w are state variables of the system, and a, b, c, d, and r are control
parameters of the system. When a = 35, b = 3, c = 12, and d = 7, r changes in the inter-
val [0, 0.1085], (0.1085, 0.1798], (0.1798, 0.1900], respectively, then the system shows
chaotic motion, hyperchaotic motion, and periodic motion, respectively. When a = 35,
b = 3, c = 12, d = 7, and r = 0.16, the hyperchaotic system has two positive Lyapunov
exponents, 0.1567 and 0.1126. The attractor of the system is shown in Figure 3.28.
60
50 (a) (b)
50
0 40
x2
x4
30
–50 20
10
–100 –50 0 –50 0 50
x1
x2
Figure 3.27: Attractors of Rössler hyperchaotic system: (a) x1 –x2 plane and (b) x2 –x4 plane.
40
(a) (b)
100
30
50
w
20
z
10
–50
0 –20 0 20
–20 0 20
x y
Figure 3.28: Attractors of Chen chaotic system: (a) x–z plane and (b) y–w plane.
3.3.3 Folded Towel Hyperchaotic Map
Folded towel map is presented by Rössler in 1979, which is described as [91]

⎧
⎪
⎨xn+1 = axn (1 – xn ) – 0.05(yn – 0.35)(1 – 2zn )
⎪
yn+1 = 0.1((yn + 0.35)(1 + 2zn ) – 1)(1 – 1.9xn ) , (3.39)
⎪
⎪
⎩z = 3.78z (1 – z ) + by
n+1 n n n
where a and b are control parameters, and x, y, z are system variables. If a = 3.8,
b = 0.2, the system is hyperchaotic, and the Lyapunov exponents are (0.418, 0.373,
–2.502) [93]. The phase diagram of the folded towel map function for a = 3.8, b = 0.2,
x0 = 0.1, y0 = 0.2, and z0 = 0.3 is shown in Figure 3.29.
The chaotic system models are much more than those models discussed earlier.
In this chapter, we mainly introduce typical chaotic systems, and they are common
models in chaotic theory and applied research. These models not only have the actual
physical meaning, but also have unique characteristics. Researchers often use them
1
0.8
0.6
z
0.4
0.2
0
0.05
1
0 0.8
y 0.6
0.4
–0.05 0.2 x
0
Figure 3.29: Chaotic attractor of folded towel map.

Questions 65
to research chaotic theory and application. With the study of chaos, researchers will
continuously find and reveal chaotic system model and its dynamics in nature.
Questions
1. Try to build a new chaos model.
2. What kind of characteristics a new chaotic system should have?
3. If xn+1 = xn2 + c and –2 < c < 0.3, calculate its bifurcation diagram, and
approximately prove Feigenbaum constant.
4. What is hyperchaotic system? What are its features?
4 Chaotic Synchronization Principle and Method
Due to the extremely sensitive initial value, researchers once thought it is impossible
to control chaos. Until 1990, American scientists, Ott, Grebogi, and Yorke, proposed
the OGY method to realize chaos control [12]. In the same year, Pecora and Car-
roll proposed a self-synchronizing method for chaos [10], and the synchronization
between two chaotic systems is realized by using the drive–response method for
the first time [94]. The breakthrough research has broken the traditional concept
that chaos movement is not controllable and dangerous. It is proved that chaos
can not only be controlled and synchronized, but also it is the basis of information
transmission and processing. So it can be used in the field of information security
communication, and thus it opened the prelude to the application of chaotic secure
communication.
Chaos synchronization, in general, belongs to chaos control. The problem of
chaos synchronization can be considered as a kind of control problem that the chaotic
orbit of the controlled system evolves as the orbit of the target system. The traditional
chaos control is generally the stable system on the unstable periodic orbits, while the
synchronization of the chaotic system is the complete reconstruction of the chaotic
state of the two systems.
In this chapter, based on the definition of chaos synchronization and the per-
formance index of chaotic synchronization system, the principle and performance of
various chaotic synchronization control methods are studied.
4.1 Definition of Chaos Synchronization
Definition 4.1 ([95]): Consider two nonlinear systems as follows:
ẋ = F(t, x), (4.1)

ẏ = F ′ (t, y) + U(t, x, y), (4.2)
where state variables x, y ∈ Rn , and F, F ′ : [R+ × Rn ] → Rn is the nonlinear map.

U : [R+ × Rn × Rn ] → Rn is the synchronization control variable of chaotic system.
R+ is the nonnegative real number set. If we have D(t0 ) ⊆ Rn , ∀x0 , y0 ∈ D(t0 ), and

limx(t; t0 , x0 ) – y(t; t0 , y0 ) = 0 when t → ∞, then systems (4.2) and (4.1) are syn-
chronous. We call system (4.1) as the drive system, and system (4.2) as the response
system, and D(t0 ) is the synchronization region. If D(t0 ) belongs to the whole space Rn ,
then the synchronization is defined as global (complete) synchronization. If D(t0 ) is a
subset of Rn , then the synchronization is defined as local (partial) synchronization. If
F = F ′ , then it is called self-synchronization.
4.2 Performance Index of Chaotic Synchronization System 67
If we set error vector e = x–y, then the synchronization error system of

systems (4.1) and (4.2) is
ė = F(t, x) – F ′ (t, y) – U(t, x, y) = G(t, e, x). (4.3)
Similarly, based on the stability of error system (4.3) at the origin of the system, the
stability conditions of the system are derived.
In the study of chaos synchronization, the stability of the synchronization system
is determined mainly by the Lyapunov stability principle or by the condition Lyapunov
exponent (CLE) criterion proposed by Pecora and Carroll. The former is the main
criterion, and the latter is a necessary condition for synchronization. But the linear
stability theorem shows that the necessary condition is also a sufficient condition for
most of the dynamic systems, and it is used as a criterion during the synchronization
research [96].
The above synchronization definition is suitable for synchronization of not only
chaotic systems, but also nonchaotic systems. It is suitable for both of the synchron-
ization of autonomous systems and nonautonomous systems, and it can be extended
to the synchronization of hyperchaotic systems and spatiotemporal chaotic systems.
4.2 Performance Index of Chaotic Synchronization System

The study of the performance of the synchronization system is directly related to the
practical application of the chaotic system, and it is of great significance to study the
mechanism of the synchronization performance. Control parameters have a great in-
fluence on the performance of the synchronization system, and how to choose the
control parameters to make the optimal synchronization performance is an import-
ant part of the study of chaos synchronization. The synchronization performance of
the system can be improved by choosing the initial value and controlling paramet-
ers of the system, such as shortening the setup time of synchronization, improving
the stability and convergence speed, and increasing the anti-disturbance ability for
the synchronization system. Here, we put forward the performance index of chaotic
synchronization system first, including synchronization stability, synchronization
robustness, synchronization steady-state error, synchronization accuracy, synchron-
ization setup time, synchronization region, synchronization dynamic characteristics,
and so on [97].
Definition 4.2: Synchronization stability refers to that the chaos synchronization sys-
tem with different initial states satisfies the Lyapunov asymptotic stability conditions,
namely the solution of system equation is convergent. It is the primary character-
istic of the chaotic synchronization system, and it is the necessary condition for the
synchronization system to work. It is the key to realize the application of the secure
68 4 Chaotic Synchronization Principle and Method
communication. Stability is divided into large-scale (global) asymptotic stability and

small-scale (local) asymptotic stability.
Definition 4.3: Synchronization robustness refers to the drive–response system that

maintains stability in the case of the system parameters mismatch. From the practical
view, if the system’s robustness is better, then the system is easier to implement. But
from the view of communication security, if the system’s robustness is better, then the
security is worse. This is a contradiction, which must be considered in the practical
application.
Definition 4.4: Synchronization steady-state error is the accurate degree of synchron-

ization of the two chaotic systems to achieve synchronization. In general, it is denoted
by B>.
Definition 4.5: Synchronization accuracy refers to the consistent degree of the chaotic
signal generated by the controlled system and the target system in the amplitude and
phase, and it is denoted by $. The relationship between the steady-state error and the
synchronization accuracy is $ = 10[lg B>] , where [B>] is the smallest integer that is not
more than B>.
Definition 4.6: Synchronization setup time is an important index for the application
of chaotic secure communication, and it is the transition time experienced by the
response signal for the given synchronization accuracy, and it is denoted by Ts . To
describe the synchronization of chaotic systems, the synchronization accuracy can be
calculated by
n
1
2
2
p= (yi – xi ) (4.4)
i=1
or
p = max{|yi – xi |, i = 1, 2, . . . , n} (4.5)
where n is the number of system variables. When p ≤ $, synchronization is achieved.
Definition 4.7: Synchronization region is the initial value range for the synchron-
ization of the chaotic system. He and Vaidya proved the solution of the system is
asymptotically stable when the initial value of chaotic system is in a certain region
(synchronization region) [98]. That is, the response system will eventually be syn-
chronized with the drive system. Obviously, a successful or practical synchronization
method should have sufficiently large synchronous region.
Table 4.1: Comparison of the maximum CLE with the synchronization setup time.
Chaotic system Drive variable Maximum CLE Theory value [99] Setup time (s)
Lorenz system 1 x –1.81 –1.86 2.8

(3 = 10, b = 8/3, r = 28) y –2.67 –9.99 1.8
Lorenz system 2 x –2.50 –2.50 2.0
(3 = 16, b = 4, r = 45.92) y –3.95 –16.00 1.4
Definition 4.8: Synchronization dynamic characteristics are also known as the

transient characteristics of the synchronization establishment. It shows as monotonic
or nonmonotonic decay. Nonmonotonic decay shows that there is a trend of nonsyn-
chronization in the synchronization process, which should be avoided.
Research shows that the control parameters are closely related to the performance in-
dex of the synchronization. Reasonable choice of control parameters can make the
CLE of the response system from positive to negative, so that the response system
can be controlled. That all CLEs of the response system are negative is a sufficient
condition for the realization of chaos synchronization control, and it is a necessary
condition in most cases. The smaller the CLE is, the faster the error convergence is,
and the shorter the synchronization setup time is. The maximum CLE in Table 4.1 is
calculated by employing the algorithm presented in Chapter 2. When the maximum
CLE of the response system is negative, and it is significantly less than zero, then it
can guarantee a large enough synchronization region. So changing the control para-
meters means changing the system’s CLE, and it can improve the performance of the
synchronous system.
4.3 Principle and Performance of Feedback Control

Synchronization
4.3.1 Multivariable drive feedback synchronization for chaotic system
Consider two continuous chaotic systems with the same structure and different initial
values as follows:

Ẋ = AX + Bf (X),
Drive system: (4.6)
X(0) ∈ Rn ,

Ẏ = AY + Bf (Y) + U(X, Y),
Response system: (4.7)
Y(0) ∈ Rn ,
where drive variables X ∈ Rn and Y ∈ Rn , and matrices A ∈ Rn×m and B ∈ Rn×m ; F:

Rn × R → Rm is the nonlinear function. AX is the linear part, and Bf (X) is the nonlinear
part. Control variable U ∈ Rm , and (A, B) is controllable. If the state variable error
E = X–Y, then the synchronization problem between response system (4.7) and drive
system (4.6) is used to find a suitable control law U(X, Y) to satisfy
lim X – Y = lim E = 0. (4.8)

t→∞ t→∞
Obviously, the synchronization control problem of two continuous chaotic systems is

transformed into the control problem of the error system.
According to Lyapunov stability theorem, we design a positive definite quadratic
function V(E) = ET PE as a Lyapunov function. If V̇(E) = –ET QE, and Q is positive
definite (that is, V̇(E) is negative definite), then the system is asymptotically stable.
The synchronous error system equation for eqs (4.6) and (4.7) is
Ė = AE + B(f (X) – f (Y)) – U(X, Y). (4.9)
So the problem of synchronization of two systems is transformed into the problem of

the control law and the choice of the control parameters. In addition, how to select
a set of control parameters from a certain range of parameters must be based on the
principle of synchronization performance optimization. It is very important whether
it is from the study of the synchronization control itself, or the practical application of
the synchronization system.
4.3.1.1 Control Law of the Feedback Controller

The multivariable drive feedback synchronization principle for continuous chaotic
system is shown in Figure 4.1. The response system is driven by all the variables of
the drive system, and the error feedback controller is composed of a linear feedback
part and a nonlinear feedback part.
−
X + + Y
∫ ∑ K ∑ ∑ ∫
+
+ +
+ +
∑ A ∑ A
+ +
B f( ) B f( )
+ −
∑
Drive system Controller Response system
Figure 4.1: Synchronization principle block diagram for continuous chaotic system.
Theorem 4.1: For the n-dimensional continuous chaotic system and its replication
system, the synchronous control method based on multivariable drive error feed-
back control is employed. The drive system is Ẋ = AX + Bf (X), and the response
system is Ẏ = AY + Bf (Y) + U, where X, Y ∈ Rn and K ∈ Rn × m . The control variable
U = K(X–Y) + B(f (X)–f (Y)). If the elements in the parameter control matrix K satisfy:
when i = j, then kij > aij ; or when i ≠ j, then kij = aij , and then the asymptotically stable
synchronization is achieved between the response system and the drive system.
Proof. Let E = X–Y, then the error system equation Ė = AE + B(f (X) – f (Y)) – U(X, Y). If
the control law of the controller is U = KE + B(f (X) – f (Y)), then Ė = (A – K)E.
Define V(E) as the Lyapunov function
V(E) = ET PE ≥ 0, (4.10)
where P is a positive definite matrix. Its differential is
V̇(E) = ĖT PE + ET PĖ = [(A – K)E]T PE + ET P[(A – K)E]

= ET [(A – K)T P + P(A – K)]E = –ET QE (4.11)
Obviously, as long as the feedback controller parameter kij meet: when i = j, kij > aij ;
when i ≠ j, kij = aij , then Q = (A – K)T P + P(A – K) is positive definite. That is, V̇(E) is
negative definite. If and only if E = 0, the equality is right. That is, the drive–response
system is asymptotically stable. ∎
Inference 4.1 When i = j, then kij > aij ; When i ≠ j, and aij = aji , then kij = kji = 0, the
asymptotic stability synchronization is realized and the structure of the controller is
simplified, which can be proved by Lyapunov stability theorem (omitted).
Because the response system is driven by all drive system variables, and the synchron-
ization of two systems is achieved by linear and nonlinear feedback controls, it is
called multivariable driving feedback synchronization control method. The method
does not need to decompose the system, which is not only suitable for simple chaotic
systems, such as the Lorenz chaotic system, but also be applied to hyperchaotic
system, such as the Rössler hyperchaotic system. In this method, the selection of feed-
back parameters is presented, and the performance of the synchronization system can
be effectively improved by choosing proper feedback control parameters.
4.3.1.2 Synchronization Simulation and Discussion
Example 1: Multivariable drive feedback synchronization control for Rössler hyper-

chaotic system.
As shown in Section 3.3.1, the Rössler hyperchaotic system equation is

⎧
⎪
⎪ ẋ1 = x2 – x3
⎪
⎪
⎪
⎨ẋ = x + ax + x
2 1 2 4
. (4.12)
⎪
⎪ ẋ = b + x x
⎪
⎪
3 1 3
⎪
⎩ẋ = –dx + cx
4 3 4
When system parameters a = 0.25, b = 3.0, c = 0.05, and d = 0.5, it is hyperchaotic.

According to Theorem 4.1 and Inference 4.1, the drive system is
⎡ ⎤ ⎡ ⎤
0 1 –1 0 0
⎢1 a 0 1⎥ ⎢0⎥
⎢ ⎥ ⎢ ⎥
Ẋ = ⎢ ⎥ X + ⎢ ⎥ (b + x1 x3 ), (4.13)
⎣0 0 0 0⎦ ⎣1⎦
0 0 –d c 0
and the response system is

⎡ ⎤ ⎡ ⎤
0 1 –1 0 0
⎢1 a 0 1⎥ ⎢0⎥
⎢ ⎥ ⎢ ⎥
Ẏ = ⎢ ⎥ Y + ⎢ ⎥ (b + y1 y3 ) + U. (4.14)
⎣0 0 0 0⎦ ⎣1⎦
0 0 –d c 0
The synchronization control law is

⎡ ⎤ ⎡ ⎤
k11 0 –1 0 0
⎢ 0 k22 1 ⎥ ⎢0⎥
⎢ 0 ⎥ ⎢ ⎥
U=⎢ ⎥ E + ⎢ ⎥ (x1 x3 – y1 y3 ). (4.15)
⎣ 0 0 k33 0 ⎦ ⎣1⎦
0 0 –d k44 0
When feedback coefficient k11 > 0, k22 > a, k33 > 0, and k44 > c, then the syn-
chronization is achieved for the drive system and the response system. Set the initial
conditions of the drive system and the response system as (–20, –30, 20, 45) and
(–40, 25, 40, 15), respectively. The control parameter kii = 1, and the simulation time
is 20 s. The synchronization phase portrait between x1 and y1 is shown in Figure 4.2.
The error curves are obtained with the same initial values and the different con-
trol parameters as shown in Figure 4.3, where |E| = [(x1 – y1 )2 + (x2 – y2 )2 + (x3 – y3 )2 +
(x4 – y4 )2 ]1/2 . The corresponding feedback control parameters of curves a, b, and
c are (0.6, 0.6, 0.6, 0.6), (1, 1, 1, 1), and (2, 2, 2, 2). Obviously, in a certain range,
the greater the control parameters, i.e., the greater the control energy, the shorter
the synchronization setup time, and the better the dynamic performance in a cer-
tain convergence error range. The simulation results show that the error feedback
control can realize the synchronization of two hyperchaotic systems. To realize the
synchronization control, the feedback control parameters must be selected correctly.
50
y1 –50
–100
–150
–150 –100 –50 0 50
x1
Figure 4.2: Synchronization phase portrait for Rössler hyperchaotic system.
80
60
40
|E|
20
a
b
c
0
0 5 10 15
t/s
Figure 4.3: Error curves with different control parameters.
The synchronization performance of the hyperchaotic synchronization system can be

improved by choosing appropriate parameters.
4.3.2 Linear and Nonlinear Feedback Synchronization for Discrete Chaotic System
Consider an m-dimensional discrete chaotic system (drive system)
Xn+1 = F(Xn ) = AXn + Bf (Xn ), (4.16)
then the response system is
Yn+1 = F(Yn ) + U(Xn , Yn ) = AYn + Bf (Yn ) + U(Xn , Yn ), (4.17)
and the control law is
U(Xn , Yn ) = K(Xn – Yn ) + P(f (Xn ) – f (Yn )) (4.18)

where Xn , Yn ∈ Rm . f is the nonlinear function and U is the feedback control law. The
feedback parameters K= diag[k1 , k2 , . . . , km ], P = diag[p1 , p2 , . . . , pm ], where ki ≥ 0
and pi ≥ 0 (i = 1, 2, . . . , m). The synchronization of the response system and the
drive system is actually the problem of the stability of the synchronous manifold

S = (Xn , Yn ) : Yn = Xn .
Here the problem is analyzed by using local linearization. Set
Yn = Xn + $n , (4.19)
where $n is a small synchronization error. From eqs (4.16) to (4.19), we obtain the linear
evolution equation of the synchronization error as

∂f
$n+1 = (A – K) + (B – P) |Y=Xn $n = Dn $n . (4.20)
∂Y
Obviously, if Dn is the constant matrix, and all the characteristic roots are within the
unit circle, then $n → 0 with n → ∞, that is, Yn → Xn . So the synchronization manifold
is stable, and the two systems will be synchronized. But if Dn in eq. (4.20) is a time-
varying matrix, then the above method does not work.
Pecora–Carroll Chaos Synchronization Theorem For all the time t, the zero solu-
tion of the nonlinear unstable system ẋ = A(t)x + O(x, t) with O(x, t) will be asymptotic-
ally stable, as long as the system meets the following conditions.

1. For time t, lim O(x, t) / x = 0.
x→0
2. For all the time t, A(t) are all bounded.
3. The zero solution of the nonlinear system ẋ = A(t)x is asymptotically stable.
That is to say, the Pecora–Carroll chaos synchronization theorem can be concluded as

follows. Only when all the Lyapunov exponents of the response system are negative,
we can achieve the chaos synchronization between the response system and drive sys-
tem. This theorem is a sufficient condition for synchronization, but the linear stability
theorem proves that the sufficient condition is also a necessary condition for many
dynamical systems, so it is used as a criterion in the study of chaos synchronization.
Define the maximum conditional Lyapunov exponent of response system (4.17) as

1 $n
+ = lim ln . (4.21)
n→∞ n $0
Obviously, + is an average statistic. It reflects the global characteristics of the system

and determines the final behavior of the system. When + > 0, the error $n will increase
by exponential law. When + < 0, the error $n will decrease by exponential law and
leads to $n = 0 finally. That is,
lim |Yn – Xn | = 0. (4.22)

n→∞
It means that the synchronization is achieved between the response system and drive
system. So, we can determine whether the synchronization is realized and obtain the
range of the feedback parameters by calculating the maximum CLE of the response
system.
The control law U(Xn , Yn ) consists of a linear feedback and nonlinear feed-
back, so it is called the linear and nonlinear feedback synchronization method. This
method has many advantages, such as simple structure, optional parameters, and fast
synchronous speed.
4.3.2.1 Control Law of the Synchronization Controller

The synchronization principle of discrete chaotic system is shown in Figure 4.4.
According to the stability criterion, the parameter selection theorem for the
synchronization of two discrete chaotic systems is given.
Theorem 4.2: For an m-dimensional discrete-time chaotic system Xn + 1 = AXn + Bf (Xn ),

its response system is Yn+1 = AYn + Bf (Yn ) + U(Xn , Yn ). By employing the linear and
nonlinear feedback control synchronization method, the control law is U(Xn ,
Yn ) = K(Xn – Yn ) + P(f (Xn ) – f (Yn )), where Xn , Yn ∈ Rm , K = diag[k1 , k2 , . . . , km ]. When
P = B, the necessary and sufficient condition for the synchronization between the re-
sponse system and the drive system is that all poles of the characteristic polynomial
T(z) = |Iz – (A – K)| are located in the unit circle.
Proof. For the drive and response systems, the error system equation is

∂f
$n+1 = (A – K) + (B – P) |Y=Xn $n . (4.23)
∂Y
X Y U
Drive system Σ Response system
− +
f( ) e f( ) +
K Σ
− +
+
Σ P
Feedback controller
Figure 4.4: Synchronization principle block diagram of discrete chaotic system.

If P = B, then we have
$n+1 = (A – K)$n . (4.24)
Obviously, the coefficient of the characteristic polynomial of the error system is a func-
tion of the linear feedback coefficient K, which is independent of the initial value. If
T(z) = |Iz – (A – K)| is a Hurwitz polynomial, then all the eigenvalues of the matrix A–K
are located in the unit circle. $n → 0 with n → ∞, that is, Yn → Xn . Synchronization
between the drive system and the response system is achieved. ∎
Theorem 4.3: For a one-dimensional discrete chaotic system xn+1 = axn + bf (xn ), its
response system is yn+1 = ayn +bf (yn )+u(xn , yn ). By employing the linear and nonlinear
feedback synchronization control method, the control law is u(xn , yn ) = k(xn – yn ) +
p(f (xn ) – f (yn )), where xn , yn ∈ R. When k ≥ 0, and p ≥ b, then the sufficient condition for
the asymptotic stability synchronization between the response system and the drive
system is |a–k| < 1.
Proof. For a one-dimensional discrete chaotic system, the error system equation is

∂f
$n+1 = (a – k) + (b – p) |y=xn $n . (4.25)
∂y
So we have

n ∂f
$n = F (a – k) + (b – p) |y=f (i) $0 , (4.26)
i=1 ∂y (x0 )
where f(x(i) ) represents the ith iteration from x0 . According to the definition in eq. (4.21),
0
the maximum CLE is

1 $n 1 n ∂f
+ = lim ln = lim ln F (a – k) + (b – p) |y=f (i) . (4.27)
n→∞ n $0 n→∞ n i=1 ∂y (x0 )
If p = b, we have

1 $n 1
+ = lim ln = lim ln [(a – k)]n = ln |a – k| . (4.28)
n→∞ n $0 n→∞ n
According to eq. (4.28), as long as k meets |a – k| < 1, then we have + < 0. ∎
Theorem 4.4: For an m-dimensional discrete chaotic system Xn+1 = AXn + Bf (Xn ), its
response system is Yn+1 = AYn + Bf (Yn ) + U(Xn , Yn ). By employing the linear and non-
linear feedback control synchronization method, the control law of the controller is
U(Xn , Yn ) = K(Xn – Yn ) + P(f (Xn ) – f (Yn )), where Xn , Yn ∈ Rm . When K = dA, and P = dB,
d > 0, then the sufficient condition for the asymptotic stability between the response
′ ′
system and the drive system is 1–e–+ < d < 1+e–+ , where +′ is the maximum Lyapunov
exponent of the drive system. Obviously, +′ > 0. (Proof omitted.)
For one-dimensional discrete chaotic system, we have the following theorem.
Theorem 4.5: For a one-dimensional discrete chaotic system xn+1 = axn + bf (xn ), its
response system is yn+1 = ayn + bf (yn ) + u(xn , yn ). By employing the linear and non-
linear feedback control synchronization method, the control law of the controller is
u(xn , yn ) = k(xn – yn ) + p(f (xn ) – f (yn )), where xn , yn ∈ R. When k = ma, p = mb, m > 0.
The sufficient conditions for the asymptotic stability of the response system and
′ ′ ′ ′
the drive system are a – ae–+ < k < a + ae–+ , b – be–+ < p < b + be–+ , that is,
′ ′
1 – e–+ < m < 1 + e–+ , where +′ is the maximum Lyapunov exponent of the drive
′
system, and + > 0.
According to the above theorems, the linear and nonlinear feedback control can be
used to realize the synchronization of two discrete chaotic systems by choosing appro-
priate feedback coefficient. Although Theorems 4.4 and 4.5 are sufficient conditions
to guarantee the synchronization of two chaotic systems, it is also a necessary con-
dition for most of the dynamical systems. It is worth noting that the synchronization
speed is very fast. When d = 1 or m = 1, the synchronization can be achieved as D-B
synchronization.
4.3.2.2 Simulation and Discussion
Example 2: Linear and nonlinear feedback synchronization for logistic map.
Logistic map is
xn+1 = ,xn (1 – xn ), (4.29)
where x ∈ (0, 1) and , ∈ (0, 4]. When , = 4, the system is chaotic. The experimental
conditions are as follows. The initial value is 0.2. The step size of parameter p and k is
0.05. The iteration number is 40,000.
The drive system is
xn+1 = F(xn ) = 4xn – 4xn2 . (4.30)
The response system is
yn+1 = F(yn ) = 4yn – 4yn2 + u(xn , yn ). (4.31)
The control law is
u(xn , yn ) = k(xn – yn ) – p(xn2 – yn2 ). (4.32)

2.5
λ 1.5
0.5
0 5 10
k (p = 0)
Figure 4.5: CLE versus feedback coefficient k.
1. When p = 0, the linear feedback control method is adopted. The maximum Lya-
punov exponent + of the response system varies with the linear feedback coeffi-
cient k as shown in Figure 4.5. Obviously, the value of + is always greater than
zero, so the two discrete systems cannot be synchronized.
2. When k = 0, the nonlinear feedback control strategy is adopted. The maximum
Lyapunov exponent + of the response system is shown in Figure 4.6 with the linear
feedback coefficient p. The value of + cannot meet the requirements of less than
zero, so the synchronization between two discrete systems cannot be achieved.
3. When k ≠ 0, p ≠ 0, it means using linear feedback and nonlinear feedback strategy
at the same time. According to Theorem 4.3, when p = 4, and 3 < k < 5, then the CLE
+ < 0, and the two systems can realize synchronization. The maximum Lyapunov
exponent of the response system varies with the linear feedback coefficient k as
2.5
1.5
λ
1
0.5
0
0 5 10
p (k = 0)
Figure 4.6: CLE versus feedback coefficient p.

0
λ
−1
−2
−3
0 2 4 6 8
k (p = 4)
Figure 4.7: CLE versus feedback coefficient k.
′ ′
4. When k = ma, p = mb, according to Theorem 4.5, and 4 – 4e–+ < k = p < 4 + 4e–+ ,
the two systems can realize synchronization. Due to +′ = 0.6931 for logistic map,
2 < k = p < 6. The maximum CLE of the response system varies with the linear
feedback coefficient as shown in Figure 4.8.
5. The synchronization performance of the system is related to the selection of the
control parameters. Experimental results show that the more negative the max-
imum CLE is by choosing the control parameters, the faster the error converges,
and the shorter the synchronization setup time is. The error convergence with dif-
ferent feedback coefficients is shown in Figure 4.9. When the initial conditions of
the drive and response systems are 0.1 and 0.8, respectively, and k = p = 3.2, then
+ = –1. The synchronization time is 4 s as shown in Figure 4.9(a). When k = p = 4,
then + = –3. The synchronization time is 1 s as shown in Figure 4.9(b). So the
synchronous speed is fast, and the D-B synchronization is realized.
λ –1
–2
–3
0 2 4 6 8
k=p
Figure 4.8: CLE versus feedback coefficient.

0.8 0.8
(a) (b)
0.6 0.6
0.4
|e|
|e|
0.4
0.2 0.2
0 0
0 2 4 6 8 0 2 4 6 8
n/s n/s
Figure 4.9: Synchronization error curves with different feedback coefficients: (a) k = p = 3.2 and
(b) k = p = 4.
Example 3: Linear and nonlinear feedback synchronization for Hénon map.
Hénon map is

xn+1 = 1 – axn2 + yn
. (4.33)
yn+1 = bxn
When a = 1.4 and b = 0.3, the system is chaotic. Next, we apply Theorem 4.2 to
Hénon map.
The drive system is

0 1 1
Xn+1 = Xn + (1 – 1.4xn2 ). (4.34)
0.3 0 0
The response system is

0 1 1
Yn+1 = Yn + (1 – 1.4yn2 ) + U(Xn , Yn ). (4.35)
0.3 0 0
The control law is

k1 0 1.4 2 2
U(Xn , Yn ) = (Xn – Yn ) + (y1n – x1n ). (4.36)
0 k2 0
Then the error system is

–k1 1
$n+1 = $n = D$n . (4.37)
0.3 –k2
Obviously, the characteristic equation of matrix D is

z + k –1
1
T(z) = |Iz – D| = = z2 + (k1 + k2 )z + k1 k2 – 0.3 . (4.38)
–0.3 z + k2
According to Jury stability criteria, the necessary and sufficient condition of all the
characteristic roots of T(z) to locate in the unit circle is
⎧ ⎧
⎪ ⎪
⎨T(1) > 0
⎪ ⎨k1 k2 + k1 + k2 > –0.7
⎪
T(–1) > 0 , that is, k1 k2 – (k1 + k2 ) > –0.7 . (4.39)
⎪
⎪ ⎪
⎪
⎩1 > |k k – 0.3| ⎩–0.7 < k k < 1.3
1 2 1 2
According to eq. (4.39), when k1 = 0 and |k2 | < 0.7, or when k2 = 0 and |k1 | < 0.7, or
when k1 = k2 = k and |k| < 0.4522, the error system is asymptotically stable. That is
to say, the response system can be synchronized with the drive system. When we set
the initial values as [0.1 1.0]T and [0 0.1]T , respectively, then the control parameters are
k1 = 0, k2 = 0.2, and P = B. The error convergence curve of the two systems is shown in
Figure 4.10(a). It shows that there exist nonsynchronization trends in the synchroniza-
tion process. When the steady-state error is B> = 10–4 , the synchronization setup time
is about 22 s. The synchronization performance of the synchronization system can be
improved by choosing the appropriate parameters. For example, when k1 = k2 = 0
and P = B, the error convergence curve is shown in Figure 4.10(b). It can be seen that
the synchronous speed is accelerated, and the synchronization time is shortened. The
simulation synchronization time is about 17 s under the same steady-state error, and
it realizes the monotonous synchronization.
By applying Theorem 4.4 to Hénon map and setting the simulation conditions, the
initial value is [0.1, 0.5]T , the step size of parameter is 0.05, and the iteration number
is 40,000.
1 1
(a) (b)
0.8 0.8
0.6 0.6
|e|
|e|
0.4 0.4
0.2 0.2
0 0
0 5 10 15 20 25 0 5 10 15 20 25
n /s n /s
Figure 4.10: Synchronization error curves with different feedback parameters: (a) k1 = 0, k2 = 0.2
and (b) k1 = k2 = 0.
0.8 0.8
(a) (b)
0.7 0.6
CLE
CLE
0.6 0.4
0.5 0.2
0.4 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
k p
Figure 4.11: Maximum CLE curves with linear or nonlinear feedback: (a) p = 0 and (b) k = 0.
From drive system (4.34) and response system (4.35), the control law is chosen as
U(Xn , Yn ) = kA(Xn – Yn ) + pB(f (Xn ) – f (Yn )) (4.40)
1. When p = 0, only the linear control method is employed. The maximum condi-
tional Lyapunov exponent varies with the linear feedback coefficient k as shown
in Figure 4.11(a). Obviously, regardless of the value of k, + is always greater than
zero, and the synchronization between the two discrete chaotic systems cannot
be realized.
2. When k = 0, only the nonlinear control method is employed. The maximum
conditional Lyapunov exponent varies with the nonlinear feedback coeffi-
cient p as shown in Figure 4.11(b). Regardless of the value of p, + cannot
meet the requirements of less than zero, so the two systems cannot achieve
synchronization.
3. When p ≠ 0 and k ≠ 0, that both the linear feedback and nonlinear feedback
strategies are used at the same time. The conditional Lyapunov exponent + is
greater than zero by rational selection of control parameters. The system can real-
ize synchronization. According to Theorem 4.4, when p = 1, the maximum CLE
varies with the linear feedback coefficient k as shown in Figure 4.12(a). It can be
seen that when 0 < k < 2, then + < 0.
4. When k = dA, p = dB, according to Theorem 4.4, if we want + < 0, then 1 –
′ ′
e–+ < d < 1 + e–+ should be satisfied. As the maximum Lyapunov exponent
of Hénon system is +′ = 0.7258, we have 0.52 < d < 1.48. The maximum CLE of
the response system varies with the feedback control parameters as shown in
Figure 4.12(b).
5. The synchronization performance of the system is related to the control
parameters. Experimental results show that the more negative the maximum
CLE is, the faster the synchronization error converges, and the shorter the
synchronization setup time is. When d = 1, then the CLE is the smallest (in-
finitesimal). At this time, only one iteration step is needed to realize the
chaos synchronization, as shown in Figure 4.13. It is equivalent to the D-B
synchronization.
Both the stability criterion of discrete system and the Pecora–Carroll asymptotic syn-
chronization theorem can synchronize two discrete chaotic systems, but they are dif-
ferent. The selection of the control parameters obtained by the stability criterion of the
discrete system is a necessary and sufficient condition for the synchronization of the
systems. While the selection of the control parameters according to the Pecora–Carroll
chaos synchronization theorem is a sufficient condition for the synchronization of the
systems. That is, it is possible to choose the control parameters to synchronize the
drive system and the response system, but the maximum CLE is not negative. For ex-
ample, for Hénon map, when K = diag[k, k] and P = B, the maximum CLE is greater
0 1
0
–1
CLE
CLE
–1
–2
–2
(a) (b)
–3 –3
0 0.5 1 1.5 2 0 0.5 1 1.5 2
k d
Figure 4.12: Maximum CLE curves with linear or nonlinear feedback: (a) p = 1, k ≠ 0 and
(b) k = dA, p = dB.
0.8
0.6
|e|
0.4
0.2
0
0 5 10
n/s
Figure 4.13: Synchronization error curve with k = A, p = B.

1.5
λ 1
0.5
0
0 1 2 3 4
k
Figure 4.14: CLE versus k.
than zero as shown in Figure 4.14. But according to Theorem 4.2, when |k| < 0.4522,
the response system and drive system can be synchronized.
4.4 Parameter Adaptive Synchronization Based on Pecora–Carroll

Synchronization Criterion
4.4.1 Principle for Adaptive Synchronization
Consider an autonomous system
ẋ = f (x, !), (4.41)
where the state variable x ∈ Rn , the system parameter ! ∈ Rm , and nonlinear function
f : Rn×m → Rn .
The given system that generates the desired orbit is
ẏ = f (y, !), (4.42)
where y ∈ Rn , ! ∈ Rm . The variable y starts from the initial and evolves with the
known parameter !∗ , and one obit O(y) is obtained, which is called as a desired or-
bit or reference orbit. System (4.42) that produces the orbit is called a target system
or a reference system. The adaptive control synchronization method is to introduce a
control mechanism for the parameter ! in system (4.41), so that the orbit O(x) of system
(4.41) from any initial value follows the orbit O(y) of the target system. System (4.41) is
called a controlled system or response system. If lim y – x = 0, then the controlled
system and the target system are synchronized.
4.4 Parameter Adaptive Synchronization Based on Pecora–Carroll Synchronization Criterion 85
The parameter !j in system (4.41) can be obtained by introducing a certain amount

of perturbation to modify the evolution of its orbit, which can make the O(x) and O(y)
realize the synchronization. For parameter !j , the control law is

dfi

!˙ j = –% h (xi – yi ), sgn – $g(!j – !∗j ), (4.43)
d!j
where i = 1, 2, . . . , L, L ≤ n, j = 1, 2, . . . , k, k ≤ m. !∗j is the parameter value to generate

the desired obit. h(⋅, ⋅) and g(⋅) are control laws, respectively, and %, $ are positive con-
trol constants. When the control law is selected, by changing % and $, the combined
system from systems (4.43) and (4.41) is
⎧
⎨ẋ = f (x, !)
. (4.44)
⎩!˙ j = –% h (xi – yi ), sgn dfi – $g(!j – !∗ )
d!j j
If all the CLEs are less than zero, the synchronization between the two autonomous
chaotic systems is realized.
4.4.2 Choose Control Law and Control Parameters
Formula (4.43) only gives a general parameter control law, and it does not make a
specific description for the choice of h and g. On the other hand, that all the Lyapunov
exponents of the combined system are less than zero is only the sufficient conditions
for the synchronization of chaotic systems. Therefore, it is necessary to make a further
discussion about the control law and control parameter for the chaos synchronization.
In the following simulation, the control law is chosen as

dfi dfi
h (xi – yi ), sgn = (xi – yi )sgn , (4.45)
d!i d!j
g(!j – !∗j ) = (!j – !∗j ). (4.46)
Therefore, the adaptive control law of the controllable parameter is

dfi
!˙ j = –% (xi – yi )sgn – $(!j – !∗j ). (4.47)
d!j
So the change of controlled parameters depends on two factors. One is the dif-
ference (xi –yi ) between the state variable x in eq. (4.41) and the corresponding
variable y of the desired orbit. The second is the difference (!j – !∗j ) between the
values of the controlled parameters and the corresponding parameter values of the
desired orbit.
Here, we take the unified chaotic system (3.35) as an example to illustrate the syn-
chronization principle. To realize the synchronization control of two unified chaotic
Y − + X
Target system ∑
Controlled system
e α
Adaptive control law
Combined system
Figure 4.15: Principle block diagram of the adaptive synchronization.
systems, the adaptive control law for parameter ! is given according to eqs (4.45) and
(4.46), that is,
!˙ = –%[(x1 – y1 )sgn(25(x2 – x1 )) + (x2 – y2 )sgn(–35x1 + 29x2 )

(4.48)
+ (x3 – y3 )sgn(–x3 /3)] – $(! – !∗ ).
According to the Pecora–Carroll synchronization stability criterion, the sufficient

condition for the synchronization of two unified chaotic systems with different para-
meters is determined by the fact that the maximum CLE of the combined system is less
than zero. The principle block diagram is shown in Figure 4.15.
4.4.3 Determine the Boundary and Range of the Control Constant
According to the analysis above, once adaptive control law (4.48) of the parameter
is determined, the size of % and $ becomes the key to achieve synchronization.
The ranges can be determined by two methods, such as numerical calculation and
theoretical analysis.
The numerical calculation method is used mainly to determine the range of % and
$ by calculating the Lyapunov exponent of the composite system which is composed
of the unified chaotic system and formula (4.48). The boundary of the constants % and
$ is determined by the fact that the maximum CLE of the composite system is equal to
zero.
The theoretical analysis method is described as follows. Assuming the reference
system is at equilibrium, set y = y∗ as the equilibrium point of the reference system.
!∗ is the parameter corresponding to y∗ . So !∗ and y∗ satisfy f (y∗ , !∗ ). Obviously, the
most basic requirement for the synchronization control is that the composite system is
stable in the neighborhood of y∗ . That is, all the eigenvalues of the Jacobian matrix of
the composite system have negative real part, or the characteristic polynomial of Jac-
obian matrix is Hurwitz polynomial. So we can obtain the range of control constant %
and $. Here the Jacobian matrix of the composite system is
4.4 Parameter Adaptive Synchronization Based on Pecora–Carroll Synchronization Criterion 87
⎡ ⎤
–a a 0 b
⎢ 28 – 35! – x c ⎥
⎢ 3 29! – 1 –x1 ⎥
J=⎢ ⎥, (4.49)
⎣ x2 x1 d e ⎦
–% sgn(b) –% sgn(c) –% sgn(e) –$
where a = 25! + 10, b = 25(x2 – x1 ), c = –35x1 + 29x2 , d = –(8 + !)/3, and e = –x3 /3. Obvi-
ously, the Jacobian matrix contains not only variables but also the symbol function, so
the characteristic polynomial is very complex, and it is difficult to determine the range
of % and $. So here the numerical calculation method of the maximum CLE presented
in the second chapter is used to determine the range of % and $.
4.4.4 Simulation Results and Discussion
When ! ∈ [0, 1], the reference system is chaotic. The initial values of the controlled
system (response system) are (10, 10, 10) and !(0) = 0.5. The initial values of the refer-
ence system (drive system) are (–5, –5, –5) and !∗ = 0.3. Let |e| = [(x1 – y1 )2 + (x2 – y2 )2 +
(x3 – y3 )2 ]1/2 , and we have the following conclusions.
1. Average evolution time 4̄ from initial deviation to synchronization is related with
the constant %, $ and the expected parameters !∗ , that is, 4̄ = 4̄(%, $, !∗ ).
2. When % is determined, there is a range for constant $ < $min or $ > $max , and
synchronization control of two chaotic systems is not achieved. When $ is determ-
ined, there is a range for constant % < %min or % > %max , and synchronization control
of two chaotic systems is not achieved either, which as shown in Table 4.2.
3. When % is determined, the larger $ is, the better the synchronization performance
is, as shown in Figure 4.16.
4. When % increases, $min and $max become larger, and the range of $ increases.
Similarly, when $ increases, %min and %max become larger, and the range of %
also increases, but the increased magnitude is small as shown in Table 4.3. It
shows that the control effect of the control constant % is better than that of control
constant $.
5. When $ is determined, the larger % is, the better the synchronization performance
is, as shown in Figure 4.17.
Table 4.2: Values of % and $ with different desired parameters for synchronization.
!∗ % $ Maximum CLE
50 161–297 –1.7894
0
39–95 230 –1.5513
50 145–328 –2.0530
0.3
35–126 230 –4.3644
50 170–267 –4.1972
0.6
14–49 230 –2.2715
25 195–340 –1.3862
0.9
18–33 230 –1.5510
30 30
(a) (b)
20 20
|e|
|e|
10 10
0 0
0 1 2 3 4 0 1 2 3 4
t/s t/s
Figure 4.16: Error curves with % = 40 and varying $: (a) $ = 140 and (b) $ = 200.
Table 4.3: Change relation between % and $ with !∗ = 0.9.
% $ △$ B % △%
10 80–110 30 200 16–26 10

20 155–255 100 210 17–27 10
30 228–380 150 220 17–28 11
40 300–530 230 230 18–30 12
50 380–700 320 240 19–31 12
30 30
(a) (b)
20 20
|e|
|e|
10 10
0 0
0 1 2 3 4 0 1 2 3 4
t/s t/s
Figure 4.17: Error curves with $ = 220 and varying %: (a) % = 35 and (b) % = 50.
4.5 Adaptive Synchronization Control Based on Lyapunov

Stable Theory
4.5.1 Adaptive Synchronization Control with Certain Parameter
To study synchronization between two unified chaotic systems with different ini-
tial values and different certain parameters, the parameters of the drive system
and response system are set to be !1 and !2 , respectively, and the drive system
equation is
⎧
⎪
⎨ẋ1 = (25!1 + 10)(y1 – x1 )
⎪
ẏ1 = (28 – 35!1 )x1 – x1 z1 + (29!1 – 1)y1 . (4.50)
⎪
⎪
⎩ż = x y – (8 + ! )z /3
1 1 1 1 1
The response system equation is

⎧
⎪
⎨ẋ2 = (25!2 + 10)(y2 – x2 )
⎪
ẏ2 = (28 – 35!2 )x2 – x2 z2 + (29!2 – 1)y2 . (4.51)
⎪
⎪
⎩ż = x y – (8 + ! )z /3
2 2 2 2 2
To realize the synchronization, the control functions u1 (t), u2 (t), and u3 (t) are added
to the right-hand side of each equation, and the response system becomes
⎧
⎪
⎨ẋ2 = (25!2 + 10)(y2 – x2 ) + u1 (t)
⎪
ẏ2 = (28 – 35!2 )x2 – x2 z2 + (29!2 – 1)y2 + u2 (t) . (4.52)
⎪
⎪
⎩ż = x y – (8 + ! )z /3 + u (t)
2 2 2 2 2 3
The error system is obtained by subtracting eq. (4.52) from eq. (4.50)
⎧
⎪
⎪ė1 = 25!1 (e2 – e1 ) + 25(!2 – !1 )(y2 – x2 ) + 10(e2 – e1 ) + u1 (t)
⎪
⎪
⎪
⎨ė = (28 – 35! )e + 29y + (29! – 1)e – 35(! – ! )x
2 1 1 2 1 2 2 1 2
, (4.53)
⎪
⎪ –x2 z2 + x1 z1 + u2 (t)
⎪
⎪
⎪
⎩ė = x y – x y – (8 + ! )e /3 – (! – ! )z /3 + u (t)
3 2 2 1 1 1 3 2 1 2 3
where synchronization error e1 = x2 – x1 , e2 = y2 – y1 , and e3 = z2 – z1 . Obviously, the

problem of synchronization between two chaotic systems with different parameter is
transformed into the stability problem of error system (4.53) at the origin e1 = 0, e2 = 0,
and e3 = 0. If the appropriate control variables u = [u1 (t), u2 (t), u3 (t)]T is chosen to
make error system (4.53) stable, then response system (4.51) can be synchronized with
drive system (4.50).
Theorem 4.6: For drive system (4.50) and response system (4.51), if we choose the
control function
⎧
⎪ ˆ 2 – e1 ) – 10e2 – 25(!2 – !1 )(y2 – x2 )
⎪u1 (t) = –25!(e
⎪
⎪
⎪
⎨u (t) = (35!ˆ – 28)e + x z – x z + 35(! – ! )x
2 1 2 2 1 1 2 1 2
, (4.54)
⎪
⎪ ˆ 2
–29(!2 – !1 )y2 – 29!e
⎪
⎪
⎪
⎩u (t) = –x y + x y + (!ˆ + 5)e /3 + (! – ! )z /3
3 2 2 1 1 3 2 1 2
where the parameter !ˆ is an estimate of the parameter !1 , and the adaptive law of the
estimate parameter is !˙ˆ = –10e1 e2 –25e21 +29e22 –e23 /3++(!1 – !).
ˆ If the control parameter
satisfy + ≥ 0, then the equilibrium point e = 0 and !1 = !ˆ exist in the error system,
and the asymptotic synchronization is realized between the response system and drive

system. That is, lim e(t) = 0 for any initial values.
t→∞
Proof. Substitute control function (4.54) into error system (4.53), then we have
⎧
⎪ ˆ 2 – e1 ) – 10e1
⎨ė1 = 25(!1 – !)(e
⎪
ˆ 1 + 29(!1 – !)e
ė2 = –35(!1 – !)e ˆ 2 – e2 . (4.55)
⎪
⎪
⎩ė = –(! – !)e
ˆ 3 /3 – e3
3 1
So error dynamics system (4.55) has no relationship with the response system para-
meter !2 . As long as the error system is stable, it can realize the synchronization of two
chaotic systems with different parameters. According to the Lyapunov stable theory,
the Lyapunov function is constructed as
1 1
˜ = eT e + !˜ 2 ,
E(e, !) (4.56)
2 2
where e = [e1 , e2 , e3 ]T , !˜ = !1 – !.
ˆ If the adaptive law of the estimate parameter is
!˙ˆ = –10e1 e2 – 25e21 + 29e22 – e23 /3 + +(!1 – !),

ˆ (4.57)
then the derivative of eq. (4.56) is
Ė = e1 ė1 + e2 ė2 + e3 ė3 – (!1 – !) ˙ˆ

ˆ !. (4.58)
Substituting eq. (4.55) and eq. (4.57) into eq. (4.58), we have
ˆ 2 ].
Ė = –[10e21 + e22 + e23 + +(!1 – !) (4.59)
According to the Lyapunov stable theory, when + ≥ 0, Ė ≤ 0, and equilibrium point

e = 0, !1 = !.ˆ The error system is asymptotically stable, so the two systems are
synchronized. ∎
We use Matlab/Simulink for numerical simulation. Set !1 = 0.8, !2 = 0.3, and

initial values x1 (0) = 10, y1 (0) = 10, z1 (0) = 10, x2 (0) = –5, y2 (0) = –5, z2 (0) = –5 and
ˆ
!(0) = 0.5, and do numerical simulation for control parameter + = –0.1, + = 1, + = 2,
and + = 5. The synchronization error curves between the two chaotic systems with
different control parameters are shown in Figure 4.18.
30 30
(a) (b)
20 20
|e|
|e|
10 10
0 0
0 5 10 0 5 10
t /s t /s
30 30
(c) (d)
20 20
|e|
|e|
10 10
0 0
0 5 10 0 5 10
t /s t /s
Figure 4.18: Synchronization error curve with different control parameters: (a) + = –0.1;
(b) + = 1; (c) + = 2; and (d) + = 5.
The following conclusions are obtained by theoretical analysis and simulations.

(1) The performance of the synchronization system can be improved by increasing
the control parameter, and the monotonous synchronization of two chaotic systems
is realized as shown in Figure 4.18(b)–(d). (2) The selection of the initial value of the
adaptive parameters has no effect on the results. For example, the synchronization
ˆ
error curve with !1 = 0.3, !(0) = 2, and + = 5 is the same with that with !1 = 0.8,
ˆ
!(0) = 0.5, and + = 5 as shown in Figure 4.18(d). (3) There is parameter + that does
not meet the synchronization conditions, but it can make the synchronization of two
systems [99], as shown in Figure 4.18(a), while the performance is poor at this time.
4.5.2 Adaptive Synchronization Control with Uncertain Parameters
Taking the unified chaotic system as an example, the synchronization of two chaotic
systems with unknown parameters is discussed.
When the two system parameters are unknown, the drive and response systems
based on unified chaotic system are
⎧
⎪
⎨ẋ1 = (25! + 10)(y1 – x1 )
⎪
ẏ1 = (28 – 35!)x1 – x1 z1 + (29! – 1)y1 , (4.60)
⎪
⎪
⎩ż = x y – (8 + !)z /3
1 1 1 1
⎧
⎪
⎨ẋ2 = (25! + 10)(y2 – x2 ) + v1 (t)
⎪
ẏ2 = (28 – 35!)x2 – x2 z2 + (29! – 1)y2 + v2 (t) , (4.61)
⎪
⎪
⎩ż = x y – (8 + !)z /3 + v (t)
2 2 2 2 3
where v = [v1 , v2 , v3 ]T is the control vector and ! is the unknown parameter. Set the
error variables as e1 = x2 –x1 , e2 = y2 –y1 , and e3 = z2 –z1 , then the error system of
systems (4.60) and (4.61) is
⎧
⎪
⎨ė1 = (25! + 10)(e2 – e1 ) + v1 (t)
⎪
ė2 = (28 – 35!)e1 + x1 z1 – x2 z2 + (29! – 1)e2 + v2 (t) . (4.62)
⎪
⎪
⎩ė = x y – x y – (8 + !)e /3 + v (t)
3 2 2 1 1 3 3
To make the error system stable and realize the synchronization between the two sys-
tems, the following theorem is proposed for the control function and the estimated
parameters.
Theorem 4.7: For drive system (4.60) and response system (4.61), if the control func-
tion is defined as
⎧
⎪
⎨v1 (t) = –(38 – 10!)e2
⎪
ˆ 2 + e1 z2 + e3 x1 ,
v2 (t) = –29!e (4.63)
⎪
⎪
⎩v (t) = e y + e x
3 1 2 2 1
the parameter !ˆ is the estimate of the parameter !. If the adaptive law of the estimate
parameter is !˙ˆ = 29e22 , then there exist the equilibrium points e = 0 and ! = !ˆ in the
error system, and the asymptotic synchronization is realized between the response

system and drive system. That is, lim e(t) = 0 for any initial values.
t→∞
Proof. Substituting control function (4.63) into error system (4.62), we have
⎧
⎪
⎪ė1 = (35! – 28)e2 – (25! + 10)e1
⎨
ˆ 2 .
ė2 = (28 – 35!)e1 + (29! – 1)e2 – 29!e (4.64)
⎪
⎪
⎩ė = –(! – !)eˆ 3 /3
3 1
As long as the error system is stable, the synchronization of two chaotic systems
with uncertain parameters is realized. According to the Lyapunov stable theory, the
Lyapunov function V is constructed as
1 1
˜ = eT e + !˜ 2 ,
V(e, !) (4.65)
2 2
where !˜ = ! – !,
ˆ and !ˆ is the estimate of uncertain !. If we choose the adaptive law of
the estimate parameter !˙ˆ = 29e22 , then the derivative of eq. (4.65) is
˜ = –(25! + 10)e21 – e22 – (8 + !)e23 /3.

V̇(e, !) (4.66)
Because V is positive, and ! ≥ 0, V̇ is negative. According to the Lyapunov stable

ˆ that is,
theory, there are equilibrium points e = 0 and ! = !,

lim e(t) = 0, ! ∈ [0, 1]. (4.67)
t→∞
The response system and the drive system are synchronized. ∎
We use Matlab/Simulink for numerical simulation. When ! ∈ [0, 1], both drive sys-
tem (4.60) and response system (4.61) are chaotic. Let the initial values x1 (0) = 10,
ˆ
y1 (0) = 10, z1 (0) = 10, x2 (0) = –5, y2 (0) = –5, z2 (0) = –5, and ! ∈ [0, 1], !(0) = 0.5.
The synchronization error curve of two chaotic systems with unknown parameters
is shown in Figure 4.19. The phase diagram of variables x1 and x2 is displayed in
Figure 4.20.
The following conclusions are obtained by theoretical analysis and simulations.
(1) The performance of the synchronization is stable as shown in Figure 4.19. When
steady-state error is B> = 10–4 , the setup time of synchronization system is short (t <
2 s), and the monotonous synchronization of two systems is achieved. (2) The selection
of the initial value of the adaptive parameters has no effect on the results. For example,
the synchronization error curve with !(0) ˆ ˆ
= 2 is the same with that with !(0) = 0.5.
(3) By the Lyapunov stability theory, the synchronization condition is only a sufficient
condition.
30
20
|e|
10
0
0 1 2 3 4
t /s
Figure 4.19: Error curve of chaotic synchronization system.

30
20
10
x2
0
Start point
–10
–20
–20 –10 0 10 20 30
x1
Figure 4.20: Phase portrait of x1 and x2 .
4.6 Intermittent Feedback Synchronization Control

Among the synchronization control methods, the intermittent feedback control
method is used to obtain the desired synchronization at low cost, and it is simple and
easy to realize. In this section, the intermittent feedback synchronization between the
unified chaotic system and its deformation system is investigated.
4.6.1 Intermittent Synchronization Control between Different Systems
Here, we discuss the problem of intermittent synchronization between two different

chaotic systems with topological similarity. For the unified chaotic system
⎧
⎪
⎨ẋ1 = (25! + 10)(x2 – x1 )
⎪
ẋ2 = (28 – 35!)x1 – x1 x3 + (29! – 1)x2 . (4.68)
⎪
⎪
⎩ẋ = x x – (8 + !)x /3
3 1 2 3
When the parameter ! ∈ [0, 1], the system is chaotic.

The deformation system of the unified chaotic system is [100]
⎧
⎪
⎨ẏ1 = (25! + 10)(y2 – y1 )
⎪
ẏ2 = (28 – 35!)y1 – y1 y3 + (29! – 1)y2 . (4.69)
⎪
⎪
⎩ẏ = y2 – (8 + !)y /3
3 1 3
Correspondingly, when ! = 0.8, systems (4.68) and (4.69) are Lü system and deform-
ation Lü system, respectively. The phase portraits with the initial value [–1, 1, 0] are
shown in Figure 4.21. Obviously, the phase track curves of Lü system and deformation
Lü system are similar, but the evolution of the deformation system in the phase space
is more divergent and the system is more complex.
(a) (b)
80 80
60 60
40 40
x3
y3
20 20
0 0
20 20
0 20 0 20
0 0
x2 –20 –20 x1 y2 –20 –20 y1
Figure 4.21: Strange attractors with ! = 0.8: (a) Lü system and (b) deformation Lü system.
To realize the synchronization between chaotic system (4.68) and deformation sys-
tem (4.69), the synchronization control signal ui (t) is added to eq. (4.69), and then eq.
(4.69) becomes
⎧
⎪
⎨ẏ1 = (25! + 10)(y2 – y1 ) + u1 (t)
⎪
ẏ2 = (28 – 35!)y1 – y1 y3 + (29! – 1)y2 + u2 (t) . (4.70)
⎪
⎪
⎩ẏ = y2 – (8 + !)y /3 + u (t)
3 1 3 3
Set ei = xi –yi , then the error system equation is

⎧
⎪
⎨ė1 = (25! + 10)(e2 – e1 ) – u1 (t)
⎪
ė2 = (28 – 35! – x3 )e1 + (29! – 1)e2 – y1 e3 – u2 (t) . (4.71)
⎪
⎪
⎩ė = (x x – x2 ) + 2x e – e2 – (8 + !)e /3 – u (t)
3 1 2 1 1 1 1 3 3
Set the control variables to

⎧
⎪
⎨u1 (t) = 0
⎪
u2 (t) = –x3 e1 p(t) + ke2 p(t) – y1 e3 p(t) , (4.72)
⎪
⎪
⎩u (t) = (x x – x2 ) + 2x e p(t) + ke p(t)
3 1 2 1 1 1 3
where p(t) is a pulse signal with period T and the ratio D of pulse length and pulse
period, and it is defined by

1 iT ≤ t ≤ iT + TD
p(t) = . (4.73)
0 iT + TD < t < (i + 1)T
It can be seen that eq. (4.72) is a composite controller composed of nonlinear continu-
ous feedback and linear intermittent feedback. So, eq. (4.71) becomes
⎧
⎪
⎨ė1 = (25! + 10)(e2 – e1 )
⎪
ė2 = (28 – 35! – x3 + x3 p(t))e1 + (29! – 1 – kp(t))e2 – y1 (1 – p(t))e3 . (4.74)
⎪
⎪
⎩ė = 2x e (1 – p(t)) – e2 – (8 + !)e /3 – ke p(t)
3 1 1 1 3 3
If it is asymptotically stable at the origin, the synchronization between systems (4.68)

and (4.70) is achieved.
Theorem 4.8: For systems (4.68) and (4.69), if it is controlled by eq. (4.72), and the
feedback parameter satisfies k > 27–6!, and D ≥ 0.4, then the asymptotically stable
synchronization will be realized between systems (4.70) and (4.68).
Proof. Decompose the error system eq. (4.74), and when iT < t < iT + TD, we have
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
ė1 e1 0
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎣ ė2 ⎦ = A1 ⎣ e2 ⎦ + ⎣ 0 ⎦, (4.75)
ė3 e3 –e12
where A1 is the Jacobian matrix of the error system when the system has a control
pulse
⎡ ⎤
–25! – 10 25! + 10 0
⎢ ⎥
A1 = ⎣ 28 – 35! 29! – 1 – k 0 ⎦. (4.76)
0 0 –(8 + !)/3 – k
When iT + TD < t < (i + 1)T, we have

⎡ ⎤ ⎡ ⎤ ⎡ ⎤
ė1 e1 0
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎣ ė2 ⎦ = A2 ⎣ e2 ⎦ + ⎣ 0 ⎦, (4.77)
ė3 e3 –e21
where A2 is the Jacobian matrix of the error system when the system has no control
pulse
⎡ ⎤
–25! – 10 25! + 10 0
⎢ ⎥
A2 = ⎣ 28 – 35! – x3 29! – 1 –y1 ⎦ (4.78)
2x1 0 (8 + !)/3
and –e2 is the nonlinear term of the error system. For all t, when e → 0,
2 1
–e / e → 0, so the evolution of errors depends on A1 and A2 . During iT < t <
1
iT + TD, the error is controlled by A1 . For ! ∈ [0, 1], according to the stability criterion
of continuous system, if k satisfies k > 27–6!, then the eigenvalues of the matrix A1
have negative real part. The zero point of the error system is asymptotically stable at
this time. During iT + TD < t < (i + 1)T, the controlled systems lose some additional
feedback, and the error is controlled by A2 . Because A2 does not satisfy the stability
condition, the trajectory of the two systems will diverge, and the synchronization of
the two systems will be lost. As long as the error is still small enough, and the ex-
ert control pulse is applied again, it can ensure that the error will not deviate too far,
so that the synchronization of the two systems will be achieved gradually. After the
continuous system is discretized, the error equation is written as
⎡ ⎤ ⎡ ⎤
e1 (i) e1 (i – 1)
⎢ ⎥ m m ⎢ ⎥
⎣ e2 (i) ⎦ = (A1 ) 1 (A2 ) 2 ⎣ e2 (i – 1) ⎦ , (4.79)
e3 (i) e3 (i – 1)
where e(i) is the error vector of the ith pulse period. DT = m1 h, (1 – D)T = m2 h, where
h is the integral step size, and m1 , m2 are the integral iteration numbers for a pulse
and no pulse, respectively. Obviously, in the process of synchronization, the system
will generate oscillation due to (A2 )m2 , and the stable synchronization process will
depend on |(A1 )m1 (A2 )m2 |. If |(A1 )m1 (A2 )m2 | = |J| < 1, then the system will become stable
[101]. The exact range of D can be obtained by numerical experiments. Simulations
show that when D ≥ 0.4, the exact synchronization can be achieved within 10 s. ∎
4.6.2 Synchronization Simulation and Performance Analysis
The synchronization control and performance improvement for the unified chaotic
system and its deformation system are studied based on Matlab simulation. Here, we
observe the changes of the time performance of the synchronization from the aspects
of the initial value, the feedback coefficient k, the proportion factor D, the step size h,
and the system parameter !.
4.6.2.1 Effect of Initial Value on Synchronization Performance

Set ! = 0.8, k = 40, T = 0.1 s, D = 0.6, h = 0.001, and the initial values are
(1) X(0) = [20, 0, 0], Y(0) = [–2, 1, 0]; (2) X(0) = [10, 0, 0], Y(0) = [–2, 1, 0]; and
(3) X(0) = [5, 0, 0], Y(0) = [–2, 1, 0]. The synchronization error curve is shown in
Figure 4.22. So, the synchronization between the two systems is achieved within 1.5 s.
The oscillation phenomenon occurs in the process of establishing the synchroniza-
tion, which is related to the characteristic value of A2 . Because at the beginning of
iT + TD < t < (i + 1)T, the error has not reached the steady state. In addition, the
oscillation for initial value (3) is minimum, while the oscillation for initial value (1) is
maximum. It can be seen that the initial value of the two systems is more; the larger
the oscillation, the more the time required to establish accurate synchronization. But
in general, it can achieve satisfactory results even when the initial value is large.
120
100
80 (1)
(2)
|e|
60
40 (3)
20
0
0 1 2 3
t/s
Figure 4.22: Synchronization error curves for different initial values.
15
10 k3
|e|
k2
5
k1
0
0 0.5 1 1.5
t/s
Figure 4.23: Synchronization error curves with different k.
4.6.2.2 Effect of different k, D, and h on synchronization performance

Set ! = 0.8, T = 0.1 s, D = 0.8, and h = 0.0001, and the initial values are X(0) =
[5, 0, 0] and Y(0) = [–2, 1, 0]. When the k values are 42, 32, and 27, the synchronization
error curves are shown in Figure 4.23. It shows that the smooth degree of the curve is
different with different k. For example, the error converges quickly for k = 27, and the
error converges the fastest for k = 42. So the establishment time of the synchronization
decreases with the increase of k, which is consistent with the theoretical analysis. In
the practical application, considering the cost and the difficulty of the realization, the
value of k should not be too large.
To achieve synchronization, the regulation of D is also important. Numerical
experiments show that, similar to k, the synchronization setup time decreases
with the increase of D. Further experiments prove that the smaller the size of
the h is, the smaller the oscillation time is, and the faster the error curve con-
verges.
4
tb/s
0
0 0.5 1
α
Figure 4.24: Relation curve of synchronization setup time tb with parameter !.
4.6.2.3 Effects of different parameter ! on synchronization performance

Let k = 45, T = 0.1 s, D = 0.6, and h = 0.0001, and the initial values are X(0) = [15, 0, 0],
Y(0) = [–15, –15, –15]. The relationship of the synchronization setup time tb with the
parameter ! is shown in Figure 4.24. It shows when ! ∈ [0, 1], the synchronization
between the unified chaotic system and its deformation system is achieved within
about 4 s, and the synchronization setup time decreases with increase of !.
4.7 Synchronization Control Based on State Observer Method

Because this synchronization method does not need to calculate the conditional
Lyapunov exponent, it is widely concerned.
4.7.1 Design Principle of State Observer
Consider a class of nonlinear feedback chaotic systems

⎧
⎪
⎨ẋ = Ax + Bu + Ew
⎪
y = CT x , (4.80)
⎪
⎪
⎩u = f (x)
where x ∈ Rn , u ∈ Rk , and w ∈ Rm are the state variables, nonlinear feedback input,

and random noise of the system, respectively. y is the output in the sender. A, B, C,
and E are known constant matrices.
In the receiver, when the nonlinear feedback input u is measurable, then the state
observer of the system can be expressed as [102]

ż = Fz + Gu + Hy
, (4.81)
x̂ = z – Jy
where z ∈ Rn and x̂ ∈ Rn . F, G, H, and J are unknown constant matrices.

Suppose In is a unit matrix with n order, the error of the state observer is defined
as e = x̂ – x = z – (In + JCT )x. Let M = In + JCT , then the dynamic equation of the error
system is
ė = Fe + (HCT – MA + FM)x + (G – MB)u – MEw. (4.82)
When error system (4.82) satisfies

⎧
⎪
⎨HC – MA + FM = 0
T
⎪
G – MB = 0 , (4.83)
⎪
⎪
⎩(I + JCT )E = 0
n
then the error system becomes ė = Fe. Obviously, if all eigenvalues of the matrix F are
less than zero, then x̂ converges to x exponentially.
Substitute M = In + JCT into the third term of HCT – MA + FM = 0, then we have
F = MA – (H + FJ)CT . (4.84)
Let
N = H + FJ (4.85)
and substitute eq. (4.84) into eq. (4.85), then we obtain
H = N + NCT J – MAJ. (4.86)
At this point, the state observer equation is changed to
ż = (MA – NCT )z + Gu + Hy. (4.87)
So, the design problem of the state observer with random disturbance is turned into
the problem of how to find the appropriate J, which satisfies (In + JCT )E = 0, and how
to choose the appropriate N, which satisfies the eigenvalues of F are less than zero.
As F = MA–NCT , the condition that the eigenvalues of F are less than zero must be
satisfied by choosing appropriate N. That is to say, in this case, we can always find a
suitable N, so that when t → ∞, x̂ converges to x exponentially.
From eq. (4.83), we obtain JCT E = –E, and according to matrix knowledge, we have
J = –E(CT E)+ + V(Ip – (CT E)(CT E)+ ), (4.88)

where (CT E)+ is the generalized inverse matrix of (CT E). V is the matrix with appropri-
ate dimensions. For the convenience of application, we often let V = 0. Then we have
J = –E(CT E)+ . At the same time, we can obtain M = Ip + JCT .
4.7.2 Design of State Observer for Unified Chaotic System
By applying the state observer method to the unified chaotic system, the sender matrix
is described as
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
–25! – 10 25! + 10 0 1 0 0 0 0
⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥
A = ⎣ 28 – 35! 29! – 1 0 ⎦ , B = ⎣ 0 –1 0 ⎦ , u = x x
⎣ 1 3⎦ , E = ⎣ 1 ⎦,
0 0 –(8 + !)/3 0 0 1 x1 x2 0
C = [0 1 0]T .
The receiver equation is

⎧ ⎡ ⎤ ⎡ ⎤
⎪
⎪ 1 0 0 25! + 10
⎪
⎪ ⎢ ⎥ ⎢ ⎥
⎪
⎪
⎪
⎪ ż = Fz + ⎢
⎣ 0 0 0⎥⎦ u+⎢
⎣ 0 ⎥y
⎦
⎪
⎪
⎪
⎪
⎨ 0 0 1 0
⎡ ⎤ , (4.89)
⎪
⎪
⎪
⎪ 0
⎪
⎪ ⎢ ⎥
⎪
⎪ ⎢ ⎥
⎪
⎪
⎪
x̂ = z – ⎣ –1 ⎦ y
⎪
⎩ 0
where
⎡ ⎤
–25! – 10 25! + 10 – n1 0
⎢ ⎥
F=⎣ 0 –n2 0 ⎦. (4.90)
0 –n3 –(8 + !)/3
Here, n1 , n2 , and n3 are the three values of the one-dimensional vector N, which can
ensure that the eigenvalues of matrix F are less than zero, and these lead the syn-
chronization error for ė = Fe to zero finally. For simple calculation, we set n1 = 25!+10,
n2 > 0, n3 = 0. When t → ∞, x̂ converges to x exponentially.
4.7.3 Simulation and Discussion
Based on Matlab simulation platform, the synchronization of the unified chaotic sys-
tem is investigated in this section. We set the initial values x1 (0) = 15, x2 (0) = 15,
x3 (0) = 15, x̂1 (0) = 10–10 , x̂2 (0) = –10, x̂3 (0) = 10–10 ; n1 = 25! + 10, n3 = 0, and let the
synchronization error
3 21

2
|e| = (x̂i – xi ) . (4.91)
i=1
Next, we study the synchronization performance of the synchronization system.
4.7.3.1 Synchronization Performance with Different Parameter !

When n2 = 3 and w(t) = 100 sin(3t), we obtain the synchronization error curves for
! = 0.5, ! = 0.8, ! = 0.9, respectively, as shown in Figure 4.25. It can be seen, al-
though the difference between the two initial values is very far, that the system can
achieve synchronization within t = 2 s for different system parameters. It shows that
the system has strong robustness to parameter variation.
4.7.3.2 Effect of Noise Interference on the Performance of System Synchronization

When n2 = 3 and ! = 0.8, we obtain the synchronization error curves for w(t) = 5
sin(3t), w(t) = 100 sin(3t), and white noise, respectively, as shown in Figure 4.26. It
can be seen that the type and size of the noise have little effect on the synchronization
error signal. The system can achieve the precise synchronization within t = 2 s. It
shows the method has strong antinoise performance.
4.7.3.3 Effect of Initial Values on the Performance of System Synchronization

When n2 = 3, ! = 0.8, and w(t) = 100 sin(3t), we obtain the synchronization error
curves for following initial values as shown in Figure 4.27.
(1) X(0) = [15, 15, 15] X̂(0) = [10–10 , –10, 10–10 ],
(2) X(0) = [50, 50, 50] X̂(0) = [10–10 , –10, 10–10 ],
(3) X(0) = [15, 15, 15] X̂(0) = [10, 5, 10].
25
20
15
|e|
10
0
0 1 2 3
t/s
Figure 4.25: Synchronization error curves for different !.

25
20
15
|e|
10
0
0 1 2 3
t/s
Figure 4.26: Synchronization error curve for different noise interferences.
80
60 (3)
40 (1)
|e|
(2)
20
0
0 1 2 3
t/s
Figure 4.27: Synchronization error curves for different initial conditions.
From Figure 4.27, the setup time for the system synchronization is the longest for (2),
and the setup time for the system synchronization is the shortest for (3). Because the
synchronization error signal e(t) decays exponentially, the time required to determine
the exact synchronization is related to the initial conditions of the transmission end
and the receiving end. It is obvious that the difference of the two initial values for (2)
between the sender and the receiver is the biggest, so it is the longest time to achieve
the exact synchronization.
4.7.3.4 Effect of Different n2 on the Performance of Synchronization System

When w(t) = 100 sin(3t), ! = 0.8, we obtain the synchronization error curves for n2 = 1,
n2 = 5, and n2 = 50, respectively, as shown in Figure 4.28. Obviously, when n2 > 0, then
the larger the n2 is, the shorter the synchronization time of the system is. But when n2
exceeds a certain value, then the value of n2 is increased, and the time needed for
synchronization is not significantly reduced.
25
20
n2 = 50
15 n2 = 5
|e|
10 n2 = 1
0
0 2 4 6
t/s
Figure 4.28: Synchronization error curves for different n2 .
4.8 Chaos Synchronization Based on Chaos Observer
4.8.1 Design Principle of Chaos Observer
For the following chaotic system

ẋ = F(x)
, (4.92)
s = h(x)
where x ∈ Rn , F: Rn → Rn ; s is the scalar output of the system. Set k(x) is a scalar dif-
ferentiable function, and V(x) is vector field, where V: Rn → Rn . Lee derivative of the
vector field V(x) along the scalar differentiable function k(x) is defined as Lv k, then
Lv k = <V(x) ● grad k(x)>, where < ● > is the Euclidean inner product. The ith Lee deriv-
ative of the vector field V(x) along the scalar differentiable function k(x) is defined as
Liv k = <V(x) ● grad Li–1
v k(x)>.
Lemma 4.1 [103]: If 6 = [h(x), Lf h(x), . . . , Ln–1

f h(x)] is diffeomorphism, then all the
T
state variables of system (4.92) are observable by using scalar output s and it is the ith
(i ≤ n – 1) Lee derivative, and x = 6–1 [s(t), s(1) (t), . . . , s(n–1) (t)]T .
If a state variable xi is selected as the output of system (4.92), then the remaining state
variables of system (4.92) can be expressed as xj = gi (xi , ẋi , . . . , xi(n–1) ), where j ≠ i. So
system (4.92) is written as
ẋ = Ax + g(xi , ẋi , . . . , xi(n–1) ), (4.93)
where A is a constant matrix, and g(xi , ẋi , . . . , xi(n–1) ) is a matrix with n× 1 order, which
includes all the nonlinear terms of system (4.92).
Lemma 4.2 [104]: Set the output of system (4.92) as
s(x) = g(xi , ẋi , . . . , xi(n–1) ) + B × [g1 (xi , ẋi , . . . , xi(n–1) ), . . . , xi , . . . , gn (xi , ẋi , . . . , xi(n–1) )]T ,
(4.94)
where B = diag[b1 , b2 , . . . , bn ]. If we select an appropriate diagonal matrix B so
that the eigenvalues of the matrix A–B have a negative real part, then system
ẏ = F(y) + s(x) – s(y) is the chaos observer of chaotic system (4.92), where s(y) =
g(yi , ẏi , . . . , yi(n–1) ) + B × [g1 (yi , ẏi , . . . , yi(n–1) ), . . . , yi , . . . , gn (yi , ẏi , . . . , yi(n–1) )]T .
4.8.2 Chaos Observer of the Unified Chaotic System
The state observer synchronization scheme is applied to the unified chaotic system,
and we have
⎡ ⎤
–25! – 10 25! + 10 0
⎢ ⎥
A = ⎣ 28 – 35! 29! – 1 0 ⎦,
0 0 –(8 + !)/3
⎧
⎪
⎪ x1 = g1 (x1 , ẋ1 , ẍ1 ) = x1
⎪
⎨ ẋ1
x2 = g2 (x1 , ẋ1 , ẍ1 ) = 25!+10 + x1 .
⎪
⎪
⎪
⎩x = g (x , ẋ , ẍ ) = –ẍ1 + 4!–11 ẋ + (27 – 6!)x
3 3 1 1 1 25!+10 25!+10 1 1 x1
Here, the drive system is driven by
s(x) = g(x1 , ẋ1 , ẍ1 ) + B × [x1 , g2 (x1 , ẋ1 , ẍ), g3 (x1 , ẋ1 , ẍ)]
⎡ ⎤ ⎡ ⎤
0 x1
⎢ ⎥ ⎢ ⎥
⎢ –ẍ1 + 4!–11 ẋ + (27 – 6!)x ⎥ ⎢ xẋ1 ⎥
=⎢ 25!+10 + x1
1 1⎥+B×⎢ ⎥,
⎢ 25!+10 25!+10 ⎥ ⎢ ⎥
⎣ x1 ẋ1 ⎦ ⎣ –ẍ ⎦
25!+10 + x1
2 1 4!–11
25!+10 + 25!+10 ẋ1 +(27–6!)x1
x1
where
⎡ ⎤
b11 b12 b13
⎢ ⎥
B = ⎣ b21 b22 b23 ⎦ .
b31 b32 b33
The state observer system in the receiver can be expressed as
ẏ = f (y) + s(x) – s(y), (4.95)
where f (y) = Ay + g(y1 , ẏ1 , ÿ1 , . . . , y1(n–1) ). s(y) is similar with s(x).
By eq. (4.95), when the appropriate B is taken, the characteristic value of the A–B
is negative, and then the accurate synchronization between the chaotic system in the
sender and the state observer in the receiver can be achieved.
4.8.3 Synchronization Simulation and Performance Analysis
The synchronization simulation of the unified chaotic system is carried out on Matlab,
and we set the system’s initial conditions x1 (0) = 1, x2 (0) = 1, x3 (0) = 1, x̂1 (0) = 15,
x̂2 (0) = 15, x̂3 (0) = 15,
⎡ ⎤
0 25! + 10 0
⎢ ⎥
B = ⎣ 28 – 35! b22 0⎦,
0 0 0
where b22 > 19! – 1, and let ei = (x̂i – xi ).

Next, we will study the synchronization performance of the synchronization
system.
4.8.3.1 Effect of Different Parameter ! on the Performance of System

Synchronization
For simulation, we set ! = 0.5, ! = 0.8, and ! = 0.9 (it means the system is the gen-
eralized Lorenz system, the generalized Lü system, and the generalized Chen system
respectively). The synchronization error curves are obtained as shown in Figure 4.29.
Obviously, for different parameter !, the unified chaotic system and the state observer
will achieve accurate synchronization, but the synchronization performance is not ex-
actly the same. This is mainly due to the use of a single parameter transmission, and
the function between x2 and x̂2 is complex. So this calculation will have a greater im-
pact on the synchronization performance between x2 and x̂2 with different ! values.
Overall, the synchronization performance is good for different ! values.
4.8.3.2 Effect of Initial Values on the Performance of System Synchronization

Setting b22 = 29 and ! = 0.8, the changes of the synchronous error signals
with the following three different initial conditions are analyzed, respectively. The
synchronization error curves of the system are shown in Figure 4.30:
Initial condition (1): X(0) = [ 15, 15, 15 ] X̂(0) = [ 10–10 , –10, 10–10 ].
Initial condition (2): X(0) = [ 50, 50, 50 ] X̂(0) = [ 10–10 , –10, 10–10 ].
Initial condition (3): X(0) = [ 15, 15, 15 ] X̂(0) = [ 10, 5, 10 ].
From Figure 4.30, when the initial values between the unified chaotic system and
the state observer are further away from each other, the oscillation is more obvi-
ous, and the time required for synchronization is longer, which is consistent with
20 20
(a) (b)
10
10
0
–10
e2
e2
0
–20
–10
–30
–20 –40
0 2 4 6 8 10 0 2 4 6 8 10
t/s t/s
20
(c)
10
–10
e2
–20
–30
–40
0 2 4 6 8 10
t/s
Figure 4.29: Synchronization error curves for different !: (a) ! = 0.5; (b) ! = 0.8; and (c) ! = 0.9.
the conclusion obtained by the previous state observer method. However, even if the
initial value of the two is far, the system can still achieve accurate synchronization.
4.8.3.3 Effect of Different b22 on the Performance of Synchronization System

When ! = 0.8, system initial values x1 (0) = 1, x2 (0) = 1, x3 (0) = 1, x̂1 (0) = 15, x̂2 (0) = 15,
x̂3 (0) = 15, and b22 = 24, b22 = 30, b22 = 50, respectively, we obtained the synchroniza-
tion error curves as shown in Figure 4.31. It shows that with the increase of b22 , the
oscillation amplitude is smaller and smaller before the system reaches synchroniz-
ation, and the time required to achieve synchronization is also less and less, but
the corresponding hardware requirements will be higher. So in practical applica-
tions, the appropriate increase of b22 is good for improving the performance of system
synchronization. In short, the system can achieve accurate synchronization in a
certain time, as long as the condition b22 > 29! – 1 is satisfied.
The chaos synchronization scheme, which is based on the state space recon-
struction technique, is relatively simple in design and calculation. In the process of
synchronization, the single parameter is transmitted, and the other parameters are de-
rived at the receiving end. In this way, the channel utilization rate is greatly improved.
30
(a) (b)
60
20
40
10
e2
e2
20
0
0
–10
0 2 4 6 8 10 0 2 4 6 8 10
t/s t/s
(c)
10
0
e2
–5
–10
0 2 4 6 8 10
t/s
Figure 4.30: Synchronization error curves for different initial conditions: (a) initial value (1);
(b) initial value (2); and (c) initial value (3).
As far as the synchronization performance, the proposed scheme has the character-
istics of short synchronization time and good synchronization performance. However,
the synchronous curve is not monotonous, and the oscillation is obvious; so, in prac-
tical application, to make the system synchronization performance optimization, the
selection of the system parameters is the key.
4.9 Projective Synchronization

In 1999, Mainieri and Rehacek [28] observed a new synchronization phenomenon
in the study of partially linear chaotic systems namely projective synchronization.
For some partially linear chaotic systems, when the proper controller is selected,
the output phase of the drive–response system state can be locked, and the amp-
litude of each corresponding state is also evolving according to a certain ratio. The
new synchronization phenomenon has been greatly concerned by researchers, and a
variety of projection synchronization schemes are proposed. Recently, a function pro-
jection synchronization method was proposed [29]. When compared with the general
10 10
(a) (b)
0
0
–10
–20 –10
e2
e2
–30
–20
–40
–50 –30
0 2 4 6 8 10 0 2 4 6 8 10
t/s t/s
10
(c)
5
0
e2
–5
–10
–15
0 2 4 6 8 10
t/s
Figure 4.31: Synchronization error curves for different b22 : (a) b22 = 24; (b) b22 = 30; and
(c) b22 = 50.
projective synchronization, the function projection synchronization means that the

drive system and the response system can be synchronized according to a certain
function, which is significant for the chaotic secure communication. In engineering
practice, the parameters of the system may be unknown. To realize the parameter
identification of chaotic systems with unknown parameters, the adaptive synchron-
ization method is applied to the synchronization of chaotic systems with unknown
parameters [29–30]. So it is important to study the projective synchronization con-
trol of chaotic systems, especially to study the synchronization of chaotic systems by
combining adaptive control and function projective synchronization.
4.9.1 Principle and Simulation of Proportional Projection Synchronization
Consider the following continuous chaotic system:
Ẋ = F(X), (4.96)
where the state variable X = (x1 , x2 , . . . , xn )T ∈ Rn , and the continuous vector function
F: Rn → Rn . The function F(X) is decomposed into
F(X) = ÃX + H̃(X), (4.97)
where ÃX is the linear part of F(X), and H̃(X) is the nonlinear part of F(X). Then, ÃX is
decomposed into
ÃX = AX + ĀX, (4.98)
where A is a full rank constant matrix, and the real part of the eigenvalue is
negative. Let
H(X) = ĀX + H̃(X), (4.99)
then eq. (4.96) can be rewritten as
Ẋ = AX + H(X). (4.100)
Now, we take system (4.100) as the drive system and employ the projective synchron-
ization control method. Then the response system is
Ẏ = AY + H(X)/!, (4.101)
where Y ∈ Rn is the n-dimensional state vector of the response system, and ! is the
proportional coefficient. If the synchronization error between drive system (4.100) and
response system (4.101) is E = X – !Y, then the synchronous error equation is
Ė = Ẋ – Ẏ = A(X – !Y) = AE. (4.102)
The real part of all eigenvalues of matrix A is negative, so synchronous error system
(4.102) is asymptotically stable at the origin according to the stability criterion of linear
system. That is, lim |E(t)| = 0. So the synchronization is achieved between response
t→∞
system (4.101) and drive system (4.100).
Applying the projective synchronization method to the simplified Lorenz chaotic
system, the drive system becomes
⎧
⎪
⎨ẋ1 = 10(x2 – x1 )
⎪
ẋ2 = (24 – 4c)x1 + cx2 – x1 x3 , (4.103)
⎪
⎪
⎩ẋ = x x – 8x /3
3 1 2 3
and the response system becomes

⎧
⎪
⎨ẏ1 = 10(y2 – y1 )
⎪
ẏ2 = (24 – 4c)y1 + by2 + !1 ((c – b)x2 – x1 x3 ) , (4.104)
⎪
⎪
⎩ẏ = –8y /3 + 1 x x
3 3 ! 1 2
where ! is the proportional coefficient (! ≠ 0), and b is the synchronization control

parameter. For the synchronization system comprising systems (4.103) and (4.104),
the following synchronization theorem is proposed.
Theorem 4.9: For the projection synchronization system composed by drive system
(4.103) and response system (4.104), if the synchronization control parameter b <
min[10, 4c–24], then the synchronization error system is asymptotically stable at the
origin point. That is, response system (4.104) will be synchronized with drive system
(4.103) according to the proportional !.
Proof. According to the principle of projective synchronization, for the projection syn-
chronization system comprising drive system (4.103) and response system (4.104), the
full rank matrix A and H(X) are, respectively,
⎡ ⎤ ⎡ ⎤
–10 10 0 0
⎢ ⎥ ⎢ ⎥
A = ⎣ 24 – 4c b 0 ⎦ , H(X) = ⎣ (c – b)x2 – x1 x3 ⎦ .
0 0 –8/3 x1 x2
If + is the eigenvalue of the matrix A, then we have

+ + 10 0
–10

|+I – A| = 4c – 24 +–b 0 = 0. (4.105)

0 0 + + 8/3
Obviously, one of the eigenvalues +1 = –8/3 and the other two characteristic roots
satisfy the equation
+2 + (10 – b)+ – 10b + 40c – 240) = 0. (4.106)
According to the Routh–Hurwitz criterion, if the synchronization control parameter

satisfies b < min[10, 4c–24], then the real part of the root of eq. (4.106) is negative. The
parameter of the system c ∈ [–1.59, 7.75], so we get b < –17.64. That is, when the control
parameter of the synchronization system is b < –17.64, the real part of all eigenvalues
of the matrix A of the synchronization error system is negative, and the synchroniza-
tion error is lim |E(t)| = 0. It means response system (4.104) will be synchronized with
t→∞
drive system (4.103) according to the ratio !. ∎
On Simulink dynamic simulation platform, we set c = 5. According to Theorem 4.9,

when b < –4, the synchronization will be achieved. Here we set b = –5 and ! = –2,
and system initial values x1 (0) = 4.4356, x2 (0) = 6.1771, x3 (0) = 7.2330; y1 (0) = 2.0883,
y2 (0) = 0.4535, y3 (0) = –5.8578. The error convergence curve, the evolution of the at-
tractor, and variable of the chaotic synchronization system are shown in Figure 4.32,
1/2
3
i=1 (xi – !yi )
where the synchronization error is defined by |E| = 2 . Figure 4.32(a)
15 20 20
(a) (b) (c)
x3
x3
|E| 10 10 10
5 0 0
y3
y3
0 –10 –10
0 5 10 –10 0 10 0 10 20
t/s x1, y1 t/s
Figure 4.32: Projective synchronization between systems (4.103) and (4.104):

(a) synchronization error curve; (b) attractors; and (c) waves of variables x3 and y3 .
shows that the two chaotic systems with different initial values can achieve the
projective synchronization at about 7.0 s, and the dynamic characteristics of the syn-
chronization is good. Figure 4.32(b) shows that the attractor of the response system
and drive system is inverse, and the size ratio is 2:1. According to Figure 4.32(c), we
can see that the response system variable y3 is synchronized with the drive system
variable x3 according to the projection ratio.
Now, we analyze the effects of different b values on the synchronization perform-
ance of the synchronization system. Set c = 5, ! = –2, and the initial values are same
with that of Figure 4.32. Simulations are carried out for b = –5, b = –6, b = –15, re-
spectively. The synchronization error curves are plotted in Figure 4.33. When b < –4,
the smaller the b, the shorter the time required to achieve synchronization, but when
b exceeds a certain value, then b reduces further, and the time required for synchron-
ization cannot be significantly reduced. By formula (4.106), the b value is smaller, and
the real part of the characteristic value of the synchronization error system is more
negative. The pole is farther from the virtual axis; the faster the convergence of the
synchronization system, the shorter the synchronization setup time.
3
|E|
2
(a)
1 (b)
(c)
0
0 5 10
t/s
Figure 4.33: Synchronization error curves for different control parameter b: (a) b = –5; (b) b = –6;
and (c) b = –15.
4.9.2 Principle and Simulation of Function Projective Synchronization
Consider the following continuous-time chaotic system as the drive system
Ẋ = AX + F(X), (4.107)
where X = (x1 , x2 , . . . , xn )T ∈ Rn is the state variable. AX is the linear part, and F(X)
is the nonlinear part. Copy eq. (4.107), and add the controller U: R2n → Rn into the
system, then the response system is
Ẏ = BY + G(Y) + U(X,Y), (4.108)
where Y = (y1 , y2 , . . . , yn )T ∈ Rn is the state vector. Set as the function matrix, and the
error function of the synchronization system is E = X – I(t)Y. By designing the con-
troller U(X, Y), we can get lim ||E|| = lim ||X – I(t)Y|| = 0, and then the generalized
t→∞ t→∞
function projective synchronization is achieved between systems (4.107) and system
(4.108).
Applying the generalized function projective synchronization method to the
simplified Lorenz system, the drive system and response system are, respectively,
⎡ ⎤⎡ ⎤ ⎡ ⎤
–10 10 0 x1 0
⎢ ⎥⎢ ⎥ ⎢ ⎥
Ẋ = AX + F(X) = ⎣ 24 – 4c c 0 ⎦ ⎣ x2 ⎦ + ⎣ –x1 x3 ⎦ , (4.109)
0 0 –8/3 x3 x1 x2
⎡ ⎤⎡ ⎤ ⎡ ⎤
–10 10 0 y1 0
⎢ ⎥⎢ ⎥ ⎢ ⎥
Ẏ = BY + G(Y) + U(X, Y) = ⎣ 24 – 4c c 0 ⎦ ⎣ y2 ⎦ + ⎣ –y1 y3 ⎦ + U(X, Y),
0 0 –8/3 y3 y1 y2
(4.110)
where c is the system parameter, and U = [u1 , u2 , u3 ]T is the nonlinear control-

ler. We have the following synchronization theorem for the synchronization system
comprising systems (4.109) and (4.110).
Theorem 4.10: For systems (4.109) and (4.110), the control law is U(X, Y) =
I–1 (t)[–İ(t)Y + F(X) – I(t)G(Y) + (A – B)X + DE], where I(t) is the projective func-
tion, and D = diag[d1 , d2 , d3 ] is the control parameter matrix. When the controller
parameter satisfies d1 + d2 > c–10, d1 d2 –d1 c + 10d2 > 240–30c, and d3 > –8/3, sys-
tems (4.109) and (4.110) can realize the function projective synchronization, that is,
lim E = 0 for any initial value.
t→∞
Proof. For the projection synchronization system comprising drive system (4.109) and
response system (4.110), the error function of the synchronization system is
E = X – I(t)Y, (4.111)
where I(t) is the diagonal function matrix. Then the error system equation is
Ė = Ẋ – İ(t)Y – I(t)Ẏ. (4.112)
Substituting eqs (4.109) and (4.110) into the above formula, then we have
Ė = AX + F(X) – BY – G(Y) – U(X, Y). (4.113)
If the controller is U(X, Y) = I–1 (t)[–İ(t)Y + F(X) – I(t)G(Y) + (A – B)X + DE], then we
obtain
Ė = (B – D)E. (4.114)
Construct Lyapunov function V(t) as
1
V = ET E. (4.115)
2
Then its differential is
V̇ = ET Ė. (4.116)
Substituting eq. (4.114) into the above formula, we have
V̇ = –ET (D – B)E. (4.117)
Set the control matrix D = diag[d1 , d2 , d3 ], when d1 +d2 > c–10, d1 d2 –d1 c+10d2 > 240–
30c, and d3 > –8/3, D–B is positive, and then V̇ ≤ 0 if and only if E = 0 for equality.
By the Lyapunov stability theorem, the error system of the drive–response system
is asymptotically stable at the origin. That is, lim ||E|| = 0. The function projection
t→∞
synchronization is implemented. ∎
On Simulink dynamic simulation platform, we set c = 5. According to Theorem 4.10,

when the control parameters are d1 = 40, d2 = 20, d3 = – 2, the synchronization will
be achieved. Set the projective function I(t) = – 3 + 2 sin t, and the initial values
x1 (0) = 4.4356, x2 (0) = 6.1771, x3 (0) = 7.2330; y1 (0) = – 1.8028, y2 (0) = – 2.3247, y3 (0) =
– 2.2974. The synchronization error convergence curve, the evolution of the attractor,
and variable of the chaotic synchronization system are shown in Figure 4.34. Fig-
ure 4.34(a) shows that the two chaotic systems with different initial values can
achieve the projective synchronization within 4.8 s, and the dynamic characteristics
1.5 20 20
(a) (b) (c)
x3
x3
1
|E|
0 0
0.5
y3
y3
0 –20 –20
0 5 10 –20 0 20 0 10 20
t/s x 1, y1 t/s

(a) Synchronization error curve; (b) attractors; and (c) waves of variables x3 and y3 .
of the synchronization are good. Figure 4.34(b) shows that the attractor of the re-
sponse system and drive system is inverse, and the size ratio satisfies the projective
function. According to Figure 4.34(c), we can see that the response system variable y3
is synchronized with the drive system variable x3 according to the projection function.
Now, we analyze the effects of different control parameter d1 , d2 , d3 on the syn-
chronization performance of the synchronization system. Set the system parameter
c = 5, the projective function I(t) = –3 + 2 sin t, and the initial values x1 (0) = 4.4356,
x2 (0) = 6.1771, x3 (0) = 7.2330; y1 (0) = –1.8028, y2 (0) = –2.3247, y3 (0) = –2.2974.
1. When d2 = 20, d3 = –2, then d1 > –7.3 according to Theorem 4.10. Synchronization
simulations are carried out for d1 = –7, d1 = –6.7, d1 = –5, respectively. The syn-
chronization error curves are plotted in Figure 4.35(a). Obviously, when d1 > –7.3,
the bigger the d1 , the shorter the time required to achieve synchronization, but
when the d1 exceeds a certain value, and then continue to increase d1 , the time
required for synchronization cannot be significantly reduced.
2. When d1 = 40 and d3 = –2, then d2 > 5.8 according to Theorem 4.10. Synchroniz-
ation simulations are carried out for d2 = 6.15, d2 = 6.5, d2 = 7, respectively. The
synchronization error curves are plotted in Figure 4.35(b). Obviously, when d2 >
5.8, the bigger the d2 , the shorter the time required to achieve synchronization,
1.5 1
(a) (b) (c)
0.4
1
|E|
|E|
|E|
0.5
0.2
0.5 1 1
1
2 2
3 32 3
0 0 0
0 5 10 15 0 5 10 15 0 5 10 15
t/s t/s t/s
Figure 4.35: Synchronization error curves for different control parameter d. (a) Curve 1: d1 = –7;
curve 2: d1 = –6.7; curve 3: d1 = –5. (b) Curve 1: d2 = 6.15; curve 2: d2 = 6.5; curve 3: d2 = 7. (c) Curve
1: d3 = –2.3; curve 2: d3 = –2; curve 3: d3 = 0.
but when the d2 exceeds a certain value, and continue to increase d2 , the time
required for synchronization cannot be significantly reduced.
3. When d1 = 40 and d2 = 20, then d3 > –8/3 according to Theorem 4.10. Synchron-
ization simulations are carried out for d3 = –2.3, d3 = –2, d3 = 0, respectively.
The synchronization error curves are plotted in Figure 4.35(c). Obviously, when
d3 > –8/3, the bigger the d3 , the shorter the time required to achieve synchroniz-
ation, but when the d2 exceeds a certain value, if we continue to increase d3 , the
time required for synchronization cannot be significantly reduced.
4.9.3 Principle and Simulation of the Adaptive Function Projective Synchronization
Consider the following drive system and response system:
Ẋ = AX + f (X) + F(X)(, (4.118)

Ẏ = BY + g(Y) + G(Y)$ + U(X, Y), (4.119)
where ( and $ are unknown parameter vectors in the drive system and the response
system, respectively. f (X) and g(Y) are vector functions of state variables X and Y.
F(X) and G(Y) are function matrices. U(X, Y) is the control law.
Supposing the projective function is I(t), and the error function of the synchron-
ization system is E = X–I(t)Y, then the synchronization error system equation is
Ė = Ẋ – İ(t)Y – I(t)Ẏ. (4.120)
Due to eqs (4.118) and (4.119), we have
Ė = Ẋ – İ(t)Y – I(t)Ẏ = M(X, Y) + F(X)( – İ(t)Y + BE – I(t)G(Y)$ – I(t)U(X, Y),

(4.121)
where M(X, Y) = f(X) – I(t)g(Y) + (A – B)X. Obviously, the adaptive function projective
synchronization problem of drive system (4.118) and response system (4.119) is trans-
formed into the stability problem of eq. (4.121). By designing the controller U(X, Y)

and the adaptive law, to make lim E lim X – I(t)Y = 0, the adaptive function
t→∞ t→∞
projective synchronization between systems (4.118) and (4.119) can be realized and
the unknown parameters can be identified.
Theorem 4.11: For the chaotic synchronization system with unknown parameter,
which is composed of drive system (4.118) and response system (4.119), if the control
function of the system is U(X, Y) = I–1 (t)[M(X, Y) + F(X)(ˆ – İ(t)Y – I(t)G(Y)$̂ + DE],
where M(X, Y) = f(X) – I(t)g(Y) + (A – B)X, and D = diag[d1 , d2 , d3 ]. If the appropri-
ate control parameters and the adaptive law of the system parameters are selected as
˙ ˙
(ˆ = F T (X)E + B(, $̂ = –I(t)GT (Y)E + B$, where (¯ = ( – (,
ˆ $̄ = $ – $̂. The adaptive func-
tion projective synchronization is realized between systems (4.118) and (4.119), and
the unknown parameters are identified correctly. That is, for any initial values, we
have lim E = 0, (ˆ = (, and $̂ = $.
t→∞
Proof. Choose U(X, Y) = I–1 (t)[M(X, Y) + F(X)(ˆ – İ(t)Y – I(t)G(Y)$̂ + DE] as the con-
troller, then the error system eq. (4.120) becomes
Ė = F(X)(¯ – I(t)G(Y)$̄ – (D – B)E. (4.122)
Constructing Lyapunov function
1 T 1 T 1 T
V= E E + (¯ (¯ + $̄ $̄, (4.123)
2 2 2
where (¯ = ( – (,
ˆ $̄ = $ – $̂, then its differential is
T˙ T˙
V̇ = ET Ė + (¯ (˙¯ + $̄ $̄˙ = ET Ė – (¯ (ˆ – $̄ $̂.
T T
(4.124)
˙ ˙
Set the adaptive law as (ˆ = F T (X)E + (¯ and $̂ = – I(t)GT (Y)E + $̄, and substitute eq.
(4.121) into eq. (4.124), then we have
T T
V̇ = – (D – B)ET E – (¯ (¯ – $̄ $̄. (4.125)
Obviously, if we choose control parameters d1 , d2 , and d3 , to make (D–B) positive def-

inite, then we have V̇ ≤ 0. If and only if E = 0, (¯ = 0, $̄ = 0 for equality. By the Lyapunov
stability theorem, the error system of the drive–response system is asymptotically
stable at the origin. That is, lim ||E|| = 0, and the adaptive function projection syn-
t→∞
chronization is implemented. At the same time, when t → ∞, we have (ˆ → ( and $̂ → $.
That is, the unknown parameters are identified correctly. ∎
Remark 1: If A = B, f (⋅) = g(⋅), and F(⋅) = G(⋅), then the two chaotic systems are ho-
mogeneous systems, and vice versa they are heterogeneous systems. Therefore, the
adaptive function projective synchronization theorem can be applied to both homo-
geneous and heterogeneous systems with unknown parameters.
Remark 2: If the parameters of the drive system or response system are known, us-
ing the above control method, the parameter adaptation law is removed, and the
corresponding generalized function projective synchronization can be realized.
4.9.3.1 Adaptive Function Projective Synchronization and Simulation

for Driving System with Unknown Parameters
Applying the adaptive function projective method to the simplified Lorenz system, the
drive system and response system are, respectively,
⎡ ⎤⎡ ⎤ ⎡ ⎤ ⎡ ⎤⎡ ⎤
–10 10 0 x1 0 0 0 0 (1
⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥
Ẋ = AX + f (X) + F(X)( = ⎣ 0 0 0 ⎦ ⎣ x2 ⎦ + ⎣ –x1 x3 ⎦ + ⎣ x1 x2 0 ⎦ ⎣ (2 ⎦ ,
0 0 –8/3 x3 x1 x2 0 0 0 (3
(4.126)
⎡ ⎤⎡ ⎤ ⎡ ⎤
–10 10 0 y1 0
⎢ ⎥⎢ ⎥ ⎢ ⎥
Ẏ = BY + g(Y) + U(X, Y) = ⎣ 24 – 4c c 0 ⎦ ⎣ y2 ⎦ + ⎣ –y1 y3 ⎦ + U(X, Y),
0 0 –8/3 y3 y1 y2
(4.127)
where c is the system parameter. C = [(1 , (2 , (3 ]T is the unknown parameter of the drive
system. U = [u1 , u2 , u3 ]T is the nonlinear controller. If we set the projective function
as I(t) = –3 + 2 sin(t), according to Theorem 4.11, the control law and the para-
meter adaptive law of the synchronization system consisting of systems (4.126) and
(4.127) are
⎡ ⎤
–İ(t)y1 + d1 e1
1 ⎢ ˆ ⎥
U= ⎣ ((1 – 24 + 4c)x1 + ((ˆ 2 – c)x2 – İ(t)y2 – x1 x3 + I(t)y1 y3 + d2 e2 ⎦ , (4.128)
I(t)
–İ(t)y3 + x1 x2 – I(t)y1 y2 + d3 e3
⎡ ⎤
x1 e2 + (1 – (ˆ 1
˙ ⎢ ⎥
(ˆ = ⎣ x2 e2 + (2 – (ˆ 2 ⎦ . (4.129)
0
There is a following inference for the synchronization system consisting of systems

(4.126) and (4.127).
Inference 4.2 For the adaptive function projective synchronization system comprising
drive system (4.126) and response system (4.127), when the controller parameters sat-
isfy d1 + d2 > c–10, d1 d2 –d1 c + 10d2 > 240–30c, and d3 > –8/3, the adaptive function
projective synchronization is achieved between systems (4.126) and (4.127). That is,
lim E = 0 for any initial values. (Proof omitted.)
t→∞
On Simulink platform, we set c = 5. According to Inference 4.2, when the con-

trol parameters are d1 = 40, d2 = 20, and d3 = – 2, the synchronization will be
achieved. Set the projective function I(t) = – 3 + 2 sin(t), and the initial values
X(0) = [4.4356, 6.1771, 7.2330]T , Y(0) = [0.5431, 0.0995, 2.8096]T , (ˆ 1 (0) = – 1, (ˆ 2 (0) = – 1.
The synchronization error convergence curve, the evolution of the attractor, and
variable waves of the chaotic synchronization system are shown in Figure 4.36.
Figure 4.36(a) shows that the two chaotic systems with different initial values can
achieve the projective synchronization in 4 s, and the dynamic characteristics of the
synchronization are good. Figure 4.36(b) shows that the attractor of the response
system and drive system is inverse phase, and the size ratio satisfies the projective
function. Figure 4.36(c) shows that in a relatively short period of time (about 2 s),
8 20 8
(a) (b) (c)
6 6
x3
4
θ1, θ2
|E|
4 0
2
2
y3
0
0 ‒20 ‒2
0 5 10 ‒20 0 20 0 5
t/s x 1, y 1 t/s

(a) Synchronization error curve; (b) attractors; and (c) waves of the estimated parameters (ˆ 1 and (ˆ 2 .
the estimated parameters (ˆ 1 and (ˆ 2 converge to the target value and realize the
identification of unknown parameters.
Now, we analyze the effects of different control parameters d1 , d2 , and d3 on
the synchronization performance of the synchronization system. When d2 = 20 and
d3 = – 2, then d1 > –7.3 according to Inference 4.2. Synchronization simulations are
carried out for d1 = – 7, d1 = – 6.7, d1 = – 5, respectively. The synchronization error
curves are plotted in Figure 4.37. Obviously, when d1 > –7.3, the bigger the d1 , the
shorter the time required to achieve synchronization, but when the d1 exceeds a cer-
tain value and continue to increase d1 , the time required for synchronization cannot
be significantly reduced. Similarly, when d1 and d3 are fixed, d2 changes, or when d1
and d2 are fixed, d3 changes, and the law is the same.
4.9.3.2 Synchronization Control and Simulation of Homogeneous

Chaotic Systems with Unknown Parameters
Applying the adaptive function projective synchronization method to two simpli-
fied Lorenz system with unknown parameters, the drive and response systems are,
respectively,
1.5
1
|E|
0.5
1
2
3
0
0 5 10 15
t/s
Figure 4.37: Synchronization error curves for different control parameter d1 .

⎡ ⎤⎡ ⎤ ⎡ ⎤ ⎡ ⎤⎡ ⎤
–10 10 0 x1 0 0 0 0 (1
⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥
Ẋ = AX + f (X) + F(X)( = ⎣ 0 0 0 ⎦ ⎣ x2 ⎦ + ⎣ –x1 x3 ⎦ + ⎣ x1 x2 0 ⎦ ⎣ (2 ⎦ ,
0 0 –8/3 x3 x1 x2 0 0 0 (3
(4.130)
⎡ ⎤⎡ ⎤ ⎡ ⎤
–10 10 0 y1 0
⎢ ⎥⎢ ⎥ ⎢ ⎥
Ẏ = BY + g(Y) + G(Y)$ + U(X, Y) = ⎣ 0 0 0 ⎦ ⎣ y2 ⎦ + ⎣ –y1 y3 ⎦
0 0 –8/3 y3 y1 y2
⎡ ⎤⎡ ⎤
0 0 0 $1
⎢ ⎥⎢ ⎥
+ ⎣ y1 y2 0 ⎦ ⎣ $2 ⎦ + U(X, Y), (4.131)
0 0 0 $3
where ( = [(1 , (2 , (3 ]T and $ = [$1 , $2 , $3 ]T are unknown parameters of the chaotic sys-
tems. U = [u1 , u2 , u3 ]T is the nonlinear controller. If the projective function is I(t) =
–3 + 2 sin(t), then according to Theorem 4.11, the control law and parameter adaptive
law of the synchronization are, respectively,
⎡ ⎤
–İ(t)y1 + d1 e1
1 ⎢ ⎥
U= ⎣ –x1 x3 + I(t)y1 y3 + x1 (ˆ 1 + x2 (ˆ 2 – İ(t)y2 – I(t)(y1 $̂1 + y2 $̂2 ) + d2 e2 ⎦ ,
I(t)
x1 x2 – I(t)y1 y2 – İ(t)y3 + d3 e3
(4.132)
⎡ ⎤
x1 e2 + (1 – (ˆ 1
˙ˆ ⎢ ⎥
( = ⎣ x2 e2 + (2 – (ˆ 2 ⎦ , (4.133)
0
⎡ ⎤
–I(t)y1 e2 + $1 – $̂1
˙ ⎢ ⎥
$̂ = ⎣ –I(t)y2 e2 + $2 – $̂2 ⎦ , (4.134)
0
On Simulink platform, we set the unknown parameter c = 5. It means the true value of
the estimated parameter is ( = $ = [0, 4, 5]T . If the initial conditions of the synchron-
ization system are X(0) = [4.4356, 6.1771, 7.2330]T , Y(0) = [–0.5431, 0.0995, –2.8096]T ,
ˆ
((0) = [–1, –1, 0]T , $̂(0) = [–2, –2, 0]T , respectively, and d1 = d2 = d3 = 5, we ob-
tain the synchronization error curves, attractors, and error curves of the estimated
parameters as shown in Figure 4.38. Figure 4.38(a) shows that the two chaotic sys-
tems with different initial values and unknown parameters can achieve the function
projective synchronization in about 2.5 s, and it has oscillation attenuation character-
istic. Figure 4.38(b) shows that the attractor of the response system and drive system
is inverse, and the size ratio satisfies the projective function. From Figure 4.38(c) and
(d), it can be seen that in a short time, the error of the estimated parameters (ˆ 1 , (ˆ 2
and $̂1 , $̂2 converges to zero, which indicates that the unknown parameters are iden-
tified. The results also show that the larger the control parameter di is, the better
4 20
(a) (b)
x3
|E|
2 0
y3
0 –20
0 5 10 –20 0 20
t /s x1, y1
8 8
(c) (d)
6 6
Δθ1, Δθ2
Δδ1, Δδ2
4 4
2 2
0 0
0 5 10 0 5 10
t /s t /s
Figure 4.38: Projective synchronization between systems (4.130) and (4.131); (a) synchronization
error curve; (b) attractors; (c) error curves for estimated parameters (ˆ 1 and (ˆ 2 ; and (d) error curves for
estimated parameters $̂1 and $̂2 .
the synchronization performance is, which is important for secure communication

application.
4.9.3.3 Synchronization Control and Simulation of Heterogeneous

Chaotic Systems with Unknown Parameters
Applying the adaptive function projective synchronization method to the simplified
Lorenz system and Chen system with unknown parameters, the drive system and the
response system are, respectively,
⎡ ⎤⎡ ⎤ ⎡ ⎤ ⎡ ⎤⎡ ⎤
–10 10 0 x1 0 0 0 0 (1
⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥
Ẋ = AX + f (X) + F(X)( = ⎣ 0 0 0 ⎦⎣ x2 ⎦ + ⎣ –x1 x3 ⎦ + ⎣ x1 x2 0 ⎦⎣ (2 ⎦,
0 0 –8/3 x3 x1 x2 0 0 0 (3
(4.135)
⎡ ⎤⎡ ⎤ ⎡ ⎤
–10 10 0 y1 0
⎢ ⎥⎢ ⎥ ⎢ ⎥
Ẏ = BY + g(Y) + G(Y)$ + U(X, Y) = ⎣ –10 0 0 ⎦ ⎣ y2 ⎦ + ⎣ –y1 y3 ⎦
0 0 –8/3 y3 y1 y2
⎡ ⎤⎡ ⎤
y2 – y1 0 0 $1
⎢ ⎥⎢ ⎥
+ ⎣ –y1 y1 + y2 0 ⎦ ⎣ $2 ⎦ + U(X, Y),
0 0 –y3 $3
(4.136)
where ( = [(1 , (2 , (3 ]T and $ = [$1 , $2 , $3 ]T are the unknown parameters of the two
systems, and U = [u1 , u2 , u3 ]T is the nonlinear controller. If we set the projective func-
tion as I(t) = –3 + 2 sin(t), according to Theorem 4.11, the control law and parameter
adaptive law of the synchronization system are
⎡ ⎤
–İ(t)y1 – I(t)(y2 – y1 )$̂ + d1 e1
1 ⎢ ⎥
U= ⎣ 10x1 – x1 x3 +I(t)y1 y3 +x1 (ˆ 1 +x2 (ˆ 2 – İ(t)y2 – I(t)[–y1 $̂1 +(y1 +y2 )$̂2 ]+d2 e2 ⎦,
I(t)
x1 x2 – I(t)y1 y2 – İ(t)y3 + I(t)y3 $̂3 + d3 e3
(4.137)
⎡ ⎤
x1 e2 + (1 – (ˆ 1
˙ˆ ⎢ ⎥
( = ⎣ x2 e2 + (2 – (ˆ 2 ⎦, (4.138)
0
⎡ ⎤
–I(t)[(y2 – y1 )e1 – y1 e2 ] + $1 – $̂1
˙ ⎢ ⎥
$̂ = ⎣ –I(t)(y1 + y2 )e2 + $2 – $̂2 ⎦. (4.139)
I(t)y3 e3 + $3 – $̂3
For simulation, we set the unknown parameter of the simplified Lorenz system c = 5,
which means that the true value of the estimated parameter is ( = [0, 4, 5]T . At the
same time, we choose the true value of the unknown parameter of the Chen system
$ = [25, 28, 1/3]T . If the initial conditions of the system and unknown parameter are
X(0) = [4.4356, 6.1771, 7.2330]T , Y(0) = [–0.62, –0.4261, –4.0838]T , ((0) ˆ = [–1, –1, 0]T ,
$̂(0) = [2, 2, 2] , and d1 = d2 = d3 = 5, then we obtain the synchronization error curves,
T
attractors, and error curves of the estimated parameters as shown in Figure 4.39.
Figure 4.39(a) shows that the two chaotic systems with different initial values and
unknown parameters can achieve the function projective synchronization in about 2
s, and it has oscillation attenuation characteristic. Figure 4.39(b) shows that the at-
tractor of the response system and drive system is inverse phase, and the size ratio
satisfies the projective function. From Figure 4.39(c) and (d), we know that in a short
time, the error of the estimated parameters (ˆ 1 , (ˆ 2 and $̂1 , $̂2 , $̂3 converges to zero, which
indicates the unknown parameters are identified. The results also show that the larger
the control parameter di is, the better the synchronization performance is. That is, the
setup time of the synchronization is shorter.
4.10 Problems and Development of Chaos Synchronization 123
15 20
(a) (b)
x3
10
0
|E|
y3
0 –20
0 5 10 –20 0 20
t /s x1, y1
10 30
(c) Δδ1, Δδ2, Δδ3 (d)
20
Δθ1, Δθ2
10
0
0
0 5 0 5
t /s t /s
Figure 4.39: Projective synchronization between systems (4.135) and (4.136): (a) synchronization
error curve; (b) attractors; (c) error curves for estimated parameters (ˆ 1 and (ˆ 2 ; and (d) error curves for
estimated parameters $̂1 , $̂2 , and $̂3 .
4.10 Problems and Development of Chaos Synchronization

Combining with the new control strategies or dynamic methods, how to improve
the performance of chaotic synchronization is key for the application research of
chaos. Under the condition of noise interference, channel distortion, and parameter
mismatch, the synchronization performance improvement of chaotic system is an im-
portant part of the research of chaos synchronization. In terms of the current research
of chaos synchronization, we should combine the robust method of control and the
stability of dynamics to improve the robustness and the synchronization performance
of the synchronization system.
We should pay more attention on the problem of hyperchaos synchronization,
because the secure communication based on hyperchaotic system has better security
performance. For example, hyperchaos system is a high-dimensional strange attractor
in the appropriate regression map. The modulated signal and the original chaotic sig-
nal are mixed together, and it cannot be resolved, so it is difficult to decryption. At
present, the research on the hyperchaotic system and its synchronization is still in the
early stage. The difficulties and hot topics in the research of chaos synchronization
theory include the criteria, the physical mechanism, and the applications of the
hyperchaotic synchronization.
It is very important to study the effects of noise interference and channel dis-
tortion on synchronization. In the real transmission process, the chaotic signal is
inevitably subject to noise interference. In addition, any transmission channel is not
ideal. So the transmitted signal will have amplitude attenuation, phase, and nonlinear
distortion. This will inevitably have a direct or indirect effect on the synchroniza-
tion of chaotic systems. It is very important to develop such kind of research for the
application of chaotic synchronization in secure communication.
Although chaos synchronization control method has been studied for more than
30 years, and a lot of physics and information scientists have done a lot of work in both
theory and experiment, it is still in the exploratory stage. We believe many problems
such as synchronization mechanism, anti-interference ability, and so on will be solved
in the future.
Questions
1. What are the performance index of chaotic synchronization system and its signi-
ficance?
2. What are the criteria or guidelines for synchronization of chaotic systems?
3. What are the key issues and challenges facing the chaos synchronization?
4. Try to construct a chaotic synchronization system and analyze its synchronization
performances by simulation.
5 Secure Communication Technology
Based on Chaos Synchronization
Chaotic system is sensitively dependent on the initial conditions, long-term unpre-

dictability, and anti-interception capability. At the same time, chaotic system itself is
deterministic, and it is completely determined by the equations, parameters, and ini-
tial conditions of the nonlinear system. Hence, it is easy to be produced and copied.
The high randomness, unpredictability, high complexity, wide frequency band char-
acter, the deterministic parameters and initial conditions of the chaotic signal, and its
easy implementation make it a hot topic in the research of chaos application.
When compared with modern communication, chaos communication has many
advantages and differences. The similarities and differences between them are shown
in Table 5.1 [18]. When compared with the traditional modern communication, the
chaos communication has obvious advantages for its potential applications and devel-
opment prospects. However, the chaotic secure communication technology, which is
still in the early stages of development, needs to solve a series of key technical issues.
Chaotic secure communication is divided into chaos analogy communication and
chaos digital communication. Chaos analog communication is usually achieved by
nonlinear circuit systems, but it requires very high standards for electrical design
and manufacture precision. In addition, it is difficult for synchronization implement-
ation. Among the existing chaotic communication schemes, chaotic masking is used
to transmit analog signals, and its disadvantage is that the fidelity and security of
the scheme are still not satisfied. Chaotic digital communication has become a hot
research topic because of its less dependence on the circuit element, easy implement-
ation in hardware, convenience in computer processing, transmission with less loss
of information, strong versatility, and wide application [112]. Chaotic digital commu-
nication mainly includes chaos shift keying (CSK), chaos spread spectrum, and chaos
modulation.
5.1 Chaotic Masking Communication

Chaotic masking is the simplest form of chaotic communication, where the inform-
ation signal is superimposed on the chaotic signal directly and it is masked by the
chaotic signal [113]. The principle block diagram is shown in Figure 5.1. Synchron-
ization is achieved between the drive subsystem and the response subsystem. At the
sender, message signal and chaotic signal with much larger amplitude are superposed
to form a noise signal. The chaotic signal completely masks the useful information
and realizes the secure communication. At the receiver, the response system is cap-
able of reproducing the chaotic signal. If the received signal is subtracted by the
126 5 Secure Communication Technology Based on Chaos Synchronization
Table 5.1: Comparison between modern communication and chaotic communication.
Content Modern communication Chaotic communication
History Matured technology for over 100 Emerging information technology

years for over 20 years
Industrial base Has become the core of the field of Research hot spot
information technology
Emission bandwidth Information bandwidth and Broadband spectral emission
broadband emission
Relationship between Category obviously, system Selection of new methods and
theory and technology integrity (from information theory new schemes to seek a
to communication technology, breakthrough. Some approved
from communication software to communication schemes, such as
circuit hardware, etc.) circuit, laser, and computer
network, have been proposed
Security technology Mathematical methods, involving Physical methods, involving chaos
foundation complex mathematical knowledge characteristics
Key security technology DES, IDEA, 3DES, RSA Chaos masking, chaos switching,
chaos modulation
Confidentiality Good performance, but there are aperiodic, unpredictable, fast
loophole due to the period of the fading correlation function, and
larger random number. It can be broadband characteristics.
decoded, but it is very Stronger confidentiality
time-consuming, and high cost.
Robustness Achieved certain robustness Application of spatiotemporal
chaos and hyperchaos can
enhance the robustness. Difficult
to be attacked
Capacity Bulk can be achieved, but there Dynamic storage capability with
are still restrictions high capacity
Power High Low power and observational
ability
Equipment cost High Relatively low
The main safety issues The fatal vulnerabilities of It is studying and is expected to
of security system software, inefficient attack address key issues, such as noise,
protection, and poor random channel distortion, and
number generator multichannel and multiuser
problems
chaotic signal generated by the response system, the message signal can be recovered
correctly.
Chaotic masking is realized by utilizing the chaos randomness and broadband
power spectrum characteristics. Taking the low-dimensional chaotic synchronization
signal as a modulation signal, the transmitted message is superposed on the chaotic
signal. From the observation of power spectrum or chaotic time series, the transmitted
message is not easy to be detected directly. To accurately extract the message signal,
Secret signal
k + – 1/k
m(t) s(t) m΄(t)
x1 y1
Driving signal
x2 y2
x3 y3
Sender Receiver
Figure 5.1: Double-channel chaotic secure communication scheme based on drive–response

synchronization.
this method requires the sender and receiver chaotic circuit parameters to match ex-
actly. That is, it requires the same initial state and the parameters. Synchronization
will be failed for tiny difference. Therefore, it is very difficult for the attacker to ana-
lyze the key parameters. On the other hand, the parameters that match exactly put
forward very high requirements to the design of the circuit.
At the sender, taking Lorenz chaotic system as an example, the mathematical
model of the modulator is described as
⎧
⎪
⎪ ẋ1 = 3(x2 – x1 )
⎪
⎪
⎪
⎨ẋ = –x x + rx – x
2 1 3 1 2
, (5.1)
⎪
⎪ẋ3 = x1 x2 – bx3
⎪
⎪
⎪
⎩s(t) = km(t) + x
1
where k is a compression ratio constant, which ensures that the useful signal m(t) is
much less than the chaotic signal. x1 is the masking signal, and x2 is the synchronous
driving signal.
At the receiver, the mathematical model of the demodulator is defined as
⎧
⎪
⎪ ẏ1 = 3(x2 – y1 )
⎪
⎪
⎪
⎨ẏ = –y y + ry – y
2 1 3 1 2
. (5.2)
⎪
⎪ ẏ = y x – by
⎪
⎪ 3 1 2 3
⎪
⎩m′ (t) = (s(t) – y )/k
1
Obviously, as long as the synchronization between the response system and drive sys-
tem is achieved completely, the useful information can be demodulated correctly. In
order to save the channel resources, this scheme is improved, and the principle is
shown in Figure 5.2. In this scheme, the driving signal consists of useful signal and
chaotic signal, that is, the drive signal is the sum of the useful signal and chaotic sig-
nal, so only one channel is required to realize the secure communication. Here, taking
Secret signal
k + – 1/k
m(t) s(t) m΄(t)
x1 y1
x2 y2
Driving signal
x3 y3
Sender Receiver
Figure 5.2: Single-channel chaotic secure communication scheme based on drive–response

synchronization.
Lorenz chaotic system as an example, the chaotic system equations for the sender and
the receiver are given by, respectively,
⎧
⎪
⎪ ẋ1 = 3(x2 – x1 )
⎪
⎪
⎪
⎨ẋ = –x x + rx – x
2 1 3 1 2
, (5.3)
⎪
⎪ ẋ = x x – bx
⎪
⎪ 3 1 2 3
⎪
⎩s(t) = km(t) + x
1
⎧
⎪
⎪ ẏ1 = 3(s(t) – y1 )
⎪
⎪
⎪
⎨ẏ = –y y + ry – y
2 1 3 1 2
. (5.4)
⎪
⎪ ẏ = y s(t) – by
⎪
⎪
3 1 3
⎪ ′
⎩ m (t) = (s(t) – y1 )/k
In this scheme, the channel resource is saved. However, one of the obvious character-
istics is that the drive signal is disturbed by the message signal, and synchronization
cannot be accurately realized, which leads to the fact that the information signal can-
not be fully recovered. Another notable feature is that the information signal doesn’t
have any impact on the dynamic characteristics of the driving system, so the security
of the secure communication system is not high enough, and it cannot effectively resist
the third-party attack. To improve the security further, an improved scheme was pro-
posed in Ref. [114] as shown in Figure 5.3. In this scheme, the sum of the information
signal and the chaotic signal feedbacks to the drive system, so that the drive system
and the response system have the same dynamic characteristics, so it can realize the
complete synchronization, and the receiving signal can be demodulated completely.
As the information signal is integrated into the chaotic system, it is more difficult
for the attacker to obtain useful information from the channel. Here, taking Lorenz
chaotic system as an example, the chaotic system equations for the sender and the
receiver is written as, respectively,
k 1/k
m(t) m΄(t)
s(t)
+ –
x1 y1
x2 y2
x3 y3
Sender Receiver
Figure 5.3: Improved single-channel chaotic secure communication scheme based on

drive–response synchronization.
⎧
⎪
⎪ẋ1 = 3(s(t) – x1 )
⎪
⎪
⎪
⎨ẋ = –x x + rx – x
2 1 3 1 2
, (5.5)
⎪
⎪ ẋ3 = x1 s(t) – bx3
⎪
⎪
⎪
⎩s(t) = m(t) + x
2
⎧
⎪
⎪ ẏ1 = 3(s(t) – y1 )
⎪
⎪
⎪
⎨ẏ = –y y + ry – s(t)
2 1 3 1
. (5.6)
⎪
⎪ẏ3 = y1 s(t) – by3
⎪
⎪
⎪ ′
⎩m (t) = s(t) – y2
To improve the secure performance of chaotic masking communication further, Car-

roll et al. [115] spread this improved chaotic masking scheme to the high-order-linked
synchronization system, and its synchronization principle and secure communication
scheme are shown in Figure 5.4. In this scheme, the receiver and the sender are com-
posed of two cascaded chaotic systems. Taking Lorenz chaotic system as an example,
the equations for the sender and the receiver are presented as, respectively,
⎧
⎪
⎪ ẋ1 = 3(s(t) – x1 )
⎪
⎪
⎪
⎪
⎪
⎪ ẋ2 = –x1 x3 + rx1 – x2
⎪
⎪
⎪
⎪
⎨ẋ3 = x1 x2 – bx3
⎪
ẋ1′ = 3(x2′ – x1′ ) , (5.7)
⎪
⎪
⎪
⎪ ′ ′
ẋ2 = –x1 x3 + rx1 – x2′
⎪
⎪
⎪
⎪
⎪
⎪ ẋ3′ = x1 x2′ – bx3′
⎪
⎪
⎪
⎩s(t) = km(t) + x′
2
k 1/k
m(t) m΄(t)
s(t)
+ –
x1 y1
xʹ1 yʹ1
x2 y2
xʹ2 yʹ2
x3 y3
xʹ3 yʹ3
Sender Receiver
Figure 5.4: Improved single-channel cascade chaotic secure communication based on

drive–response synchronization.
⎧
⎪
⎪ ẏ1 = 3(s(t) – y1 )
⎪
⎪
⎪
⎪
⎪
⎪ ẏ2 = –y1 y3 + ry1 – y2
⎪
⎪
⎪
⎪
⎨ẏ3 = y1 y2 – by3
⎪
ẏ1′ = 3(y2′ – y1′ ) . (5.8)
⎪
⎪
⎪
⎪ ′ ′
ẏ = –y1 y3 + ry1 – y2′
⎪
⎪
⎪ 2
⎪
⎪
⎪ ′ ′ ′
⎪ẏ3 = y1 y2 – by3
⎪
⎪
⎩m′ (t) = (s(t) – y′ )/k
2
5.2 Chaos Shift Keying Communication

First, in 1993, Parlitz et al. [116] and Dedieu et al. [41] proposed the digital modulation
method of CSK. Soon, Kolumban et al. [117, 118] successively put forward the improved
CSK digital communication system, which mainly includes chaotic on–off keying
(COOK), differential chaos shift keying (DCSK), frequency modulation differential
chaos shift keying (FM-DCSK), quadrature chaos shift keying (QCSK) [119], and so on.
Among these schemes, DCSK and FM-DCSK have better antinoise performance [120].
FM-DCSK was listed as a long-term research program by European Commission. In
2000, IEEE Transactions on CAS-I published the 12th special issue about chaos to dis-
cuss the performance of a class of digital modulation schemes. Compared with DCSK,
QCSK has the same bandwidth and bit error rate (BER) performance, but the data
transmission rate is higher. There are two demodulation methods, such as coherent
demodulation and noncoherent demodulation.
CSK is a kind of chaotic modulation technology for transmitting digital informa-
tion. The basic idea is to map the different information code of binary (or multinary)
digital signal to the piecewise chaotic waveform. The modulated signal is detected in
the receiver based on the correlation reception theory. It needs a chaotic basis func-
tion or multiple chaotic basis functions for CSK modulation. A basis function of CSK
has two types at least, such as COOK and CSK. The antinoise performance of bipolar
coherent demodulation of CSK is better than that of COOK. If the bit energy remains
constant and the chaotic basis function is recovered accurately at the receiver, the
antinoise performance of the CSK is equivalent to that of Binary Phase Shift Keying
(BPSK). The noncoherent detection cannot detect the bipolar CSK, and the antinoise
performance of the noncoherent COOK is worse than that of the coherent COOK, but
the COOK can achieve the best antinoise performance when the bit energy keeps
constant. The decision threshold of the demodulation of reverse-phase CSK and COOK
depends on the signal-to-noise ratio (SNR). When two basis functions are adopted,
this problem can be solved. When symbols “0” and “1” are mapped to two different
chaotic waveforms, the CSK of the two basis functions is formed, and the antinoise
performance of coherent demodulation is better than that of the noncoherent de-
modulation. Especially, through choice of orthonormal basis functions, the decision
threshold level becomes zero, and the performance has nothing to do with the SNR,
and its performance is similar with that of coherent Frequency Shift Keying (FSK).
5.2.1 COOK Communication
COOK is the most simple CSK communication, and the structure of basic principle is
shown in Figure 5.5. {bi } is a binary information signal with “1” and “–1.” When bi = 1,
the switch is turned on. When bi = –1, the switch is turned off. At the receiver, the
information signal is recovered by employing the correlation algorithm. If the input
signal is of equal probability distribution, and the average bit energy for a single sym-
bol to Eb , then the code energies of “1” and “–1” are 2Eb and 0, respectively. Obviously,
the larger the bit energy gap, the better the antinoise performance of the signal. The
structure of this scheme is simple. For example, it only needs a chaotic oscillator, but
the security of the system is not high.
Correlator
Decision
Chaotic ri(t) T zi
oscillator Channel ∫ 0 .dt Decoder
T {bî}
{bi} cp
n(t)
Figure 5.5: System principle block diagram of noncoherent COOK modulation/demodulation.

5.2.2 CSK Communication
Next, we introduce the principle of CSK communication based on two chaotic signal
generators. The structure of the transmitter is shown in Figure 5.6. {bi } is a binary
sequence of information signal, which only includes “1” and “–1”. After modulated
by the modulator, the signal transmitted in the channel is a noise-like chaotic signal,
whose mathematical description is

g1 (t), bi = 1
si (t) = . (5.9)
g2 (t), bi = –1
In the receiver, coherent or noncoherent mode is used to recover the information sig-
nal, and the principle of coherent demodulation is shown in Figure 5.7. The output of
the correlator is
T T T T
zi1 = ri (t)ĝ1 (t)dt = [gi (t) + n(t)]ĝ1 (t)dt = gi (t)ĝ1 (t)dt + n(t)ĝ1 (t)dt, (5.10)
Ts Ts Ts Ts
T T T T
zi2 = ri (t)ĝ2 (t)dt = [gi (t) + n(t)]ĝ2 (t)dt = gi (t)ĝ2 (t)dt + n(t)ĝ2 (t)dt. (5.11)
Ts Ts Ts Ts
Obviously, if bi = 1, then si (t) = g1 (t), ri (t) = g1 (t) + n(t). We obtain zi1 > zi2 , and the
recovered signal is b̂i = 1. If bi = –1, then si (t) = g2 (t), ri (t) = g2 (t) + n(t). We have
zi1 < zi2 , and the recovered signal is b̂i = –1.
Chaotic osillator 1
g1 (t)
si (t) = {gi (t)}
g2 (t) {bi}
Chaotic osillator 2
Figure 5.6: Block diagram of CSK modulator.
Correlator
T zi1
∫ Ts .dt
gˆ 1 (t) T
si (t) ri (t) Chaotic oscillator 1
+ Decoder
– {bî}
Chaotic oscillator 2 gˆ 2 (t)
n(t) T Comparator
∫ Ts .dt
zi2
T
Correlator
Figure 5.7: Principle block diagram of CSK coherent receiver.

Correlator Decision
si (t) zi {bî}
T .dt
∫0 Decoder
T
cp
Figure 5.8: Principle block diagram of CSK noncoherent receiver.
By using the noncoherent demodulation, the demodulation principle is shown in

Figure 5.8, and the output of the correlator is
T T T T T
zi = s2i (t)dt = [gi (t) + n(t)]2 dt = gi2 (t)dt + 2 gi (t)n(t)dt + n2 (t)dt. (5.12)
0 0 0 0 0
For the CSK noncoherent demodulation, the receiver need not recover the chaotic
carrier, and the detector is based on the bit energy of the transmission signal. If the
decision threshold is set to the median of the two bit energies, then the received sig-
nal can be demodulated correctly. Obviously, the selection of the optimal threshold is
closely related to the SNR of the system. So the threshold shift is the main defect of the
noncoherent demodulation.
Next, we take CSK digital secure communication system based on the unified
chaotic system as an example to illustrate, and the system structure is shown in Fig-
ure 5.9. The chaotic systems of the transmitter and receiver are unified chaotic system.
The chaotic system 11 in the transmitting end and the chaotic system 21 in the receiv-
ing end have the same parameter !1 , and the chaotic signal generated by them are
x11 (t) and x21 (t), respectively. The chaotic system 12 in the transmitting end and the
chaotic system 22 in the receiving end have the same parameter !2 , and the chaotic
signals generated by them are x12 (t) and x22 (t), respectively. By employing the syn-
chronization strategy, the unified chaotic systems 11 and 12 at the sending end are
synchronized with the unified chaotic systems 21 and 22 at the receiving end, respect-
ively, then we have x11 (t) = x21 (t), x12 (t) = x22 (t). The binary digital signal a(t) = {an } is
used to control the chaotic signal. When a(t) is of high voltage, x11 (t) is sent. When a(t)
is of low voltage, x12 (t) is sent. The transmitted signal in the channel is expressed as

x11 (t), a(t) = 1
s(t) = x11 (t) ⋅ a(t) + (1 – a(t)) ⋅ x12 (t) = . (5.13)
x12 (t), a(t) = 0
Obviously, the signal transmitted in the channel is chaotic. If the noise in the channel
is Gaussian white noise n(t), then the signal at the receiving end is expressed as

x11 (t) + n(t), a(t) = 1
r(t) = s(t) + n(t) = x11 (t) ⋅ a(t) + (1 – a(t)) ⋅ x12 (t) + n(t) = (5.14)
x12 (t) + n(t), a(t) = 0
Z1
α1 Chaotic
x11(t) ʃT dt Sampling
system 11
a(t) x21 (t)
α1 Chaotic
system 21
s(t) r(t) b(t)
Channel Quenching Sampling
Decision
pulse pulse
α2 Chaotic
system 22
Chaotic n(t) x22 (t)
α2 Z2
system 12
x12(t) ʃT dt Sampling
Figure 5.9: Principle block diagram of CSK digital secure communication system based on unified
chaotic system.
For bipolar CSK, due to the correlation integral, two signal energies cannot be dis-
tinguished, so we can only use coherent demodulation method to demodulate the
signal.
The chaotic signals are broadband nonperiodic signals, so the chaotic signals
obtained by the same chaotic system with different initial conditions or gener-
ated by the same system with different system parameters are almost orthogonal.
The correlation properties of chaotic signals are similar to those of Gauss white
noise. Generally, the cross-correlation of the chaotic signal approximates to zero,
and the autocorrelation approximates to the impulse function. The correlation prop-
erties of chaotic signals generated by the unified chaotic system are shown in
Figure 5.10.
The receiver is composed of a chaotic synchronization system, a correlator (in-
cluding multiplier and integrator), and a sampling decision device. If the outputs of
correlator 1 and correlator 2 are z1 and z2 , respectively, and their outputs are
1 1
Normalization coefficient ρ
Normalization coefficient ρ
0.5 0.5
0 0
–0.5 –0.5
–1 –0.5 0 0.5 1 –1 –0.5 0 0.5 1
Delay time τ/s Delay time τ/s
(a) (b)
Figure 5.10: Correlation characteristic of chaotic signals generated by the unified chaotic system:
(a) autocorrelation and (b) cross-correlation.

T (x1 (t) + n(t))x21 (t)dt = Eb + T n(t)x21 (t)dt, a(t) = 1
z1 = r(t)x21 (t)dt = , (5.15)
T T (x2 (t) + n(t))x21 (t)dt = 1Eb + T n(t)x21 (t)dt, a(t) = 0

+ n(t))x22 (t)dt = 1Eb + T n(t)x22 (t)dt,
T (x1 (t) a(t) = 1
z2 = r(t)x22 (t)dt = , (5.16)
T T (x2 (t) + n(t))x22 (t)dt = Eb + T n(t)x22 (t)dt, a(t) = 0

where 1 = T x11 (t)x12 (t)dt /Eb is the correlation coefficient of chaotic signals. When
2 2
the two signals are completely orthogonal, 1 = 0. Eb = T x11 (t)dt = T x12 (t)dt is the bit
energy of the chaotic signal. Obviously, if the correlator output signals, z1 and z2 , are
sampled to be decided, then the original signals can be recovered by the decision rules

z1 > z2 decision for “1”
b(t) = . (5.17)
z1 < z2 decision for “0”
Obviously, ignoring the effect of noise, the noise tolerance of coherent demodulation
in the receiver system is Nc = (1 – 1)Eb . The output signal of the decision device is
b(t). It can be seen that the demodulation performance of the system is related to the
characteristics of the chaotic signal. Because the autocorrelation of the chaotic signal
generated by the unified chaotic system with different parameters is close to $(t), and
its cross-correlation is very weak, the system has a good correlation demodulation
performance and the original signal can be recovered, that is b(t) = a(t).
Next, we discuss the BER performance and security performance of CSK secure
communication based on unified chaotic system. Let the two possible deterministic
signals arriving at the receiver be x1 (t) and x2 (t), respectively, and their durations are
(0, T), and have the same energy Eb . The noise n(t) at the input of the receiver is Gauss
white noise with zero mean, and its unilateral power spectral density is n0 . According
to the principle of the best reception, in the case of equal priori probabilities, the limit
BER performance is
⎡ ⎡ ⎤⎤
1⎣ E (1 – 1)
1 – erf ⎣ ⎦⎦ ,
b
Pe = (5.18)
2 2n0
where erf(x) is the error function, and 1 is the cross-correlation coefficient between
signals x1 (t) and x2 (t), and its range is (–1, 1). Because x1 (t) and x2 (t) are orthogonal
signals, the cross-correlation coefficient is zero. The BER performance of the system is
rewritten as
⎡ ⎡ ⎤⎤
1⎣ ⎣ E b ⎦⎦ 1 Eb
Pe = 1 – erf = erfc . (5.19)
2 2n0 2 2n0
The BER simulation result of the CSK system is shown in Figure 5.11. When the SNR is
greater than 10 dB, the BER is less than 10–6 .
100
BER Pe
10–10
10–20
0 5 10 15
SNR /dB
Figure 5.11: BER performance of the CSK system.
The security of CSK communication system is good, and it is mainly manifested in the
following aspects. First, the system has strong anticrack ability, because the signal
transmitted in channel is chaotic. With the evolution of the chaotic system, different
chaotic signals represent the same two value signals. It is equal to “one time pad,” so
it is unbreakable. From the view of cryptography, the system has a huge key space. The
chaotic system is sensitively dependent on initial conditions. The unified chaotic sys-
tems with different parameters and different initial conditions can produce different
chaotic signals, so chaotic system parameters and initial value can be used as keys.
Different chaotic dynamical equations can produce different chaotic signals, and dif-
ferent chaotic variables also can be modulated. In addition, different synchronous
mode also affects the demodulated signal. So the key space of the system is composed
of parameters, initial values, chaotic systems, chaotic variables, and synchronization
modes. The larger the key space is, the higher the security performance is.
It is worth noting that the good anti-interference and security of the system is at
the cost of the complexity of the system. In practice, we should consider the trade-off
between these two aspects. Because chaotic signal is nonperiodic, the bit energy of
output information signal is not entirely consistent. Even in the noise-free case, there
is also a deviation for the results related to the estimation, so it will affect the BER per-
formance of the system. To solve this problem, we can use the frequency modulation
(FM) to modulate CSK, namely FM-DCSK modulation.
According to the principle shown in Figure 5.9, the simulation model is obtained
as shown in Figure 5.12. If the parameter ! = 1 for the chaotic system 11 and the chaotic
system 21, then initial values are (–5, –5, –5). At the same time, if the parameter ! = 0.3
for the chaotic system 12 and the chaotic system 22, then initial values are (10, 10, 10).
The chaotic variable x is modulated by the information signal a(t). When a(t) = 1, the
modulated signal s(t) transmitted in the channel is the chaotic signal x11 generated
by the unified system 11. When a(t) = 0, the modulated signal s(t) transmitted in the
channel is the chaotic signal x12 generated by the unified system 12. By using the chaos
synchronization control strategy, the synchronization between the unified chaotic sys-
tems is achieved as shown in Figure 5.13. Bernoulli random binary sequence is used as
x 1
S/H
1 Parameter y s
z Product
Integrator Sample
Constant1
Unified chaotic system 1 x and hold
1 Parameter y
z
Bernoulli bin AWGN Constant2
Unified chaotic system 2
Bernoulli Random Switch
AWGN x Pulse Relational Scope1
Binary Generator Channel 0.3 Parameter y generator operator
z
x Constant4
0.3 Parameter y
z 1
Constant3 S/H
s
Product1 Integrator1 Sample
and hold1
Figure 5.12: Simulation model of CSK digital secure communication based on unified chaotic system.
40 40
20 20
x21 0 x22 0
–20 –20
–40 –40
–40 –20 0 20 40 –40 –20 0 20 40
x11 x11
(a) x11(t)~x21(t) (b) x11(t)~x22(t)
Figure 5.13: Synchronization phase diagram of the variables between the sender and receiver:
(a) x11 (t)–x21 (t) and (b) x11 (t)–x22 (t).
the information source. The integrator is triggered by the rising edge of external clock
pulse. The sampling–holder is triggered by the falling edge of external clock. Decision
is realized by the relevant logic comparator. The simulation results are shown in Fig-
ure 5.14, where T(t) is the system clock signal; a(t) is the modulation signal; s(t) is the
modulated signal transmitted in channel; and b(t) is the recovered information signal
at the receiver. Obviously, b(t) = a(t), that is, chaos signal demodulates correctly. Sim-
ulation results show that the original binary digital signal can be demodulated in this
CSK secure communication scheme correctly.
5.2.3 DCSK Communication
In 1996, Kolumban et al. [117] proposed the DCSK scheme, and it solved the problem
of CSK decision threshold value depending on the SNR. The basic principle is that
2
Clock signal
T(t) 1
0
0 10 20 30 40 50 60
2
Modulating signal
a(t)
0
0 10 20 30 40 50 60
20 Modulated signal
s(t)
–20
0 10 20 30 40 50 60
2
Recovery signal
b(t)
0
0 10 20 30 40 50 60
t/s
Figure 5.14: Simulation results of the CSK secure communication system.
each digital symbol is represented by two chaotic segments signal. The first segment
is called “reference signal,” and the second segment is named “useful signal.” When
we send “1,” two same segments are sent (same phase). When we send “0,” the two
inverse phase segments are sent (inverse phase). In the receiver, the correlation of
these two segments is calculated, and the information signal is recovered according
to the correlation value. When the output of the correlator is more than 1, the output
signal is “1.” When the output of the correlator is less than 1, the output signal is “0.”
The principle of DCSK modulator is shown in Figure 5.15.
1
T/2
2 si (t)
Chaos generator
T/2
3
–1 {bi}
Figure 5.15: Principle block diagram of the DCSK modulator.

Correlator
Decision device
ri(t) = si(t) + n(t) Zi
T {bˆ i}
–dt Decoder
T/2
T
T/2 cp
Figure 5.16: Principle block diagram of the DCSK coherent demodulation.
From Figure 5.15, the signal transmitted in channel is presented as

x(t), tk ≤ t ≤ tk + T/2
si (t) = for bi = 1, (5.20)
+x(t – T/2), tk + T/2 < t ≤ tk + T

x(t), tk ≤ t ≤ tk + T/2
si (t) = for bi = 0. (5.21)
–x(t – T/2), tk + T/2 < t ≤ tk + T
Because each bit is mapped to two continuous segments with the length T/2, the
information signal at the receiver can be recovered by coherent demodulation. The
demodulation principle of the DCSK scheme is shown in Figure 5.16. In this demodu-
lation scheme, the output signal of the correlator is
T T
zi = ri (t)ri (t – T/2)dt = [si (t) + n(t)][si (t – T/2) + n(t – T/2)]dt
T/2 T/2
T T T
. (5.22)
= s2i (t)dt + si (t)(n(t) + n(t – T/2)dt + n(t)n(t – T/2)dt
T/2 T/2 T/2
Compared with CSK and COOK, the advantage of the DCSK scheme is that the de-
cision threshold is zero whether there is noise or not. The reference signal and carrying
information signal go through the same channel, so it is not sensitive to channel dis-
tortion, and it has a good antinoise performance. The shortage of DCSK is that it can
only be used in the chaotic digital communication with low bit rate. To solve this
problem, in 1998, Kolumban et al. proposed a DCSK scheme based on FM-DCSK [118].
5.2.4 FM-DCSK Communication
FM-DCSK is constituted by DCSK and a frequency modulator. The output of FM is a

chaotic signal with band-limited and uniform power spectrum density. The structure
of FM-DCSK modulator is shown in Figure 5.17. It employs the chaotic analog phase-
locked loop (APLL) to generate band-pass chaotic signal, and then FM is applied
for chaotic signals. The subsequent parts are consistent with DCSK, which transmits
1 T/2
Reference
si(t)
frequency 2
PD F(s) FM T/2
fR K
3
–1 {bi}
VCO
Figure 5.17: Principle block diagram of FM-DCSK modulation.
Correlator
Decision device
FM-DCSK T Zi {b̂i}
–dt Decoder
ri(t) T/2
T
T/2 cp
Figure 5.18: Principle block diagram of FM-DCSK demodulation.
reference signal in [0, T/2] and transmits same phase chaotic FM signal or inverse
phase chaotic FM signal in [T/2, T].
The demodulation principle of the FM-DCSK is shown in Figure 5.18. Similarly
with that of DCSK, it employs the correlation receiving principle to demodulate the
received information {bi }. For noise channel, the output of the FM-DCSK correlator has
zero variance. Its data rate is not limited by the characteristics of the chaotic signal,
and its decision threshold is identically zero. Thus, FM-DCSK can improve the symbol
rate and reduce errors. When APLL is designed correctly, we can obtain the chaotic
band-limited signal with uniform power density, which makes the output (bit energy)
of the receiver is constant, regardless of the data rate. Thus, FM-DCSK has higher data
rates, and its antinoise performance is the same with that of DCSK.
When compared with other chaotic modulation schemes, FM-DCSK and DCSK
have better robustness to multipath interference and nonideal channel. FM-DCSK
modulation is suitable for wireless local area networks, indoor wireless commu-
nications, mobile communications, and others which are sensitive to multipath
interference.
5.2.5 QCSK Communication
QCSK is a kind of chaotic communication scheme, which is similar to the conven-

tional quadrature phase modulation. When compared with DCSK, it has the same
bandwidth, approximate BER characteristics, but higher data transmission rate. The
basic principle of QCSK is to generate the orthogonal chaotic signal at the specified
time interval. In QCSK, the linear combination of two chaotic basis functions is used
for coding four symbols. QCSK can be regarded as a combination of two DCSK systems.
A double data rate is obtained by increasing complexity of the QCSK.
To explain the principle of QCSK, first, the concept of orthogonal chaotic sig-
nal is presented. If two independent chaotic signals c1 (t) and c2 (t) (or even different
segments of the same chaotic signal) exhibit very low cross-correlation, they are
orthogonal in a time interval [0, 4], namely
4
c1 (t)c2 (t)dt = 0. (5.23)
0
If x(t) is a chaotic reference signal defined in the interval t ∈ [0, 4], and it has a zero
mean. In the interval t ∈ [0, 4], x(t) can be expanded as follows by Fourier series
(A0 = 0):
∞

x(t) = Ak sin(k9t + >k ), (5.24)
k=1
where 9 = 20/4. Correspondingly, the average power of x(t) in the interval t ∈ [0, 4] is
∞
1 4
1 2
px(t) = x2 (t)dt = Ak . (5.25)
4 0 2
k=1
The orthogonal signal y(t), in the interval t ∈[0, 4], can be obtained by each Fourier
frequency component of x(t) with the phase shift 0/2, namely
∞

y(t) = Ak sin(k9t + >k – 0/2), (5.26)
k=1
Then the signals x(t) and y(t) are orthogonal over the interval t ∈ [0, 4] and their power
is equal, i.e.

1 4
x⊥y⇔ x(t)y(t)dt = 0, (5.27)
4 0
4
1 4 2 1
px(t) = py(t) ⇔ x (t)dt = y2 (t)dt. (5.28)
4 0 4
0
The above properties are equivalent to

4
2 Ak
1 2
1 cos(! + "), k=m
Ak sin(k9t + >k – !)Am sin(m9t + >m – ")dt = . (5.29)
4 0 0, k≠m
According to definition formula (5.26) of the reference quadrature signal y(t), the x(t)
and y(t) are extended into the periodic signal with the periodic 4. It is found that y(t)
is the Hilbert transform of x(t), and the Hilbert transform of the given signal can be
calculated by the following methods:
1. Frequency domain method: It is basically the same as the method of generating
Hilbert transform in Amplitude Modulation - Single Side Band modulation. A fast
Fourier transform (FFT) is applied to a chaotic sequence with length k, and then
remove the negative frequency (signal energy) corresponding to FFT coefficients.
It is obtained by inverse FFT transformation. At present, the algorithm can be
realized by a special digital signal processing (DSP) chip.
2. Time domain method: It is achieved by the design of digital Hilbert converter with
finite impulse response (FIR) or infinite impulse response (IIR). A simple method
for calculating the Hilbert transformation of a given discrete signal is to design
an FIR filter which is approximated by Hilbert transform. The FIR filter can be
implemented by the ideal impulse response with appropriate width:

2 sin2 (0k/2)
0 k , k≠0
h(k) = . (5.30)
0, k=0
In the interval I = [0, 4], a chaotic sampling orthogonal base function is defined as

1 1
c = [cx (t), cy (t)] = √ x(t), √ y(t) , (5.31)
E4 E4
where E4 = Px 4 is the energy of x or y over the interval I. So we have

4 4
c2x (t)dt = c2y (t)dt = 1. (5.32)
0 0
The orthogonal condition (5.32) indicates that the energy of cx and cy is constant
at each interval I. The constant is one in special case. The chaotic signal can be
represented by a linear combination of cx and cy , namely ms (t) = as cx (t) + bs cy (t),
where s represents a signal space symbol. Equivalently, the symbol s can be
4
expressed in plural form, namely, ms = as + jbs and 0 m2s (t)dt = 1.
Being similar to DCSK and QCSK, the modulator structure is shown in Figure 5.19. It
includes a chaotic generator, which generates the signal cx at each interval [0, T/2],
and the corresponding quadrature signal cy is generated by the filter. For simplicity,
it is assumed that the system has no time delay. According to the symbol s, the en-
coder generates a linear combination of cx and cy . Information symbol is provided by
a bit/symbol converter, and the converter maps each pair of information bits into a
symbol. Thus, each symbol period transmits 2 bits information. Then, the QCSK sig-
nal is formed by the two-channel analog multiplexer with the switching time T/2. Note
that although there is no filter in the figure, the QCSK scheme must contain a low-
pass band-limiting filter to limit the bandwidth of transmission signal. The chaotic
d(k) Bit/symbol converter
cx
T/2
sQCSK
cx
Chaotic signal generator T/2
Encoder
ms = ascx + bscy
cy
Hilbert wave filter
Figure 5.19: Principle block diagram of QCSK modulator.
reference signal segment r(t) = Eb cx (t) is transmitted in the first half symbol period of
the transmission symbol s, and the information segment i(t) = Eb ms (t) is transmitted
in the second half of the symbol period, where s represents the symbol that needs to
be transmitted, while Eb represents the energy per bit. The orthogonal condition (5.32)
shows that the energy per transmitted bit is a constant. If the symbol period is T = 24,
then the QCSK signal is
√
Eb cx (t), 0 ≤ t < T/2
SQCSK (t) = √ , (5.33)
Eb (as cx (t – T/2) + bs cy (t – T/2)), T/2 ≤ t < T
where s = 0, 1, 2, 3.
The simplified block diagram of QCSK demodulator is shown in Figure 5.20. QCSK
signal may be demodulated by the method of differential coherent detection. During
the process of demodulation, the symbol period is assumed to be synchronized. That
is, the beginning of each symbol is taken as the starting of the coherent process at the
receiver.
A
~
ms T za
ʃ T/2
–dt
~
~ d (k)
cx
T/2 Decision Bit/symbol
module converter
~
ms T T
~ ~
ʃ T/2
–dt
zb
sQCSK cy
B
Hilbert wave
filter
Figure 5.20: Principle block diagram of QCSK demodulation.

In theory, by using the chaos basis signals cx and cy , carrier information segment i(t) =
Eb ms (t) is processed in the interval [T/2, T], and we can recover the plural ms = as +jbs .
In fact, according to formulas (5.27) and (5.28), we can obtain
T/2
as = ms (t)cx (t)dt, (5.34)
0
T/2
bs = ms (t)cy (t)dt. (5.35)
0
At the receiver, we can observe noise, filtered reference signal r(t), and information
segment i(t). If the signal is only affected by the additive Gaussian white noise, then
the input of the correlator A in Figure 5.20 is

r̃(t – T/2) = Eb (cx (t – T/2)) + n(t – T/2), t ∈ [T/2, T], (5.36)

ĩ(t – T/2) = Eb [as cx (t – T/2) + bs cy (t – T/2)] + n(t), t ∈ [T/2, T]. (5.37)
The estimated value of the quadrature signal H[r̃] is generated by the disturbed chaotic
reference signal, where H is the Hilbert transform operator:

H[r̃(t – T/2)] = H[ Eb cx (t – T/2) + n(t – T/2)] = Eb cy (t – T/2) + n′ (t – T/2), (5.38)
where n′ (t) = H[n(t)]. With the ideal Hilbert transform, the statistical properties of n′ (t)
are the same as that of n(t).
Considering the orthogonality of cx and cy , the output of the correlator is
T
T
T T T
za = r̃ t – ĩ(t)dt = as Eb + Eb aS cx t – n t– dt
T
2
2 T
2
2 2
T T T ,
T T T
+ Eb bS cy t – n t– dt + Eb cx (t)n(t)dt + n(t)n t – dt
T
2
2 2 T
2
T
2
2
(5.39)
T T
T T ′ T
zb = H(r̃ t – ĩ(t)dt = bs Eb + )n t – Eb asdt cy (t –
T
2
2 2 2 T
2
T T T .
T ′ T ′ T
+ Eb bs cy t – n t– dt + Eb cy (t)n(t)dt + n(t)n t – dt
T
2
2 2 T
2
T
2
2
(5.40)
CSK belongs to chaotic digital secure communication. When the basis function is
strictly recovered, the noise performance of the coherent reception is better than that
of the noncoherent reception. If the orthogonal basis function is chosen, the chaotic
keying and the traditional digital modulation can achieve similar noise performance.
However, at present, it is difficult to meet the requirements of the strict recovery

basis function, and it is more difficult to regenerate the chaotic basis function than to
regenerate the periodic basis function. As a result, the performance of the coherent
chaotic keying communication is slightly less than that of the same kind of tradi-
tional digital communication. But CSK belongs to broadband communication, and
the performance of chaos communication shows more superiority in multipath com-
munication environment. Among CSK communication schemes, QCSK and FM-DCSK
have better performance and application prospect.
5.3 Chaos Parameter Modulation

Chaos parameter modulation is also known as wide spectrum transmission. The basic
idea is that an information signal is injected into the transmitter, and thus it changes
the dynamic characteristics of the original chaotic system. So the information signal
is modulated. Compared with CSK and chaotic masking, chaos parameter modula-
tion has the following advantages. First, it employs the whole range of the chaotic
signal spectrum to hide information. Second, it increases the sensitivity to parameter
variation, thereby enhancing confidentiality.
For CSK technology, no matter when the value of binary information signal
changed, there is always a synchronization waiting time because of the difference
of both initial conditions. While for chaos parameter modulation, receiver tracks
transmitter continuously, and two chaotic system states would surely be the same.
Compared with chaotic shift keying, it is expected to get a higher bit rate. In addi-
tion, with further development of chaos modulation technology, the direct chaotic
spectrum expansion technique is also used as a chaos modulation technique.
Research results show that the main drawback of chaos communication scheme
is being susceptible to channel noise and distortion. The traditional masking com-
munication requirements of the power information signal are far lower than that
of chaotic signal, which makes chaotic masking communication more difficult to
be used in practice. A new method of chaotic pulse position modulation (CPPM) is
proposed in Ref. [121], which shows good performance in the case of channel distor-
tion. This scheme is based on the chaotic pulse sequence, and the interval between
the pulses is determined by the chaotic characteristics of the nonlinear function.
The pulse sequence of chaotic pulse interval modulation is used as a carrier. Binary
information is modulated onto a carrier by pulse position modulation (PPM). As a
result, each pulse is unchanged or delayed some time depending on the transmission
information “0” or “1.” The pulse duration of “0” or “1” is estimated by chaotic pulse
sequence synchronization at the receiver, and thus the transmission information is
demodulated in this way.
The block diagram of chaotic pulse position modulator is shown in Figure 5.21.
The chaotic pulse signal generated by chaotic pulse generator is expressed as
Comparator
Integrator S&H F(x)
V(t)
Delay modulation Pulse generator1
Pulse generator 2 Output Binary source
Figure 5.21: Principle block diagram of chaotic pulse position modulator.
∞

U(t) = W(t – tj ), (5.41)
j=0
where W(t – tj ) represents a pulse waveform generated at time tj , and tj is expressed as
j

tj = t0 + Tn , (5.42)
n=0
where Tn is the time interval between the nth pulse and the (n–1)th pulse. The time
interval sequence Ti represents the iteration of the chaotic process. For the design
convenience, we consider the chaos is generated by a one-dimensional map Tn =
F(Tn–1 ), where F(⋅) is a nonlinear function.
The information is encoded in the chaotic pulse signal by adding the delay time
in the pulse interval Tn . Thus, the resulting pulse sequence can be defined as
Tn = F(Tn–1 ) + d + mSn , (5.43)
where Sn is the information signal. Only binary information is considered here, so Sn

is equal to 0 or 1. m represents the modulation degree, and the parameter d is the time
delay constant for the modulation and demodulation. When we design chaotic pulse
generator, the choice of the nonlinear function F(⋅), and parameters d and m should
ensure the chaotic performance of the map.
The transmission signal is a CPPM signal in the channel, which is denoted by
⎛ ⎞
∞
j

U(t) = W ⎝t – t0 – Tn ⎠, (5.44)
j=0 n=0
where Tn is generated by formula (5.43), and we assume that the duration of each
pulse W(t) of the pulse sequence is much less than the minimum value of the pulse
interval Tn . When detecting information at the receiver, the demodulator is triggered
by receiving pulse U(t). After measuring successive time intervals Tn and Tn–1 , the
information signal is recovered as
Sn = (Tn – F(Tn–1 ) – d)/m. (5.45)
If the nonlinear function F(⋅), parameters d and m at the receiver are same with
those of the sender, then the encrypted information Sn can be easily recovered. When
the matching accuracy of nonlinear functions is not sufficient, large decoding error
occurs.
The demodulator principle is shown in Figure 5.22. When the demodulator is
synchronized with the pulse modulator, to detect a single bit of the transmission in-
formation, the demodulator must simply determine whether the desired position of
the transmitted pulse is delayed. In the figure, the integrator, sample and hold circuit,
and a nonlinear threshold function module are reset or triggered by the pulse from the
transmitter, not by the feedback loop within the receiver. To be exact, only when the
input signal U(t) from the input channel exceeds the threshold, they are triggered. For
each received pulse, the result of V(tn+1 ) – F(vn ) is calculated and used to determine
whether the pulse is delayed. Assuming that the difference is less than the reference
value "(d + m/2), the data bit Sn+1 is determined to be “1,” otherwise, it is determined
to be “0.”
Another important part of the receiver is the window selection module. Once the
receiver has correctly received two consecutive pulses, the receiver can predict the
fastest time to receive the next pulse. It means that the input of the demodulator can be
Subtractor
Output
Comparator
Integrator S&H F(x) S&H
Window selection
Threshold pulse
modulator
Modulated pulse signal input
Figure 5.22: Principle block diagram of the chaotic pulse position demodulator.
masked at this time, which is done by the window selection module. In the experiment
circuit, the input of the receiver is opened by a window control pulse generated by the
comparator at the moment tn+1 ′ = tn + "F(vn ), and the input end is turned on until the
demodulator is triggered by the first received pulse. So the window selection module
greatly reduces the probability of the receiver triggered by the noise, interference, or
the pulse of other users.
Although the performance of CPPM is slightly worse than that of BPSK, noncoher-
ent FSK, and ideal PPM, it has the following characteristics:
1. Low probability intercept and low probability decoding ability.
2. Although the complexity of the circuit is increased slightly, the security of com-
munication system is improved.
3. Compared with the majority of the communication schemes based on chaotic
transformation, the system presents very good performance.
4. CPPM can be used for multiuser communication.
5. Compared with other pulse system, it eliminates any traces of periodicity of the
transmission signal spectrum due to its independent period clock.
5.4 Chaotic Spread Spectrum Communication

Chaotic spread spectrum communication is divided into direct sequence spread spec-
trum (DS/SS) and frequency hopping spread spectrum (FH/SS). The principle is to
use the chaotic sequence as the pseudorandom sequence. Common pseudorandom
sequences include m sequence, Gold sequence, and Bent function sequence, but these
sequences have a certain periodicity, limit numbers, and poor anti-interception ability.
Chaotic sequences are nonperiodic sequences, which have the statistical properties of
approaching Gauss white noise. The chaotic sequences generated by different systems
or with different phases are different sequences.
5.4.1 Principle of Chaotic DS/SS Communication
A chaotic DS/SS communication system is built instead of pseudorandom codes with

the chaotic sequences [122], and its principle diagram is shown in Figure 5.23.
Here, assume each user is synchronized, spread spectrum codes are synchron-
ized, and the signal power of each user is same. If there are M users working at the
same time totally, and the message data of each user is ai (k) (i = 1, 2, . . . , M), the
original data signal of the ith user is expressed as
KN+(N–1)

ai (k) = di (n), (5.46)
n=KN
where di (n) is the information bit value of the ith user, di (n) ∈ { – 1, 1}. Let xi (n) be the
chaotic spread spectrum sequence of the sender of the ith user, and its expression is
s1(k) y1(n)
a1(k) ʃT Decision b1(k)
N(t)
x1(n) f (ω0, 1) f (ω0, 1) x 1(n) cp
s2(k) s(n) r(n) y1(n)
a2(k) Channel ʃT Decision b2(k)
.
. x2(n) f (ω0, 2) f (ω 0 , 2) x 2(n) cp .
. . . . . . . .
. . . . . . .
. . . . . .
sm(k) y1(n)
am(k) ʃT Decision bm(k)
xm(n) f (ω0, f (ω 0, m) x m(n) cp

m)
Figure 5.23: Principle block diagram of multiuser chaotic sequence spread spectrum communication
system.

xi (n) = KN+(N–1)
n=KN gi (n), where gi (n) is the spread spectrum characteristic sequence of
the ith user.
The input signal r(n) of the receiver is expressed as
M

r(n) = Ak (n)dk (n)gk (n) + N(n), (5.47)
k=1
where Ai (n) is the carrier signal amplitude of the ith user. N(n) is the zero mean Gauss
white noise. So, the output of the correlator of the receiver of the ith user is
KN+(N–1)

yj (k) = r(n)xi′ (n), (5.48)
n=KN
where xi′ (n) is the chaotic spread spectrum sequence of the receiver of the ith user, and
the system synchronizes between the sender and the receiver, that is, xi (n) = xi′ (n). Let
Tb be the time width of each original information bit, then
K

1 1
yi (k) = r(t)gi (t)dt = Ai di + 1i,k Ai di + n(t)gi (t)dt
Tb Tb Tb Tb
i=1
i≠k
= Ai di + MAIi + Zi (5.49)
For formula (5.49), the first item is the desired information component. The second
item is the recovery data obtained by the received signal with the kth signal itself, and
multiple access interference (MAI) caused by other users. Obviously, the MAI term
is related to the user number k, the amplitude of the signal of other user, and the
T
cross-correlation coefficient 1i,k , where 1i,k = T1 0 b gi (t)gk (t)dt. The third item is noise
b
interference.
After the integral (correlation operation) of the receiver for the ith user, the signal
is decided by the decision threshold, which is zero. The data is demodulated by the
ith user at the receiver according to

–1, yi (k) < 0
bi (k) = . (5.50)
1, yl (k) > 0
To sum up, whether the user information is correctly recovered is greatly related with
the characteristics of the spread spectrum code.
5.4.2 Generation of Chaotic Spread Spectrum Code
Taking three typical chaotic maps, including logistic, Chebyshev, and Tent map as
examples, the binary chaotic spread spectrum sequence is generated [123]. In the mul-
tiuser chaotic sequence spread spectrum communication system, spread spectrum
sequence is a binary sequence with value ±1, namely, the real values of chaotic map
sequences are quantized into binary to generate spread spectrum sequence.
5.4.2.1 Logistic Chaotic Map

Logistic map equation is
xn+1 = rxn (1 – xn ), 0 < xn < 1, (5.51)
where r is the system parameter, and r ∈ (0, 4]. When 3.5699 . . . < r ≤ 4, the system is
chaotic. The mean value of the sequence is 0.5.
5.4.2.2 Chebyshev Chaotic Map

The 9-order Chebyshev chaotic map is defined as
xk+1 = cos(9 cos–1 xk ), –1 ≤ xk ≤ 1, (5.52)
where 9 is usually taken as an integer power of 2, and the mean value of the
sequence is 0.
5.4.2.3 Tent Chaotic Map

The Tent chaotic map is defined as

xk /a, 0 < xk < a
xk+1 = , (5.53)
(1 – xk )/(1 – a), a ≤ xk ≤ 1
where 0 < a < 1, and the mean value of the sequence is 0.5.
Next, we discuss the balance and correlation of the three kinds of chaotic
sequences.
5.4.2.3.1 Balance of the Sequences

The balance of a sequence is determined by the statistical distribution about “1” and
“–1” in the sequence. If P and Q denote the number of “1” and “–1” in the chaotic
sequence, respectively, then the balance degree of the sequence E is defined as
E = (P – Q)/N, (5.54)
where N is the total number of symbol sequence. The sequence balance is related with
the carrier suppression degree of the DS/SS system. The imbalance of spread spectrum
sequence will lead the leakage of carrier information, or be prone to error or lost. Thus,
it is important to study the balance of the direct spread spectrum system (DS/SS).
Applying formula (5.54) to analyze the impact of the initial value of three kinds of
chaotic maps, the results are shown in Figure 5.24. When r = 4 for the logistic map,
the initial value x0 is one of the 9,999 different numbers between 0 and 1. From Fig-
ure 5.24(a), we can obtain that the effect of x0 on the balance is small, but there are
three peaks, and the initial value x0 = 0.25, 0.5, and 0.75, respectively. When 9 = 20 for
the Chebyshev map, the initial value x0 belongs to one of the 9,999 different numbers
between 0 and 1. From Figure 5.24(b), we found that the effect of x0 on the balance is
very small, and there are no peaks. When a = 0.49 for the Tent map, the initial value
x0 forms the 9,999 different numbers between 0 and 1. From Figure 5.24(c), we find
that the effect of x0 on the balance is small, and there are three peaks, and the ini-
tial value x0 = 0.2401, 0.49, and 0.7501, respectively. It is worth pointing out that the
peak point of Tent map is related with the initial value selection. If the accuracy of the
initial value is not enough, the peak of this map may be missed.
5.4.2.3.2 Correlation of the Sequence

From formula (5.49), in the case of same SNR, and without consideration of multipath
interference, MAI becomes the major factor in the transmission performance of the
system. Root mean square (RMS) of cross-correlation of the sequence characterizes the
size of the MAI of the spreading sequence, so RMS of a sequence is important for ana-
lyzing the transmission performance. As a result, the actual simulation of the chaotic
1 1 1
Balance degree E
(a) (b) (c)

Balance degree E
Balance degree E
0.5 0.5 0.5
0 0 0
–0.5 –0.5 –0.5
–1 –1 –1
0 0.25 0.5 0.75 1 0 0.5 1 0 0.2401 0.5 0.7501 1
Initial value x0 Initial value x0 Initial value x0
Figure 5.24: Relationship between the balance E and the initial value x0 , for different maps:
(a) Logistic map; (b) Chebyshev map; and (c) Tent map.
sequence has a deviation with the theoretical value. Here, we change the length of
the chaotic sequence N and maintain the relevant range M, to study the effect of the
length of the sequence on the deviation. The RMS 3R12 is defined as

M
1
3R12 = [R12 (m)]2 . (5.55)
2M + 1
m=–M
Here, we use formula (5.55) to calculate the RMS values of the cross-correlation of the
sequence with length N = 2,000 generated by the above three kinds of chaotic maps.
The results are shown in Table 5.2.
From Table 5.2, after the frame processing, RMS value of cross-correlation for
each chaotic sequence has been greatly affected, and its change is different. Besides,
different frame lengths correspond to different RMS values, which destroy the ori-
ginal relevant characteristics. From Table 5.2, after the frame processing, the cross-
correlation of three chaotic sequences becomes smaller, and it also shows that the
chaotic sequence with a certain length has a better cross-correlation compared with
the original sequence. But there is no fixed relationship between the frame length and
the RMS value, that is to say, after the frame processing, chaotic sequence destroys
the original conclusion. That is, the longer the sequence, the smaller the deviations of
correlation.
5.4.3 Simulation Module Design of Multiuser Chaotic Spread Spectrum System
Multiuser chaotic spread spectrum system involves the design of multiple modules,
including the spread spectrum sequence generator module, the correlation receiving
module, the multipath channel module, the polarity transformation module, and the
frame sampling module. The next sections introduce each of these.
5.4.3.1 Spread Spectrum Sequence Generator Module

The chaotic spread spectrum code is generated by Chebyshev, improved logistic, lo-
gistic, Tent, and other iterative chaotic maps. In addition, the real value is transformed
into the binary sequence by setting a decision threshold . , and the real value is
Table 5.2: RMS values of three chaotic sequences.
Frame length Chebyshev Tent Logistic
64 0.2450 0.2769 0.2783

950 0.2034 0.2292 0.2306
1,000 0.2031 0.2300 0.2307
Processing without frames 0.2488 0.2816 0.2824
1
z̅
Unit delay ++ 1
0.5
Out1
Sign Gain
f(u)
a 0.5
Fcn
Constant Constant 1
Figure 5.25: Simulation model of chaotic sequence generator based on Tent map.
converted by the decision function T(x) = [1 + sgn(x – . )]/2. Here, two design methods
of the chaotic sequence generator module are introduced. One is the design of spread
spectrum sequence generator based on Tent sequence, and the other is the design of
spread spectrum sequence generator module based on series chaotic map.
The simulation model of the spread spectrum sequence generator module based
on Tent chaotic sequence is shown in Figure 5.25. In the figure, the initial value and
the sampling time of the whole sequence generator are controlled by the initial value
and the sampling time of the unit delay module, respectively. The value of the constant
module corresponds to the bifurcation parameter of the Tent chaotic map. The para-
meter of the Fcn module indicates the chaotic map, and finally the chaotic sequence
is converted to binary sequence by the sign quantization module. It is worth noting
that the Tent map from the mathematical model to the computer simulation model
is more difficult to achieve, because the expression of the chaotic map is a piecewise
function, and the bifurcation parameters are divided into two segments. Therefore,
the expression Fcn is
Fcn = u(1) ⋅ (u(1) ≥ 0) ⋅ (u(1) < u(2))/u(2)

, (5.56)
+ (1 – u(1)) ⋅ (u(1) ≥ u(2)) ⋅ (u(1) ≤ 1)/(1 – u(2))
where u(1) represents xn , and u(2) represents the bifurcation parameter a.

The simulation model of the spread spectrum sequence generator module based
on series chaotic map is shown in Figure 5.26. The first chaotic sequence of series
sequence adopts improved logistic map, and the second chaotic sequence uses Cheby-
shev map. Generally, the sampling time of the first chaotic sequence is 1/n times the
sampling time of the second chaotic sequence. In this system, we set the sampling
time of the first chaotic sequence as 1/192 s, and then the sampling time of the
second chaotic sequence is 1/192,000 s. Let the sampling time be 1 s, namely x1 = y1 ,
x2 = y1,000 , . . . , x192 = y191,000 , that is, the first chaotic sequence should control the
second chaotic sequence to generate 1,000 chaotic sequences per second. For simula-
tion, the initial values of the two chaotic sequences are set first. Because x1 = y1 , the
two chaotic sequences must be set to the same initial value in the series simulation,
1
z
Unit Delay1
f(u)
2
8 1
Fcn
Constant z
Constant 1
Fcn1 1
Out1
Pulse f(u)
Generator
Figure 5.26: Simulation model of spread spectrum sequence generator based on series chaotic map.
namely, the initial values of two unit delay modules are the same. Multiple n of the
sampling time of the two-stage sequence is controlled by a periodic sequence gener-
ator, and the sampling time of the periodic sequence generator can be the same as
the sampling time of a chaotic sequence generator. In the multiuser chaotic spread
spectrum communication system, when the frames are sampled, it is required that
the values of chaotic sequences are discrete. So when the 192nd value controlled by
clock arrives, we obtain the chaotic sequence yn , rather than yn+1 , and at this moment
yn = y192,000 . Otherwise chaotic sequence generator cannot work due to the value of
the continuous chaotic sequence. Fcn1 in Figure 5.26 is expressed as
Fcn1 = (cos(u(2) ⋅ a cos(u(1)))) ⋅ (abs(u(4) – 1)) + cos(u(2) ⋅ a cos(u(3))) ⋅ u(4), (5.57)
where u(1) represents xn of the Chebyshev map, u(2) is the bifurcation parameter q of
the Chebyshev map, u(3) represents xn+1 of improved logistic map which belongs to
the first chaotic sequence, and u(4) is the sampling time of multiple n.
5.4.3.2 Correlation Receiving Module

The block diagram of traditional correlation detection at receiver is shown in
Figure 5.27. The dashed box in the figure corresponds to a match filter.
Integral module is the core part of the spread spectrum system, which consists
of unbuffer module, integral sampling module, zero-order hold module, and frame
sampling module, as shown in Figure 5.28. It works as follows. Each frame data (N ∗ 1)
from In1 module is decomposed to (1 ∗ N) sequences by unbuffer module, and then do
sampling integral. The zero-order hold module makes the time duration, hold time,
and information code sampling time of each binary symbol to be the same after the in-
tegration. Finally, they are converted into frame format by the sampling module. The
integrator parameter settings are as follows: Output sampling time is the sampling
γ1 d1
1
T sign
Tb b t = iTb
g1(t)
γ2 d2
r(t) 1
T sign
Tb b t = iTb
g2(t)
γm dm
1
T sign
Tb b t = iTb
gm(t)
Figure 5.27: Principle block diagram of the detector of the traditional DS-Code Division Multiple
Access system.
Integrate To
1 and Dump 1
Frame
In1 Out1
Figure 5.28: Simulation block diagram of correlation receiver.
time of information code, and it is the same with simulation step. The sampling
number of the buffer module is N.
5.4.3.3 Rake Receiving Module

In the environment of multipath transmission, the performance of the conventional
correlation receiver is not ideal, while the rake receiver is a good choice. Rake receiver
technology is a multipath diversity reception technology. It collects all the delay com-
ponents (multipath components) of the original transmitted signals by the relevant
receiving for each multipath signal. Therefore, multipath diversity is also known as
code diversity.
The principle of rake receiver is that the relevant receiving units are applied to a
correlation receiver for each path, where the relevant receiver is relevant to a delayed
version of the signal (that is the multipath components of desired signal) with the
same expectations (received). Then it is weighted based on the relative intensity of
output (called rake tooth output) of these correlation receivers, and all the weighted
outputs from various quarters are synthesized into a single output. The selection
principle of weighted coefficient is to make the output SNR maximum. The theoretical
basis is that when the multipath time delay exceeds a chip period, multipath signal is
taken as uncorrelated.
More specifically, we assume that the L correlators in the CDMA receiver are used
to capture the L strongest multipath components of the desired codes, and then a
weighted network is used to obtain the combination of the outputs of the L correlators,
which is applied as bits detection of the desired codes as shown in Figure 5.29.
From Figure 5.29, the outputs ci (t) of L correlators are multiplied by weighted coef-
ficients di (t) (where i = 1, 2, . . . , L), respectively, and they are summed to obtain the
total output signal, which is denoted as
L

e(t) = di (t)ci (t), (5.58)
i=1
where the weighted coefficients di (t) are defined as
c2 (t)
di (t) = L i , i = 1, . . . , L (5.59)
i=1 ci (t)
2
Obviously, when the correlator (e.g. correlator i) fades, the energy of output ci is small,
and the corresponding weighted coefficients di also decrease automatically, and it
leads the component in formula (5.59) to decrease.
The simulation model of the Rake receiver is shown in Figure 5.30. Assuming that
the data transmitted in channel appears four multipath transmissions, and the time
delay is 0, 2, 6, and 7, respectively. At the receiver, these multipath signals are done in
the time period corresponding 0, 2, 6, and 7 in advance, and then sum them. Finally,
the equal-gain combining will decrease the BER significantly. The equal-gain combin-
ing means that the various branch signals are summed in the same phase, and the
weighted factors of each branch are equal.
c1(t)
Correlator 1
d1(t)
r(t) c2(t) E(t) bm

Correlator 2 Σ (∙)dt Decision
d2(t) 0
.
.
.
cp
cL(t)
Correlator L
dL(t)
Figure 5.29: Structure block diagram of rake receiver with L channels.

1
Data Integrate
z–0 and Dump
Integrate +
z–2 and Dump
+
-K- 1
2 -K- +
Integrate Out1
Code and Dump + Normalize
Normalize z–6 Path Gain3
Path Gain2
Integrate
z–7 and Dump
DeSpreader
Figure 5.30: Simulation model of Rake receiver.
5.4.3.4 Multipath Channel Module

The tapped delay line model is adopted for the multipath fading channel, and its
impulse response is
L

h(t) = !l $(t – 4l )ej6l , (5.60)
l=1
where L is the number of decomposable paths in the multipath channel. !l , 6l , 4l are

the gain, the phase, and the time delay of the lth decomposable channel, respect-
ively. $(t) is the delta function. Under the condition of Rayleigh channel, !l obeys the
Rayleigh distribution, and 6l obeys the uniform distribution in [0, 20).
The simulation model of multipath channel is shown in Figure 5.31. Through the
multipath channel, signal becomes four columns delayed signal, of which the time
delay is 0, 2, 6, and 7, respectively, and the frame sampling format is used. Note that the
sampling time of multipath channel should be the same as that of information code.
5.4.3.5 Polarity Transformation Module

The operator of spreading spectrum and de-spreading spectrum can be realized by the
XOR operator of the unipolar binary code, but it is difficult to process when the result
is zero sometimes. When the signal is superimposed noise, it is no longer a binary
1 Matrix To Row
In1 multiply frame z–[0267] sum -K- 1
Out1
Normalize
Ones(1,p) Separate Path delays Combine path gain
paths paths
Paths
Figure 5.31: Multipath channel simulation model.

code, and we cannot use XOR operator at this time. Employing multiplication of bi-
polar binary symbols can also realize spreading and de-spreading operation, and can
overcome the shortcomings of the above-mentioned method. During simulation, the
polarity conversion is realized by bipolar to unipolar converter module and unipolar
to bipolar converter module.
5.4.3.6 Frame Sampling Module

Taking into account the inherent requirements of the DSP and data communication ex-
penses, the frame-based processing is needed. Obviously, the signal processing based
on frame is much more complex than that based on sampling. The sampled signal
processes a sampled point at each time step, while a frame-based signal processes
one frame with N sample points at each time step. But by using the matrix feature
of MATLAB, Simulink greatly improves the processing efficiency. The communication
between the blocks is reduced by using the frame-based processing, and the sim-
ulation is much faster than that of the signal based on the sampling. In short, the
simulation speed is improved by using the frame-based signal. Therefore, we employ
the frame-based method.
In Simulink, the conversion between the sampled signal and the frame signal is
achieved by the buffer module (buffer). There are two functions in buffer module. The
first is to accept the sampling signal input and generate a frame input with certain
frame size. The second is to accept the frame input and modify the size of the frame,
and the buffer module must be used in this situation. Both cases involve the over-
lap between frames, and the problem of setting the initial value of the frame. When
a frame is generated by sampling, the buffer will use the input scalar to generate a
column vector. If you need to generate a sample from a frame signal, you should use
the unbuffer module.
5.4.4 Design and Simulation of Multiuser Chaotic Spread Spectrum

System Based on Rake Receiver
According to Figure 5.23, multiuser chaotic spread spectrum communication simula-

tion system with five users is built based on the rake receiver as shown in Figure 5.32.
In the figure, the subsystem of users 2–5 at the sender is formed by multiplication
of user information with spread spectrum code. There are corresponding subsystems
of users 2–5 at the receiver. User 1 and receiver 1 within the dotted line are composed
of unpackaged modules.
In addition, according to the setting of the parameters for the multiple paths
module, the type of spread spectrum communication can be changed flexibly. Setting
parameters are shown in Table 5.3.
All the initial value of Chebyshev, improved logistic, logistic, tent, and series
chaotic sequence is set to 0.63. Parameters r, q, and a are equal to 4, 8, and 0.49,
Unipolar to User 1 Receive 1

Sequerce Error rate Tx [in1]
bipolar
converter Calculation Rx
Bipolar to
unipolar
converter
BPSK Bipolor to
AWGN BPSK Error rate unipolar
converter
Unipolar to
Random bipolar Multiple
integer converter paths Rake
+ Unipolor combiner
out1 to bipolar
converter
+
out2 out4 [In4]
User 2 Receive
User 4 [In2] Receive
+ 4
2
+
out3 [In3]
User 3 out5 Receive [In5] Receive
User 5 + 3 5
Figure 5.32: Simulation model of multiuser chaotic spread spectrum communication system.
Table 5.3: Setting parameter of the multiple paths module.
Number of paths Path delays Types of spread spectrum communication system
0 0 Synchronous chaotic spread spectrum communication system

3 [0, 3, 5] Asynchronous chaotic spread spectrum communication system
respectively. The output range of the random integer generator is 2, and sampling time
is 1/192 s, and the sampling number for each frame is 1. The sampling time of chaotic
spread spectrum code is 1/192,000 s, and the sampling number for each frame is 1,000.
The multipath number is 3, and the delay time is 0, 3, and 5, respectively, and the sym-
bol period time is 1/192 s. The symbol period of additive white Gaussian noise channel
is 1/192 s, and the delay number in the receiver is 1. The number of users is 5, and the
simulation time is 10 s. By simulation of the chaotic spread spectrum communication
system, the results are shown in Figure 5.33.
In Figure 5.33, the transmitted signal (a) and received signal (b) are almost
identical, except a little bit delay for the received signal. It is shown that the simulation
system can receive signal correctly.
5.4.5 Performance Analysis of Multiuser Chaotic Spread Spectrum

System Based on Rake Receiver
Next, we analyze the performance of the multiuser spread spectrum communication

system from the aspects of the initial value, the SNR, the number of users, the length
of the spreading sequence, and so on.
2
(a)
1
–1
2
(b)
1
–1
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Figure 5.33: Waveforms of (a) transmitted signal and (b) received signal.
100 100
Chebyshev Logistic 100
Tent
10–1 10–1 10–1

BER
BER
BER
10–2 10–2 10–2
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x0 x0 x0
Figure 5.34: Relationship curves between the BER of spread spectrum system and the initial value x0
of different maps.
5.4.5.1 Influence of Initial Value x0 of Chaotic Map on System Error Rate

Without loss of generality, we set SNR Es /E0 to 1 dB. The initial value changes between
0.0001 and 0.9999 as shown in Figure 5.34. For Chebyshev chaotic map, the error rate
is less affected by the initial value, and there is no peak point.
The peak value of the balance degree of the Logistic chaotic map (corresponding
to the initial values are 0.25, 0.5, and 0.75, respectively) results in the peak value of
BER of the system. When the other initial values are selected, the change of the BER is
relatively flat. Thus, when we choose the initial value, the three initial values of 0.25,
0.5, and 0.75 should be avoided.
From Figure 5.34, we find that there are three peak values of the balance degree of
the tent chaotic map, and the corresponding initial values are 0.2401, 0.49, and 0.7501,
respectively. When x0 = 0.49, the fractal parameter is also 0.49. If the value 0.49 is sub-
stituted into formula (5.53), we find all the elements are zeros except the first element
1 in the real value sequence. In this case, the result of formula (5.54) is E = –0.998. It
means there is a peak point at the initial value 0.49, which has a great impact on the
balance degree, as shown in Figure 5.24(c). But we cannot simply say that the element
of a sequence should avoid this point when constructing the sequence. From eq. (5.53),
it can be seen that the range of sequence element is (0, 1], namely the sequence ele-
ment of the tent chaotic map cannot be zero. Obviously, the obtained binary sequence
is meaningless at x0 = 0.49, so we do not discuss the case of x0 = 0.49. However,
when the initial value x0 = 0.2401 and x0 = 0.7501, there are peaks that show that the
tent map does have a balance degree peak point and it verifies the conclusion in Fig-
ure 5.24(c). So when the initial value of the sequence is selected, the two values 0.2401
and 0.7501 should be avoided.
5.4.5.2 Influence of SNR on BER

Relationship between BER and SNR of spread spectrum system with five kinds of
spreading codes is shown in Figure 5.35. Obviously, when the value of SNR Es /E0 var-
ies from 0 to 8, the BER performance of multiuser logistic, Chebyshev, and tent chaotic
spread spectrum communication system is similar, and is better than that of m se-
quence and Gold sequence. It indicates that the capability of the anti-MAI of chaotic
spread spectrum signal is better than that of the m sequence and Gold sequence. When
the SNR is smaller, the system exhibits mainly noise interference. At this time, the BER
performance of Chebyshev map is the best, and it possesses good antinoise perform-
ance. When the SNR is larger, the system exhibits mainly MAI. Then, logistic map has
good resistance to MAI.
5.4.5.3 Influence of User Number i on BER

Without loss of generality, the user number i = 1–10, and we set the SNR Es /E0 1 dB.
From Figure 5.36, it can be seen that the BER of chaotic spread spectrum system with
100
Logistic
Chebyshev
Tent
Gold
m
BER
10–2
10–4
0 2 4 6 8
SNR
Figure 5.35: BER of spread spectrum system with five kinds of spreading codes.
100
Logistic
Tent
Chebyshev
BER
10¯1
10¯2
0 2 4 6 8 10
User number i
Figure 5.36: BER of the system with three kinds of chaos spreading code for different users.
three kinds of spreading codes has little change with the increase in the number of
users. Where the BER of spread spectrum system of Chebyshev map is minimum, the
BER of tent map is maximum. It indicates that the system based on Chebyshev map
has good antinoise and anti-MAI.
5.4.5.4 Influence of the Sequence Length N on BER

Without loss of generality, the SNR is 8 dB and the sequence length N varies from
100 to 3,500. The other conditions remain the same. When N varies from 0 to 1,050,
Figure 5.37 shows that the effect of logistic map on the BER of the system is better than
that of Chebyshev map or tent map. When N varies from 1,050 to 3,500, the BER of
Chebyshev map is the best, and logistic map follows, while tent map is the worst. Also,
from Figure 5.37, it is easy to see that the BER of the logistic map changes suddenly at
N = 1, 100. The main reason is that the correlation properties of chaotic maps may
10–1
Logistic
Chebyshev
Tent
10–2
BER
10–3
10–4
0 1,000 2,000 3,000 4,000
Sequence length after interception
Figure 5.37: BER of the system for different sequence lengths.

Questions 163
be destroyed after the sequence is truncated, which also confirms the conclusion of
Table 5.2.
The simulation results show that the multiuser chaotic spread spectrum com-
munication system based on rake receiver has the characteristics of strong anti-
interference ability, low BER, high subscriber capacity, good security and easy to
implement, and it is flexible for model transformation.
The BER performance of the chaotic spread spectrum communication system in
series chaotic map is better than that of a single chaotic sequence and m sequence
of IS-95 standard. Further, the generation of series chaotic sequence is simple, and
it has rich source codes, high security, strong antijamming performance, and similar
random states, which are good to increase the system capacity and reduce the system
complexity.
Questions
1. What are the key technologies in chaotic secure communication?
2. What are the factors that affect the performance of chaotic secure
communication?
3. Try to compare the principle and characteristics of different chaotic switched
communication schemes.
4. Describe the construction and simulation of a multiuser chaotic spread spectrum
communication system.
6 Data Encryption Based on Chaotic Sequence
Chaotic signal has great potential in cryptography with special features like unpre-
dictability, high complexity, easy implementation, and so on. Compared to other
encryption methods, chaotic encryption is a kind of dynamic encryption. The pro-
cessing speed is not related with the key length, so it has high efficiency. It can be
applied in real-time signal processing or static encryption.
Chaos theory began to attract the attention of cryptographic community in the late
1980s. The idea of chaotic encryption was proposed by British mathematician Mat-
thews [49]. On one hand, due to the loss of information in the iteration, the amount
of information is gradually lost to zero, so it is impossible to predict the chaotic se-
quences correctly. On the other hand, the chaotic synchronization is the key to achieve
the correct decryption. Because the computer system can accurately reproduce all
the starting states, the synchronization algorithm of chaotic iterative system can be
realized by computer process synchronization. This is a feature that a physical chaos
model does not have. There is no need to use complex physical chaotic synchroniza-
tion and control method. Therefore, since the 1990s, the chaotic sequence encryption
has become an important research frontier of modern cryptography [112, 124–127].
There are two research directions in chaotic cryptography. One is the chaotic se-
cure communication based on chaotic synchronization. The other is to construct a
new stream cipher and block cipher using chaotic system. In this chapter, we first
discuss the theory of chaotic cryptography, and then we study the data encryption al-
gorithm based on chaotic sequences and explore the application of chaotic sequences
in digital watermarking and image encryption.
6.1 Chaos Cryptography
6.1.1 Traditional Cryptography
According to different methods of the text data encryption, traditional cryptography

can be divided into the sequence cipher system and the block cipher system [128].
For sequence cipher, the original plaintext is transferred into plaintext data se-
quence first, and then the ciphertext sequence is obtained by employing the XOR
operation with the key sequences. The generated ciphertext sequence is sent to
the receiver. At the receiver, the plaintext sequence is recovered by using the XOR
operation again. The sequence cipher encryption model is shown in Figure 6.1.
There is no data expansion and error transmission in sequence cipher. Its advant-
ages include good real-time performance and easy implementation for encryption
Binary sequence… m2, m1, m0

Key sequence… c2, c1, c0
Channel
… m2, m1, m0
Key generator
k … z2, z1, z0
Key k
Key source Key sequence generator
Figure 6.1: Principle diagram of sequence cipher system.
k = (k1, k2 …, ki) k = (k1, k2 …, ki)
Plaintext Ciphertext Plaintext

Encryption Decryption
algorithm algorithm
x = (x1, x2 …, xm) y = (y1, y2 …, yn) x = (x1, x2 …, xm)
Figure 6.2: Principle diagram of block cipher system.
and decryption. The security of the sequence is mainly dependent on the key se-
quence. If the key sequence is completely random, it is “one time pad,” which
is not to be deciphered. However, in practice, one-time pad is very difficult to be
realized, so the general encryption and decryption key sequence is pseudo-random
sequence. One-time pad requires the key sequence has a sufficiently long period,
good random statistical property, and greater linear unpredictability, and it also
needs to meet the cipher principle about “diffusion” and “confusion” in the algorithm
design.
For block cipher system, the data sequence x1 , x2 , . . ., is divided into x =
(x1 , x2 , . . . , xm ) groups with the length m, and each group (a vector with the length
m) is converted into equal length data sequence y = (y1 , y2 , . . . , yn ) (a vector with
the length n) with the control of key sequence k = (k1 , k2 , . . . , kt ). The principle dia-
gram of block cipher is shown in Figure 6.2. The difference between block cipher
and the sequence cipher is that each bit of the output is not only related to the in-
put bit of the corresponding time, but also related to the plaintext digital with the
length m. The block cipher has the following advantages. First, the block cipher is
easy to be standardized. Second, synchronization is easy to be achieved for block
cipher. A transmission error of a group data will not affect the others. The loss of a
plaintext–ciphertext group will not affect the decryption of its subsequent groups.
However, the block cipher cannot hide data patterns, and it cannot resist the attack
of replay, embedding, and deletion. General block cipher algorithms include IDEA
and DES.
166 6 Data Encryption Based on Chaotic Sequence
6.1.2 Relationship between Chaos Cryptography and Traditional Cryptography
Relation between chaos theory and traditional cryptography is summarized in

Figure 6.3. The chaotic system is corresponding to cryptography algorithm. The
extreme sensitivity of chaotic system to initial conditions and control parameters is
corresponding to the diffusion of cryptography. Random and long periodic orbits of
chaotic system can replace the pseudo-random signal in the cipher system. An ini-
tial value of chaotic map is spread to the whole phase space through the iteration,
while the cryptographic algorithm generates the expected diffusion and confusion.
Diffusion and confusion are the basic principles of design encryption algorithm. Dif-
fusion means that the effects of each plaintext numbers are spread to multiple output
cipher as much as possible so that the statistics characteristic of the plaintext can be
concealed easily. Confusion means the statistics relationship between the plaintext
and the ciphertext is complex. The parameters and initial values of the chaotic system
are equivalent to the keys in the cipher algorithm. One important difference between
chaotic system and cryptographic algorithm is that the traditional cryptographic al-
gorithm is defined on finite set, while chaos cryptograph is defined on the real set,
which means the traditional chaos theory is adapted to the continuous system, and
this is the essential difference between them. In the chaotic system, Lyapunov ex-
ponent is a measure of the initial condition sensitivity, and Roland pointed out the
possible link between it and security [129].
Chaotic encryption is different from the traditional encryption algorithm. The
main difference is that the random signal used in chaotic encryption is generated
by chaotic system. Chaotic signals are classified into continuous chaotic flows and
discrete chaotic sequences. The former is used for the analog chaotic secure commu-
nication, and the latter is used for encrypted communication. The continuous signal
Chaotic system Cipher algorithm
Sensitivity to initial Diffusion
Iteration Circulation
Parameter Key
Real number Finite set
Figure 6.3: Relationship between chaos theory and traditional cryptography.

generated by chaotic system can also be used as a discrete signal after it is made
discrete, and the new Field Programmable Gate Array (FPGA) technology is used to
design the hardware encryption system, which has become a new research hotspot
[130]. In this section, we will discuss the information encryption based on discrete
chaotic system.
6.1.3 Principle of Chaos Sequence Encryption
Chaos sequence encryption belongs to the sequence cipher system, and its encryp-
tion principle is similar to that of the sequence cipher encryption. The difference is
that the general sequence cipher uses the pseudo-random sequence generated by
the circuit based on the shift registers, while the chaos sequence encryption uses the
chaos sequence generated by chaotic system. In the early stage, the pseudo-random
sequence generator is used to generate the key stream. An attacker can use the cor-
relation analysis method to analyze and reconstruct the pseudo-random sequence
effectively. Now, chaotic system is used to replace the pseudo-random sequence gen-
erator, which destroys the adaptive condition of the correlation analysis, and the
security of the encryption algorithm is enhanced. The encryption principle is shown
in Figure 6.4.
Encryption process is
Y = K ⊕ X, (6.1)
and decryption process is
X = K ⊕ Y, (6.2)
where K is the key stream generated by chaotic system. X is plaintext, and Y is

ciphertext, and ⊕ is XOR operation.
It is a good encryption method by taking the chaotic sequence as key stream and
using a one-time pad cryptosystem, but due to computer finite precision effect, chaotic
sequence will eventually be periodic. How to produce a longer cycle, better random
chaotic sequence is the key for an encryption algorithm, therefore, researchers have
proposed a variety of schemes [131–133].
Key Chaotic system

ki
Plaintext xi Ciphertext yi
(Ciphertext yi) (Plaintext xi)
Figure 6.4: Block diagram of chaos sequence encryption/decryption.

The security of chaotic sequence encryption is mainly dependent on the key stream.
In the chaotic encryption system, the random sequence {xi } generated by the chaotic
system is taken as the key flow {ki }, and it does XOR operation with the plaintext
data flow {mi }; thereby the encrypted data stream {ci } is generated. The plaintext
data stream is binary, while there are two ways to produce the key stream {ki }. One
is to use the original chaotic sequence {xi } directly, and the other is to use the pro-
cessed data of the sequence {xi }. The former is actually the direct simulation of chaotic
modulation communication technology, and the latter is the improvement of the
former. There are a variety of processing methods for the original chaotic data by
employing computer technology. For example, thousands of bytes at the beginning
are discarded to overcome the transition process, or the sequence is obtained from
the chaotic sequence according to a certain law, or a compound iterative algorithm is
used to generate random sequences. So the randomness of the processed sequence
will be greatly improved, which improves the anticracking ability of the encryption
system.
The security performance of the encryption system is directly related to the size
of the key space. For chaotic encryption system, many values can be used as a key,
including nonlinear equations, initial values, system parameters, data-processing
mode, and data encoding way. Obviously, the key space is huge, and the security is
good.
6.2 Data Encryption Algorithm Based on Chaotic Sequence
6.2.1 Basic Requirements of Chaotic Encryption Algorithm
A good encryption algorithm not only can resist the brute-force attacks, but also has
the ability of anti-plaintext attack, and superior temporal and spatial performance.
Resisting brute-force attack. The encryption system should have enough key space
to resist the brute-force attack. Taking into account the extreme case, the user pass-
word is assumed to be a byte of all values, and then the number of possible values of
the password is only 256. Assuming that a possible password needs only a second,
it needs only 256 seconds to try all possible passwords. Obviously, the greater the
password scope is, the more difficult the brute-force attack is.
Anti-chosen plaintext attack. Many cryptanalysts previously obtain the plaintext

and ciphertext pairs through a variety of ways. Especially for data transmission on
the network, it is easy to speculate a few hundred bytes of encrypted TCP data pack-
ets, which contain user name and password. Because a few beginning bytes of the
data packets are often fixed, the encryption algorithm requires that it is still difficult
to analyze the remaining plaintext and key, even when the cryptanalyst has a certain
number of plaintext/ciphertext pairs.
Superior temporal and spatial performance. If an encryption algorithm consumes

a lot of time, or consumes a large amount of computer memory, then it is not desirable.
In fact, research results show that longer encryption time does not necessarily lead to
greater strength of the encryption.
To ensure the security of the algorithm, we think that if a chaotic system is used
in the encryption algorithm, it must be secure chaotic system.
Definition 6.1 Secure chaotic system: A chaotic system with discrete, zero correla-
tion, mixed, robust and large parameter sets is defined as a secure chaotic system. It
should be pointed out that for a chaotic encryption algorithm, the encryption system
based on a secure chaotic system with the above properties is not necessarily safe, but
when a chaotic system without the above properties is used for data encryption, the
chaotic encryption system is bound to be weak. Next, we discuss it in detail.
1. Discrete. It means that the chaotic system should be a discrete map, and the
phase trajectory is not continuous. If the phase trajectory is continuous, the evol-
ution trend of the system will be implied by any small neighborhood of the track,
which may reduce the security of the algorithm.
2. Zero correlation. It means that the cross-correlation of chaotic sequence is zero,
and it is useful to resist the cipher correlation analysis. Zero correlation is associ-
ated with statistical properties of the system, and it indicates that the probability
density of the system is uniform.
3. Mixed. If plaintext is considered to be the initial condition, the mixed means that
the effect of a single plaintext symbol is spread to many ciphertext symbols. Ob-
viously, this property corresponds to the diffusion property in cryptography. A
chaotic system with mixed property has better statistics, and when iteration num-
ber increases, the statistics of ciphertext is not dependent on that of plaintext. The
statistics of ciphertext cannot be obtained by that of plaintext.
4. Robustness. Robustness means that chaotic system can still keep the character-
istics with small parameter perturbation. Most chaotic attractors are not stable, so
a chaos encryption system based on the chaotic system without robustness has
weak key property.
5. Large parameter sets. Chaotic system should have large parameters and initial
value space, and a global chaotic system is the best choice. In addition, an im-
portant index of cryptographic system is the Shannon entropy, which is used to
measure the key space. For discrete time chaotic system, it is approximately log2 k,
where k is the number of keys. The larger the parameter space of the dynamical
system is, the greater the k of discrete system is.
In summary, when we design a chaotic encryption system, the discrete chaotic system
with robustness, zero correlation, mixed property, and large parameter set should be
given priority. Although these five conditions cannot contain all security content, it
is very high threshold for chaotic encryption. Strictly speaking, among the existing
chaotic systems, no one is the safe chaotic system. Sheng et al. [88] proposed the
elliptic chaotic system based on tangent delay, which has the above five conditions.
However, this chaotic system is too complicated, and its mechanism also needs to be
studied further. At present, Logistic map and its improvement have very good security
features.
6.2.2 Logistic Map and Its Statistics
One-dimensional discrete time nonlinear system is defined as
xn+1 = T(xn ), (6.3)
where xn ∈ V and n = 0, 1, 2, 3, . . . , xn is the system state. T: V → V is a map. If

the initial value is x0 , then we can get a sequence {xn , n = 0, 1, 2, 3, . . .} by iteration
operation.
In this section, Logistic map is employed to design an encryption algorithm, and
its equation is
xn+1 = f (xn ) = ,xn (1 – xn ), (6.4)
where , ∈ (0, 4], xn ∈ (0, 1). The Lyapunov exponent of the system with the parameter
varying is shown in Figure 3.2. When, , ∈ (3.5699 . . . , 4], it is chaotic.
For an encryption communication, both the initial value sensitivity and random-
ness of the signal are important. The stronger sensitivity and better randomness can
lead to higher security. These characteristics can be described by probability statistics,
mean value, autocorrelation, and cross-correlation, as shown in Section 3.1.1.
For Logistic map, when , = 4, the mean value of the chaotic sequence is constant,
and the autocorrelation is delta function and the cross-correlation is zero. Its probab-
ility statistics property is similar to that of the white noise with zero mean (as shown
in Figure 6.5). So the Logistic map is suitable for information encryption. In addition,
Logistic map basically satisfies the definition of the secure chaotic system. Therefore,
Logistic map can be applied in digital communication and multimedia security.
6.2.3 Chaotic Encryption Model and Algorithm
To encrypt digital signal, we need to turn the real number sequence {xk } to pseudo-
random sequence [134], which is taken as a key sequence. The simplest way is to
obtain a new integer by selecting a few digits after the decimal point of the xk . The
encryption/decryption model based on Logistic map is shown in Figure 6.6.
Based on the model above, an encryption algorithm is designed, and the flow
chart is shown in Figure 6.7. Taking into account the randomness and encryption
speed, we chose the interval of pseudo-random sequence number N = 5. yi is 4th,
15
10
ρ(x)
0
0 0.5 1
x
Figure 6.5: Probability density distribution of Logistic map.
Plaintext Bi
Xi Yi Ki Ci
Iterative Get new integer by Yi mod 256 XOR
M times taking out N bits
Encryption
Channel
Decryption
Secret channel
Setting initial Generate key XOR
value (key) sequences
Plaintext
Figure 6.6: Chaotic encryption/decryption model.
5th, 6th, bit following the decimal point of xi , which can improve the ability to resist
plaintext attack. In order to decrease the influence of the transition process, it is ap-
propriate to discard the previous 2,000 iterations. In addition, it is difficult for users
to remember the password with two floating-point values, so the first step includes a
process of mapping the strings remembered by the user to x0 and ,. The encryption
algorithm is presented as follows:
1. Initialize the Logistic map, and set the key x0 and ,.
2. Iterate Logistic map for 2,000 times.
3. yi is gotten by taking the 4th–6th bit of xi .
4. Key sequence is obtained by yi mod 256.
5. Do XOR operation with the plaintext or ciphertext.
6. Judge whether encryption/decryption is complete; if it is incomplete, repeat steps
3–5 after 5 iterations of Logistic map. Otherwise, exit encryption system.
Start
Set initial value x0, μ
Iteration 2000 times
Generate key sequence
XOR operation
Y
Completed?
N
Read laintext/ciphertext
Iteration 5 times
Take out the 4th, 5th, 6th bit
mod 256
XOR and output
Exit
End
Figure 6.7: Flow chart of chaotic encryption/decryption algorithm.
6.2.4 Implement of Encryption Algorithm Based on Chaotic Sequence
Based on the proposed encryption model and encryption algorithm, we developed the
data encryption software by using VC.Net in Visual Studio.NET tool. VS.Net is a com-
plete tools for the generation of Web ASP applications, Web services XML, desktop
applications, mobile applications, etc. Programming languages use the functions of
the .NET framework, which provides access to key technologies to simplify the de-
velopment of Web ASP applications and Web services XML. VS.Net employs MFC
(Microsoft Foundation Classes) program method. It makes full use of the advantages
Main control module

(user interface)
Password file name Error code
Encryption/decryption
module (EncFile.dll)
The converted μ and x0 Key Ki
Generate key sequence
Figure 6.8: Relationship diagram of encryption software modules.
of oriented object technology for which there is little need to be concerned about the
implementation details of an object when we are programming. At the same time,
the strong function of the various objects in the class library can complete most
of the program functions, and it effectively ensures that the program is good for
debugging.
We use Microsoft Visual C++ to develop the core algorithm. The software size is
controlled within 1.5M for facilitating users to download. This software is composed
of the main control module, the chaotic key sequence generation module, and the file
encryption/decryption module. The secret relationship among the modules is shown
in Figure 6.8.
6.2.4.1 File Encryption/Decryption Module

This module provides three functions: TranslateKey, EncryptFile, and DencryptFile.
The three interfaces provide a certain degree of abstraction to the main control
module. The main control module does not need to be concerned about the details
of the underlying file encryption, and it only needs to get the encrypted file and the
user password. The specific encryption/decryption process is completed by the file
encryption module.
The test procedure of the parameters is accomplished by the algorithm using these
parameters, and several standard error codes are defined by the main control program.
When an error occurs, the algorithm can return the standard error code and its specific
error information code. The format of the error code is DWORD iError. The high sixteen
bits are standard error code, which can be extracted by the HIWORD of the main con-
trol program, while the low 16 bits are custom error information codes. In addition, a
standard procedure for obtaining the error description information from the error in-
formation code is also included in order to report to the user and realize the software’s
“self description.”
After the standard encryption/decryption file method is provided, it is free to

change the underlying encryption algorithm and will not affect the main control
module, which makes the software version more flexible to upgrade.
6.2.4.2 Chaotic Sequence Generation Module

This module includes two functions: MapKey and GetNextKey. The former can set two
parameters of the logistic map and check the parameters to ensure they are within
reasonable range. The latter is used to get the next encryption key. This module is
transparent to the main control module, and it is also the general module of chaotic
encryption/decryption. It contains two global variables to record the parameters of
logistic map set by the encryption/decryption module. One key is obtained by 5 itera-
tions of the generating function. Experiments show that the design is very successful.
6.2.4.3 Main Control Module

It is the most challenging part in the whole design because the software does not
contain too many user interaction processes, and users need not submit too much in-
formation. The software is suitable to design to be anti-traditional with relatively new
interface form. At the same time, because this program is user oriented, and it costs
a long time when large data files are encrypted, the multithread technology should
be employed to separately implement the user interface process and the encryption
process. The interface display thread can query the working state and progress of the
underlying work thread at scheduled time and report them to the user.
On the Windows platform, the basic unit of interaction with the user is the
application window, so we design a corresponding window in each interactive
stage. This independent design for each state can make the process of design and
implementation relatively intuitive, and maintenance and optimization of the process
will become easier.
The interaction between the users and the main control module includes the
following several aspects: the users start the application; the users choose the file
encryption or the file decryption; the users select the plain text file and the cipher
text file; the users enter the password; and the encryption/decryption process is com-
pleted.
The window procedure mentioned above can be described by a state transition
diagram as shown in Figure 6.9. Each state can be represented by a window, and each
state conversion can be achieved with displaying or hiding the window. For the three
groups, key input 1 and key input 2, file input 1 and file input 2, as well as the en-
cryption progress 1 and the encryption progress 2, each group can be achieved with
the same window, and we only need to record the current state. To display and hide
the window, two functions (FadeIn() and FadeOut()) are added to each window class.
FadeIn() is used to display the window, while FadeOut() is used to hide the window.
The window fade and other special effects can be achieved by modifying the two
standard processes.
Initial state Final state

Encryption
Cancel
Decryption Cancel
File input 1 File input 2
Cancel
Ensure
Ensure
Cancel
Decrypted
Encrypted
Key input 1 Key input 2

Ensure
Ensure
Error
Error
Encryption progress Encryption progress

display 1 display2
Figure 6.9: State transition diagram of the main control module.
6.2.5 System Performance Analysis
6.2.5.1 Analysis of Anticracking Ability

The key of cryptanalysis is to obtain the initial value x0 or parameter , set by the
user. The two values will be taken as floating-point number. If the effective value of a
computer floating-point number is 16, then 30 bits are uncertain for the two numbers,
and the possible combinations are 1030 , while key combination of DES encryption al-
gorithm is 256 . Obviously 30 > 56 log 2. So the proposed algorithm can effectively resist
brute-force attack.
To improve the adaptability of chosen plaintext attack, we use interval sampling
method. If the interval number is one, then the relationship between xn and xn+1 is
z(i) = (x1 (i) + x2 (i)) mod 232 . (6.5)
It is a quadratic equation with one unknown. If the interval number is N, then the
relationship between xn and xn+1 becomes one-dimensional equation of N degrees.
6.2.5.2 Statistics Distribution of Ciphertext

Whether the statistics distribution of encrypted data is uniform is an important index
to evaluate the encryption algorithm.
Theorem 6.1: If the chaotic sequence is completely random, the encrypted data is
subject to the uniform distribution.
Proof. The plaintext and chaotic (key) sequence is considered as a byte of data stream,
and we take a byte of plaintext m and key k randomly from them. Supposing that the
value of m is different, that is, the probability is unequal for one bit to be “0” or “1,”
then the appearance of “0” or “1” is independent. If the probability of “1” is p, then the
probability of “0” is 1–p. The chaotic (key) sequence satisfies the performance of white
noise for the ideal state; that is, “0” or “1” appearing at each bit has the same probab-
ility of 0.5, and then the probability of appearing “1” in the ith bit of ciphertext c is
P(Ci=1) = P(mi=1,ki=0) + P(mi=0,ki=1) = p × 0.5 + (1 – p) × 0.5 = 0.5. (6.6)
The probability of appearing “0” in the ith bit of ciphertext c is
P(ci=0) = 1 – P(ci=1) = 0.5. (6.7)
Obviously, the probability distribution of the ciphertext is uniform. ∎
6.2.5.3 Experiment Results

The chaotic encryption software is designed by employing modular design method
and multithread technology. If both communication sides have the same key and the
encryption software, they can realize the encryption transmission of the information.
The software can encrypt not only the text file, but also the bitmap file. Figure 6.10
is the result of a bitmap encryption. Figure 6.10(a) is the plaintext image. Figure 6.10(b)
is the encrypted image. Figure 6.10(c) is the decrypted image. Compared with Fig-
ure 6.10(a), Figure 6.10(c) is clear and accurate without distortion. Figure 6.10(d) is the
decryption image with small deviation of key, and obviously it cannot be decrypted
correctly.
Data encryption standard (DES) is a relatively good symmetric encryption al-
gorithm. Rivest-Shamir-Adleman (RSA) proposed by Rivest, Shamir, and Ademanis
is an outstanding representative of the public key algorithm. Now, the chaotic en-
cryption algorithm is compared with the DES and RSA encryption algorithm by
encrypting/decrypting Lena image, and the results are shown in Table 6.1. Obviously,
chaotic encryption has advantages in encryption speed and security.
It should be pointed out that the software is mainly suitable for static data en-
cryption. In order to realize the real-time data stream encryption, we must solve the
problems, such as fault tolerance, synchronization, delay, and so on, which will be
researched in the seventh chapter.
6.3 Binary Watermark Image Encryption Algorithm Based

on Chaos
Chaotic sequence is a kind of pseudo-random sequence. It has a simple form and is
very sensitive to initial conditions and has statistical properties of white noise. These
(a) (b)
(c) (d)
Figure 6.10: Image encryption results: (a) original image; (b) encrypted image; (c) correct decryption
image; and (d) decryption image with key change for 10–14 .
Table 6.1: Performances of DES, RSA, and chaotic encryption algorithm.
Algorithm Encryption/ Advantages Defects

decryption time (s)
DES 0.12/0.15 Implementation by software Small key space; insecurity

and hardware; complexity of
encryption/decryption is same
RSA 92/120 High security; large key space Slow encryption speed;
decryption time is longer
than encryption time; used
in digital signature with
small data
Chaos 1.9/2.1 Implementation by software Immature algorithm;
and hardware; complexity of further study about security
encryption/decryption is
same; large key space
characteristics can meet the requirements of the security of digital watermarking

technology. Therefore, chaos-based digital watermarking is becoming a hot topic.
6.3.1 Improvement of Chaotic Pseudo-Random Sequence Generator
Aiming at the defects of various chaotic sequence generators, many researchers have
put forward the improvement schemes, which mainly include the following two
methods.
6.3.1.1 Improvement of Traditional Chaotic Sequence Generator

This method focuses on improving the traditional chaotic sequence implementation
algorithm, and it includes two types. One is to improve the single traditional chaotic
sequence, such as piecewise linear chaotic sequence, Chebyshev chaotic sequence,
and so on. In order to enhance security, we can consider using multiple chaotic sys-
tems. For example, the output of the two chaotic systems can be combined to generate
pseudo-random bit stream, such as the combination of logistic map and tent map.
In addition, the new algorithm for generating uniform pseudo-random number is
studied [135], and spatiotemporal chaotic system is tried to generate pseudo-random
numbers [136].
6.3.1.2 New Secure Chaotic System

At present, along with the emergence of the improved algorithm of the traditional
chaotic sequences, some new chaotic sequence generators have been put forward.
Among them, the TD-ERCS chaotic system has good cryptographic properties due
to its discreteness, global chaos, global zero correlation, huge parameter and initial
value space, and strong anti-degradation ability [137, 138]. So TD-ERCS chaotic system
is a chaotic system with great potential application.
6.3.2 Binary Watermark Image Encryption Algorithm Based on TD-ERCS Map
Binary image processing algorithm is simple and easy to understand and realize, and
has fast calculation speed and other features. A lot of digital watermarking system
uses binary image as the watermark information, so the watermark is more intuitive.
Here, TD-ERCS chaotic map is used to encrypt the watermark image.
6.3.2.1 TD-ERCS Map

TD-ERCS map is a new global two-dimensional discrete chaotic system. It is an ideal
model for information encryption with zero correlation, stable probability distribu-
tion, and good randomness of the sequence. Set the system parameter ,(0 < , < 1),
initial values x0 (–1 ≤ x0 ≤ 1) and !(0 < ! < 0), tangent delay m(m = 2, 3, 4, 5, 6, . . .).
TD-ERCS map is described in Section 3.1.4. The system can generate two independent
sequences xn (|xn | < 1) and kn (|kn | < 1).
6.3.2.2 Design of Encryption Algorithm

Double-precision floating-point number is used in TD-ERCS iteration, and the data
format is shown in Figure 6.11. The data occupies 8 bytes of storage space, and it is
composed of the following three parts:
1. Sign bit s. s represents the sign of the number; it is one bit by which 0 means
positive, and 1 means negative.
2. Exponent bit e. e represents exponential order code, and it is 11 bits. It is unsigned
number (0 < e < 2047).
3. Tail bit f . f is the fractional part of the tail number, 52 bits.
If W is a binary watermark image, and the pixel size is M ×N, the encryption algorithm
is as follows:
1. Taking TD-ERCS system parameters (,, x0 , !, m) as the key, chaotic sequence is

generated by eq. (3.17). Sequence {xn } is composed of M × N/52 numbers gen-
erated by the chaotic sequence (Considering security, the initial segment of the
sequence is not used).
2. Binary sequence {bi } is obtained from f , 1 ≤ i ≤ M × N.
3. The binary watermark image is encrypted with the binary sequence {bi }; encryp-
tion operation is
W ′ = W(i) ⊕ b(i) 1 ≤ i ≤ M × N, (6.8)
where ⊕ is XOR operation. At the receiving end, input the same key, and by
repeating the encryption algorithm, the image can be decrypted correctly.
6.3.3 Performance Analysis of Encryption Algorithm
The security of encryption algorithm mainly depends on the key, while the key has a
great relationship with the key space, the key sensitivity, and the complexity of the
algorithm. Here we discuss the security of the algorithms.
6.3.3.1 Key Sensitivity Analysis

To confirm the effectiveness of the algorithm, a watermark image “ ” with
the size of 64 × 64 is selected. The key is set as , = 0.7123, x0 = 0.7654, ! = 1, 000,
and m = 9. The encryption effect is shown in Figure 6.12. Figure 6.12(a) is the original
1 bit
11 bits 52 bits
s e f
0 63
Figure 6.11: IEEE 754 double-precision binary floating-point format.

(a) (b) (c)
Figure 6.12: Image encryption and decryption: (a) original image; (b) encrypted image; and
(c) decrypted image with correct key.
(a) (b) (c) (d)
Figure 6.13: Decrypted image with wrong keys: (a) , deviation with 10–16 ; (b) x0 deviation with 10–16 ;
(c) ! = 1, 001; and (d) m = 10.
image. Figure 6.12(b) is the encrypted image, and Figure 6.12(c) is the decrypted im-
age with correct key. Figure 6.13 shows the sensitivity to the key. For decryption, the
original watermark images are unable to be decrypted with small deviation of the key.
It can be seen from the experiment results that this algorithm is very sensitive to
the key. If a slightly modified key is used to decrypt the cipher image, the decryption
fails completely. The key contains four dependent parameters: ,, x0 , !, m. ,, and x0
are in real numbers. So the algorithm is secure with large key space.
6.3.3.2 Scrambling Degree Analysis

The gray difference between pixels and their adjacent pixels is defined as

G(x, y) – G(xi′ , yj′ )
i,j
GD(x, y) = , (6.9)
4
where G(x, y) is the gray value of (x, y). G(xi′ , yj′ ) is the gray value of (x, y) at the four
positions: up and down, and left and right.
If the size of the image is M × N, the average adjacent pixel difference of the image is
described as
N–1
M–1
GD(x, y)
x=2 y=2
E(GD(x, y)) = , (6.10)
(M – 2) × (N – 2)
and adjacent pixel scrambling degree is defined by
E′ (GD(x, y)) – E(GD(x, y))

GDD(I, I ′ ) = , (6.11)
E′ (GD(x, y)) + E(GD(x, y))
where E and E′ are the average adjacent pixel difference before and after scrambling,
respectively. The range of GDD is (–1, 1). The larger the range is, the better the image
scrambling is.
Figure 6.14 presents three binary images, and the sizes of them are 64 × 64. The
evaluation of adjacent pixel scrambling is listed in Table 6.2.
E(GD(x, y)) is the average adjacent pixel difference of original image I. The value
of E(GD(x, y)) is about 0.1, which shows that the original image has large pixel val-
ues in close smooth area, and the average gray difference of adjacent pixels is small.
E′ (GD(x, y)) is the average adjacent pixel difference of encrypted image I ′ . In Table 6.2,
E′ (GD(x, y)) is greatly different from E (GD(x, y)), which indicates each pixel tends to
have random distribution, and the average adjacent pixel difference varies greatly.
GDD (I, I ′ ) is larger than 0.5, which illustrates that the adjacent pixel scrambling
degree is bigger, and the encryption effect is very well.
(a) (b) (c)
Figure 6.14: Binary images: (a) image a; (b) image b; and (c) image c.
Table 6.2: Evaluation of adjacent pixel scrambling.
Image a Image b Image c
E(GD(x, y)) 0.0731 0.0695 0.0419

E ′ (GD(x, y)) 0.4750 0.4779 0.4690
GDD(I, I′ ) 0.7333 0.7461 0.8360
6.3.3.3 Analysis of Anti-Brute-Force Attack

The user password is binary sequence K = k, kx0 k! km . k, , kx0 , and k! have the same
data length. They are used to generate floating-point numbers ,, x0 , and !. km is used
to generate integer m. Let the length of km be 12 bits, and then m is generated by
⎧
⎪
⎨ 2 (km = 0)
m= 4 (km = 1) . (6.12)
⎪
⎩ k + 3 (2 ≤ k ≤ 4095)
m m
If the computer word is 32 bits, and the precision (floating-point number) is 10–9 , then
the key space is about 1027 × 212 ≈ 1030 , which is large enough to resist brute-force
attacks.
The results of experiments and performance analysis show that the binary im-
age watermarking encryption algorithm based on the TD-ERCS map is good with large
key space, high key sensitivity, and high security. Therefore, if the encryption al-
gorithm is applied to the watermarking algorithm, it can ensure the security of the
watermark.
6.4 Color Image Watermarking Algorithm Based on Chaotic Map

and Wavelet Transform
As a new technology of copyright protection and content authentication, digital wa-
termarking has become a hot topic. The application of chaos in digital watermarking
technology mainly includes two aspects. One is the pretreatment of watermark in-
formation, such as the following: (1) digital watermark information is embedded
into the image directly, (2) the watermark information is encrypted by chaotic se-
quence, and (3) binary image watermarking is encrypted or scrambled. The other is
the watermark embedding process encryption, such as the choice of the position,
intensity, etc.
In this section, a novel color image watermark algorithm based on discrete
chaotic map and 2-D discrete wavelet transform is proposed. The proposed algorithm
uses a meaningful binary image as the watermark, which increases the water-
mark provability. TD-ERCS discrete chaotic system and the commonly generalized
Arnold map are used to encrypt the watermark image and determine the embed-
ding position, which improves the security of the watermark image. The watermark
is embedded by wavelet transform. Simulation experiments and performance ana-
lysis show that the proposed algorithm not only can resist some common image
processing and noise interference, but also can resist the image shear, small rota-
tion, histogram modification, and joint attack with good invisibility, robustness, and
security.
6.4 Color Image Watermarking Algorithm Based on Chaotic Map and Wavelet Transform 183
6.4.1 Wavelet Transformation Algorithm and Its Application
Wavelet transform is an ideal mathematical tool for local analysis, which has been
widely used in signal analysis, speech synthesis, image recognition, computer vision,
data compression, seismic exploration, and atmosphere and ocean analysis.
6.4.1.1 Continuous Wavelet Transform

If 8(x) is a real-valued function, and its spectrum 8̂(w) satisfies the condition
2

∞ 8̂(w)
C8 = dw < ∞, (6.13)
–∞ |w|
then 8(x) is called a basic wavelet. 8(x) is similar to impulse response of a band-pass
filter. Assuming that a and b denote expansion and translation operation of 8(x), re-
spectively, then by the translation and expansion of the basic wavelet, the wavelet
basis functions can be generated as
1 x–b
8a,b (x) = √ 8( ), (6.14)
a a
where a ∈ R, b ∈ R, a > 0.
If f (x) is a square integrable function, and f (x) ∈ L2 (R), then the continuous wavelet
transform of f (x) is defined as

∞
Wf (a, b) = f , 8a,b (x) = f (x)8a,b (x)dx. (6.15)
–∞
It indicates the decomposition of f (x) on function Wf (a, b), and it also can be con-
sidered as f (x), which is filtered by using a band-pass filter. The wavelet inverse
transform definition is

∞
∞
1 da
f (x) = wf (a, b)8a,b (x)db 2 . (6.16)
C8 0 –∞ a
It should be noted that 8a,b (x) is not orthogonal, and it has very large redundancy. In
addition, the basic wavelet is not unique.
6.4.1.2 Two-Dimensional Discrete Wavelet Transform of Image

Digital image is a discrete two-dimensional signal. The common method for two-
dimensional discrete wavelet transform is to let the two-dimensional scaling function
be separated; that is,
6(x, y) = 6(x)6(y), (6.17)

where 6(x) is one-dimensional scaling function. If 8(x) is the corresponding wavelet,

then three two-dimensional basic wavelets are
⎧
⎪
⎪81 (x, y) = 6(y)8(x)
⎨
82 (x, y) = 8(x)6(y) . (6.18)
⎪
⎪
⎩83 (x, y) = 8(x)8(y)
The scaling function 6(x) can be treated as a low-pass filter, while the wavelet function
8(x) is the high-pass filter of the same layer, and the function set is

8lj,m,n (x, y) = 2j 8l (x – 2j m, y – 2j n) , j ≥ 0, l = 1, 2, 3. (6.19)
It is the standard orthogonal on the basis of L2 (R2 ), and j, l, m, and n are integers.
f0 (x, y) is an image with the size N × N, and N is the power of 2. j is resolution
parameter, and the scale is 2j . When j = 0, the scale of the original image is 1, and it
is a resolution signal for zeroth layer. The resolution reduces by halve with the value j
increasing by 1.
The two-dimensional discrete wavelet transform of an image is carried out as fol-
lows. In each layer of the transformation, the image is decomposed into four small
images with same size. Each of them is obtained by the inner product of original im-
age with a basic wavelet and 2 times interval sampling in x and y directions. The first
layer (j = 1) wavelet transform is
⎧
⎪
⎪ f10 (m, n) =< f0 (x, y), 6(x – 2m, y – 2n) >
⎪
⎪
⎪
⎨f 1 (m, n) =< f (x, y), 81 (x – 2m, y – 2n) >
1 0
. (6.20)
⎪
⎪ f1 (m, n) =< f0 (x, y), 82 (x – 2m, y – 2n) >
2
⎪
⎪
⎪
⎩f 3 (m, n) =< f (x, y), 83 (x – 2m, y – 2n) >
1 0
For the transformation of subsequent levels (j > 1), four same interval samplings and
filter operation are performed at each layer. fj0 (x, y) is decomposed into four smal-
ler images on the scale of 2j+1 in the same way. The inner product in the form of
convolution is
⎧
⎪
⎪ fj+1
0
(m, n) = [fj0 (x, y)6(–x, –y)](2m, 2n)
⎪
⎪
⎪
⎪
⎨f 1 (m, n) = [f 0 (x, y)81 (–x, –y)](2m, 2n)
j+1 j . (6.21)
⎪
⎪ f 2 (m, n) = [f 0 (x, y)82 (–x, –y)](2m, 2n)
⎪
⎪ j+1 j
⎪
⎪
⎩f 3 (m, n) = [f 0 (x, y)83 (–x, –y)](2m, 2n)
j+1 j
Because the scaling function and the wavelet function are separable, each convolu-
tion can be decomposed into one-dimensional convolution on the row and column
f 0 (x, y)
j+1
h0(–x) 2
h0(–x) 2
f 1j + 1(x, y)
h1(–x) 2
f 0j (x, y)
f 2 (x, y)
j+1
h0(–x) 2
h1(–x) 2
f 3j + 1(x, y)
h1(–x) 2
Figure 6.15: Diagram of wavelet decomposition.
f 0j (x, y)
+1
2 h0( x )
+ 2 h0( x )
f 1j + 1(x, y)
2 h1( x ) f 0j (x, y)
4
f 2j + 1(x, y)
2 h 0( x )
+ 2 h1( x )
f 3j + 1(x, y)
2 h1( x )
Figure 6.16: Diagram of inverse wavelet transform.
of fj0 (x, y), as shown in Figure 6.15. h0 is the low-pass filter, and h1 is the high-pass
filter. ↓2 is two times sampling, and fj+10
(x, y) is the low-frequency components after
wavelet decomposition.fj+1 (x, y), fj+1 (x, y), and fj+1
1 2 3
(x, y) are the high-frequency detail
components of horizontal, vertical, and diagonal directions, respectively.
Obviously, the wavelet transform of the image is to filter the image in the hori-
zontal and vertical directions by using the filter module. At each level, fj0 (x, y) contains
the low-frequency approximation components of the previous layer. fj1 (m, n), fj2 (m, n),
and fj3 (m, n) contain the high-frequency detail components of horizontal, vertical, and
diagonal directions, respectively.
Similarly, the two-dimensional inverse wavelet transform of the image is shown
in Figure 6.16, where ⊕ represents convolution.
6.4.1.3 Application of Wavelet Transform for Image Watermarking

The original image, three-level wavelet decomposition diagram, and structure dia-
gram are shown in Figure 6.17. Three high-frequency bands HLi , LHi , and HHi (i =
1, 2, 3) and a low-frequency band LL3 are obtained by three-level wavelet decom-
position of the original image. According to the characteristics of image wavelet
LL3 HL3
HL2
LH3 HH3
HL1
LH2 HH2
LH1 HH1
(a) (b) (c)
Figure 6.17: Three-level wavelet decomposition of image; (a) original image; (b) three-level wavelet
decomposition; and (c) three-level wavelet decomposition structure.
decomposition, the low-frequency band LL3 is the best approximation of the original
image under the largest scale and minimum resolution decided by wavelet trans-
form decomposition series. Most energy of the image is centered in the low-frequency
band. High-frequency band series are the details of image at different scales and
different resolution. These bands are not unrelated from the perspective of multiresol-
ution. For each high-frequency band, they are described from being coarse to fine
for the same edge, contour and texture information of an image in different direc-
tions, different scales, and different resolutions, and there are some relations between
them.
Compared with other transforms, the image watermarking algorithm based on
wavelet transform has many good characteristics, such as spatial scale localization,
multiresolution representation, adaptation to human visual system, different spa-
tial support region, large selectivity, meeting the demand of security, and simple
calculation.
6.4.2 Design and Implementation of Watermark Embedding and Extraction

Algorithm
6.4.2.1 Principle of Watermarking Algorithm

The principle of watermark algorithm is shown in Figure 6.18. To improve the secur-
ity of the embedded watermark image, the chaotic sequence generated by TD-ERCS
discrete chaotic system is used to encrypt the watermark image. According to the
visual characteristics of the human eyes, the human eyes are not sensitive to the
blue component of the RGB image, and thus the watermark is embedded in the blue
component of the color image.
Based on the advantages of embedded watermark of high- and low-frequency
parts in the wavelet domain, the encrypted watermark information frequency is
embedded into the four sub-bands of the blue component of the host image.
The embedding position is determined by the chaotic sequence generated by the
Original color image
Extract the blue component
Three levels wavelet decomposition
Watermark Encrypted General

Watermarking
image watermark Arnold
embedded algorithm
image map
Wavelet inverse transform
TD-ERCS
Color image embedded
with watermarking
Extracted Decrypt Original

watermark watermark Extract watermark image
image image
Figure 6.18: Block diagram of watermark embedding and extraction.
generalized Arnold map. Watermark information is spread to the low-frequency and

high-frequency coefficients of the three-level decomposition. Because the original
image is needed for watermark extraction, it belongs to the nonblind extraction.
6.4.2.2 Choice of Watermark Embedding Position

To embed the watermark into the color image, the first question is to embed the water-
mark into which component of the color image. There are many color image models,
including the RGB model, YUV model, YIQ model, YCbCr model, etc. The RGB model is
the most commonly used. A pixel of true color image has three bytes of R, G, and B data
structures, which represent the size of red, green, and blue. Bmp image is presented
with the RGB method. Based on the brightness equation, gray image transformed from
color image is defined by
Igrey = 0.299IR + 0.587IG + 0.114IB . (6.22)
According to eq. (6.22), the watermark is embedded into the blue component of the
color image.
After the blue component of the color image is decomposed by three-level wavelet,
a low-frequency band and nine high-frequency bands are obtained. The watermark
can be embedded into the low-frequency part or the high-frequency part of the wavelet
domain [139]. Here, the encrypted watermark is randomly embedded into the four sub-
bands of three-level wavelet decomposition in the blue component. The watermark is
spread to the deepest frequency and high-frequency coefficients of the three level-
wavelet decomposition.
6.4.2.3 Selection of Wavelet Base Function

In the design of digital watermarking algorithm based on wavelet transform, the first
question is how to choose proper wavelet base function to ensure the robustness and
invisibility of the watermarking system. The wavelet base function determines the ef-
ficiency and effect of wavelet transform. Different wavelet has different properties,
which will lead to the different distribution characteristics and the energy conver-
gence characteristics of wavelet coefficients, and the robustness of the watermark is
also different.
At present, biorthogonal wavelet is widely used in the image processing due to the
characteristics of regularity, compactness, and symmetry. The research shows that the
Harr wavelet has good performance, and it is a good choice in the digital watermarking
algorithm. The Harr wavelet is defined as
⎧
⎪
⎨ 1, 0 ≤ t < 0.5
8(t) = –1, 0.5 ≤ t ≤ 1 . (6.23)
⎪
⎩ 0, others
Compared to other orthogonal wavelet bases, the Harr wavelet has some advantages
as follows:
1. The support length is the shortest.
2. The computation complexity of decomposition and reconstruction is lower than
that of other wavelets, and it results in the rapid implementation.
3. When the Mallat algorithm is used, the boundary extension is not required. There
is no edge effect due to the limited length of the image signal.
4. When the image is divided into several mutual disjoint image blocks with the same
size, each image block has a corresponding independent wavelet block, and these
wavelet blocks do not intersect each other.
Based on the discussion above, we select Haar wavelet as the wavelet basis function.
6.4.2.4 General Arnold Map

Arnold map is described as

xn+1 = (xn + yn ) mod 1
, (6.24)
yn+1 = (xn + 2yn ) mod 1
where x mod 1 means only the decimal part of x is obtained. The phase space of (xn ,
yn ) is limited in [0, 1] × [0, 1]. The matrix form of eq. (6.24) is

xn+1 11 xn xn
= mod 1 = C mod 1, (6.25)
yn+1 12 yn yn
where C is a matrix, and |C| = 1. So Arnold map is a guaranteed area map. At the same
time, the Arnold map is the one-to-one map. That is, each point in the unit matrix
uniquely transforms to the other point within the unit matrix.
Now, we generalize the phase space of the Arnold map to be {0, 1, 2, . . . , M – 1} ×
{0, 1, 2, . . . , M – 1} and generalize the equation to be two-dimensional reversible area
preserving equation, which is described as

xn+1 ab xn xn
= =C mod M, (6.26)
yn+1 cd yn yn
where a, b, c, and d are positive integers, and the area preserving requires |C| = ad–
bc = 1. Based on the condition, only three parameters of a, b, c, and d are independent.
The generalized Arnold map is still chaotic. We use the generalized Arnold map
to generate the embedding position of the watermark. Taking the pixel coordinates
(x, y) of the watermark image as initial values, and taking the three independent para-
meters and iteration number k of the coefficient matrix C as the key, the generated
iterative results (x′ , y′ ) are used as the embedding position of the watermark in (x, y).
When iteration number is large enough, the corresponding embedding position will
be greatly separated for any two adjacent watermark pixels.
6.4.2.5 Watermark Embedding and Extracting Algorithm

H is the original host image with the size N × N, and W is binary watermark im-
age with the size M × M. The watermark embedding algorithm is described as
follows.
1. Blue component is extracted from the original color image. Wavelet decompos-
ition coefficient d(l (i, j) is obtained by the three-level wavelet decomposition of
the blue component. For d(l (i, j), l is wavelet decomposition level (l = 1, 2, 3), and
( = 0, 1, 2, 3 represent the wavelet coefficients of the low-frequency, horizontal,
vertical, and diagonal directions, respectively. A new matrix with size (N/4)×(N/4)
is got by merging the third-level coefficient matrix of the low-frequency, hori-
zontal, vertical, and diagonal directions, and it is marked as F = {f (x, y), 1 ≤ x ≤
N/4, 1 ≤ y ≤ N/4}.
2. The watermark image W is encrypted by encryption algorithm based on
TD-ERCS map. The encrypted watermark image is denoted by W ′ =
′
w (i, j), 1 ≤ i ≤ M, 1 ≤ j ≤ M .
3. Set the three independent parameters of generalized Arnold map and iteration
numbers k.
4. For the watermark pixel w′ (i, j), let x0 = i, y0 = j. The embedding position (p, q) in
the matrix F is obtained by iteration k times of eq. (6.26).
5. Embed watermark. The watermark pixels are embedded into the selected wavelet
coefficients according to

′ f (p, q) + ! ⋅ w′ (i, j), (=0
f (p, q) = , (6.27)
f (p, q) + " ⋅ w′ (i, j), ( = 1, 2, 3
where f (p, q) is the original wavelet coefficient. f ′ (p, q) is the wavelet coeffi-
cient after embedding watermark. ( = 0, 1, 2, 3 represent wavelet coefficients of
low-frequency, horizontal, vertical, and diagonal directions, respectively. ! and "
indicate the intensity of embedding watermark, which is determined by the spe-
cific watermark and the original image, and it needs trade-off between invisibility
and robustness of the watermark.
6. Repeat steps (4) and (5) until all the watermark pixels are embedded. After the
three-level wavelet inverse transform is carried out by employing all the wavelet
coefficients, the blue component image with embedded watermark is obtained.
Finally the color image embedded with watermark is obtained.
The watermark extraction algorithm is described as follows.

H ∗ is the image waiting for detecting. First, we extract blue component of H ∗ ,
and then the blue component is decomposed by three-level wavelet. A new matrix
marked as F ∗ = {f (x, y)} (where 1 ≤ x ≤ N/4, 1 ≤ y ≤ N/4) is obtained by merging the
coefficient matrix of the low-frequency, horizontal, vertical, and diagonal directions
in the third level. Refer to steps (3)–(5) of the watermark embedding algorithm; the
embedding position of the watermark is obtained by using the same key. The formula
for extracting watermark is
⎧
⎪
⎪ 1 ( = 0 , f ∗ (p, q) – f (p, q) ≥ !/2
⎪
⎨0
′ ( = 0 , f ∗ (p, q) – f (p, q) < !/2
wm = , (6.28)
⎪
⎪ 1 ( = 1, 2, 3 , f ∗ (p, q) – f (p, q) ≥ "/2
⎪
⎩
0 ( = 1, 2, 3 , f ∗ (p, q) – f (p, q) < "/2
where wm ′ is the mth extraction watermark pixel. f (p, q) is the wavelet coefficients
of the original image. ( = 0, 1, 2, 3 represent wavelet coefficients of low-frequency,

horizontal, vertical, and diagonal directions, respectively. !, " indicate the intensity
of embedding watermark. After extracting the watermark information, it is decrypted
by the TD-ERCS map and the correct watermark is recovered.
6.4.3 Simulation and Performance Analysis
Based on Matlab platform, the watermark embedding and extraction is carried out on
24 bits true color Lena image with the size 512 × 512. The watermark image is the binary
image “ ” with the size 64 × 64. The embedding strength is ! = 41 and " = 59.
The performances of the algorithm are analyzed, including invisibility, robust-

ness, and security. To evaluate the concealment of the algorithm, we employ the peak
signal-noise ratio (PSNR). The PSNR of the color images is defined as

2552
PSNR = 10 log10 , (6.29)
(MSE(R) + MSE(G) + MSE(B))/3
where
N N
1 ′
MSE = (X (i, j) – X(i, j))2 , (6.30)
N×N
i=1 j=1
where X ′ (i, j) is the image with embedded watermark, and X(i, j) is the original image,
0 < i, j ≤ N. The bigger the PSNR is, the better the invisibility is.
The robustness evaluation is between the original watermark W and the extracted
watermark W ′ . Here, the methods of visibility and normalized correlation coefficient
(NC) are employed.
Visibility is evaluated by observing the difference between the original watermark
and the extracted watermark. Normalized correlation coefficient is defined as
′
W (i, j)W(i, j)
i,j
NC = , (6.31)
W 2 (i, j)
i,j
where W ′ (i, j) is the extracted watermark, and W(i, j) is the original image. Obviously,
0 ≤ NC ≤ 1. The greater the NC is, the better the similarity between the original
watermark and the extracted watermark is.
6.4.3.1 Invisibility
First, we test the invisibility of the watermark. Figure 6.19(a) is the original image,
and Figure 6.19(c) is the embedded watermark image. Obviously, human eyes cannot
distinguish the difference between them. The embedded watermark image has good
subjective quality. The PSNR of the embedded watermark image is 38.8074 dB with
good objective quality. Figure 6.19(b) is the original watermark, and Figure 6.19(d)
is the extracted watermark. The normalized correlation coefficient NC = 1, which
illustrates that the two watermarks are identical.
6.4.3.2 Robustness
The robustness of watermarking algorithm is to test the anti-attack ability. The
sharpening, distortion, histogram equalization, contrast enhancement, and mosaic
effect are completed by Photoshop CS3 Adobe.
(1) Anti-JPEG compression attack

JPEG lossy compression is applied to Figure 6.19(c). The test results of anti-JPEG
compression attack capability are shown in Table 6.3. Q is the quality factor, and
(a) (b)
(c) (d)
Figure 6.19: Watermark invisibility experiment: (a) original image; (b) original watermark;
(c) embedded watermark image; and (d) extracted watermark.
Table 6.3: Test of anti-JPEG compression attack capability.
Parameter Experimental results
Q 100 90 80 70 60 50 40 30 20
NC 1 1 0.9489 0.9221 0.8759 0.8005 0.7178 0.6350 0.4915
its range is 0 ∼ 100. The larger the Q is, the less compression distortion the im-
age is. NC is the normalized correlation coefficient. The watermark images with
Q = 70, 50, and 30 are extracted as shown in Figure 6.20. It can be seen that
when Q varies from 100 to 50, NC is bigger than 0.8, which means the watermark
image and the original watermark image are basically the same. The extracted
watermark is still clearly distinguished when Q = 30. So the algorithm has strong
ability of resisting JPEG compression.
(2) Antinoise attack
Figure 6.21 shows the noise attack test results. Figure 6.21(a) illustrates the wa-
termark image and the extracted watermark from the 2% Gauss noise attack.
(a) Q = 70 (b) Q = 50 (C) Q = 30
Figure 6.20: Watermark image extracted from JPEG compression: (a) Q = 70; (b) Q = 50; and
(c) Q = 30.
(a)
(b)
Figure 6.21: Experiments of antinoise attack: (a) watermark image and the extracted watermark from
the 2% Gauss noise attack and (b) watermark image and the extracted watermark from the 4%
salt-and-pepper noise attack.
The watermark image and the extracted watermark from the 4% salt-and-pepper
noise attack are shown in Figure 6.21(b). The quality of the image has been
seriously degraded, but we can still extract a clear watermark image. So the
algorithm has a strong ability to resist noise attack.
(a) (b) (c)
Figure 6.22: Anti-filtering attack experiment: (a) extracted watermark after low-pass filtering
experiment; (b) extracted watermark after median filtering experiment; and (c) extracted watermark
after high-pass filtering experiment.
(3) Anti-filtering attack
The result of an anti-filtering attack is shown in Figure 6.22. When low-pass fil-
tering, median filtering, and high-pass filtering are finished, the NC values are 1,
0.9976, and 0.9951, respectively.
(4) Anti-cutting attack
This algorithm is strongly robust to cutting. Figure 6.23 is the watermark image
and the extracted watermark after cutting 1/4 or central 15% cutting. The NCs of
Figure 6.23(a) and Figure 6.23(b) are 0.7956 and 0.8637, respectively.
(5) Anti-rotation attack
Rotation attack is the most representative geometric attack. As long as attacker

makes a small geometric transform on the original image, such as the rotation of
0.1 degree, the pixel position almost changed completely. The owners cannot ex-
tract the watermark in the image, while people cannot see the difference between
the attacked image and the original image. When the rotation angle is greater
than 1, the human eyes can recognize the difference between the attacked image
and the original image. At this situation, we can identify that the image has been
tampered. The anti-rotation attack experiment results are shown in Figure 6.24.
This algorithm can resist rotation attack with 0 to 0.6 degrees.
Besides the JPEG compression, noise attack, filtering attack, shear attack, rotation at-
tack, and other common noise processing and attack, this algorithm can also resist
histogram equalization, contrast enhancement, USM sharpening, distortion, mosaic
effect, and some joint attacks. The experiment results are shown in Table 6.4.
6.4.3.3 Security Analysis

The proposed watermarking algorithm has double security. On one hand, the attack-
ers do not know the 3 independent parameters (a, b, and c) and iteration number k
of the generalized Cat map. On the other hand, even though the correct watermark
(a)
(b)
Figure 6.23: Anti-cutting attack experiment: (a) watermark image and the extracted watermark with
1/4 cutting of upper left corner and (b) watermark image and the extracted watermark with central
15% cutting.
(a) (b) (c)
Figure 6.24: Anti-rotation attack experiments: (a) watermark extracted from rotation with 0.2
degree; (b) watermark extracted from rotation with 0.4 degree; and (c) watermark extracted from
rotation with 0.6 degree.
information is extracted, the original watermark image cannot be recovered without

the TD-ERCS parameters (,, x0 , !, and m). The key is composed of (a, b, and c) and (,,
x0 , !, and m), so the key space is large enough to resist the brute-force attack, and the
algorithm has high security.
Table 6.4: Tests of anti-attack ability.
Attack method PSNR (dB) NC Attack method PSNR (dB) NC
Gauss low-pass filter 35.8808 1 15% central cutting 33.6596 0.8637

Median filter 35.6995 0.9976 1/4 upper left corner 30.0935 0.7956
cutting
High-pass filter 28.3325 0.9951 0.1 degree rotation 35.8816 1
Histogram equalization 27.7572 0.8662 0.3 degree rotation 34.3736 0.9781
Contrast enhancement 29.0262 0.9051 0.5 degree rotation 32.9006 0.8783
2% Gauss noise 28.7673 0.9440 0.6 degree rotation 32.4629 0.7762
4% salt-and-pepper noise 35.5972 0.9805 JPEG compression 35.4051 0.9221
(Q = 70)
USM sharpening 28.8351 0.9854 JPEG compression 34.6336 0.8005
(Q = 50)
Distortion (ripple) 34.3541 0.9684 JPEG compression 33.9761 0.6350
(Q = 30)
Mosaic effect 32.8215 0.7786 Gauss low-pass 33.9952 0.9127
filtering, then zoom
Narrow 1/4, then 4 times 33.3291 0.9659 1% Gauss noise and 31.7662 0.9202
magnification median filter
6.5 Chaotic Image Encryption Algorithm

From the view of the data-processing methods, the image encryption technology can
be divided into three categories: position replacement, value transformation, and its
combination. From the view of signal processing, image encryption algorithm can
be divided into spatial domain algorithm and frequency domain algorithm. Image
encryption based on chaos is a kind of transformation of image data by employ-
ing chaotic sequence, which makes the visual difference between the transformed
image and the original image. The visual difference here refers to the qualitative or
quantitative differences in color, brightness, or outline. In this section, we focus on
the chaos-based image encryption technique.
6.5.1 Chaos-Based Image Encryption Technology
6.5.1.1 Image Scrambling Based on Chaotic Map

Digital image scrambling transformation includes one-dimensional, two-
dimensional, three-dimensional, and high-dimensional transformations. Objectively
they should be one-to-one map, which means they are reversible transformations.
The encryption can be achieved through the transformation, and the decryption can
be achieved through the inverse transform. In order to provide a larger key space and
a more efficient way to change the correlation between pixels, the encryption method
based on high-dimensional chaotic systems has become a new research hotspot in
recent years. For designing a chaotic scrambling scheme, we should choose a suitable
chaotic map. We think the following factors should be considered:
1. Lyapunov exponent should be as large as possible. If the Lyapunov exponent is
larger, the system is more sensitive to the initial value, and it is more suitable for
image encryption.
2. Control parameters should be as much as possible. For the chaotic cipher system,
parameters are often used as key. Large parameter space can ensure a large key
space to resist the brute-force attack.
3. Chaos map should be a reversible one-to-one map to ensure the scrambling
transformation is reversible.
4. High-dimensional map should be given priority. If the system structure of high-
dimensional chaotic system is more complex, then it can change the correlation
between pixels more effectively.
When choosing the encryption parameters, the structure and characteristics of the
chaotic system need to be fully investigated. For example, the weak key should be re-
moved, and the parameters range should be determined, and the key space should
be as large as possible. The chaotic maps usually applied in image encryption in-
clude Kolmogorov flow, two-dimensional Cat map, and two-dimensional Baker map.
Some scholars extend the two-dimensional map to the three-dimensional map. When
the three-dimensional map is used to scramble pixel, the scrambling speed becomes
faster.
In the chaotic scrambling, the sensitivity to initial conditions corresponds to
chaotic plaintext sensitivity. The larger the location of the plaintext varies, the smal-
ler the correlation between the two adjacent pixels is. So it is more difficult to attack
by analyzing the correlation of pixels. Besides chaos-based pixel scrambling or spa-
tial scrambling, we can also use chaotic sequences to scramble rows and columns of
the image. Most image scrambling is carried out in the space domains. Scrambling
damages the relationship between pixels in the image encryption, but it increases
the encoding difficulty and reduces the compression rate. So there is a problem:
How to combine image scrambling encryption with image compression effectively at
present?
6.5.1.2 Image Diffusion Based on Chaotic Sequences

Pixel position transformation does not change the pixel values. The statistics of the
encryption image is poor, so it is not suitable to be used individually. Traditional
encryption theory proposed a scheme of changing the image statistics by diffusion
operation and combining the diffusion operation with the pixel scrambling. The dif-
fusion operation can increase the ability of resisting differential attack and the chosen
plaintext attack.
Diffusion is to transform change of a pixel to the other pixels, and it leads to
the high sensitivity of the encryption system. There are two common methods about
diffusion. One is the diffusion of 4 adjacent gray values in the image. The diffusion
range increases by the power exponent of 4 with the increase of scrambling cycles.
The other is the raster scanning. The diffusion speed is fast. It is realized by “mod and
add” operation. The “mod operation” is designed to control the results in the range of
pixel values, while the “add operation” is to achieve the transmission function by tak-
ing the former change to the next operation. The diffusion process can be expressed
as
Ci = (Pi + Ci–1 ) mod L, (6.32)
where Pi is the pixel value before diffusion. Ci is the pixel value after diffusion. L is the
pixel gray value. To enhance the randomness of bit distribution, the XOR operation
and chaotic sequence are generally introduced in diffusion algorithm, and then eq.
(6.32) is rewritten as
Ci = li ⊕ (Pi + li ) mod L ⊕ Ci–1 , (6.33)
where li (i = 1, 2, . . .) is the processed chaotic sequence, which limits the range of pixel
gray values. In recent years, the encryption scheme based on chaotic sequence is also
developed rapidly. One-dimensional chaotic system is used widely, such as logistic
map and PLCM map. But research results show that the security of the encryption
based on low-dimensional chaotic systems is not high enough [140, 141]. Researchers
have begun to focus on the high-dimensional chaotic systems.
Decryption is the symmetric inverse transform of encryption. The random dis-
tribution of the diffused image makes statistics attacks difficult. Plaintext sensitivity
corresponds to the diffusion rate, and key sensitivity corresponds to the initial value
sensitivity of the diffusion function.
6.5.1.3 Comprehensive Selection of Chaotic System in Image Encryption

When we design a chaotic image encryption system, the selected chaotic system
should be better than other chaotic systems in the following aspects:
1. Initial value sensitivity. It is beneficial to realize the key sensitivity of the security
system, the sensitivity of the original text, and the sensitivity of the cipher text.
2. The uniform distribution of the chaotic states. Chaotic systems can traverse all
possible states randomly. It can enhance the security of the secret key and the
cipher text and can effectively improve the ability of anti-statistics attack.
3. Non periodicity. After the chaotic system is made discrete, it will not be repeated in
the case of limited number of iterations, thus increasing the difficulty of statistical
attack.
4. Nonlinearity. The linear structure of the chaotic system increases the information
correlation. The higher the nonlinearity of chaotic system is, the more random the
state vibration is, and the more secure the encryption system is.
After constructing a chaotic cipher system, it is necessary to carry out the cipher ana-
lysis and use cipher attacks to test the security of the system. Even passing all the
known cipher attacks, the cipher system is not completely secure.
6.5.2 A New Chaotic Image Encryption Scheme
In this section, a new image encryption scheme based on one-way coupled map lattice
(OCML) and TD-ERCS chaotic system is studied. It combines pixel position scrambling
with gray value substitution.
6.5.2.1 Spatiotemporal Chaotic System

One-way coupled map lattice is a spatiotemporal chaos system, and it can be de-
scribed as

xn+1 (i) = (1 – %i )f (xn (i)) + 2%i f [xn (i – 1)] , (6.34)
where n is discrete time coordinate, n = 1, 2, 3, . . . , T (T is the length of time series).

i is discrete space coordinate, i = 1, 2, 3, . . . , L (L is OCML lattice number). f is local
chaotic map. %i is the coupling coefficient, 0 < %i < 1. In this section, we choose logistic
map as the local chaotic map, and it is expressed as
f (x) = (3.75 + +/4)x(1 – x), (6.35)
where 0 ≤ + ≤ 1, the initial value is a random number at (0, 1). If the entire OCML
system satisfies the periodic boundary conditions xn (0) = xn (L), the space sequences
and time sequences of the whole system are chaotic.
6.5.2.2 Scrambling Image by Spatiotemporal Chaotic Sequences

Ref. [142]. proposed a chaotic image scrambling algorithm based on rank transform.
The replacement address code is generated by the chaotic analog sequence without
quantification of the chaotic sequence. It provides a new way for chaotic scrambling.
If the pixel is the basic unit, the scrambling is carried out by the mapping relation-
ship between the pixel and x(i, n). The replacement address code requires to traverse
all the addresses, so row-and-column scrambling method is applied to scramble the
image. Suppose an image with size M × N, and let L = M, T = N, then a spatiotemporal
matrix x with size M × N is obtained. Steps of row scrambling image matrix by x are as
follows:
1. A chaotic sequence {a1 , a2 , . . . , aN } is obtained from one row of the spatiotemporal
matrix x. {ã1 , ã2 ⋯ãN } is obtained by ascending (descending) of {a1 , a2 , . . . , aN }.
2. Permutation address set {p1 , p2 , . . . , pN } is formed by determining the position of
each element of the sequence {a1 , a2 , . . . , aN } in {ã1 , ã2 ⋯ãN }.
3. The same row of the image matrix is replaced according to the permutation
address set {p1 , p2 , . . . , pN }.
(a) (b)
(c) (d)
Figure 6.25: Image scrambling experiments: (a) original image; (b) scrambled image; (c) original
image; and (d) scrambled image.
The column scrambling of the image matrix is finished by the same way. Because spa-
tiotemporal chaos is chaotic in time and space, each scrambled row and column are
satisfied with a good random. Figure 6.25 shows the scrambling results of two different
images. It can be seen that the scrambled image does not have any outline of the
original image by a round row-and-column scrambling.
6.5.2.3 Initial Conditions and the Key Scheme

Chaotic system is sensitive to initial values and parameters, which is consistent
with key sensitivity of the traditional encryption algorithm. The initial values and
parameters are often used as keys of the chaotic encryption system. In this scheme,
the TD-ERCS chaotic system, with large parameter space and good statistics, is ap-
plied to generating two independent random sequences, which are used as the initial
value sequence and coupling coefficient %i of OCML, respectively.
For the TD-ERCS map, given ,(0 < , ≤ 1), initial value x0 (–1 ≤ x0 ≤ 1) and !(0 <
! < 0), tangent delay m (m = 2, 4, 5, 6), two independent real sequences xn and kn
are generated. (,, x0 , !, m) are seed parameters. The two real sequences xn and kn are
normalized, and then they are assigned to the initial value and coupling coefficient of
the OCML system, respectively. The whole encryption system has five keys +, ,, x0 , !,
and m, so the key space is large.
6.5.2.4 Chaotic Diffusion Algorithm

To improve the security of the image, the gray value diffusion is very important. Spati-
otemporal chaotic sequence is high-dimensional pseudo-random sequence with good
cryptographic properties, so the OCML chaotic system is directly used in the diffusion
operation. The security and the encryption speed are greatly improved at the same
time.
In this diffusion algorithm, two real sequences with length M × N are got from
spatiotemporal matrix x. Integer sequences L1 and L2 in [0, N1 – 1] are obtained by
dealing with the two values sequences. N1 is the image gray level. Of course, we can
do the appropriate transformation for the space-time matrix x and then select the L1
and L2 from the time and space, respectively. The selection methods are as follows.
L1 is obtained by quantification after progressive scanning for spatiotemporal matrix
x. L2 , first by scanning all the even columns and then scanning all the odd columns,
is obtained by quantification and rounding finally.
If L1 (i) ⊕ L2 (i) is an even number, the encryption process is
C(i) = (I(i) + L2 (i)) mod N1 ⊕ C(i – 1) ⊕ L1 (i). (6.36)
If L1 (i) ⊕ L2 (i) is an odd number, the encryption process is
C(i) = (I(i) + L1 (i)) mod N1 ⊕ C(i – 1) ⊕ L2 (i), (6.37)
where I(i) is pixel gray value of the image. C(i – 1) is the former encrypted ciphertext
pixel value, and C(i) the later encrypted ciphertext pixel value.
If L1 (i) ⊕ L2 (i)is an even number, the decryption process is
I(i) = (C(i) ⊕ L1 (i) ⊕ C(i – 1) + N – L2 (i)) mod N1 . (6.38)
If L1 (i) ⊕ L2 (i) is an odd number, the decryption process is
I(i) = (C(i) ⊕ L2 (i) ⊕ C(i – 1) + N – L1 (i)) mod N1 . (6.39)
6.5.2.5 Image Encryption Scheme

The image encryption scheme is shown in Figure 6.26.
Step 1 Input seed parameters and two normalized sequences are generated by M
iterations of the TD-ERCS system (pre iteration 100 times).
TD-ERCS
OCML
(M iterations) λ
Seed
parameters
Time
Spatial sequence N L1, L2
sequence M
Plaintext Ciphertext
Row Column
Diffusion
scrambling scrambling
Figure 6.26: Block diagram of image encryption.
Step 2 Set +, and assign the two normalized sequences generated in Step 1 to ini-
tial values and coupling coefficients of OCML system with M order. The OCML
system iterates N times.
Step 3 Row scrambling by time series with the length M.
Step 4 Column scrambling by space series with the length N.
Step 5 Get L1 and L2 from the OCML system. Do diffusion by employing the algorithm
above.
Steps 3 and 5 can be repeated several times to get higher security. In fact, it can get a
good effect with one-time encryption.
Decryption is the inverse process of the steps above.
6.5.2.6 System Security Analysis

(1) Statistical analysis. In order to resist statistical attack, the encrypted image needs
to be kept in certain diffusion and disturbance. In this section, we take Lena
image encryption as an example to analyze the security of the algorithm.
(a) Histogram analysis. Histogram analysis can be used to measure the disturb-
ing of encryption algorithm. The histograms of Lena and its encryption image
are shown in Figure 6.27. The histogram of the encrypted image is uniform,
so it does not retain any information about the plaintext.
(b) Correlation analysis. Correlation is generally measured by the correlation
coefficient r as follows:
cov(p, q)
r= , (6.40)
D(p)D(q)
where p and q are gray values of the two adjacent pixels in the image, and
cov (p, q) = E(p–E(p)(q–E(q)) is the covariance, where D(p) and D(q) are the
variances. The correlations are calculated by randomly selecting 1,000 pair
adjacent pixels from horizontal, vertical, and diagonal directions of the im-
age, respectively. Table 6.5 lists the correlation coefficients of the original
800
700
600
500
400
300
200
100
0
0 50 100 150 200 250
(a) (b)
600
500
400
300
200
100
0 50 100 150 200 250
(c) (d)
Figure 6.27: (a) Original image; (b) histogram of the original image; (c) encrypted image; and
(d) Histogram of the encrypted image.
Table 6.5: Correlation coefficients of two adjacent pixels.
Original image Encrypted image
Horizontal 0.9487 0.0207

Vertical 0.9218 0.0106
Diagonal 0.9492 0.0400
image and its encrypted image. The correlation coefficients of the encrypted
images are very small, which indicates that no detectable correlations ex-
ist between the original image and its corresponding cipher image. The
diffusion of the algorithm is good.
(2) Key sensitivity analysis. A good image encryption algorithm should be suffi-
ciently sensitive to the encryption key and the plaintext. We test several images
600
500
400
300
200
100
0 50 100 150 200 250

(a) (b)
Figure 6.28: Key sensitivity analysis results: (a) encrypted image with , = 0.73120000001 and
(b) histogram of the decrypted image.
to prove the sensitivity of the algorithm to the key. In this encryption experi-
ment, the Lena image is encrypted with the initial values + = 4, , = 0.7312,
x0 = 0.5654, ! = 0.9876, and m = 9. Figure 6.28(a) is the decrypted image with
, = 0.73120000001, the other keys being unchanged. Figure 6.28(b) is the histo-
gram of the decrypted image. It can be seen that the decrypted image is totally
different from the original image with the key , varying by 10–11 . The histogram
is still uniform. So this algorithm is sensitive enough to the key.
For a given key, the encryption of different images will generate different
ciphertext. The diffusion operation ensures that even the change of a single
pixel will cause a dramatic effect on the whole text, and the encrypted im-
age is completely different. So the algorithm is also sensitive enough to the
plaintext.
(3) Analysis of anti-brute-force attack. The password is binary sequence K, K =
k+ ku kx0 k! km , where the k+ , ku , kx0 , k! , and km are used to generate floating-point
number +, ,, x0, and ! and integer m. If the length of km is 12 bits, then m is
obtained by
⎧
⎪
⎨ 2 (km = 0)
m= 4 (km = 1) . (6.41)
⎪
⎩ k + 3 (2 ≤ k ≤ 4095)
m m
If the word “length” in the computer is 32 bytes, and the precision is 10–14 , then the
key space is 1056 × 212 ≈ 1059 . The key space is large enough to resist the brute-force
attack.
The experiments above show that the new image encryption scheme based on
OCML and TD-ERCS chaotic system has excellent diffusion and disturbance character-
istics. In this scheme, the security of the algorithm can be ensured by the combination
Questions 205
of the random row–column scrambling with changing pixel gray value. The key space
is large enough to resist the brute-force attack, and the encrypted image is greatly
sensitive to the keys and the plaintext. The algorithm is able to encrypt the images of
any size quickly.
Questions
1. What is chaotic cryptography? What are the differences between chaotic crypto-
graphy and traditional cryptography?
2. Analyze the principle of chaotic encryption/decryption.
3. Design a chaotic pseudo-random sequence generator.
4. What are the performance-testing methods of chaotic pseudo-random sequence?
7 Audio and Video Chaotic Encryption
and Communication Technology
With the development of network communication, Internet provides an amount of

information services to users every day. Because the foundation protocol of Internet
is not completely secure, when unencrypted information transmits on the network,
it will be directly exposed to the entire network, which is a great threat to informa-
tion security, especially for audio and video communications, such as eavesdropping,
call tracking, phone hijacking, and denial of service. Therefore, we should improve
network design and also take a variety of technologies, such as voice encryption
technology, to ensure the safety of audio and video communications. With the deep-
ening of chaotic applied research, promising progresses have been made on chaotic
cryptosystem [143–146], and a new path has been opened to improve the information
security. Thus, it is significant to combine the traditional data encryption algorithm,
video image standard MPEG and H.264 with chaos and to apply it in real-time
communication. This chapter focuses on audio and video real-time encryption and
communication technology based on chaos.
7.1 Principle of Real-Time Voice Encryption Communication

In Chapter 6, we studied the chaotic encryption principle of static data file, and it
mainly used the idea of computer process synchronization to realize synchroniza-
tion of chaotic iterative system. The starting parameters and synchronization strategy
themselves are passwords, so it need not use complex physical chaotic synchroniza-
tion or control method. But if you want to achieve real-time transmission of encrypted
data stream, chaotic synchronization is an unavoidable problem. Here, we adopt the
network adaptive communication environment (ACE), which is a cross-platform, to
implement the real-time transport protocol (RTP)/RTP control protocol (RTCP) and
combine the chaotic encryption algorithm with the extended Data Encryption Stand-
ard (DES) encryption algorithm. A voice encryption communication system, which
is adaptable for network real-time transmission, is developed by equal interval drive
synchronous method.
7.1.1 Principle of Digital Encryption Communication System
Although the sequence cipher system, block cipher system, and public key cryptosys-
tem can be used for encryption of digital signal, researchers have certain requirements
for the real time and quality (measured by bit error rate) of information transmission
in practice. Therefore, an appropriate cryptosystem must be selected according to the
specific digital communication system and the corresponding transmission quality of

channel [147].
Generally speaking, because there are error diffusion and delay for block cipher
system, block cipher system is applied in the occasions of good channel transmis-
sion quality or containing data retransmission function. Due to its huge amount of
computation for public key system, it is difficult to meet the real-time requirements
of voice communication system in the present technology level. Therefore, the public
key system is usually not directly applied to real-time data stream encryption. Further-
more, because of the characteristics of small delay and no error diffusion, sequence
cipher system is widely used in digital communication system, which is suitable for
the quality of all kinds of channels [148]. The principle of digital encryption and
communication system based on the sequence cipher system is shown in Figure 7.1.
Digital voice encryption communication system mainly includes source cod-
ing/decoding, encryption/decryption operator, cipher sequence generation, cipher
synchronization control, and synchronization insertion/detection.
Through sampling, quantization, and encoding (including compressed encod-
ing), source coding/decoding module at the sender transforms the original analog
signal into binary digital sequence and inputs them to the encryption device. This
module recovers the binary sequence to analog signals at the receiver.
Encryption/decryption operation generally uses the XOR operation. At the sender,
ciphertext is obtained by the XOR operation of the binary digital sequence from
source encoding module with secret key sequence. At the receiver, the binary plaintext
sequence is recovered by the XOR operation to secrete key sequence.
The main function of the cipher sequence generation module is to generate secrete
key sequences for encryption or decryption, which is the key part of reliability in di-
gital encryption and communication system, including the encryption sequence gen-
eration module (it is used to encrypt the plaintext on the sender) and the decryption
sequence generation module (it is used to decrypt the ciphertext at the receiver).
Synchronous insertion Synchronous detection

Source coding/decoding unit
Source coding/decoding unit
Encryption sequence Decryption sequence

generation generation
Channel
Cipher synchronization Cipher synchronization

control control
Decryption sequence Encryption sequence

generation generation
Synchronous detection Synchronous insertion
Figure 7.1: Principle block diagram of digital encryption and communication system.
208 7 Audio and Video Chaotic Encryption and Communication Technology
The cipher synchronization control module mainly generates a cipher synchroniz-

ation signal to update the status of secrete key sequence at both the receiver and
the sender. To ensure the cipher synchronization in the encryption and decryption
processes, the secrete key (initial conditions or parameters of chaotic system) must
be the same.
At the sender, synchronous insertion/detection module inserts the cipher syn-
chronization signal into the corresponding position of ciphertext stream. At the
receiver, digital information stream is detected in real time. Once the cipher synchron-
ization signal is found, it is extracted from the information stream, and its message
key is sent to the decryption sequence generation module to update its status and the
noncipher synchronization information is transmitted transparently at the same time.
7.1.2 Real-Time Voice Encryption and Communication System
Point-to-point (P2P) real-time voice chaotic encryption and communication system is

shown in Figure 7.2. At the transmitter, first, analog signal is transformed into binary
plaintext data stream through the A/D converter. Then, chaotic sequence generator
generates a chaotic pseudorandom sequence to XOR the plaintext data stream, and
a ciphertext data stream is produced. Subsequently, ciphertext is sent to the network
channel according to the network protocol. At the receiver, a chaotic sequence, which
is synchronous with the sender, is used to decrypt the ciphertext sequence by the XOR
operation to restore the binary voice stream. Finally, through D/A converter, the binary
voice stream is returned to the analog voice signal and it is broadcasted.
The encryption process at the sender is expressed as
S(n) = M(n) ⊕ E(n), (7.1)
where M(n) is the voice sequence, E(n) is the chaotic sequence, S(n) is the ciphertext
sequence, and ⊕ is the XOR operator.
The decryption process at the receiver is the inverse process of encryption, i.e.
M ′ (n) = R(n) ⊕ D(n), (7.2)
M(t) M(n) S(n) R(n) M’(n) M’(t)

Voice A/D Encryption Network Decryption D/A Voice
E(n) D(n)
Chaotic sequence Chaotic sequence

generator generator
Figure 7.2: Principle block diagram of the voice chaotic encryption system.
where R(n) is the ciphertext on the receiving side and R(n) = S(n) for an ideal channel.
D(n) is the chaotic sequence on the receiving side. M ′ (n) is the decryption output.
Obviously, when both sides of the system are synchronized, i.e. D(n) = E(n), then
M ′ (n) = M(n). So the voice encryption communication is achieved.
Obviously, the key of realizing the encryption communication is to realize the
synchronization of chaotic sequence generators between the receiver and the sender.
Because the object of computer processing is digital signal, the existing analog
synchronization method cannot meet the requirements of network transmission. Com-
bined physical synchronization with computer programming techniques, the equal
interval drive synchronization method is proposed, which is suitable for computer
network communication. Theory analysis and experimental results show that this
method has strong fault tolerance capability.
7.1.3 Cipher Synchronization Technology
Cipher synchronization is the key technology in the digital real-time encryption and
communication system. In practice, to ensure the strict consistency of the start-
ing points of system communication on the encryption and decryption sides, the
cipher synchronization must be established before encryption and communication.
The order of cipher synchronization is shown in Figure 7.3.
First, bit synchronization is set up for the system. It can be realized by the
clock provided by the source terminal or channel transmission device, and also can
be achieved by extracting the clock from the source terminal or the digital stream
provided by the transmission device. Second, cipher synchronization is established.
After the establishment of bit synchronization of communication system, the sender
sends the cipher synchronization signal (including the synchronization head and the
message key) to the other side to make the cipher generator of two sides to work
synchronously. For some real-time encryption communication systems, it needs to es-
tablish frame synchronization (group synchronization) before the establishment of
cipher synchronization. Finally, ciphertext sequence is transmitted. After the code
synchronization signal of the sending side is transmitted, the ciphertext sequence is
transmitted immediately. The encrypted sequence is obtained by the corresponding
encryption algorithm, which is based on the same message key as the receiving side.
Receiver extracts the message key from the cipher synchronization signal and injects
it into the cipher generator. Simultaneously, the cipher generator generates the cipher
... ... ... ...
Bit Frame Cipher Ciphertext

synchronization synchronization synchronization synchronization
Figure 7.3: Order of the cipher synchronization.

sequence, which makes the decryption cipher sequence and the encryption cipher
sequence to work in the synchronized state.
Cipher synchronization methods include continuous synchronization, unique
initial synchronization, self-synchronization, and special synchronization. In this sec-
tion, a continuous synchronization method is adopted to realize the cipher synchron-
ization of encryption communication. The continuous cipher synchronization is to
send a cipher synchronization signal with a fixed interval, and it has advantages, such
as allowing the lag to enter, plaintext beyond function and multiple communication
ways. It is worth noting that the transmission cipher synchronization information may
reduce the transmission rate, but the security and reliability of system are improved.
7.2 Real-Time Data Stream Encryption Algorithm

In real-time voice encryption algorithm, the traditional DES cryptography is no longer
safe [149]. Researchers proposed various improved algorithms, such as 3DES based on
secrete key expansion, but its key space is still limited. Therefore, 3DES encryption
method based on chaotic map is adopted to guarantee the security of data stream.
Specifically, synchronized chaotic map systems are set on the encryption and decryp-
tion sides, the secrete key of each data packet is generated by chaotic system to realize
real-time encryption and transmission of voice data.
7.2.1 3DES Encryption Algorithm
3DES is defined by the National Institute of Standards of Technology (NIST) in 1999

[150]. 3DES defines two kinds of work modes. In the first mode, secrete key k1 is used
to encrypt the 64-bit plaintext. Then secrete key k2 is used for the second encryption.
Finally, secrete key k3 is used for the third encryption, and the ciphertext is outputted.
This pattern is called DES-EEE, which is expressed as
C = Ek3 (Ek2 (Ek1 (M))), (7.3)
where M is the plaintext and C is the ciphertext. This encryption mode extends the
key space into 168 bits. In the second mode, k1 is used to encrypt the 64-bit plaintext.
Then the result is decrypted by k2 . Finally, k3 is used to encrypt the decryption result.
This pattern is called DES-EDE, which is expressed as
C = Ek3 (Dk2 (Ek1 (M))) (7.4)
where M is the plaintext and C is the ciphertext. This encryption mode extends the
key space into 112 bits. In the encryption process, secrete keys k1 , k2 , and k3 have three
options as follows.
1. k1 , k2 , and k3 are independent of each other, and they are different from each other.
2. k1 and k2 are different, but k3 is same with k1 .
3. k1 , k2 , and k3 are same, and this way is equivalent to the standard DES.
Here, we adopt the second mode of 3DES, and the second way of secrete key. The
encryption model is shown in Figure 7.4. 64-bit data is encrypted with k1 to obtain the
ciphertext first. Then k2 is used to decrypt the result to obtain “plaintext,” and then
k3 (k1 ) is used to encrypt the decryption results. Obviously, after removing the bits
of eight times from the 128-bit secrete key, there are 112-bit secrete key, and the key
space is 2112 . The only disadvantage of this method is that three times ordinary DES
encryption of the algorithm spends about three times implementation time. But it is
acceptable for very cheap computing resource at present.
7.2.2 Encryption Algorithm Based on 3DES and Chaotic Map
Chaotic dynamics system can be used in many application fields, such as digital
communication and multimedia data security [151].
Similarly, logistic map studied in the sixth chapter is defined as
xn+1 = , × xn × (1 – xn ), (7.5)
where , ∈ (0, 4] and xn ∈ (0, 1). When 3.5699 . . . < , ≤ 4, it is chaotic. Accord-
ing to the statistical analysis in Section 3.1.1, the chaotic sequence of logistic map
{xn , n = 0, 1, 2, 3, . . .} is nonperiodic and nonconvergent, and it is very sensitive to
initial conditions.
DES algorithm with key expansion (3DES) is the core of real-time voice encryp-
tion algorithm, and the secrete keys of encryption and decryption are generated by
two synchronized chaotic systems (logistic map). The encryption model is shown in
Figure 7.5. Ki is the secrete key of encryption and decryption of each data packet, which
is generated by the logistic map. x0 and , are the initial condition and parameter of
logistic map, and that is the secrete key transmitted in the secret channel.
When the logistic map is used to generate secrete key, two problems need to be
solved. One is how to transform the generated sequence into a 64-bit secrete key, and
another is how to synchronize two chaotic maps to ensure that each data packet to
DES encryption DES decryption DES encryption

Plaintext Ciphertext
k1 k2 k3(k1)
Figure 7.4: 3DES encryption model.

3DES encryption 3DES decryption
DES DES DES DES DES DES

Data Public Data
packet channel packet
Ki Ki
Logistic map Logistic map
x0 or u0
Secrete channel
Figure 7.5: Encryption model based on 3DES and Logistic map.
i 11-bit order code 52-bit tail code
0 1 11 12 63
Figure 7.6: Storage format for double type data.
obtain the right secrete key for decryption. Here, the double precision type is used to
store the initial condition x0 and parameter , of logistic map. The double type is 8
bytes (64 bits) in Win32. So it can be used as the secrete key of DES. The secrete key
of 3DES is obtained by iterating the logistic map two times to produce 64-bit double
data. The storage format of double data is shown in Figure 7.6.
In this figure, the 0th bit is sign bit. The first bit to the 11th bit is 11-bit order code,
and the 12th bit to the 63rd bit is 52-bit mantissa. The calculation method of double
type data is
x0 = (–1)i × 2e–1023 × 1.f , (7.6)
where i is the sign bit, e is the order code, and f is the mantissa.
The range of iteration value is (0, 1), and thus e ∈ [0, 1022]. The number of its valid
bits is still 11, and its sign bit is unchanged. So the total number of valid bits in double
type is 63. The iteration space is 2 × (263 –252 ), in which multiplying by 2 represents two
64-bit iterations, and minus 252 means to subtract it at e = 1, 023. Because the number
of 8 times in DES algorithm is only used as check bit, the key space of an actual data
packet is about 2112 .
To realize synchronization between two chaotic systems, the iteration number of
system is saved in the header of each encrypted data packet as plaintext. According
to the iteration number in the header, the receiver iterates the logistic map to produce
decryption key and uses it to decrypt. In practice, the former packet of secrete key can
be memorized to reduce iterations. Since the number of iterations is transmitted as
plaintext in the header data, it may be intercepted by third party. But the decryption
key cannot be obtained without the initial condition and parameter of this map. As
long as the synchronous data in the header is received correctly, the synchronization
can be achieved.
Shannon proves that one-time pad is unable to be deciphered, and 3DES has high
security level. So, the encryption algorithm has a high security. Corresponding to real-
time voice communication, if a single data packet is deciphered, it will not affect the
entire transmission process. Because the pseudorandomness of chaos leads to the
condition that secrete keys are relatively independent, so that each data packet has
the same crack strength to ensure the security of data communication to the maximum
extent.
7.2.3 Network Synchronization Method
Synchronization is the key to achieve real-time data stream encryption. If transmis-

sion errors of the sequence generated by the logistic map occur, it will cause the failure
of synchronization of encryption and decryption. Here, a synchronization scheme
is designed, which fully depends on the network to achieve self-synchronization as
It can be seen that the processes of encryption and decryption are symmetrical,
and the synchronization between them is completely dependent on the error-free
transmission on the network. Tests show that the synchronization of encryption and
decryption is good in the beginning of a few minutes. However, when the time is longer
(usually between 5 and 10 min), the phenomenon of out of step will suddenly happen,
so that the entire decryption is destroyed, and it cannot be restored. On one hand, the
transmission control protocol (TCP) cannot be fully realized error-free transmission.
On the other hand, both the encryption and decryption sides do not have fault-tolerant
ability, and a little mistake could waste all the previous efforts. After further analysis,
it is found that the above scheme is not without merit, because it is synchronized at
the beginning of a few minutes. That is to say, we could establish an error tolerance
and error correction mechanism to make the less probability of data error. But how do
you know when the data is wrong, unless the server sends the voice data and error
correction code at the same time. Otherwise, it is not known. However, considering
Encryption side Decryption side

Voice decoding unit
Voice encoding unit
Send Receive
Network
Encryption Decryption
sequence sequence
generator generator
Figure 7.7: Network self-synchronization scheme.

Encryption side Decryption side
Voice decoding unit

Voice encoding unit
Synchronous Synchronous
Network
Encryption insertion detection
Decryption
sequence sequence
generator generator
Figure 7.8: Equal interval drive network synchronization scheme.
Encryption side D1 E1 D2 E2 D3 E3 D4 E4 D5 E5
Decryption side Dx E1 D2 Ex D3 E3 E4 D5 E5
Dec Dec Dec Dec Dec
Ex Ex E3 Ex E5
Error Error Correct Error Correct
Figure 7.9: Block diagram of fault tolerance capacity analysis.
the particularity of voice information, it does not necessarily have the ability of error
correction as long as the fault is tolerant. In real life, when a sentence does not listen
to the other side, he can ask the other side to say it again. Based on the above analysis,
an equal interval drive synchronization scheme is designed in Figure 7.8.
Compared with Figure 7.7, the synchronous insertion and synchronous detection
module are added to the system in Figure 7.8. In the encryption side, the drive code of
chaotic sequence (number of iterations) is inserted into the header of ciphertext data
packet, and it is sent with the ciphertext data packet. In decryption side, the driver
code is taken out by the synchronous detection module, and it is used to drive the
chaotic sequence generator to generate secrete key to decrypt the cipher sequence,
and the secrete key is consistent with that of the sender.
The fault tolerance capability of the above synchronization scheme is analyzed.
We will consider all errors that may occur in the network transmission. The transmis-
sion process is shown in Figure 7.9.
The upper part of Figure 7.9 is the data sent to the network, where Di is the drive
code and Ei is encrypted voice stream. The middle data is read from the Internet on
the decryption side, where Di is the drive code and Ei is the encrypted voice stream.
The bottom data is the decrypted voice data stream. Obviously, transmission errors are
classified into the following categories:
1. Drive code is transmitted in error. For instance, D1 is transmitted to Dx .
2. Encrypted voice data is transmitted in error. For instance, E2 is transmitted to Ex .
3. Drive code is lost. For instance, D4 is lost.
As shown in Figure 7.9, the first and second errors only affect one data packet of voice
stream. When D2 or D3 is received on the decryption side, all contacts between Ex and
Dx will be cut off. D3 is used to iterate two values to generate two double precision de-
cryption sequences to complete decryption, while the encryption side also use D3 to
generate two double-precision encryption sequence to accomplish encryption. So the
encryption and decryption sequences are same, and the synchronization of encryp-
tion and decryption is realized. For the third case, it only affects one voice data packet
too. When the correct driver code is received, it will return to the state of synchroniz-
ation. After a long period of testing, it is proved that this scheme is feasible and can
keep synchronization of encryption and decryption in infinite time. Although trans-
mission errors occur accidentally in the middle, they will not affect the whole process
of call. There is no security risk to expose the drive code to the outside because the fol-
lowing n-bit sequence does not have the possibility to be predicted without knowing
the values of parameter , and initial condition x0 .
In general, the main idea of this synchronization scheme is to make use of the
tiny bit error rate provided by network transmission protocol. Regardless of whether
there is a bit error, the decryption side is forced to make a new synchronization at set
intervals (it is very small). That is to say, the synchronization of the entire call time
is changed to the synchronization of many times. Even if a period of time is not syn-
chronized, it will only affect the data in this segment, and the entire call is not affected.
7.2.4 Security Analysis of Real-Time Voice Encryption System
Dachselt et al. [152] studied the cryptanalysis method of an encryption system based
on discrete chaotic map. Yang et al. [153] thought that analog chaotic encryption sys-
tem is difficult to resist the return mapping attack. Zhou et al. [154] reported that
simple chaotic encryption system did not have a high security level and proposed an
improved algorithm by increasing the number of iterations. Here, we design an en-
cryption system, which combines the block encryption algorithm with chaotic encryp-
tion algorithm, and the security of system is analyzed by the following cryptanalysis
methods:
1. The system has the ability to resist the brute-force attack. Because the encryption
system is not dependent on the encryption scheme or algorithm, but relies on the
secrete key. It is worth noting that secrete key signal generator could not be open,
and some literatures about cryptanalysis often adopted this design principle. This
is particularly important for the advantages of chaotic encryption. The previous

analysis indicates that the system adopts the improved DES algorithm, i.e. 3DES,
to encrypt data packets, and its key space is 2112 . Since the secrete key of sys-
tem is generated by chaotic map, the secrete key of each data packet is alterative.
Therefore, the key space of each data packet is 2112 , which is large enough to resist
brute-force attack. Even if the breaker acquires the equation of chaotic map, it is
difficult to determine its parameters and initial conditions.
2. The system has the ability to resist the chosen plaintext and chosen ciphertext at-
tacks. As the secrete key of system is generated by the chaotic map, and the secrete
key sequence is unpredictable, there is no one corresponding to the plaintext
and the ciphertext. The ciphertexts corresponding to the same text are different
and not related. So, the chaotic encryption system is equivalent to one-time pad
mechanism, and Shannon has proved that it is unable to be deciphered.
3. The system has the ability to resist spectrum analysis. Spectrum analysis is one
of cipher cracking methods often adopted by cryptanalysts. However, because
chaotic signal with wide frequency spectrum is similar to the noise, attackers
cannot get any useful information about the encrypted signals through the spec-
trum analysis method. Therefore, this method has no effect on the output signal of
chaotic encryption system. In addition, there is no correlation between the chaotic
encrypted signal and the original signal, which ensures the security of encryption
system. Any deciphered error to the ciphertext is amplified by the chaotic system,
and it also improves the encryption intensity.
7.3 Realization of Real-Time Voice Encryption System
7.3.1 Structure of Real-Time Voice Encryption System
Based on ACE, RTP, and real-time encryption scheme, a real-time network voice
encryption system is developed. Its structure is shown in Figure 7.10. The whole
system consists of the following modules: voice acquisition/broadcast module, en-
coding/decoding module, encryption/decryption module, and transmission module.
As a real-time system, it is necessary to have a real-time voice acquisition and broad-
cast module. Encryption operation is achieved by using real-time encryption scheme.
The communication protocol is based on RTP, which is transmitted on the Internet
through the network layer and physical layer.
Figure 7.10 is a P2P communication model. Multicast function is realized by using
RTP in voice transmission module. So, it is not only used in P2P communication, but
also can be used in many-to-many communication. The bidirectional arrows in the
figure represent the bidirectional data stream.
The realization of each module is independent of each other. It communic-
ates with other module by interfaces. Each module is realized by the static library
(.lib file), such as VoiceIO.lib of voice capture/broadcast module, Codec.lib of voice
Voice play and acquisition
Voice decoding and encoding
Voice encryption and decryption
Network transmission
Voice encryption and decryption
Voice encoding and decoding
Voice acquisition and play
Figure 7.10: Structure of real-time voice encryption system.
encoding/decoding module, Crypto.lib of voice encryption/decryption module, and

RTPLib.lib of voice transmission module. These static libraries can be used inde-
pendently or in combination with external applications, which greatly enhances the
reusability of each module.
Speech acquisition module records the voice data collected by microphone
through sound card, while speech broadcast outputs the speech data by sound card
and speaker. The signals via microphone and speaker are analog signals, and signals
of computer processing and transmission are digital signals. So the role of sound card
is to achieve the mutual conversion of the two kinds of signals. The conversion process
is related to the sampling frequency, encoding bits, and channels number. Human
voice frequency is about 10 kHz. According to the sampling theorem, the sampling
frequency is chosen as 22.050 kHz. Here, the system uses 16-bit memory Pulse Code
Modulation (PCM) encoding, single channel.
Encoding/decoding module employs A-law compression encoding. Through this
method, 16-bit linear code obtained by sampling can be converted to 8-bit nonlinear
encoding. In this way, the transmission rate can be reduced by half.
Encryption/decryption module uses the encryption scheme based on the 3DES
(extended DES) and chaotic system (logistic map), and block encryption algorithm
is used in the stream communication. Logistic map, synchronization process, and
3DES encryption algorithm are packaged together, and only two interfaces of en-
cryption/decryption are exposed to external. As described in Section 7.2.3, chaotic
synchronization is accomplished by the header of data packets, which include the
Application layer
RTP/RTCP
UDP
IP
Physical layer
Figure 7.11: Application model of network voice stream based on RTP.
iteration number of logistic map, and the iteration number is directly used to generate
the secrete key for encryption/decryption.
Network transmission module is based on RTP, and the network application
model is shown in Figure 7.11. User Datagram Protocol (UDP) protocol is used in the
RTP transport layer to achieve simple multicast function. The speech transmission
module exposes the RTP_Session class to outside, and inside of this module is the
specific implementation of RTP.
The RTP is a transport protocol for multimedia data stream on Internet. RTP is defined
in the transmission conditions of P2P or one-to-many, and its purpose is to provide
time information and achieve flow synchronization. RTP usually uses UDP protocol to
transmit data, but RTP can also work on other protocols such as TCP or Asynchron-
ous Transfer Mode (ATM). When an application starts an RTP session, two ports are
used, one for RTP and the other for RTCP. RTP itself does not provide flow control or
congestion control, but it provides these services by RTCP. RTP algorithm is usually
not implemented as an independent network layer, but as a part of the application
code. During the RTP session, the participants periodically transmit RTCP packets.
The RTCP packet contains some statistical information such as the number of sending
packets and lost packets, so that the server can use these information to dynamically
change the transmission rate, and even change the effective load type. With the use
of RTP and RTCP, they can optimize the transmission efficiency by effective feedback
and minimum overhead, which is especially suitable for the transmission of real-time
data on Internet.
The software running platform of voice encryption system based on RTP is Win-
dows 2000 professional or Red Hat Linux 9.2. The development tools are Rose Rational
2000, C++ Visual 6, g++, and emacs. The third library is ACE 5.4 and Standard
Template Library. The software framework is shown in Figure 7.12.
7.3.2 Real-Time Transmission Protocol and Its Data Structure
The RTP data packet is composed of two parts, the RTP header and the continuous
media data with indefinite length, and its format is shown in Figure 7.13, where V
represents the version number of RTP specified as 2. P is the padding bit. When P = 1,
Voice input/output
Voice compression encoding/decoder
Voice encryption/decryption
RTP/RTCP
ACE network communication library
Socket interface
Figure 7.12: Software architecture diagram.
it indicates that the data packet tail is loaded with a nonload information, which can
be used to encrypt or notify the bottom layer protocol. A data unit has several RTP
packages. X is an extension bit, and when X = 1, it indicates that the RTP header is
attached to an extended header with variable length. M is used to mark the important
event in the data stream, such as the frame boundary. Load type is used to identify
the type of media in the receiver. If it is the encoding data, it determines the decoder
of the receiver. The serial number is used to identify the order of the RTP packets sent
by the sender, and the serial number increases by 1 when an RTP packet is sent. The
receiver detects the loss situation of data packet by serial number, and then the packet
is recovered after reordering. The time stamp reflects the sampling interval of RTP
packet in the first byte. The sampling frequency must be derived from a clock with
linear growth, and the clock frequency usually depends on the data transmission
format. If the RTP packet is periodically generated, then the sampling frequency is
determined by the sampling clock, rather than the system clock. The synchronization
source identifier (SSRC) identifies the synchronization source, which is randomly
generated, and the SSRC identifier of two synchronization sources in the same RTP
session is different. Contributing source identifier (CSRC) can contain 0–15 items.
Contributing source refers to a new combination of RTP packets after the procession
of mixer when one or more of RTP data packets are received from the synchronous
source. The mixer is used as the SSRC of the combination RTP packet, and all the
SSRC are transmitted to the receiver as a special source identifier, so that the receiver
knows all the SSRC of RTP packets.
Another protocol is RTCP, which is usually used with RTP, and it can be used to
carry out the feedback about transmission quality (such as the jitter and average loss
rate) and transfers some identity information of participants. RTCP packet contains
the following formats:
SR: source report package; it is used to send and receive the statistics of active
source.
RR: receiver report package; it is used to receive the statistics of nonactive
station.
V P X CC M Load type Serial number
Timestamp
SSRC
CSRC
Data payload
Figure 7.13: RTP packet format.
SDES: source site description information packet; it is used to report the related
information of site, including the specification name CNAME.
BYE: send this packet when you want to leave.
APP: application package; it contains custom functions.
An RTCP composite package usually contains several RTCP packages. To perform this
protocol, the following two points must be observed. First, in the case of the band-
width allowed, receiving statistics (RR or SR) must be transmitted as soon as possible,
so that the accuracy of statistics is improved. Therefore, each transmitted RTCP com-
posite package must contain a report package. Second, the new source site must
receive CNAME as soon as possible to verify the other source sites. So each RTCP
composite package must contain a CNAME SDES too.
With the aid of the above control packet, RTCP can complete four control func-
tions, including quality of service monitoring control, congestion control, flagging
media synchronization, providing identification information, and estimation and
planning of session size.
7.3.3 Implementation of RTP Based on Adaptive Communication Environment
ACE is an object-oriented (OO) toolkit, which implements many basic design patterns
of communication software. The target user of ACE is the developer who develops
high-performance communication services and applications on UNIX or Win32 plat-
form. ACE simplifies the development of communication between processes, event
channel separation, display dynamic link, and concurrent OO network applications
and services. By dynamically linking services and applications at run time and ex-
ecuting these services in one or more processes or threads, ACE enables the system’s
configuration and reconfiguration to be automated.
As shown in Figure 7.14, the hierarchical architecture of ACE consists of three
basic layers: uses layer of operating system (OS), C++ packet layer, and frame com-
ponent layer [155]. ACE contains a lot of reusable packaging, framework, and classes,
GATEWAY TOKEN LOGGING NAME TIME

ACE frame
SERVER SERVER SERVER SERVER SERVER
component
layer
ACCEPTOR CONNECTOR SERVICE HANDLER CORBA HANDLER
ADAPTIVE SERVICE EXECUTIVE (ASX)
PROCESS/ THREAD MANAGERS LOG MSG SERVICE SHARED MALLOC

REACTOR/ CONFIG-
C++ SOCK-SAP FIFO MEM SYS V
SYNCH SPIPE PROACTOR URATOR
Packet WRAPPERS SAP TLI-SAP SAP MAP WRAPPERS
class
Operating system adaptation layer
C PROCESSES/ STREAM SOCKETS/ NAMED SELECT/ DYNAMIC MEMORY SYSTEM

THREADS PIPES TLI PIPES IO COMP LINKING MAPPING V IPC
APIs
Process/threadsubsystem Communicationsubsystem Virtual memory subsystem
Public POSIX and Win32 services
Figure 7.14: Structure of ACE and components.
which can be used to perform common network programming tasks across many
OSs. ACE components can be used to achieve the following objectives: event de-
multiplexing and event processor allocation, connection establishment and service
initialization, inter-process communication and shared memory management, dy-
namic configuration of distributed communication services, concurrency/concurrent,
and synchronization.
The important ACE modules in this system are presented as follows:
1. SOCK_SAP. It is inherited from the IPC SAP, and it is the OO package of the bottom
layer network programming interface.
2. ACE reactor. It is the OO framework for event demultiplexing.
3. ACE_Task class. It is the center of the active object and the passive object, which
is used to create a user definition and process application messages in ACE.
The structure of RTP library is shown in Figure 7.15. Among them, RTP_Session module
is the core module of the whole protocol library and is also the only interface ex-
posed to the upper-level users. Users can control this module to start real-time data
communication, and it is responsible for creating the RTCP_Recv_Handler module
and RTCP_Timer_Handler modules. RTP_Sender is used to send RTP data packets.
RTP_Reactor_Task is a thread object. It is the central hub for the protocol part, and
it is also an I/O event dispatcher. RTP_Reactor_Task is used to create three classes,
including RTCP_Recv_Handler, RTCP_Timer_Handler, and RTP_Recv_Handler. Then
it records the arrival events of RTP and RTCP packets and the timing events of RTCP
transmission. Finally, it calls these modules when these events come.
Sender _ report RTP_Time stamp SDES_table
RTP_Sender Dst _ I P_Table
RTCP_Queue RTP_Session RTP_Playload_Queue
RTP_Reactor _Task RTP_Buffer_Stradegy
RTCP_Recv_Handler RTCP_Timer _Handler RTP_Recv_Handler
RTCP_Processor RTCP_Packet _Creator RTP_Hdr - Processor
Sender _ report RTP_Time stamp SDES_ table
Figure 7.15: Structure of RTP library based on ACE.
RTP_Recv_Handler module is used to process the arrival event of RTP data packets.
It will receive the RTP message when the event arrives. RTP_Hdr_Processor module
is used to process the RTP message and extracts the valid data. RTCP_Recv_Handler
module is used to receive RTCP message, which has a similar function with
RTP_Recv_Handler. RTCP_Processor module is used to process the RTCP message
and obtain its statistical information. RTCP_Timer_Handler module is used to send
RTCP message. RTCP_Packet_Creator module is used to generate RTCP message. The
Sender_Report module is used to set up and save some statistical information in
the RTCP message. SDES_Table module is used to save and set SDES information.
DST_IP_Table module is used to save the IP information of source site that is involved
in the communication. RTP_Timestamp module provides some operations related
with time, such as accessing system time or RTP timestamp. RTP_Payload_Queue
module is used to save the received RTP data. RTCP_Queue module is used to save
the received statistical information of RTCP.
7.3.4 System Debugging and Performance Testing
Test project RTVC (real time) is applied to debug each module of the system. CContoller
class is used to combine and coordinate each module, which is responsible for re-
ceiving user’s parameter settings (audio parameters, secrete key, and communication
members) and coordinating the transmission of data stream among each module.
The problems encountered during the debugging process as well as their solution
methods are illustrated as follows.
During the test of speech acquisition/playback module, when you stop record-
ing, you are prompted that recording buffer still has data, and the stop recording
operation failed, which results in the failure of recording operation again. The error
reason is that the recording buffer is released in the message processing function
WM_MIM_DONE, and it is not fully released when the recording is stopped. Therefore,
the recording buffer memory is managed automatically. When you stop recording, the
recording thread is hanged up (also the execution of message processing function
WM_MIM_ DONE is suspended), and then the recording buffer is released explicitly.
The recording thread is restored when you restart recording.
In addition, communication often automatically terminates when the system
works for a period of time. Through testing and tracking, it is found that the output rate
of data queue is faster than that of input, which results in that the playback module
occasionally does not get data, and then the playback module will be shut down auto-
matically. So communication is terminated. To solve this problem, the whole process
is controlled manually. When the play module cannot obtain the data, this module is
still working, but it does not fill data to the playing buffer.
The communication system is tested by the release version. The results are shown
in Table 7.1. In the conditions of 22.050 kHz sampling frequency, 16-bit quantization
data, single channel, and 1,024 bytes recording buffer (sampling time is approxim-
ately 22.68 ms), the effect of common mode (no encryption) is best, and the delay
cannot be felt. The effect of encryption mode is second, and there is a little noise, but
still can be understood. For the use of equal interval drive of the continuous chaotic
network synchronization method, the system works in synchronized state. Even if a
data packet header is damaged, it has no effect on communication process.
In the conditions of 22.050 kHz sampling frequency, 8-bit quantization data,
single channel, and 1,024 bytes recording buffer (the sampling time is 45 ms), the
communication effect of common mode is apparently worse than the above mode,
and there are some time delay. The reason is that its sampling time is too long. The en-
cryption mode of communication is not as good as the common mode, but its security
is superior to the common mode.
Table 7.1: Results of system test using the published version (release).
Mode Quantization Sampling Processing Transmission Voice Safety

bits delay delay delay quality performance
Common 16 22.68 0.125 0 4.5 Poor

Encryption 16 22.68 ≈8 0 4.2 Excellent
Common 8 45 0 0 4.0 Poor
Encryption 8 45 ≈8 0 3.9 Excellent
In this section, we study and develop a real-time voice encryption system based
on chaotic map and 3DES encryption algorithm. The system has the characteristics
of fast encryption speed, less encryption time, and strong ability of anti-breaking. In
the implementation of the whole system, the modular design and multi-thread tech-
nology are used to minimize the coupling of each packaging module and improve
their reusability. In the implementation of RTP/RTCP, the use of cross-platform ACE
greatly improves the cross-platform characteristics of RTP/RTCP module (RTPLib.lib),
and it can be ported to Windows 2000 Professional, Linux RedHad 9.2, and other OSs.
Through the network test, the encryption system can realize the real-time encryption
and transmission of voice signals, and it has good communication quality and high
security. However, it is necessary to improve and optimize the encryption speed and
signal delay for commercial use.
7.4 MPEG Video Encryption Algorithm Based on Chaotic Sequence

Video image data has a strong correlation, i.e. there is a large amount of redundant
information. The redundant information can be divided into spatial redundant
information and temporal redundant information. Compression technology is used to
remove the redundant information in data stream (to remove the correlation between
data), and it includes the intra-image data compression technology, inter-image data
compression technology, and entropy encoding compression technology. At present,
the commonly used video compression algorithms are MPEG, H.26X based on dis-
crete cosine transform (DCT), and compression encoding based on discrete wavelet
transform. H.26X standard is mainly used for real-time video communication, such as
video conference and video phone. MPEG standard is mainly used for video storage
(DVD), video broadcast and video streaming media, such as Digital Subscriberloop
(DSL), Internet video, and wireless video. Here, we focus on MPEG video encryption
algorithm based on chaos.
In recent years, the encryption algorithm combined with the compression al-
gorithm is applied to MPEG video stream, which has received extensive attention.
Bhargava et al. proposed an encryption algorithm, which the secrete key is used to
vary the coefficients of DCT and sign bit of motion vector [156]. Based on this method,
researchers proposed an algorithm to further encrypt the intra-DC coefficients. A per-
ceptual video encryption algorithm (PVEA) is proposed by encrypting the fixed length
code data of all compressed video stream. Due to the smaller amount of encrypted
data, the compression algorithm has little effect on video coding and decoding and
will not affect the compressed format. But there is still a defect about the security of
MPEG. To maintain a fast encryption speed, the above algorithms often use a single
stream cipher to carry out data encryption. But the stream cipher cannot resist known
plaintext attack. Other optimization measures, such as the scrambling and the plain-
text or ciphertext feedback mechanism, are gradually adopted. Considering that both
the DC and AC coefficients have a very important effect on encryption, researchers ad-
vised to shuffle all DCT coefficients of the same frequency location [157]. To maintain
the size of data, some researchers proposed a scheme to shuffle the variable length
code or run-length encoded (RLE) event [158, 159] before the entropy encoding, and
the same or better results can be obtained from these cryptographic algorithms.
7.4.1 Encryption Principle and Procedure of MPEG Video Data
A practical MPEG video encryption algorithm should meet the requirements of the
invariance of compression format and ratio, the less computation quantity, and the
widely adjustable security level. Encryption of fixed length code data can meet these
requirements at utmost extreme, but the encryption degree and security need to be
improved further. The analysis and experiment results show that the effective tex-
ture information is distributed in all the DCT coefficients. Scrambling for internal
macro-block of each video frame is an effective means to hidden information, which
is added to the fixed length data encryption algorithm. In general, the entire MPEG
video encryption scheme consists of two parts as follows [160].
Step 1: The data are encrypted including the intra-DC coefficients, non-intra-DC
coefficients, sign bit of AC coefficients, ESCAPE DCT coefficients, motion
vector symbol, and residual fixed length code.
Step 2: The bit stream of macro-block is scrambled, and each frame is assigned a
different scrambling table.
As all the operations are done on the video stream, the entire encryption algorithm has
no effect on the compression ratio. Stream cipher and scrambling table are generated
by the chaotic map. The principle block diagram of encryption algorithm is shown
in Figure 7.16. It shows the complete process of chaotic MPEG video encryption. To
reduce the chaotic iterations, we uses the same Cat map (with the same initial con-
ditions and parameters) to generate the stream cipher and permutation table. So, the
sequence cipher and block cipher (permutation) will share the same set of secrete key
parameters {x1 (0), y1 (0), u, v}. Of course, to enhance the security, we can use chaotic
map with different initial conditions and parameters to generate chaotic sequences
for different encryption processes. In this case, the stream encryption and scrambling
have independent secrete key, but it requires more chaos iteration and the amount of
computation.
From Figure 7.16, the key space of whole encryption system is composed of {x1 (0),
y1 (0), u, v, p, x2 (0)}. In the experiments, secrete key is reset for each image group. You
can change the secrete key to improve the security of system by setting the step size
vector d = {du , dv , dp , dx1 , dy1 , dx2 }. The decryption process is the symmetric operation
of the encryption process.
The details of encryption process will be presented in the following sections.
x2 (0), p x1 (0), y1 (0),u,v
Discrete PLCM Discrete cat map
x2
x 1 , y1 x1
z(i) = (x1 (i) + x2 (i)) mod232

Sort transformation
w(i) = y1(i)XORz(i)
z,w ã
Extraction of
Extraction of bits
scrambling table
s p
Video stream Encryption of Macro-block Encrypted data

FLC data scrambling
Figure 7.16: Principle block diagram of MPEG encryption based on chaos.
7.4.2 Design of Stream Cipher Based on Chaos
Chaos-based cryptography has been widely used in video encryption. To design a fast
and secure chaotic cipher, we should using fixed point operation instead of floating
point operation, using the simple chaotic system, a large encryption unit, and so on.
So two simple chaotic maps are selected to design the stream cipher: they are cat map
and piecewise linear chaotic map.
Cat map is defined by

x1 (k + 1) 1 u x1 (k)
= mod 1, (7.7)
y1 (k + 1) v uv + 1 y1 (k)
where u and v are control parameters.

Piecewise linear chaotic map is defined by
⎧
⎪
⎪ x2 (k)/p, 0 ≤ x2 (k) < p
⎪
⎨
(x2 (k) – p)/(0.5 – p), p ≤ x2 (k) < 0.5
x2 (k + 1) = , (7.8)
⎪
⎪ (1 – p – x2 (k))/(0.5 – p), 0.5 ≤ x2 (k) < 1 – p
⎪
⎩
(1 – x2 (k))/p, 1 – p ≤ x2 (k) < 1
where p is the control parameter.

To avoid floating point operations, the above two maps are discretized to operate
in the integer domain. The discretized cat map is described by

x1 (k + 1) 1 u x1 (k)
= mod 232 , (7.9)
y1 (k + 1) v uv + 1 y1 (k)
where u and v are unsigned integers, and initial conditions x1 (0) and y1 (0) are re-
stricted to unsigned integers. Obviously, after the discretization, x1 (k), y1 (k) ∈ [0, 232 ),
and they are integers, which ensure that all operations are fixed point operation. The
discretized piecewise linear chaotic map is described by
⎧
⎪
⎪ 232 ⋅ x2 (k)/p , 0 ≤ x2 (k) < p
⎪
⎨
232 ⋅ (x2 (k) – p)/(231 – p) , p ≤ x2 (k) < 231
x2 (k + 1) = (7.10)
⎪
⎪ 232 ⋅ (232 – p – x2 (k))/(231 – p) , 231 ≤ x2 (k) < 232 – p
⎪
⎩ 32
2 ⋅ (232 – x2 (k))/p , 232 – p ≤ x2 (k) < 232
where x represents an integer that is no more than x, and integer p ∈ (0, 231 ).
Correspondingly, x2 (k) is an integer that belongs to [0, 232 ).
Thus, x1 , x2 , and y1 are three different chaotic integer sequences generated by two
chaotic maps. To avoid the degradation of chaotic characteristics after discretization,
the following improvement measures are taken. First, two chaotic maps are iterated
for 50 times as pretreatment. Then, starting from the 51th iteration, the generated data
is processed to obtain two new integer sequences z and w as follows:
z(i) = (x1 (i) + x2 (i)) mod 232 , (7.11a)

w(i) = y1 (i)XORz(i), (7.11b)
where XOR is the XOR operation between bits, and mod is modulo operation. This
operation not only reduces the degradation of original chaotic systems but also im-
proves the complexity of the generated chaotic sequence. The new sequences, z =
[z(0), z(1), . . . , z(n – 1)] and w = [w(0), w(1), . . . , w(n – 1)], have better random prop-
erties. Then, each of them is transformed into the corresponding binary form, i.e.
z(i) = b0 b1 . . . b31 and w(i) = b′0 b′1 . . . b′31 (b′j and bj are binary numbers, j = 0, 1, . . . , 31).
A binary random sequence s is obtained by extracting the bits of integer random
sequence z and w, and alternately permuting them. Therefore, when the Cat and Piece-
wise Linear Chaotic Map (PLCM) maps iterate one time, a 64-bit secrete key stream
b0 b′0 b1 b′1 . . . b31 b′31 is produced.
Through a series of random tests, it is proved that the sequence s has good ran-
domness. First, set the initial conditions of u = 20, v = 2, p = 220 , x1 (0) = 600,
y1 (0) = 100, and x2 (0) = 1, 000. Then, for each of iteration, bits of z(i) and w(i) are al-
ternates extracted to obtain three binary random sequences s1 , s2 , and s3 with length
of 200, 2,000, and 20,000, respectively. These binary random sequences are tested
Table 7.2: Random test results for different length sequences.
Sequence length Runs test

Balance test
1-Run 2-Run 3-Run 4-Run 5-Run
200 0.0552 0.4754 0.2452 0.1321 0.0632 0.0352

2,000 0.0215 0.4895 0.2485 0.1239 0.0624 0.0315
20,000 0.0047 0.4977 0.2496 0.1248 0.0627 0.0313
1 1
(a) (b)
R (k)
R (k)
0.5 0.5
0 0
–1 0 1 2 –1 0 1 2
k X 104 k X 104
1 1
(c) (d)
R (k)
R (k)
0.5 0.5
0 0
–1 0 1 2 –1 0 1 2
k X 104 k X 104
Figure 7.17: Autocorrelation and cross-correlation tests of chaotic sequence: (a) autocorrelation;
(b) cross-correlation of two sequences with different parameter p; (c) cross-correlation of two
sequences with different x1 (0); and (d) cross-correlation of two sequences with different parameter u.
randomly, and the results are shown in Table 7.2. As shown, all of them pass the test
and show excellent performance.
To ensure the security of the cipher, it is necessary to carry out the correlation test.
The sequences with different lengths are chosen to carry out the autocorrelation and
cross-correlation tests. Test results are shown in Figure 7.17.
Obviously, the results of autocorrelation tests are ideal as shown in Figure 7.17(a). Us-
ing two different values of p, x1 (0), and u, respectively, two different chaotic sequences
with the same length are obtained to make the cross-correlation test, and the res-
ults are shown in Figure 7.17(b–d). As shown, the test results are also satisfied, which
means that the chaotic binary sequence has high sensitivity to parameters of chaotic
system.
In this scheme, the sequence s as the stream cipher is used to encrypt the fixed
length data of video stream, and the ciphertext feedback mode is adopted to resist the
known plaintext attack in the encryption process.
7.4.3 Design of Scrambling Algorithm Based on Chaos
The algorithm requires that each frame takes different scrambling tables to scramble
macro-block. Here, the sorting-based method is adopted, and it has advantages on
simple operation and fast running speed. We suppose that each frame is made up of
N macro-blocks. The specific scrambling process for each frame is as follows.
1. A sequence with length N, a = {a1 , a2 , . . . , aN }, is extracted from the chaotic
integer sequence. Here, the chaotic sequence is generated by the same chaotic
map with the upper stream cipher. That is to say, a sequence with length N can
be obtained from the chaotic sequence x1 , y1 or x2 , and x1 is selected in this
experiment.
2. Rearrange the sequence in ascending or descending order, and then a new se-
quence is generated as ã = {ã1 , ã2 , . . . , ãN }. This method is feasible because there
do not exist the same elements of chaotic sequence.
3. Record the location in new sequence {ã1 , ã2 , . . . , ãN } of each element in the ori-
ginal sequence a = {a1 , a2 , . . . , aN } to get the corresponding address sequence
p = {p1 , p2 , . . . , pN }.
4. Use the scrambling addresses p = {p1 , p2 , . . . , pN } to scramble the macro-block
after the current frame is scanned.
The above scrambling method does not increase the redundancy iteration, which
makes the whole process simple and quick.
7.4.4 Encryption Experiment and Encryption Effect Analysis
The experimental environment are the open-source codec of MPEG-4, xvidcore-1.1.2.

XVID has high encoding speed, and it is considered to be the fastest MPEG-4 encoder
at present. Moreover, the XVID encoder is designed for Windows users. The source
code is completely open, and the quality of image compression is close to DVD quality.
But the data quantity is only one-eighth of DVD. So XVID encoder is the best choice
for Windows users.
XVID encoder is developed in the UNIX OS and C++6.0 Visual environment. When
it is used in the Windows system, you must remove the UNIX format first. The spe-
cific method is to use the writing board to open the ∗ .dsw, ∗ .dsp, and Makefile files,
and then save them directly. So it does not appear that there is no phenomenon of
engineering loading when opening the xvidcore.dsw file.
This encryption algorithm is combined with compression algorithm. The basic
idea of encoding includes initializing the encoder, creating an encoding instance
handle, circular encoding for each frame data, and destroying the existing encoding
instance handle to stop encoding. Encryption algorithm is embedded in compressed
encoding. For each frame of the image data, the process is shown in Figure 7.18. First,
Sampling
DCT transform
Quantification
N
Fixed length encoding data?
Stream encrption
Entropy encoding
Macro-block
scrambling
Output video
stream
Figure 7.18: Flowchart of the encryption algorithm combined with compression encoding.
the encoder samples the pixels, and then do DCT. The encryption of the fixed length
data is carried out after quantification, and the scrambling of macro-block is done
after the entropy encoding. To do so, it will not affect encoding.
According to this procedure, a series of standard test sequences are tested. Figure 7.19
is the encryption effect diagram for two different standard test sequences. Fig-
ure 7.19(a) and 7.19(c) is the first frame image of foreman and akiyo, respectively.
Figure 7.19(b) is the first frame decoding image of foreman only using the macro-block
scrambling method, and Figure 7.19(d) is the first frame decoding image of akiyo using
the complete encryption algorithm.
Therefore, attackers cannot get any valuable information from the scrambled and
encrypted image. This scheme has good property of format invariance and its control
parameters can be further set up to achieve different security levels.
7.4.5 Performance Analysis of MPEG Video Encryption System
The performances of encryption system are analyzed mainly from the following two
aspects.
(a) (b)
(c) (d)
Figure 7.19: Video encryption effects: (a) foreman plain image; (b) effect of macro-block scrambling;
(c) akiyo plain image; and (d) full encryption effect of this scheme.
7.4.5.1 Analysis of the System Security

As discussed previously, for a group of images, key space of the chaotic encryp-
tion system consists of {u, v, p, x1 (0), y1 (0), x2 (0)}, which is large enough to resist a
brute-force attack. Considering the scrambling process with the size N, for frame p,
the majority of macro-blocks come from the corresponding prediction of the previous
frame. Because the previous frame has been scrambled, attackers are unable to restore
the current image. The same results can be obtained for frame B. For frame I, even in
the qicf video clips, the size of scrambling is N = 176 × 144/162 = 99. So the search
space is 99! which is large enough to resist the brute-force attack too.
In the fixed length code data encryption, the feedback operation makes secrete
key stream to depend on the whole plaintext, and thus can resist the known/chosen
plaintext attack. In the scrambling process, if an attacker obtains a pair of plaintext
and ciphertext, it is clear that the scrambling table can be obtained by comparing.
However, in this system, the scrambling table changes according to the chaotic se-
quence, and the evolution of chaotic sequence is difficult to be predicted. Therefore, it
is not possible to infer the original chaotic sequence from the scrambling algorithm. In
addition, the secrete key is changed in real time with the change of images, resulting
in that the attack is more difficult.
Experiments show that this algorithm can also resist error concealment attack
(ECA). In our experiments, different encryption algorithms are used to encrypt a
176 × 144 akiyo video fragment, and then they are attacked by ECA as shown in
(a) (b)
Figure 7.20: Results of ECA attack on the first frame of “akiyo”: (a) encryption algorithm with fixed
length data and (b) new algorithm.
Table 7.3: Results of encryption speed test.
Video clips Size Ratio (%)
Akiyo 176 × 144 2.5

Carphone 176 × 144 1.9
Foreman 176 × 144 2.1
Miss America 176 × 144 1.6
Dancer 352 × 288 3.1
Singer 352 × 288 2.4
Figure 7.20. As can be seen, if a single compressed video stream is encrypted by using
fixed length data, a basic outline can be seen after the ECA attack. But it is completely
unable to be identified by using our encryption scheme.
7.4.5.2 Analysis of Encryption Speed

Experiments show that the new encryption algorithm has a fast speed. By employing
two simple chaotic maps and iterating the chaotic maps in the integer domain, the
computation time of encryption system is greatly reduced. On the other hand, because
of the selected finite fixed length code data to be encrypted and simple scrambling
scheme, the average number of iterations is greatly reduced. Taking several 90 frames
video fragments as example, the ratios of encryption time and compression time are
tested and recorded. Results are shown in Table 7.3. As can be seen, the time cost of
encryption process has little effect on the MPEG4 compression, and this scheme can
meet the real-time communication requirements.
7.5 H.264 Video Encryption Algorithm Based on Chaos

H.264 is a video encoding standard with extremely low bit rate, which is adopted by
the International Telecommunication Union. It is the tenth part of MPEG-4 and is also
called advanced video coding by the International Organization for Standardization

(ISO). H.264 standard has better image quality, higher compression ratio, and better
network adaptability than that of the previous video encoding standard. However,
its good encoding effect is at the cost of a huge amount of computing. Therefore,
the encryption of H.264 should be considered as a simplified encryption algorithm,
which can reduce the computational complexity and the cost of secrete key, and
simultaneously the compression effect does not be affected.
In this section, a new chaotic video encryption scheme is designed.
7.5.1 Chaotic Stream Cipher Based on Tangent Delay-Ellipse Reflecting

Cavity System
Chaotic stream cipher is constructed by tangent delay-ellipse reflecting cavity system

(TD-ERCS) chaotic system. In fact, the TD-ERCS chaotic system is an iterative map,
which is discussed in Section 3.1.4. Iterate the chaotic map for one round at each
time, and then the tail numbers of results are extracted to construct pseudorandom
binary sequence. The generated sequences x and k are denoted as {xj |j = 1, 2, 3, . . . }
and {kj |j = 1, 2, 3, . . .}, where xj and kj represent the iteration values x and k in round
m + j – 1. The binary codes of kj and xj mantissa are extracted from low to high and
are denoted as {KjP |P = 1, 2, . . . , 51, 52} and {XjP |P = 1, 2, . . . , 51, 52}, respectively. The
binary pseudorandom sequence s of TD-ERCS is obtained by alternant arranging KjP
and XjP . As it can be seen, when the TD-ERCS goes into normal state, each round of
iteration can generate 104-bit binary sequence, which is 13 bytes.
According to the characteristics of H.264 standard, the selective encryption al-
gorithm is investigated based on chaotic pseudorandom sequence. The main feature
of this algorithm is that chaos is introduced to enhance the security of system and the
encryption speed is improved. The encryption algorithm has low complexity and good
performance on security and real time. It has little influence on the compression per-
formance of encoding and has certain data operability, which is consistent with the
basic requirements of video data encryption.
7.5.2 H.264 Video Encryption Scheme
To reduce the amount of data to be encrypted, the most effective data must be selected
for encryption. Because the human eye is more sensitive to the brightness information
than the color information, and the brightness information and color information are
contained in the compressed file separately. Here a few key data are selected to en-
crypt. The basic encryption framework is shown in Figure 7.21. The encryption content
of this scheme includes all sign bits of DC and AC coefficients in 4 × 4 luminance
block, the sign bits of DC coefficients in chroma block, and the sign bits in motion
vector of P/SP frame and B frame. The reason for choosing these data to encrypt is
Video signal
Intra-prediction Inter-prediction
Integer change/ Motion Motion

quantization compensation estimation
Motion
Sign bit of coefficients? vector
N
Y
Chaotic stream
encryption
Entropy encoding
Pack
Encrypted
video stream
Figure 7.21: Flowchart of H.264 encryption process.
because those data can maximize the impact of visual effects and maintain the basic
compression ratio and signal-to-noise ratio (SNR) of encoding at the minimum cost.
The chaotic stream cipher is designed by using the above pseudorandom se-
quence s. To improve the security, the following ciphertext feedback mode is adopted.
Encryption process is described by
C(i) = XOR(s(i), C(i – 1 – s(i))), (7.12)
where XOR is the bit XOR operation; s is the secrete key stream; and C(i) is the
ciphertext.
Experiments are carried out on the H.264 open-source codec JM. JM is one of the
three official encoders of H.264. Although the algorithm is complex and its speed is
slow, it is suitable for the second development. To run JM in the VC6.0 compiler, you
(a) (b)
(c) (d)
(e) (f)
(g) (h)
Figure 7.22: H.264 video encryption effects. (a) First frame of foreman. (b) First frame full encryption
of foreman. (c) Encryption symbol bit of brightness of DCT coefficient. (d) Encryption symbol bit of
chroma DC coefficient. (e) mv encryption for frame 89 of foreman. (f) Full encryption for frame 89 of
foreman. (g) Mobile plain image. (h) Full encryption of mobile image.
have to set up configuration file before the coding and decoding. For example, the
encoding configuration file can be set to encoder_baseline.cfg.
In this experiment, the encryption effect of different data in the standard video se-
quence is recorded as shown in Figure 7.22. As it can be seen, different data can be en-
crypted to achieve different results. For the encryption of motion vector, an inevitable
result is that the encryption effect is not obvious for a set of video sequences at the be-
ginning as shown in Figure 7.22(b). With the accumulation of motions, the ciphertext
information of encrypted motion vector will continue to be accumulated, and the
effect is more and more obvious as shown in Figure 7.22(e) and 7.22(f). After full encryp-
tion, the image has become very blurred as shown in Figure 7.22(f) and 7.22(h). To im-
prove the security of encryption further, we can scramble the frequency coefficient, 4×
4 change block or Flexible Macroblock Ordering (FMO)-based macro-block. Of course,
those algorithms are based on the cost of compression ratio and encryption delay.
7.5.3 Performance Analysis of Video Encryption
7.5.3.1 Security Analysis of the Encryption System

On one hand, for the chaotic pseudorandom sequence generator based on the TD-
ERCS map, the total secrete key parameters are (x0 , tg!, ,, m), and small changes in
parameters will generate a completely different sequence. Obviously, the key space is
large enough to resist the brute-force attack. At the same time, because of the real-time
variation of the chaotic iteration parameters and initial conditions in each image, the
algorithm has the ability to resist the chosen plaintext attack.
On the other hand, for the selective encryption scheme, encrypted data includes
the DCT coefficients, which represent the texture information, and the motion vector,
which represents the motion information. To do so, the readability of video data can
be greatly damaged.
7.5.3.2 Real-Time Analysis of the Encryption System

The encryption speed depends on the producing speed of the chaotic stream cipher
and the amount of encrypted video data. For TD-ERCS, the system can generate 104-
bit secrete key stream for each round of iteration. For encrypted data, the proposed
encryption scheme only selects the key data of video stream and only involves the sign
bit. For example, considering that the majority of AC coefficients of chroma are zero,
the encryption effect is not affected. So the DC coefficients of chroma are chosen to be
encrypted. Experiments are carried out on several standard sequences with Quarter
Common Intermediate Format (QCIF) format, and the real-time performance is shown
Table 7.4: Results of encryption speed.
Video clips Size Frame number Ratio (%)
Carphone 176 × 144 90 1.26

Foreman 176 × 144 90 1.82
Mobile 176 × 144 300 2.76
Dancer 352 × 288 150 3.05
Questions 237
in Table 7.4. The percentage of encryption time accounting for encoding time shows
that the encryption system has good real-time performance.
Any encryption method may have an impact on the encoding effect of video se-
quences. Practical encryption algorithm should reduce the influence of compression
ratio and SNR of video sequences. Because only the sign bits of video data are en-
crypted, the encryption algorithm has little impact on the SNR and compression ratio
of video sequence. This algorithm takes the macro-block, the frame, and the image
group as the encoding unit, so it can keep the frame synchronization information of
the video data to be invariable, and it has certain data operability, such as supporting
the operations of position index, pasting, and cutting. For the encryption to DCT and
quantization coefficients, it does not change any information about data format. The
experimental results show that the SNR and compression ratio are not affected.
Questions
1. What are the technical difficulties of real-time data stream encryption?
2. How to evaluate the security of real-time data stream encryption?
3. How to deal with the relationship between encryption and compression in video
transmission?
4. Design a wireless transmission scheme based on chaos.
8 Analysis and Simulation of Fractional-Order
Chaotic System
With the further research of nonlinear chaotic theory, fractional-order chaotic system
has become a hot topic. To apply the fractional-order chaotic system in the field of in-
formation security, it is necessary to reveal its characteristics. Moreover, key technical
problems are required to be solved quickly, such as accurate solution of fractional
differential equations, digital signal processing (DSP) circuit realization, and so on.
In this chapter, we focus on solution algorithm and simulation technology. It is the
theoretical and experimental basis for applications of fractional-order chaotic system.
8.1 Development of Fractional-Order Chaotic System

Theoretical research of fractional calculus has more than 300 years history, and it has
the same long history as the integer-order calculus. However, due to the lack of ap-
plication background, its development is relatively slow. Actually, fractional dynamic
behavior can be observed in many physical systems, such as viscous system, dielectric
polarization, colored noise, electromagnetic waves, and so on. Compared to integer-
order calculus equation, fractional-order equation can describe nature phenomenon
more accurately in applied science areas [161] such as memory of various materi-
als, mechanics and electrical characterization, viscoelastic dampers, power fractal
network, fractional-order sinusoidal oscillator, robots, electronic circuits, electro-
lytic chemical, fractional capacitance theory, fractal theory, fractional proportional–
integral–derivative controller, viscoelastic vibration control system, flexible configur-
ation object, and so on. Since Mandelbort first reported that there are numerous fractal
phenomena in nature in 1983 [162], fractional calculus has become a hot topic and has
developed quickly.
At the beginning, the investigation of fractional-order chaotic system was based
on the research of integer-order chaotic system. When fractional differential operator
is introduced to nonlinear dynamic system, the fractional-order system still exhibits
chaotic behavior, such as Chua circuit [163], a nonautonomous Duffing system [164],
improved Duffing system [165], fractional Jerk system [166], Rössler system [167], and
so on. Although dynamic characteristics of some fractional-order chaotic system have
been investigated, there still remain many other fractional-order chaotic systems to
be studied.
Up to now, there are many definitions about fractional-order calculus [168]. Li and
Deng studied three definitions of fractional-order calculus deeply [169]. Among these
definitions, Riemann–Liouville (R-L) definition and Caputo definition are usually
used in practice. Although the R-L definition can simplify the calculations of frac-
tional differential, its physical meaning is not clear and its value is difficult to be
8.1 Development of Fractional-Order Chaotic System 239
measured. However, when using Caputo definition, its physical meaning of fractional
differential is clear and its value is easy to be measured. Based on these two defini-
tions, people proposed two different approximate methods for solving the fractional
differential equation. One is time–frequency method based on R-L definition [170].
First, the fractional integral operator in the time domain is transformed into transfer
function in frequency domain. Then piecewise linear approximation method is used
to realize approximation calculation in frequency domain. The other is predictor–
corrector method based on Caputo definition [171–173]. It is a time domain method.
The time series of the fractional-order system are obtained by Adams–Bashforth
prediction equation and Adams–Moulton correction equation. However, frequency
domain method cannot reflect the dynamic characteristic of fractional-order chaotic
system accurately. As for the time domain method, the calculation step can be set
small enough for more accuracy, but its calculation is more complex relatively.
By far, the chaotic characteristic study of fractional-order system is mainly based
on existing integer-order chaotic system. The main idea is to observe whether the sys-
tem is still chaotic when the system order changed from integer to fractional. However,
the system order when system entering into chaotic still needs further research, as well
as dynamic characteristic of fractional-order chaotic system. For example, Ahmad re-
ported that when the order of system is 2 + % (0 < % < 1), third-order chaotic systems
still perform chaotic behavior [6]. And fourth-order chaotic systems still perform hy-
perchaotic behavior when the system order is 3 + % (0 < % < 1) [174]. Obviously, the
range of % is too big in this case. Is it possible to narrow? For example, fractional-order
Chua system performs chaotic behavior when the order is 2.7. People get the conclu-
sion only through observing system behavior when the system order is 2.7 and 2.6. So
is it chaotic when the order is 2.65? As a result, smaller step size should be chosen
and smaller error fractional differential approximation algorithm model should be
designed. Besides, dynamic characteristics such as Lyapunov exponent spectrum,
bifurcation, and dimension remain further research.
Synchronization technology is the key of application of chaotic system to secure
communication. In 1990, Pecora and Corroll first proposed the concept of chaotic syn-
chronization [10]. They also realized synchronization of two mutually coupled chaotic
systems in the laboratory by using the same signal circuits driven [11]. In the next
20 years, chaotic synchronization developed a lot. Even though synchronization of
integral chaotic system has achieved much research results, research of fractional-
order system synchronization is still in the beginning because it is too complex and
the effective mathematical tool is deficient. If we apply integral-order control method
to fractional-order system, true characteristic will be ignored. As a result, studying
control method of fractional-order chaotic system has practical significance.
By far, more effective synchronization control method has not been proposed
for fractional-order chaotic system. In most cases, we develop integral chaotic syn-
chronization method to fractional-order chaotic, so its synchronization control mech-
anism and performance need further study. Studies show that the mutual coupling
240 8 Analysis and Simulation of Fractional-Order Chaotic System
synchronization, projective synchronization, and adaptive method have unique ad-

vantages on improve synchronization performance. The target of fractional-order
system synchronization research is to improve system performance and finding gen-
eral law when changing order and parameter. The most suitable synchronization
method should be chosen through comparing synchronization establishment time
and synchronization accuracy. It is significantly important to lay technical foundation
for the application of fractional-order chaotic system in secure communication.
Application of fractional-order chaotic systems is becoming a new direction of
nonlinear applied science research area [175], especially in the aspects of inform-
ation secure communication. Complexity of fractional-order chaotic sequence is a
key point of the system safety performance. It is directly related to the perform-
ance of cryptography of fractional-order chaotic encryption system. For example,
to ensure maximum communication capacity of spread spectrum communication,
practical chaotic pseudorandom code sequence should have the biggest complex-
ity. Complexity of fractional-order chaotic system contains two meanings. One is
for chaotic system, which is always related to nonlinear term of system equation.
As for fractional-order system, it may also be associated with order. This kind of
complexity is mainly measured by calculating system fractal. The other meaning
is for complexity of time series generated by fractional-order chaotic system. It
is mainly used to measure the similarity degree of time signal and random sig-
nal. Some parameters, for example, information entropy, Lyapunov exponent, and
fractal dimension, are adopted to measure the degree of system disorder. But these
methods are not suitable to describe the complexity of time sequence generated
by chaotic system. Therefore, the phase space observation method and structural
complexity algorithm should be combined to analyze the fractional-order chaotic
system and corresponding chaotic pseudorandom sequence. It will provide basis
for the application of fractional-order chaotic system to cryptography and secure
communication.
Circuit implementation of fractional-order chaotic system is another key issue of
chaotic secure communication. In 2006, Bohannan et al. [176] applied for a patent to
design fractional devices. Liu led his team to achieve circuit simulation for a number of
fractional-order chaotic system through designing equivalent circuit of fractional in-
tegral [177–179]. But these methods are all based on frequency domain method. Even
though the time domain method has significant advantages in studying fractional-
order chaotic system, its circuit implementation is quite difficult. However, as the
development of microelectronics, it is possible to realize time domain approximate
method based on Field Programmable Gate Array (FPGA) or DSP. This method is a
new way to realize the application of fractional-order chaotic system.
In all, research of fractional-order chaotic system is still at the initial stage. There
still remain a lot of theoretical and practical issues that need further study. In this
chapter, we will analyze the relationship of dynamic characteristics and fractional
order. Synchronization of fractional-order chaotic system will also be studied.
8.2 Definition and Physical Meaning of Fractional Calculus 241
8.2 Definition and Physical Meaning of Fractional Calculus

q
Generally, fractional calculus is denoted by the operator ! Dt , which indicates both
of the fractional differential and the fractional integration. When q is positive, the
operator represents fractional differential. When q is negative, it indicates fractional
integration. Here, we introduce several common definitions about fractional calculus.
Definition 8.1: Gamma function A(x) is a basic function for fractional calculus, and its
definition is
∞
A(x) = tx–1 e–t dt. (8.1)
0
When x = n, A(n) = (n–1)!
Definition 8.2: Grünwald–Letnikov fractional calculus is defined as

N–1
q dq f (t) t – a –q j q t–a
a Dt f (t) = = lim (–1) j f t – j . (8.2)
d(t – a)q N→∞ N N
j=0
Definition 8.3: Riemann–Liouville fractional calculus is defined as

⎧ t
⎪
⎨ A(–q) a (t – 4)
1 –q–1 f (4)d4, q < 0
⎪
q
a Dt f (t) = f (t) q = 0, (8.3)
⎪
⎪
⎩Dn [ Dq–n f (t)] q>0
a t
where q ∈ R, n is the first integer greater than q, which means n – 1 ≤ q < n, A(⋅)
is Gamma function. According to the above definition, q-order differential of power
function and constant is
q A(r + 1)
Dt0 tr = (t – t0 )r–q , (8.4)
A(r + 1 – q)
q C
Dt0 C = (t – t0 )–q . (8.5)
A(1 – q)
The Laplace transform of R-L fractional calculus is defined as
q
L{0 Dt f (t)} = sq F(s), q ≤ 0, (8.6)
n–1

q q–k–1
L{0 Dt f (t)} = sq F(s) – s k 0 Dt f (0), n – 1 < q ≤ n ∈ N. (8.7)
k=0
As the Laplace transform needs the value of noninteger-order differential at t = 0, it is

difficult to solve Laplace transform of R-L fractional-order differential. However, it can
be solved according to the Caputo definition of fractional differential
Definition 8.4: Caputo fractional differential is defined as

⎧ t f (m) (4)
⎨ 1
, m–1<q<m
q A(m–q) 0 (t–4)q+1–m
0 Dt f (t) = m , (8.8)
⎩ m f (t),
d
q=m
dt
where m is the first integer greater than q. The Laplace transform of Caputo fractional
differential is
n–1

q
L{0 Dt f (t)} = sq F(s) – sq–1–k f (k) (0), n – 1 < q ≤ n ∈ N. (8.9)
k=0
When compared with Laplace transform of R-L fractional differential, for Caputo
fractional differential, only integer-order differential is needed in Laplace transform.
When the initial value of operator is zero, its corresponding Laplace transform is
q
L{0 Dt f (t)} = sq F(s). (8.10)
Two basic properties of Caputo fractional differential definition are
∗ q q
Dt0 Jt0 f (t) = f (t), (8.11)
m–1

q q q m–q (t – t0 )k
Jt0 (∗ Dt0 )f (t) = Jt0 Jt0 Dm m m
t0 f (t) = Jt0 Dt0 f (t) = x(t) – x(k) (t0+ ) . (8.12)
k!
k=0
Thus, the q-order differential of the power function and constant are defined as
∗ q r A(r + 1)
Dt0 t = (t – t0 )r–q , (8.13)
A(r + 1 – q)
∗ q
Dt0 C = 0. (8.14)
Integer-order calculus has clear geometric interpretation and physical meaning. How-
ever, because of the complexity of fractional calculus, it is difficult to understand the
concept of fractional calculus. Therefore, there exist some obstacles in practical ap-
plications. At present, fractional calculus does not have universal and unified physical
meaning and geometric interpretation yet [180].
It is easy to get the physical meaning of fractional integral according to its defin-
ition. If we suppose integral as storage, then the fractional integral can be taken as
memory storage. The fractional integral only store near information and abandon
the past information gradually. Professor Liu et al. in Peking University described the
fractional differential as a “bridge between weather and climate,” which means the
fractional differential of climate is weather. It is the fractional differential that makes
climate has greater memory than weather [181].
We can get some conclusions according to the definition of fractional differen-

tial. (1) The initial value of input function is added to the output as attenuate form.
(2) When the initial value is zero, the fractional differential is formed as convolution.
Thus, the fractional differential is actually an integral and it forgets characteristic
gradually (or time decay of memory). (3) Introduction of the definition of R-L fractional
differential can simplify the calculation of fractional derivative. (4) Introduction of
the definition of Caputo fractional differential makes Laplace transform more concise,
which benefits the calculation of fractional differential equation [182].
8.3 Solution Methods of Fractional-Order Chaotic System
8.3.1 Frequency Domain Method for Solving Fractional-Order Chaotic System
There are several methods and algorithms for solving fractional differential equations
and fractional-order systems, for example, Laplace transform, Fourier transform al-
gorithm, and numerical algorithms. However, these methods have some drawbacks.
At present, people are most interested in using integer-order model to approximate
the fractional-order system.
Bode graphics approximating method of the fractional-order system is realized
by the time–frequency conversion. The transfer functions are calculated by piecewise
linear approximation in frequency domain. This method is in favor of the project im-
plementation. The idea is to obtain frequency domain expansion of fractional integral
operator by solving 1/s! in frequency domain. And then the frequency domain func-
tion is transformed to the integer-order state equation [170]. This approximation error
is within the permissible range and can fully meet the needs of engineering.
If the initial condition is set to zero, then the Laplace transform of R-L in formula
(8.7) becomes to

dq f (t)
L = sq L{f (t)}. (8.15)
dtq
Therefore, the fractional integral operator in frequency domain with q-order can be
described by the transfer function F(s) = 1/sq . To analyze dynamics of fractional-order
chaotic system effectively, standard integral-order operator is required to approxim-
ate fractional-order operators (Bode plot approximation). In this case, approximation
error will definitely be within the permissible range and can fully meet engineering
requirement. Fractional integral operators are defined as
H(s) = 1/s! , (8.16)
where s = j9 is complex frequency, and ! is a positive fraction which is also called

fractional power. Since the magnitude of this function is relatively big at low frequency
and the system is of multifrequency, a fractional-order system with multifrequency

can be described as
1
H(s) = !i , 0 < !i < 1; i = 1, 2, . . . , n, (8.17)
n s
i=1 1+ pT
i
where 1/pTi is relaxation time and !i is fractional power factor. Formula (8.16) is called
single structural model of single fractional-order system. Through a fractional extreme
representation of single frequency, the transfer function in frequency domain can be
represented as
1
H(s) = ! , (8.18)
s
1+ pT
where 1/pT is relaxation time and ! is fractional power factor (0 < ! < 1).
By using –20 dB segments with “zigzag” type to approximate, formula (8.18) can
be approximated as [170]
N–1 s

1 i=0 1+ zi
H(s) = ! ≈ lim , (8.19)
s N→∞ N s
1+ pT i=0 1+ pi
where N represents the number of poles. Since working frequency in engineering prac-
tice is limited, the number of the above formula can be limited to N. Then the above
formula can be further approximated as
N–1 s

1 i=0 1+ zi
H(s) = ! ≈ . (8.20)
s N s
1+ pT i=0 1+ pi
As a result, the problem of getting the fractional function expression is transformed

into the problem of finding zero-pole pair. If the approximation error is controlled
within y dB (y is positive), the pole-zero pair is determined as follows:
The first pole: p0 = pT 10y/(20!)

The first zero: z0 = p0 10y/(10(1–!)) .
The second pole: p1 = z0 10y/(10!) .
The second zero: z1 = p1 10y/(10(1–!)) .
..
.
The Nth zero: zN+1 = pN+1 10y/(10(1–!)) .
The (N+ 1)th pole: pN = zN+1 10y/(10(!)) .
Here PT is the angular frequency and p0 is the first singular value which is determined
by the setting error y. PN is the last singular value which is determined by N.
Now, let a = 10y/(10(1–!)) , b = 10y/(10!) , then we have
ab = 10y/(10!(1–!)) . (8.21)
So the distribution relationship of zeros and poles can be written as

z0 z1 zN–1
= =⋯= = a, (8.22)
p0 p1 pN–1
p1 p2 pN–1
= =⋯= = b, (8.23)
z0 z1 zN–1
z1 z2 zN–1
= =⋯= = ab, (8.24)
z0 z1 zN–2
p1 p2 pN
= =⋯= = ab. (8.25)
z0 p1 pN–1
According to the above relationship and p0 , we can obtain all zero-pole values:
pi = (ab)i p0 , i = 1, 2, 3, . . . , (8.26)
zi = (ab)i ap0 , i = 1, 2, 3, . . . . (8.27)
N of formula (8.19) is deduced by the following procedure. The constraint relationship

of system frequency bandwidth is
pN–1 < 9max < pN , (8.28)
(ab)N–1 p0 < 9max < (ab)N p0 , (8.29)

9max
(ab)N–1 < < (ab)N , (8.30)
p0

9max
(N – 1) log(ab) < log < N log(ab). (8.31)
p0
Therefore,

9max
log p0
N–1< < N, (8.32)
log(ab)
⎡ ⎤
log 9pmax
N = int ⎣ ⎦ + 1.
0
(8.33)
log(ab)
Generally, we set y = 2 dB or y = 3 dB, 9max = 0.01–100 rad/s. According to the above

formula, 1/s! in frequency domain is obtained. The expansion form of 1/s! is shown in
Table 8.1 [166], where ! changes from 0.1 to 0.9 and approximation error is 2 dB. When
the approximation error is 3 dB, the expansion form of 1/s! is shown in Table 8.2 [163].
Table 8.1: !-Order fractional integral linearity conversion function approximation.

(Approximation error is 2 dB.)
! 1/sa Transfer functions
1 220.4s4 + 5004s3 + 5038s2 + 234.5s + 0.4840

0.1
s0.1 s5+ 359.8s4 + 5742s3 + 4247s2 + 147.7s + 0.2099
1 60.95s4 + 816.9s3 + 582.8s2 + 23.24s + 0.04934
0.2
s0.2 s5+ 134.0s4 + 956.5s3 + 383.5s2 + 8.953s + 0.01821
1 23.76s4 + 224.9s3 + 129.1s2 + 4.733s + 0.01052
0.3
s0.3 s5 + 64.51s4 + 252.2s3 + 63.61s2 + 1.104s + 0.002267
1 25.00s4 + 558.5s3 + 664.2s2 + 44.15s + 0.1562
0.4
s0.4 s5 + 125.6s4 + 840.6s3 + 317.2s2 + 7.428s + 0.02343
1 15.97s4 + 593.2s3 + 1080s2 + 135.4s + 1
0.5
s0.5 s + 134.3s4 + 1072s3 + 543.4s2 + 20.10s + 0.1259
5
1 8.579s4 + 255.6s3 + 405.3s2 + 35.93s + 0.1696

0.6
s0.6 s5 + 94.22s4 + 472.9s3 + 134.8s2 + 2.639s + 0.009882
1 5.406s4 + 177.6s3 + 209.6s2 + 9.197s + 0.01450
0.7
s0.7 s5 + 88.12s4 + 279.2s3 + 33.30s2 + 1.927s + 0.0002276
1 5.235s3 + 1453s2 + 5306s + 254.9
0.8
s0.8 s4 + 658.1s3 + 5700s2 + 658.2s + 1
1 1.766s3 + 38.27s2 + 4.914
0.9
s0.9 s + 36.15s2 + 7.789s + 0.01000
3
Table 8.2: !-Order fractional integral linearity conversion function approximation.

(Approximation error is 3 dB.)
! 1/sa Transfer functions
1 501.14(s + 0.16811)
0.1
s0.1 (s + 0.3162)(s + 681.1)
1 141.2538(s + 0.1334)(s + 10)
0.2
s0.2 (s + 0.035623)(s + 4.217)(s + 316.2)
1 125.8925(s + 0.08483)(s + 2.276)(s + 61.05)
0.3
s0.3 (s + 0.03162)(s + 0.8483)(s + 22.76)(s + 610.5)
1 26.6073(s + 0.07499)(s + 1.334)(s + 23.71)
0.4
s0.4 (s + 0.02371)(s + 0.4217)(s + 7.499)(s + 133.4)
1 50.1187(s + 0.07943)(s + 1.259)(s + 19.95)(s + 316.2)
0.5
s0.5 (s + 0.01995)(s + 0.3162)(s + 5.021)(s + 79.43)(s + 12.59)
1 28.1838(s + 0.1)(s + 1.778)(s + 31.62)(s + 562.3)
0.6
s0.6 (s + 0.01778)(s + 0.3162)(s + 5.623)(s + 100)(s + 1778)
1 7.9433(s + 0.1638)(s + 4.394)(s + 117.9)
0.7
s0.7 (s + 0.016389)(s + 0.4393)(s + 11.79)(s + 100)(s + 316.2)
1 8.1752(s + 0.487)(s + 36.52)(s + 2, 738)
0.8
s0.8 (s + 0.0154)(s + 1.155)(s + 86.6)(s + 6, 494)
1 4.2987(s + 14.68)(s + 31, 620)
0.9
s0.9 (s + 0.01468)(s + 31.62)(s + 68, 130)
8.3.2 Time Domain Method for Solving Fractional-Order Chaotic System
8.3.2.1 Adams–Bashforth–Moulton Predictor–Corrector Scheme

A typical fractional-order time domain approximation method is Adams–Bashforth–
Moulton predictor–corrector scheme [171–173]. This method is described as follows.
For the differential equation

dq x
dtq = f (t, x), 0≤t≤T
. (8.34)
xk (0) = x0(k) , k = 0, 1, 2, . . . , q – 1
It equals Volterra integral equation
n–1
k t
t 1
x(t) = x0(k) + (t – 4)q–1 f (4, x(t))d4. (8.35)
k! A(q)
k=0 0
Obviously, the two parts of the right side in the above equation totally determined
by the initial value. A typical case is 0 < q < 1, and the Volterra equation has weak
singularity. In Refs [171–173], a predictor–corrector scheme for eq. (8.35) is reported. It
can be regarded as a similar algorithm of Adams–Moulton algorithm with one step.
Let h = T/N, tj = jh (j = 0, 1, 2, . . . , N), where T is the integral upper limit of this
equation. Then the corrector formula of eq. (8.35) is given by
q–1 n
k
tn+1 hq p hq
xh (tn+1 ) = x0(k) + f (tn+1 , xh (tn+1 ))+ aj,n+1 f (tj , xh (tj )), (8.36)
k! A(q + 2) A(q + 2)
k=0 j=0
where

nq+1 – (n – q)(n + 1)q , j=0
aj,n+1 = . (8.37)
(n – j – 2)q+1 + (n – j)q+1 – 2(n – j + 1)q+1 , 1≤j≤n
Then we use one-step Adams–Bashforth rule to replace one-step Adams–Moulton

p
rule. The predictor term xh (tn+1 ) is given by
n–1
n
p
k
tn+1 1
xh (tn+1 ) = x0(k) + bj,n+1 f (tj , xh (tj ), (8.38)
k! A(q)
k=0 j=0
where
hq
bj,n+1 = ((n – j + 1)q – (n – j)q ), 0 ≤ j ≤ n. (8.39)
q
The basic algorithm of fractional-order Adams–Bashforth–Moulton method is shown

in formulas (8.36) and (8.38), where aj,n+1 and bj,n+1 are presented in formulas (8.37)
and (8.39), respectively. The estimate error of this method is e = max |x(tj ) – xh (tj )| =
i=0,1,...,N
o(hp ), where p = min (2, 1 + q). Although the time domain method is more complex
than the frequency domain method, the step size can be smaller, and it is much better
for analysis of dynamics of fractional-order system.
8.3.2.2 Adomian Decomposition Algorithm

Adomian decomposition algorithm (Adomian decomposition method, ADM) is a time
domain approximation method for solving fractional-order system [183]. This method
is used to solve linear and nonlinear problems. The corresponding approximate nu-
merical solution is highly accurate with fast convergence. Moreover, it needn’t be
discreted and too many computer memories. The numerical solutions of fractional-
order Chen system and fractional-order Rössler system based on ADM are studied
[184, 185], which proves the effectiveness of the algorithm.
q
If the fractional-order chaotic system is ∗ Dt0 x(t) = f (x(t))+ g(t), where x(t) = [x1 (t),
x2 (t), . . . , xn (t)]T is the functional variable and g(t) = [g1 (t), g2 (t), . . . , gn (t)]T is the con-
stant term for autonomous system and f includes both linear and nonlinear terms,
then the system can be separated into three parts by
⎧
⎪∗ q
⎨ Dt0 x(t) = Lx(t) + Nx(t) + g(t)
⎪
x(k) (t0+ ) = bk , k = 0, . . . , m – 1 , (8.40)
⎪
⎪
⎩m ∈ N, m – 1 < q ≤ m
q
where ∗ Dt0 represents q-order Caputo differential operator. L represents the linear
term. N represents the nonlinear term, and bk is the initial value. Applying the integral
operator to both sides of the equation, we have
q q q
x = Jt0 Lx + Jt0 Nx + Jt0 g + I, (8.41)
(t–t0 )k
where I = m–1 k=0 bk k! is the initial value. According to the Adomian decomposi-

tion algorithm, the solution of the equation can be represented as x = ∞ i
i=0 x . The
nonlinear term can be decomposed as
⎧
i di
⎪
⎨A j =
1
i! d+i
N(vji (+))
+=0
⎪ i , (8.42)
⎩vji (+) = k
(+) xj k
k=0
where i = 0, 1, . . . , and j = 1, 2, . . . , n. The nonlinear term is
∞

Nx = Ai (x0 , x1 , . . . , xi ). (8.43)
i=0
Then, the solution of the equation is

∞
∞
∞

q q q
x= xi = Jt0 L xi + Jt0 Ai + Jt0 g + I, (8.44)
i=0 i=0 i=0
where its iterative relationship is

⎧
⎪ q
⎪
⎪ x0 = Jt0 g + I
⎪
⎪
⎪
⎪ q q
x1 = Jt0 Lx0 + Jt0 A0 (x0 )
⎪
⎪
⎪
⎪
⎪x2 = J q Lx1 + J q A1 (x0 , x1 )
⎨ t0 t0
.. . (8.45)
⎪
⎪ .
⎪
⎪
⎪
⎪ i q i–1 q i–1 0 1 i–1
⎪
⎪x = Jt0 Lx + Jt0 A (x , x , . . . , x )
⎪
⎪
⎪
⎪
⎩...
Next, we take fractional-order simplified Lorenz system as an example to explain this

method further.
The differential equations of fractional-order simplified Lorenz system with a
parameter c is [186]
⎧
⎪ ∗ q
⎨ Dt0 x1 = 10(x2 – x1 )
⎪
∗ Dq x = (24 – 4c)x – x x + cx . (8.46)
⎪ t0 2 1 1 3 2
⎪
⎩∗ Dq x = x x – 8x /3
t0 3 1 2 3
The linear and nonlinear terms of system are decomposed as

⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
L x1 10(x2 – x1 ) Nx1 0 g1 0
⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥
L
⎣ x2 ⎦ ⎣= (24 – 4c)x 1 + cx , N
2 ⎦ ⎣ x2 ⎦ = –x x , g
⎣ 1 3 ⎦ ⎣ 2 ⎦ ⎣0⎦.
= (8.47)
Lx3 –8x3 /3 N x3 x1 x2 g3 0
According to eq. (8.44), the first seven items of nonlinear term are [187]
⎧
⎪
⎪ A02 = –x10 x30
⎪
⎪
⎪
⎪
⎨A2 = –x1 x3 – x1 x3
1 1 0 0 1
⎪
A32 = –x12 x30 – x11 x31 – x10 x32 , (8.48)
⎪
⎪
⎪
⎪
⎪A2 = –x1 x3 – x12 x31 – x12 x32 – x1 x3
4 3 0 0 3
⎪
⎪
⎩ 5
A2 = –x1 x3 – x1 x3 – x1 x3 – x13 x32 – x12 x33 – x10 x35
5 0 4 1 1 4
⎧
⎪
⎪ A03 = x10 x20
⎪
⎪
⎪
⎪
⎨A3 = x1 x2 + x1 x2
1 1 0 0 1
⎪
A33 = x12 x20 + x11 x21 + x10 x22 (8.49)
⎪
⎪
⎪
⎪ A43 = x13 x20 + x12 x21 + x12 x22 + x10 x23
⎪
⎪
⎪
⎩ 5
A3 = x15 x20 + x14 x21 + x11 x24 + x13 x22 + x12 x23 + x10 x25 .
According to the given initial values, we have

⎧
⎪
⎨x1 = x1 (t0 )
0
⎪
x20 = x2 (t0 ) , (8.50)
⎪
⎪
⎩x0 = x (t )
3 3 0
q
Let c01 = x10 , c02 = x20 , c03 = x30 , which means c0 = [c01 , c02 , c03 ]. According to x1 = Jt0 Lx0 +
q
Jt0 A0 in formula (8.45) and properties of fractional calculus, we have
⎧
⎪ 0 (t–t0 )q
⎨x1 = 10(c2 – c1 ) A(q+1)
1 0
⎪
q
x21 = [(24 – 4c)c01 + cc02 – c01 c03 ] (t–t 0) . (8.51)
⎪ A(q+1)
⎩x1 = – 8 c0 + c0 c0 (t–t0 )
⎪
3 3 1 1 2 A(q+1)
If c11 = 10(c02 – c01 ), c12 = (24 – 4c)c01 + cc02 + c04 – c01 c03 , c13 = –(8/3)c01 + c01 c02 , then
x1 = c1 (t – t0 )q /A(q + 1). So we only need to solve the corresponding coefficient of each
term. According to formula (8.46), the other five coefficients are obtained as
⎧
⎪
⎨c1 = 10(c2 – c1 )
2 1 1
⎪
c22 = (24 – 4c)c11 + cc12 – c11 c03 – c01 c13 , (8.52)
⎪
⎪
⎩c2 = – 8 c1 + c1 c0 + c0 c1
3 3 1 1 2 1 2
⎧
⎪
⎪ c3 = 10(c22 – c21 )
⎨ 1
c32 = (24 – 4c)c21 + cc22 – c21 c03 – c11 c13 AA(2q+1)
2 (q+1) – c1 c3 ,
0 2
(8.53)
⎪
⎪
⎩c3 = – 8 c2 + c2 c0 + c1 c1 A(2q+1) + c0 c2
3 3 1 1 2 1 2 A2 (q+1) 1 2
⎧
⎪
⎨c1 = 10(c2 – c1 )
4 3 3
⎪
A(3q+1)
c42 = (24 – 4c)c31 + cc32 – c31 c03 – (c21 c13 + c11 c23 ) A(q+1)A(2q+1) – c01 c33 , (8.54)
⎪
⎪
⎩c4 = – 8 c3 + c3 c0 + (c1 c2 + c2 c1 ) A(3q+1) + c0 c3
3 3 1 1 2 1 2 1 2 A(q+1)A(2q+1) 1 2
⎧
⎪
⎪c51 = 10(c42 – c41 )
⎪
⎪
⎪
⎨c5 = (24 – 4c)c4 cc4 – c4 c0 – (c3 c1 + c1 c3 ) A(4q+1) –
2 1 2 1 3 1 3 1 3 A(q+1)A(3q+1)
, (8.55)
⎪
⎪
⎪ c21 c23 AA(4q+1)
2 (2q+1) – c1 c3
0 4
⎪
⎪
⎩c5 = – 8 c4 + c4 c0 + (c1 c3 + c3 c1 ) A(4q+1) + c2 c2 A(4q+1) + c0 c4
3 3 1 1 2 1 2 1 2 A(q+1)A(3q+1) 1 2 A2 (2q+1) 1 2
⎧
⎪
⎪c61 = 10(c52 – c51 )
⎪
⎪
⎪
⎪ A(5q+1)
⎪
⎨c2 = (24 – 4c)c1 + cc2 – c1 c3 – (c1 c3 + c1 c3 ) A(q+1)A(4q+1) –
6 5 5 5 0 4 1 1 4
⎪
A(5q+1)
(c21 c33 + c31 c23 ) A(2q+1)A(3q+1) – c01 c53 . (8.56)
⎪
⎪
⎪
⎪ 8 5 5 4 4 A(5q+1)
c63 = – 3 c1 + c1 c02 + (c11 c2 + c1 c12 ) A(q+1)A(4q+1) +
⎪
⎪
⎪
⎪
⎩ (c2 c3 + c3 c2 ) A(5q+1)
1 2 A(2q+1)A(3q+1) + c1 c2
0 5
1 2
15 20
(a) (b)
10
15
5
x3
0
x2
–5
5
–10
–15 0
–10 –5 0 5 10 –10 –5 0 5 10
x1 x1
Figure 8.1: Attractor of fractional-order simplified Lorenz system: (a) x1 –x2 plane and (b) x1 –x3 plane.
So the solution of system equation is
(t – t0 )q (t – t0 )2q (t – t0 )3q
x̃j (t) = c0j + c1j + c2j + c3j
A(q + 1) A(2q + 1) A(3q + 1)
, (8.57)
(t – t0 )4q (t – t0 )5q (t – t0 )6q
+ c4j + c5j + c6j
A(4q + 1) A(5q + 1) A(6q + 1)
where j = 1, 2, 3. Then we can get the analytic solution of system variable. In practical
calculation, the time period [t0 , t] should be divided into smaller time interval [tk , tk+1 ].
In each interval, t0 is set as tk , and then the value of corresponding tk+1 is obtained. The
attractor phase diagram of fractional-order simplified Lorenz system based on formula
(8.57) is shown in Figure 8.1. In this figure, the length of small time interval is 0.01 s.
The initial values are [0.1, 0.2, 0.3]. Parameter c = 5 and fractional order q = 0.98.
8.4 Simulation Method for Fractional-Order Chaotic System
8.4.1 Dynamic Simulation Method for Fractional-Order Chaotic System
Dynamic simulation analysis method is to use the dynamic model to represent the
state change over time, and the dynamical characteristics of the system are analyzed
by observing the real-time evolution of the variables. By employing this method,
we can directly observe the simulation results and also can make a quantitative
analysis for the simulation output data. It has the characteristics of intuition and flex-
ibility. Based on frequency domain analysis method of fractional calculus operator,
fractional-order system simulation model is designed through state-space module in
Simulink. State-space module is a state-space description module for input–output
variables, which is defined as

ẋ = Ax + Bu
, (8.58)
y = Cx + Du
where x is the state vector, u is the input vector, and y is the output vector. A, B, C, and
D are coefficient matrices. Obviously, those modules receive inputs and then generate
outputs. Parameters of coefficient matrices A, B, C, and D are corresponding to that of
transform function, respectively, which can be calculated by tf2ss transform function
in Matlab.
Steps of the dynamic simulation are as follows:
1. Determine the function modules listed according to the differential equation of
fractional-order chaotic system. The simulation system is built by connecting
these modules.
2. Set simulation parameters. It includes system operating parameters (e.g., sys-
tem operating time, integration step size, tolerance error, integration algorithms,
etc.) and function modules operating parameters (such as integrator initial value,
signal amplitude, frequency, etc.).
3. Set observation module. It is used to output data and observe operation status of
simulation system.
4. Analyze the simulation results. Simulation results are analyzed by output data
and waveform, and compared with that of theoretical analysis. If it is not consist-
ent with each other, the model and parameters should be revised.
During the simulation, Scope module is used to show real-time output signal dynamic-
ally. Phase diagram can be observed by XY graph module. In phase space, convergence
motion corresponds to the equilibrium point, periodic motion corresponds to a closed
trajectory, and chaotic motion corresponds to random separation of the never closed
trajectory in a certain region (strange attractor). So, the output phase diagram of XY
graph module is used to preliminarily judge whether the system is chaotic. Out module
can save results to workspace automatically, so the saved data can be output by yout()
function, and it is used to calculate the power spectrum and the maximum Lyapunov
exponent of the system.
How to set the state-space module is the key of realizing dynamic simulation of
fractional-order chaotic system. Parameter matrices A, B, C, and D in the property dia-
log box are determined by the order of fractional-order system. To obtain each matrix
element, we can call command [A,B,C,D] = TF2SS(NUM,DEN) (where NUM represents
the molecular coefficient matrix and DEN represents the denominator coefficient mat-
rix) or [A,B,C,D] = ZP2SS(Z,P,K) (where Z represents the zeros matrix of transform
function, P represents the poles matrix, and K represents the coefficient).
For example, when q = 0.95, the transform function of fractional-order integral
operator is written as [167]
1 1.2831s2 + 18.6004s + 2.0833

H(s) = ≈ . (8.59)
s0.95 s3 + 18.4738s2 + 2.6547s + 0.003
For formula (8.59), calling format of the function TF2SS is [A,B,C,D] = TF2SS
(1.2831,18.6004,2.0833], [1,18.4738,2.6547,0.003]). Its corresponding return value is
⎡ ⎤ ⎡ ⎤
–18.4738 –2.6547 –0.003 1
⎢ ⎥ ⎢ ⎥
A=⎣ 1 0 0 ⎦ , B = ⎣ 0 ⎦, C = [1.2831, 18.6004, 2.0833140], D = 0.
0 1 0 0
Next, we take the fractional-order diffusionless Lorenz system as an example to

explain further. The diffusionless Lorenz system is
⎧
⎪
⎨ẋ = –x – y
⎪
ẏ = –xz , (8.60)
⎪
⎪
⎩ż = xy + R
where R is the parameter of system. When R ∈ (0, 5), the system is chaotic. Its
corresponding fractional-order diffusionless Lorenz system is written as
⎧ !
⎪ d x
⎪
⎪ = –x – y
⎪
⎪ dt!
⎪
⎪
⎪
⎨
d" y
= –xz , (8.61)
⎪
⎪ dt"
⎪
⎪
⎪
⎪
⎪ #
⎩ d z = xy + R
⎪
dt#
where 0 < !, ", # ≤ 1 are fractional orders. Dynamic simulation model of the system
is constructed as shown in Figure 8.2. The initial values of system are set as [0.3782,
0.6748, 2.8831]. Let ! = " = # = 0.95 and simulation time is 200 s. By using ODE45
differential equation solvers, the strange attractor of diffusionless Lorenz system in
x–z plane is shown in Figure 8.3.
8.4.2 Circuit Simulation Method of Fractional-Order Chaotic System
Design of fractional integral operator circuit is the key of Electronics Workbench cir-
cuit simulation for fractional-order chaotic system. Because the fractional integral
operator with q-order is described by transform function F(s) = 1/sq in frequency do-
main, the unit circuit as shown in Figure 8.4 is used to achieve the expansion of 1/sq
[177]. The nonlinear terms are realized by multipliers. Then the equivalent circuit of
the fractional-order system is constructed.
In Figure 8.4, system function of the equivalent circuit between A and B is
described as
R1 R2 R3 Rn
H(s) = + + +⋯+ , (8.62)
sR1 C1 + 1 sR2 C2 + 1 sR3 C3 + 1 sRn Cn + 1
Scope
xʹ = Ax + Bu
y = Cx + Du
α = 0.95
XYGraph
–1 xʹ = Ax + Bu
y = Cx + Du
Product2 Gain β = 0.95 XYGraph1
xʹ= Ax + Bu
y = Cx + Du
Product1 XYGraph2
γ = 0.95
5 1
Out2
R
Figure 8.2: Simulation model of the fractional-order diffusionless Lorenz system.
20
10
0
z
–10
–20
–10 –5 0 5 10
x
Figure 8.3: Attractor of diffusionless Lorenz system in the x–z plane when q = 0.95.
R1 R2 R3 Rn
...
A B
C1 C2 C3 Cn
Figure 8.4: Unit circuit of fractional-order integrator.

where n is the highest order of s in denominator of 1/s! when q varied from 0.1 to 0.9.
For example, when q = 0.95, n = 3, according to eq. (8.59), we have
R1 R2 R3
F(s) = + + . (8.63)
sR1 C1 + 1 sR2 C2 + 1 sR3 C3 + 1
Comparing eq. (8.63) with eq. (8.59), we have R1 = 15.1 kK, R2 = 1.51 kK, R3 = 692.9 MK,
C1 = 3.616 ,F, C2 = 4.602 ,F, C3 = 1.267 ,F. Resistors and capacitors of 1/sq calculated
in Ref. [166] are shown in Tables 8.3 and 8.4, respectively [177].
The circuit of fractional-order diffusionless Lorenz system is shown in Figure 8.5,
in which the resistance values are R1 = R2 = R3 = R4 = R7 = 10 kK, R6 = R8 = R9 = 100 kK,
Table 8.3: Resistance of fractional-order integrator unit circuit with order !.
1
sa R1 /MK R2 /MK R3 /MK R4 /MK R5 /MK R6 /MK
1
s0.1
0.636 0.3815 0.5672
1
s0.2
1.130 0.6070 0.3500 0.425 0.2498
1
s0.3
2.050 0.9380 0.4830 0.262 0.1430 0.106
1
s0.4
3.744 1.3920 0.6310 0.294 0.1230 0.068
1
s0.5
6.824 1.9440 0.7440 0.296 0.0754 0.030
1
s0.6
12.330 2.4480 0.7380 0.233 0.0754 0.006
1
s0.7
21.900 2.6000 0.5260 0.113 0.0246
1
s0.8
37.850 1.7540 0.1700 0.017 0.0018
1
s0.9
62.840 0.2500 0.0025
Table 8.4: Resistance of fractional-order integrator unit circuit with order !.
1
sa C1 /,F C2 /,F C3 /,F C4 /,F C5 /,F C6 /,F
1
s0.1
15.720 0.1572 0.0006335
1
s0.2
27.990 2.9300 0.285 0.0132
1
s0.3
22.640 5.5200 1.200 0.2460 0.029
1
s0.4
15.020 5.9260 1.920 0.6050 0.183 0.036
1
s0.5
9.246 5.1450 2.129 0.8480 0.324 0.925
1
s0.6
5.527 4.0850 1.990 0.9260 0.420 0.156
1
s0.7
3.284 3.1390 1.700 0.8860 0.454 0.207
1
s0.8
1.980 2.4000 1.390 0.7800 0.420
1
s0.9
1.232 1.8400 1.100
R1 R4 fraction
R5
R2
R3
AD633 fraction
Y R6
X
R9
AD633 fraction
Y R7
X R10
R8
5V
Figure 8.5: Circuit simulation model of diffusionless Lorenz system with q = 0.95.
10
5
Voltage(z)
–5
–10
–10 –5 0 5 10
Voltage(x)
Figure 8.6: Attractor of fractional-order diffusionless Lorenz system based on circuit simulation.
R5 = R10 = 1 MK. “Fraction” is the integrator unit circuit of 0.95-order after packaged.
When R = 5 and q = 2.85, its attractor of x–z plane is shown in Figure 8.6. Obviously, it
is consistent with that of simulation result as shown in Figure 8.3.
8.4.3 Numerical Simulation Method of Fractional-Order Chaotic System
Numerical simulation method is to transform the mathematic model of fractional-

order chaotic system into recursive calculation form which is suitable for processing
on a computer. The main advantage of this method is that it can solve complex
problems analytically. The basic steps are as follows [188]:
1. Establish mathematical model for fractional-order chaotic system.
2. According to the characteristic of this model, choose suitable software as simula-
tion tool and design the computer program according to the mathematical model.
3. Run the program and get the numerical solution of the system.
4. According to the numerical solution, analyze dynamic characteristic such as
calculating Lyapunov exponent and fractal dimension.
For example, according to the approximation formula based on Adams–Bashforth–

Moulton time domain method, the fractional-order diffusionless Lorenz chaotic
system is presented as
⎧
⎪
⎪ h! p p n
⎪
⎪ xn+1 = x0 + [–yn+1 – xn+1 ] + a1,j,n+1 (–yj – xj )
⎪
⎪ A(! + 2)
⎪
⎪ j=0
⎪
⎨ h" p p n
yn+1 = y0 + (–xn+1 zn+1 ) + a2,j,n+1 (–xj zj ) , (8.64)
⎪
⎪ A(" + 2) j=0
⎪
⎪
⎪
⎪ h# n
⎪
⎪ p p
⎪
⎩ zn+1 = z0 + [xn+1 yn+1 + R] + a3,j,n+1 (xj yj + R)
A(# + 2) j=0
where
⎧
⎪ p 1 n
⎪
⎪xn+1 = x0 + b (–yj – xj )
⎪
⎪ A(!) j=0 1,j,n+1
⎪
⎪
⎨ 1 n
p
yn+1 = y0 + b (–xj zj ) , (8.65)
⎪
⎪ A(") j=0 2,j,n+1
⎪
⎪
⎪
⎪ p 1 n
⎪
⎩zn+1 = z0 + b (xj yj + R)
A(#) j=0 3,j,n+1
⎧
⎪ h!
⎪
⎪ b1,j,n+1 = ((n – j + 1)! – (n – j)! ), 0 ≤ j ≤ n
⎪
⎪ !
⎨ h"
b2,j,n+1 = ((n – j + 1)" – (n – j)" ), 0 ≤ j ≤ n , (8.66)
⎪
⎪ "
⎪
⎪ h#
⎪
⎩b3,j,n+1 = ((n – j + 1)# – (n – j)# ), 0 ≤ j ≤ n
#
⎧
⎪
⎪ n! – (n – !)(n + 1)! j=0
⎪
⎪ a =
⎪
⎪ 1,j,n+1
⎪
⎪ (n – j + 2)!+1 + (n – j)!+1 – 2(n – j + 1)!+1 , 1≤j≤n
⎪
⎪
⎨ n" – (n – ")(n + 1)" j=0
a2,j,n+1 = . (8.67)
⎪
⎪ (n – j + 2)"+1 + (n – j)"+1 – 2(n – j + 1)"+1 , 1≤j≤n
⎪
⎪
⎪
⎪
⎪
⎪ n# – (n – #)(n + 1)# j=0
⎪
⎩a3,j,n+1 =
⎪
(n – j + 2)#+1 + (n – j)#+1 – 2(n – j + 1)#+1 , 1≤j≤n
Based on Matlab, we let ! = " = # = 0.95, R = 5, h = 0.01, T = 200, and the computer
program is designed. The chaotic attractor of fractional-order diffusionless Lorenz
20
10
0
z
–10
–20
–10 –5 0 5 10
x
Figure 8.7: Attractor of fractional-order diffusionless Lorenz system based on numerical simulation.
chaotic system with q = 0.95 is shown in Figure 8.7. Obviously, it is consistent with
the results of dynamic simulation and circuit simulation.
8.4.4 Comparison of Simulation Methods
1. Comparison among dynamic simulation, circuit simulation, and numerical simu-

lation based on predictor–corrector method. The three simulation methods have
their own advantages and disadvantages. We summarized their features as shown
in Table 8.5 from the aspects of analysis method, fractional order change, res-
ult output, and simulation speed. Dynamic simulation method can construct
dynamic system model quickly by employing abundant functional blocks in Sim-
ulink. You don’t need to write a line of codes. When simulating complicated
fractional-order nonlinear chaotic system, this method is more simple and intuit-
ive. However, its limitation is that the value of q can only take finite value. Circuit
simulation has similar characteristic with dynamic simulation, but because the
circuit simulation is more close to the realistic engineering system, it is more im-
portant for the application of the fractional-order chaotic system. Programming
is required for numerical simulation, and dynamic output process of variable
evolution cannot be observed directly. But when the value of fractional-order
changes continuously, it is more convenient to analyze the chaotic characteristic
of the fractional-order system.
Table 8.5: Comparison of simulation methods.
Simulation methods Analysis methods Variation of q Output results Speed
Dynamic simulation Frequency domain Finite Dynamic Fast

Circuit simulation Frequency domain Finite Dynamic Fast
Numerical simulation Time domain Infinity Static Slow
8.5 Dynamics of Fractional-Order Unified System Based on Frequency Domain Method 259
Table 8.6: Comparison between numerical simulation methods for fractional-

order chaotic system.
Adomian method Predictor–corrector scheme
Time complexity O(n), fast O(n2 ), slow

Space complexity O(1), small O(n), large
N = 1, 000 0.9701 s 2.0154 s
N = 2, 000 1.8403 s 7.7051 s
N = 5, 000 4.5162 s 51.3995 s
2. Comparison between ADM and predictor–corrector scheme. Both of them are time
domain methods. Comparison of time complexity and space complexity of the
two methods is shown in Table 8.6. It can be seen that the ADM is better than
predictor–corrector method from both of these two aspects. The data in last three
lines of Table 8.6 is the requirement time for getting time series with the differ-
ent lengths (the results are different with different CPU frequencies. Here, the
computer parameter is Intel Dual E2180 2.0 GHz). It can be seen that the growth
rate of time required for predictor–corrector method is much faster than that of
Adomian decomposition algorithm. So, we can get the conclusion that Adomian
decomposition algorithm is better than predictor–corrector scheme when solving
fractional-order chaotic system.
In this section, we analyzed three simulation methods of fractional-order chaotic

system. For different methods, design methods and simulation steps are described.
Furthermore, to show the effectiveness of different simulation methods, the fractional-
order diffusionless Lorenz chaotic system is simulated based on Matlab, Simulink,
and EWB.
8.5 Dynamics of Fractional-Order Unified System Based

on Frequency Domain Method
In this section, we use dynamic simulation method to analyze fractional-order unified
system based on Matlab/Simulink.
8.5.1 Fractional-Order Unified Chaotic System Mode
In 2002, a new chaotic system is proposed [84], which connects Lorenz attractor and
Chen attractor. Meanwhile, Lü system is a special case of this system. So it is called
unified chaotic system [189], and its mathematic model is
⎧
⎪
⎨ẋ = (25a + 10)(y – x)
⎪
ẏ = (28 – 35a)x – xz + (29a – 1)y , (8.68)
⎪
⎪
⎩ż = xy – (a + 8)/3z
where a is the system parameter. When a ∈ [0,1], it is chaotic. This system represents a
family of chaotic system. Chaotic attractor in x–z plane of unified chaotic system with
different parameters is shown in Figure 3.20.
After replacement of integer-order differential operator with fractional-order dif-
ferential operator, fractional-order unified system becomes
⎧ !
⎪ d x
⎨ dt! = (25a + 10)(y – x)
⎪
d" y
= (28 – 35a)x – xz + (29a – 1)y , (8.69)
⎪
⎪ dt"
⎩ d# z a+8
dt# = xy – 3 z
where !, ", and # are fractional orders and system parameter a ∈ [0,1]. The fractional
order for different variables x, y, and z may be different. The principle is the same for
different fractional orders or the same fractional order. Here, we set 0 < ! = " = # =
q ≤ 1 to analyze fractional-order unified system.
8.5.2 Dynamic Simulation of Fractional-Order Unified Chaotic System
In this section, we focus on the dynamics of the fractional-order unified chaotic system
by Simulink simulation
8.5.2.1 Simulation Model

When a = 1, eq. (8.69) becomes
⎧ q
⎪ d x
⎨ dtqq = 35(y – x)
⎪
d y
dtq = –7x – xz + 28y
. (8.70)
⎪
⎪
⎩ dq z = xy – 3z
dtq
According to eq. (8.70), when we design the model of fractional-order system, some
modules are required, including state-space module, gain module, sum module, and
produce module. Based on Simulink, we copy the required modules from module lib-
rary to edit window, and then connecting these modules according to system (8.70) as
8.5.2.2 Setting Simulation Parameter

After the model is set up simulation environment and parameters are required to be
configured. By choosing menu “Simulation: Simulation Parameters” in the model
window, the simulation configuration dialog will pop up. Then we modify simula-
tion configuration. Almost all of the modules have dialog box for property setting. By
double-clicking the module or clicking “Edit: BLOCK Propertied” after choosing the
module, a basic properties dialog box will be opened. In this dialog, users can set the
basic properties of the module.
State-Space1
xʹ = Ax + Bu
y = Cx + Du
35
State-Space2
Gain
xʹ = Ax + Bu
y = Cx + Du
–7
28
Gain2
Gain1
State-Space3
xʹ = Ax + Bu
y = Cx + Du
Product Gain3
Product1
Figure 8.8: Simulation model of fractional-order unified system with a = 1.
In state-space module, the values of parameters A, B, C, and D are determined by the

value of fractional order as explained in Section 8.4.1. The initial values of system are
set by the initial condition. The other parameters are set as default values. Here, the
initial values are set as (0.1, 0.1, 0.1) and simulation time is set as 30 s. The ODE45
differential equation solver is used to solve the equation.
8.5.3 Dynamics of Fractional-Order Unified System with Fixed Parameter
According to eq. (8.70), we analyze dynamics of the fractional-order unified system

with q = 0.1–1, changing parameter or order.
8.5.3.1 Direct Observation Analysis

According to direct observation method, period motion in the phase space cor-
responds to a closed curve, while chaotic behavior corresponds to a never closed
trajectory which is random separated in a certain area (strange attractor). After we
finish the simulation of fractional-order chaotic system, the time sequence diagram
and system phase diagram are plotted. Then we can preliminarily determine whether
the fractional-order system is chaotic.
According to the fractional-order unified system model, the Scope module and
XYgraph module are used to observe simulation results.
60 40
40
40
20
z
z
20
20
(a) (b) (c)
0 0 0
–20 0 20 –20 0 20 –20 0 20
x x x
40 40 40
20 20 20
z
z
(d) (e) (f)
0 0 0
–20 0 20 –20 0 20 –20 0 20
x x x
40 40 40
20 20 20
z
z
(g) (h) (i)
0 0 0
–20 0 20 –20 0 20 –20 0 20
x x x
Figure 8.9: Attractors in the x–z plane of fractional-order unified system with different orders:
(a) q = 0.1, (b) q = 0.2, (c) q = 0.3, (d) q = 0.4, (e) q = 0.5, (f) q = 0.6, (g) q = 0.7, (h) q = 0.8, and (i)
q = 0.9.
Let q = 0.1–1 and set step size as 0.1. Simulation time is set to 40 s. For different frac-
tional orders, we get the trajectory in x–z phase plane as shown in Figure 8.9. It can be
seen that there exist the random separate states in the phase space when q = 0.1–0.3
and q = 0.8–1. It means chaotic attractor exists and the system is chaotic in this case.
When q = 0.4–0.7, the phase curve finally converges to one point, which means the
system is in periodic state.
By observing phase diagrams of the fractional-order unified system directly, we
get the following conclusions. When q = 0.1–0.3, the system is chaotic. When
q = 0.4–0.7, the system is periodic. When q = 0.8–1, the system shows chaotic behavior
again. Therefore, fractional-order unified system is piecewise chaotic.
8.5.3.2 Power Spectrum Analysis

In this section, we will use power spectrum analysis method mentioned in Section 2.2
to analyze the fractional-order unified system. Power spectrum analysis can provide
frequency information of signals. Discrete power spectrum corresponds to periodic
behavior, while continuous power spectrum corresponds to chaotic behavior. So we
1 1 1
p(t)
p(t)
p(t)
0.5 0.5 0.5
(a) (b) (c)

0 0 0
0 0.5 0 0.5 0 0.5
f f f
1 1 1
p(t)
p(t)
p(t)
0.5 0.5 0.5
(d) (e) (f)

0 0 0
0 0.5 0 0.5 0 0.5
f f f
1 1 1
p(t)
p(t)
p(t)
0.5 0.5 0.5
(g) (h) (i)

0 0 0
0 0.5 0 0.5 0 0.5
f f f
Figure 8.10: Normalized power spectrum density of fractional-order unified system with different
orders: (a) q = 0.1, (b) q = 0.2, (c)q = 0.3, (d) q = 0.4, (e) q = 0.5, (f) q = 0.6, (g) q = 0.7, (h) q = 0.8,
and (i) q = 0.9.
can tell whether the system is chaotic or not by observing power spectrum. Based on
Matlab, we calculated normalized power spectrum when fractional order q = 0.1–1,
which is shown in Figure 8.10.
From Figure 8.10, we get the following conclusions. When q = 0.8–1 and
q = 0.1–0.3, the power spectrum is continuous, which means the system is chaotic.
When q = 0.4–0.7, the power spectrum is discrete, which means the system is peri-
odic. Therefore, by observing power spectrum, the fractional-order unified system
is piecewise chaotic, which is consistent with the conclusion obtained by direct
observation.
8.5.3.3 Analysis of the Largest Lyapunov Exponent

Lyapunov exponent represents the average ratio of convergence or divergence of
the system variables in the phase space. It is an important parameter to determine
whether the system is chaotic or not. If the largest Lyapunov exponent is greater than
zero, the system is chaotic definitely.
Table 8.7: Largest Lyapunov exponent of unified system with different order q.
q System order The largest LE
0.1 0.3 0.2180

0.2 0.6 0.1616
0.3 0.9 0.1249
0.4 1.2 –0.0050
0.5 1.5 –1.4618
0.6 1.8 –0.0032
0.7 2.1 –0.0078
0.8 2.4 0.0199
0.9 2.7 0.0756
1 3 0.0263
Here, we use Wolf method [190] to calculate the largest Lyapunov exponent of
fractional-order unified system. When the fractional order q = 0.1–1 with step size
of 0.1, the results are shown in Table 8.7. When q = 0.8–1 and q = 0.1–0.3, the largest
Lyapunov exponent is positive, which means the system is chaotic. When q = 0.4–0.7,
the largest Lyapunov exponent is zero, which means the system is periodic. Thus, from
the largest Lyapunov exponent, we can get the conclusion that the system is piecewise
chaotic [191].
8.5.4 Dynamics of Fractional-Order Unified System with Different Parameter
8.5.4.1 Dynamics with Different Parameter and Order

We analyzed chaotic characteristic of fractional-order unified system with fixed para-
meter (a= 1). Next, we use the above methods to analyze dynamics of fractional-order
unified system when parameter a = 0–1 and step size is 0.1. The variation of fractional
order is still between 0.1 and 1 and its step size is 0.1. Dynamics of fractional-order
unified system with different parameters is shown in Table 8.8. However, when q = 1,
the system is integer-order unified system. When parameter varies from 0 to 1, the
system is always chaotic. From Table 8.8, we can get the following conclusion. When
a = 0, system changes from chaotic to nonchaotic with the increase of order q. More
specifically, system is chaotic at q = 1 and nonchaotic at the range of 0 < q ≤ 0.9.
When a = 0.1–0.3, the system is chaotic at q = 1 and q = 0.1. It is nonchaotic at the
range of 0.2 ≤ q ≤ 0.9. When a = 0.4–0.7, the system is chaotic at q = 1 and at the
range of 0.1 ≤ q ≤ 0.3. It is nonchaotic at the range of 0.4 ≤ q ≤ 0.9. When a = 0.8–1,
the system is chaotic at the range of 0.8 ≤ q < 1 and 0.1 ≤ q ≤ 0.3, and it is nonchaotic
at the range of 0.4 ≤ q ≤ 0.7.
Obviously, dynamics of fractional-order unified chaotic system is related to both
the system parameter and the fractional order. When the system parameter is fixed,
Table 8.8: Status of fractional-order unified system with different parameter.
Parameter a Fractional order Largest LE Status
0 < q ≤ 0.9 [–341.0837, –0.0459] Nonchaotic

0
q=1 0.0521 Chaotic
q = 0.1 [0.1418, 0.0605] Chaotic
0.1 ≤ a ≤ 0.3 0.2 ≤ q ≤ 0.9 [–339.9469, –0.0005] Nonchaotic
q=1 [0.0730, 0.0886] Chaotic
0.1 ≤ q ≤ 0.3 [0.2250, 0.0694] Chaotic
0.4 ≤ a ≤ 0.7 0.4 ≤ q ≤ 0.9 [–331.3635, –0.0085] Nonchaotic
q=1 [0.0509, 0.0742] Chaotic
0.1 ≤ q ≤ 0.3 [0.1206, 0.0686] Chaotic
0.8 ≤ a ≤ 1 0.4 ≤ q ≤ 0.7 [–330.7911, –0.0028] Nonchaotic
0.8 ≤ q ≤ 1 [0.0104, 0.0756] Chaotic
the system is piecewise chaotic. When the fractional order is fixed, it is possible for
the system to change from nonchaotic to chaotic. Besides, it can be seen that with the
increase of parameter and order, the chaotic range of the system expands, while the
nonchaotic range of order narrows.
8.5.4.2 Routes to Chaos with Fixed Fractional Order and Parameter Varying
We take q = 0.3 as an example to discuss the routes to chaos of the fractional-
order system. When the initial values are (–4.7508, –7.3886, 12.3605) and parameter
a = 0–1, the phase diagram on the x–z plane of fractional-order unified system is
shown in Figure 8.11. When the system parameter a is relatively small, the system
40 40 40
(a) (b) (c)
20 20 20
0 0 0
–20 0 20 –20 0 20 –20 0 20
40 40 40
(d) (e) (f)
20 20 20
0 0 0
–20 0 20 –20 0 20 –20 0 20
Figure 8.11: Attractors in the x–z phase of unified system with different parameter and q = 0.3:
(a) a = 0, (b) a = 0.1, (c) a = 0.2, (d) a = 0.3, (e) a = 0.45, and (f) a = 0.49.
converges to the equilibrium point (as shown in Figure 8.11(a), 8.11(b), and 8.11(c)).
However, with the increase of a, the evolution time of state variable tending to the
equilibrium point becomes longer (as shown in Figure 8.11(d) and 8.11(e)). As a further
increase to a certain value, such as a = 0.49, system reaches a border of attrac-
tion basin. At this time, system enters into chaos through boundary crisis bifurcation
(as shown in Figure 8.11(f)). However, if we pursue accurate analysis for routes to
chaos of fractional-order system, complex mathematical analysis and calculation are
required.
By observing system phase diagrams and calculating power spectrum and the
largest Lyapunov exponent, we analyzed dynamical characteristic of fractional-order
unified system in detail. We also find that system shows piecewise chaotic behavior
with the variation of system parameter and differential order.
8.6 Dynamics of Fractional-Order Diffusionless Lorenz System

Based on Time Domain Method
8.6.1 Fractional-Order Diffusionless Lorenz System Model
The mathematic model of diffusionless Lorenz system is

⎧
⎪
⎨ẋ = –y – x
⎪
ẏ = –xz , (8.71)
⎪
⎪
⎩ż = xy + R
where R is the system parameter. When R ∈ (0, 5), there exist equilibrium points
√ √
(x∗ , y∗ , z∗ ) = (± R, ∓ R, 0). The eigenvalues of this system satisfy the equation
+3 + +2 + R+ + 2R = 0. When R = 3.4693, the chaotic attractor of diffusionless Lorenz
system is shown in Figure 8.12(a). Its maximum Kaplan–Yorke fractal is 2.23542 [192].
10 (a) 10 (b)
0 0
z
–10 –10
10 10 10 10
0 0 0 0
y –10 y –10
–10 x –10 x
Figure 8.12: Attractors of diffusionless Lorenz system with R = 3.4693: (a) ! = " = # = 1 and
(b) ! = " = # = 0.95.
8.6 Dynamics of Fractional-Order Diffusionless Lorenz System Based 267
Correspondingly, the fractional-order diffusionless Lorenz system is

⎧ !
⎪ d x
⎪
⎪ = –y – x
⎪
⎪ !
⎨ dt"
d y
= –xz , (8.72)
⎪ "
⎪ ddt# z
⎪
⎪
⎪
⎩ = xy + R
dt#
where !, ", and # are the fractional order and 0 < !, ", # ≤ 1. When ! = " =
# = 0.95 and R = 3.4693, the chaotic attractor of diffusionless Lorenz system is shown
in Figure 8.12(b).
According to the previous discussion about time domain solution algorithm of the
fractional-order system, we can get the iterative calculation equation of fractional-
order diffusionless system as shown in eq. (8.64). Next, we analyze its dynamic
characteristic in detail.
8.6.2 Chaotic Characteristic of Fractional-Order Diffusionless Lorenz System
Here, we adopt Wolf algorithm [190] calculating the largest Lyapunov exponent. Let
! = " = # = q and control the variation of parameter, we can get the effective chaotic
range of frictional-order diffusionless system as shown in Figure 8.13. Obviously, there
exist three different dynamic statuses: limit circle, chaos, and convergence. If the
integer-order system is chaotic, the corresponding chaotic range of fractional-order
system will narrow slowly with the increase of system control parameter. If the integer-
order system shows limit circle, the corresponding fractional-order system is chaotic.
But the chaotic range will narrow quickly with the increase of system parameter. When
R = 8, the largest Lyapunov exponent of fractional-order diffusionless Lorenz system
1.05
1
Limit circle
0.95 Chaotic
q
0.9
Converge
0.85
0.5 2 4 6 8 10
R
Figure 8.13: Chaos range of fractional-order diffusionless Lorenz system.

0.1
0.05
The largest LE
–0.05
–0.1
0.9 0.95 1
q
Figure 8.14: The largest Lyapunov exponent of fractional-order diffusionless Lorenz system
with R = 8.
is shown in Figure 8.14. Obviously, its chaotic characteristic is consistent with the
chaotic range as shown in Figure 8.13.
8.6.3 Bifurcation Analysis with Varying System Parameter R
Here, we set the fractional order ! = " = # = 0.95 and let the system control parameter
R vary from 0.5 to 11 with the step size of 0.01. The initial values are x(0) = –0.0249,
y(0) = –0.1563, and z(0) = 0.9441, and the simulation time is 140 s. The bifurc-
ation diagram of fractional-order system is shown in Figure 8.15. Obviously, when
the order of fractional-order diffusionless system is 2.85 and R varies, the system is
chaotic with a periodic window. When the control parameter R decreased from 11, the
30
25
20
15
Zmax
10
5
0
–5
–10
0 1 2 3 4 5 6 7 8 9 10 11
R
Figure 8.15: Bifurcation diagram of fractional-order diffusionless system with R varying at q = 0.95.
18
14
Zmax
10
2
9 9.4 9.8 10.2 10.6 11
R
Figure 8.16: Periodic window of fractional-order diffusionless system with R varying at q = 0.95.
fractional-order diffusionless system enters into chaotic by period-doubling bifurc-

ation as shown in Figure 8.16 (the step size is 0.005). In periodic window, several
bifurcations are observed, such as an interior crisis bifurcation at R ≅ 5.57, a flip
bifurcation at R ≅ 6.02, and a tangent bifurcation at R ≅ 6.55. In the interior crisis,
the chaotic attractor collides with an unstable periodic orbit or limit cycle within its
basin of attraction. When the collision occurs, the attractor suddenly expands in size
but remains bounded. For the tangent bifurcation, a saddle point and a stable node
coalesce and annihilate one another, producing an orbit that has periods of chaos in-
terspersed with periods of regular oscillation [193]. This same behavior occurs in the
periodic windows of the logistic map, including the small windows within the lar-
ger windows [194]. In this periodic window, we also observe the route to chaos by a
period-doubling bifurcation.
8.6.4 Bifurcation Analysis with Varying Fractional Order q
Now, we set ! = " = # = q, R = 8 and q varying from 0.9 to 1. The initial val-
ues and running time kept the same as above, but the step size of R is 0.0005. The
system bifurcation diagram is shown in Figure 8.17. The system enters into chaos by
period-doubling bifurcation with decreasing fractional order. It is interesting to note
that a chaotic transient is observed when q is less than 0.912. The state-space traject-
ory is shown in Figure 8.18(a) for q = 0.911, which suddenly switches to a pattern
of oscillation that decays to the right equilibrium point. The time domain waveform
of variation z(t) shown in Figure 8.18(b) eventually converges to the fixed point. On
the average, the transform from chaotic behavior into damped state requires about
70 oscillations. When q is relatively greater, its chaotic behavior will remain longer.
Similar phenomenon has been reported in the standard Lorenz system [195]. The
fractional-order system enters to chaos by a period-doubling bifurcation as shown
25
20
15
Zmax
10
0
0.9 0.92 0.94 0.96 0.98 1
q
Figure 8.17: Bifurcation diagram of fractional-order diffusionless system with q varying at R = 8.
20
(b)
20
(a) 10
0 0
z
–10
–20
20 Chaotic oscillation Decay
10 –20
0 0 50 100
y 0
–20 –10 x t/s
Figure 8.18: Transient chaotic behavior in system (8.72) with R = 8 and q = 0.911:
(a) attractor diagram in the x–y–z plane and (b) time domain waveform.
in Figure 8.19(a) with steps in R of 0.0002. System is chaotic when q ∈ [0.92,0.962].

To observe dynamics in the periodic window, the periodic window is expanded as
shown in Figure 8.19(b) with steps of 0.0001. Obviously, there exist three kinds of
bifurcations, such as a tangent bifurcation, a flip bifurcation, and an interior crisis.
The fractional order parameter can be taken as a bifurcation parameter, just like the
control parameter.
8.6.5 Bifurcations with Different Fractional Order
1. Fix " = # = 1, R = 8 and let ! vary. The system is calculated numerically for
! ∈ [0.4,1] with an increment of 0.002. The bifurcation diagram is shown in Fig-
ure 8.20(a). Obviously, the fractional-order system is chaotic when ! ∈ [0.43,0.94)
20 25
(a) (b)
16 20
12 15
Zmax
Zmax
8 10
4 5
0 0
0.95 0.96 0.97 0.98 0.99 1 0.93 0.934 0.938 0.942 0.946 0.95
q q
Figure 8.19: Bifurcation diagram of fractional-order diffusionless Lorenz system at R = 8:

(a) q ∈ [0.95, 1] and (b) q ∈ [0.93, 0.95].
25 18
(a) (b)
20 16
15 14
10
12
5
Zmax
Zmax
10
0
–5 8
–10 6
–15 4
–20 2
0.4 0.5 0.6 0.7 0.8 0.9 1 0.48 0.49 0.5 0.51 0.52
α(β = γ = 1) α(β = γ = 1)
Figure 8.20: Bifurcation diagrams of fractional-order diffusionless Lorenz system with ! at " = # = 1:
(a) ! ∈ [0.4, 1] and (b) ! ∈ [0.48, 0.52].
and it is periodic when ! ∈ (0.484,0.503). The periodic window is expanded

as shown in Figure 8.20(b) with steps of 0.0002. When ! increases from 0.4, or
decreases from 1, a period-doubling bifurcation route to chaos is observed.
2. Fix ! = # = 1, R = 8 and let " vary. The step of " is 0.001. The system is calculated
numerically for " ∈ [0.75,1] with an increment of " equal to 0.001. The bifurca-
tion diagram is shown in Figure 8.21. Obviously, when " decreased from 1, the
fractional-order system enters into chaos by period-doubling bifurcation. When "
is less than 0.8, the system converges to a fixed point. Thus, the minimum order
of fractional-order diffusionless Lorenz system is 2.8 in this case.
3. Fix ! = " = 1, R = 8 and let # vary. The step of # is 0.001. The system is calculated
numerically for # ∈ [0.8,1]. The bifurcation diagram is shown in Figure 8.22(a).
It has similar dynamics with the previous two cases, which enters into chaotic
by period-doubling bifurcation. Finally, the system converges to a fixed point at
25
20
15
Zmax
10
0
0.75 0.8 0.85 0.9 0.95 1
β (α = γ = 1)
Figure 8.21: Bifurcation diagram of fractional-order diffusionless system with " at ! = # = 1.
25
(a) 24 (b)
20 22
20
15
18
Zmax
Zmax
10 16
5 14
12
0
10
–5 8
0.8 0.84 0.88 0.92 0.96 1 0.81 0.82 0.83 0.84 0.85 0.86
γ (α = β = 1) γ (α = β = 1)
Figure 8.22: Bifurcation diagrams of fractional-order diffusionless Lorenz system with # at ! = " = 1:
(a) # ∈ [0.8, 1] and (b) # ∈ [0.81, 0.86].
# = 0.816. However, its range of chaotic order is smaller. The dynamic behaviors
in the periodic window are similar with the case of (2) as shown in Figure 8.22(b)
with steps of 0.0002.
The analysis results mentioned above show the fractional-order diffusionless Lorenz
system has rich dynamical behaviors. It provides a theoretical and simulation basis
for practical application of fractional-order chaotic system.
Questions
1. What is the difference and relationship between fractional-order chaotic system
and integer-order system?
Questions 273
2. For fractional-order chaotic system, which solving method is the best?

3. What are the methods of dynamic analysis for fractional-order chaotic system?
4. Compare characteristics among different simulation methods for fractional-order
chaotic system.
5. Choose an integer-order chaotic system to analyze its corresponding fractional-
order dynamic characteristic.
9 Simulation and Hardware Implementation
of Chaotic System
Chaos system can be applied in many fields, such as secure communications, in-
formation encryption, system control, system optimization, and so on. To develop the
applications of chaos theory, the hardware design and implementation of chaotic sys-
tem must be solved, including the design and implementation of chaotic analog circuit
and chaotic digital circuit. In this chapter, we focus on the simulation and hardware
implementation of chaotic system.
9.1 Dynamic Simulation of Chaotic System
9.1.1 Dynamic Simulation Steps on Simulink
Dynamic simulation steps of Simulink are shown in Figure 9.1. The steps include
establishing simulation model, setting system parameters and observation window,
analyzing simulation results, and modifying model and parameters [188]. A detailed
description is presented as follows:
1. Establishing simulation mathematical model. It is the key for the simulation of
chaos system. According to the chaotic system equation, the whole system is sim-
plified to the source system. Functions of the system are determined, and each
part of the function is modeled. We should find out the relationship between
the various parts and draw the system flow diagram model. With the increase
of the modules and complexity of the system, the model will become more and
more complicated. To reduce the number of modules in the model window, we
can group the same function modules and simplify the complex problem by
packaging subsystem. For example, to observe the time evolution of the syn-
chronization error signal (|e| = (x2 – x1 )2 + (y2 – y1 )2 + (z2 – z1 )2 ), the module
is constructed as shown in Figure 9.2(a), which can be packaged into a subsys-
tem as shown in Figure 9.2(b), which simplifies the structure of the simulation
system.
2. Building simulation system model. According to the established mathematical
model, the function module is dragged into the editing window from the Simulink
model library. After the modules are connected, the simulation model of chaotic
system is constructed.
3. Setting simulation parameters. It is an important part of the simulation process,
which may directly affect the simulation times and results. Parameters setting in-
clude setting operating system parameters (such as the system running time, the
integration step size, the tolerable error, the integral algorithm) and the operation
9.1 Dynamic Simulation of Chaotic System 275
Establish Set Analyze

Establish Set system
simulation observation simulation
math model parameters
system window results
Modify model and the parameters
Figure 9.1: Matlab/Simulink simulation steps.
1
x
2 Product1
x'
MATLAB
3 1 x
y Function
d x'
sprt
4 Product2 y
y' d
y'
5
z z
6 Product3 z'
z' distance
(a) (b)
Figure 9.2: Error calculation submodule for synchronization system: (a) error signal function module
and (b) packaged subsystem.
parameters of the function modules (such as the initial value, signal amplitude,
frequency). The former is finished in the parameter submenu of the Simulink
menu, and the latter is finished in the pop-up window by double-clicking the
module.
4. Setting observation window. The observation module is used to observe the op-
eration of the simulation system, so as to adjust the parameters and analyze
the results. The observation modules include the oscilloscope (Scope), XYgraph,
Display, to workspace, and so on.
5. Analyzing simulation results and correcting model or parameter. We can judge
whether the simulation results are consistent with that of the theoretical ana-
lysis by analyzing the output waveform and output data of the simulation system.
Generally, it is difficult to succeed at a time. Many factors affect the simulation
results. First, you need to check whether the simulation system is consistent
with the theoretical model or whether the parameter settings are correct. If the
correct simulation results cannot be obtained by checking and modifying the
simulation parameters, you need to consider whether the simulation model is
correct. We need to modify the model and then simulate it until the results are
satisfactory.
276 9 Simulation and Hardware Implementation of Chaotic System
9.1.2 Simulation Result Analysis and Performance Improvement
Outputting simulation results (graphics) to the word document is an important way to

observe the results. First, the output signal is written into the return variable. That is
to say, the Out module in the sink library is connected as the output. After choosing
the variable time (I/O) and output (Output) box and setting the corresponding time
variable tout() and output variable yout(), the simulation is operated. Finally, we can
run the Matlab command plot(tout, yout) in the command window. In addition, by
using “Figure Copy” in the Edit Windows of the submenu, one can get the copied
graph, and then paste it into the drawing software window. To improve the resolution,
the graph is generally processed by graphics processing software and then be pasted
into the document.
Note that the simulation time is different from the time of simulation running.
The simulation time is defined as the difference between the Start time and the Stop
time. The time of simulation running is the time to run a simulation. In general, the
two are not equal. The time required to run a simulation depends on many factors, in-
cluding the complexity of the model, the maximum and minimum step size, computer
performance, and so on.
Several reasons will lead to simulation time too long. For example, the simulation
step size and the tolerance error are too small. When the model contains Fcn Matlab
module, the simulation of each step needs to explain the program, which will greatly
reduce the simulation speed, so it is best to use the S-function M file. S-function of M
file also allows Simulink to call the Matlab interpreter at each simulation time, so it
can be replaced by the C-mex function or equivalent subsystem.
9.2 Circuit Simulation of Chaotic System

Circuit simulation experiment method is used to simulate the chaotic system by the
electronic circuit diagram and to study the characteristics of the chaotic system by
analyzing the experimental results. For example, the Electronics Workbench (EWB)
software platform is used to design the chaotic system.
EWB is a virtualization working platform that simulates the analog circuit. It has
a more perfect model library and several common analytical instruments. It can be
used to analyze the electronic circuit and directly output the designed circuit files to
common electronic circuit layout software, such as Protel. The experimental analysis
steps of the chaotic system are shown in Figure 9.3 [188].
1. According to the chaotic system, design the EWB simulation circuit of a chaotic
system.
2. Choose the components in the window of EWB and connect them.
3. Determine the component parameters, including the label of the components and
its parameter values.
Circuit diagram
EWB simulation circuit
Component parameters
Observe the
Adjust circuit
experimental results
N
Results are
right?
Data analysis
Figure 9.3: EWB simulation procedure of chaotic system.
4. Use EWB virtual instrument to observe the experimental results, analyze data, and
compare whether the results are consistent with that of the theoretical analysis.
Otherwise we should adjust the circuit.
9.3 Design of Chaotic Analog Circuit
9.3.1 Procedure of Chaotic Circuit Design
There are three kinds of chaotic circuit design methods, such as individual design,
modular design, and improved modular design. The advantage of the individual
design is that the number of circuit components is the least, but it needs a rich prior
knowledge and circuit design skills. The modular design is based on the nondimen-
sional state equation, which is universal but needs more components. The improved
modular design is applied to determine the system parameters by comparing the state
differential equation with the actual circuit state differential equation so that the num-
ber of components is the minimum. Chaotic circuit modular design flowchart is shown
in Figure 9.4. First, based on the dimensionless chaotic state equation, the phase
Chaotic state equation
Range of state variables
N
Beyond component range?
Variable ratio compression transform
Differential to integral conversion
Time scale transformation
EWB simulation test
Circuit design and hardware

implementation
Figure 9.4: Chaotic circuit modular design flowchart.
diagrams are calculated to determine the range of the state variables. If the variable is
too large to beyond the scope of the dynamic variables of components, the proportion
of compression transform is needed, and all variables will be reduced N times. If the
variables are within the range, it is not necessary to carry out the compression trans-
formation. Then the EWB circuit simulation is carried out, and the circuit is designed
finally.
Circuit design process mainly includes following five steps.
1. Variable ratio compression transformation. Because the common power supply
voltage is 15 V, the linear dynamic range of an operational amplifier (OA) is 13.5 V.
According to the simulation graph, one needs to analyze whether the variable
range is beyond. If it is beyond, it needs to do compression transformation.
2. Do timescale transformation.
3. Do differential integral transformation.
4. The equation was standardized to adapt to the inverse phase adder in the module
circuit.
5. According to the standard integral equation, the corresponding modular circuit is
designed.
Select needed signal

Connect the input
× Reverse phase Inverse integrator1 Inverter1

additive ratio1 F1 –x1 x1
f1
Reverse phase
× Inverse integrator2 Inverter2
additive ratio2 F2 –x2 x2
f2
∙ ∙ ∙
∙ ∙ ∙
∙ ∙ ∙
× Reverse phase Inverse

InverterN
additive ratioN integratorN –xN xN
fN FN
Figure 9.5: Block diagram of chaotic circuit design.
9.3.2 Modular Design of Chaotic Circuit
The modular design of chaotic circuit is designed mainly based on dimensionless

equation. The general principle block diagram is shown in Figure 9.5 [196]. The cir-
cuit consists of four modules, including nonlinear function, reverse phase additive
proportional arithmetic unit, reverse integrator, and inverter. A nonlinear function in-
cludes multiplication function, the absolute value function, and so on. Circuits of the
latter three modules mainly are built based on amplifiers.
9.3.3 Basic Components of Chaotic Circuit
In general, a chaotic circuit contains a nonlinear unit and a linear unit. A nonlinear
unit circuit includes analog multiplier, Chua’s diode circuit, the absolute value of the
arithmetic circuit and index operation circuit, and so on. Based on the OA device, the
common linear basic unit circuits are shown in Figure 9.6, and the input and output
expressions of each unit circuit are presented as follows.
1. Reverse phase proportion amplifier: vo = – RR21 vi as shown in Figure 9.6(a).

2. Reverse phase adder: vo = – RR31 vi1 + RR32 vi2 as shown in Figure 9.6(b).
R2
3. Subtractor: vo = R1 (vi2 – vi1 ) as shown in Figure 9.6(c).
R1 +R2
4. Proportional amplifier: vo = R1 vi as shown in Figure 9.6(d).
dv
5. Reverse phase differentiator: vo = –RC dti as shown in Figure 9.6(e).

6. 1
Reverse phase integrator: vo = – RC vi dt as shown in Figure 9.6(f).
R2 νi1 R1 R3
νi R1 νi2 R2
νo νo
(a) (b)
R2
νi
νi R1 νo
νo
νi2 R1 R1 R2
R2
(c) (d)
R C
νi C νi R
νo νo
(e) (f)
Figure 9.6: Linear unit circuits: (a) reverse phase proportion amplifier; (b) reverse phase adder;
(c) subtractor; (d) proportional amplifier; (e) inverse phase differentiator; and (f) reverse phase
integrator.
Now taking the simplified Lorenz system as an example, the modular design
process is described in detail. The mathematical model of simplified Lorenz
system is [56]
⎧
⎪
⎨ẋ = 10(y – x)
⎪
ẏ = –xz + (24 – 4c)x + cy , (9.1)
⎪
⎪
⎩ż = xy – 8z/3
where c is the system parameter. When c ∈ [–1.59, 7.75], it is chaotic.

1. When c = 5, by numerical simulation, all the state variables of the system are
beyond the dynamic range of the device, so it is necessary to change the system
variables. Let x′ = 0.25x, y′ = 0.25y, and z′ = 0.25z, then system (9.1) becomes to
⎧
⎪ ′ ′ ′
⎨ẋ = 10(y – x )
⎪
′
ẏ = –4x z + 4x′ + 5y′ .
′ ′ (9.2)
⎪
⎪
⎩ż′ = 4x′ y′ – 8z′ /3
2. Do timescale transformation to system (9.2) and transform t to 40 t′ , where 40 is the

time-scaling factor. Let 40 = 1/(R0 C0 ), then we have
⎧
⎪ ′ ′ ′
⎨ẋ = R0 C0 10(y – x )
1
⎪
ẏ′ = 1 ′ ′
R0 C0 (–4x z + 4x +
′ 5y′ ) . (9.3)
⎪
⎪
⎩ż′ = ′ ′ ′
R0 C0 (4x y – 8z /3)
1
3. After the differential is transformed to integral, the system becomes

⎧
⎪ ′ ′ ′ ′
⎨x = – R0 C0 10((–y ) + x )dt
1
⎪
y′ = – 1 ′ ′ ′ ′
R0 C0 (4x z + 4(–x ) + 5(–y ))dt
′ . (9.4)
⎪
⎩z ′ = –
⎪ ′ ′ ′ ′
R0 C0 (–4x y + 8z /3)dt
1
4. According to system (9.4), the basic circuit units include inverter, inverting adder,
integrator, and multiplier. The circuit is shown in Figure 9.7.
In Figure 9.7, x, y, and z represent the output voltage values, respectively. According
to the circuit theory, the state equation of the circuit is obtained:
⎧
⎪ R3
y – RR31 x
⎪ẋ =
1
⎪ R4 C1
⎨ R2
ẏ = 1
R11 C2 – R10 0.1xz + RR107 x + R10
R8 y . (9.5)
⎪
⎪ R9
⎪
⎩ż = 1 R16 R16
R17 C3 R14 0.1xy – R15 z
R1 R3 C1
x R6
R2
y R4
R5
x x
R7
x
R8 R10 C2
x Y y R13
R9
–z X R11
R12
y y
x Y
R14 R16 C3
X
y R19
R15
z R17
R18
z z
Figure 9.7: Circuit diagram of simplified Lorenz system.

Comparing eq. (9.4) with eq. (9.5), the value of the components is obtained as
Ri = 10 kK (i = 1, 2, 4, 5, 6, 11, 12, 13, 17, 18, 19), Ri = 100 kK (i = 3, 10, 16), R7 = 20 kK,
R8 = 25 kK, R9 = R14 = 2.5 kK, R15 = 37.5 kK, Ci = 100 nF (i = 1, 2, 3).
9.4 Improved Modular Design of Chaotic Circuit

The modular design of chaotic circuit is intuitive, simple, and easy to debug, but it has
a large number of components. The circuit based on the improved modular design has
fewer components, and its precision is higher than that of the standard-type chaotic
circuit. The basic unit circuit module is shown in Figure 9.8. Here taking the multiscroll
Jerk system as an example, we illustrate the improved modular design in detail.
The equation of multiscroll Jerk system is [197]
⎧
⎪
⎨ẋ = y
⎪
ẏ = z , (9.6)
⎪
⎪
⎩ż = –x – y – !z + f (x)
where ! is the system parameter. When ! = 0.6 and f (x) = sgn(x)+sgn(x+2)+sgn(x–2),

it can produce a four-scroll chaotic attractor. According to the results of the numerical
simulation, the system variables are not beyond the dynamic range of the device, so it
is not necessary to do proportion transformation for variables.
Do timescale transformation for system (9.6), let t = 40 t′ , 40 = 10, 000, we obtain
⎧
⎪ dx
⎪
⎪ ′ = 10, 000y
⎪
⎪ dt
⎪
⎪
⎨
dy
= 10, 000z . (9.7)
⎪
⎪ dt ′
⎪
⎪
⎪
⎪
⎩ dz = –10, 000x – 10, 000y – 6, 000z + 10, 000f (x)
⎪
dt′
The circuit units for system (9.7) include basic unit circuit module, inverter module,
and nonlinear circuit module. Next, we design the system circuit based on OA and
current conveyor (CC).
R1
νi1
R2 C
νi2
R3
νi3
νo
if
Figure 9.8: General basic unit circuit module based on improved modular design.
9.4.1 Design of Multiscroll Jerk Circuit Based on OA
Four-scroll Jerk circuit based on OA devices is shown in Figure 9.9.

In Figure 9.9, x, y, and z represent the output voltage, respectively, the output
current value of the subcircuit f (x) is: i(x) = –sgn(x–2)/Rx1 –sgn(x)/Rx2 –sgn(x+2)/Rx3 .
According to the circuit theory, we can obtain the state equation of the circuit
⎧
⎪ 1
⎪ẋ = R1 C1 y
⎨
ẏ = R21C2 z . (9.8)
⎪
⎪
⎩ż = – 1 x
+ y
+ z
+ i(x)
C3 R5 R6 R7
Comparing eq. (9.7) with eq. (9.8), the values of the components are obtained as fol-
lows: Ri = 10 kK (i = 1, 2, 3, 4, 5, 6), Rxi = 135 kK (i = 1, 2, 3), R7 = 16.7 kK, Ci = 10
nF (i = 1, 2, 3). According to Figure 9.9, we do simulations on EWB, and the results are
9.4.2 Design of Multiscroll Jerk Circuit Based on CC
The CC is the active device commonly used in current mode circuit, and it has strong
versatility, high precision, less external devices, and good high-frequency character-
istics. It has been widely used in active power filter, analog signal processing, and
so on. Among the CCs, the second-generation current conveyor (CCII) is widely ap-
plied. Its circuit symbol is shown in Figure 9.11. Characteristics of the ports are iy = 0,
Vx = Vy , and iz = ±ix . When iz = ix , it is the second-generation current conveyor
(CCII+). When iz = –ix , it is the reverse second-generation current conveyor (CCII–).
(a) (b)
i (x)
f (x)
R5 R4
Rx1
R6 +2 V
R7 y x
Rx2 i(x)
R3
C3
C2
C1
Rx3
R2
R1 ‒2 V
z
y x
Figure 9.9: Multiscroll Jerk circuit based on OA devices: (a) main circuit and
(b) subcircuit of function f (x).
(a) (b)
2 2
1 1
0 0
z
y
‒1 ‒1
‒2 ‒2
‒5 ‒3 ‒1 1 3 5 ‒5 ‒3 ‒1 1 3 5
x x
Figure 9.10: Jerk circuit simulation with four scrolls based on the OA: (a) phase diagram on x–y plane
and (b) phase diagram on x–z plane.
iy
Vy Y iz
ix CC Z Vz
Vx X
Figure 9.11: CCII circuit symbol.
iy C C II +
Vy Y iy CC II –
Vz Vy Y
ix W Y Vz
Vx X Z iz ix W W
Vx X Z X Z iz
(a) C C
(b) CC II–
Figure 9.12: CCII implementation circuit : (a) CCII+ and (b) CCII–.
CCII+ can be implemented by employing an AD844 device, while CCII– can be realized
by cascading two AD844 devices as shown in Figure 9.12.
Based on CCII+, the common linear basic unit circuits are shown in Figure 9.13;
the input and output expressions of each unit circuit are shown as follows [198]:
1. Proportional amplifier: vo = RR21 vi , as shown in Figure 9.13(a).
2. Reverse proportional amplifier: vo = – RR21 vi , as shown in Figure 9.13(b).
dv
3. Inverting differential device: vo = –RC dti , as shown in Figure 9.13(c).

4. Inverting adder: vo = – RR31 vi1 + RR32 vi2 , as shown in Figure 9.13(d).

5. Reverse integrator: vo = – RC 1
vi dt, as shown in Figure 9.13(e).

6. 1
Integrator: vo = RC vi dt, as shown in Figure 9.13(f).
νi Y W νo Y W νo
CC CC
X Z νi X Z
R1 R2 R1 R2
(a) (b)
Y W νo Y W νo
CC CC
νi X Z νi1 X Z
R1 R3
C
νi2
R R2
(c) (d)
Y W νo νi Y W νo
CC CC
νi X Z X Z
R C R C
(e) (f)
Figure 9.13: Basic unit circuits: (a) proportional amplifier; (b) reverse proportional amplifier;
(c) reverse differentiator; (d) inverting adder; (e) reverse integrator; and (f) integrator.
i(x)
f (x)
Rwx N1
R3 W Y 2V Y W
CC CC
Z X X Z
R4 Ra Rb
y Rwx
R5 R7 Y W
R6 x x CC i(x)
Y W X Z
R1 CC Ra Rb
Y W y X Z Rwx
R2 CC ‒2V Y W
X Z C1
Y W CC
CC z C2 X Z
X Z Ra Rb
C3
(a) (b)
Figure 9.14: Four-scroll Jerk circuit based on CCII device: (a) main circuit and (b) subcircuit f (x).
Based on the mathematical models and system (9.7), the four-scroll Jerk circuit is
designed based on CCII as shown in Figure 9.14.
Nonlinear function f (x) is a step function, and a saturation function is achieved
based on the N1 network as shown in Figure 9.14(b). The saturation function sequence
is realized by parallel network. If Rb /Ra is large enough, the sequence of saturation
function can be approximated by infinite step function. The output current is i(x) =
–[sgn(x –2)+sgn(x)+sgn(x +2)]/Rwx . According to the circuit theory, the state equation
of the circuit is
(a) (b)
2 2
1 1
0 0
y
z
‒1 ‒1
‒2 ‒2
‒5 ‒3 ‒1 1 3 5 ‒5 ‒3 ‒1 1 3 5
x x
Figure 9.15: Four-scroll Jerk chaotic attractor based on CCII: (a) phase diagram on x–y plane and
(b) phase diagram on x–z plane.
(a) (b)
Figure 9.16: Phase diagram on x–y plane: (a) circuit implement based on OA and (b) circuit
implement based on CC.
⎧
⎪
⎨ẋ = R1 C1 y
1
⎪
ẏ = R21C2 z . (9.9)
⎪
⎪
⎩ż = – 1 x
+ y
+ z
+ i(x)
C3 R3 R4 R5
When comparing eq. (9.7) with eq. (9.9), the value of each component was obtained
as follows: Ri = 10 kK (i = 1, 2, 3, 4, 6), R5 = 16.7 kK, R7 = 20 kK, Rwx = 102 kK,
Ra = 470 K, Rb = 1 MK, Ci = 10 nF (i = 1, 2, 3). The circuit in Figure 9.14 is simulated on
EWB. According to Figure 9.14, we do simulations on EWB, and the results are shown
in Figure 9.15.
9.4.3 Multiscroll Jerk Circuit Implementation
According to Figures 9.9 and 9.14, the four-scroll Jerk circuit is designed based on
CC and OA devices, and the phase diagram is observed by oscilloscope as shown in
9.5 Digital Signal Processor Design and Implementation of Chaotic Systems 287
Figure 9.16. When compared with Figures 9.10 and 9.15, respectively, we concluded that
the results of actual circuit experiment and circuit simulation experiment are consist-
ent, and it verified the correctness of the circuit design and the physical realization of
the four-scroll Jerk system.
9.5 Digital Signal Processor Design and Implementation

of Chaotic Systems
Hardware implementation of chaotic system is a key problem for the applications
of chaotic system in engineering practice. With the development of microelectron-
ics and digital signal processing technology, the digital circuit realization of chaotic
system has become a research hotspot [199–201]. Compared with the analog devices,
digital signal processor (DSP) device has the characteristics of stability, repeatabil-
ity, programmable, and easy to implement. It brings about great developments of
digital signal processing in the fields of aviation, aerospace, radar, sonar, commu-
nication, domestic appliance, and so on and becomes the core of electronic system.
With increase in the calculation speed of DSP, the bandwidth of the real-time pro-
cessing is greatly increased, and the DSP is widely used in the application of chaotic
system.
Here, taking simplified Lorenz system and fractional-order simplified Lorenz sys-
tem as examples, we will study the DSP implementation of the chaotic system based
on DSP TMS320F2812.
9.5.1 DSP Experimental Platform of Chaotic System
The hardware platform to implement chaotic system based on DSP is shown in

Figure 9.17, in which the D/A converter is the 16-bit dual-channel DAC8552 chip, and
the interface chip is MAX3232. Their pin distribution is shown in Figure 9.18.
According to the system requirements, we can choose DSP chip with fixed point
or floating point. The main frequency of DSP chip will affect the operation speed,
and consequently affect the generating speed of the chaotic sequence. In addition,
the width of the data bus will affect the numerical accuracy of the chaotic sequence.
We need to consider the DSP chip peripherals to connect with other parts of the
Communication imterface
Computer
(MAX3232)
DSP
(TMS320F2812)
DA conversion
Oscilloscope
(DAC8552)
Figure 9.17: Hardware block diagram of chaotic system based on DSP.

C 1+ 1 16 VCC
V+ 2 15 GND
C1– 3 14 T1OUT
C2– 4 MAX3232 13 R1IN
VDD 1 8 GND C2+ 5 12 R1OUT
VREF 2 7 DIN V– 6 11 T1IN
DAC8552
VOUTB 3 6 SCLK 10 T2IN
T2OUT 7
VOUT B 4 5 SYNC R2IN 8 9 R2OUT
(a) (b)
Figure 9.18: Pin distribution of DAC8552 and MAX3232.
system conveniently. Here, we choose TMS320F2812 with 32-bit fixed point DSP chip,
amount of peripheral interfaces, and 150 MHz main frequency. Universal Asynchron-
ous Receiver/Transmitter (UART) serial interface is generally used in communication
interface, which transmits chaos sequences to the computer in real time. The D/A con-
verter converts the chaotic sequence generated by DSP into analog signals, which can
be observed on the oscilloscope. So it demands that the D/A conversion has the dual-
channel conversion function for synchronous output, and its conversion speed and
precision are high enough. The operation command words of DSP for the DAC8552 are
shown in Figure 9.19, where DB15–DB0 is the converted digital signal.
By employing CCS4.2 as the DSP integrated development environment, the exper-
imental platform of chaotic system based on DSP is shown in Figure 9.20.
The software flowchart for the realization of the chaos system on DSP is shown in
Figure 9.21, which includes DSP initialization, setting of the clock frequency, initializ-
ation of the related interface, and so on. Different chaotic system has different system
parameters, and the iteration step size, iteration number, and iteration initial value
also need to be determined. According to different solution algorithm, we can derive
the iterative calculation formula. Each iteration calculation results needs to be stored
as the initial value of the next iteration. If data output to D/A or PC, the corresponding
data processing is needed, as shown in Figures 9.22 and 9.23.
24 bits in the A data buffer written by the first step

Buffer
Res Res LDB LDA DC PD1 PD0 DB15 — DB1 DB0
select
0 0 0 0 × 0 0 0 D15 — D1 D0
24 bits in the B data buffer written by the second step
Buffer
Res Res LDB LDA DC PD1 PD0 DB15 — DB1 DB0
select
0 0 1 1 × 1 0 0 D15 — D1 D0
Figure 9.19: Control word outputted by DAC8552 dual channel.

Oscilloscope
D/A converter
TMS320F2812
UART experimental platform
Figure 9.20: Experimental platform of chaotic system based on DSP.
Start
DA To PC or PC
DSP initialization
DA ?
Parameters and initial values

Data processing1 Data processing2
Iterative computation
To DA To PC
Data entry
Data push N Finish?
Replace iteration Y
initial value End
Figure 9.21: Software flow of chaos system based on DSP.
Generation of chaotic sequence
Turn to positive number
Proportional amplification
Output to D/A converter
Figure 9.22: Data processing 1.
1. Data processing 1: To realize the D/A conversion correctly, a large enough positive
number is added to all of the output value of chaotic sequence, which is equival-
ent to coordinate translation on the output of the system. DAC8552 is a 16-bit D/A
converter, and the converted digital quantity must be positive integers between 0
Chaotic sequence value
Turn to positive number
The number of decimal

places to keep
Bit by bit turn to

ASCII code
Output to PC
Figure 9.23: Data processing 2.
and 65,535, while the majority value of the chaos sequence generated by DSP is
too small. So we need to multiply the sequence value with a positive integer so
that the digital numbers are distributed in the whole range of conversion as far as
possible.
2. Data processing 2: Similarly, the sequence value of variable is turned to positive.
To facilitate the computer display and process, each bit of the chaotic sequence
value is converted to ASCII code. Namely, each decimal adds 48 and transmits
to the computer bit by bit. Each complete sequence is separated with commas
(ASCII 0 × 2C). The length of the data can be chosen. When we analyzing it on
computer, the chaotic sequence value can be restored to the values calculated
by DSP.
The DSP implementation steps are presented by taking integer-order simplified Lorenz
system and its fractional-order counterpart as examples.
9.5.2 DSP Implementation of Integer-Order Chaotic System
Taking the simplified Lorenz system as an example, the DSP implementation of the
chaotic system based on the fourth-order Runge–Kutta algorithm is described. The
simplified Lorenz system equation is [56]
⎧
⎪
⎪ẋ = 10(y – x)
⎨
ẏ = (24 – 4c)x – xz + cy , (9.10)
⎪
⎪
⎩ż = xy – 8z/3
where c is the system parameter. When c ∈ [–1.59, 7.75], it is chaotic. By solving the
differential equations based on the fourth-order Runge–Kutta method, we obtain
⎧
⎪
⎨xn+1 = xn + (K11 + 2K12 + 2K13 + K14 )/6
⎪
yn+1 = yn + (K21 + 2K22 + 2K23 + K24 )/6 . (9.11)
⎪
⎪
⎩z = z + (K + 2K + 2K + K )/6
n+1 n 31 32 33 34
The recursive parameters are as follows:
⎧
⎪
⎨K11 = 10[(yn – xn )]T
⎪
K21 = [(24 – 4c)xn – xn zn + cyn ]T , (9.12)
⎪
⎪
⎩K = [x y – 8z /3]T
31 n n n
⎧
⎪
⎨K12 = 10[(yn + 0.5K21 ) – (xn + 0.5K11 )]T
⎪
K22 = [(24 – 4c)(xn + 0.5K11 ) – (xn + 0.5K11 )(zn + 0.5K31 ) + c(yn + 0.5K21 )]T , (9.13)
⎪
⎪
⎩K = [(x + 0.5K )(y + 0.5K ) – 8(z + 0.5K )/3]T
32 n 11 n 21 n 31
⎧
⎪
⎨K13 = 10[(yn + 0.5K22 ) – (xn + 0.5K12 )]T
⎪
K = [(24 – 4c)(xn + 0.5K12 ) – (xn + 0.5K12 )(zn + 0.5K32 ) + c(yn + 0.5K22 )]T , (9.14)
⎪ 23
⎪
⎩K = [(x + 0.5K )(y + 0.5K ) – 8(z + 0.5K )/3]T
33 n 12 n 22 n 32
⎧
⎪
⎨K14 = 10[(yn + K23 ) – (xn + K13 )]T
⎪
K24 = [(24 – 4c)(xn + K13 ) – (xn + K13 )(zn + K33 ) + c(yn + K23 )]T , (9.15)
⎪
⎪
⎩K = [(x + K )(y + K ) – 8(z + K )/3]T
34 n 13 n 23 n 33
where T is the iteration step size. According to eqs (9.11)–(9.15), setting initial values
x = 0.1, y = 0.2, z = 0.3, c = 5, and T = 0.01, the attractor phase diagram is generated
by the hardware implementation as shown in Figure 9.24.
Taking 4 bits after decimal point of the chaotic sequence generated by DSP,
and transmitting a total of 10,000 sequence values between 1,001 and 20,000 to the
computer, the phase diagram of the attractor is plotted as shown in Figure 9.25.
(a) (b) (c)

y
z
z
x x y
Figure 9.24: Attractors in an oscilloscope for the Lorenz system based on the DSP: (a) x–y plane;
(b) x–z plane; and (c) y–z plane.
(a) (b) (c)

15 18 18
0 9 9
y
z
‒15 0 0
‒15 0 15 ‒15 0 15 ‒15 0 15
x x y
Figure 9.25: Attractors of the simplified Lorenz system by experiment data from DSP: (a) x–y plane;
(b) x–z plane; and (c) y–z plane.
9.5.3 DSP Implementation of Fractional-Order Chaotic System
Except the difference of the iterative algorithm, the realization of fractional-order

chaotic system based on DSP is similar to that of the integer-order chaotic system
based on DSP. Fractional-order simplified Lorenz system is described as [186]
⎧
⎪ ∗ q
⎨ Dt0 x = 10(y – x)
⎪
∗ Dq y = (24 – 4c)x – xz + cy , (9.16)
⎪ t0
⎪
⎩∗ Dq z = xy – 8z/3
t0
where q is the order of fractional differential and c is the system parameter. Choosing
suitable q and c, it is chaotic. According to the Adomian decomposition algorithm, the
fractional differential equations are solved, and the first seven items are obtained as
follows [202]:
⎧
⎪
⎨C10 = xm
⎪
C20 = ym , (9.17)
⎪
⎪
⎩C = z
30 m
⎧
⎪
⎪ C = 10(C20 – C10 )
⎪
⎨ 11
C21 = (24 – 4c)C10 + cC20 – C10 C30 , (9.18)
⎪
⎪
⎪C = – 8 C + C C
⎩ 31 30 10 20
3
⎧
⎪
⎪ C = 10(C21 – C11 )
⎪
⎨ 12
C22 = (24 – 4c)C11 + cC21 – C11 C30 – C10 C31 , (9.19)
⎪
⎪ 8
⎪
⎩C32 = – C31 + C11 C20 + C10 C21
3
⎧
⎪
⎪C = 10(C22 – C12 )
⎨ 13
C23 = (24 – 4c)C12 + cC22 – C12 C30 – C11 C31 AA(2q+1)
2 (q+1) – C10 C32 , (9.20)
⎪
⎪
⎩C = – 8 C + C C + C C A(2q+1) + C C
33 3 32 12 20 11 21 A2 (q+1) 10 22
⎧
⎪
⎨C14 = 10(C23 – C13 )
⎪
A(3q+1)
C24 = (24 – 4c)C13 + cC23 – C13 C30 – (C12 C31 + C11 C32 ) A(q+1)A(2q+1) – C10 C33 , (9.21)
⎪
⎪
⎩C = – 8 C + C C + (C C + C C ) A(3q+1) + C C
34 3 33 13 20 12 21 11 22 A(q+1)A(2q+1) 10 23
⎧
⎪
⎪ C15 = 10(C24 – C14 )
⎪
⎪
⎪
⎪ A(4q+1)
⎪
⎪C25 = (24 – 4c)C14 + cC24 – C14 C30 – (C13 C31 + C11 C33 ) A(q+1)A(3q+1)
⎨
–C12 C32 AA(4q+1)
2 (2q+1) – C10 C34 , (9.22)
⎪
⎪
⎪
⎪C35 = – C34 + C14 C20 + (C13 C21 + C11 C23 )
8 A(4q+1)
⎪
⎪ A(q+1)A(3q+1)
⎪
⎪
3
⎩+C12 C22 A(4q+1) + C10 C24
A2 (2q+1)
⎧
⎪
⎪C16 = 10(C25 – C15 )
⎪
⎪
⎪
⎪ A(5q+1)
⎪
⎨C26 = (24 – 4c)C15 + cC25 – C15 C30 – (C14 C31 + C11 C34 ) A(q+1)A(4q+1)
⎪
A(5q+1)
–(C13 C32 + C12 C33 ) A(2q+1)A(3q+1) – C10 C35 , (9.23)
⎪
⎪
⎪
⎪ A(5q+1)
⎪C36 = – 3 C35 + C15 C20 + (C14 C21 + C11 C24 ) A(q+1)A(4q+1)
8
⎪
⎪
⎪
⎩+(C13 C22 + C12 C23 ) A(5q+1) + C10 C25
A(2q+1)A(3q+1)
⎡ ⎤ ⎡ ⎤
xm+1 C10 C11 C12 C13 C14 C15 C16
⎢ ⎥ ⎢ ⎥
⎣ ym+1 ⎦ = ⎣ C20 C21 C22 C23 C24 C25 C26 ⎦
zm+1 C30 C31 C32 C33 C34 C35 C36
T
hq h2q h3q h4q h5q h6q
1 . (9.24)
A(q + 1) A(2q + 1) A(3q + 1) A(4q + 1) A(5q + 1) A(6q + 1)
For all the above-mentioned equations, h is the iterative step size and A(⋅) is the gamma
function. According to the calculating process, A(⋅) and hnq are unchanged in the iter-
ative process, so to speed up the calculation, we need to calculate the relevant A(⋅) and
hnq before iteration, taking them as constants in iterative process. The initial values are
x0 = 0.1, y0 = 0.2, z0 = 0.3, c = 5, q = 0.9, and h = 0.01, and hardware implementation
of the attractor phase diagram is shown in Figure 9.26.
Taking 4 bits after decimal point of the chaotic sequence generated by DSP and
transmitting a total of 10,000 sequence values between 10,001 and 20,000 to the
computer, the phase diagram of the attractor is plotted as shown in Figure 9.27.
In this chapter, we study the simulation and circuit design techniques of the chaotic
system, including the dynamic simulation, circuit simulation, and realization of the
chaotic system. The realization of analog circuit and digital circuit of chaotic system
has laid an experimental foundation for the applications of chaotic system.
(a) (b) (c)
z
z
x x y
Figure 9.26: Attractors in an oscilloscope for the fractional-order simplified Lorenz system based on
the DSP: (a) x–y plane; (b) x–z plane; and (c) y–z plane.
(a) (b) (c)

15 18 18
0 9 9
y
z
‒15 0 0
‒15 0 15 ‒15 0 15 ‒15 0 15
x x y
Figure 9.27: Attractors of the fractional-order simplified Lorenz system by experiment data from DSP:
(a) x–y plane; (b) x–z plane; and (c) y–z plane.
Questions
1. Try to study a chaotic system by using different simulation methods and compare
the results.
2. What are the similarities and differences of the chaotic systems based on the OA
device and the CC device? What is the difference between the chaotic signals
generated by them?
3. To achieve the digital circuit implementation of a chaotic system, what are the
demands for solving algorithm of differential equations?
4. Is there finite precision effect in DSP implementation of the chaotic system, and
how to overcome it?
5. Design and implement a chaotic system circuit.
Appendix A: Important Academic Journals
in the Field of Chaos
No. Journals ESI subject Influence factor
1 Physical Review Letter Physics 7.943

2 Communications in Nonlinear Science and Physics 2.773
Numerical Simulation
3 Physical Review E Physics 2.313
4 Chaos Physics 2.188
5 Physics Letters A Physics 1.766
6 Physica D – Nonlinear Phenomena Physics 1.669
7 Chaos, Solitons and Fractals Physics 1.246
8 Chinese Physics B Physics 1.148
9 Chinese Physics Letters Physics 0.811
10 Journal of Nonlinear Mathematical Physics Physics 0.569
11 Acta Physica Sinica Physics 1.016
12 Nonlinear Analysis – Theory, Methods & Mathematics 1.64
Applications
13 Nonlinearity Mathematics 1.602
14 International Journal of Bifurcation and Chaos Mathematics 0.921
15 Nonlinear Analysis: Modelling and Control Mathematics 0.861
16 Advanced Nonlinear Studies Mathematics 0.538
17 Nonlinear Dynamics Engineering 3.009
18 IEEE Transactions on Circuits and Systems I/II Engineering 2.24/1.327
19 Electronics Letters Engineering 1.038
20 International Journal of Nonlinear Sciences and Engineering 0.622
Numerical Simulation
Appendix B: Matlab Source Programs
of Chaos Characteristic Analysis
Taking the simplified Lorenz system as an example, Matlab simulation programs are
presented. First, the analysis method of integer-order system is introduced, includ-
ing Runge–Kutta solving algorithm, Euler solving algorithm, Lyapunov exponents
(LE) calculation, plotting phase diagram, bifurcation diagram, Poincaré section, and
0–1 test method. Second, the programs about solving and analyzing algorithm of the
fractional-order chaotic system are designed, including frequency domain approach
(Simulink simulation), Adams–Bashforth–Moulton approach, Adomian decomposi-
tion method, C0 complexity algorithm, and chaos diagram method. Finally, Matlab
programs about chaotic synchronization algorithm are introduced for readers’ refer-
ence. Of course, readers can modify the relevant codes according to their own needs
to meet the system or research.
If a document begins with “function,” then it is a function. The reader needs to
be entered and saved in a separate .m file, so that you can call it when you need it.
Design program by employing function belongs to the module design approach, which
is beneficial to improve the programming efficiency.
B.1 Mode of the Simplified Lorenz System

The simplified Lorenz system is defined by
⎧
⎪
⎪ẋ = 10 (y – x)
⎪
⎨
ẏ = (24 – 4c) x – xz + cy , (B.1)
⎪
⎪
⎪
⎩
ż = xy – 8z/3
where c is the system parameter. When c ∈ [–1.59 7.75], the system is chaotic.
The fractional-order simplified Lorenz system is defined as
⎧ q
⎪ ∗ D x = 10(y – x)
⎪
⎪ t0
⎨
∗ Dq y = (24 – 4c)x – xz + cy . (B.2)
⎪ t0
⎪
⎪
⎩∗ q
Dt0 z = xy – 8z/3
In this system, there are two parameters, order q and control parameter c.
Appendix B: Matlab Source Programs of Chaos Characteristic Analysis 297
B.2 Programs about Characteristic Analysis of Integer-Order

Simplified Lorenz System
B.2.1 Simulation Codes for Matlab Toolbox with Runge–Kutta Algorithm
First, program the system function according to system (B.1):
function f=chao_SimpleLorenz(t,y)
global c
f=zeros(3,1);
f(1)=10*(y(2)-y(1));
f(2)=(24-4*c)*y(1)+c*y(2)-y(1)*y(3);
f(3)=y(1)*y(2)-8/3*y(3);
Running the following codes in Matlab window, the attractor phase diagram is
obtained as shown in Figure B.1.
clc
clear
global c
c=2;
[T,Y]=ode45(@chao_SimpleLorenz,0:0.01:500,[1 2 3 ]);
figure
plot(Y(30000:end,1),Y(30000:end,3))
xlim([-15 15])
xlabel(‘\itx’)
ylabel(‘\itz’)
B.2.2 Solution Based on Euler and Runge–Kutta Algorithm

In practical applications, for example, digital signal processing or Field Program-
mable Gate Array (FPGA) digital circuit implementation, we need to prepare the
35
30
25
20
y
15
10
0
‒15 ‒10 ‒5 0 5 10 15
x
Figure B.1: Attractor of the simplified Lorenz system (ode45).

298 Appendix B: Matlab Source Programs of Chaos Characteristic Analysis
relevant code to solve the numerical solution algorithm, and cannot use the Matlab
toolbox program. So solution algorithm based on the Euler algorithm and Runge–
Kutta method has a good application value.
1. Numerical solution program based on Euler algorithm:
function [x y z]=EulerSolve(c,x0,h,N)
% c is the system parameter; x0 is the ini-
tial value; h is the
iteration step size; N is the iteration number.
x=zeros(1,N);
y=zeros(1,N);
z=zeros(1,N);
x(1)=x0(1)+h*(10*(x0(2)-x0(1)));
y(1)=x0(2)+h*((24-4*c)*x0(1)-x0(1)*x0(3)+c*x0(2));
z(1)=x0(3)+h*(x0(1)*x0(2)-(8/3)*x0(3));
for i=2:N
x(i)=x(i-1)+h*(10*(y(i-1)-x(i-1)));
y(i)=y(i-1)+h*((24-4*c)*x(i-1)-x(i-1)*z(i-
1)+c*y(i-1));
z(i)=z(i-1)+h*(x(i-1)*y(i-1)-(8/3)*z(i-1));
end
[x y z]= EulerSolve (2,[1 2 3],0.01,20000);

plot(x(10000:end),z(10000:end))
xlabel(‘\itx’)
ylabel(‘\itz’)
2. Numerical solution program based on fourth-order Runge–Kutta algorithm:

function [x y z]=RungeSolve(c,xin,h,N)
% c is the system parameter; x0 is the ini-
tial value; h is the
iteration step size; N is the iteration number.
x=zeros(1,N);
y=zeros(1,N);
z=zeros(1,N);
x0=xin(1);y0=xin(2);z0=xin(3);
k11=10*(-x0+y0);
k21=(24-4*c)*x0-x0*z0+c*y0;
k31=x0*y0-(8/3)*z0;
k12=10*(-(x0+0.5*h*k11)+(y0+0.5*h*k21));
k22=(24-4*c)*(x0+0.5*h*k11)-(x0+0.5*h*k11)*(z0+0.5*h*k31)
+c*(y0+0.5*h*k21);
k32=(x0+0.5*h*k11)*(y0+0.5*h*k21)-(8/3)*(z0+0.5*h*k31);
k13=10*(-(x0+0.5*h*k12)+(y0+0.5*h*k22));
k23=(24-4*c)*(x0+0.5*h*k12)-(x0+0.5*h*k12)*(z0+0.5*h*k32)
+c*(y0+0.5*h*k22);
k33=(x0+0.5*h*k12)*(y0+0.5*h*k22)-(8/3)*(z0+0.5*h*k32);
k14=10*(-(x0+h*k13)+(y0+h*k23));
k24=(24-4*c)*(x0+h*k13)-(x0+h*k13)*(z0+h*k33)+c*(y0+h*k23);
k34=(x0+h*k13)*(y0+h*k23)-(8/3)*(z0+h*k33);
x0=x0+h*(k11+2*k12+2*k13+k14)/6;
y0=y0+h*(k21+2*k22+2*k23+k24)/6;
z0=z0+h*(k31+2*k32+2*k33+k34)/6;
x(1)=x0;y(1)=y0;z(1)=z0;
for i=2:N
k11=10*(-x0+y0);
k21=(24-4*c)*x0-x0*z0+c*y0;
k31=x0*y0-(8/3)*z0;
k12=10*(-(x0+0.5*h*k11)+(y0+0.5*h*k21));
k22=(24-4*c)*(x0+0.5*h*k11)-(x0+0.5*h*k11)
*(z0+0.5*h*k31)+c*(y0+0.5*h*k21);
k32=(x0+0.5*h*k11)*(y0+0.5*h*k21)-
(8/3)*(z0+0.5*h*k31);
k13=10*(-(x0+0.5*h*k12)+(y0+0.5*h*k22));
k23=(24-4*c)*(x0+0.5*h*k12)-(x0+0.5*h*k12)
*(z0+0.5*h*k32)+c*(y0+0.5*h*k22);
k33=(x0+0.5*h*k12)*(y0+0.5*h*k22)-
(8/3)*(z0+0.5*h*k32);
k14=10*(-(x0+h*k13)+(y0+h*k23));
k24=(24-4*c)*(x0+h*k13)-(x0+h*k13)*(z0+h*k33)
+c*(y0+h*k23);
k34=(x0+h*k13)*(y0+h*k23)-(8/3)*(z0+h*k33);
x0=x0+h*(k11+2*k12+2*k13+k14)/6;
y0=y0+h*(k21+2*k22+2*k23+k24)/6;
z0=z0+h*(k31+2*k32+2*k33+k34)/6;
x(i)=x0;y(i)=y0;z(i)=z0;
end
[x y z]=RungeSolve(2,[1 2 3],0.01,20000);
plot(x(10000:end),z(10000:end))
xlabel(‘\itx’)
ylabel(‘\itz’)
B.2.3 Program of Calculating LE

1. Program based on tool box:
The web site of the tool box to download is http://www.math.rsu.ru/mexmat/
kvm/matds/. Download the toolbox, as long as preparation of the following
functions according to the simplified Lorenz system.
function f=SimLorenz_ly(t,X)
global c
x=X(1); y=X(2); z=X(3);
Y= [X(4), X(7), X(10);
X(5), X(8), X(11);
X(6), X(9), X(12)];
f=zeros(12,1);
f(1)=10*(y-x);
f(2)=-x*z+(24-4*c)*x+c*y;
f(3)=x*y-8/3*z;
Jac=[-10,10,0;
24-4*c-z, c, -x;
y, x, -8/3];
f(4:12)=Jac*Y;
You can calculate the system’s LE spectrum by building a new .m file, and write
codes as follows:
clc
clear
global c
c=2;
[T,Res]=lyapunov(3,@SimLorenz_ly,@ode45,0,0.5,500,
[ 0.1 0 0.1],5);=Ly=Res(end,:)
The running result is
Ly =
0.8486 0.0014 -11.5161
2. Program of calculating LEs based on Euler and QR decomposition:

function ly=oulaJianhua(c)
h=0.01;
x0=1;y0=2;z0=3;
Q=[1 0 0;0 1 0; 0 0 1];
ly1=0;ly2=0;ly3=0;m=20000;
I=eye(3);
for i=1:m
x1=x0+h*(10*(y0-x0));
y1=y0+h*((24-4*c)*x0-x0*z0+c*y0);
z1=z0+h*(x0*y0-(8/3)*z0);
x0=x1;y0=y1;z0=z1;
J=I+h*[-10 10 0; (24-4*c)-z1 c -x1; y1 x1 -8/3];
B=J*Q;
[Q,R]=qr(B);
ly1=ly1+log(abs(R(1,1)));
end
ly=[ly1/(m*h) ly2/(m*h) ly3/(m*h)];
Running on Matlab
ly=oulaJianhua(c)
The result is
ly =
0.9996 -0.0013 -11.7466
3. Program of calculating LEs based on Runge–Kutta and QR decomposition:
function ly=RungeJianhua(c)
h=0.01;
x0=1;y0=1;z0=2;
Q=eye(3);
ly1=0;ly2=0;ly3=0;m=50000;
I=eye(3);
for i=1:m
k11=10*(-x0+y0);
k21=(24-4*c)*x0-x0*z0+c*y0;
k31=x0*y0-(8/3)*z0;
k12=10*(-(x0+0.5*h*k11)+(y0+0.5*h*k21));
k22=(24-4*c)*(x0+0.5*h*k11)-(x0+0.5*h*k11)
*(z0+0.5*h*k31)+c*(y0+0.5*h*k21);
k32=(x0+0.5*h*k11)*(y0+0.5*h*k21)-(8/3)
*(z0+0.5*h*k31);
k13=10*(-(x0+0.5*h*k12)+(y0+0.5*h*k22));
k23=(24-4*c)*(x0+0.5*h*k12)-(x0+0.5*h*k12)
*(z0+0.5*h*k32)+c*(y0+0.5*h*k22);
k33=(x0+0.5*h*k12)*(y0+0.5*h*k22)-
(8/3)*(z0+0.5*h*k32);
k14=10*(-(x0+h*k13)+(y0+h*k23));
k24=(24-4*c)*(x0+h*k13)-(x0+h*k13)*(z0+h*k33)
+c*(y0+h*k23);
k34=(x0+h*k13)*(y0+h*k23)-(8/3)*(z0+h*k33);
x0=x0+h*(k11+2*k12+2*k13+k14)/6;
y0=y0+h*(k21+2*k22+2*k23+k24)/6;
z0=z0+h*(k31+2*k32+2*k33+k34)/6;
J1=[-10 10 0;(24-4*c)-z0 c -x0;y0 x0 -8/3];
J2=[-10 10 0;(24-4*c)-(z0+0.5*h*k31) c -
(x0+0.5*h*k11);y0+0.5*h*k21x0+0.5*h*k11 -8/3];%
J3=[-10 10 0; (24-4*c)-(z0+0.5*h*k32) c -
(x0+0.5*h*k12); y0+0.5*h*k22x0+0.5*h*k12 -8/3];%
J4=[-10 10 0;(24-4*c)-(z0+h*k33) c -
(x0+h*k13);y0+h*k23 x0+h*k13-8/3];%
J=I+h*(J1+2*J2*(I+0.5*h*J1)+2*J3*(I+0.5*h*J2*(I+0.5*h*J1))
+J4*(I+h*J3*(I+0.5*h*J2*(I+0.5*h*J1))))/6;
B=J*Q;
[Q,R]=qr(B);
end
ly=[ly1/(m*h) ly2/(m*h) ly3/(m*h)];
Running in the Matlab window
ly=RungeJianhua(2)
The result is
ly=
0.8620 -0.0041 -11.5245
4. Program of calculating LEs with parameter varying:
Here, program is presented for the calculation of LEs of the simplified Lorenz sys-
tem with parameter c. LE diagram as shown in Figure B.2 is obtained by employing
Lyapunov.m:
clc
clear
–5
Lyapunov
–10
–15
–2 0 2 4 6 8
c
Figure B.2: Lyapunov exponents of the simplified Lorenz system.
global c
C=-2:0.05:8;
L=length(C);
ly=zeros(3,L);
for i=1:L
c=C(i);
[T,Res]=lyapunov(3,@SimLorenz_ly,@ode45,0,0.5,500,
[ 0.1 0 0.1],5);ly(:,i)=Res(end,:);
end
plot(C,ly)
xlabel(‘\itc’)
ylabel(‘Lyapunov’)
B.2.4 Program of Calculating the Maximum LE

The maximum LE of chaotic system has several calculation methods, and this program
is one of them. It must be used with f=chao_SimpleLorenz(t,y), and the specific code
is as follows:
function Ly=MaxLypunov(cc)
% simplified Lorenz system
global c;
c=cc;
T,Y]=ode45(@chao_SimpleLorenz,200:0.01:202,[0.1 0.1 0.1]);
lyap=0;
tao=2; % time interval
x0=Y(end,:);
y0=[x0(1)-3e-2,x0(2)-4e-2,x0(3)-5e-2];
d0=sqrt((y0(1)-x0(1))^2+(y0(2)-x0(2))^2+(y0(3)-x0(3))^2);
N=1000;
d=zeros(1,N);
for i=1:N
[T,Y]=ode45(@chao_SimpleLorenz,0:0.01:tao,x0);
x0=Y(end,:);
[T,Y]=ode45(@chao_SimpleLorenz,0:0.01:tao,y0);
y1=Y(end,:);
d(i)=sqrt((y1(1)-x0(1))^2+(y1(2)-x0(2))^2+(y1(3)-
x0(3))^2);
z=d0/d(i);
y0=x0+(y1-x0)*z;
if mod(i,40)==0
lyap1=lyap;
lyap=0;
for j=1:i
lyap=lyap+log(d(i)/d0);
end
lyap=lyap/(j*tao);
if abs(lyap-lyap1)<0.001
break;
end
end
end
Ly=0;
for j=1:i
Ly=Ly+log(d(i)/d0);
end
Ly=Ly/(j*tao);
Input in the Matlab window
Ly=MaxLypunov(2)
The result is
Ly =
0.6559
B.2.5 Program for Calculating Bifurcation Diagram
clc
clear
global c
C=-2:0.025:8;
L=length(C);
for i=1:L
c=C(i);
[T,Y]=ode45(@chao_SimpleLorenz,0:0.01:200,[1 2 3]);
data=Y(10000:end,3);
for j=2:(length(data)-1)
if data(j)>data(j-1)&&data(j)>data(j+1)
plot(c,data(j),‘.’,‘markersize’,1);
hold on;
if j==20
break;
end
end
end
end
xlabel(‘\itc’)
ylabel(‘\itzmax’)
Running the above codes, the bifurcation diagram of the simplified Lorenz system
can be obtained as shown in Figure B.3. In practice, the step size of parameter c is
appropriate to take a smaller one.
B.2.6 Program for Plotting Poincaré Section
clc
clear
global c
c=2;
[t,Y]=ode45(@chao_SimpleLorenz,0:0.01:500,[1 2 3]);
A=3; % z
B=2; % y
D=1; % x
m=22; %value of the section
50
40
30
zmax
20
10
0
‒2 0 2 4 6 8
c
Figure B.3: Bifurcation of the simplified Lorenz system.

N=length(Y);
for k=1:N-1
if (Y(k,A)>m&&Y(k+1,A)<m)||(Y(k,A)<m&&Y(k+1,A)>m)
%choose a plane on z axis
a=(Y(k+1,A)-Y(k,A))/(t(k+1)-t(k));
% the chosen plane
b=(Y(k,A)*t(k+1)-Y(k+1,A)*t(k))/(t(k+1)-t(k));
% t=0, Intercept of z line (coordinates)
c=(Y(k+1,B)-Y(k,B))/(t(k+1)-t(k));
% slope of y line
d=(Y(k,B)*t(k+1)-Y(k+1,B)*t(k))/(t(k+1)-t(k));
% t=0 Intercept of y line (coordinates)
e=(Y(k+1,D)-Y(k,D))/(t(k+1)-t(k));
% slope of x line
f=(Y(k,D)*t(k+1)-Y(k+1,D)*t(k))/(t(k+1)-t(k));
% t=0 Intercept of x line (coordinates)x
g=(m-b)/a; % z=10, the time needed
q=e*g+f; % z=10, the value of x
p=c*g+d; % z=10, the value of y plot(q,p);hold on;
end
end
xlabel(‘y’)
ylabel(‘x’)
Running the above codes, the Poincaré section of the simplified Lorenz system can be
20
10
0
x
–10
–20
–15 –10 –5 0 5 10 15
y
Figure B.4: Poincaré section of the simplified Lorenz system.

B.2.7 Program for 0–1 Test

Here is the program for plotting p–s diagram of 0–1 test:
clc
clear
global c
c=2;
[T,Y]=ode45(@chao_SimpleLorenz,0:0.01:1500,[1 2 3]);
xt=Y(10000:end,3);
N=length(xt)-5;
n=floor(N/15);
th=zeros(1,N+1);
p=th;
s=p;
o=p;
M=p;
for i=2:N+1
th(i)=c+th(i-1)+xt(i);%
end
for i=2:N+1
p(i)=p(i-1)+xt(i)*cos(th(i));
end
for i=2:N+1
s(i)=s(i-1)+xt(i)*sin(th(i));
end
plot(p,s)
xlabel(‘p’)
ylabel(‘s’)
Running the above codes, the p–s diagram of the simplified Lorenz system with 0–1
test can be obtained as shown in Figure B.5.
1000
–1000
–2000
S
–3000
–4000
–5000
–6000
–5000–4000–3000–2000–1000 0 1000 2000
p
Figure B.5: p–s diagram of the simplified Lorenz system with 0–1 test.
B.3 Solution Program for the Fraction-Order Simplified Lorenz

System
B.3.1 Solution Program Based on Predictor–Corrector Approach
function [x y z]=jianhua(c,q1,q2,q3)
if nargin==0
c=5;
q1=0.98;q2=0.98;q3=9.98;
end
h=0.01; % iteration step size
N=5000; % iteration number
x0=-6.7781;y0=-8.3059;z0=9.2133; % initial val-
ues of the system values;
M1=0;M2=0;M3=0;
x(N+1)=[0];y(N+1)=[0];z(N+1)=[0];
x1(N+1)=[0];y1(N+1)=[0];z1(N+1)=[0];
x1(1)=x0+h^q1*10*(y0-x0)/(gamma(q1)*q1);
y1(1)=y0+h^q2*((24-4*c)*x0-x0*z0+c*y0)/(gamma(q2)*q2);
z1(1)=z0+h^q3*(x0*y0-8*z0/3)/(gamma(q3)*q3);
x(1)=x0+10*h^q1*(y1(1)-x1(1)+q1*(y0-x0))/gamma(q1+2);
y(1)=y0+h^q2*((24-4*c)*x1(1)-x1(1)*z1(1)+c*y1(1)+q2*((24-
4*c)*x0-x0*z0+c*y0))/gamma(q2+2);
z(1)=z0+h^q3*(x1(1)*y1(1)-8*z1(1)/3+q3*(x0*y0-
8*z0/3))/gamma(q3+2);
for n=1:N
M1=(n^(q1+1)-(n-q1)*(n+1)^q1)*10*(y0-x0);
M2=(n^(q2+1)-(n-q2)*(n+1)^q2)*((24-4*c)
*x0-x0*z0+c*y0);
M3=(n^(q3+1)-(n-q3)*(n+1)^q3)*(x0*y0-8*z0/3);
N1=((n+1)^q1-n^q1)*10*(y0-x0);
N2=((n+1)^q2-n^q2)*((24-4*c)*x0-x0*z0+c*y0);
N3=((n+1)^q3-n^q3)*(x0*y0-8*z0/3);
for j=1:n
M1=M1+((n-j+2)^(q1+1)+(n-j)^(q1+1)-2*(n-j+1)^(q1+1))
*10*(y(j)-x(j));
M2=M2+((n-j+2)^(q2+1)+(n-j)^(q2+1)-2*(n-j+1)^(q2+1))
*((24-4*c)*x(j)-x(j)*z(j)+c*y(j));
M3=M3+((n-j+2)^(q3+1)+(n-j)^(q3+1)-2*(n-j+1)^(q3+1))
*(x(j)*y(j)-8*z(j)/3);
N1=N1+((n-j+1) q1-(n-j)^q1)*10*(y(j)-x(j));
N2=N2+((n-j+1)^q2-(n-j)^q2)*((24-4*c)*x(j)
-x(j)*z(j)+c*y(j));
N3=N3+((n-j+1) q3-(n-j)^q3)*(x(j)*y(j)-8*z(j)/3);
end
x1(n+1)=x0+h^q1*N1/(gamma(q1)*q1);
y1(n+1)=y0+h^q2*N2/(gamma(q2)*q2);
z1(n+1)=z0+h^q3*N3/(gamma(q3)*q3);
x(n+1)=x0+h^q1*(10*(y1(n+1)-x1(n+1))+M1)/gamma(q1+2);
y(n+1)=y0+h^q2*((24-4*c)*x1(n+1)-x1(n+1)*z1(n+1)
+c*y1(n+1)+M2)/gamma(q2+2);
z(n+1)=z0+h^q3*(x1(n+1)*y1(n+1)-8*z1(n+1)/3+M3)
/gamma(q3+2);
end
Running following codes in Matlab window:
[x y z]=jianhua(2,0.98,0.98,0.98);
plot(x,z)
xlabel(‘\itx’)
ylabel(‘\itz’)
the phase diagram of the fractional-order simplified Lorenz system is obtained as

shown in Figure B.6.
B.3.2 Solution Program Based on Adomian Decomposition Approach
function dy=SimpleFratral(x0,h,q,c)
dy=zeros(1,3);
b=8/3;
a=10;
if q<=1 && q>0
c10=x0(1);
c20=x0(2);
30
25
20
15
z
10
0
–15 –10 –5 0 5 10 15
x
Figure B.6: Phase diagram of the fractional-order simplified Lorenz system with q = 0.98, c = 2.
c30=x0(3);
end
if q>1&& q<2
c10=x0(1)+x0(1)*h;
c20=x0(2)+x0(2)*h;
c30=x0(3)+x0(3)*h;
end
c11=a*(c20-c10);
c21=(24-4*c)*c10+c*c20-c10*c30;
c31=-b*c30+c10*c20;
c12=a*(c21-c11);
c22=(24-4*c)*c11+c*c21-c11*c30-c10*c31;
c32=-b*c31+c11*c20+c10*c21;
c13=a*(c22-c12);
c23=(24-4*c)*c12+c*c22-c12*c30-c11*c31*(gamma(2*q+1)
/gamma(q+1)^2)-c10*c32;
c33=-b*c32+c12*c20+c11*c21*(gamma(2*q+1)/gamma(q+1)^2)+c10*c22;
c14=a*(c23-c13);
c24=(24-4*c)*c13+c*c23-c13*c30-(c12*c31+c11*c32)*(gamma(3*q+1)
/(gamma(q+1)*gamma(2*q+1)))-c10*c33;
c34=-b*c33+c13*c20+(c12*c21+c11*c22)*(gamma(3*q+1)/(gamma(q+1)
*gamma(2*q+1)))+c10*c23;
c15=a*(c24-c14);
c25=(24-4*c)*c14+c*c24-c14*c30-(c13*c31+c11*c33)*(gamma(4*q+1)
/(gamma(q+1)*gamma(3*q+1)))-c12*c32*(gamma(4*q+1)
/gamma(2*q+1)^2)-c10*c34;
c35=-b*c34+c14*c20+(c13*c21+c11*c23)*(gamma(4*q+1)
/(gamma(q+1)*gamma(3*q+1)))+c12*c22*(gamma(4*q+1)
/gamma(2*q+1)^2)+c10*c24;
c16=a*(c25-c15);
c26=(24-4*c)*c15+c*c25-c15*c30-(c14*c31+c11*c34)
*(gamma(5*q+1)/(gamma(q+1)*gamma(4*q+1)))
-(c13*c32+c12*c33)
*(gamma(5*q+1)/gamma(2*q+1)*gamma(3*q+1))-c10*c35;
c36=-b*c35+c15*c20+(c14*c21+c11*c24)*(gamma(5*q+1)
/(gamma(q+1)*gamma(4*q+1)))+(c13*c22+c12*c23)
*(gamma(5*q+1)/gamma(2*q+1)*gamma(3*q+1))+c10*c25;
dy(1)=c10+c11*(h^q/gamma(q+1))+c12*h^(2*q)/gamma(2*q+1)+c13
*h^(3*q)/gamma(3*q+1)+c14*h^(4*q)/gamma(4*q+1)+c15*h^(5*q)
/gamma(5*q+1)+c16*h^(6*q)/gamma(6*q+1);
dy(2)=c20+c21*(h^q/gamma(q+1))+c22*h^(2*q)/gamma(2*q+1)+c23
*h^(3*q)/gamma(3*q+1)+c24*h^(4*q)/gamma(4*q+1)+c25*h^(5*q)
/gamma(5*q+1)+c26*h^(6*q)/gamma(6*q+1);
dy(3)=c30+c31*(h^q/gamma(q+1))+c32*h^(2*q)/gamma(2*q+1)
+c33*h^(3*q)/gamma(3*q+1)+c34*h^(4*q)/gamma(4*q+1)+c35
*h^(5*q)/gamma(5*q+1)+c36*h^(6*q)/gamma(6*q+1);
function [t,y]=FratalSim(h,N,x0,q,c)
y=zeros(3,N);
t=zeros(1,N);
tt=0;
for i=1:N
t(i)=tt;
dy=SimpleFratral(x0,h,q,c);
y(:,i)=dy;
x0=dy;
tt=tt+h;
end
Running in Matlab window,
clc
clear
h=0.01;
c=5;
N=10000;
q=0.95;
x0=[0.1 0.2 0.3];
[t,y]=FratalSim(h,N,x0,q,c);
plot(y(1,:),y(3,:))
xlim([-12 12])
xlabel(‘\itx’)
ylabel(‘\itz’)
The attractor (similar to Figure B.6) and the chaotic sequences of the fractional-order
simplified Lorenz system are obtained.
B.3.3 Analysis Program of Complexity and Chaos Diagram

1. Program for calculating C0 complexity:
function CO=COFuZadu(x,r)
% function name: COFuZadu
% function: calculating C0 complexity
% input parameter x and r. r is the toler-
ance limit ; x is the chaotic sequence
% output parameters: C0.
Y=fft(x);
Gn=mean(abs(Y).^2);
YY=zeros(1,length(x));
for i=1:length(x)
if abs(Y(i))^2>r*Gn
YY(i)=Y(i);
end
end
xx=ifft(YY);
Ssum=0;
for i=1:length(x)
Ssum=Ssum+(xx(i)-x(i))^2;
end
CO=Ssum/sum(x.^2);
2. Program for calculating spectral entropy (SE) complexity:
function SE=SEShannon(x)
%function name: SEShannon
%function: calculating the SE complex-
ity of chaotic sequence
%input parameter x. x is the chaotic sequence
%output parameter: SE.
N=length(x);
flag=0;
x=x-mean(x);
for i=1:N
if x(i)~=0
flag=1;
end
end
if flag==0
SE=0;
return;
end
Y=fft(x);
Xk=(abs(Y).^ 2)./N;
Xk=Xk(1:(floor(N/2)));
ptot=sum(Xk);
PK=Xk./ptot;
P=0;
for i=1:N/2
if PK(i)~=0
P=P-PK(i)*log(PK(i));
end
end
se2=sum(P);
SE=se2/log(N/2);
3. Program for complexity with order q varying
clc
clear
L=201;
c=5;
Q=linspace(0.5,1,L);
C0=zeros(1,L);
SE=zeros(1,L);
for i=1:L
q=Q(i);
[t,y]=FratalSim(0.01,11000,[1 2 3],q,c);
temp=y(1,1001:end);
C0(i)=COFuZadu(temp,15);
SE(i)=SEShannon(temp);
disp(i)
end
figure
plot(Q,SE)
xlabel(‘q’)
ylabel(‘SE’)
figure
plot(Q,C0)
xlabel(‘q’)
ylabel(‘C0’)
Running the above codes, the complexity of the fractional-order simplified Lorenz
system with varying q can be obtained as shown in Figure B.7.
0.8 0.7
(a) (b)
0.6
0.6
0.5
0.4
SE
0.4
C0
0.3
0.2
0.2
0.1
0 0
0.5 0.6 0.7 0.8 0.9 1 0.5 0.6 0.7 0.8 0.9 1
q q
Figure B.7: Complexity of the fractional-order simplified Lorenz system with varying q(c = 2):
(a) SE complexity and (b) C0 complexity.
4. Program for complexity with order c varying:

clc
clear
L=201;
q=0.98;
C=linspace(-2,8,L);
C0=zeros(1,L);
SE=zeros(1,L);
for i=1:L
c=C(i);
[t,y]=FratalSim(0.01,11000,[1 2 3],q,c);
temp=y(1,1001:end);
C0(i)=COFuZadu(temp,15);
SE(i)=SEShannon(temp);
disp(i)
end
figure
plot(C,SE)
xlabel(‘c’)
ylabel(‘SE’)
figure
plot(C,C0)
xlabel(‘c’)
ylabel(‘C0’)
Running the above codes, the complexity of the fractional-order simplified Lorenz
system with c varying can be obtained as shown in Figure B.8.
0.6 0.2
(a) (b)
0.15
0.4
0.1
C0
SE
0.2
0.05
0 0
–2 0 2 4 6 8 –2 0 2 4 6 8
c c
Figure B.8: Complexity of the fractional-order simplified Lorenz system with c varying (q= 0.98):
(a) SE complexity and (b) C0 complexity.
5. Program for complexity-based chaos diagram:

clc
clear
L=11;
Q=linspace(0.65,1,L);
C=linspace(-2,8,L);
C0=zeros(L,L);
SE=zeros(L,L);
for i=1:L
q=Q(i);
for j=1:L
c=C(j);
[t,y]=FratalSim(0.01,12000,[1 2 3],q,c);
temp=y(1,2001:end);
C0(i,j)=COFuZadu(temp,15);
SE(i,j)=SEShannon (temp);
end
disp(i)
end
[X,Y]=meshgrid(C,Q);
figure
colormap(flipud(hot));
contourf(X,Y,SE);
xlabel(‘c’)
ylabel(‘q’)
figure
colormap(flipud(hot));
contourf(X,Y,C0);
(a) (b)
1 0.6 1 0.6
0.95 0.5 0.95 0.5

0.9 0.9
0.4 0.4
0.85 0.85
0.3 0.3
q
q
0.8 0.8
0.2 0.2
0.75 0.75
0.7 0.1 0.7 0.1

(a) (b)
0.65 0.0 0.65 0.0
–2 0 2 4 6 8 –2 0 2 4 6 8
c c
Figure B.9: Chaos diagram of the fractional-order simplified Lorenz system on q–c plane:
(a) based on SE complexity and (b) based on C0 complexity.
xlabel(‘c’)
ylabel(‘q’)
Running the above codes, the complexity-based chaos diagram on q–c plane of the
fractional-order simplified Lorenz system can be obtained as shown in Figure B.9.
It is worth pointing out that the greater the value of L, the more the mesh of the
parameters plane, and the more details can be reflected. But the code running time is
relatively long, and it needs to wait patiently. We suggest to choose L = 101 at least.
At the same time, the length of the sequence is also important when we calculate the
complexity. Because both the SE algorithm and C0 algorithm are based on fast Fourier
transformation, the length of the sequence is suggested to choose N = 104. For test,
we choose L = 51 and N = 104.
B.4 Simulation of Chaotic Synchronization

B.4.1 Drive–Response Synchronization
⎧
⎪
⎨ẋ = 10 (y – x)
⎪
Drive system: ẏ = (24 – 4c) x – xz + cy , (B.3)
⎪
⎪
⎩ż = xy – 8z/3
⎧
⎪
⎨ẋ1 = 10 (y1 – x1 )
⎪
Response system: ẏ1 = (24 – 4c) x – xz1 + cy1 , (B.4)
⎪
⎪
⎩ż = xy – 8z /3
1 1 1
⎧
⎪
⎨ė1 = ẋ1 – ẋ
⎪
Error system: ė2 = ẏ1 – ẏ . (B.5)
⎪
⎪
⎩ė = ż – ż
3 1
It will be very convenient to solve this problem by using the Matlab Ode45 function, if
the drive system and the response system are considered as a big system. Of course,
readers can also solve the drive system and input the drive system sequence into the
response system, but the program may be relatively complex:
function dy = SimplePC(t,y)
%Function name: SimplePC
%function: the drive-response synchronization of the simpli-
fied Lorenz system
global c;
dy = zeros(6,1);
dy(1)=10*(y(2)-y(1));
dy(2)=(24-4*c)*y(4)-y(4)*y(3)+c*y(2);
dy(3)=y(4)*y(2)-8*y(3)/3;
dy(4)=10*(y(5)-y(4));
dy(5)=(24-4*c)*y(4)-y(4)*y(6)+c*y(5);
dy(6)=y(4)*y(5)-8*y(6)/3;
Setting c = –1, initial values of the drive system are [5, 8, –10] and the initial values
of the response system are [5, –5, 20]. Running the below Matlab codes, the syn-
chronization error curves of the synchronization system are obtained as shown in
Figure B.10:
clc
clear
global c
c=-1;
[T,Y]=ode45(@SimplePC,0:0.01:15,[5, -5, 20, 15, 8, -10 ]);
figure
plot(T,Y(:,1)-Y(:,4))
xlabel(’t’)
ylabel(’e1’)
figure
plot(T,Y(:,2)-Y(:,5))
xlabel(‘t’)
ylabel(‘e2’)
figure
plot(T,Y(:,3)-Y(:,6))
xlabel(‘t’)
ylabel(‘e3’)
15 30 30
(a) (b) (c)
10 20 20
5
10 10
0
e3
0
e2
e1
0
–5
–10 –10
–10
–15 –20 –20
–20 –30 –30
0 5 10 15 0 5 10 15 0 5 10 15
t t t
Figure B.10: Synchronization error curves of the simplified Lorenz system: (a) error curve of
sequence x; (b) error curve of sequence y; and (c) error curve of sequence z.
B.4.2 Coupling Synchronization
⎧
⎪
⎨ẋ = 10 (y – x)
⎪
Drive system: ẏ = (24 – 4c) x – xz + cy , (B.6)
⎪
⎪
⎩ż = xy – 8z/3
⎧
⎪
⎨ẋ1 = 10 (y1 – x1 ) + k1 (x – x1 )
⎪
Response system: ẏ1 = (24 – 4c) x1 – x1 z1 + cy1 + k2 (y – y1 ) , (B.7)
⎪
⎪
⎩ż = x y – 8z /3 + k (z – z )
1 1 1 1 3 1
⎧
⎪
⎨ė1 = ẋ1 – ẋ
⎪
Error system: ė2 = ẏ1 – ẏ . (B.8)
⎪
⎪
⎩ė = ż – ż
3 1
function dy = SimpleOH(t,y)
%function name: SimpleOH %function: the coupling synchroniza-
tion of the simplified Lorenz system
global c;
global k1;
global k2;
global k3;
dy = zeros(6,1);
dy(1)=10*(y(2)-y(1))+k1*(y(4)-y(1));
dy(2)=(24-4*c)*y(1)-y(1)*y(3)+c*y(2)+k2*(y(5)-y(2));
dy(3)=y(1)*y(2)-8*y(3)/3+k3*(y(6)-y(3));
dy(4)=10*(y(5)-y(4));
dy(5)=(24-4*c)*y(4)-y(4)*y(6)+c*y(5);
dy(6)=y(4)*y(5)-8*y(6)/3;
3 8 4
(a) (b) (c)
2 6 2
1 4 0
0 2 –2
e3
e2
e1
–1 0 –4
–2 –2 –6
–3 –4 –8
–4 –6 –10
0 5 10 15 0 5 10 15 0 5 10 15
t t t
Figure B.11: Coupling synchronization error curves of the simplified Lorenz system: (a) error curve of
sequence x; (b) error curve of sequence y; and
(c) error curve of sequence z.
Setting c = 2, k1 = 4, k2 = 2, and k3 = 3, the initial values of the drive system are [4, 5, 6]
and the initial values of the coupling system are [1, 2, 3]. Running the Matlab codes
below, the synchronization error curves of the synchronization system are obtained
as shown in Figure B.11:
clc
clear
global c
global k1
global k2
global k3
c=2;
k1=4;k2=2;k3=3;
[T,Y]=ode45(@SimpleOH,0:0.01:15,[1 2 3 4 5 6 ]);
figure
plot(T,Y(:,1)-Y(:,4))
xlabel(‘t’)
ylabel(‘e1’)
figure
plot(T,Y(:,2)-Y(:,5))
xlabel(‘t’)
ylabel(‘e2’)
figure
plot(T,Y(:,3)-Y(:,6))
xlabel(‘t’)
ylabel(‘e3’)
B.4.3 Tracking Synchronization
⎧
⎪
⎨ẋ = 10 (y – x) + u1
⎪
ẏ = (24 – 4c) x – xz + cy + u2 . (B.9)
⎪
⎪
⎩ż = xy – 8z/3 + u
3
Design the controller:

⎧
⎪
⎨u1 = ṙ1 + r1 – 10y + 9x
⎪
u2 = ṙ2 + r2 – (24 – 4c) x + xz – (c + 1)y , (B.10)
⎪
⎪
⎩u = ṙ + r – xy + 5z/3
3 3 3
where r1 , r2 , and r3 are signals to be tracked.

Next, we let the simplified Lorenz system to track the sin curve, the constant
curve, and the x sequence of the hyperchaotic Lorenz system, respectively,
⎧ ⎧
⎪ r t ⎪
⎪
⎨ 1 = sin ⎨ṙ1 = cos t
⎪
r2 = 4 and ṙ2 = 0 , (B.11)
⎪
⎪ ⎪
⎪
⎩r = x ⎩ṙ = 10(y – x )
3 1 3 1 1
⎧
⎪
⎪ ẋ1 = 10 (y1 – x1 )
⎪
⎪
⎪
⎨ẏ = 28x – x z + y – w
1 1 1 1 1 1
, (B.12)
⎪
⎪ ż = x y – 8z /3
⎪
⎪
1 1 1 1
⎪
⎩ẇ = ky z
1 1 1
clc; clear
L=10000;
c=2;h=0.001;k=0.21;
y0=[1 2 1];
yy0=[5 6 7 8];
data=zeros(3,L);
T=zeros(1,L);
R1=zeros(1,L);
R2=zeros(1,L);
R3=zeros(1,L);
t=0;
for i=1:L
t=t+h;
yy0(1)=yy0(1)+h*(10*(yy0(2)-yy0(1)));
yy0(2)=yy0(2)+h*(-yy0(1)*yy0(3)+28*yy0(1)+yy0(2)-yy0(4));
yy0(3)=yy0(3)+h*(yy0(1)*yy0(2)-8/3*yy0(3));
yy0(4)=yy0(4)+h*(k*yy0(2)*yy0(3));
r1=sin(t);dr1=cos(t);
r2=4;dr2=0;
r3=yy0(1);dr3=10*(yy0(2)-yy0(1));
u1=dr1+r1-10*y0(2)+9*y0(1);
u2=dr2+r2-(24-4*c)*y0(1)+y0(1)*y0(3)-(c+1)*y0(2);
u3=dr3+r3-y0(1)*y0(2)+5*y0(3)/3;
y0(1)=y0(1)+h*(10*(y0(2)-y0(1))+u1);
y0(2)=y0(2)+h*((24-4*c)*y0(1)-y0(1)*y0(3)+c*y0(2)+u2);
y0(3)=y0(3)+h*(y0(1)*y0(2)-8*y0(3)/3+u3);
data(1,i)=y0(1);data(2,i)=y0(2);data(3,i)=y0(3);
T(i)=t;
R1(i)=r1;
R2(i)=r2;
R3(i)=r3;
end
figure
plot(T,R1,‘-.’)
hold on
plot(T,data(1,:),‘--’)
hold off
legend(‘r1’,‘x’)
xlabel(‘t’)
ylabel(‘r1,x’)
figure
plot(T,R2,‘-.’)
hold on
hold off
xlabel(‘t’)
ylabel(‘r2,x’)
figure
plot(T,R3,‘-.’)
hold on
hold off
xlabel(‘t’)
1.5 4.5 20
r1 (a) (b) r3 (c)
1 x 4
10
0.5 3.5 r2
r2,y
r1,x
r2,z
y 0
0 3
–0.5 2.5 –10

z
–1 2 –20
0 2 4 6 8 10 0 2 4 6 8 10 0 2 4 6 8 10
t t t
Figure B.12: Tracking curves of the simplified Lorenz system: (a) sequence x; (b) sequence y; and
(c) sequence z.
ylabel(‘r3,x’)
Running the above Matlab codes, the tracking curves of the simplified Lorenz system
are obtained as shown in Figure B.12. Of course, readers can also plot the tracking
error curves.
References
[1] Ford J. Chaos: Solving the unsolvable, predicting the unpredictable, in chaotic dynamic and
fractal. New York, Academic Press, 1986.
[2] Lorenz EN. Deterministic nonperiodic flows. J Atmos Sci 1963, 20, 130–41.
[3] Li TY, Yorke JA. Period three implies chaos. Am Math Mon 1975, 82(10), 985–92.
[4] May R. Simple mathematical models with very complicated dynamics. Nature 1976, 261,
459–67.
[5] Feigenbaum MJ. Quantitative universality for a class of non-linear transformations. J Stat Phys
1978, 19(1), 25–52.
[6] Feigenbaum MJ. The universal metric properties of non-linear transformations. J Stat Phys
1979, 21(6), 669–706.
[7] Takens F. “Detecting strange attractors in turbulence” in Dynamical systems and turhulence,
Warwick, 1980, Lecture Notes in Mathematics, vol. 898. D. A. Rand and L. S. Young, Eds.
Berlin, Springer, 1981, 366–81.
[8] Glass L, Guevara MR, Shrier A, Perez R. Bifurcation and chaos in a periodically stimulated
cardiac oscillator. Physica D 1983, 7, 89–101.
[9] Grassberger P. Toward a quantitative theory of self-generated complexity. Int J Theor Phys
1986, 25(9), 907–38.
[10] Pecora LM, Carroll TL. Synchronization in chaotic systems. Phys Rev Lett 1990, 64(8), 821–4.
[11] Pecora LM, Carroll TL. Driving systems with chaotic signals. Phys Rev A 1991, 44(4), 2374–83.
[12] Ott E, Grebogi C, Yorke JA. Controlling chaos. Phys Rev Lett 1990, 64(11), 1196–9.
[13] Ding J. Puzzles of the wise men-a brief history of chaos and fractals. Beijing, Higher Education
Press, 2013.
[14] Liang ML, Wang ZK. Talks about chaos and balance. Dalian, Dalian University of Technology
Press, 2008.
[15] Matsumoto T, Chua LO, Tanaka S. Simplest chaotic nonautonomous circuit. Phys Rev A 1984,
30, 1155–7.
[16] Chen GR, Ueta T. Yet another chaotic attractor. Int J Bifurcation Chaos 1999, 9(7), 1465–6.
[17] Wu CW, Chua LO. A unified framework for synchronization and control of dynamical systems.
Int J Bifurcation Chaos 1994, 4(4), 979–98.
[18] Fang JQ. Controlling chaos and developing high and new technology. Beijing, Atomic Energy
Press, 2002.
[19] Cuomo KM, Oppenheim AV, Strogatz SH. Synchronization of Lorenz-based chaotic circuits with
applications to communications. IEEE Trans Circuits Syst II 1993, 40(10), 626–33.
[20] Carroll TL, Pecora LM. Cascading synchronized chaotic systems. Physica D 1993, 67(1),
126–40.
[21] Kocarev L, Parlitz U. General synchronization with application to communication. Phys Rev Lett
1995, 74(25), 5028–31.
[22] Kapitaniak T, Chua LO. Hyperchaotic attractors of unidirectionally-coupled Chua’s circuits. Int J
Bifurcation Chaos 1994, 4(2), 477–82.
[23] Pyragas K, Tamasevicius A. Experimental control of chaos by delayed self-controlling
feedback. Phys Lett A 1993, 180(1–2), 99–102.
[24] Huberman BA, Lumer E. Dynamics of adaptive systems. IEEE Trans CAS-I 1990, 37(4), 547–50.
[25] John JK, Amritkar RE. Synchronization of unstable orbits using adaptive control. Phys Rev E
1994, 49(6), 4843–8.
[26] Yang T, Chua LO. Impulsive stabilization for control and synchronization of chaotic systems:
Theory and application to secure communication. IEEE Trans CAS-I 1997, 44(10), 976–88.
[27] Yang T, Chua LO. Impulsive control and synchronization of nonlinear dynamical systems and
application to secure communication. Int J Bifurcation Chaos 1997, 7(3), 645–64.
324 References
[28] Mainieri R, Rehacek J. Projective synchronization in three-dimensional chaotic systems. Phys

Rev Lett 1999, 82(15), 3042–5.
[29] Du HY, Zeng QS, Wang CH. Function projective synchronization of different chaotic systems
with uncertain parameters. Phys Lett A 2008, 372, 5402–10.
[30] Park JH. Adaptive control for modified projective synchronization of a four-dimensional chaotic
system with uncertain parameters. J Comput Appl Math 2008, 213, 288–93.
[31] Angeli AD, Genesio R, Tesi A. Dead-beat chaos synchronization in discrete time system. IEEE
Trans CAS-I 1995, 42(1), 54–6.
[32] Vasegh N, Khellat F. Takagi-Sugeno fuzzy modeling and chaos control of partial differential
systems. Chaos 2013, 23(4), 042101.
[33] Shieh CS, Hung RT. Hybrid control for synchronizing a chaotic system. Appl Math Model 2011,
35(8), 3751–8.
[34] Sun KH, Mou J, Zhou JL. Synchronization control and its application for the unified chaotic
system based on chaos observer. IET Control Theory Appl 2008, 25(4), 794–8.
[35] Sun KH, Zhang TS. Synchronization control method of hyperchaotic systems by multi variables
driving and error feedback. J Cent South Univ (Science and Technology) 2005, 36(4), 653–7.
[36] Kennedy M, Kolumbon G. Special issue on noncoherent chaotic communication. IEEE Trans
CAS-I 2000, 47(12), 1661–732.
[37] Kocarev L, Maggio G, Ogorzalerk M. Special issue on applications of chaos in modern
communication systems. IEEE Trans CAS-I 2001, 48(12), 1385–527.
[38] Hasler M, Mazzini G. Special issue on applications of nonlinear dynamics to electronic and
information engineering. Proc IEEE 2002, 90(5), 631–40.
[39] Wang XY. Synchronization of chaotic system and its application in secure communication.
Beijing, Science Press, 2012.
[40] Yin Y. Experimental demonstration of chaotic synchronization in the modified Chua’s
Oscillators. Int J Bifurcation Chaos 1997, 7(6), 1401–10.
[41] Dedieu HD, Kennedy MP, Hasler M. Chaos shift-keying: Modulation and demodulation of a
chaotic carrier using self-synchronizing Chua’s circuits. IEEE Trans CAS-II 1993, 40(10),
634–42.
[42] Yang T, Chua LO. Secure communication via chaotic parameter modulation. IEEE Trans CAS-I
1996, 43(9), 817–9.
[43] Batini GH, McGillem CD. A chaotic direct sequence spread spectrum communication system.
IEEE Trans Commun 1994, 42(2, 3, 4), 1524–7.
[44] Itoh M. Spread spectrum communication via chaos. Int J Bifurcation Chaos 1999, 9(1), 155–213.
[45] Cuomo KM, Oppenheim AV. Circuit implementation of synchronized chaos with application to
communications. Phys Rev Lett 1993, 71, 65–8.
[46] Kennedy MP, Kolumban G. Digital communications using chaos. CRC PRESS’99, 1–24.
[47] Kolumban G, Kis G, Jako Z, Kennedy MP. FM-DCSK: A robust modulation scheme for chaotic
communications. IEICE Trans Fundam 1998, E81–A, 1798–1802.
[48] Xu WK, Wang L, Kolumban G. A novel differential chaos shift keying modulation scheme. Int J
[49] Mathews R. On the derivation of a chaotic encryption algorithm. Cryptrologia 1989, 29–42.
[50] Baptista MS. Cryptography with chaos. Phys Lett A 1998, 240, 50–4.
[51] Wong WK. A fast chaotic cryptographic with dynamic look-up table. Phys Lett A 2002, 298(4),
238–42.
[52] Wong WK, Ho SW, Yung CK. A chaotic cryptography scheme for generating short ciphertext.
Phys Lett A 2003, 310, 67–73.
[53] Akgul A, Calgan H, Koyuncu I, Pehlivan I, Istanbullu A. Chaos-based engineering applications
with a 3D chaotic system without equilibrium points. Nonlinear Dynamics 2016, 84(2):
481–495.
References 325
[54] Utami D, Suwastio H, Sumadjudin BB. FPGA implementation of digital chaotic cryptography.
Lect Notes Comput Sci 2002, 25(10), 239–47.
[55] Sun KH, Sprott JC. Periodically forced chaotic system with signum nonlinearity. Int J Bifurcation
Chaos 2010, 20(5), 1499–507.
[56] Sun KH, Sprott JC. Dynamics of a simplified Lorenz system. Int J Bifurcation Chaos 2009, 19(4),
1357–66.
[57] Gottwald GA, Melbourne I. A new test for chaos in deterministic systems. Proc R Soc Med
2004, 460, 603–11.
[58] Gottwald GA, Melbourne I. Testing for chaos in deterministic systems with noise. Physica D
2005, 212, 100–10.
[59] Sun KH, Liu X, Zhu CX. The 0–1 test algorithm for chaos and its applications. Chin Phys B 2010,
19(11), 110510.
[60] Horgan J. From complexity to perplexity. Sci Am 1995, 272(6), 104–9.
[61] Lempel A, Ziv J. On the complexity of finite sequence. IEEE Trans 1976, IT-22, 75–9.
[62] Pincus SM. Approximate entropy (ApEn) as a complexity measure. Chaos 1995, 5(1), 110–17.
[63] Sun KH, Tan GQ, Sheng LY. Complexity analysis of TD-ERCS discrete chaotic pseudo-random
sequences. Acta Phys Sin 2008, 57(6), 3359–68.
[64] Chen WT, Zhuang J, Yu WX, Wang ZZ. Measuring complexity using FuzzyEn, ApEn, and SampEn.
Med Eng Phys 2009, 31(1), 61–8.
[65] Sun KH, He SB, Yin LZ, A Di-Li⋅Duo Li-Kun. Application of FuzzyEn algorithm to the analysis of
complexity of chaotic sequence. Acta Phys Sin 2012, 61(13), 130507.
[66] Larrondo HA, González CM, Martin MT, Plastino A, Rosso OA. Intensive statistical complexity
measure of pseudorandom number generators. Phys A Stat Mech Appl 2005, 356, 133–8.
[67] Sun KH, He SB, Sheng LY. Complexity analysis of chaotic sequence based on the intensive
statistical complexity algorithm. Acta Phys Sin 2011, 60(2), 020505
[68] Rajeev KA, Subba JR, Ramakrishna Ramaswamy. Information-entropic analysis of chaotic time
series: Determination of time-delays and dynamical coupling. Chaos Solitons Fractals 2002,
14(4), 633–41.
[69] Malihe S, Serajeddin K, Reza B. Entropy and complexity measures for EEG signal classification
of schizophrenic and control participants. Artif Intell Med 2009, 47(3), 263–74.
[70] Abdulnasir Y, Mehmet A, Mustafa P. Application of adaptive neuro-fuzzy inference system for
vigilance level estimation by using wavelet-entropy feature extraction. Expert Syst Appl 2009,
36(4), 7390–9.
[71] Chen F, Xu JH, Gu FJ. Dynamic process of information transmission complexity in human
brains. Biol Cybern 2000, 83 (4), 355–66.
[72] Shen EH, Cai ZJ, Gu FJ. Mathematical foundation of a new complexity measure. Appl Math
Mech 2005, 26(9), 1188–96.
[73] Cai ZJ, Sun J. Modified C0 complexity and applications. Journal of Fudan University (Natural
Science), 2008, 47(6), 791–96
[74] Sun KH, He SB, He Y, Yin LZ. Complexity analysis of chaotic pseudo-random sequences based
on spectral entropy algorithm. Acta Phys Sin 2013, 62(1), 010501.
[75] Sun KH, He SB, Zhu CX, He Y. Analysis of chaotic complexity characteristics based on C0
algorithm. Acta Electronica Sin 2013, 41(9), 1765–71
[76] Hénon M. A two-dimensional mapping with a strange attractor. Commun Math Phys 1976, 50,
291–312.
[77] Sheng LY, Sun KH, Li CB. Study of a discrete chaotic system based on tangent-delay for elliptic
reflecting cavity and its properties. Acta Phys Sin 2004, 53(9), 2871–6.
[78] Duffing G. Erzwungene Schwingungen bei Veränderlicher Eigenfrequenz. Braunschweig,
Vieweg, 1918.
326 References
[79] van der Pol B. On relaxation-oscillations. The London, Edinburgh, and Dublin Philosophical
Magazine and Journal of Sciences Series 1926, 7(2), 978–92.
[80] Rössler OE. An equation for continuous chaos. Phys Lett A 1976, 57(5), 397–8.
[81] Zhong GQ, Tang KS. Circuitry implementation and synchronization of Chen’s attractor. Int J
[82] Lü JH, Chen GR. A new chaotic attractor coined. Int J Bifurcation Chaos 2002, 12(3), 659–61.
[83] Vanĕček A, Čelikovský S. Control systems: From linear analysis to synthesis of chaos. London,
Pretice-Hall, 1996.
[84] Lü JH, Chen GR, Cheng DZ, Čelikovský S. Bridge the gap between the Lorenz system and the
Chen system. Int J Bifurcation Chaos 2002, 12(12), 2917–26.
[85] Lü JH, Zhou TS, Chen GR, Zhang SC. The compound structure of Chen’s attractor. Int J
[86] Zhang ZQ, Park J H; Shao HY. Adaptive synchronization of uncertain unified chaotic systems via
novel feedback controls. Nonlinear Dyn 2015, 81(1–2), 695–706.
[87] Cheng CJ. Robust synchronization of uncertain unified chaotic systems subject to noise and its
application to secure communication. Appl Math Comput 2012, 219(5), 2698–712.
[88] Sun KH, Yang JL, Ding JF, Sheng LY. Circuit design and implementation of Lorenz chaotic system
with one parameter. Acta Phys Sin 2010, 59(12), 8385–92.
[89] Sprott JC. A proposed standard for the publication of new chaotic systems. Int J Bifurcation
Chaos 2011, 21(9), 2391–4.
[90] Saltzman B. Finite amplitude free connection as an initial value problem-I. J Atmos Sci 1962 19,
329–41.
[91] Rössler OE. An equation for hyperchaos. Phys Lett A 1979, 71(2–3), 155–7.
[92] Li Y, Chen G, Tang KS. Generating hyperchaos via state feedback control. Int J Bifurcation
Chaos 2005, 15(10), 3367–75.
[93] Zgliczynski P. Symbolic dynamics for the Rössler folded towel map. Banach Center Publ 1999,
47(1), 253–8.
[94] Carroll TL, Pecora LM. Synchronization chaotic circuits. IEEE Trans CAS-I 1991, 38(4), 453–6.
[95] Fang JQ. Control and synchronization of chaos in nonlinear systems and its application
prospect (I). Prog Phys 1996, 16(1), 1–74.
[96] Fang JQ. Control and synchronization of chaos in nonlinear systems and its application
prospect (II). Prog Phys 1996, 16(2), 137–201.
[97] Sun KH. Chaos synchronization control theory and its application in information encryption.
Central South University, 2005.
[98] He R, Vaidya PG. Analysis and synthesis of synchronous periodic and chaotic systems. Phys
Rev A 1992, 46(12), 7387–92.
[99] Zhou CS, Lai CH. Analysis of spurious synchronization with positive conditional Lyapunov
exponents in computer simulations. Physica D 2000, 135(1–2), 1–23.
[100] Sun KH, Shang F, Zhong K, Sheng Z. Synchronous control between unified chaotic system and
its modified system based on intermittent feedback. J Electron Inf Technol 2009, 31(1), 71–4.
[101] Ma J, Liao GH, Mo XH. Hyperchaos synchronization and control using intermittent feedback.
Acta Phys Sin 2005, 54(12), 5585–90.
[102] Guan XP, Fang ZP, Chen CL, Hua CC. Chaos control and its application in secure
communication. Beijing, National Defence Industry Press, 2002.
[103] Zhou P. Observers for a class of 3D continuous chaotic systems. Acta Phys Sin 2003, 52(5),
1108–11.
[104] Yang XS. On observability of 3D continuous-time autonomous chaotic systems based on scalar
output measurement. Int J Bifurcation Chaos 2002, 12(5), 1159–62.
[105] Xu DL, Li ZG. Controlled projective synchronization in nonpartially-linear chaotic systems. Int J
References 327
[106] Li GH. Modified projective synchronization of chaotic system. Chaos Solitons & Fractals 2007,
32 (5), 1786–90.
[107] Cafagna D, Grassi G. Observer-based projective synchronization of fractional systems via a
scalar signal: Application to hyperchaotic Rössler systems. Nonlinear Dyn 2012, 68(1–2),
117–28.
[108] Farivar F, Nekoui MA, Shoorehdeli MA, Teshnehlab M. Modified projective synchronization of
chaotic dissipative gyroscope systems via backstepping control. Indian J Phys 2012, 86(10),
901–6.
[109] Wang SB, Wang XY, Zhou YF. A memristor-based complex Lorenz system and its modified
projective synchronization. Entropy 2015, 17(11), 7628–44.
[110] Kim DW, Chang PH, Kim SH. A new chaotic attractor and its robust function projective
synchronization. Nonlinear Dyn 2013, 73(3), 1883–93.
[111] Liu J, Liu ST, Sprott JC. Adaptive complex modified hybrid function projective synchronization
of different dimensional complex chaos with uncertain complex parameters. Nonlinear Dyn
2016, 83(1–2), 1109–21.
[112] Santo B. Chaos synchronization and cryptography for secure communications. New York,
Hershey, 2010.
[113] Cuomo KM, Oppengeim AV, Strogatz SH. Synchronization of Lorenz-based chaotic circuits with
applications to communications. IEEE Trans CAS-II 1993, 140(10), 626–33.
[114] Mianovic V, Zaghloul ME. Improved masking algorithm for chaotic communication systems.
Electron Lett 1996, 32(1), 11–12.
[115] Carroll TC, Heagy JF, Pecora LM. Transforming signals with chaotic synchronization. Phys Rev E
1996, 54(5), 4676–80.
[116] Parlitz U, Chua LO, Kocarev L, Halle KS, Shang A. Transmission of digital signals by chaotic
synchronization. Int J Bifurcation Chaos 1992, 2(2), 973–7.
[117] Kolumban G, Kennedy MP, Jako Z, Kis G. Chaotic communications with correlator receivers:
Theory and performance limits. Proc IEEE 2002, 90(5), 711–32.
[118] Kolnmban G, Kis G, Kennedy MP, Já Z. FM-DCSK: A new and robust solution to chaos
communications. In: Proceedings of International Symposium Nonlinear Theory and Its
Applications, Hawaii, 1997, 117–20.
[119] Galias Z, Maggio GM. Quadrature chaos-shift keying: Theory and performance analysis. IEEE
Trans CAS-I 2001, 48(12), 1510–19.
[120] Kennedy MP, Kolumbán G. Digital communications using chaos. Signal Process 2000, 80(7),
1307–20.
[121] Rulkov NF, Sushchik MM, Tsimring LS, et al. Digital communication using
chaotic-pulse-position modulation. IEEE Trans CAS-I 2001, 48(12), 1436–44.
[122] Heidari-Bateni G, McGillem CD. A chaotic direct-sequence spread-spectrum communication
system. IEEE Trans Commun 1994, 42(2/3/4), 1524–7.
[123] Sun KH, Zhou JL, Mou J. Design and performance analysis of multi-user chaotic sequence
spread-spectrum communication system. J Electron Inf Technol 2007, 29(10),
2436–40.
[124] Liao XF, Xiao D, Chen Y, Xiang T. Principle of chaos cryptography and its application. Beijing,
Science Press, 2009.
[125] Li CB, He B. Chaotic encryption algorithm and hash function construction research. Beijing,
Publishing House of Electronic Industry, 2011.
[126] Su M, Wen WY, Zhang Y S Security evaluation of bilateral-diffusion based image encryption
algorithm. Nonlinear Dyn 2014, 77(1–2), 243–6
[127] Dong C. Color image encryption using one-time keys and coupled chaotic systems. Signal
Process Image Commun 2014, 29(5), 628–40.
328 References
[128] Yang YX, Lin XD. Coding cryptology. Beijing, People’s Posts and Telecommunications
Publishing House, 1992.
[129] Roland S. Use of chaotic dynamical systems in cryptography. J Franklin Inst 2001, 338(4),
429–41.
[130] Li P, Zhang W, Li Z. FPGA implementation of a coupled-map-lattice-based cryptosystem. Int J
Circuit Theor Appl 2010, 38(1), 85–98.
[131] Wang FL. A new pseudo-random number generator and application to a digital secure
communication scheme based on compound symbolic chaos. Acta Phys Sin 2011, 60(11),
110517.
[132] Hu YH, Luo J, Su H. A new method for improving the limited precision effect of chaotic
sequences. Syst Simul 2012, 24 (11), 2349–52.
[133] Wang XS. A chaotic sequence encryption method. Chin J Comput 2002, (4), 351–6.
[134] Liu WH, Sun, KH, Zhu CX. A fast image encryption algorithm based on chaotic map. Optics and
Lasers in Engineering 2016, 84: 26–36.
[135] Wang L. A novel chaos-based pseudo-random number generator. Acta Phys Sin 2006, 55(8),
3964–8.
[136] Wang XY, Bao XM. A novel block cryptosystem based on the coupled chaotic map lattice.
Nonlinear Dyn 2013, 72(4), 707–15.
[137] Sheng LY, Wen J, Cao LL, Xiao YY. Differential cryptanalysis of TD-ERCS chaos. Acta Phys Sin
2007, 56(1), 78–83.
[138] Sheng LY, Xiao YY, Sheng Z. A universal algorithm for transforming chaotic sequences into
uniform pseudo-random sequences. Acta Phys Sin 2008, 57(7), 4007–13
[139] McLauchlan L, Mehruebeoglu M. DWT and DCT embedded watermarking using Chaos theory.
Conference on Mathematics of Data/Image Coding, Compression, and Encryption with
Applications XII, San Diego, CA, 2010, 8.
[140] Millerioux G. Crypyanalysis of hybrid cryptosystems. Int J Bifurcation Chaos 2013, 23(10),
1350173.
[141] Gangadhar C, Rao KD. Hyperchaos based image encryption. Int J Bifurcation Chaos 2009,
19(11), 3833–9.
[142] Cao GH. Image scrambling algorithm based on chaotic weighted sampling theory and sorting
transformation. J Beijing Univ Aeronaut Astronaut 2013, 39(1), 67–72
[143] Li HJ, Zhang JS. Embedding adaptive arithmetic coder in chaos-based cryptography. Chin Phys
B 2010, 19(5), 050508.
[144] Rahman SMM, Hossain MA, Mouftah H. Chaos-cryptography based privacy preservation
technique for video surveillance. Multimedia Syst 2012, 18(2), 145–55.
[145] Vidal G, Baptista MS, Mancini H. Fundamentals of a classical chaos-based Cryptosystem with
some quantum cryptography features. Int J Bifurcation Chaos 2012, 22(10), 1250243.
[146] Fouda JSAE, Effa JY, Sabat SL, Ali M. A fast chaotic block cipher for image encryption. Commun
Nonlinear Sci Numer Simul 2014, 19(3), 578–88.
[147] Wang YM, Liu JW. Security of communication network-theory and technology. Xi’an, Xi’an
Electronic and Science University Press, 2000.
[148] Zhang FX, Zheng YJ. Communication secure technology. Beijing, National Defence Industry
Press, 2003.
[149] Feng DG. Research status and development trend of domestic and foreign cryptography.
J Commun 2002, 23(5), 18–26.
[150] American National Standards Institute, Triple Data Encryption Algorithm Modes of Operation,
ANSI standard X9.52, 1998.
[151] Kocarev L. Chaos-based cryptography: A brief overview. IEEE Circuit Syst Mag 2001, 1(3), 6–21.
[152] Dachselt F, Kelber K, Schwartz W. Discrete-time chaotic encryption system-part III:
Cryptographical analysis. IEEE Trans CAS-I 1998, 45(9), 1–19.
References 329
[153] Yang T, Yang LB, Yang CM. Cryptanalyzing chaos secure communication using return maps.
Phys Lett A 1998, 245(6), 495–510.
[154] Zhou LH, Peng ZW, Feng ZJ, Zhong TX. Security property of chaotic encryption systems.
J Shanghai Jiaotong Univ 2001, 35(1), 133–8.
[155] Syyid Umar. The adaptive communication environment: “ACE”. Hughes network systems,
1998.
[156] Bhargava B, Shi CG, Wang SY. MPEG video encryption algorithms. Multimedia Tool Appl 2004,
24(1), 57–79.
[157] Zeng W, Lei S. Efficient frequency domain selective scrambling of digital video. IEEE Trans
Multimedia 2003, 5(1), 118–29.
[158] Liu Z, Li X, Dong ZY. Enhancing security of frequency domain video encryption. Proceedings of
ACM Multimedia, New York, USA, Poster Session, 2004, 304–7.
[159] Liu G, Ikenaga T, Goto S, Baba T. A selective video encryption scheme for MPEG compression
standard. IEICE Trans Fundam Electron Commun Comput Sci 2006, E89–A(1), 194–202.
[160] Shang F, Sun K H, Cai YQ. A new efficient MPEG video encryption system based on chaotic
maps. Proceedings-1st International Congress on Image and Signal Processing, CISP, 2008, v
3, 12–16, Haikou.
[161] Hilfer R. Applications of fractional calculus in physics. New Jersey, World Scientific, 2001.
[162] Mandelbort BB. The fractal geometry of nature. New York, Freeman, 1983.
[163] Hartley TT, Lorenzo CF, Qammer HK. Chaos in a fractional order Chua’s system. IEEE Trans
Circuits Syst- 1995, 42(8), 485–90.
[164] Arena P, Caponetto R, Fortuna L, Porto D. Chaos in fractional order Duffing system.
Proceedings of the European Conference on Circuit Theory and Design. (ECCTD, 97), Budapest.
Hungary, September 1997, 1258–72.
[165] Ge ZM, Ou CY. Chaos in a fractional order modified Duffing system. Chaos Solitons Fractals
2007, 34(2), 262–91.
[166] Ahmad WM, Sprott JC. Chaos in fractional-order autonomous nonlinear system. Chaos
Solitons Fractals 2003, 16(2), 339–51.
[167] Li CG, Chen GR. Chaos and hyperchaos in the fractional-order Rössler equations. Phys A: Stat
Mech Appl 2004, 341(1–4), 55–61.
[168] Podlubny I. Fractional differential equations. New York, Academic Press, 1999.
[169] Li CP, Deng WH. Remarks on fractional derivatives. Appl Math Comput 2007, 187(2), 777–84.
[170] Charef A, Sun HH. Fractal system as represented by singularity function. IEEE Trans Autom
Control 1992, 37(9), 1465–70.
[171] Kai D. An algorithm for the numerical solution of differential equations of fractional order.
Electron Trans Numer Anal 1998, 5(Mar.), 612–7.
[172] Kai D, Ford NJ, Freed AD. A predictor-corrector approach for the numerical solution of fractional
differential equations. Nonlinear Dyn 2002, 29(1–4), 3–22.
[173] Kai D, Ford NJ. Analysis of fractional differential equations. J Math Anal Appl 2002, 265(2),
229–48.
[174] Ahmad W. Hyperchaos in fractional order nonlinear systems. Chaos Solitons Fractals 2005,
26(5), 1459–65.
[175] Arman KB, Kia F, Naser P, Henry L. A chaotic secure communication scheme using fractional
chaotic systems based on an extended fractional Kalman filter. Commun Nonlinear Sci Numer
Simul 2009, 14(3), 863–79.
[176] Bohannan GW, Hurst SK, Spangler L. Electrical component with fractional order impedance.
2006, No. PCT/US2006/008821.
[177] Wang FQ, Liu CX. Study on the critical chaotic system with fractional order and circuit
experiment. Acta Phys Sin 2006, 55(8), 3922–7
330 References
[178] Liu CX. A hyperchaotic system and its fractional-order circuit simulation. Acta Phys Sin 2007,
56(12), 6865–73.
[179] Chen XR, Liu CX, Wang FQ. Study on the fractional-order Liu chaotic system with circuit
experiment and its control. Acta Phys Sin 2008, 57(3), 1416–22.
[180] Zhu CX, Zou Y. Summary of research on fractional-order control. Control Decis 2009, 24(2),
161–9.
[181] Liu SD, Shi SY, Liu SS. Bridge between weather and climate: Fractional derivatives. Meteor Sci
Technol 2007, 35(1), 15–19.
[182] Zhao CN, Li YS, Lu T.Analysis and design of fractional-order system. Beijing, National Defense
Industry Press, 2011.
[183] Adomian G. A new approach to nonlinear partial differential equations. J Math Anal Appl 1984,
102(2), 420–34.
[184] Cafagna D, Grassi G. Bifurcation and chaos in the fractional-order Chen system via a
time-domain approach. Int J Bifurcation Chaos 2008, 18(18), 1845–63.
[185] Cafagna D, Grassi G. Hyperchaos in the fractional-order Rössler system with lowest-order. Int J
[186] Sun KH, Wang X, Sprott JC. Bifurcations and chaos in fractional-order simplified Lorenz
system. Int J Bifurcation Chaos 2010, 20(4), 1208–19.
[187] He SB, Sun KH, Wang HH. Solution of the fractional-order chaotic system based on Adomian
decomposition algorithm and its complexity analysis. Acta Phys Sin 2014, 63(3), 030502.
[188] Sun KH, Yang JL, Qiu SS. Study of Simulation approaches for fractional-order chaotic systems.
J Syst Simul 2011, 23(11), 2361–5, 2370.
[189] Chen GR, Lü JH. Dynamic analysis, control and synchronization of lorenz family. Beijing,
Science Press, 2006.
[190] Wolf A, Swift JB, Swinney HL, Vastano JA. Determining Lyapunov exponents from a time series.
Physica D 1985, 16(3), 285–317.
[191] Sun KH, Ren J, Qiu SS. Chaotic dynamic characteristics of fractional-order unified system.
J South China Univ Technol (Natural Science Edition) 2008, 36(8), 6–10.
[192] Sun KH, Sprott JC. Bifurcations of fractional-order diffusionless Lorenz system. Electron J Theor
Phys 2009, 6(22), 123–34.
[193] Pomeau Y, Manneville P. Intermittent transition to turbulence in dissipative dynamical
systems. Commun Math Phys 1980, 74(2), 189–97.
[194] Sprott JC. Chaos and time-series analysis. New York, Oxford University Press, 2003.
[195] Yorke JA, Yorke ED. Metastable chaos: The transition to sustained chaotic behavior in the
Lorenz model. J Stat Phys 1979, 21(3), 263–77.
[196] Yang JL. Research on circuit design and implementation of simplified Lorenz system. Central
South University, 2011.
[197] Sprott JC. Some simple chaotic Jerk functions. Am J Phys 1997, 65(6), 537–43.
[198] Zuo T, Sun KH, Ai XX, Wang HH. High-order grid multi-scroll chaotic attractors generated by the
second generation current conveyor circuit. IEEE Trans Circuits Syst-II 2014, 61(10), 818–22.
[199] Merah L, Ali-Pacha A, Said N H, Mamat M. Design and FPGA implementation of Lorenz chaotic
system for information security issues. Appl Math Sci 2013, 7(5), 237–46.
[200] Xu X, Yu S, Xu L. Chaotic digital image encryption and its hardware implementation based on
DSP technology. Electron Sci Technol 2011, 24(3), 1–5.
[201] Yu SM. Chaotic systems and chaotic circuits: Principle, design and its application in
communications. Xi’an, Xi’an Electronic and Science University Press, 2011.
[202] Wang HH, Sun KH, He SB. Characteristic analysis and DSP realization of fractional-order
simplified Lorenz system based on Adomain decomposition method. Int J Bifurcation Chaos
2015, 25(6), 155008.
Index
0-1 test 30, 33 Chaotic sequence encryption 168
3DES 210 Chaotic spread spectrum code 150
Chaotic spread spectrum communication 148
Active-passive synchronization 12 Chaotic spread-spectrum communication 17
Adaptive communication environment 216, 218, Chaotic stream cipher 233
220 Chen attractor 55
Adaptive synchronization 12, 84 Chen hyperchaotic system 63
Adomian decomposition algorithm 248, 292 Chen’s attractor 9
Adomian decomposition method 259 Chosen-ciphertext 216
Anti brute force attack 182, 204 Chosen-plaintext 216
Anti-chosen plaintext attack 168 Chua’s circuit 8, 51
Auto-correlation 228 Cipher synchronization 209, 210
Autonomous equation 50 Circuit design process 278
Circuit implementation 240
Balance degree 151
Circuit simulation 253, 258, 276
Behavior complexity 35
Circuit simulation method 253
Bifurcation analysis 268
Color image watermark 182
Biorthogonal wavelet 188
Complexity 35, 240
Block cipher system 165, 207
Condition Lyapunov exponent 29, 69
Brute-force attack 215, 231
Continuous time chaotic system 47
Butterfly effect 3, 4
Continuous variable feedback
C 0 algorithm 36 synchronization 12
C 0 complexity 38, 39 COOK 131
Caputo fractional differential 242 Correlation analysis 202
Cat map 226 Correlation coefficient 202
Chaos 2, 3, 6, 9 Correlator 134
Chaos communication 125 Coupling synchronization 12
Chaos control 11 Current conveyor 282
Chaos cryptography 18
Chaos encryption 18 Data encryption standard 176, 210
Chaos observer 104 Dead-Beat Synchronization 13
Chaos parameter modulation 17, 145 Diffusionless Lorenz system 253
Chaos phenomenon 2 Digital voice encryption communication
Chaos self-synchronization 11 system 207
Chaos sequence encryption 167 Direct observation analysis 261
Chaos shift keying 16, 130 Discrete cosine transform 224
Chaos synchronization 11, 66 Discrete time chaotic map 41
Chaos theory 2, 164 Dividing frequency sampling method 33
Chaotic cryptography 164 Double-scroll attractor 52
Chaotic diffusion 201 Drive feedback synchronization 69
Chaotic encryption 164, 166 Drive-response synchronization 11
Chaotic masking 16, 125, 126 DSP implementation 290, 292
Chaotic pseudo random sequence generator 178 Duffing oscillator 48
Chaotic pulse position modulation 148 Dynamic characteristic 239
Chaotic scrambling 197 Dynamic simulation 252, 253, 258, 274
Chaotic secure communication 15, 125 Dynamic simulation analysis method 251
332 Index
Encryption algorithm 172, 179 Maximum condition Lyapunov exponent 86

Equal gain combining 156 Maximum conditional Lyapunov exponent 82
Equal interval drive synchronization 209, 214 Multi scroll Jerk system 282
Equivalent circuit 253 Multipath channel 157
Ergodicity 10 Multipath diversity reception 155
Error curve 72 Multipath transmission 155, 156
Multiple variable drive feedback
Feigenbaum constant 6, 10 synchronization 70
Five-scroll chaotic attractor 53 Multi-user chaotic spread spectrum
FM-DCSK 139 communication 158
Folded towel map 64
Fractal dimension 10, 30 Non-coherent demodulation 133
Fractional calculus 238, 241 Normalized correlation coefficient 191
Fractional integral 242, 243 Numerical simulation 256, 258
Fractional-order chaotic system 238 Numerical solution 248
Fractional-order diffusionless Lorenz chaotic
One-time pad 214, 216
system 257
One-way coupled map lattice 199
Fractional-order diffusionless Lorenz
Operational amplifier 282
system 266
Fractional-order simplified Lorenz system 249 Peak signal-noise ratio 191
Fractional-order unified system 260 Period-doubling bifurcation 7
Frame synchronization 209 Phase diagram analysis 21
Frequency modulator 139 Piecewise linear chaotic map 226
Function projective synchronization 113 Piecewise linear function 52
Poincaré section 19, 24
Gamma function 241 Power spectral density 19
Grunwald-Letnikov fractional calculus 241 Power spectral density function 22
Power spectrum analysis 262
H.264, 232
Projective Synchronization 13, 108, 110
H.264 standard 233 Pseudo phase space method 34
Hénon map 44, 80, 81 Pulse position modulation 145
Histogram analysis 202 Pulse synchronization 13
Hyperchaotic system 62
Quadrature chaotic shift keying 140
Image diffusion 197
Image scrambling 196 Rake receiver 155, 156, 158
Intermittent feedback synchronization 94 Randomness 10
Resisting brute force attack 168
Jury stability criteria 81 Riemann-Liouville fractional calculus 241
Root mean square 151, 152
Key sensitivity analysis 179, 203 Rössler equation 50
Rössler hyperchaotic system 62, 71
Logistic map 41, 211 RTP protocol 218
Lorenz system 49
Lyapunov exponent 10, 25, 28, 197 Scaling property 10
Lyapunov exponent spectrum 27 Scrambling algorithm 229
Lyapunov function 93, 117 Scrambling degree 181
Lyapunov stability principle 67 Scrambling table 231
Lyapunov stable theory 92 Second generation current conveyor 283
Index 333
Secure chaotic system 169 Synchronization robustness 68

Sequence cipher 164 Synchronization setup time 68, 72
Sequence cipher system 207 Synchronization stability 67
Seven-scroll chaotic attractor 54 Synchronization steady state error 68
Simplified Lorenz chaotic system 110
Simplified Lorenz system 27, 33, 59 TD-ERCS 46, 178
Spectral entropy complex 36 Tent map 44
Spectrum analysis 216 The largest Lyapunov exponent 263
State observer 101 Three-scroll chaotic attractor 52
State observer synchronization 105 Time domain approximation 247
Statistical analysis 202
Strange attractor 28 Unified chaotic system 57, 91, 259
Stream cipher 225 Uniform distribution 176
Structure complexity 36
Synchronization dynamic characteristics 69 Watermark algorithm 186
Synchronization error 74 Watermark embedding algorithm 189
Synchronization performance 67, 82, 123 Watermark extraction algorithm 190
Synchronization region 68 Wavelet transform 183

Chaotic Communication

Uploaded by

Copyright:

Available Formats

Chaotic Communication

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

Chaotic Communication

Uploaded by

Copyright:

Available Formats

Kehui Sun

Chaotic Secure Communication

Coding and Cryptography

Modern Communications Technology

A Course in Mathematical Cryptography

Computational Methods in Applied Mathematics

Principles and Technologies

Library of Congress Cataloging-in-Publication Data

Bibliographic information published by the Deutsche Nationalbibliothek

© 2016 Walter de Gruyter GmbH, Berlin/Boston

G Y from Hangzhou University of Electronic Science and Technology, Professor Zhang

2 Characteristic Analysis Methods for Nonlinear System 21

3 Typical Chaotic Systems 41

4 Chaotic Synchronization Principle and Method 66

4.6 Intermittent Feedback Synchronization Control 94

5 Secure Communication Technology

5.4.5 Performance Analysis of Multiuser Chaotic Spread

7 Audio and Video Chaotic Encryption

7.1.3 Cipher Synchronization Technology 209

8 Analysis and Simulation of Fractional-Order Chaotic System 238

8.4.2 Circuit Simulation Method of Fractional-Order Chaotic

9 Simulation and Hardware Implementation of Chaotic System 274

9.5.1 DSP Experimental Platform of Chaotic System 287

1.1 Linear, Nonlinear, and Chaos

1.2 Development of Chaos

deterministic, which is completely determined by the equations, nonlinear system

1.3 Famous Scientists and Important Events

1.3.1 Lorenz and Butterfly Effect

1.3.2 Li Tien-Yien and the Concept of Chaos

1.3.3 Feigenbaum and Feigenbaum Constant

Feigenbaum M. J. (1944–), a professor of physics at Cornell University, entered the

Bifurcation width wi also forms a geometric sequence

Calculation of Feigenbaum constant is shown in Figure 1.1.

k Bifurcation Bifurcation value +k Spacing ratio (+k – +k–1 )/(+k+1 – +k )

Figure 1.1: Calculation of Feigenbaum constant.

1.3.4 Leon Ong Chua and Chua’s Circuit

Leon Ong Chua (1936–) is a professor at Department of Electrical Engineering and

Figure 1.2: Chua’s circuit.

1.3.5 Guanrong Chen and Chen’s Attractor

Guanrong Chen (1948–) is a professor in the Department of Electronic Engineering in

Figure 1.3: Chen attractor (x–z plane).

1.4 Definition and Characteristics of Chaos

Chaos has aperiodic bounded dynamic behaviors in deterministic nonlinear dynam-

1.5 Overview of Chaos Synchronization Method

self-synchronization method [10]. Synchronization between two chaotic systems was

1.5.1 Chaos Synchronization Methods and Characteristics

Table 1.2: Chaos synchronization methods.

Time chaotic synchronization Spatiotemporal chaos Hyperchaos synchronization

Drive-response synchronization Feedback technology Drive-response synchronization

nonlinear systems, they cannot be decomposed due to physical nature or natural

Active-Passive Synchronization: In 1995, Kocarev and Parlitoz proposed active-

Coupling Synchronization: In 1990, Winful and Rahman theoretically studied the

Continuous Variable Feedback Synchronization: In 1993, German scholars Pyragas

Adaptive Synchronization: In 1990, Huberman and Lumer employed adaptive

Projective Synchronization: In 1999, Mainieri and Rehacek observed a new phe-

Performance of different synchronization methods is different. Based on Lorenz

1.5.2 Other Synchronization Methods and Problems

of chaotic secure communication to improve the stability of synchronization and

1.6 Summary of Chaos Secure Communication