Control Engineering Series 84

An adaptive system for linear systems with unknown parameters is a

nonlinear system. The analysis of such adaptive systems requires similar
techniques to analyse nonlinear systems. Therefore it is natural to treat
adaptive control as a part of nonlinear control systems.
Nonlinear and Adaptive Control Systems treats nonlinear control
and adaptive control in a unified framework, presenting the major
results at a moderate mathematical level, suitable for MSc students
and engineers with undergraduate degrees. Topics covered include
introduction to nonlinear systems; state space models; describing
functions for common nonlinear components; stability theory; feedback
linearization; adaptive control; nonlinear observer design; backstepping
design; disturbance rejection and output regulation; and control
applications, including harmonic estimation and rejection in power
distribution systems, observer and control design for circadian rhythms,
and discrete-time implementation of continuous-time nonlinear control
laws.

Zhengtao Ding is a Senior Lecturer in Control

Engineering and Director for MSc in Advanced Control
and Systems Engineering at the Control Systems
Centre, School of Electrical and Electronic Engineering,
The University of Manchester, UK. His research interests
focus on nonlinear and adaptive control design. He
pioneered research in asymptotic rejection of general
periodic disturbances in nonlinear systems and
produced a series of results to systematically solve this
problem in various situations. He also made significant
contributions in output regulation and adaptive control
of nonlinear systems with some more recent results on
observer design and output feedback control as well.
Dr Ding has been teaching Nonlinear and Adaptive
Control Systems to MSc students for 9 years, and he
has accumulated tremendous experiences in explaining
difficult control concepts to students.

Nonlinear and Adaptive

Control Systems

Nonlinear and Adaptive

Control Systems

Nonlinear and
Adaptive Control
Systems

Ding

Zhengtao Ding

The Institution of Engineering and Technology

www.theiet.org
978-1-84919-574-4

Nonlinear and Adaptive Control.indd 1

04/03/2013 09:32:20

IET CONTROL ENGINEERING SERIES 84

Nonlinear and
Adaptive Control
Systems

Other volumes in this series:

Volume 8
Volume 18
Volume 20
Volume 28
Volume 33
Volume 34
Volume 35
Volume 37
Volume 39
Volume 40
Volume 41
Volume 42
Volume 44
Volume 47
Volume 49
Volume 50
Volume 51
Volume 52
Volume 53
Volume 54
Volume 55
Volume 56
Volume 57
Volume 58
Volume 59
Volume 60
Volume 61
Volume 62
Volume 63
Volume 64
Volume 65
Volume 66
Volume 67
Volume 68
Volume 69
Volume 70
Volume 71
Volume 72
Volume 73
Volume 74
Volume 75
Volume 76
Volume 77
Volume 78

A history of control engineering, 18001930 S. Bennett

Applied control theory, 2nd Edition J.R. Leigh
Design of modern control systems D.J. Bell, P.A. Cook and N. Munro (Editors)
Robots and automated manufacture J. Billingsley (Editor)
Temperature measurement and control J.R. Leigh
Singular perturbation methodology in control systems D.S. Naidu
Implementation of self-tuning controllers K. Warwick (Editor)
Industrial digital control systems, 2nd Edition K. Warwick and D. Rees (Editors)
Continuous time controller design R. Balasubramanian
Deterministic control of uncertain systems A.S.I. Zinober (Editor)
Computer control of real-time processes S. Bennett and G.S. Virk (Editors)
Digital signal Processing: principles, devices and applications N.B. Jones and
J.D.McK. Watson (Editors)
Knowledge-based systems for industrial control J. McGhee, M.J. Grimble and A. Mowforth
(Editors)
A History of control engineering, 19301956 S. Bennett
Polynomial methods in optimal control and ltering K.J. Hunt (Editor)
Programming industrial control systems using IEC 1131-3 R.W. Lewis
Advanced robotics and intelligent machines J.O. Gray and D.G. Caldwell (Editors)
Adaptive prediction and predictive control P.P. Kanjilal
Neural network applications in control G.W. Irwin, K. Warwick and K.J. Hunt (Editors)
Control engineering solutions: a practical approach P. Albertos, R. Strietzel and N. Mort
(Editors)
Genetic algorithms in engineering systems A.M.S. Zalzala and P.J. Fleming (Editors)
Symbolic methods in control system analysis and design N. Munro (Editor)
Flight control systems R.W. Pratt (Editor)
Power-plant control and instrumentation: the control of boilers and HRSG systems
D. Lindsley
Modelling control systems using IEC 61499 R. Lewis
People in control: human factors in control room design J. Noyes and M. Bransby (Editors)
Nonlinear predictive control: theory and practice B. Kouvaritakis and M. Cannon (Editors)
Active sound and vibration control M.O. Tokhi and S.M. Veres
Stepping motors, 4th edition P.P. Acarnley
Control theory, 2nd Edition J.R. Leigh
Modelling and parameter estimation of dynamic systems J.R. Raol, G. Girija and J. Singh
Variable structure systems: from principles to implementation A. Sabanovic, L. Fridman
and S. Spurgeon (Editors)
Motion vision: design of compact motion sensing solution for autonomous systems
J. Kolodko and L. Vlacic
Flexible robot manipulators: modelling, simulation and control M.O. Tokhi and
A.K.M. Azad (Editors)
Advances in unmanned marine vehicles G. Roberts and R. Sutton
(Editors)
Intelligent control systems using computational intelligence techniques A. Ruano
(Editor)
Advances in cognitive systems S. Nefti and J. Gray (Editors)
Control theory: a guided tour, 3rd Edition J.R. Leigh
Adaptive sampling with Mobile WSN K. Sreenath, M.F. Mysorewala, D.O. Popa
and F.L. Lewis
Eigenstructure control algorithms: applications to aircraft/rotorcraft handling
qualities design S. Srinathkumar
Advanced control for constrained processes and systems F. Garelli, R.J. Mantz and
H. De Battista
Developments in control theory towards glocal control L. Qiu, J. Chen, T. Iwasaki and
H. Fujioka
Further advances in unmanned marine vehicles G.N. Roberts and R. Sutton (Editors)
Frequency-domain control design for high-performance systems J. OBrien

Nonlinear and
Adaptive Control
Systems
Zhengtao Ding

The Institution of Engineering and Technology

Published by The Institution of Engineering and Technology, London, United Kingdom

The Institution of Engineering and Technology is registered as a Charity in England &
Wales (no. 211014) and Scotland (no. SC038698).
2013 The Institution of Engineering and Technology
First published 2013
This publication is copyright under the Berne Convention and the Universal Copyright
Convention. All rights reserved. Apart from any fair dealing for the purposes of research or
private study, or criticism or review, as permitted under the Copyright, Designs and Patents Act
1988, this publication may be reproduced, stored or transmitted, in any form or by any means,
only with the prior permission in writing of the publishers, or in the case of reprographic
reproduction in accordance with the terms of licences issued by the Copyright Licensing
Agency. Enquiries concerning reproduction outside those terms should be sent to the
publisher at the undermentioned address:
The Institution of Engineering and Technology
Michael Faraday House
Six Hills Way, Stevenage
Herts, SG1 2AY, United Kingdom
www.theiet.org
While the author and publisher believe that the information and guidance given in
this work are correct, all parties must rely upon their own skill and judgement when
making use of them. Neither the author nor the publisher assumes any liability to
anyone for any loss or damage caused by any error or omission in the work, whether
such an error or omission is the result of negligence or any other cause. Any and all
such liability is disclaimed.
The moral rights of the author to be identied as author of this work have been
asserted by him in accordance with the Copyright, Designs and Patents Act 1988.

British Library Cataloguing in Publication Data

A catalogue record for this product is available from the British Library

ISBN 978-1-84919-574-4 (hardback)

ISBN 978-1-84919-575-1 (PDF)

Typeset in India by MPS Limited

Printed in the UK by CPI Group (UK) Ltd, Croydon

To my family, Yinghong, Guangyue and Xiang

Contents

Preface
1 Introduction to nonlinear and adaptive systems
1.1
Nonlinear functions and nonlinearities
1.2
Common nonlinear systems behaviours
1.3
Stability and control of nonlinear systems

ix
1
1
4
5

2 State space models

2.1
Nonlinear systems and linearisation around an operating point
2.2
Autonomous systems
2.3
Second-order nonlinear system behaviours
2.4
Limit cycles and strange attractors

9
9
11
12
18

3 Describing functions
3.1
Fundamentals
3.2
Describing functions for common nonlinear components
3.3
Describing function analysis of nonlinear systems

25
26
29
34

4 Stability theory
4.1
Basic denitions
4.2
Linearisation and local stability
4.3
Lyapunovs direct method
4.4
Lyapunov analysis of linear time-invariant systems

41
41
45
46
51

5 Advanced stability theory

5.1
Positive real systems
5.2
Absolute stability and circle criterion
5.3
Input-to-state stability and small gain theorem
5.4
Differential stability

55
55
59
65
71

6 Feedback linearisation
6.1
Inputoutput linearisation
6.2
Full-state feedback linearisation

75
75
83

7 Adaptive control of linear systems

7.1
MRAC of rst-order systems
7.2
Model reference control
7.3
MRAC of linear systems with relative degree 1
7.4
MRAC of linear systems with high relatives
7.5
Robust adaptive control

89
90
94
99
102
103

viii

Nonlinear and adaptive control systems

8 Nonlinear observer design

8.1
Observer design for linear systems
8.2
Linear observer error dynamics with output injection
8.3
Linear observer error dynamics via direct state transformation
8.4
Observer design for Lipschitz nonlinear systems
8.5
Reduced-order observer design
8.6
Adaptive observer design

109
109
111
120
122
127
136

9 Backstepping design
9.1
Integrator backstepping
9.2
Iterative backstepping
9.3
Observer backstepping
9.4
Backstepping with ltered transformation
9.5
Adaptive backstepping
9.6
Adaptive observer backstepping

141
141
144
147
152
159
167

10

Disturbance rejection and output regulation

10.1 Asymptotic rejection of sinusoidal disturbances
10.2 Adaptive output regulation
10.3 Output regulation with nonlinear exosystems
10.4 Asymptotic rejection of general periodic disturbances

175
175
186
194
204

11

Control applications
11.1 Harmonics estimation and rejection in power distribution
systems
11.1.1 System model
11.1.2 Iterative observer design for estimating frequency
modes in input
11.1.3 Estimation of specic frequency modes in input
11.1.4 Rejection of frequency modes
11.1.5 Example
11.2 Observer and control design for circadian rhythms
11.2.1 Circadian model
11.2.2 Lipschitz observer design
11.2.3 Phase control of circadian rhythms
11.3 Sampled-data control of nonlinear systems
11.3.1 System model and sampled-data control
11.3.2 Stability analysis of sampled-data systems
11.3.3 Simulation

219
219
220
224
232
234
235
238
239
241
243
247
249
251
260

Bibliographical Notes

263

References

268

Index

275

Preface

This book is intended for the use as a textbook at MSc and senior undergraduate
level in control engineering and related disciplines such as electrical, mechanical,
chemical and aerospace engineering and applied mathematics. It can also be used as
a reference book by control engineers in industry and research students in automation
and control. It is largely, although not entirely, based on the course unit bearing the
same name as the book title that I have been teaching for several years for the MSc
course at Control Systems Centre, School of Electrical and Electronic Engineering,
The University of Manchester. The beginning chapters cover fundamental concepts
in nonlinear control at moderate mathematical level suitable for students with a rst
degree in engineering disciplines. Simple examples are used to illustrate important
concepts, such as the difference between exponential stability and asymptotic stability.
Some advanced and recent stability concepts such as input-to-state stability are also
included, mainly as an introduction at a less-demanding mathematical level compared
with their normal descriptions in the existing books, to research students who may
encounter those concepts in literature. Most of the theorems in the beginning chapters
are introduced with the proofs, and some of the theorems are simplied with less
general scopes, but without loss of rigour. The later chapters cover several topics
which are closely related to my own research activities, such as nonlinear observer
design and asymptotic disturbance rejection of nonlinear systems. They are included
to demonstrate the applications of fundamental concepts in nonlinear and adaptive
control to MSc and research students, and to bridge the gap between a normal textbook
treatment of control concepts and that of research articles published in academic
journals. They can also be used as references for the students who are working on
the related topics. At the end of the book, applications to less traditional areas such
as control of circadian rhythms are also shown, to encourage readers to explore new
applied areas of nonlinear and adaptive control.
This book aims at a unied treatment of adaptive and nonlinear control.
It is well known that the dynamics of an adaptive control system for a linear dynamic
system with unknown parameters are nonlinear. The analysis of such adaptive systems requires similar techniques to the analysis for nonlinear systems. Some more
recent control design techniques such as backstepping relies on Lyapunov functions
to establish the stability, and they can be directly extended to adaptive control of nonlinear systems. These techniques further reduce the traditional gap between adaptive
control and nonlinear control. Therefore, it is now natural to treat adaptive control
as a part of nonlinear control systems. The foundation for linear adaptive control
and nonlinear adaptive control is the positive real lemma, which is related to passive
systems in nonlinear control and Lyapunov analysis. It is decided to use the positive

Nonlinear and adaptive control systems

real lemma and related results in adaptive control and nonlinear control as the
main theme of the book, together with Lyapunov analysis. Other important results
such as circle criterion and backstepping are introduced as extensions and further
developments from this main theme.
For a course unit of 15 credits on nonlinear and adaptive control at the Control
Systems Centre, I normally cover Chapters 14, 6 and 7, and most of the contents of
Chapter 5, and about half of the materials in Chapter 9. Most of the topics covered
in Chapters 8, 10 and 11 have been used as MSc dissertation projects and some of
them as PhD projects. The contents may also be used for an introductory course
on nonlinear control systems, by including Chapters 15, 8 and the rst half of
Chapter 9, and possibly Chapter 6. For a course on adaptive control of nonlinear
systems, an instructor may include Chapters 1, 2, 4, 5, 7 and 9. Chapter 8 may be
used alone as a brief introduction course to nonlinear observer design. Some results
shown in Chapters 8, 10 and 11 are recently published, and can be used as references
for the latest developments in related areas.
Nonlinear and adaptive control is still a very active research area in automation
and control, with many new theoretic results and applications continuing to merge.
I hope that the publication of this work will have a good impact, however small,
on students interests to the subject. I have been beneted from my students, both
undergraduate and MSc students, through my teaching and other interactions with
them, in particular, their questions to ask me to explain many of the topics covered
in this book with simple languages and examples. My research collaborators and
PhD students have contributed to several topics covered in the book through joint
journal publications, whose names may be found in the references cited at the end of
the book. I would like to thank all the researchers in the area who contributed to the
topics covered in the book, who are the very people that make this subject fascinating.

Chapter 1

Introduction to nonlinear and adaptive systems

Nonlinearity is ubiquitous, and almost all the systems are nonlinear systems. Many
of them can be approximated by linear dynamic systems, and signicant amount of
analysis and control design tools can then be applied. However, there are intrinsic
nonlinear behaviours which cannot be described using linear systems, and analysis
and control are necessarily based on nonlinear systems. Even for a linear system,
if there are uncertainties, nonlinear control strategies such as adaptive control may
have to be used. In the last two decades, there have been signicant developments in
nonlinear system analysis and control design. Some of them are covered in this book.
In this chapter, we will discuss typical nonlinearities and nonlinear behaviours, and
introduce some basic concepts for nonlinear system analysis and control.

1.1

Nonlinear functions and nonlinearities

A dynamic system has its origin from dynamics in classic mechanics. The behaviour of
a dynamic system is often specied by differential equations. Variables in a dynamic
system are referred to as states, and they can be used to determine the status of a
system. Without external inuences, the state variables are sufcient to determine the
future status for a dynamic system. For a dynamic system described by continuous
differential equations, state variables cannot be changed instantly, and this reects the
physical reality. Many physical and engineering systems can be modelled as dynamic
systems, using ordinary differential equations. Application areas of dynamic systems
have expanded rapidly to other areas such as biological systems, nancial systems,
etc. Analysis of the behaviours of dynamic systems is essential to the understanding
of various applications in many science and engineering disciplines. The behaviour
of a dynamic system may be altered by exerting external inuences, and quite often
this kind of inuences is based on knowledge of the current state. In this sense, the
dynamic system is controlled to achieve certain behaviours.
State variables are denoted by a vector in an appropriate dimension for convenience, and the dynamic systems are described by rst-order differential equations of
the state vector. A linear dynamic system is described by
x = Ax + Bu
y = Cx + Du,

(1.1)

Nonlinear and adaptive control systems

where x Rn is the state, u Rm is the external inuence, which is referred to as

the input, and y Rs is a vector that contains the variables for measurement, which
is referred to as the output, and A Rnn , B Rnm , C Rsn and D Rsm are
matrices that may depend on time. If the system matrices A, B, C and D are constant
matrices, the system (1.1) is a linear time-invariant system. For a given input, the state
and the output can be computed using the system equation (1.1). More importantly,
the superposition principle holds for linear dynamic systems.
Nonlinear dynamic systems are the dynamic systems that contain at least one
nonlinear component, or in other words, the functions in the differential equations
contain nonlinear functions. For example we consider a Single-Input-Single-Output
(SISO) system with input saturation
x = Ax + B (u)
y = Cx + Du,

(1.2)

where : R R is a saturation function dened as

1
(u) = u

for u < 1,
otherwise,
for u > 1.

(1.3)

The only difference between the systems (1.2) and (1.1) is the saturation function .
It is clear that the saturation function is a nonlinear function, and therefore this
system is a nonlinear system. Indeed, it can be seen that the superposition principle
does not apply, because after the input saturation, any increase in the input amplitude
does not change the system response at all.
A general nonlinear system is often described by
x = f (x, u, t)
y = h(x, u, t),

(1.4)

where x Rn , u Rm and y Rs are the system state, input and output respectively,
and f : Rn Rm R Rn and h : Rn Rm R Rs are nonlinear functions.
Nonlinearities of a dynamic system are described by nonlinear functions. We may
roughly classify nonlinear functions in nonlinear dynamic systems into two types.
The rst type of nonlinear functions are analytical functions such as polynomials,
sinusoidal functions and exponential functions, or composition of these functions.
The derivatives of these functions exist, and their Taylor series can be used to obtain
good approximations at any points. These nonlinearities may arise from physical
modelling of actual systems, such as nonlinear springs and nonlinear resistors, or due
to nonlinear control design, such as nonlinear damping and parameter adaptation law
for adaptive control. There are nonlinear control methods such as backstepping which
requires the existence of derivatives up to certain orders.

Introduction to nonlinear and adaptive systems

Other nonlinearities may be described by piecewise linear functions, but with a

nite number of points where the derivatives do not exist or the functions are not even
continuous. The saturation function mentioned earlier is a piecewise linear function
that is continuous, but not smooth at the two joint points. A switch, or an idea relay,
can be modelled using a signum function which is not continuous. There may also be
nonlinearities which are multi-valued, such as relay with hysteresis, which is described
in Chapter 3 in detail. A nonlinear element with multi-valued nonlinearity returns a
single value at any instance, depending on the history of the input. In this sense,
the multi-valued nonlinearities have memory. Other single-valued nonlinearities are
memoryless. The nonlinearities described by piecewise linear functions are also called
hard nonlinearities, and the common hard nonlinearities include saturation, relay, dead
zone, relay with hysteresis, backlash, etc. One useful way to study hard nonlinearities
is by describing functions. Hard nonlinearities are also used in control design, for
example signum function is used in siding mode control, and saturation functions
are often introduced for control inputs to curb the peaking phenomenon for the semiglobal stability. In the following example, we show a nonlinearity that arises from
adaptive control of a linear system.
Example 1.1. Consider a rst-order linear system
x = ax + u,
where a is an unknown parameter. How to design a control system to ensure the
stability of the system? If a range a < a < a+ is known, we can design a control
law as
u = cx a+ x
with c > 0, which results in the closed-loop system
x = cx + (a a+ )x.
Adaptive control can be used in the case of completely unknown a,
u = cx a x
a = x2 .
If we let a = a a , the closed-loop system is described by
x = cx + a x
a = x2 .
This adaptive system is nonlinear, even though the original uncertain system is linear.
This adaptive system is stable, and it does need the stability theory introduced later
in Chapter 7 of the book.

Nonlinear and adaptive control systems

1.2 Common nonlinear systems behaviours

Many nonlinear systems may be approximated by linearised systems around operating
points, and their behaviours in close neighbourhoods can be predicted from the linear
dynamics. This is the justication of applying linear system theory to control study,
as almost all the practical systems are nonlinear systems.
There are many nonlinear features that do not exist in linear dynamic systems,
and therefore linearised dynamic models cannot describe the behaviours associated
with these nonlinear features. We will discuss some of those nonlinear features in the
following text.
Multiple equilibrium points are common for nonlinear systems, unlike the linear
system. For example
x = x + x2 .
This system has two equilibria at x = 0 and x = 1. The behaviours around these equilibrium points are very different, and they cannot be described by a single linearised
model.
Limit cycles are a phenomenon that periodic solutions exist and attract nearby
trajectories in positive or negative time. Closed curve solutions may exist for linear
systems, such as solution to harmonic oscillators. But they are not attractive to nearby
trajectories and not robust to any disturbances. Heart beats of human body can be
modelled as limit cycles of nonlinear systems.
High-order harmonics and subharmonics occur in the system output when subject
to a harmonic input. For linear systems, if the input is a harmonic function, the
output is a harmonic function, with the same frequency, but different amplitude
and phase. For nonlinear systems, the output may even have harmonic functions with
fractional frequency of the input or multiples of the input frequency. This phenomenon
is common in power distribution networks.
Finite time escape can happen in a nonlinear system, i.e., the system state tends
to innity at a nite time. This will never happen for linear systems. Even for an
unstable linear system, the system state can only grow at an exponential rate. The
nite time escape can cause a problem in nonlinear system design, as a trajectory
may not exist.
Finite time convergence to an equilibrium point can happen to nonlinear systems.
Indeed, we can design nonlinear systems in this way to achieve fast convergence.
This, again, cannot happen for linear systems, as the convergence rate can only be
exponential, i.e., a linear system can only converge to its equilibrium asymptotically.
Chaos can only happen in nonlinear dynamic systems. For some class of nonlinear systems, the trajectories are bounded, but not converge to any equilibrium
or limit cycles. They may have quasi-periodic solutions, and the behaviour is very
difcult to predict.
There are other nonlinear behaviours such as bifurcation, etc., which cannot
happen in linear systems. Some of the nonlinear behaviours are covered in detail in
this book, such as limit cycles and high-order harmonics. Limit cycles and chaos are

Introduction to nonlinear and adaptive systems

discussed in Chapter 2, and limit cycles also appear in other problems considered
in this book. High-order harmonics are discussed in disturbance rejection. When the
disturbance is a harmonic signal, the internal model for disturbance rejection has to
consider the high-order harmonics generated due to nonlinearities.

1.3 Stability and control of nonlinear systems

Nonlinear system behaviours are much more complex than those of linear systems.
The analytical tools for linear systems often cannot be applied to nonlinear systems.
For linear systems, the stability of a system can be decided by eigenvalues of system
matrix A, and obviously for nonlinear systems, this is not the case. For some nonlinear
systems, a linearised model cannot be used to determine the stability even in a very
small neighbourhood of the operating point. For example for a rst-order system
x = x3 ,
linearised model around the origin is x = 0. This linearised model is critically stable,
but the system is unstable. In fact, if we consider another system
x = x3 ,
which is stable, but the linearised model around the origin is still x = 0. Frequency
domain methods also cannot be directly applied to analysing inputoutput relationship
for nonlinear systems, as we cannot dene a transfer function for a general nonlinear
system.
Of course, some basic concept may still be applicable to nonlinear systems,
such as controllability, but these systems are often in different formulation and
use different mathematical tools. Frequency response method can be applied to
analysing a nonlinear system with one nonlinear component based on approximation
in frequency domain using describing function method. High gain control and zero
dynamics for nonlinear systems originate from their counterparts in linear systems.
One important concept that we need to address is stability. As eigenvalues are
no longer a suitable method for nonlinear systems, stability concepts and methods
to check stability for nonlinear systems are necessary. Among various denition,
Lyapunov stability is perhaps the most fundamental one. It can be checked by using a
Lyapunov function for stability analysis. Some of the other stability concepts such as
input-to-state stability may also be interpreted using Lyapunov functions. Lyapunov
functions can also provide valid information in control design. Lyapunov stability
will be the main stability concept used in this book.
Compared with the stability concepts of nonlinear systems, control design methods for nonlinear systems are even more diversied. Unlike control design for linear
systems, there is a lack of systematic design methods for nonlinear systems. Most of
the design methods can only apply to specic classes of nonlinear systems. Because
of this, nonlinear control is often more challenging and interesting. It is impossible
for the author to cover all the major areas of nonlinear control design and analysis,
partially to authors knowledge base, and partially due to the space constraint.

Nonlinear and adaptive control systems

People often start with linearisation of a nonlinear system. If the control design
based on a linearised model works, then there is no need to worry about nonlinear
control design. Linearised models depend on operating points, and a switching strategy might be needed to move from one operating point to the others. Gain scheduling
and linear parameter variation (LPV) methods are also closely related to linearisation
around operating points.
Linearisation can also be achieved for certain class of nonlinear systems through
a nonlinear state transformation and feedback. This linearisation is very much different from linearisation around operating points. As shown later in Chapter 6, a number
of geometric conditions must be satised for the existence of such a nonlinear transformation. The linearisation obtained in this way works globally in the state space,
not just at one operating point. Once the linearised model is obtained, further control
design can be carried out using design methods for linear systems.
Nonlinear functions can be approximated using articial neuron networks, and
fuzzy systems and control methods have been developed using these approximation
methods. The stability analysis of such systems is often similar to Lyapunov functionbased design method and adaptive control. We will not cover them in this book. Other
nonlinear control design methods such as bandband control and sliding mode control
are also not covered.
In the last two decades, there were developments for some more systematic
control design methods, such as backstepping and forwarding. They require the system
to have certain structures so that these iterative control designs can be carried out.
Among them, backstepping method is perhaps the most popular one. As shown in
Chapter 9 in the book, it requires the system state space function in a sort of lowertriangular form so that at each step a virtual control input can be designed. Signicant
amount of coverage of this topic can be found in this book. Forwarding control design
can be interpreted as a counterpart of backstepping in principle, but it is not covered
in this book.
When there are parametric uncertainties, adaptive control can be introduced to
tackle the uncertainty. As shown in a simple example earlier, an adaptive control system is nonlinear, even for a linear system. Adaptive technique can also be introduced
together with other nonlinear control design methods, such as backstepping method.
In such a case, people often give it a name, adaptive backstepping. Adaptive control
for linear systems and adaptive backstepping for nonlinear systems are covered in
details in Chapter 7 and Chapter 9 in this book.
Similar to linear control system design, nonlinear control design methods can
also be grouped as state-feedback control design and output-feedback control design.
The difference is that the separation principle is not valid for nonlinear control design
in general, that is if we replace the state in the control input by its estimate, we would
not be able to guarantee the stability of the closed-loop system using state estimate.
Often state estimation must be integrated in the control design, such as observer
backstepping method.
State estimation is an important topic for nonlinear systems on its own. Over the
last three decades, various observer design methods have been introduced. Some of
them may have their counterparts in control design. Design methods are developed

Introduction to nonlinear and adaptive systems

for different nonlinearities. One of them is for systems with Lipschitz nonlinearity,
as shown in Chapter 8. A very neat nonlinear observer design is the observer design
with output injection, which can be applied to a class of nonlinear systems whose
nonlinearities are only of the system output.
In recent years, the concept of semi-global stability is getting more popular.
Semi-global stability is not as good as global stability, but the domain of attraction
can be as big as you can specify. The relaxation in the global domain of attraction does
give control design more freedom in choosing control laws. One common strategy is
to use high gain control together with saturation. We will not cover it in this book,
but the design methods in semi-global stability can be easily followed once a reader
is familiar with the control design and analysis methods introduced in this book.

Chapter 2

State space models

The nonlinear systems under consideration in this book are described by differential
equations. In the same way as for linear systems, we have system state variables, inputs
and outputs. In this chapter, we will provide basic denitions for state space models of
nonlinear systems, and tools for preliminary analysis, including linearisation around
operating points. Typical nonlinear behaviours such as limit cycles and chaos will
also be discussed with examples.

2.1 Nonlinear systems and linearisation around

an operating point
A system is called a dynamic system if its behaviours depend on its history. States
of a system are the variables that represent the information of history. At any time
instance, the current state value decides the systems future behaviours. In this sense,
the directives of the state variables decide the systems behaviours, and hence we
describe a dynamic system by a set of rst-order differential equations in vector
form as
x = f (x, u, t),

x(0) = x0 ,

(2.1)

where x Rn is the state of the system, f : Rn Rn is a continuous function and

u Rm denotes the external inuence, which is usually referred to as the input to the
system.
For the differential equation (2.1) to exist as a unique solution for a given initial
condition, we need to impose a restriction on the nonlinear function f that f must be
Lipschitz with respect to the variable x. The denition of Lipschitz condition is given
below.
Denition 2.1. A function f : Rn Rm R Rn is Lipschitz with a Lipschitz
constant if for any vectors x, x Dx Rn , and u Du Rm and t It R, with
Dx , Du being the regions of interest and It being an time interval,
f (x, u, t) f (x, u, t) x x ,

(2.2)

with > 0.
Note that Lipschitz condition implies continuity with respect to x. The existence
and uniqueness of a solution for (2.1) are guaranteed by the function f being Lipschitz
and being continuous with respect to t.

10

Nonlinear and adaptive control systems

Remark 2.1. The continuity of f with respect to t and state variable x might be
stronger than we have in real applications. For example, a step function is not continuous in time. In the case that there are nite number of discontinuities in a given
interval, we can solve the equation of a solution in each of the continuous region, and
join them together. There are situations of discontinuity with state variable, such as
an ideal relay. In such a case, the uniqueness of the solution can be an issue. Further
discussion on this is beyond the scope of this book. In the systems considered in
the book, we would assume that there would be no problem with the uniqueness of
a solution.

The system state contains the whole information of the behaviour. However,
for a particular application, only a subset of the state variables or a function of state
variables is of interest, which can be denoted as y = h(x, u, t) with h : Rn Rm
R Rs , normally with s < n. We often refer to y as the output of the system. To
write them together with the system dynamics, we have
x = f (x, u, t),

x(0) = x0

y = h(x, u, t).
In this book, we mainly deal with time-invariant systems. Hence, we can drop the
variable t in f and h and write the system as
x = f (x, u),

x(0) = x0

(2.3)

y = h(x, u),

where x Rn is the state of the system, y Rs and u Rm are the output and the
input of the system respectively, and f : Rn Rm Rn and h : Rn Rm Rs are
continuous functions.
Nonlinear system dynamics are much more complex than linear systems in
general. However, when the state variables are subject to small variations, we would
expect the behaviours for small variations to be similar to linear systems, based on
the fact that
f (x + x, u + u) f (x, u) +

f
f
(x, u)x +
(x, u)u,
x
u

when x and u are very small.

An operating point at (xe , ue ) is taken with x = xe and u = ue being constants
such that f (xe , ue ) = 0. A linearised model around the operation point can then be
obtained. Let
x = x xe ,
u = u ue ,
y = h(x, u) h(xe , ue ),
then the linearised model is given by

State space models

x = Ax + Bu
y = C x + Du ,

11
(2.4)

where A Rnn , B Rnm , C Rsn , D Rsm matrices with elements ai,j , bi,j , ci,j
and di,j respectively shown by, assuming that f and h are differentiable,
ai,j =

fi
(xe , ue ),
xj

bi,j =

fi
(xe , ue ),
uj

ci,j =

hi
(xe , ue ),
xj

di,j =

hi
(xe , ue ).
uj

Remark 2.2. For a practical system, a control input can keep the state in an equilibrium point, i.e., at a point such that x = 0, and therefore it is natural to look at the
linearisation around this point. However, we can obtain linearised model at points that
are not at equilibrium. If (xe , ue ) is not an equilibrium point, we have f (xe , ue ) = 0.
We can carry out the linearisation in the same way, but the resultant linearised system
is given by
x = Ax + Bu + d
y = C x + Du ,
where d = f (xe , ue ) is a constant vector.

2.2 Autonomous systems

For a system in (2.3), the external inuence can only be exerted through the input u.
If a system does not take any input, its future state only depends on the initial state.
It means that the system is not inuenced by external factors, and the behaviours are
completely determined by the system state. Such a system is often referred to as an
autonomous system. A denition is given below.
Denition 2.2. An autonomous system is a dynamic system whose behaviour does
not explicitly depend on time.
In terms of differential equations, an autonomous system can be expressed as
x = f (x),
where x Rn is the state of the system, and f : Rn Rn .

(2.5)

12

Nonlinear and adaptive control systems

Remark 2.3. It is easy to see that for a system in (2.3), if the control input remains
constant, then it is an autonomous system. We only need to re-dene the function f as
fa (x) := f (x, uc ) where uc is a constant input. Even if the inputs are polynomials and
sinusoidal functions of time, we can convert the system to the autonomous system by
modelling the sinusoidal and polynomial functions as the state variables of a linear
dynamic system, and integrate this system into the original system. The augmented
system is then an autonomous system.

Denition 2.3. For an autonomous system (2.5), a point xe Rn is a singular point
if f (xe ) = 0.
It is easy to see that singular points are equilibrium points. Singular points are
more preferred for autonomous systems, especially for second-order systems.
Since autonomous system do not have external input, the set of all the trajectories provides a complete geometrical representation of the dynamic behaviour.
This is often referred to as the phase portrait, especially for second-order systems
in the format
x 1 = x2
x 2 = (x1 , x2 ).
In the above system, if we interpret x1 as the displacement, then x2 is the velocity. The
state variables often have clear physical meanings. Phase portraits can be obtained by
a number of methods, including analysing the behaviours near the singular points. In
fact, singular points might get the name from their positions in the phase portrait. In a
phase portrait, the lines usually do not intercept each other due to the uniqueness of the
solutions. However, they meet at the points where f (x) = 0, seemingly intercepting
each other. Those points are singular in this sense.

2.3 Second-order nonlinear system behaviours

For a second-order system, we can write the system equation as
x 1 = f1 (x1 , x2 )
x 2 = f2 (x1 , x2 ).
For an equilibrium point (x1e , x2e ), the linearised model is given by
 
 
x 1
x 1
=
A
,
e
x 2
x 2
where

f1
x1 (x1e , x2e )
Ae =
f2
(x1e , x2e )
x1

f1
(x1e , x2e )
x2
.

f2
(x1e , x2e )
x2

(2.6)

State space models

13

Therefore, the behaviour about this equilibrium or singular point is determined by

the properties of matrix Ae . Based on the eigenvalues of Ae , we can classify the
singular points in the following six different cases. We use 1 and 2 to denote the
two eigenvalues of Ae .
Stable node, for 1 < 0, 2 < 0. This is a case when both eigenvalues are negative
real numbers. The linearised model is stable, and a typical phase portrait around this
singular point is shown below.
x2

x1

Stable node

Unstable node, for 1 > 0, 2 > 0. This singular point is unstable, and the trajectories
diverge from the point, but not spiral around it.
x2

x1

Unstable node

Saddle point, for 1 < 0, 2 > 0. With one positive and one negative eigenvalues, the
hyperplane in three dimensions may look like a saddle. Some trajectories converge
to the singular point, and others diverge, depending on the directions of approaching
the point.
x2

x1

Saddle point

Stable focus, for 1,2 = j, ( < 0). With a negative real part for a pair of conjugate poles, the singular point is stable. Trajectories converge to the singular point,

14

Nonlinear and adaptive control systems

spiralling around. In time domain, the solutions are similar to decayed sinusoidal
functions.
x2

x1

Stable focus

Unstable focus, for 1,2 = j, ( > 0). The real part is positive, and therefore
the singular point is unstable, with the trajectories spiralling out from the singular
point.
x2

x1

Unstable focus

Centre, for 1,2 = j, ( > 0). For the linearised model, when the real part is zero,
the norm of the state is constant. In phase portrait, there are closed orbits around the
singular point.
x2

x1

Centre

To draw a phase portrait, the analysis of singular points is the rst step. Based
on the classication of the singular points, the behaviours in neighbourhoods of these
points are more or less determined. For other regions, we can calculate the directions
of the movement from the directives.
At any point, the slope of trajectory can be computed by
f2 (x1 , x2 )
dx2
.
=
dx1
f1 (x1 , x2 )

State space models

15

With enough points in the plane, we should be able to sketch phase portraits connecting the points in the directions determined by the slopes.
Indeed, we can even obtain curves with constant slopes, which are named as
isoclines. An isocline is a curve on which (f2 (x1 , x2 )/f1 (x1 , x2 )) is constant. This, again,
can be useful in sketching a phase portrait for a second-order nonlinear system.
It should be noted that modern computer simulation can provide accurate solutions to many nonlinear differential equations. For this reason, we will not go to
further details of drawing phase portraits based on calculating slope of trajectories
and isoclines.
Example 2.1. Consider a second-order nonlinear system
x 1 = x2
x 2 = x2 2x1 + x12 .
Setting
0 = x2 ,
0 = x2 2x1 + x12 ,
we obtain two singular points (0, 0) and (2, 0).
Linearised system matrix for the singular point (0, 0) is obtained as


0
1
A=
,
1 2
and the eigenvalues are obtained as 1 = 2 = 1. Hence, this singular point is a
stable node.
For the singular point (2, 0), the linearised system matrix is obtained as


0
1
A=
,
2 1
and the eigenvalues are 1 = 2 and 2 = 1. Hence, this singular point is a saddle
point. It is useful to obtain the corresponding eigenvectors to determine which direction is converging and which is diverging. The eigenvalues v1 and v2 , for 1 = 2
and 2 = 1, are obtained as


 
1
1
v1 =
, v2 =
2
1
This suggests that along the direction of v1 , relative to the singular point, the state
converges to the singular point, while along v2 , the state diverges. We can clearly
see from Figure 2.1 that there is a stable region near the singular point (0, 0).
However, in the neighbourhood of (2, 1) one part of it is stable, and the other part
is unstable.


16

Nonlinear and adaptive control systems

1
0.5
0
0.5

x2

1
1.5
2
2.5
3
3.5
4
4

x1

Figure 2.1 Phase portrait of Example 2.1

Example 2.2. Consider the swing equation of a synchronous machine

H = Pm Pe sin (),

(2.7)

where H is the inertia, is the rotor angle, Pm is the mechanical power and Pe is the
maximum electrical power generated. We may view Pm as the input and Pe sin as
the output. For the convenience of presentation, we take H = 1, Pm = 1 and Pe = 2.
The state space model is obtained by letting x1 = and x2 = as
x 1 = x2
x 2 = 1 2 sin (x1 ).
For the singular points, we obtain
5
1
or ,
6
6
= 0.

x1e =
x2e

Note that there are an innite number of singular points, as x1e = 2k + 16 and
x1e = 2k + 56 are also solutions for any integer value of k.

State space models

17

Let us concentrate on the analysis of the two singular points ( 16 , 0) and ( 56 , 0).
The linearised system matrix is obtained as

A=


1
.
0

0
2 cos (x1e )

For ( 61 , 0), the eigenvalues are 1,2 = 31/4 j, and therefore this singular point it a
centre.
For ( 56 , 0), the eigenvalues are 1 = 31/4 and 2 = 31/4 . Hence, this singular
point it a saddle point. The eigenvalues v1 and v2 , for 1 = 31/4 and 2 = 31/4 , are
obtained as

v1 =


1
,
31/4


v2 =


.

31/4

x2

3
2

x1

Figure 2.2 Phase portrait of Example 2.2

Figure 2.2 shows a phase portrait obtained from computer simulation. The centre at
( 16 , 0) and the saddle point at ( 56 , 0) are clearly shown in the gure. The directions
of the ow can be determined from the eigenvectors of the saddle point. For example
the trajectories start from the points around (5, 3) and move upwards and to the left,
along the direction pointed by the eigenvector v1 towards the saddle point. Along the
direction pointed by v2 , the trajectories depart from the saddle point.


18

Nonlinear and adaptive control systems

2.4 Limit cycles and strange attractors

Some trajectories appear as closed curves in phase portrait. For autonomous systems,
they represent periodic solutions, as the solutions only depend on the states, and they
are referred to as cycles. For linear systems, periodic solutions appear in harmonic
oscillators. For example a harmonic oscillator described by
x 1 = x2
x 2 = x1
has solutions as sinusoidal functions, and the amplitude is determined by the initial
values. It is easy to see that the function V = x12 + x22 remains a constant, as we have
V = 2x1 (x2 ) + 2x2 x1 = 0.
Therefore, the solutions of this oscillator are circles around the origin. For a given
initial state, the radius does not change. When two initial points are very close, their
solutions will be very close, but they will not converge to one cycle, no matter how
close the initial values. Also the solution for this harmonic oscillator is not robust, as
any small disturbance will destroy the cycle.
There are cycles in the phase portrait shown in Figure 2.2. Even though they are
solutions to a nonlinear dynamic system, they are similar to the cycles obtained from
the harmonic oscillator in the sense that cycles depend on the initial values and they
do not converge or attract to each other, no matter how close the two cycles are.
The cycles discussed above are not limit cycles. For limit cycles we have the
following denition.
Denition 2.4. A closed curve solution, or in other word, a cycle, of an autonomous
system is a limit cycle, if some non-periodic solutions converge to the cycle as t
or t .
A limit cycle is stable if nearby trajectories converge to it asymptotically, unstable
if move away. One property of a limit cycle is that amplitude of the oscillation may
not depend on the initial values. A limit cycle may be attractive to the nearby region.
One of the most famous ones is van der Pol oscillator. This oscillator does have a
physical meaning. It can be viewed as a mathematical model of an RLC circuit, with
the resistor being possible to take negative values in certain regions.
Example 2.3. One form of van der Pol oscillator is described in the following
differential equation:
y (1 y2 )y + y = 0,
where is a positive real constant.
If we take x1 = y and x2 = y , we obtain the state space equation
x 1 = x2
x 2 = x1 + (1 x12 )x2 .

(2.8)

State space models

19

From this state space realisation, it can be seen that when = 0, van der Pol oscillator
is the same as a harmonic oscillator. For with small values, one would expect that
it behaves like a harmonic oscillator.
A more revealing state transformation for with big values is given by
x1 = y
1
x2 = y + f (y),

where f (y) = y3 /3 y. Under the above transformation, we have the system as
x 1 = (x2 f (x1 ))
x 2 = 1 x1 .

(2.9)

It can be obtained that

x1
dx2
(x2 f (x1 )) = 2
dx1

(2.10)

2
= 0 or x2 = f (x1 ). This can be
This equation suggests that as , we have dx
dx1
seen from the phase portrait for very big values of in Figure 2.4.
Let us stick with the state space model (2.9). The only singular point is at the
origin (0, 0). The linearised system matrix at the origin is obtained as


A=


1
.
0

From the eigenvalues of A, we can see that this singular point is either an unstable
node or an unstable focus, depending on the value of . Phase portrait of van der
Pol oscillator with = 1 is shown in Figure 2.3 for two trajectories, one with initial
condition outside the limit cycle and one from inside. The broken line shows x2 =
f (x1 ). Figure 2.4 shows the phase portrait with = 10. It is clear from Figure 2.4 that
the trajectory sticks with the line x2 = f (x1 ) along the outside and then moves almost
horizontally to the other side, as predicted in the analysis earlier.

Limit cycles also exist in high-order nonlinear systems. As seen later in Chapter 11, circadian rhythms can also be modelled as limit cycles of nonlinear dynamic
systems. For second-order autonomous systems, limit cycles are very typical trajectories. The following theorem, PoincareBendixson theorem, describes the features
of trajectories of the second-order systems, from which a condition on the existence
of a limit cycle can be drawn.
Theorem 2.1. If a trajectory of the second-order autonomous system remains in a
nite region, then one of the following is true:

Nonlinear and adaptive control systems

3

x2

3
2.5

1.5

0.5

0
x1

0.5

1.5

2.5

Figure 2.3 Phase portrait of van der Pol oscillator with = 1

x2

20

3
2.5

1.5

0.5

0
x1

0.5

1.5

Figure 2.4 Phase portrait of van der Pol oscillator with = 10

2.5

State space models

21

The trajectory goes to an equilibrium point.

The trajectory tends to an asymptotically stable limit cycle.
The trajectory is itself a cycle.

For high-order nonlinear systems, there are more complicated features if the
trajectories remain in a bounded region. For the asymptotic behaviours of dynamic
systems, we dene positive limit sets.
Denition 2.5. Positive limit set of a trajectory is the set of all the points for which
the trajectory converges to, as t .
Positive limit sets are also referred to as -limit sets, as is the last letter of
Greek letters. Similarly, we can dene negative limit sets, and they are called -limit
sets accordingly. Stable limit cycles are positive limit sets, so do stable equilibrium
points. The dimension for -limit sets is zero or one, depending on singular points or
limit cycles.
Strange limit sets are those limit sets which may or may not be asymptotically
attractive to the neighbouring trajectories. The trajectories they contain may be locally
divergent from each other, within the attracting set. Their dimensions might be fractional. Such structures are associated with the quasi-random behaviour of solutions
called chaos.
Example 2.4. The Lorenz attractor. This is one of the most widely studied examples
of strange behaviour in ordinary differential equations, which is originated from
studies of turbulent convection by Lorenz. The equation is in the form
x 1 = (x2 x1 )
x 2 = (1 + x3 )x1 x2

(2.11)

x 3 = x1 x2 bx3 ,
where
, and b are
positive
constants. There are three equilibrium points (0, 0, 0),

( b, b, ) and ( b, b, ). The linearised system matrix around the origin
is obtained as

A = +1
0

1
0

0
0 ,
b

and its eigenvalues are obtained as 1,2 = ( 1) ( 1)2 + 4 /2 and

3 = b. Since the rst eigenvalue is positive, this equilibrium is unstable. It can
be shown that the other equilibrium points are unstable when the parameters satises
> b + 1,
( + 1)( + b + 1)
.
>
b1

Nonlinear and adaptive control systems

35
30
25

x3

20
15
10
5
0
20
10
0
x2

10
20

10

15

10

15

x1

Figure 2.5 A typical trajectory of Lorenz attractor

20
15
10
5
x2

22

0
5
10
15
20
15

10

0
x1

Figure 2.6 Projection to (x1 , x2 )

10

15

State space models

35

30

25

x3

20

15

10

0
15

10

0
x1

10

15

Figure 2.7 Projection to (x1 , x3 )

15

10

x1

10

15

20

40

60
Time (s)

Figure 2.8 Time trajectory of x1

80

100

23

24

Nonlinear and adaptive control systems

It can be established that the trajectories converge to a bounded region specied by

( + 1)x12 + x22 + (x3 2( + 1))2 C
for a positive constant C. When all the three equilibria are unstable, the behaviour of
Lorenz system is chaotic. A trajectory is plotted in Figure 2.5 for b = 2, = 10 and
= 20. Projections to (x1 , x2 ) and (x1 , x3 ) are shown in Figures 2.6 and 2.7. The time

trajectory of x1 is shown in Figure 2.8.

Chapter 3

Describing functions

In classical control, frequency response is a powerful tool for analysis and control
design of linear dynamic systems. It provides graphical presentation of system dynamics and often can reect certain physical features of engineering systems. The basic
concept of frequency response is that for a linear system, if the input is a sinusoidal
function, the steady-state response will still be a sinusoidal function, but with a
different amplitude and a different phase. The ratio of the input and output amplitudes and the difference in the phase angles are determined by the system dynamics.
When there is a nonlinear element in a control loop, frequency response methods
cannot be directly applied. When a nonlinear element is a static component, i.e., the
input and output relationship can be described by an algebraic function, its output
to any periodic function will be a periodic function, with the same period as the
input signal. Hence, the output of a static nonlinear element is a periodic function
when the input is a sinusoidal function. It is well known that any periodic function with piece-wise continuity has its Fourier series which consists of sinusoidal
functions with the same period or frequency as the input with a constant bias, and
other sinusoidal functions with high multiple frequencies. If we take the term with
the fundamental frequency, i.e., the same frequency as the input, as an approximation, the performance of the entire dynamic system may be analysed using frequency
response techniques. Describing functions are the frequency response functions of
nonlinear components with their fundamental frequency terms as their approximate
outputs. In this sense, describing functions are rst-order approximation in frequency
domain. It can also be viewed as a linearisation method in frequency domain for
nonlinear components.
Describing function analysis remains as an important tool for analysis of nonlinear systems with static components despite several more recent developments in
nonlinear control and design. It is relatively easy to use, and closely related to frequency response analysis of linear systems. It is often used to predict the existence
of limit cycles in a nonlinear system, and it can also be used for prediction of subharmonics and jump phenomena of nonlinear systems. In this chapter, we will present
basic concept of describing functions, calculation of describing functions of common
nonlinear elements and how to use describing functions to predict the existence of
limit cycles.

26

Nonlinear and adaptive control systems

3.1 Fundamentals
For a nonlinear component described by a nonlinear function f : R R, its
A sin(wt)
f (x)

w(t)

output
w(t) = f (A sin(t))
to a sinusoidal input A sin (t) is a periodical function, although it may not be sinusoidal in general. Assuming that the function f is piecewise-continuous, w(t) is a
piecewise-continuous periodic function with the same period as the input signal. A
piecewise periodical function can be expanded in Fourier series
a0 
(an cos(nt) + bn sin(nt)),
+
2
n=1

w(t) =
where

1
a0 =

an =

bn =

(3.1)

w(t)d(t)
w(t) cos(nt)d(t)
w(t) sin(nt)d(t).

Remark 3.1. For a piecewise-continuous function w(t), the Fourier series on the
right-hand side of (3.1) converges to w(t) at any continuous point, and to the average
of two values obtained by taking limits from both sides at a dis-continuous point. If
we truncate the series up to order k,
a0 
wk (t) =
(an cos(nt) + bn sin(nt)),
+
2
n=1
k

where wk is the best approximation in least squares, i.e., in L2 .

Taking the approximation to the rst order, we have

w1 =

a0
+ a1 cos(t) + b1 sin(t).
2

If a0 = 0, which can be guaranteed by setting the nonlinear function f to an odd

function, we have the approximation
w1 = a1 cos(t) + b1 sin(t)

(3.2)

which is an approximation at the fundamental frequency. The above discussion shows

that for a nonlinear component described by the nonlinear function f , the approximation at the fundamental frequency, i.e., the frequency of the input signal, to an input

Describing functions

27

signal A sin(t), is a sinusoidal function in (3.2) with the Fourier coefcients a1 and
b1 shown in (3.1). Hence, we can analyse the frequency response of this nonlinear
component.
We can rewrite w1 in (3.2) as
w1 = M sin(t + ),
where
M (A, ) =

(3.3)


a21 + b21 ,

(A, ) = arctan(a1 /b1 ).

In complex expression, we have
w1 = Mej(t+) = (b1 + ja1 )ejt .
The describing function is dened, similar to frequency response, as the complex ratio
of the fundamental component of the nonlinear element against the input by
N (A, ) =

Mejt+
b1 + ja1
=
.
Aejt
A

(3.4)

Remark 3.2. A clear difference between the describing function of a nonlinear element and the frequency response of a linear system is that the describing function
depends on the input amplitude. This reects the nonlinear nature of the describing
function.

Remark 3.3. If f is a single-valued odd function, i.e., f (x) = f (x), we have

1
a1 =
f (A sin(t)) cos(t)d(t)



1
1 0
f (A sin(t) cos(t)d(t) +
f (A sin(t) cos(t)d(t)
=

0


1
1
f (A sin(t) cos(t)d(t) +
f (A sin(t) cos(t)d(t)
=
0
0
= 0.
If a1 = 0, the describing function is a real value.
Example 3.1. The characteristics of a hardening spring are given by
x3
.
2
Given the input A sin(t), the output is
f (x) = x +

w(t) = f (A sin(t))
= A sin(t) +

A3
sin3 (t).
2

28

Nonlinear and adaptive control systems

Since f is an odd function, we have a1 = 0. The coefcient b1 is given by


 
A3
1
b1 =
A sin(t) +
sin3 (t) sin(t)d(t)

2

 /2 
4
A3
2
4
=
A sin (t) +
sin (t) d(t).
0
2
Using the integral identity

 /2
n 1 /2 n2
sinn (t)d(t) =
sin (t)d(t)
n
0
0

for n > 2,

we have
3
b1 = A + A3 .
8
Therefore, the describing function is
N (A, ) = N (A) =

3
b1
= 1 + A2 .
A
8

Alternatively, we can also use the identity

sin(3t) = 3 sin(t) 4 sin3 (t)
to obtain
A3
sin3 (t)
2


1
A3 3
sin(t) sin(3t)
= A sin(t) +
2 4
4


3 3
1 3
= A + A sin(t) A sin(3t).
8
8

w(t) = A sin(t) +

Hence, we obtain b1 = A + 38 A3 from the rst term.

Through the above discussion, describing functions are well dened for nonlinear
components whose inputoutput relationship can be well dened by piecewisecontinuous functions. These functions are time-invariant, i.e., the properties of
nonlinear elements do not vary with time. This is in line with the assumption for frequency response analysis, which can only be applied to time-invariant linear systems.
We treat describing functions as the approximations at the fundamental frequencies,
and therefore in our analysis, we require a0 = 0 which is guaranteed by odd functions
for the nonlinear components. With the describing function of a nonlinear component, we can then apply analysis in frequency responses for the entire system. For the
convenience of this kind of analysis, we often assume that the nonlinear component
for which the describing function is used to approximate its behaviours is the only

Describing functions
r +

f (x)

29

G(s)

Figure 3.1 Block diagram for describing function analysis

nonlinear component in the system, as shown in Figure 3.1. Hence, in the remaining part of this chapter, we use the following assumptions for describing function
analysis:

There is only a single nonlinear component in the entire system.

The nonlinear component is time-invariant.
The nonlinearity is odd.

3.2 Describing functions for common nonlinear components

In this section, we will calculate the describing functions of common nonlinear
elements in a number of examples.
Example 3.2. Saturation. A saturation function shown in Figure 3.2 is described by

f (x) =

kx,

for |x| < a,

sign(x)ka,

otherwise.

(3.5)

The output to the input A sin(t), for A > a, is symmetric over quarters of a period,
and in the rst quarter,
f(x)

a
a

Figure 3.2 Saturation

w(t) =

kA sin(t),

0 t ,

ka,

< t /2,

(3.6)

where = sin1 (a/A). The function is odd, hence we have a1 = 0, and the symmetry
of w1 (t) implies that

30

Nonlinear and adaptive control systems

4
b1 =

/2

w1 sin(t)d(t)
0

kA sin2 (t)d(t) +

/2

ka sin(t)d(t)



1
4ka
2kA
sin(2 ) +
cos( )
=

 4ka
2kA

a
=
cos( ) +
cos( )


2kA

a
=
+ cos( )

A



2kA
a
a2
=
+
1 2 .

A
A
Note that we have used sin =
describing function is given by
2k
b1
=
N (A) =
A


sin

a
A

and cos =


1



a
a2
a
1 2 .
+
A A
A

a2
.
A2

Therefore, the

(3.7)


Example 3.3. Ideal relay. The output from the ideal relay shown in Figure 3.3
(signum function) is described by, with M > 0,

M , t < 0,
w(t) =
(3.8)
M,
0 t < .
It is again an odd function, hence we have a1 = 0. The coefcient b1 is given by
b1 =


0

M sin(t)d(t) =

4M

f(x)
M

x
M

Figure 3.3 Ideal relay

Describing functions

31

f(x)

a
a

Figure 3.4 Dead zone

and therefore the describing function is given by
N (A) =

4M
.
A

(3.9)


Example 3.4. Dead zone. A dead zone is a complement to saturation. A dead zone
shown in Figure 3.4 can be described by a nonlinear function

k(x a), for x > a,

for |x| < a,

f (x) = 0,

k(x + a), for x < a.

(3.10)

The output to the input A sin(t), for A > a, is symmetric over quarters of a
period, and in the rst quarter,

0,
0 t ,
w(x) =
(3.11)
k(A sin(t) a), < t /2,
where = sin1 (a/A). The function is odd, hence we have a1 = 0, and the symmetry
of w(t) implies that
4
b1 =

/2

w(t) sin(t)d(t)
0

/2

k(A sin(t) a)d(t)



 1
4ka
2kA

+ sin(2 )
cos( )

2
2


2kA

a
= kA
+ cos( )

A



2kA
a2
a
= kA
1 2 .
+

A
A

32

Nonlinear and adaptive control systems

Similar to the calculation

 of the describing function for saturation, we have used
2
a
sin = A and cos = 1 Aa 2 . The describing function for a dead zone is given by
b1
2k
N (A) =
=k
A


sin1



a
a2
a
1 2 .
+
A A
A

(3.12)


Remark 3.4. The dead-zone function shown in (3.10) complements the saturation
function shown in (3.5) in the sense that if we use fs and fd to denote the saturation
function and dead-zone function, we have fs + fd = k for the describing functions
shown in (3.7) and (3.12), the same relationship holds.

Example 3.5. Relay with hysteresis. Consider a case when there is a delay in the
ideal relay as shown in Figure 3.5. The nonlinear function for relay with hysteresis
can be described by

M,
for x a,

M , for |x| < a, x > 0,

f (x) =
for |x| < a, x < 0,
M,

M , for x a.

(3.13)

When this nonlinear component takes A sin (t) as the input with A > a, the
output w(t) is given by

for t < ( ),
M,
w(t) = M , for ( ) t < ,

M,
for t < ,

(3.14)

where = sin1 ( Aa ). In this case, we still have a0 = 0, but not a1 . For a1 we have

f(x)
M

x
M

Figure 3.5 Relay with hysteresis

Describing functions
a1 =

( )

1
+

M cos(t)d(t) +

( )

33

M cos(t)d(t)

M cos(t)d(t)

4M
sin( )

4M a
=
.
A

Similarly, we have
b1 =

( )

1
+

M sin(t)d(t) +

( )

M sin(t)d(t)

M sin(t)d(t)

4M
cos( )

4M
a2
=
1 2.

From
N (A, ) =

b1 + ja1
,
A

we have
4M
N (A) =
A


a2
a
1 2 j
.
A
A

(3.15)

Using the identity cos( ) + j sin( ) = ejr , we can rewrite the describing function as
N (A) =

4M j arcsin(a/A)
e
.
A


Remark 3.5. Comparing the describing function of the relay with hysteresis with
that of ideal relay in (3.9), the describing functions indicate that there is a delay in the
relay with hysteresis by arcsin(a/A) in terms of phase angle. There is indeed a delay
of = arcsin(a/A) in the time response w(t) shown in (3.14) with that of the ideal
relay. In fact, we could use this fact to obtain the describing function for the relay
with hysteresis.


34

Nonlinear and adaptive control systems

3.3 Describing function analysis of nonlinear systems

One of the most important applications of describing functions is to predict the existence of a limit cycle in a closed-loop system that contains a nonlinear component
with a linear transfer function, as shown in Figure 3.1. Consider a system with a linear transfer function G(s) and a nonlinear element with describing function N (A, )
in the forward path, under unit feedback. The inputoutput relations of the system
component by setting r = 0 can be described by
w = N (A, )x
y = G(j)w
x = y,
with y as the output and x as the input to the nonlinear component. From the above
equations, it can be obtained that
y = G(j)N (A, )(y),
and it can be arranged as
(G(j)N (A, ) + 1)y = 0.
If there exists a limit cycle, then y  = 0, which implies that
G(j)N (A, ) + 1 = 0,

(3.16)

or
G(j) =

1
.
N (A, )

(3.17)

Therefore, the amplitude A and frequency of the limit cycle must satisfy the above
equation. Equation (3.17) is difcult to solve in general. Graphic solutions can be
found by plotting G(j) and 1/N (A, ) on the same graph to see if they intersect
each other. The intersection points are the solutions, from which the amplitude and
frequency of the oscillation can be obtained.
Remark 3.6. The above discussion is based on the assumption that the oscillation,
or limit cycle, can be well approximated by a sinusoidal function, and the nonlinear
component is well approximated by its describing function. The describing function
analysis is an approximate method in nature.

Only a stable limit cycle may exist in real applications. When we say stable limit
cycle, we mean that if the state deviates a little from the limit cycle, it should come
back. With the amplitude as an example, if A is perturbed from its steady condition,
say with a very small increase in the amplitude, for a stable limit cycle, the system
will decay to its steady condition.

Describing functions

35

As describing functions are rst-order approximations in the frequency domain,

stability criteria in the frequency domain may be used for the stability analysis of
limit cycles. Nyquist criterion can be extended to give the conditions for stability of
limit cycles.
Recall the case for a linear system with the forward transfer function G(s) with
unit feedback. The characteristic equation is given by
G(s) + 1 = 0, or G(s) = 1.
The Nyquist criterion determines stability of the closed-loop system from the number
of encirclements of the Nyquist plot around point 1, or (1, 0) in the complex plain.
In the case that there is a control gain K in the forward transfer function, the
characteristic equation is given by
KG(s) + 1 = 0, or G(s) =

1
.
K

In this case, the Nyquist criterion can be extended to determine the stability of the
closed loop by counting the encirclements of the Nyquist plot around (1/K, 0) in the
complex plain in the same way as around (1, 0). The Nyquist criterion for non-unity
forward path gain K is also referred to as the extended Nyquist criterion. The same
argument holds when k is a complex number.
We can apply the extended Nyquist criterion to determine the stability of a limit
cycle. When the condition specied in (3.17) is satised for some (A0 , 0 ), A0 and
0 are the amplitude and frequency of the limit cycle respectively, and N (A0 , 0 )
is a complex number. We can use the extended Nyquist criterion to determine the
stability of the limit cycle with the amplitude A0 and frequency 0 by considering a
perturbation of A around A0 .
To simplify our discussion, let us assume that G(s) is stable and minimum phase.
It is known from the Nyquist criterion that the closed-loop system with constant
gain K is stable if the Nyquist plot does not encircle (1/K, 0). Let us consider a
perturbation in A to A+ with A+ > A0 . In such a case, 1/N (A+ , 0 ) is a complex
number in general. If the Nyquist plot does not encircle the point 1/N (A+ , 0 ), we
conclude that the closed-loop system is stable with the complex gain 1/N (A+ , 0 ).
Therefore, in a stable closed-loop system, the oscillation amplitude decays, which
makes A+ return to A0 . This implies that the limit cycle (A0 , 0 ) is stable. Alternatively,
if the Nyquist plot encircles the point 1/N (A+ , 0 ), we conclude that the closedloop system is unstable with the complex gain 1/N (A+ , 0 ). In such a case, the
oscillation amplitude may grow even further, and does not return to A0 . Therefore,
the limit cycle is unstable.
Similar arguments can be made for the perturbation to a smaller amplitude. For
an A < A0 , if the Nyquist plot does encircle the point 1/N (A , 0 ), the limit cycle
is stable. If the Nyquist plot does not encircle the point 1/N (A , 0 ), the limit cycle
is unstable.
When we plot 1/N (A, 0 ) in the complex plane with A as a variable, we obtain
a line with direction of the increment of A. Based on the discussion above, the way

36

Nonlinear and adaptive control systems

Im
G( jw )
Re
A
A+

Stable

N(A,w 0)

Figure 3.6 Digram for stable limit cycle

Im
G( jw )
Re
A+
A
Unstable
1

N(A,w 0)

Figure 3.7 Digram for unstable limit cycle

of the line for 1/N (A, 0 ) intersects with the Nyquist plot determines the stability
of the limit cycle. Typical Nyquist plots of stable minimum-phase systems are shown
in Figures 3.6 and 3.7 for stable and unstable limit cycles with nonlinear elements
respectively.
We can summarise the above discussion for the stability criterion of limit cycles
using describing function.
Theorem 3.1. Consider a unity-feedback system with the forward path with stable
minimum phase transfer function G(s) and a nonlinear component with the describing
function N (A, ), and suppose that the plots, 1/N and G(j) intersect at the point
with A = A0 and = 0 . The limit cycle at (A0 , 0 ) is stable if the plot of 1/N (A, 0 )
crosses the Nyquist plot from the inside of the encirclement to the outside of the
encirclement as A increases. The limit cycle at (A0 , 0 ) is unstable if the plot of
1/N (A, 0 ) crosses the Nyquist plot from the outside of the encirclement to the
inside of the encirclement as A increases.
Remark 3.7. Theorem 3.1 requires the transfer function to be stable and minimum
phase, for the simplicity of the presentation. This theorem can be easily extended to

Describing functions

37

the case when G(s) is unstable or has unstable zeros by using corresponding stability
conditions based on the Nyquist criterion. For example if G(s) is stable and has one
unstable zero, then the stability criterion for the limit cycle will be opposite to the
condition stated in the theorem, i.e., the limit cycle is stable if the plot of 1/N (A, 0 )
crosses the Nyquist plot from the outside of the encirclement to the inside of the
encirclement as A increases.

K
with K a
s(s + 1)(s + 2)
positive constant and an ideal relay in a closed loop, as shown in Figure 3.8. We will
determine if there exists a limit cycle and analyse the stability of the limit cycle.

Example 3.6. Consider a linear transfer function G(s) =

K
s(s + 1)(s + 2)

Figure 3.8 Closed-loop system for Example 3.6

For the ideal relay, we have N =
obtain that
G(j) =

4M
. For the transfer function, we can
A

K
j(j + 1)(j + 2)

=K

32 j(2 2 )
.
(32 )2 + 2 (2 2 )2

From
G(j) =

1
,
N

we obtain two equations for real and imaginary parts respectively as

(G(j)) = 0,
A
.
4M
From the equation of the imaginary part, we have
(G(j)) =

(2 2 )
= 0,
(32 )2 + 2 (2 2 )2

which gives = 2. From the equation of the real part, we have

K

A
32
,
=
2
2
2
2
2
(3 ) + (2 )
4M

which gives A = 2KM /3.

38

Nonlinear and adaptive control systems

Hence, we have shown

that there exists a limit cycle with amplitude and frequency
at (A, ) = (2KM /3 , 2).
The plot of (1/N (A)) = ( A/4M ) overlaps with the negative side of the real
axis. As A increases from 0, (1/N (A)) moves from the origin towards left. Therefore,
as A increases, (1/N (A)) moves from inside of the encirclement of the Nyquist plot
to outside of the encirclement, and the limit cycle is stable, based on Theorem 3.1.
A simulation result for K = M = 1 is shown in Figure 3.9 with the amplitude
A = 0.22 and period T = 4.5 s, not far from the values A = 0.2212 and T = 4.4429,
predicted from the describing function analysis.

Example 3.7. In this example, we consider a van der Pol oscillator described by
y + (3y2 1)y + y = 0.

(3.18)

We will use describing function analysis to predict the existence of a limit cycle, and
compare the predicted amplitudes and periods for different  values with the simulated
ones.
To use the describing analysis, we need to formulate the system in the format of
one linear transfer function and a nonlinear element. Rearranging (3.18), we have
y  y + y = 

d 3
y .
dt

0.25
0.2
0.15
0.1

0.05
0
0.05
0.1
0.15
0.2
0.25
34

36

38

40

42

44

Time (s)

Figure 3.9 The simulated output for Example 3.6

46

Describing functions

39

Hence, the system (3.18) can be described by a closed-loop system with a nonlinear
component
f (x) = x3
and a linear transfer function
G(s) =

s
.
s2 s + 1

Using the identity

sin(3t) = 3 sin(t) 4 sin3 (t)
in a similar way as in Example 3.1, we obtain the describing function for
f (x) = x3 as
N (A) =

3 2
A.
4

Setting
G(j) = 

1
N


= 0,

we have
(1 2 )
= 0,
(1 2 )2 +  2 2
which gives = 1. From the equation for the real part, we obtain
4
 2 2
= 2
(1 2 )2 +  2 2
3A

which gives A = 2 3/3.

The linear part of the transfer function has one unstable pole. We need to take
this into consideration for the stability of the limit cycle. As A increases, 1/N (A)
moves from the left to the right along the negative part of the real axis, basically
from the outside of the encirclement of the Nyquist plot to the inside of the encirclement. This suggests that the limit cycle is stable, as there is an unstable pole in
the linear transfer function. The simulation results for  = 1 and  = 30 are shown in
Figures 3.10 and 3.11. In both cases, the amplitudes are very close to the predicted
one from the describing function analysis. For the period, the simulation result for
 = 1 in Figure 3.10 is very close to 2, but the period for  = 30 is much better than
2 . This suggests that the describing function analysis gives a better approximation
for the case of  = 1 than  = 30. In fact, for a small value of  the oscillation is
very similar to a sinusoidal function. With a big value of , the wave form is very
different from a sinusoidal function, and therefore the describing function method
cannot provide a good approximation.


Nonlinear and adaptive control systems

0.5

0.5

10

15

20

25

30

35

Time (s)

Figure 3.10 The simulated output for Example 3.7 with  = 1

0.5

40

0.5

30

40

50

60

70

80

90

100

110

Figure 3.11 The simulated output for Example 3.7 with  = 30

120

Chapter 4

Stability theory

For control systems, design, one important objective is to ensure the stability of the
closed-loop system. For a linear system, the stability can be evaluated in time domain
or frequency domain, by checking the eigenvalues of the system matrix or the poles
of the transfer function. For nonlinear systems, the dynamics of the system cannot be
described by equations in linear state space or transfer functions in general. We need
more general denitions about the stability of nonlinear systems. In this chapter, we
will introduce basic concepts of stability theorems based on Lyapunov functions.

4.1 Basic denitions

Consider a nonlinear system
x = f (x),

(4.1)

where x D R is the state of the system, and f : D R R is a continuous

function, with x = 0 as an equilibrium point, that is f (0) = 0, and with x = 0 as an
interior point of D. Here we use D to denote a domain around the equilibrium x = 0.
This domain can be interpreted as a set with 0 as its interior point, or it can also
be simplied as D = {x|x < r} for some positive r. In the remaining part of this
chapter, we will use D in this way.
When we say the stability of the system, we refer to the behaviour of the system
around the equilibrium point. Here we assume that x = 0 is an equilibrium point
without loss of generality. In case that the system has an equilibrium point at x0 , we
can always dene a state transformation with x x0 as the new state, to shift the
equilibrium point to the origin.
The system in (4.1) is referred to as an autonomous system, as it does not depend
on the signals other than the system state. For nonlinear control systems, we can write
n

x = f (x, u),

(4.2)

where u R is the control input. With u as an external signal, the system (4.2) is
not autonomous. However, for such a system, if we design a feedback control law
u = g(x) with g : Rn Rm as a continuous function, the closed-loop system
m

x = f (x, g(x))
becomes an autonomous system.

42

Nonlinear and adaptive control systems

In this chapter, we will present basic denitions and results for stability of
autonomous systems. As discussed above, control systems can be converted to
autonomous systems by state feedback control laws.
There are many different denitions of stability for dynamics systems. Often
different denitions are needed for different purposes, and many of them are actually
the same when the system is linear. Among different denitions, the most fundamental
one is the Lyapunov stability.
Denition 4.1 (Lyapunov stability). For the system (4.1), the equilibrium point x = 0
is said to be Lyapunov stable if for any given positive real number R, there exists a
positive real number r to ensure that x(t) < R for all t 0 if x(0) < r. Otherwise
the equilibrium point is unstable.
x(t)
r

x(0)

Figure 4.1 Lyapunov stability

The denition of Lyapunov stability concerns with the behaviours of a dynamic
system with respect to the initial state. If a system is Lyapunov stable, we can impose
a restriction on the initial state of the system to make sure that the state variables stay
in a certain region. For the two positive numbers in the denition, R and r, the denition did not explicitly require R r. However, if we set r > R, from the continuity
of the solution, we cannot ensure x(t) < R for t close to 0, because of x(0) > R.
Therefore, when using this denition, we need r R.
Example 4.1. Consider a linear system
x = Ax,

x(0) = x0 ,

where

A=

with > 0. For this linear system, we can explicitly solve the differential equation
to obtain

x(t) =

cos t
sin t


sin t
x.
cos t 0

Stability theory

43

It is easy to check that we have x(t) = x0 . Hence, to ensure that x(t) R, we

only need to set r = R, i.e., if x0  R, we have x(t) R for all t > 0.
Note that for the system in Example 4.1, the system matrix has two eigenvalues
on the imaginary axis, and this kind of systems is referred to as critically stable in
many undergraduate texts. As shown in the example, this system is Lyapunov stable.
It can also be shown that for a linear system, if all the eigenvalues of the system
matrix A are in the closed left half of the complex plane, and the eigenvalues on the
imaginary axis are simple, the system is Lyapunov stable. However, if the system
matrix has multiple poles on the imaginary axis, the system is not Lyapunov stable.
For example let x 1 = x2 , and x 2 = 0 with x1 (0) = x1,0 , x2 (0) = x2,0 . It is easy to obtain
that x1 (t) = x1,0 + x2,0 t and x2 (t) = x2,0 . If we want x(t) R, there does not exist
a positive r for x(0) r to guarantee x(t) R. Therefore, this system is not
Lyapunov stable.
For linear systems, when a system is stable, the solution will converge to the equilibrium point. This is not required by Lyapunov stability. For more general dynamic
systems, we have the following denition concerning with the convergence to the
equilibrium.
Denition 4.2 (Asymptotic stability). For the system (4.1), the equilibrium point
x = 0 is asymptotically stable if it is stable (Lyapunov) and furthermore limt
x(t) = 0.
x(t)
r

x(0)

Figure 4.2 Asymptotic stability

Linear systems with poles in the open left half of the complex plane are asymptotically stable. The asymptotic stability only requires that a solution converges the
equilibrium point, but it does not specify the rate of convergence. In the following
denition, we specify a stability property with an exponential rate of convergence.
Denition 4.3 (Exponential stability). For the system (4.1), the equilibrium point
x = 0 is exponentially stable if there exist two positive real numbers a and such that
the following inequality holds:
x(t) < ax(0)et
for t > 0 in some neighbourhood D Rn containing the equilibrium point.

(4.3)

44

Nonlinear and adaptive control systems

For linear systems, the stability properties are relatively simple. If a linear system
is asymptotically stable, it can be shown that it is exponentially stable. Of course,
for nonlinear systems, we may have a system that is asymptotically stable, but not
exponentially stable.
Example 4.2. Consider a nonlinear system
x(0) = x0 > 0,

x = x3 ,

where x R. Let us solve this differential equation. From the system equation we
have

dx
= dt,
x3

which gives
1
1
= 2t

x2 (t) x02
and
x(t) = 

x0
1 + 2x02 t

It is easy to see that x(t) decreases as t increases, and also limt x(t) = 0. Therefore,
this system is asymptotically stable. However, this system is not exponentially stable,
as there does not exist a pair of a and to satisfy


x0
1+

2x02 t

ax0 e t .

Indeed, if there exist such constants a and , we have



1 + 2x02 te t

1
,
a

which is not satised for any choices of a and , because the left-hand side converges
to zero. Hence, the system considered in this example is asymptotically stable, but
not exponentially stable.

Lyapunov, asymptotic and exponential, stabilities are dened around equilibrium
points. If the properties hold for any initial points in the entire state space, they are

Stability theory

45

referred to as global stability properties. In the following denitions, we give their

global versions.
Denition 4.4 (Globally asymptotic stability). If the asymptotic stability dened in
Denition 4.2 holds for any initial state in Rn , the equilibrium point is said to be
globally asymptotically stable.
Denition 4.5 (Globally exponential stability). If the exponential stability dened
in Denition 4.3 holds for any initial state in Rn , the equilibrium point is said to be
globally exponentially stable.
The stability property discussed in Example 4.2 is globally and asymptotically
stable. In the two examples shown in this section, the stability properties are checked
based on the actual solutions of the systems. In general, explicit solutions of nonlinear
systems are difcult to obtain, and it is expected to check the stability properties of
nonlinear systems without knowing the solutions. In the later part of this chapter, we
will show a number of results to establish stability properties without their solutions.

4.2 Linearisation and local stability

In this section, we introduce a result for checking the stability of nonlinear systems
based on its linearised model.
Theorem 4.1 (Lyapunovs linearisation method). For a linearised model, there are
three cases:

If the linearised system has all the systems poles in the open left half of the complex
plane, the equilibrium point is asymptotically stable for the actual nonlinear
system.
If the linearised system has poles in the open right half of the complex plane, then
the equilibrium point is unstable.
If the linearised system has poles on the imaginary axis, then the stability of the
original system cannot be concluded using the linearised model.

We do not show a proof of this theorem here. It is clear that this theorem can
be applied to check local stabilities of nonlinear systems around equilibrium points.
For the case that the linearised model has poles on the imaginary axis, this theorem
cannot give conclusive result about the stability. This is not a surprise, because stable
and unstable systems can have the same linearised model. For example the systems
x = x3 and x = x3 have the same linearised model at x = 0, that is x = 0, which is
marginally stable. However, as we have seen in Example 4.2, the system x = x3 is
asymptotically stable, and it is not difcult to see that x = x3 is unstable. For both the
stable and unstable cases of linearised models, the linearised model approximates the
original system better when the domain around the equilibrium point gets smaller.
Hence, the linearised model is expected to reect on the stability behaviours around
the equilibrium point.

46

Nonlinear and adaptive control systems

Example 4.3. Consider a nonlinear system

x 1 = x2 + x1 x13 ,
x 2 = x1 .
It can be seen that x = (0, 0) is an equilibrium point of the system. The linearised
model around x = (0, 0) is given by
x = Ax
where
A=

1
1

1
0

The linearised system is unstable as (A) = 12 3j . Indeed, this nonlinear system is a

van der Pols system, and the origin is unstable. Any trajectories that start from initial
point close to the origin and within the limit cycle will spiral out, and converge to the
limit cycle.


4.3 Lyapunovs direct method

Lyapunovs linearisation method can only be used to check local stabilities, and also
there is a limitation in the case of marginal stability. Fortunately, there is a direct
method to check the stability of dynamic systems. This method is based on Lyapunov
functions. We need a few denitions before we can show some of the results on
stability based on Lyapunov functions.
Denition 4.6 (Positive denite function). A function V (x) : D Rn R is said to
be locally positive denite if V (x) > 0 for x D except at x = 0 where V (0) = 0. If
D = Rn , i.e., the above property holds for the entire state space, V (x) is said to be
globally positive denite.
There are many examples of positive denite functions, such as xT Px for P being
a positive denite matrix, or even x.
Denition 4.7 (Lyapunov function). If in D Rn containing the equilibrium point
x = 0, the function V (x) is positive denite and has continuous partial derivatives,
and if its time derivative along any state trajectory of system (4.1) is non-positive,
i.e.,
V (x) 0
then V (x) is a Lyapunov function.

(4.4)

Stability theory

47

Stability analysis based on a Lyapunov function is probably the most commonly

used method to establish the stability of nonlinear dynamic systems. A fundamental
theorem on Lyapunov function is given below.
Theorem 4.2 (Lyapunov theorem for local stability). Consider the system (4.1). If
in D Rn containing the equilibrium point x = 0, there exists a function V (x) :
D Rn R with continuous rst-order derivatives such that

V (x) is positive denite in D

V (x) is non-positive denite in D

then the equilibrium point x = 0 is stable. Furthermore, if V (x) is negative denite,

i.e., V (x) is positive denite in D, then the stability is asymptotic.
Proof. We need to nd a value for r such that when x(0) < r, we have x(t) < R.
Dene
BR := {x|x R} D,
and let
a = min V (x).
x=R

Since V (x) is positive denite, we have a > 0. We then dene the level set within BR
c := {x BR |V (x) < c},
where c is a positive real constant and c < a. The existence of such a positive real
constant c is guaranteed by the continuity and positive deniteness of V . From the
denition of c , x c implies that x < R. Since V 0, we have V (x(t))
V (x(0)). Hence, for any x(0) c , we have
V (x(t)) V (x(0)) < c,
which implies
x(t) < R.
Since c contains 0 as an interior point, and V is a continuous function, there
must exist a positive real r such that
Br := {x|x < r} c .
Hence, we have
Br c BR .

48

Nonlinear and adaptive control systems

Therefore, for any x(0) Br , we have

V (x(t)) V (x(0)) < c,
and x(t) < R. We have established that if V is non-positive, the system is Lyapunov
stable.
Next, we will establish the asymptotic stability from the negative deniteness of
V . For any initial point in D, V (x(t)) monotonically decreases with time t. Therefore,
there must be a lower limit such that
lim V (x(t)) = 0.

The asymptotic stability can be established if we can show that = 0. We can

prove it by seeking a contradiction. Suppose > 0. Let
= min ( V (x)),
xD 

where  := {x D|V (x) < }. Since V is negative denite, we have > 0. From
the denition of , we have
V (x(t)) V (x(0)) t.
The right-hand side turns to negative when t is big enough, which is a contradiction.
Therefore, we can conclude that limt V (x(t)) = 0, which implies limt x(t) = 0.
2
Example 4.4. A pendulum can be described by
+ + sin = 0,
we re-write the dynamic system as
where is the angle. If we let x1 = and x2 = ,
x 1 = x2
x 2 = sin x1 x2 .
Consider the scalar function
V (x) = (1 cos x1 ) +

x22
.
2

(4.5)

The rst term (1 cos x1 ) in (4.5) can be viewed as the potential energy and the
x2
second term 22 as the kinetic energy. This function is positive denite in the domain
D = {|x1 | , x2 R}. A direct evaluation gives

Stability theory

49

V (x) = sin x1 x 1 + x2 x 2
= x22 .
Hence, the system is stable at x = 0. However, we cannot conclude the asymptotic
stability of the system from Theorem 4.2. This system is in fact asymptotically stable
by using more advanced stability theorem such as invariant set theorem, which is not
covered in this book.

When establishing global stability using Lyapunov functions, we need the function V (x) to be unbounded as x tends to innity. This may sound strange. The reason
behind this point is that we need the property that if V (x) is bounded, then x is
bounded, in order to conclude the boundedness of x from the boundedness of V (x).
This property is dened in the following function as the radial unboundedness of V .
Denition 4.8 (Radially unbounded function). A positive denite function V (x) :
Rn R is said to be radially unbounded if V (x) as x .
Theorem 4.3 (Lyapunov theorem for global stability). For the system (4.1) with D =
Rn , if there exists a function V (x) : Rn R with continuous rst order derivatives
such that

V (x) is positive denite

V (x) is negative denite
V (x) is radially unbounded

then the equilibrium point x = 0 is globally asymptotically stable.

Proof. The proof is similar to the proof of Theorem 4.2, except that for any given
point in Rn , we need to show that there is a level set dened by
c = {x Rn |V (x) < c}
to contain it.
Indeed, since the function V is radially unbounded, for any point in Br with any
positive real r, there exists a positive real constant c such that Br c . It is clear
that the level set c is invariant for any c, that is, for any trajectory that starts in c
remains in c . The rest of the proof follows the same argument as in the proof of
Theorem 4.2.
2
Example 4.5. Consider the nonlinear system
x = x3 ,

x(0) = x0 > 0,

50

Nonlinear and adaptive control systems

where x R. In Example 4.2, we have shown that the equilibrium point x = 0 is

asymptotically stable by checking the solution of the differential equation. In this
example, we use a Lyapunov function.
Let
V =

1 2
x
2

and it is easy to see that this function is globally positive denite. Its derivative is
given by
V = x4
which is negative denite. Hence, from Theorem 4.3, we conclude x = 0 is
asymptotically stable.

To conclude this section, we have another result for exponential stability.
Theorem 4.4 (Exponential stability). For the system (4.1), if there exists a function
V (x) : D Rn R with continuous rst-order derivatives such that
a1 xb V (x) a2 xb ,
V
f (x) a3 xb ,
x

(4.6)
(4.7)

where a1 , a2 , a3 and b are positive real constants, the equilibrium point x = 0 is

exponentially stable. Furthermore, the conditions hold for the entire state space, then
the equilibrium point x = 0 is globally exponentially stable.
The proof of this theorem is relatively simple, and we are going to show it here.
We need a technical lemma, which is also needed later for stability analysis of robust
adaptive control systems.
Lemma 4.5 (Comparison lemma). Let g, V : [0, ) R. Then
V (t) aV (t) + g(t),

t 0

(4.8)

implies that
V (t) eat V (0) +

t
0

for any nite constant a.

e(t ) g( )d ,

t 0

(4.9)

Stability theory

51

Proof. From the derivative of Veat , we have

d
(Veat ) = V eat + aVeat .
dt
Substituting V from (4.8) in the above equation, we have
d
(Veat ) eat g(t).
dt

(4.10)

Integrating (4.10), we have

 t
at
V (t)e V (0) +
ea g( )d.

(4.11)

Multiplying both sides of (4.11) by ea gives (4.8). This completes the proof.

Now we are ready to prove Theorem 4.4.

Proof. From (4.6) and (4.7), we have
a3
V V .
a2
Applying the comparison lemma (Lemma 4.5), we have
V (t) V (0)e(a3 /a2 )t .
Then from (4.6) and (4.12), we have
1/b

1
V (t)
x(t)
a1

1/b
1

V (0)
e(a3 /a2 b)t
a1
 1/b
a2

x(0)e(a3 /a2 b)t .

a1

1/b
and =
Hence, (4.3) is satised with a = aa21
exponentially stable.

(4.12)

a3
,
a2 b

and the equilibrium point is

2

4.4 Lyapunov analysis of linear time-invariant systems

In this section, we will apply Lyapunov stability analysis to linear time-invariant (LTI)
systems. Consider an LTI system
x = Ax

(4.13)

52

Nonlinear and adaptive control systems

where x Rn and A Rnn , x is the state variable, and A is a constant matrix. From
linear system theory, we know that this system is stable if all the eigenvalues of A are
in the open left half of the complex plane. Such a matrix is referred to as a Hurwitz
matrix. Here, we would like to carry out the stability analysis using a Lyapunov
function. We can state the stability in the following theorem.
Theorem 4.6. For the linear system shown in (4.13), the equilibrium x = 0 is globally
and exponentially stable if and only if there exist positive denite matrices P and Q
such that
AT P + PA = Q

(4.14)

holds.
Proof. For sufciency, let
V (x) = xT Px,

(4.15)

and then the direct evaluation gives

V = xT AT Px + xT PAx = xT Qx.

(4.16)

Let us use max () and min () to denote maximum and minimum eigenvalues of a
positive denite matrix. From (4.15), we have
min (P)x2 V (x) max (P)x2 .

(4.17)

From (4.17) and (4.16), we obtain

V min (Q)x2

min (Q)
x2 .
max (P)

(4.18)

Now we can apply Theorem 4.4 with (4.17) and (4.18) to conclude that the equilibrium
point is globally and exponentially stable. Furthermore, we can identify a1 = min (P),
a2 = max (P), a3 = min (Q) and b = 2. Following the proof of Theorem 4.4, we have

x(t)

(Q)
max (P)
min
t
x(0)e 2max (P) .
min (P)

For the necessary part, we have

x(t) ax(0)et

(4.19)

Stability theory

53

for some positive real constants a and , which implies limt x(t) = 0. Since
x(t) = eAt x(0),
we can conclude limt eAt = 0. In such a case, for a positive denite matrix Q, we
can write


d[exp (AT t)Q exp (At)] = Q.

(4.20)

For the left-hand side, we can obtain



d[exp (AT t)Q exp (At)]

= AT


exp (AT t)Q exp (At)t +

exp (AT t)Q exp (At)tA

Let

P=

exp (AT t)Q exp (At)t

and if we can show that P is positive denite, then we obtain (4.14), and hence
complete the proof. Indeed, for any z Rn  = 0, we have

z T Pz =

z T exp (AT t)Q exp (At)zdt.

Since Q is positive denite, and eAt is non-singular for any t, we have z T Pz > 0, and
therefore P is positive denite.
2

Chapter 5

Advanced stability theory

Lyapunov direct method provides a tool to check the stability of a nonlinear system
if a Lyapunov function can be found. For linear systems, a Lyapunov function can
always be constructed if the system is asymptotically stable. In many nonlinear systems, a part of the system may be linear, such as linear systems with memoryless
nonlinear components and linear systems with adaptive control laws. For such a system, a Lyapunov function for the linear part may be very useful in the construction
for the Lyapunov function for the entire nonlinear system. In this chapter, we will
introduce one specic class of linear systems, strict positive real systems, for which,
an important result, KalmanYakubovich lemma, is often used to guarantee a choice
of the Lyapunov function for stability analysis of several types of nonlinear systems.
The application of KalmanYakubovich lemma to analysis of adaptive control systems will be shown in later chapters, while in this chapter, this lemma is used for
stability analysis of systems containing memoryless nonlinear components and the
related circle criterion. In Section 5.3 of this chapter, input-to-state stability (ISS) is
briey introduced.

5.1 Positive real systems

Consider a rst-order system
y = ay + u,
where a > 0 is a constant, and y and u R are the output and input respectively. This
is perhaps the simplest dynamic system we could possibly have. The performance
of such a system is desirable in control design. One of the characteristics for such a
1
is with positive real part if s = + j with
system is that its transfer function s+a
> 0. If such a property holds for other rational transfer functions, they are referred
to as positive real transfer functions. For this, we have the following denition.
Denition 5.1. A rational transfer function G(s) is positive real if

G(s) is real for real s

(G(s)) 0 for s = + j with 0

For analysis of adaptive control systems, strictly positive real systems are more
widely used than the positive real systems.

56

Nonlinear and adaptive control systems

Denition 5.2. A proper rational transfer function G(s) is strictly positive real if
there exists a positive real constant  such that G(s ) is positive real.
Example 5.1. For the transfer function G(s) =
s = + j,
G(s) =

1
,
s+a

with a > 0, we have, for

1
a + j
=
a + + j
(a + )2 + 2

and
(G(s)) =
Hence, G(s) =

a+
> 0.
(a + )2 + 2

1
s+a

is positive real. Furthermore, for any  (0, a), we have

(G(s )) =
and therefore G(s) =

a+
>0
(a  + )2 + 2
1
s+a

is also strictly positive real.

Denition 5.1 shows that a positive real transfer function maps the closed right
half of the complex plane to itself. Based on complex analysis, we can obtain the
following result.
Proposition 5.1. A proper rational transfer function G(s) is positive real if

all the poles of G(s) are in the closed left half of the complex plane
any poles on the imaginary axis are simple and their residues are non-negative
for all R, (G(j)) 0 when j is not a pole of G(s)

1
It can be seen that G(s) = s+a
, with a < 0, is not positive real. If G(s) is positive real, we must have (G(j)) 0. Similarly, other necessary conditions can be
obtained for a transfer function G(s) to be positive real. We can state those conditions
in an opposite way.

Proposition 5.2. A transfer function G(s) cannot be positive real if one of the
following conditions is satised:

The relative degree of G(s) is greater than 1.

G(s) is unstable.
G(s) is non-minimum phase (i.e., with unstable zero).
The Nyquist plot of G(j) enters the left half of the complex plane.

Advanced stability theory

Based on this proposition, the transfer functions G1 =

s2

57

s1
, G2 =
+ as + b

s+1
1
and G3 = 2
are not positive real, for any real numbers a
s2 s + 1
s + as + b
and b, because they are non-minimum phase, unstable, and with relative degree 2
s+4
respectively. It can also be shown that G(s) = 2
is not positive real as
s + 3s + 2

G(j) < 0 for > 2 2.

One difference between strictly positive real transfer functions and positive real
transfer functions arises due to the poles on imaginary axis.
1
Example 5.2. Consider G(s) = . For s = + j, we have?
s



1
= 2
.
(G(s)) = 
+ j
+ 2
Therefore, G(s) =

1
s

is positive real. However, G(s) =

1
s

is not strictly positive real.



For the stability analysis later in the book, we only need the result on strictly
positive real transfer functions.
Lemma 5.3. A proper rational transfer function G(s) is strictly positive real if and
only if

G(s) is Hurwitz, i.e., all the poles of G(s) are in the open left half of the complex
plane.
The real part of G(s) is strictly positive along the j axis, i.e.,
0,

(G(j)) > 0,

lims G(s) > 0, or in case of lims G(s) = 0, lim 2 (G(j)) > 0.

Proof. We show the proof for sufciency here, and omit the necessity, as it is more
involved. For sufciency, we only need to show that there exists a positive real constant
 such that G(s ) is positive real.
Since G(s) is Hurwitz, there must exist a positive real constant such that for
(0, ], G(s ) is Hurwitz. Suppose (A, b, cT , d) is a minimum state space
realisation for G(s), i.e.,
G(s) = cT (sI A)1 b + d.
We have
G(s ) = cT (sI I A)1 b + d
= cT (sI A)1 ((sI I A) + I )(sI I A)1 b + d
= G(s) + E(s),

(5.1)

58

Nonlinear and adaptive control systems

where
E(s) = cT (sI A)1 (sI I A)1 b.
E(s) is Hurwitz, and strictly proper. Therefore, we have
(E(j)) < r1 ,

R, (0, ]

(5.2)

for some positive real r1 and the existence of lim 2 (E(j)), which implies
2 (E(j)) < r2 ,

for || > 1 , (0, ]

(5.3)

for some 1 > 0.

If lim (G(j)) > 0, we have
(G(j)) > r3 ,

(5.4)

for some r3 > 0. Hence, combining (5.2) and (5.4), we obtain, from (5.1), that
(G(j )) > r3 r1 ,

R.

(5.5)

Then we have (G(j )) > 0, by setting < rr31 .

In the case that lim (G(j)) = 0, the condition
lim 2 (G(j)) > 0

implies that
2 (G(j)) > r4 ,

for || > 2

(5.6)

for some positive reals r4 and 2 . From (5.1), (5.3) and (5.6), we obtain that
2 (G(j )) > r4 r2 ,

for || > 3

(5.7)

where 3 = max{1 , 2 }. From the second condition of the lemma, we have, for some
positive real constant r5 ,
(G(j)) > r5 ,

for || 3 .

(5.8)

Then from (5.1), (5.2) and (5.8), we obtain that

(G(j )) > r5 r1 ,

for || 3 .

(5.9)

Combining the results in (5.7) and (5.9), we obtain that (G(j )) > 0 by setting
= min{ rr42 , rr51 }. Therefore, we have shown that there exists a positive real such that
G(s ) is positive real.
2

Advanced stability theory

59

The main purpose of introducing strictly positive real systems is for the following
result, which characterises the systems using matrices in time domain.
Lemma 5.4 (KalmanYakubovich lemma). Consider a dynamic system
x = Ax + bu
y = cT x,

(5.10)

where x Rn is the state variable; y and u R are the output and input respectively;
and A, b and c are constant matrices with proper dimensions, and (A, b, cT ) is
controllable and observable. Its transfer function G(s) = cT (sI A)1 b is strictly
positive real if and only if there exist positive denite matrices P and Q such that
AT P + PA = Q,
Pb = c.

(5.11)

Remark 5.1. We do not provide a proof here, because the technical details in the
proof such as nding the positive denite P and the format of Q are beyond the scope
of this book. In the subsequent applications for stability analysis, we only need to
know the existence of P and Q, not their actual values for a given system. For example
in the stability analysis for adaptive control systems in Chapter 7, we only need to
make sure that the reference model is strictly positive real, which then implies the
existence of P and Q to satisfy (5.11).


5.2 Absolute stability and circle criterion

In this section, we will consider a dynamic system which consists of a linear part
and a memoryless nonlinear component. Surely some engineering systems can be
modelled in this format, such as linear systems with nonlinearity in sensors. Let us
consider a closed-loop system
x = Ax + bu
y = cT x

(5.12)

u = F(y)y,
where x Rn is the state variable; y and u R are the output and input respectively;
and A, b and c are constant matrices with proper dimensions. The nonlinear component
is in the feedback law. Similar systems have been considered earlier using describing
functions for approximation to predict the existence of limit cycles. Nonlinear elements considered in this section are sector-bounded, i.e., the nonlinear feedback gain
can be expressed as
< F(y) <
for some constants and , as shown in Figure 5.2.

(5.13)

60

Nonlinear and adaptive control systems

r=0 +

.
x = Ax + bu

y = cTx

F(y)

Figure 5.1 Block digram of system (5.12)

F(y)y
b

Figure 5.2 Sector-bounded nonlinear feedback gain

The absolute stability refers to the globally asymptotic stability of the equilibrium point at the origin for the system shown in (5.12) for a class of sector-bounded
nonlinearities shown in (5.13). We will use KalmanYakubovich lemma for the stability analysis. If the transfer function for the linear part is strictly positive real, we
can establish the stability by imposing a restriction on the nonlinear element.
Lemma 5.5. For the system shown in (5.12), if the transfer function cT (sI A)1 b
is strictly positive real, the system is absolutely stable for F(y) > 0.
Proof. The proof is straightforward by invoking KalmanYakubovich lemma. Since
the linear part of the system is strictly positive real, then there exist positive denite
matrices P and Q such that (5.11) is satised.
Consider a Lyapunov function candidate
V = xT Px.

Advanced stability theory

61

Its derivative is given by

V = xT (AT P + PA)x + 2xT PBbu
= xT Qx 2xT PbF(y)y
= xT Qx 2xT cF(y)y
= xT Qx 2F(y)y2 ,
where in obtaining the third line of equation, we used Pb = c from the Kalman
Yakubovich lemma. Therefore, if F(y) > 0, we have
V xT Qx,
and then system is exponentially stable by Theorem 4.4.

Note that the conditions specied in Lemma 5.4 are the sufcient conditions.
With the result shown in Lemma 5.5, we are ready to consider the general case
for < F(y) < . Consider the function dened by
F
F =
F

(5.14)

and obviously we have F > 0. How to use this transformation for analysis of systems
stability?
With G(s) = cT (sI A)1 b, the characteristic equation of (5.12) can be
written as
G(s)F + 1 = 0.

(5.15)

Manipulating (5.15) by adding and subtracting suitable terms, we have

G(s)(F ) = G 1,

(5.16)

G(s)( F) = G + 1.

(5.17)

With (5.16) being divided by (5.17), we can obtain that

1 + G F

+ 1 = 0.
1 + G F

(5.18)

:= 1 + G
G
1 + G

(5.19)

Let

and we can write (5.18) as

F + 1 = 0
G

(5.20)

62

Nonlinear and adaptive control systems

which implies that the stability of the system (5.12) with the nonlinear gain shown in

(5.13) is equivalent to the stability of the system with the forward transfer function G

and the feedback gain F. Based on Lemma 5.5 and (5.20), we can see that the system
is strictly positive real.
(5.12) is stable if G
in (5.19) cannot deal with the case = .
The expressions of F in (5.14) and G
In such a case, we re-dene
F = F

(5.21)

we can obtain the manipulated characteristic

which ensures that F > 0. With this F,
equation as
G
(F ) + 1 = 0
1 + G
which enables us to re-dene
:=
G

G
.
1 + G

(5.22)

We summarise the results in the following theorem.

Theorem 5.6 (Circle criterion). For the system (5.12) with the feedback gain satisfying
dened by
the condition in (5.13), if the transfer function G
1 + G(s)

G(s)
=
1 + G(s)
or in case of = , by

G(s)
=

G(s)
1 + G(s)

is strictly positive real, with G(s) = cT (sI A)1 b, the system is absolutely stable.
is strictly positive real? Let us assume that
What is the condition of G if G
> > 0. Other cases can be analysed similarly. From Lemma 5.3, we know that
to be strictly positive real, we need G
to be Hurwitz, and (G(j))

for G
> 0, that is



1 + G(j)
1 + G(j)


> 0,

R,

which is equivalent to



1/ + G(j)
1/ + G(j)


> 0,

R.

(5.23)

Advanced stability theory

If

+ G(j) = r1 ej1 and

63

+ G(j) = r2 ej2 , the condition in (5.23) is satised by

< 1 2 < ,
2
2

which is equivalent to the point G(j) that lies outside the circle centered at
( 12 (1/ + 1/), 0) with radius of 12 (1/ 1/) in the complex plane. This circle intersects the real axis at ( 1 , 0) and ( 1 , 0). Indeed, 1 + G(j) is represented
as a vector from the point ( 1 , 0) to G(j), and 1 + G(j) as a vector from the
point ( 1 , 0) to G(j). The angle between the two vectors will be less than 2 when
G(j) is outside the circle, as shown in Figure 5.3. Since the condition must hold

for all R, the condition (G(j))

> 0 is equivalent to the Nyquist plot of G(s)
is strictly positive real requires that
that lies outside the circle. The condition that G
the Nyquist plot of G(s) does not intersect with the circle and encircles the circle
counterclockwise the same number of times as the number of unstable poles of G(s),
as illustrated in Figure 5.4.
Alternatively, the circle can also be interpreted from complex mapping. From
(5.19), it can be obtained that
G=

1
G
.

(5.24)

The mapping shown in (5.24) is a bilinear transformation, and it maps a line to a line
or circle. For the case of > > 0, we have


1
1
1
/
G=
.
(5.25)

G
The function
/

G /
Im

G( jw)
q2
1
a

q1

Re

1
b

Figure 5.3 Diagram of |1 2 | <

64

Nonlinear and adaptive control systems

Im

G( jw)

1
a

1
b

Re

Figure 5.4 Circle criterion

maps the imaginary axis to a circle centred as (1/2, 0) with the radius 1/2, i.e., the
line from (1, 0) to (0, 0) on the complex plane is the diameter of the circle. Then
the function


1
/
1

G
maps the imaginary axis to a circle with the diameter on the line from (0, 0) to

( 1 1 , 0) on the complex plane. Finally, it can be seen from (5.25) that this map of G
1
to G maps the imaginary axis to the circle with the diameter on the line from ( , 0)
to ( 1 , 0), or in other words, the circle centered as ( 12 (1/ + 1/), 0) with radius
of 12 (1/ 1/). It can also be shown that the function maps the open left-hand
complex plane to the domain inside the circle.
Indeed, we can evaluate the circle directly from (5.24). Let u and v denote the
real and imaginary parts of the mapping of the imaginary axis, and we have
u = 

j 1
+ 2
= 2
,
j
+ 2 2

v = 

j 1
( )
.
= 2
j
+ 2 2

Denoting = , we obtain




1/ + (1/)2
1
1 1
1 1 2
1 1
+
+

,
1 + 2
2

2
1 + 2


(1/ 1/)
1 1
1
2
v =
=
.

2
1+
2
1 + 2

u =

Advanced stability theory

65

It is now easy to see that



1
u+
2

1
1
+

2

 
2
1 1
1
+v =
.

which describes the circle that is discussed before.

5.3 Input-to-state stability and small gain theorem

In this section, we continue to consider the stability properties for systems with inputs.
For linear systems, if a system is asymptotically stable, the state will remain bounded
when the input is bounded. Does this property still hold for nonlinear systems?
Consider a nonlinear system
x = f (x, u)

(5.26)

where x Rn is the state of the system, and u Rm is a bounded input, and f :

Rn Rm Rn is a continuous and locally Lipschitz function. If the autonomous
system
x = f (x, 0)
is asymptotically stable, will the state remain bounded for a bounded input signal u?
Example 5.3. Consider a nonlinear system
x = x + (1 + 2x)u,

x(0) = 0

where x R. When u = 0, we have

x = x
which is asymptotically (exponentially) stable. However, the state of this system may
not remain bounded for a bounded input. For example if we let u = 1, we have
x = x + 1
of which the state variable x is unbounded.

From the above example, it can be seen that even the corresponding autonomous
system is asymptotically stable, the state may not remain bounded subject to a bounded
input. We introduce a denition for systems with the property of bounded state with
bounded input.
To show a denition of input-to-state stable (ISS), we need to use comparison
functions, which are dened below.

66

Nonlinear and adaptive control systems

Denition 5.3. A function : [0, a) [0, ) is a class K function if is continuous, and strictly increasing with (0) = 0. If a = and limr (r) = , the
function is a class K function.
Denition 5.4. A function : [0, a) [0, ) [0, ) is a class KL function if it
is continuous, for a xed t = t0 , (, t0 ) is a class K function, and for a xed x(0),
limt (x(0), t) = 0.
Denition 5.5. The system (5.26) is ISS if there exist a class KL function and a
class K function such that
x(t) (x(0), t) + (u ),

t > 0.

(5.27)

There is an alternative denition, which may be convenient to use in some

situations. Here, we state it as a proposition.
Proposition 5.7. The system shown in (5.26) is ISS if and only if there exist a class
KL function and a class K function such that
x(t) max{(x(0), t), (u )},

t > 0.

(5.28)

Proof. From (5.28), we have

x(t) (x(0), t) + (u )
and hence the system is ISS. From (5.27), there exist a class KL function 1 and a
class K function 1
x(t) 1 (x(0), t) + 1 (u )
max{21 (x(0), t), 21 (u )}
and therefore the system satises (5.28) with = 21 and = 21 .

Example 5.4. Consider a linear system

x = Ax + Bu,
where x Rn and u Rm are the state and input respectively, and A and B are matrices
with appropriate dimensions and A is Hurwitz. For this system, when u = 0, the
system is asymptotically stable, as A is Hurwitz. From its solution


x(t) = eAt x(0) +

0

eA(t ) Bu( )d ,

Advanced stability theory

67

we have


x(t) eAt x(0) +

eA(t ) d Bu .

Since A is Hurwitz, there exist positive real constants a and such that eA(t) 
aet . Hence, we can obtain
 t
x(t) aet x(0) +
ae(t ) d Bu
0

ae

a
x(0) + Bu

It is easy to see that the rst term in the above expression is a KL function of t and
x(0) and the second term is a K function of u . Therefore, the linear system with
Hurwitz system matrix A is ISS.

Next, we show a result on establishing ISS property from a Lyapunov function.
Theorem 5.8. For the system (5.26), if there exists a function V (x) : Rn R with
continuous rst-order derivatives such that
a1 xb V (x) a2 xb ,
V
f (x, u) a3 xb ,
x

x (u)

(5.29)
(5.30)

where a1 , a2 , a3 and b are positive real constants and is a class K function, the
system (5.26) is ISS.
Proof. From Theorem 4.4, we can see that the system is exponentially stable when
u = 0, and V (x) is a Lyapunov function for the autonomous system x = f (x, 0). To
consider the case for non-zero input, let us nd a level set based on V , dened by
c := {x|V (x) c}
and determine the constant c such that for x
/ c , we have x > (u ). Indeed,
let
c = a2 ((u ))b .
In this case, from V (x) > c, we have
V (x) > a2 ((u ))b ,
which implies that
a2 xb > a2 ((u ))b
and hence x > (u ).

68

Nonlinear and adaptive control systems

Therefore, for any x outside c , we can obtain from (5.29) and (5.30), in a similar
way to the proof of Theorem 4.4,
a3
V V
a2
and then

x(t)

a2
a1

1/b
x(0)e

a
2

a 3b t

There will be a time, say t1 , at which the trajectory enters c .

We can show that c is invariant, i.e., the trajectory starts from c will remain
in c , and this is shown by the fact that on the boundary of c , V 0. Therefore,
after t1 , the trajectory will remain in c . Furthermore, for x c , we have,
V (x) < a2 ((u ))b
which implies that

x

a2
a1

1/b
(u ).

Combining the cases for x outside and in c , we conclude that


x max

a2
a1

1/b
x(0)e

a
2

a 3b t

a2
,
a1

1/b


(u ) .

Hence, the system is ISS with the gain function () = ( aa21 )1/b ().
There is a more general result than Theorem 5.8 that requires x = f (x, 0) to be
asymptotically stable, not necessarily exponential stable. The proof of that theorem
is beyond the level of this text, and we include it here for completeness.
Theorem 5.9. For the system (5.26), if there exists a function V (x) : Rn R with
continuous rst-order derivatives such that
1 (x) V (x) 2 (x),
V
f (x, u) 3 (x),
x

x (u),

(5.31)
(5.32)

where 1 , 2 , 3 are K functions and is a class K function, the system (5.26) is

ISS with the gain function () = 11 (2 (())).
Note that class K functions are class K functions that satisfy the property
lim (r) = .

Advanced stability theory

69

There is a slightly different version of Theorem 5.9 which is shown below.

Corollary 5.10. For the system (5.26), if there exists a function V (x) : Rn R with
continuous rst-order derivatives such that
1 (x) V (x) 2 (x),

(5.33)

V
f (x) (x) + (u),
x

(5.34)

where 1 , 2 , are K functions and is a class K function, the system (5.26)

is ISS.
The function which satises (5.31) and (5.32) or (5.33) and (5.34) is referred to as
ISS-Lyapunov function. In fact, it can be shown that the existence of an ISS-Lyapunov
function is also a necessary condition for the system to be ISS. Referring to (5.34),
the gain functions and characterise the ISS property of the system, and they are
also referred to as an ISS pair. In other words, if we say a system is ISS with ISS pair
(, ), we mean that there exists an ISS-Lyapunov function that satises (5.34).
Example 5.5. Consider a nonlinear system
x = x3 + u,
where x R is the state, and u is input. The autonomous part x = x3 is considered
in Example 4.2, and it is asymptotically stable, but not exponentially stable. Consider
an ISS-Lyapunov function candidate
V =

1 2
x .
2

Its derivative is given by

V = x4 xu
1
1
x4 |x|(|x|3 2|u|)
2
2
1
x4 , for|x| (2|u|)1/3 .
2
Hence the system is ISS with the gain function (|u|) = (2|u|)1/3 , based on
Theorem 5.9. Alternatively, using Youngs inequality, we have
|x||u|

1 4 3 4/3
|x| + |u|
4
4

70

Nonlinear and adaptive control systems

which gives
3
3
V x4 + |u|4/3 .
4
4
Therefore, the system is ISS based on Corollary 5.10 with () = 34 ()4 and

() = 34 ()4/3 .
ISS property is useful in establishing the stability of interconnected systems. We
include two results here to end this section.
Theorem 5.11. If for the cascade connected system
x 1 = f1 (x1 , x2 ),

(5.35)

x 2 = f1 (x2 , u),

(5.36)

the subsystem (5.35) is ISS with x2 as the input, and the subsystem (5.36) is ISS with
u as input, the overall system with state x = [x1T , x2T ]T and input u is ISS.
Theorem 5.12 (ISS small gain theorem). If for the interconnected system
x 1 = f1 (x1 , x2 ),

(5.37)

x 2 = f1 (x2 , x2 , u),

(5.38)

the subsystem (5.37) is ISS with x2 as the input with 1 as the ISS gain for x2 , and
the subsystem (5.38) is ISS by viewing x1 and u as the inputs, with the ISS input gain
function 2 for x1 , the overall system with state x = [x1T , x2T ]T and input u is ISS if
1 (2 (r)) < r,

r > 0.

(5.39)

From Theorem 5.11, it can be seen that if the subsystem x2 is globally asymptotically stable when u = 0, the overall system is globally asymptotically stable. Similarly,
Theorem 5.12 can be used to establish the stability of the following system:
x 1 = f1 (x1 , x2 )
x 2 = f1 (x2 , x2 ),
and the global and asymptotic stability of the entire system can be concluded if the
gain condition shown in Theorem 5.12 is satised.
Example 5.6. Consider the second-order nonlinear system
x 1 = x13 + x2 ,
2/3

x 2 = x1 x2 3x2 .

Advanced stability theory

71

From Example 5.5, we know that x1 -subsystem is ISS with the gain function
1 () = (2)1/3 . For the x2 -subsystem, we choose
V2 =

1 2
x
2 2

and we have
5/3
V 2 = 3x22 + x1 x2

x22 |x2 |5/3 (2|x2 |1/3 |x1 |)



|x1 | 3
2
x2 , for |x2 | >
2
Hence, the x2 -subsystem is ISS with the gain function 2 () = ( 2 )3 . Now we have,
for r > 0,
   1/3  1/3
r 3
1
1 (2 (r)) = 2
=
r < r.
2
4
Therefore, the system is globally asymptotically stable.

5.4 Differential stability

Consider a nonlinear system
x = f (x, u),

(5.40)

where x Rn is the state vector, u Rs is the input and f : Rn Rs Rn is a

nonlinear smooth vector eld with f (0, u) = 0.
Denition 5.6. This system (5.40) has differential stability if there exists a Lyapunov
function V (x) such that V : Rn R for all x, x Rn , u Rs , satises
1 (x) V (x) 2 (x),
V (x x )
(f (x, u) f (x, u)) 3 (x x ),
x


V (x) c2
3 (x),

c1
x

(5.41)

where i , i = 1, 2, 3, are K functions and ci , i = 1, 2, are positive real constants with

c2 > 1.

72

Nonlinear and adaptive control systems

Remark 5.2. The conditions specied in (5.41) are useful for observer design, in
particular, for the stability analysis of the reduced-order observers in Chapter 8.
A similar denition to differential stability is incremental stability. However, the
conditions specied in (5.41) are not always satised by the systems with incremental
stability. When x = 0, i.e., in the case for one system only, the conditions specied
in (5.41) are then similar to the properties of the nonlinear systems with exponential
stability. The last condition in (5.41) is specied for interactions with other systems.
This condition is similar to the conditions for the existence of changing the supply
functions for inter-connection of ISS systems.

We include two illustrative examples below for the properties of differential
stability.
Example 5.7. A linear system is differentially stable if the system is asymptotically
stable. Consider
x = Ax,
where A Rnn . If the system is asymptotically stable, A must be Hurwitz. Therefore
there exist positive denite matrices P and Q such that
AT P + PA = Q.
Let V (x) = xT Px. In this case, the conditions (5.41) are satised with
1 (x) = min (P)x2 ,
2 (x) = max (P)x2 ,
3 (x) = min (Q)x2 ,
c1 =

min (Q)
, c2 = 2,
4(max (P))2

where min () and max () denote the minimum and maximum eigenvalues of a
positive denite matrix.

The differential stability is closely related to observer design. Consider a system
with input u Rs
x = Ax + Bu.

(5.42)

If x = Ax is differentially stable, an observer for (5.42) can be designed as

x = Ax + Bu,

x (0) = 0.

(5.43)

It is easy to see that x x converges to 0 exponentially.

However, for nonlinear systems, differential stability is not guaranteed by the
asymptotic or even exponential stability of a system.

Advanced stability theory

73

Example 5.8. We consider a rst-order nonlinear system

x = x 2 sin x.
Take V = 12 x2 , and we have
V = x2 2x sin x.
For |x| , we have x sin x 0, and therefore
V x2 .
For |x| > , we have
V x2 2x sin x
x2 + 2|x|




2
|x|
2
= 1
x 2|x|
1



2
1
x2 .

Combining both the cases, we have



2

V 1
x2 .

Hence, the system is exponentially stable. But this system is not differentially stable.
Indeed, let e = x x . We have
e = e 2( sin (x) sin (x + e)).
By linearising the system at x = and x = , and denoting the error at this point by
el , we have
e l = el 2 cos (x)|x= (x ) + 2 cos (x)|x = (x )
= el
and the system is unstable in a neighbourhood of this point.

Chapter 6

Feedback linearisation

Nonlinear systems can be linearised around operating points and the behaviours in
the neighbourhoods of the operating points are then approximated by their linearised
models. The domain for a locally linearised model can be fairly small, and this may
result in that a number of linearised models are needed to cover an operating range
of a system. In this chapter, we will introduce another method to obtain a linear
model for nonlinear systems via feedback control design. The aim is to convert a
nonlinear system to a linear one by state transformation and redening the control
input. The resultant linear model describes the system dynamics globally. Of course,
there are certain conditions for the nonlinear systems to satisfy so that this feedback
linearisation method can be applied.

6.1 Inputoutput linearisation

The basic idea for inputoutput linearisation is fairly straightforward. We use an
example to demonstrate this.
Example 6.1. Consider the nonlinear system
x 1 = x2 + x13
x 2 = x12 + u
y = x1 .
For this system, taking the derivatives of y, we have
y = x2 + x13 ,
y = 3x12 (x2 + x13 ) + x12 + u.
Now, let us dene
v = 3x12 (x2 + x13 ) + x12 + u,
and we obtain
y = v.

(6.1)

76

Nonlinear and adaptive control systems

Viewing v as the new control input, we see that the system is linearised. Indeed, let
us introduce a state transformation
1 := y = x1 ,
2 := y = x2 + x13 .
We then obtain a linear system
1 = 2
2 = v.
We can design a state feedback law as
v = a1 1 a2 2
to stabilise the system with a1 > 0 and a2 > 0. The control input of the original system
is given by
u = 3x12 (x2 x13 ) x12 + v.

(6.2)

We say that the system (6.1) is linearised by the feedback control law (6.2). Notice
that this linearisation works for the entire state space.

As shown in the previous example, we can keep taking the derivatives of the
output y until the input u appears in the derivative, and then a feedback linearisation
law can be introduced. The derivatives of the output also introduce a natural state
transformation.
Consider a nonlinear system
x = f (x) + g(x)u
y = h(x),

(6.3)

where x D Rn is the state of the system; y and u R are output and input respectively; and f and g : D Rn Rn are smooth functions and h : D Rn R is a
smooth function.
Remark 6.1. The functions f (x) and g(x) are vectors for a given point x in the state
space, and they are often referred to as vector elds. All the functions in (6.3) are
required to be smooth in the sense that they have continuous derivatives up to certain
orders when required. We use the smoothness of functions in the remaining part of
the chapter in this way.

The inputoutput feedback linearisation problem is to design a feedback control
law
u = (x) + (x)v
with (x)  = 0 for x D such that the inputoutput dynamics of the system

(6.4)

Feedback linearisation
x = f (x) + g(x)(x) + g(x)(x)v
y = h(x)

77
(6.5)

are described by
y() = v

(6.6)

for 1 n.
For the system (6.3), the rst-order derivative of the output y is given by
h(x)
(f (x) + g(x)u)
x
:= Lf h(x) + Lg h(x)u,

y =

where the notations Lf h and Lg h(x) are Lie derivatives.

For any smooth function f : D Rn Rn and a smooth function
h : D Rn R, the Lie derivative Lf h, referred to as the derivative of h along
f , is dened by
Lf h(x) =

h(x)
f (x).
x

This notation can be used iteratively, that is

Lf (Lf h(x)) = Lf2 h(x),
Lfk (h(x)) = Lf (Lfk1 (h(x))),
where k 0 is an integer.
The solution to this problem depends on the appearance of the control input in
the derivatives of the output, which is described by the relative degree of the dynamic
system.
Denition 6.1. The dynamic system (6.3) has relative degree at a point x if the
following conditions are satised:
Lg Lfk h(x) = 0,

for k = 0, . . . , 2,

Lg Lf1 h(x)  = 0.
Example 6.2. Consider the system (6.1). Comparing it with the format shown in
(6.3), we have
 3

 
x1 + x 2
0
f (x) =
, g(x) =
, h(x) = x1 .
1
x12

78

Nonlinear and adaptive control systems

Direct evaluation gives

Lg h(x) = 0,
Lf Lg h(x) = 1.
Therefore, the relative degree of the system (6.1) is 2.

For SISO linear systems, the relative degree is the difference between the orders
of the polynomials in the numerator and denominator of the transfer function. With
the denition of the relative degree, we can present the inputoutput feedback
linearisation using Lie derivatives.
Example 6.3. Consider the system (6.1) again, and continue from Example 6.2. With
Lg h(x) = 0, we have
y = Lf h(x)
where
Lf h(x) = x13 + x2 .
Taking the derivative of Lf h(x), we have
y = Lf2 h(x) + Lg Lf h(x)u
where
Lf2 h(x) = 3x12 (x2 + x13 ) + x12 ,
Lg Lf h(x) = 1.
Therefore, we have
v = Lf2 h(x) + Lg Lf h(x)u
or
u=

Lf2 h(x)
Lg Lf h(x)

1
v
Lg Lf h(x)

which gives the same result as in Example 6.1.

The procedure shown in Example 6.3 works for systems with any relative degrees.
Suppose that the relative degree for (6.3) is , which implies that Lg Lfk h(x) = 0 for
k = 0, . . . , 2. Therefore, we have the derivatives of y expressed by
y(k) = Lfk h(x),

for k = 0, . . . , 1,

y() = Lf h(x) + Lg Lf1 h(x)u.

(6.7)
(6.8)

Feedback linearisation

79

If we dene the input as

u=

1
Lg Lf1 h(x)

(Lf h(x) + v)

it results in
y() = v.
We can consider to use i := y(i1) = Lfi1 h as coordinates for the linearised
has full
inputoutput dynamics. The only remaining issue is to establish that
x
rank. To do that, we need to introduce a few notations.
For any smooth functions f , g : D Rn Rn , the Lie bracket [f , g] is
dened by
[f , g](x) =

g(x)
g(x)
f (x)
f (x)
x
x

and we can introduce a notation which is more convenient for high-order Lie
brackets as
adf0 g(x) = g(x),
adf1 g(x) = [f , g](x),
adfk g(x) = [f , adfk1 g](x).
For the convenient of presentation, let us denote
dh =

h
x

which is a row vector. With this notation, we can write

Lf h = < dh, f > .
Based on the denition of Lie Bracket, it can be obtained by a direct evaluation
that, for a smooth function h : Rn R,
L[f , g] h = Lf Lg h Lg Lf h = Lf < dh, g > < dLf h, g >,
that is
< dh, [f , g] > = Lf < dh, g > < dLf h, g > .
Similarly it can be obtained that, for any non-negative integers k and l,
< dLfk h, adfl+1 g > = Lf < dLfk h, adfl g > < dLfk+1 h, adfl g >

(6.9)

80

Nonlinear and adaptive control systems

By now, we have enough tools and notations to show that

the denition of the relative degree,
< dLfk h, g > = 0,

has full rank. From

for k = 0, . . . , 2

and by using (6.9) iteratively, we can show that

< dLfk h, adfl g > = 0,

for k + l 2,

(6.10)

and
< dLfk h, adfl g > = (1)l < dLf1 h, g >,

for k + l = 1.

(6.11)

From (6.10) and (6.11), we have

dh(x)
dLf h(x)

g(x) adf g(x) . . . ad 1 g(x)

..
f

1
dLf h

0
...
0
(1)1 r(x)
0

. . . (1)2 r(x)

= .

.
.
.
..
..
..
..

r(x) . . .

(6.12)

where r(x) = < dLf1 h, g >. Therefore, we conclude that

dh(x)
dLf h(x)

..
=

x
.
dLf1 h
has full rank. We summarise the result about inputoutput feedback linearisation in
the following theorem.
Theorem 6.1. If the system in (6.3) has a well-dened relative degree in D, the
inputoutput dynamics of the system can be linearised by the feedback control law
u=

1
Lg Lf1 h(x)

(Lf h(x) + v)

and the linearised inputoutput dynamics are described by

(6.13)

Feedback linearisation
1 = 2
..
.
1 =

81

(6.14)

= v
with a partial state transformation
i = Lfi1 h(x) for i = 1, . . . , .

(6.15)

When = n, the whole nonlinear system dynamics are fully linearised.

The results shown in (6.10) and (6.11) can also be used to conclude the following
result which is needed in the next section.
Lemma 6.2. For any functions f , g : D Rn Rn , and h : D Rn R, all
differentiable to certain orders, the following two statements are equivalent for r > 0:

Lg h(x) = Lg Lf h(x) = = Lg Lfr h(x) = 0,

Lg h(x) = L[f , g] h(x) = = Ladfr g h(x) = 0.

Remark 6.2. The inputoutput dynamics can be linearised based on Theorem 6.1. In
the case of < n, the system for < n can be transformed under certain conditions
to the normal form
z = f0 (z, ),
1 = 2 ,
..
.
1 = ,
= Lf h + uLg Lf1 h,
y = 1
where z Rn is the part of the state variables which are not in the inputoutput
dynamics of the system, and f0 : Rn Rn is a smooth function. It is clear that
when < n, the inputoutput linearisation does not linearise the dynamics
z = f0 (z, ). Also note that the dynamics z = f0 (z, 0) are referred to as the zero
dynamics of the system.

To conclude this section, we use an example to demonstrate the inputoutput
linearisation for the case < n.

82

Nonlinear and adaptive control systems

Example 6.4. Consider the nonlinear system

x 1 = x13 + x2
x 1 = x12 + x3 + u
x 2 = x12 + u
y = x1 .
For this system we have

3
x1 + x2
f (x) = x12 + x3 ,
x12

0
g(x) = 1 ,
1

h(x) = x1 .

It is easy to check that

Lf h = x13 + x2 ,
Lg h = 0,
Lg Lf h = 1,
Lf2 h = 3x12 (x13 + x2 ) + x12 + x3
and therefore the relative degree = 2. For inputoutput linearisation, we set
u =

1
(Lf2 h + v)
Lg Lf h

= 3x12 (x13 + x2 ) x12 x3 + v.

Introduce the partial state transformation
1 = h = x1 ,
2 = Lf h = x13 + x2 ,
and it is easy to see that dh and dLf h are linearly independent. The linearised
inputoutput dynamics are described by
1 = 2 ,
2 = v.
If we like to transform the system to the normal form shown in Remark 6.2, we need
to introduce another state, in addition to 1 and 2 . For this, we have the additional
state
z = x3 x2 .
The inverse transformation is given by
x1 = 1 ,
x2 = 2 13 ,
x3 = z + 2 13 .

Feedback linearisation

83

With the coordinates z, 1 and 2 , we have the system in the normal form
z = z 2 + 13
1 = 2
2 = z + 2 + 12 13 + 312 2 + u.
It is clear that the inputoutput linearisation does not linearise the dynamics of z. Also
note that the zero dynamics for this system are described by
z = z.


6.2 Full-state feedback linearisation

Consider a nonlinear system
x = f (x) + g(x)u,

(6.16)

where x D Rn is the state of the system; u R is the input; and f , g :

D Rn Rn are smooth functions. The full-state linearisation problem is to nd a
feedback control design
u = (x) + (x)v

(6.17)

with (x)  = 0 for x D such that the entire system dynamics are linearised by a state
transformation with v as the new control input.
It is clear from the results shown in the inputoutput linearisation that the complete linearisation can only be achieved when the relative degree of the system equals
the order of the system. If we can nd an output function h(x) for the system (6.16),
the inputoutput linearisation result shown in the previous section can be applied
to solve the full-state feedback linearisation problem. Therefore, we need to nd an
output function h(x) such that
Lg Lfk h(x) = 0

for k = 0, . . . , n 2

Lg Lfn1 h(x)  = 0

(6.18)
(6.19)

Based on Lemma 6.2, the condition specied in (6.18) is equivalent to the condition
Lad k g h(x) = 0
f

for k = 0, . . . , n 2

(6.20)

and furthermore, the condition in (6.19) is equivalent to

Lad n1 g h(x)  = 0.

(6.21)

The output function h(x) that satises the condition shown in (6.20) is a solution of
the partial differential equation
[g, adf g, . . . , adfn2 g]

h
= 0.
x

(6.22)

84

Nonlinear and adaptive control systems

To discuss the solution of this partial differential equation, we need a few notations
and results. We refer to a collection of vector elds as a distribution. For example if
f1 (x), . . . , fk (x) are vector elds, with k a positive integer,
 = span{f1 (x), . . . , fk (x)}
is a distribution. The dimension of distribution is dened as
dim ((x)) = rank[f1 (x), . . . , fn (x)].
A distribution  is said to be involutive, if for any two vector elds f1 , f2 , we
have [f1 , f2 ] . Note that not all the distributions are involutive, as shown in the
following example.
Example 6.5. Consider the distribution
 = span{f1 (x), f2 (x)}
where

2x2
f1 (x) = 1 ,
0

1
f2 (x) = 0 .
x2

A direct evaluation gives

f2
f1
x

0 0
= 0 0
0 1

0
= 0.
1

[f1 , f2 ] =

f1
f2
x

0
2x2
0
0 1 0
0
0
0

2
0
0

0
1
0 0
0
x2

/ , and therefore  is not involutive. Indeed, the rank

It can be shown that [f1 , f2 ]
of the matrix [f1 , f2 , [f1 , f2 ]] is 3, which means that [f1 , f2 ] is linearly independent of f1
and f2 , and it cannot be a vector eld in . The rank of [f1 , f2 , [f1 , f2 ]] can be veried
by its non-zero determinant as

 2x2

|[f1 , f2 , [f1 , f2 ]]| =  1
 0

1
0
x2


0 
0  = 1.
1 


Feedback linearisation

85

We say that a distribution  is integrable if there exists a non-trivial h(x) such

that for any vector eld f , < dh, f > = 0. The relationship between an involutive
distribution and its integrability is stated in the following theorem.
Theorem 6.3 (Frobenius theorem). A distribution is integrable if and only if it is
involutive.
Now we are ready to state the main result for full-state feedback linearisation.
Theorem 6.4. The system (6.16) is full-state feedback linearisable if and only if
x D

the matrix G = [g(x), adf g(x), . . . , adfn1 g(x)] has full rank
the distribution Gn1 = span{g(x), adf g(x), . . . , adfn2 g(x)} is involutive

Proof. For the sufciency, we only need to show that there exists a function h(x) such
that the relative degree of the system by viewing h(x) as the output is n, and the rest
follows from Theorem 6.1. From the second condition that Gn1 is involutive, and
Frobenius theorem, there exists a function h(x) such that
h
[g, adf g, . . . , adfn2 g] = 0
x
which is equivalent to, from Lemma 6.2,
Lg Lfk h(x) = 0,

for k = 0, . . . , n 2.

From the condition that G is full rank, we can establish

Lg Lfn1 h(x)  = 0.
In fact, if Lg Lfn1 h(x) = 0, then from Lemma 6.2, we have
h
G=0
x
which is a contradiction as G has full rank.
For necessity, we show that if the system is full-state feedback linearisable, then
the two conditions hold. From the problem formulation, we know that the full-state
linearisability is equivalent to the existence of an output function h(x) with relative
degree n. From the denition of the relative degree and Lemma 6.2, we have
Lad k g h(x) = 0,
f

for

k = 0, . . . , n 2,

which implies
h
[g, adf g, . . . , adfn2 g] = 0.
x

86

Nonlinear and adaptive control systems

From Frobenius theorem, we conclude that Gn1 is involutive. Furthermore, from the
fact that the system with h(x) as the output has a relative degree n, we can show, in
the same way as the discussion that leading to Lemma 6.2, that

dh(x)
dLf h(x)

n1
..

g(x) adf g(x) adf g(x)

.
dLfn1 h

0
...
0
(1)n1 r(x)
0

. . . (1)n2 r(x)

= .

.
.
.
..
..
..
..

r(x) . . .

where r(x) = Lg Lfn1 h(x). This implies that G has rank n. This concludes the
proof.
2
Remark 6.3. Let us see the conditions in Theorem 6.4 for linear systems with
f (x) = Ax, g(x) = b.
where A is a constant matrix and b is a constant vector. A direct evaluation gives
[f , g] = Ab,
and
adfk g = (1)k Ak b
for k > 0. Therefore, we have
G = [b, Ab, . . . , (1)n1 An1 b].
It can be seen that the full rank condition of G is equivalent to the full controllability
of the linear system.

In the next example, we consider the dynamics of the system that was considered in Example 6.4 for inputoutput linearisation for the input h(x) = x1 . We will
show that the full-state linearisation can be achieved by nding a suitable output
function h(x).

Feedback linearisation

87

Example 6.6. Consider the nonlinear system

x 1 = x13 + x2
x 1 = x12 + x3 + u
x 2 = x12 + u,
for full-state feedback linearisation. With

3
0
x1 + x2
f (x) = x12 + x3 , g(x) = 1 ,
1
x12
we have

1
[f , g] = 1 ,
0

and

3x12 + 1
adf2 g = 2x1 .
2x1

Hence, we have

1
0
G2 = span 1 , 1

1
0

and

0 1
G = 1 1
1 0

3x12 + 1
2x1 .
2x1

The distribution G2 is involutive, as it is spanned by constant vectors. The matrix G

has full rank, as shown by its determinant
|G| = 3x12 + 1  = 0.
Hence, the conditions in Theorem 6.4 are all satised, and the system is full-state
linearisable. Indeed, we can nd
h(x) = x1 x2 + x3
by solving

0 1
h
1 1 = 0.
x 1 0
For this h(x), it is easy to check that

88

Nonlinear and adaptive control systems

Lg h = 0,
Lf h = x2 + x13 x3 ,
Lg Lf h = 0,
Lf2 h = 3x12 (x13 + x2 ) + x3 ,
Lg Lf2 h = 3x12 + 1,
Lf3 h = (15x14 + 6x1 x2 )(x13 + x2 ) + 3x12 (x3 + x12 ) + x12 .

It can be seen that the relative degree, indeed, equals 3 as

Lg Lf2 h = 3x12 + 1  = 0.
For the full-state linearisation, we have the state transformation
1 = x1 x2 + x3 ,
2 = x2 + x13 x3 ,
2 = 3x12 (x13 + x2 ) + x3 ,
and the feedback law
u=



1
v (15x14 + 6x1 x2 )(x13 + x2 ) 3x12 (x3 + x12 ) x12 .
+1

3x12

Chapter 7

Adaptive control of linear systems

The principle of feedback control is to maintain a consistent performance when there

are uncertainties in the system or changes in the setpoints through a feedback controller using the measurements of the system performance, mainly the outputs. Many
controllers are with xed controller parameters, such as the controllers designed by
normal state feedback control, and H control methods. The basic aim of adaptive
control also is to maintain a consistent performance of a system in the presence of
uncertainty or unknown variation in plant parameters, but with changes in the controller parameters, adapting to the changes in the performance of the control system.
Hence, there is an adaptation in the controller setting subject to the performance of the
closed-loop system. How the controller parameters change is decided by the adaptive
laws, which are often designed based on the stability analysis of the adaptive control
system.
A number of design methods have been developed for adaptive control. Model
Reference Adaptive Control (MRAC) consists of a reference model which produces the desired output, and the difference between the plant output and the
reference output is then used to adjust the control parameters and the control
input directly. MRAC is often in continuous-time domain, and for deterministic
plants. Self-Tuning Control (STC) estimates system parameters and then computes
the control input from the estimated parameters. STC is often in discrete-time and for
stochastic plants. Furthermore, STC often has a separate identication procedure for
estimation of the system parameters, and is referred to as indirect adaptive control,
while MRAC adapts to the changes in the controller parameters, and is referred to
as direct adaptive control. In general, the stability analysis of direct adaptive control is less involved than that of indirect adaptive control, and can often be carried
out using Lyapunov functions. In this chapter, we focus on the basic design method
of MRAC.
Compared with the conventional control design, adaptive control is more
involved, with the need to design the adaptation law. MRAC design usually involves
the following three steps:

Choose a control law containing variable parameters.

Design an adaptation law for adjusting those parameters.
Analyse the stability properties of the resulting control system.

90

Nonlinear and adaptive control systems

7.1 MRAC of rst-order systems

The basic design idea can be clearly demonstrated by rst-order systems. Consider a
rst-order system
y + ap y = bp u,

(7.1)

where y and u R are the system output and input respectively, and ap and bp are
unknown constant parameters with sgn(bp ) known. The output y is to follow the output
of the reference model
y m + am ym = bm r.

(7.2)

The reference model is stable, i.e., am > 0. The signal r is the reference input. The
design objective is to make the tracking error e = y ym converge to 0.
Let us rst design a Model Reference Control (MRC), that is, the control design
assuming all the parameters are known, to ensure that the output y follows ym .
Rearrange the system model as


a p am
y + am y = bp u
y
bp
and therefore we obtain


a p am
bm
y r
e + am e = bp u
bp
bp
:= bp (u au y ar r) ,
where
ay =

ap a m
,
bp

ar =

bm
.
bp

If all the parameters are known, the control law is designed as

u = ar r + ay y

(7.3)

and the resultant closed-loop system is given by

e + am e = 0.
The tracking error converges to zero exponentially.
One important design principle in adaptive control is the so-called the certainty
equivalence principle, which suggests that the unknown parameters in the control
design are replaced by their estimates. Hence, when the parameters are unknown,
let a r and a y denote their estimates of ar and ay , and the control law, based on the
certainty equivalence principle, is given by
u = a r r + a y y.

(7.4)

Adaptive control of linear systems

91

Note that the parameters ar and ay are the parameters of the controllers, and they
are related to the original system parameters ap and bp , but not the original system
parameters themselves.
The certainty equivalence principle only suggests a way to design the adaptive
control input, not how to update the parameter estimates. Stability issues must be
considered when deciding the adaptive laws, i.e., the way how estimated parameters
are updated. For rst-order systems, the adaptive laws can be decided from Lyapunov
function analysis.
With the proposed adaptive control input (7.4), the closed-loop system dynamics
are described by
e + am e = bp ( a y y a r r),

(7.5)

where a r = ar a r and a y = ay a y . Consider the Lyapunov function candidate

V =

1 2 |bp | 2 |bp | 2
a +
a ,
e +
2
2r r
2y y

(7.6)

where r and y are constant positive real design parameters. Its derivative along the
trajectory (7.5) is given by




y
r
a
a
V = am e2 + a r |bp | ebp r + a y |bp | ebp y .
r
y
If we can set
a r
|bp | ebp r = 0,
r
a y
|bp | ebp y = 0,
y

(7.7)
(7.8)

we have
V = am e2 .

(7.9)

Noting that a r = a r and a y = a y , the conditions in (7.7) and (7.8) can be satised
by setting the adaptive laws as
a r = sgn(bp )r er,

(7.10)

a y = sgn(bp )y ey.

(7.11)

The positive real design parameters r and y are often referred to as adaptive gains,
as they can affect the speed of parameter adaptation.
From (7.9) and Theorem 4.2, we conclude that the system is Lyapunov stable
with all the variables e, a r and a y bounded, and hence the boundedness of a r and a y .
However, based on the stability theorems introduced in Chapter 4, we cannot
conclude anything about the tracking error e other than its boundedness. In order
to do it, we need to introduce an important lemma for stability analysis of adaptive
control systems.

92

Nonlinear and adaptive control systems

Lemma 7.1 (Barbalats

 lemma). If a function f (t) : R R is uniformly continuous
for t [0, ), and 0 f (t)dt exists, then limt f (t) = 0.
From (7.9), we can show that

V (0) V ()
e2 (t)dt =
< .
am
0

(7.12)

Therefore, we have established that e L2 L and e L . Since e and e are

bounded, e2 is uniformly continuous. Therefore, we can conclude from Barbalats
lemma that limt e2 (t) = 0, and hence limt e(t) = 0.
We summarise the stability result in the following lemma.
Lemma 7.2. For the rst-order system (7.1) and the reference model (7.2), the adaptive control input (7.4) together with the adaptive laws (7.10) and (7.11) ensures the
boundedness of all the variables in the closed-loop system, and the convergence to
zero of the tracking error.
Remark 7.1. The stability analysis ensures the convergence to zero of the tracking
error, but nothing can be told about the convergence of the estimated parameters. The
estimated parameters are assured to be bounded from the stability analysis. In general,
the convergence of the tracking error to zero and the boundedness of the adaptive
parameters are stability results that we can establish for MRAC. The convergence
of the estimated parameters may be achieved by imposing certain conditions of the
reference signal to ensure the system is excited enough. This is similar to the concept
of persistent excitation for system identication.

Example 7.1. Consider a rst-order system
b
,
s+a
where b = 1 and a is an unknown constant parameter. We will design an adaptive
controller such that the output of the system follows the output of the reference model
Gp =

1
.
s+2
We can directly use the result presented in Lemma 7.2, i.e., we use the adaptive
laws (7.10) and (7.11) and the control input (7.4). Since b is known, we only have one
unknown parameter, and it is possible to design a simpler control based on the same
design principle.
From the system model, we have
Gm =

y + ay = u,
which can be changed to
y + 2y = u (a 2)y.

Adaptive control of linear systems

93

Subtracting the reference model

y m + 2ym = r,
we obtain that
e + 2e = u ay y r.
where ay = a 2. We then design the adaptive law and control input as
a y = y ey,
u = a y y + r.
The stability analysis follows the same discussion that leads to Lemma 7.2. Simulation
study has been carried out with a = 1, = 10 and r = 1. The simulation results are
shown in Figure 7.1. The gure shows that the estimated parameter converges to the
true value ay = 3. The convergence of the estimated parameters is not guaranteed
by Lemma 7.2. Indeed, some strong conditions on the input or reference signal are
needed to generate enough excitation for the parameter estimation to achieve the
convergence of the estimated parameters in general.


y
ym

y and ym

0.8
0.6
0.4
0.2
0
0

10

10

t (s)

Estimated parameter

0
1
2
3
4

4
t (s)

Figure 7.1 Simulation results of Example 7.1

94

Nonlinear and adaptive control systems

7.2 Model reference control

It is clear from MRAC design for rst-order systems that an MRC input is designed
rst which contains unknown parameters, and the adaptive control input is then
obtained based on the certainty equivalence principle. Hence, MRC design is the
rst step for MRAC. Furthermore, MRC itself deserves a brief introduction, as it is
different from the classical control design methods shown in standard undergraduate
texts. In this section, we will start with MRC for systems with relative degree 1, and
then move on to MRC of systems with high-order relative degrees.
Consider an nth-order system with the transfer function
y(s) = kp

Zp (s)
u(s),
Rp (s)

(7.13)

where y(s) and u(s) denote the system output and input in frequency domain; kp is the
high frequency gain; and Zp and Rp are monic polynomials with orders of n and
n respectively with as the relative degree. The reference model is chosen to have
the same relative degree of the system, and is described by
ym (s) = km

Zm (s)
r(s),
Rm (s)

(7.14)

where ym (s) is the reference output for y(s) to follow; r(s) is a reference input; and
km > 0 and Zm and Rm are monic Hurwitz polynomials.
Remark 7.2. A monic polynomial is a polynomial whose leading coefcient, the
coefcient of the highest power, is 1. A polynomial is said to be Hurwitz if all its roots
are with negative real parts, i.e., its roots locate in the open left half of the complex
plane. The high-frequency gain is the leading coefcient of the numerator of a transfer
function.

The objective of MRC is to design a control input u such that the output of the
system asymptotically follows the output of the reference model, i.e., limt (y(t)
ym (t)) = 0.
Note that in this chapter, we abuse the notations of y, u and r by using same
notations for the functions in time domain and their Laplace transformed functions
in the frequency domain. It should be clear from the notations that y(s) is the Laplace
transform of y(t) and similarly for u and r.
To design MRC for systems with = 1, we follow a similar manipulation to the
rst-order system by manipulating the transfer functions. We start with
y(s)Rp (s) = kp Zp (s)u(s)
and then
y(s)Rm (s) = kp Zp (s)u(s) (Rp (s) Rm (s))y(s).

Adaptive control of linear systems

95

R (s) R (s)

Note that Rp (s) Rm (s) is a polynomial with order n 1, and m Zm (s) p is a proper
transfer function, as Rp (s) and Rm (s) are monic polynomials. Hence, we can write


Zp (s)
Rm (s) Rp (s)
y(s)Rm (s) = kp Zm (s)
u(s) +
y(s) .
Zm (s)
Zm (s)
If we parameterise the transfer functions as
Zp (s)
T (s)
= 1 1
,
Zm (s)
Zm (s)
Rm (s) Rp (s)
T (s)
y(s) = 2
y(s) 3 ,
Zm (s)
Zm (s)
where 1 Rn1 , 2 Rn1 and 3 R are constants and
(s) = [sn2 , . . . , 1]T ,
we obtain that


Zm (s)
1T (s)
2T (s)
u(s)
u(s)
y(s) 3 y(s) .
y(s) = kp
Rm (s)
Zm (s)
Zm (s)

(7.15)

Hence, we have the dynamics of tracking error given by



Zm (s)
1T (s)
2T (s)
u(s)
u(s)
y(s) 3 y(s) 4 r ,
e1 (s) = kp
Rm (s)
Zm (s)
Zm (s)

(7.16)

where e1 = y ym and 4 = kkmp .

The control input for MRC is given by
u(s) =

T (s)
1T (s)
u(s) + 2
y + 3 y + 4 r(s)
Zm (s)
Zm (s)

:= T ,
where
T = [1T , 4T , 3 , 4 ],
= [1T , 2T , y, r]T ,
with
1 =

(s)
u,
Zm (s)

2 =

(s)
y.
Zm (s)

(7.17)

96

Nonlinear and adaptive control systems

Remark 7.3. The control design shown in (7.17) is a dynamic feedback con(s)
troller. Each element in the transfer matrix
is strictly proper, i.e.,
Zm (s)
with relative degree greater than or equal to 1. The total number of parameters in
equals 2n.

Lemma 7.3. For the system (7.13) with relative degree 1, the control input (7.17)
solves MRC problem with the reference model (7.14) and limt (y(t) ym (t)) = 0.
Proof. With the control input (7.17), the closed-loop dynamics are given by
e1 (s) = kp

Zm (s)
(s),
Rm (s)

where (s) denotes exponentially convergent signals due to non-zero initial values.
The reference model is stable, and then the track error e1 (t) converges to zero
exponentially.
2
Example 7.2. Design MRC for the system
y(s) =

s+1
u(s)
s2 2s + 1

with the reference model

ym (s) =

s2

s+3
r(s).
+ 2s + 3

We follow the procedures shown early to obtain the MRC control. From the transfer
function of the system, we have
y(s)(s2 + 2s + 3) = (s + 1)u(s) + (4s + 2)y(s),
which leads to
y(s) =

s+3
2
s + 2s + 3

s+1
4s + 2
u(s) +
y(s)
s+3
s+3



10
s+3
2
u(s)
y(s) + 4y(s) .
= 2
u(s)
s + 2s + 3
s+3
s+3
Subtracting it by the reference model, we have


s+3
2
10
e1 (s) = 2
u(s)
u(s)
y(s) + 4y(s) r(s) ,
s + 2s + 3
s+3
s+3
which leads to the MRC control input
u(s) =

2
10
u(s) +
y(s) 4y(s) + r(s)
s+3
s+3

= [2 10 4 1][1 (s) 2 (s) y(s) r(s)]T ,

Adaptive control of linear systems

97

where
1
u(s),
s+3
1
2 (s) =
y(s).
s+3
1 (s) =

Note that the control input in the time domain is given by

u(s) = [2 10 4 1][1 (t) 2 (t) y(t) r(t)]T ,
where
1 = 31 + u,
2 = 32 + y.

For a system with > 1, the input in the same format as (7.17) can be obtained.
The only difference is that Zm is of order n < n 1. In this case, we let P(s) be a
monic and Hurwitz polynomial with order 1 so that Zm (s)P(s) is of order n 1.
We adopt a slightly different approach from the case of = 1.
Consider the identity


Zm (s) Rm (s)P(s)
y(s) =
y(s)
Rm (s) Zm (s)P(s)


Zm (s) Q(s)Rp (s) + (s)
=
y(s) .
(7.18)
Rm (s)
Zm (s)P(s)
Note that the second equation in (7.18) follows from the identity
Rm (s)P(s) = Q(s)Rp (s) + (s),
where Q(s) is a monic polynomial with order n 1, and (s) is a polynomial
with order n 1. In fact Q(s) can be obtained by dividing Rm (s)P(s) by Rp (s) using
long division, and (s) is the remainder of the polynomial division. From the transfer
function of the system, we have
Rp (s)y(s) = kp Zp (s)u(s).
Substituting it into (7.18), we have


kp1 (s)
Zm (s) Q(s)Zp (s)
y(s) = kp
u+
y(s) .
Rm (s) Zm (s)P(s)
Zm (s)P(s)
Similar to the case for = 1, if we parameterise the transfer functions as
Q(s)Zp (s)
T (s)
= 1 1
,
Zm (s)P(s)
Zm (s)P(s)
kp1 
Zm (s)P(s)

2T (s)
3
Zm (s)P(s)

98

Nonlinear and adaptive control systems

where 1 Rn1 and 2 Rn1 and 3 R are constants and

(s) = [sn2 , . . . , 1]T ,
we obtain that
y(s) = kp



Zm (s)
T (s)
T (s)
u(s) 1
u(s) 2
y(s) 3 y(s) .
Rm (s)
Zm (s)P(s)
Zm (s)P(s)

Hence, we have the dynamics of tracking error given by



Zm (s)
T (s)
T (s)
u(s) 1
u(s) 2
y(s) 3 y(s) 4 r ,
e1 (s) = kp
Rm (s)
Zm (s)P(s)
Zm (s)P(s)
where e1 = y ym and 4 =
u =

km
. The
kp

control input is designed as

1T (s)
T (s)
u+ 2
y + 3 y + 4 r
Zm (s)P(s)
Zm (s)P(s)

:= T

(7.19)

with the same format as (7.17) except

1 =

(s)
u,
Zm (s)P(s)

2 =

(s)
y.
Zm (s)P(s)

Remark 7.4. The nal control input is in the same format as shown for the case
= 1. The lters for w1 and w2 are in the same order as in the case for = 1, as the

order of Zm (s)P(s) is still n 1.
Lemma 7.4. For the system (7.13) with relative degree > 1, the control input (7.19)
solves MRC problem with the reference model (7.14) and limt (y(t) ym (t)) = 0.
The proof is the same as the proof for Lemma 7.3.
Example 7.3. Design MRC for the system
y(s) =

s2

1
u
2s + 1

with the reference model

ym (s) =

1
r.
s2 + 2s + 3

The relative degree of the system is 2. We set P = s + 1. Note that

(s2 + 2s + 3)(s + 1) = (s + 5)(s2 2s + 1) + (14s 2).

Adaptive control of linear systems

99

From the reference model, we have

y(s) =

1
s2 + 2s + 3


(s + 5)(s2 2s + 1)y(s) + (14s 2)y(s)
s+1


1
(s + 5)u(s) + (14s 2)y(s)
= 2
s + 2s + 3
s+1


1
4
1
= 2
u(s) +
u(s) 16
y(s) + 14y(s) .
s + 2s + 3
s+1
s+1

1
2
s + 2s + 3

(s2 + 2s + 3)(s + 1)
y(s)
s+1

The dynamics of the tacking error are given by



1
4
1
e1 (s) = 2
u(s) +
u(s) 16
y(s) + 14y(s) r(s) .
s + 2s + 3
s+1
s+1
We can then design the control input as
u = [1, 16, 14, 1][1 , 2 , y, r]T
with
1
u,
s+1
1
2 =
y.
s+1
1 =

7.3 MRAC of linear systems with relative degree 1

Adaptive control deals with uncertainties in terms of unknown constant parameters.
It may be used to tackle some changes or variations in model parameters in adaptive
control application, but the stability analysis will be carried under the assumption the
parameters are constants. There are other common assumptions for adaptive control
which are listed below:

the known system order n

the known relative degree
the minimum phase of the plant
the known sign of the high frequency gain sgn(kp )

In this section, we present MRAC design for linear systems with relative
degree 1.
Consider an nth-order system with the transfer function
y(s) = kp

Zp (s)
u(s),
Rp (s)

(7.20)

100

Nonlinear and adaptive control systems

where y(s) and u(s) denote the system output and input in frequency domain; kp is
the high frequency gain; and Zp and Rp are monic polynomials with orders of n 1
and n respectively. This system is assumed to be minimum phase, i.e., Zp (s) is a
Hurwitz polynomial, and the sign of the high-frequency gain, sgn(kp ), is known. The
coefcients of the polynomials and the value of kp are constants and unknown. The
reference model is chosen to have the relative degree 1 and strictly positive real, and
is described by
ym (s) = km

Zm (s)
r(s),
Rm (s)

(7.21)

where ym (s) is the reference output for y(s) to follow, r(s) is a reference input, and
Zm (s) and Rm (s) are monic polynomials and km > 0. Since the reference model is
strictly positive real, Zm and Rm are Hurwitz polynomials.
MRC shown in the previous section gives the control design in (7.17). Based on
the certainty equivalence principle, we design the adaptive control input as
u(s) = T ,

(7.22)

where is an estimate of the unknown vector R2n , and is given by

= [1T , 2T , y, r]T
with
1 =

(s)
u,
Zm (s)

2 =

(s)
y.
Zm (s)

With the designed adaptive control input, it can be obtained, from the tracking
error dynamics shown in (7.16), that
Zm (s) T
( T )
Rm (s)


kp T
Zm (s)

= km

Rm (s)
km

e1 (s) = kp

(7.23)

where = .
To analyse the stability using a Lyapunov function, we put the error dynamics in
the state space form as


k
e = Am e + bm kmp T
(7.24)
e1 = cmT e
where (Am , bm , cm ) is a minimum state space realisation of km RZmm (s)
, i.e.,
(s)
cmT (sI Am )1 bm = km

Zm (s)
.
Rm (s)

Adaptive control of linear systems

101

Since (Am , bm , cm ) is a strictly positive real system, from KalmanYakubovich lemma

(Lemma 5.4), there exist positive denite matrices P and Q such that
ATm Pm + Pm Am = Qm ,
Pm bm = cm .

(7.25)
(7.26)

Dene a Lyapunov function candidate as

1 T
1

kp

T 1

V = e Pm e +

,
2
2 km
where R2n is a positive denite matrix. Its derivative is given by



kp

kp T
1 T T
T

V = e (Am Pm + Pm Am )e + e Pm bm +

T 1
2
km
km
Using the results from (7.25) and (7.26), we have



kp

kp
1
V = eT Qm e + e1 T +

T 1
2
km
km

kp

1
= eT Qm e +

T 1 sgn(kp )e1 .
2
km
Hence, the adaptive law is designed as
= sgn(kp ) e1 ,

(7.27)

which results in
1
V = eT Qm e.
2
Furthermore it can be shown that
We can now conclude the boundedness of e and .
e L2 and e 1 L . Therefore, from Barbalats lemma we have limt e1 (t) = 0.
The boundedness of other system state variables can be established from the minimumphase property of the system.
We summarise the stability analysis for MRAC of linear systems with relative
degree 1 in the following theorem.
Theorem 7.5. For the rst-order system (7.20) and the reference model (7.21), the
adaptive control input (7.22) together with the adaptive law (7.27) ensures the boundedness of all the variables in the closed-loop system, and the convergence to zero of
the tracking error.
Remark 7.5. The stability result shown in Theorem 7.5 only guarantees the convergence of the tracking error to zero, not the convergence of the estimated parameters.
In the stability analysis, we use KalmanYakubovich lemma for the denition of
Lyapunov function and the stability proof. That is why we choose the reference model
to be strictly positive real. From the control design point of view, we do not need
to know the actual values of Pm and Qm , as long as they exist, which is guaranteed

102

Nonlinear and adaptive control systems

by the selection of a strictly positive real model. Also it is clear from the stability
analysis, that the unknown parameters must be constant. Otherwise, we would not

have = .


7.4 MRAC of linear systems with high relatives

In this section, we will introduce adaptive control design for linear systems with their
relative degrees higher than 1. Similar to the case for relative degree 1, the certainty
equivalence principle can be applied to the control design, but the designs of the
adaptive laws and the stability analysis are much more involved, due to the higher
relative degrees. One difculty is that there is not a clear choice of Lyapunov function
candidate as in the case of = 1.
Consider an nth-order system with the transfer function
y(s) = kp

Zp (s)
u(s),
Rp (s)

(7.28)

where y(s) and u(s) denote the system output and input in frequency domain, kp is
the high frequency gain, Zp and Rp are monic polynomials with orders of n and
n respectively, with > 1 being the relative degree of the system. This system is
assumed to be minimum phase, i.e., Zp (s) is Hurwitz polynomial, and the sign of the
high-frequency gain, sgn(kp ), is known. The coefcients of the polynomials and the
value of kp are constants and unknown. The reference model is chosen as
ym (s) = km

Zm (s)
r(s)
Rm (s)

(7.29)

where ym (s) is the reference output for y(s) to follow; r(s) is a reference input; and
Zm (s) and Rm (s) are monic polynomials with orders n and n respectively and
km > 0. The reference model (7.29) is required to satisfy an additional condition that
there exists a monic and Hurwitz polynomial P(s) of order n 1 such that
ym (s) = km

Zm (s)P(s)
r(s)
Rm (s)

(7.30)

is strictly positive real. This condition also implies that Zm and Rm are Hurwitz
polynomials.
MRC shown in the previous section gives the control design in (7.19). We design
the adaptive control input, again using the certainty equivalence principle, as
u = T ,
where is an estimate of the unknown vector R2n , and is given by
= [1T , 2T , y, r]T

(7.31)

Adaptive control of linear systems

103

with
1 =

(s)
u,
Zm (s)P(s)

2 =

(s)
y.
Zm (s)P(s)

The design of adaptive law is more involved, and we need to examine the dynamics
of the tacking error, which are given by
Zm
(u T )
Rm

Zm P(s) 
= km
k(uf T ) ,
Rm

e1 = kp

(7.32)

where
k=

kp
,
km

uf =

1
u
P(s)

and =

1
.
P(s)

An auxiliary error is constructed as

Zm P(s) 
Zm P(s)  2 
k(uf T ) km
ns ,
 = e1 km
Rm
Rm

(7.33)

where k is an estimate of k, n2s = T + uf2 . The adaptive laws are designed as

= sgn(bp ) ,

(7.34)

k = (uf T ).

(7.35)

With these adaptive laws, a stability result can be obtained for the boundedness of
parameter estimates and the convergence of the tracking error. For the completeness,
we state the theorem below without giving the proof.
Theorem 7.6. For the system (7.28) and the reference model (7.29), the adaptive
control input (7.31) together with the adaptive laws (7.34) and (7.35) ensures the
boundedness of all the variables in the closed-loop system, and the convergence to
zero of the tracking error.

7.5 Robust adaptive control

Adaptive control design and its stability analysis have been carried out under the
condition that there is only parametric uncertainty in the system. However, many
types of non-parametric uncertainties do exist in practice. These include

high-frequency unmodelled dynamics, such as actuator dynamics or structural

vibrations
low-frequency unmodelled dynamics, such as Coulomb frictions

104

Nonlinear and adaptive control systems

measurement noise
computation roundoff error and sampling delay

Such non-parametric uncertainties will affect the performance of adaptive control

systems when they are applied to practical systems. They may cause instability. The
difference between adaptive control of linear systems and other design methods is
parameter estimation. Without specic requirement for input signals, such as persistent excitation, we can only establish the boundedness of estimated parameters, and
the asymptotic convergence to zero of the output error. Therefore, the closed-loop
adaptive system is not even asymptotically stable. For other design methods for linear
systems without parameter adaptation, the closed-loop systems are normally exponentially stable. For a linear system with exponential stability, the state inherently
remain bounded under any bounded input. This is not the case for a linear system
under adaptive control, due to the difference in the stability properties.
Many non-parametric uncertainties can be represented by a bounded disturbance
to the nominal system. Even a bounded disturbance can cause serious problem in
parameter adaptation. Let us consider a simple example.
Consider the system output is described by
y = .

(7.36)

The adaptive law

= ,

(7.37)

where
 = y
will render the convergence of the estimate by taking
V =

1 2

as a Lyapunov function candidate and the analysis

)
V = (y
= 2 2 .

(7.38)

The boundedness of can then be concluded, no matter what the signal is.
Now, if the signal is corrupted by some unknown bounded disturbance d(t),
i.e.,
y = + d(t).

Adaptive control of linear systems

105

the same adaptive will have a problem. In this case,

V = (y )
= ( + d )
= 2 2 d
2 2
d2
1
=
( + d)2 + .
2
2
2
From the above analysis, we cannot conclude the boundedness of even though is
bounded. In fact, if we take = 2, = 1 and = (1 + t)1/2 L and let


5
d(t) = (1 + t)1/4
2(1 + t)1/4 ,
4
it can then be obtained that
5
y(t) = (1 + t)1/4 , 0 as t ,
4
= 5 (1 + t)3/4 (1 + t)1
4
which has a solution
= (1 + t)1/4 as t .

(7.39)

In this example, we have observed that adaptive law designed for the disturbancefree system fails to remain bounded even though the disturbance is bounded and
converges to zero as t tends to innity.
Remark 7.6. If is a constant, then from (7.38) we can show that the estimate
exponentially converges to the true value. In this case, there is only one unknown
parameter. If is a vector, the requirement for the convergence is much stronger. In
the above example, is bounded, but not in a persistent way. It does demonstrate that
even a bounded disturbance can cause the estimated parameter divergent.

Robust adaptive control issue is often addressed by modifying parameter adaptive
laws to ensure the boundedness of estimated parameters. It is clear from the example
shown above that bounded disturbance can cause estimated parameters unbounded.
Various robust adaptive laws have been introduced to keep estimated parameters
bounded in the presence of bounded disturbances. We will show two strategies using
the following simple model:
y = + d(t)

(7.40)

and
with d as a bounded disturbance. In the following we keep using  = y
2
V = 2 . Once the basic ideas are introduced, it is not difcult to extend the robust
adaptive laws to adaptive control of dynamic systems.
Dead-zone modication is a modication to the parameter adaptive law to stop
parameter adaptation when the error is very close to zero. The adaptive law is
modied as

106

Nonlinear and adaptive control systems

=


0

|| > g
|| g

(7.41)

where g is a constant satisfying g > |d(t)| for all t. For  > g, we have
V = 
= ( )
= (y d(t) )
= ( d(t))
< 0.
Therefore, we have

< 0, || > g

V
= 0, || g
and we can conclude that V is bounded. Intuitively, when the error  is small, the
bounded disturbance can be more dominant, and therefore, the correct adaptation
direction is corrupted by the disturbance. In such a case, a simple strategy would be
just to stop parameter adaptation. The parameter adaptation stops in the range || g,
and for this reason, this modication takes the name dead-zone modication. The
size of the dead zone depends on the size of the bounded disturbances. One problem
with the dead-zone modication is that the adaptive law is discontinuous, and this
may not be desirable in some applications.
-Modication is another strategy to ensure the boundedness of estimated parameters. The adaptive law is modied by adding an additional term to the normal
adaptive law as
= 

(7.42)

where is a positive real constant. In this case, we have

V = ( d(t)) +
=  2 + d(t) 2 +
2
d2
2
2
+ 0
+
2
2
2
2
2
d02
+ ,
V +
2
2

(7.43)

where d0 |d(t)|, t 0. Applying Lemma 4.5 (comparison lemma) to (7.43),

we have
 2

 t
d0
2
t
(t )
V (t) e
+ V (0)
e
+
d ,
2
2
0

Adaptive control of linear systems

107

and therefore we can conclude that V L , which implies the boundedness of the
estimated parameter. A bound can be obtained for the bounded parameter as


2
1 d02
+
.
(7.44)
V ()
2
2
Note that this modication does not need a bound for the bounded disturbances, and
also it provides a continuous adaptive law. For these reasons, -modication is one
of the most widely used modications for parameter adaptation.
Remark 7.7. We re-arrange the adaptive law (7.42) as
+ = .
Since ( ) is a positive constant, the adaptive law can be viewed as a stable rst-order
dynamic system with () as the input and as the output. With a bounded input,
obviously remains bounded.

The robust adaptive laws introduced here can be applied to various adaptive
control schemes. We demonstrate the application of a robust adaptive law to MRAC
with = 1. We start directly from the error model (7.24) with an additional bounded
disturbance
e = Am e + bm (k T + d(t))
e1 = cmT e,

(7.45)

where k = kp /km and d(t) are a bounded disturbance with the bound d0 , which represents the non-parametric uncertainty in the system. As discussed earlier, we need a
robust adaptive law to deal with the bounded disturbances. If we take -modication,
then the robust adaptive law is
= sgn(kp ) e1 .

(7.46)

We will show that this adaptive law will ensure the boundedness of the variables.
Let
V =

1 T
1
e Pe + |k| T 1 .
2
2

Similar to the analysis leading to Theorem 7.5, the derivative of V is obtained as

1
V = eT Qe + e1 ( k T + d) + |k| T 1
2
1
min (Q)
e
2 + e1 d + |k| T
2
1
min (Q)
e
2 + |e1 d| |k|

2 + |k| T .
2

108

Nonlinear and adaptive control systems

Note that
1
d02
min (Q)
e
2 +
,
4
min (Q)
1
1
| T |

2 +

2 .
2
2
Hence, we have
|e1 d|

|k|
d02
|k|
1

2 +
+

2
V min (Q)
e
2
4
2
min (Q)
2
V +

|k|
d02
+

2 ,
min (Q)
2

where is a positive real and

=

min{(1/2)min (Q), |k| }

.
max{max (P), |k|/min ( )}

Therefore, we can conclude the boundedness of V from Lemma 4.5 (comparison

lemma), which further implies the boundedness of the tracking error e1 and the

estimate .
From the above analysis, it is clear that the adaptive law with -modication
ensures the boundedness of all the variables in the closed-loop adaptive control system.
It is worth noting that the output tracking error e1 will not asymptotically converge
to zero, even though the bounded disturbance d(t) becomes zero. That is the price to
pay for the robust adaptive scheme.

Chapter 8

Nonlinear observer design

Observers are needed to estimate unmeasured state variables of dynamic systems.

They are often used for output feedback control design when only the outputs are
available for the control design. Observers can also be used for other estimation purposes such as fault detection and diagnostics. There are many results on nonlinear
observer design in literature, and in this chapter, we can introduce only a number of
results. Observer design for linear systems is briey reviewed before the introduction
of observers with linear error dynamics. We then introduce another observer design
method based on Lyapunovs auxiliary theorem, before the observer design for systems with Lipschitz nonlinearities. At the end of this chapter, adaptive observers are
briey described.

8.1 Observer design for linear systems

We briey review the results for linear systems. Consider
x = Ax
y = Cx,

(8.1)

where x Rn is the state; y Rm is the system output with m < n; and A Rnn
and C Rmn are constant matrices. From linear system theory, we know that this
system, or the pair (A, C), is observable if the matrix

C
CA

Po = .
..
CAn1
has rank n. The observability condition is equivalent to that the matrix


I A
C
has rank n for any value of C.
When the system is observable, an observer can be designed as
x = Ax + L(y C x ),

(8.2)

110

Nonlinear and adaptive control systems

where x Rn is the estimate of the state x, and L Rnm is the observer gain such
that (A LC) is Hurwitz.
For the observer (8.2), it is easy to see the estimate x converges to x asymptotically. Let x = x x , and we can obtain
x = (A LC)x.
Remark 8.1. In the observer design, we do not consider control input terms in the
system in (8.1), as they do not affect the observer design for linear systems. In fact, if
Bu term is added to the right-hand side of the system (8.1), for the observer design,
we can simply add it to the right-hand side of the observer in (8.2), and the observer
error will still converge to zero exponentially.

Remark 8.2. The observability condition for the observer design of (8.1) can be
relaxed to the detectability of the system, or the condition that (A, C) is detectable,
for the existence of an observer gain L such that (A LC) is Hurwitz. Detectability is
weaker than observability, and it basically requires the unstable modes of the system
observable. The pair (A, C) is detectable if the matrix


I A
C
has rank n for any in the closed right half of the complex plan. Some other design
methods shown in this chapter also need only the condition of detectability, although,
for simplicity, we state the requirement for the observability.

There is another approach to full-state observer design for linear systems (8.1).
Consider a dynamic system
z = Fz + Gy,

(8.3)

where z Rn is the state; F Rnn is Hurwitz; and G Rnm . If there exists an

invertible matrix T Rnn such that Z converges to Tx, then (8.3) is an observer with
the state estimate given by x = T 1 z. Let
e = Tx z.
A direction evaluation gives
e = TAx (Fz + GCx)
= F(Tx z) + (TA FT GC)x
= Fe + (TA FT GC)x.
If we have
TA FT GC = 0,
the system (8.3) is an observer with an exponentially convergent estimation error. We
have the following lemma to summarise this observer design.

Nonlinear observer design

111

Lemma 8.1. The dynamic system (8.3) is an observer for the system (8.1) if and only
if F is Hurwitz and there exists an invertible matrix T such that
TA FT = GC.

(8.4)

Proof. The sufciency has been shown in the above analysis. For necessity, we only
need to observe that if any of the conditions is not satised, we cannot guarantee the convergence of e to zero for a general linear system (8.1). Indeed, if (8.4)
is not satised, then e will be a state variable with a non-zero input, and we can set
up a case such that e does not converge to zero. So does for the condition that F is
Hurwitz.
2
How to nd matrices F and G such that the condition (8.4) is satised? We list
the result in the following lemma without the proof.
Lemma 8.2. Suppose that F and A have exclusively different eigenvalues. The necessary condition for the existence of a non-singular solution T to the matrix equation
(8.4) is that the pair (A, C) is observable and the pair (F, G) is controllable. This
condition is also sufcient when the system (8.1) is single output, i.e., m = 1.
This lemma suggests that we can choose a controllable pair (F, G) and make sure
that the eigenvalues of F are different from those of A. An observer can be designed if
there is a solution of T from (8.4). For single output system, the solution is guaranteed.

8.2 Linear observer error dynamics with output injection

Now consider
x = Ax + (y, u)
y = Cx,

(8.5)

where x Rn is the state, y Rm is the system output with m < n, u Rs is the

control input, or other known variables, A Rnn and C Rmn are constant matrices
and : Rm Rs Rn is a continuous function. This system is a nonlinear system.
However, comparing with the system (8.1), the only difference is the additional term
(y, u). The system (8.5) can be viewed as the linear system (8.1) perturbed by the
nonlinear term (y, u).
If the pair (A, C) is observable, we can design an observer as
x = Ax + L(y C x ) + (y, u),

(8.6)

where x Rn is the estimate of the state x and L Rnm is the observer gain such
that (A LC) is Hurwitz. The only difference between this observer and the one in
(8.2) is the nonlinear term (y, u). It can be seen that the observer error still satises
x = (A LC)x.

112

Nonlinear and adaptive control systems

Note that even the system (8.5) is nonlinear, the observer error dynamics are
linear. The system in the format of (8.5) is referred to as the system with linear
observer errors. More specically, if we drop the control input u in the function ,
the system is referred to as the output injection form for observer design.
Let us summarise the result in the following proposition.
Proposition 8.3. For the nonlinear system (8.5), a full-state observer can be designed
as in (8.6) if (A, C) is observable. Furthermore, the observer error dynamics are linear
and exponentially stable.
For a nonlinear system in a more general form, there may exist a state transformation to put the system in the format of (8.5), and then the proposed observer can
be applied.
We will introduce the conditions for the existence of a nonlinear state transformation to put the system in the format shown in (8.6). Here, we only consider single
output case, and for the simplicity, we do not consider the system with a control input.
The system under consideration is described by
x = f (x)
y = h(x),

(8.7)

where x Rn is the state vector, y R is the output, f : Rn Rn and h : Rn R

are continuous nonlinear functions with f (0) = 0, and h(0) = 0. We will show that
under what conditions there exists a state transformation
z = (x),

(8.8)

where  : Rn Rn , such that the transformed system is in format

z = Az + (Cz)
y = Cz,

(8.9)

where A Rnn , C R1n , (A, C) is observable and : R Rn . Without loss of

generality, we take the transformed system (8.9) as
z1 = z2 + 1 (y)
z2 = z3 + 2 (y)
...
zn1 = zn + n1 (y)
zn = n (y)
y = z1 .

(8.10)

This system is in the output injection form for nonlinear observer design.
Remark 8.3. Assume that the system (8.9) is different from (8.10) with (A, C) as
a general observable pair, instead of having the special format implied by (8.10).

Nonlinear observer design

113

In this case, we use a different variable z to denote the state for (8.10), and it can be
written as
z + (y)

z = A
(8.11)
y = C z .
It is clear that there exists a linear transformation
z = Tz,
which transforms the system (8.9) to the system (8.11), because (A, C) is observable.
The transformation from (8.7) to (8.11) is given by

z = T (x) := (x).
Therefore, if there exists a nonlinear transformation from (8.7) to (8.9), there must
exist a nonlinear transformation from (8.7) to (8.11), and vise versa. That is why
we can consider the transformed system in the format of (8.10) without loss of
generality.

If the state transformation transforms the system (8.7) to (8.9), we must have


(x)
f (x)
= Az + (Cz)
x
x=(z)
h((z)) = Cz,

where  = 1 . For notational convenience, let us denote

f (z) = Az + (Cz)

h(z)
= Cz.
From the structure shown in (8.10), we can obtain

h(z)
= z1 ,

Lf h(z)
= z2 + 1 (z1 ),
1
(z2 + 1 (z1 ))
z1
:= z3 + 2 (z1 , z2 ),

Lf2 h(z)
= z3 +
...

Lfn1
h(z) = zn +

n2

n2
k=1

zk

(zk+1 + k (z1 ))

:= zn + n1 (z1 , . . . , zn1 ).

(8.12)

114

Nonlinear and adaptive control systems

From the above expression, it is clear that

h
z

Lf h(z)

..

n1
Lf h(z)

=.
.
.

0
1
..
.

0
0

..
.

...
...
..
.

...

z
This implies that
dLf h,
. . . , dLn1 h
d h,
f
are linearly independent. This property is invariant under state transformation, and
therefore, we need the condition under the coordinate x, that is
dh, dLf h, . . . , dLfn1 h
are linearly independent. Indeed, we have

Lfn1 h(x) = (Lfn1

h(z))z=(x) ,
and therefore

h(x)
x
Lf h(x)
x
..
.

Ln1 h(x)
f
x

h(z)
z

Lf h(z)

=
z

..

.

n1
Lf h(z)
z

for k = 0, 1, . . . , n 1

(x)
.
x

z=(x)

The linear independence of dh, dLf h, . . . , dLfn1 h is a consequence of the observability of (A, C) in (8.9). In some literature, this linear independence condition is
dened as the observability condition for nonlinear system (8.7). Unlike linear systems, this condition is not enough to design a nonlinear observer. We state necessary
and sufcient conditions for the transformation to the output injection form in the
following theorem.
Theorem 8.4. The nonlinear system (8.7) can be transformed to the output injection
form in (8.10) if and only if

the differentials dh, dLf h, . . . , dLfn1 h are linearly independent

there exists a map  : Rn Rn such that

Nonlinear observer design


(z)  n1
,
= adf r, . . . , adf r, r
x=(z)
z

115
(8.13)

where r is a vector eld solved from

dh
dLf h
..
.

dLfn1 h

0
..

= . .
0
1

(8.14)

Proof. Sufciency. From the rst condition and (8.14), we can show, in a similar way
as for (6.12), that

dh(x)
dLf h(x)


n1
..

adf r(x) . . . adf r(x) r(x)

.
dLf1 h

=.
..

0
1
..
.

...
...
..
.

...

0
0

.. .
.

(8.15)

n1

r(x) . . . adf r(x) r(x) has full rank. This implies that there exists
Therefore, adf
an inverse mapping for . Let us denote it as  =  1 , and hence we have

(x) n1
adf r(x) . . . adf r(x) r(x) = I .
(8.16)
x
Let us dene the state transformation as z = (x) and denote the functions after this
transformation as


f (z) = (x) f (x)
,
x
x=(z)

h(z)
= h((z)).
We need to show that the functions f and h are in the format of the output injection
form as in (8.10). From (8.16), we have
(x) nk
adf r(x) = ek ,
x

for k = 1, . . . , n,

where ek denotes the kth column of the identity matrix. Hence, we have, for
k = 1, . . . , n 1,

116

Nonlinear and adaptive control systems



(x) nk
adf r(x)
x


=

x=(z)

(x)
n(k+1)
[ f (x), adf
r(x)]
x

x=(z)

(x)
(x) n(k+1)
=
r(x)
f (x),
adf
x
x


= f (z), ek+1
=

f (z)
.
zk+1

This implies that

f (z)
= ek ,
zk+1
i.e.,

for k = 1, . . . , n 1,

f (z) .
= ..

1
0
..
.

0
1
..
.

0
0

0
0

0
0

.. .
.

1
0

...
...
..
.

...
...

Therefore, we have shown that f is in the output injection form.

From the second condition, we have
(z)
nk
= [adf
(x) r(x)]x=(z)
zk

for k = 1, . . . , n.

Hence, we obtain that, for k = 1, . . . , n,



h(z)
(z)
h(x)
=
zk
x x=(z) zk


h(x)
nk
=
[adf
(x) r(x)]x=(z)
x x=(z)
= [Lad nk

f (x) r(x)

h(x)]x=(z) .

Furthermore from (8.15), we have

Lad n1

h(x) = 1,

Lad nk

h(x) = 0,

f (x) r(x)
f (x) r(x)

for k = 2, . . . , n.

Therefore, we have

h(z)
= [1, 0, . . . , 0].
z
This concludes the proof for sufciency.

x=(z)

Nonlinear observer design

117

Necessity. The discussion prior to this theorem shows that the rst condition is
necessary. Assume that there exists a state transformation z = (x) to put the system
in the output injection form, and once again, we denote
f (z) =

(x)
f (x)
x

x=(z)

h(z)
= h((z)),
where  = 1 . We need to show that when the functions f and h are in the format
of the output injection form, the second condition must hold. Let

g(x) =

(z)
zn

.
z=(x)

From the state transformation, we have



(z)
f (x) =
f (z)
z

.
z=(x)

Therefore, we can obtain

[f (x), g(x)] =
=
=
=
=


 



(z)
(z)
,
f (z)

z
zn z=(x)
z=(x)




(z)
f (z), en z=(x)
z z=(x)


f (z)
(z)
z z=(x) zn


(z)
en1
z z=(x)


(z)
.
zn1 z=(x)

Similarly, we can show that


nk
g =
adf

(z)
znk

for k = n 2, . . . , 1.

z=(x)

Hence, we have established that


(z) n1
= adf g, . . . , adf g, g x=(z) .
z

(8.17)

118

Nonlinear and adaptive control systems

The remaining part of the proof is to show that g(x) coincides with r(x) in (8.14).
From (8.17), we have

h(x) (z)
h(z)
=
z
x
z

h(x) n1
=
adf g, . . . , adf g, g x=(z)
x
= [Lad n1 g h(x), . . . , Ladf g h(x), Lg h(x)]x=(z) .
f

is in the output feedback form, the above expression implies that

Since h(z)
Lad n1 g h(x) = 1,
f

Lad nk g h(x) = 0,
f

for k = 2, . . . , n,

which further imply that

Lg Ln1 h(x) = 1,
Lg Lnk h(x) = 0,
i.e.,

dh
dLf h
..
.

dLfn1 h

for k = 2, . . . , n,

0
..

= . .
0
1

This completes the proof of necessity.

Remark 8.4. It would be interesting to revisit the transformation for linear singleoutput systems to the observer canonical form, to reveal the similarities between the
linear case and the conditions stated in Theorem 8.4. For a single output system
x = Ax
y = Cx,

(8.18)

where x Rn is the state; y R is the system output; and A Rnn and C R1n
are constant matrices. When the system is observable, we have Po full rank.
Solving r from Po r = e1 , i.e.,

0
C
..
CA

.. r = . ,
0
.
n1
1
CA
and the state transformation matrix T is then given by
T 1 = [An1 r, . . . , Ar, r].

(8.19)

Nonlinear observer design

119

We can show that the transformation z = Tx which transforms the system to the
observer canonical form

z = TAT 1 z := Az
1

y = CT z := Cz
where

a1
a2
..
.

A =

an1
an

1
0
..
.

0
1
..
.

0
0

0
0

...
...
..
.
...
...

0
0

.. ,
.

1
0

C = [1 0 . . . 0 0],
with constants ai , i = 1, . . . , n, being the coefcients of the characteristic polynomial
|sI A| = sn + a1 sn1 + + an1 s + an .
Indeed, from (8.19), we have
AT 1 = [An r, . . . , A2 r, Ar]

(8.20)

Then from CayleyHamilton theorem, we have

An = a1 An1 an1 A an I .
Substituting this into the previous equation, we obtain that

a1 1 0 . . . 0
a2 0 1 . . . 0

.. .. . .
.. = T 1 A,
1
n1

AT = [A r, . . . , Ar, r] ...
. .
. .

an1 0 0 . . . 1
an 0 0 . . . 0
which gives

TAT 1 = A.
Again from (8.19), we have

CT 1 = [CAn1 r, . . . , CAr, Cr] = [1, 0, . . . , 0] = C.

If we identify f (x) = Ax and h(x) = Cx, the rst condition of Theorem 8.4 is that Po
has full rank, and the condition in (8.13) is identical as (8.19).

When the conditions for Theorem 8.4 are satised, the transformation z = (x)
exists. In such a case, an observer can be designed for the nonlinear system (8.7) as
z = Az + L(y C z ) + (y)
x = (z ),

(8.21)

120

Nonlinear and adaptive control systems

where x Rn is the estimate of the state x of (8.7); z Rn is an estimate of z in (8.10);

and L Rn is the observer gain such that (A LC) is Hurwitz.
Corollary 8.5. For nonlinear system (8.7), if the conditions for Theorem 8.4 are
satised, the observer in (8.21) provides an asymptotic state estimate of the system.
Proof. From Theorem 8.4, we conclude that (A, C) is observable, and there exists an
L such that (A LC) is Hurwitz. From (8.21) and (8.10), we can easily obtain that
z = (A LC)z
where z = z z . We conclude that x asymptotically converges to x as limt
z (t) = 0.
2

8.3 Linear observer error dynamics via direct state

transformation
Observers presented in the last section achieve the linear observer error dynamics by
transforming the nonlinear system to the output injection form, for which an observer
with linear observer error dynamics can be designed to estimate the transformed state,
and the inverse transformation is used to transform the estimate for the original state. In
this section, we take a different approach by introducing a state transformation which
directly leads to a nonlinear observer design with linear observer error dynamics, and
show that this observer can be directly implemented in the original state space.
We consider the same system (8.7) as in the previous section, and describe it here
under a different equation number for the convenience of presentation
x = f (x)
y = h(x),

(8.22)

where x Rn is the state vector, y R is the output, f : Rn Rn and h : Rn R

are continuous nonlinear functions with f (0) = 0, and h(0) = 0. We now consider if
there exists a state transformation
z = (x),

(8.23)

where  : Rn Rn , such that the transformed system is in format

z = Fz + Gy,

(8.24)

for a chosen controllable pair (F, G) with F Rnn Hurwitz, and G Rn . Comparing
(8.24) with (8.9), the matrices F and G in (8.24) are chosen for the observer design,
while in (8.9), A and C are any observable pair which depends on the original system.
Therefore, the transformation in (8.24) is more specic. There is an extra benet
gained from this restriction in the observer design as shown later.

Nonlinear observer design

121

From (8.22) and (8.24), the nonlinear transformation must satisfy the following
partial differential equation
(x)
f (x) = F(x) + Gh(x).
x

(8.25)

Our discussion will be based on a neighbourhood around the origin.

Denition 8.1. Let i (A) for i = 1, . . . , n are the eigenvalues of a matrix A Rn . For
eigenvalue of F is resonant with the eigenvalues of A if
another matrix F Rn , an
there exists an integer q = ni=1 qi > 0 with qi being non-negative integers such that
for some j with 1 j n
j (F) =

qi i (A).

i=1

The following theorem states a result concerning with the existence of a nonlinear
state transformation around the origin for (8.22).
Theorem 8.6. For the nonlinear system (8.22), there exists a state transformation,
i.e., a locally invertible solution to the partial differential equation (8.25) if

the linearised model of (8.22) around the origin is observable

the eigenvalues of F are not resonant with the eigenvalues of f
(0)
x

 


f
f
the convex hall of 1 x (0) , . . . , n x (0) does not contain the origin

Proof. From the non-resonant condition and the exclusion of the origin of the convex
hall of the eigenvalues of f
(0), we can establish the existence of a solution to the
x
partial differential equation (8.25) by invoking Lyapunov Auxiliary Theorem. That
the function  is invertible around the origin is guaranteed by the observability of the
linearised model

 and the the controllability of (F, G). Indeed, with the observability of
f
h
(0), x (0) and the controllability of (F, G), we can apply Lemma 8.2 to establish
x
that


(0)
x

is invertible.

With the existence of the nonlinear transformation to put the system in the form
of (8.24), an observer can be designed as
z = F z + Gy
x =  1 (z ),

(8.26)

where  is the inverse transformation of . It is easy to see that the observer error
dynamics are linear as
z = F z .

122

Nonlinear and adaptive control systems

Note that once the transformation is obtained, the observer is directly given without
designing observer gain, unlike the observer design based on the output injection
form.
The observer can also be implemented directly in the original state as
x = f (x) +

1

(x)
G(y h(x)),
x

(8.27)

which is in the same structure as the standard Luenberger observer for linear systems

1
by viewing 
(x) G as the observer gain. In the following theorem, we show that
x
this observer also provides an asymptotic estimate of the system state.
Theorem 8.7. For the nonlinear system (8.22), if the state transformation in (8.25)
exists, the observer (8.27) provides an asymptotic estimate. Furthermore, the
dynamics of the transformed observer error ((x) (x)) are linear.
Proof. Let e = (x) (x). Direct evaluation gives



1
(x)
(x)

e =
f (x)
f (x) +
G(y h(x))
(x)
x
x
x
(x)
(x)
f (x)
f (x) G(y h(x))
x
x
= F(x) + Gy (F(x) + Gh(x)) G(y h(x))
=

= F((x) (x))
= Fe.
Therefore, the dynamics of the transformed observer error are linear, and the transformed observer error converges to zero exponentially, which implies the asymptotic
convergence of x to x.
2
Remark 8.5. For linear systems, we have briey introduced two ways to design
observers. For the observer shown in (8.2), we have introduced the observer shown
in (8.6) to deal with nonlinear systems. The nonlinear observer in (8.27) can be
viewed as a nonlinear version of (8.3). For both cases, the observer error dynamics
are linear.


8.4 Observer design for Lipschitz nonlinear systems

In this section, we will deal with observer design for nonlinear systems with Lipschitz
nonlinearity. We introduced this denition for a time-varying function in Chapter 2 for
the existence of a unique solution for a nonlinear system. For a time-invariant function,
the denition is similar, and we list below for the convenience of presentation.

Nonlinear observer design

123

Denition 8.2. A function : Rn Rs Rn is Lipschitz with a Lipschitz constant

if for any vectors x, x Rn and u Rs
(x, u) (x, u) x x ,

(8.28)

with > 0.
Once again we consider linear systems perturbed by nonlinear terms as
x = Ax + (x, u)
y = Cx,

(8.29)

where x Rn is the state; y Rm is the system output with m < n; u Rs is the control
input, or other known variables, A Rnn and C Rmn are constant matrices with
(A, C) observable; and : Rn Rs Rn is a continuous function with Lipschitz
constant with respect to the state variable x. Comparing with the system (8.5), the
only difference is the nonlinear term (x, u). Here, it is a function of state variable,
not only the output, as in (8.5), and therefore the observer design by output injection
does not work.
We can still design an observer based on the linear part of the system, and replace
the unknown state in the nonlinear function by its estimate, that is,
x = Ax + L(y C x ) + (x, u),

(8.30)

where x Rn is the estimate of the state x and L Rnm is the observer gain. However,
the condition that (A LC) is Hurwitz is not enough to guarantee the convergence
of the observer error to zero. Indeed, a stronger condition is needed, as shown in the
following theorem.
Theorem 8.8. The observer (8.30) provides an exponentially convergent state estimate if for the observer gain L, there exists a positive denite matrix P Rnn such
that
(A LC)T P + P(A LC) + 2 PP + I + I = 0,
where is the Lipschitz constant of and is any positive real constant.
Proof. Let x = x x . From (8.29) and (8.31), we have
x = (A LC)x + (x, u) (x, u).
Let
V = x T P x .
Its derivative along the observer error dynamics is obtained as
V = x T ((A LC)T P + P(A LC))x + 2xT P((x, u) (x, u)).

(8.31)

124

Nonlinear and adaptive control systems

For the term involved with the nonlinear function , we have

2xT P((x, u) (x, u))
1
2 x T PP x + 2 ((x, u) (x, u))T ((x, u) (x, u))

1
= 2 x T PP x + 2 (x, u) (x, u)2

2 x T PP x + x2
= x T ( 2 PP + I )x.
Applying the condition shown in (8.31), we have the derivative of V satisfying
V x T ((A LC)T P + P(A LC) + 2 PP + I )x
= x T x .
This implies that the observer error converges to zero exponentially.

In the condition (8.31), is an arbitrary positive real number. In this case, we

can use inequality to replace the equality, that is
(A LC)T P + P(A LC) + 2 PP + I < 0.

(8.32)

Using the same Lyapunov function candidate as in the proof of Theorem 8.8, we can
show that
V < 0
which implies that the observer error converges to zero asymptotically. We summarise
this result below.
Corollary 8.9. The observer (8.30) provides an asymptotically convergent state estimate if for the observer gain L, there exists a positive denite matrix P Rnn that
satises the inequality (8.32).
Remark 8.6. By using the inequality (8.32) instead of the equality (8.31), we only
establish the asymptotic convergence to zero of the observer error, not the exponential
convergence that is established inTheorem 8.8 using (8.31). Furthermore, establishing
the asymptotic convergence of the observer error from V < 0 requires the stability
theorems based on invariant sets, which are not covered in this book.

The condition shown in (8.32) can be relaxed if one-side Lipschitz constant is
used instead of the Lipschitz constant.
Denition 8.3. A function : Rn Rs Rn is one-sided Lipschitz with respect to
P and one-sided Lipschitz constant if for any vectors x, x Rn and u Rs
(x x )T P((x, u) (x, u)) x x 2
where P R

nn

is a positive real matrix, and is a real number.

(8.33)

Nonlinear observer design

125

Note that the one-sided Lipschitz constant can be negative. It is easy to see from
the denition of the one-sided Lipschitz condition that the term (x x )T P((x, u)
(x, u)) is exactly the cross-term in the proof Theorem 8.8 which causes the term
2 PP + I in (8.32). Hence, with the Lipschitz constant with respect to P, the
condition shown in (8.32) can be replaced by
(A LC)T P + P(A LC) + 2I < 0.

(8.34)

This condition can be further manipulated to obtain the result shown in the following
theorem.
Theorem 8.10. The observer (8.30) provides an asymptotically convergent state
estimate if the following conditions hold:
L = P 1 C T ,

(8.35)

A P + PA + 2I 2 C C < 0,
T

(8.36)

where P Rnn is a positive real matrix, is a positive real constant and is the
one-sided Lipschitz constant of with respect to x and P.
Proof. From (8.35), we have
T 



LT
LT
1
1
CP
CP
= 0,

which gives
P 1 C T LT + LCP 1 = P 1 C T CP 1 +

LLT
.

Using (8.35) and multiplying the above equation by P on both sides, we obtain the
identity
C T LT P + PLC = 2 C T C.
From this identity and (8.36), we can easily obtain the inequality (8.34). Similar to
the proof of Theorem 8.8, we let
V = x T P x ,
and obtain, using the one-sided Lipschitz condition of ,
V = x T ((A LC)T P + P(A LC))x + 2xT P((x, u) (x, u))
x T ((A LC)T P + P(A LC))x. + 2 x T x .
Applying the inequality (8.34) to the above expression, we have
V < 0,
which implies that the observer error asymptotically converges to zero.

126

Nonlinear and adaptive control systems

To end this section, we consider a class of systems with nonlinear Lipschitz output
function
x = Ax
(8.37)
y = h(x),
where x Rn is the state vector, y Rm is the output, A Rnn is a constant matrix
and h : Rn Rm is a continuous function. We can write the nonlinear function h as
h = Hx + h1 (x) with Hx denoting a linear part of the output, and the nonlinear part
h1 with Lipschitz constant .
An observer can be designed as
x = Ax + L(y h(x)),
(8.38)
where the observer gain L Rnm is a constant matrix.
Theorem 8.11. The observer (8.38) provides an exponentially convergent state
estimate of (8.37) if the observer gain L can be chosen to satisfy the following
conditions:
1
L = 2 P 1 H T ,
(8.39)

PA + AT P

HTH
+ (1 + )I = 0,
2

(8.40)

where P Rnn is a positive denite matrix and is a positive real constant.

Proof. Let x = x x . From (8.37) and (8.38), we have
x = (A LH )x + L(h1 (x) h1 (x)).
Let
V = x T P x ,
where x = x x . It can be obtained that
V = x T ((A LH )T P + P(A LH ))x + 2xT PL(h1 (x) h1 (x))
x T ((A LH )T P + P(A LH ))x + x T (I + 2 PLLT P)x



T 

H
H
HTH
T
T
T
= x T AT P + PA
+
I
x

+
x

P
P
x

L
2

= x T x .
Therefore, we can conclude that x converges to zero exponentially.

Remark 8.7. The nonlinearity in the output function with linear dynamics may
occur in some special cases such as modelling a periodic signal as the output of a
second-order linear system. This kind of formulation is useful for internal model
design to asymptotically reject some general periodic disturbances, as shown in
Chapter 10.


Nonlinear observer design

127

8.5 Reduced-order observer design

The observers introduced in the previous sections all have the same order as the
original systems. One might have noticed that an observer provides estimate of the
entire state variables, which include the system output. Only the variables which
are not contained in the output are needed for state estimation from an observer.
This was noticed at the very early stage of the development of observer design of
linear systems, and leads to the design of observers with less order than the original
systems. The observers with less order than the original system are referred to as
reduced-order observers. This section devotes to reduced-order observer design for
nonlinear systems.
Consider a nonlinear system
x = f (x, u)

(8.41)

y = h(x),

where x Rn is the state vector, u Rs is the known input, y Rm is the output

and f : Rn Rs Rn is a nonlinear smooth vector eld. To design a reduced-order
observer, we need to study the dynamics of other state variables other than the output.
We can dene a partial-state transformation.
Denition 8.4. A function g : Rn Rnm is an output-complement transformation
if the function T given by

T (x) :=

h(x)
g(x)

is a diffeomorphism, where h(x) is the output function. The transformed states are
referred to as output-complement states.
Clearly an output-complement transformation denes a state transformation
together with the output function. If the output-complement states are known, then
the state variables can be fully determined. In fact, if we can design an observer for
the output-complement state, this observer is a reduced-order observer.
Denition 8.5. The dynamic model
z = p(z, y) + q(y, u)

(8.42)

is referred to as reduced-order observer form for the system (8.41) if the z = g(x) is
output-complement transformation, and the dynamic system
z = p(z, y)
is differentially stable.

128

Nonlinear and adaptive control systems

For a system which can be transformed to a reduced-order observer form, we can

propose a reduced-order observer design as
z = p(z , y) + q(y, u)
 
y
x = T 1
.
z

(8.43)

Theorem 8.12. If the system (8.41) can be transformed to the reduced-order observer
form (8.42), the state estimate x provided by the reduced-order observer in (8.43)
asymptotically converges to the state variable of (8.41).
Proof. First, let us establish the boundedness of z. Since z = p(z, y) is differentially
stable, from the denition of the differential stability in Chapter 5, there exists a
Lyapunov function, V (z), such that the conditions (5.41) are satised. Take the same
function V (z) here as the Lyapunov function candidate with z = 0. From (5.41) and
(8.42), we have
V
V 3 (z) +
q(y, u)
z
V
3 (z) + 
q(y, u).
(8.44)
z
Let us recallYoungs inequality in a simplied form that for any a R and b R,
and a pair of constants p > 1 and q > 1 with 1p + q1 = 1, we have
|ab|

p p
1
|a| + q |b|q
p
q

for any positive real constant .

Applying Youngs inequality to the second term on the right-hand side of (8.44)
gives


c2
c2 
 V (z)  




 q(y, u) c4  V (z)  + 1 c q(y, u)c3 ,
(8.45)
 z 

c2
z 
c3 c43
2
where c3 = c2c1
, and c4 is an arbitrary positive real constant. We set c4 = ( c12c2 )1/c2 ,
which results in



c2
 V (z) 



 q(y, u) c1  V (z)  + 1 c5 q(y, u)c3 ,
(8.46)
 x 

2
z 
2

where c5 =

3
1 1
( c c ) c2
c3 2 1 2

. Substituting (8.46) into (8.44), we have

1
1
V 3 (z) + c5 q(y, u)c3 .
(8.47)
2
2
Since q(y, u) is continuous, there exists a class K function g such that, for all y Rm
and u Rs , we have
q(y, u) g(y

+ u).

Nonlinear observer design

129

c3

). For z (y + u), we have

We then choose () = 31 (2c5 (g())

+ u))c3
3 (z) 2c5 (g(y
2c5 q(y, u)c3 ,
which further implies that
1
V 3 (z).
4

(8.48)

Hence, V (z) is an ISS-Lyapunov function, and therefore z is bounded when y is

bounded.
The dynamics of e are given by
e = p(z(t), y) p((z(t) e), y).
Taking V (e) as the Lyapunov function candidate, we have
V
(p(z(t), y) p((z(t) e), y))
V =
e
3 (e).
Therefore, we can conclude that the estimation error asymptotically converges to zero.
With z as a convergent estimate of z, we can further conclude that x is an asymptotic
estimate of x.
2
To demonstrate the proposed reduced-order observer design, let us consider an
example.
Example 8.1. Consider a second-order system
x 1 = x12 3x12 x2 x13
x 2 = x2 + x12 6x2 x12 + 3x22 x1 x23
y = x1 .
Let us check if the system can be transformed to the reduced-order observer
form. For this, we need to nd g(x). Take
z = g(x) = x2 x1 .
We have
z = x2 (x2 x1 )3
= (1 + z 2 )z + y.
Comparing with the format shown in (8.42), we have p(z, y) = (1 + z 2 )z + y. Note
that for the system without input, we always have z = p(z, y). Let V = 12 z 2 . It is easy
to see the rst and the third conditions in (5.41) are satised. For the second condition,
we have

130

Nonlinear and adaptive control systems

V (z z )
(p(z) p(z ))
z
= (z z )(z z + z 3 z 3 )
= (z z )2 (1 + z 2 zz + z 2 )
1
= (z z )2 (1 + (z 2 + z 2 + (z z )2 ))
2
(z z )2 .

Therefore, the system satises the conditions specied in (5.41). We design the
reduced-order observer as
z = (1 + z 2 )z + y
x 2 = z + y.
Simulation study has been carried out, and the simulation results are shown in
Figures 8.1 and 8.2.


0.7
x1
x2

0.6

x1 and x2

0.5

0.4

0.3

0.2

0.1

Time (s)

Figure 8.1 State variables

10

Nonlinear observer design

131

0.4
x2
Estimate of x2

0.35

x2 and estimate

0.3
0.25
0.2
0.15
0.1
0.05
0

10

Time (s)

Figure 8.2 Unmeasured state and its estimate

There are lack of systematic design methods for nonlinear observers with global
convergence when there are general nonlinear terms of unmeasured state variables
in the systems. With the introduction of the reduced-order observer form, we like to
further explore the class of nonlinear systems which can be transformed to the format
shown in (8.42), and therefore a nonlinear observer can then be designed accordingly.
We consider a multi-output (MO) nonlinear system
x = Ax + (y, u) + E(x, u)
y = Cx,

(8.49)

where x Rn is the state; y Rm is the output; u Rs is the control; is a known

nonlinear smooth vector eld; : Rn Rs Rm is a smooth nonlinear function;
C Rmn , E Rnm and A Rnn are constant matrices, with (A, C) observable.
When = 0, the system (8.49) degenerates to the well-known form of the nonlinear systems with the linear observer error dynamics, and nonlinear observer can be
easily designed by using nonlinear output injection. With this additional term , the
nonlinear output injection term can no longer be used to generate a linear observer
error dynamics. We will convert the system to the reduced-order observer form
considered, and then apply the reduced-order observer design for state observation.

132

Nonlinear and adaptive control systems

Without loss of generality, we can assume that C has full row rank. There exists
a nonsingular state transformation M such that
CM 1 = [Im , 0m(nm) ].
If span{E} is a complement subspace of ker{C} in Rn , we have (CM 1 )(ME) = CE
invertible. If we partition the matrix ME as

E1
ME :=
E2
with E1 Rmm , then we have CE = E1 . Furthermore, if we partition Mx as

1
Mx :=
2
with 1 Rm , we have
z = g(x) = 2 E2 E11 1 .

(8.50)

Note that we have 1 = y. With the partition of


A1,1 A1,2
MAM 1 :=
A2,1 A2,2
and


M :=

1
,
2

we can write the dynamics 1 and 2 as

1 = A1,1 1 + A1,2 2 + 1 + E1
2 = A2,1 1 + A2,2 2 + 2 + E2 .
Then we can obtain the dynamics of z as
z = A2,1 1 + A2,2 2 + 2 E2 E11 (A1,1 1 + A1,2 2 + 1 )
= (A2,2 E2 E11 A1,2 )2 + (A2,1 E2 E11 A1,1 )1 + 2 E2 E11 1
= (A2,2 E2 E11 A1,2 )z + q(y, u),

(8.51)

where
q(y, u) = (A2,2 E2 E11 A1,2 )E2 E11 y + (A2,1 E2 E11 A1,1 )y
+ 2 (y, u) E2 E11 1 (y, u).
Note that the nonlinear function (x, u) does not appear in the dynamics of z
in (8.51) due to the particular choice of z in (8.50).
Remark 8.8. After the state transformation, (8.51) is in the same format as (8.42).
Therefore, we can design a reduced-order observer if z = (A2,2 E2 E11 A1,2 )z is differentially stable. Notice that it is a linear system. Hence, it is differentially stable if it

Nonlinear observer design

133

is asymptotically stable, which means the eigenvalues of (A2,2 E2 E11 A1,2 ) are with
negative real parts.

Following the format shown in (8.43), a reduced-order observer can then be
designed as
z = (A2,2 E2 E11 A1,2 )z + q(y, u)
and the estimate of x is given by


y
.
x = M 1
z + E2 E11 y

(8.52)

(8.53)

Theorem 8.13. For a system (8.49), if

C has full row rank, and span{E} is a complement subspace of ker{C} in Rn

all the invariant zeros of (A, E, C) are with negative real parts

it can be transformed to the reduced-order observer form. The estimates z given in

(8.52) and x (8.53) converge to the respective state variables z and x exponentially.
Proof. We only need to show that z = p(z, y) is differentially stable, and then we can
apply Theorem 8.12 to conclude the asymptotic convergence of the reduced-order
observer error. The exponential convergence comes as a consequence of the linearity
in the reduced-order observer error dynamics. For (8.52), we have
p(z, y) = (A2,2 E2 E11 A1,2 )z,
which is linear in z. Therefore, the proof can be completed by proving that the
matrix (A2,2 E2 E11 A1,2 ) is Hurwitz. It can be shown that the eigenvalues of
(A2,2 E2 E11 A1,2 ) are the invariant zeros of (A, E, C). Indeed, we have


 1

M 0
0
sI A E
M
0 Im
0
Im
C
0


sI MAM 1 ME
=
CM 1
0

sIm A1,1
A1,2
E1
sInm A2,2 E2 .
= A2,1
(8.54)
Im
0
0
Let us multiply the above matrix in the left by the following matrix to perform a row
operation:

0
0
Im
E2 E11 Inm 0 ,
0
0
Im

134

Nonlinear and adaptive control systems

we result in the following matrix:

A1,2
E1
sIm A1,1


sInm (A2,2 E2 E11 A1,2 ) 0
Im
0
0
with
 = E2 E11 (sIm A1,1 ) A2,1 .
Since E1 is invertible, any values of s which make the matrix


sI A E
C
0
rank decient must be the eigenvalues of (A2,2 E2 E11 A1,2 ). From the second condition that all the invariant zeros of (A, E, C) are with negative real part, we can conclude
that (A2,2 E2 E11 A1,2 ) is Hurwitz.
2
Example 8.2. Consider a third-order system
x 1 = x1 + x2 y1 + u + x2 x3 x1 x3
x 2 = x1 + x2 + x3 2y1 + u + y1 y2 + x2 x3
x 3 = y12 + x1 x3 + x2 x3
y1 = x 1
y2 = x1 + x2 .
We can identify

y1 + u
= 2y1 + y1 y2 + u ,
y12

and we have

A = 1
0

1
1
0

0
1,
0


=

E= 1
1

x2 x3 x1 x3
,
x1 x3

0
1,
2


C=

(8.55)

1
1

0
1

0
.
0

(8.56)

It can be easily checked that (A, C) is observable and the invariant zero of
(A, E, C) is at 2. Therefore, the conditions in Theorem 8.13 are satised. It is
also easy to see that x2 = y2 + y1 . Therefore, the only unmeasured state variable is
x3 . Indeed, following the procedures introduced earlier, we have



1
0 0


1 0
, E2 1 2 ,
M = 1 1 0 , E1 =
0 1
0
0 1


x1
, 2 = x3 ,
1 =
x2 x 1

Nonlinear observer design

and
z = x3 y1 2y2 = x3 + x1 2x2 .
With

A1,1 =

1 =

0
0

1
,
0

A1,2

y1 + u
,
y1 + y1 y2


0
=
,
1

A2,1 0


0 ,

A2,2 = 0,

2 = y12 ,

we have the dynamics of z as

z = 2z + q(y, u),
where
q(y, u) = y1 5y2 y12 2y1 y2 u.
A reduced-order observer can then be designed as
z = 2z + q(y, u)
x 3 = z + y1 + 2y2 .

1
x1
x2
x3

6
0

10

12

Time (s)

Figure 8.3 State variables

14

16

18

20

135

136

Nonlinear and adaptive control systems

0.6
x3
0.4

Estimate of x3

0.2

x3 and estimate

0
0.2
0.4
0.6
0.8
1
1.2
0

10

12

14

16

18

20

Time (s)

Figure 8.4 Unmeasured state and its estimate

Simulation study has been carried with x(0) = [1, 0, 0.5]T and u = sin t. The
plots of the state variables are shown in Figure 8.3 and the estimated state is shown
in Figure 8.4 .


8.6 Adaptive observer design

When there are unknown parameters in dynamic systems, observers can still be
designed under certain conditions to provide estimates of unknown states by using
adaptive parameters. These observers are referred to as adaptive observers. One would
expect more stringent conditions imposed on nonlinear systems for which adaptive
observers can be designed. In this section, we will consider two classes of nonlinear
systems. The rst one is based on the nonlinear systems in the output injection form
or output feedback form with unknown parameters and the other class of the systems
is nonlinear Lipschitz systems with unknown parameters.
Consider nonlinear systems which can be transformed to a single-output
system
x = Ax + 0 (y, u) + b T (y, u)
y = Cx,

(8.57)

Nonlinear observer design

137

where x Rn is the state, y R is the system output, u Rs is the control input, or

other known variables, A Rnn , b Rn1 and C R1n are constant matrices, with
(A, C) observable, Rr is an unknown constant vector, 0 : Rm Rs Rn and
: Rm Rs Rr are continuous functions. If the parameter vector is known, the
system is in the output-injection form.
One would expect that additional conditions are required for the design of
an adaptive observer. If the linear system characterised by (A, b, C) is of relative
degree 1 and minimum phase, we can design an observer gain L Rn such that the
linear system characterised by (A LC, b, C) is a positive real system. In such a case,
there exist positive real matrices P and Q such that
(A LC)T P + P(A LC) = Q
Pb = C T .

(8.58)

Let us consider the observer designed as

x = Ax + 0 (y, u) + b T (y, u) + L(y C x ),

(8.59)

where x is the estimate of x and is the estimate of . We need to design an adaptive

law for . Let x = x x . The observer error dynamics are obtained as
x = (A LC)x + b T (y, u) ,

(8.60)

where = .
Consider a Lyapunov function candidate

V = x T P x + T  1 ,
where  Rnn is a positive denite matrix. From (8.58) and (8.60), we have
V = x T ((A LC)T P + P(A LC))x + 2xT Pb(y, u)T + 2 T  1
= xT Qx + 2xT C T (y, u)T 2 T  1
= xT Qx 2( (y C x )(y, u))T  1
If we set the adaptive law as
= (y C x )(y, u),
we obtain
V = xT Qx.
Then similar to stability analysis of adaptive control systems, we can conclude that
limt x (t) = 0, and is bounded. The above analysis leads to the following theorem.
Theorem 8.14. For the observable single-output system (8.57), if the linear system
characterised by (A, b, C) is minimum phase and has relative degree 1, there exists an

138

Nonlinear and adaptive control systems

observer gain L Rn that satises the conditions in (8.58) and an adaptive observer
designed as
x = Ax + 0 (y, u) + b T (y, u) + L(y C x )
= (y C x )(y, u),

(8.61)

where  Rnn is a positive denite matrix and provides an asymptotically

convergent state estimate with adaptive parameter vector remaining bounded.
Remark 8.9. In this remark, we will show a particular choice of the observer
gain for the adaptive control observer for a system that satises the conditions in
Theorem 8.14. Without loss of generality, we assume

0 1 0 ... 0
b1
0 0 1 ... 0
b2

..

..
..
.. , b = .. ,
..
A=.

.
.
.
.

.
0 0 0 ... 1
bn1
bn
0 0 0 ... 0
C = [1

...

0],

where b1  = 0, since the relative degree is 1. Indeed, for an observer system {A, b, C}
with relative degree 1, there exists a state transformation, as shown in Remark 8.4, to
transform the system to the observable canonical form. Furthermore, if we move the
rst column of A in the canonical form, and combine it with 0 (y, u), we have A as in
the format shown above. Since the system is minimum phase, b is Hurwitz, i.e.,
B(s) := b1 sn1 + b2 sn2 + + bn1 s + bn = 0
has all the solutions in the left half of the complex plane. In this case, we can design
the observer gain to cancel all the zeros of the system. If we denote L = [l1 , l2 , . . . , ln ],
we can choose L to satisfy
sn + l1 sn1 + l2 sn2 + + ln1 s + ln = B(s)(s + ),
where is a positive real constant. The above polynomial equation implies
L = (I + A)b.
This observer gain ensures
1
,
s+
which is a strict positive real transfer function.
C (sI (A LC))1 b =

Now we will consider adaptive observer design for a class of Lipschitz nonlinear
systems. Consider
x = Ax + 0 (x, u) + b T (x, u)
y = Cx,

(8.62)

Nonlinear observer design

139

where x Rn is the state, y R is the system output, u Rs is the control input,

or other known variables, A Rnn , b Rn1 and C R1n are constant matrices,
with (A, C) observable, Rr is an unknown constant vector, 0 : Rn Rs Rn
and : Rm Rs Rr are Lipschitz nonlinear functions with Lipschitz constants 1
and 2 with respect to x.
This system is only different in nonlinear functions from the system (8.57) considered earlier for adaptive observer design, where the nonlinear functions are restricted
to the functions of the system output only, but the functions do not have to be globally
Lipschitz for a globally convergent observer.
We propose an adaptive observer for (8.62) as
x = Ax + 0 (x, u) + b T (x, u) + L(y C x )
= (y C x )(x, u),

(8.63)

where L Rn is an observer gain and  Rnn is a positive denite matrix.

When state variables other than the system output are involved in the nonlinear
functions, we often impose the conditions that the nonlinear function are Lipschitz,
and the conditions are more stringent for designing an observer than the nonlinear
systems with only nonlinear functions of the system output. We have considered
adaptive observer design for a class of nonlinear systems (8.57) with only the nonlinear
functions of the system output. For the adaptive observer (8.63) to work for the system
(8.62), we will expect stronger conditions for adaptive observer design. For this, we
state the following result.
Theorem 8.15. The adaptive observer proposed in (8.63) provides an asymptotically
convergent state estimate for the system (8.62) if there exists a positive denite matrix
P such that
(A LC)T P + P(A LC) + (1 + 2 3 b)(PP + I ) + I 0
Pb = C T ,

(8.64)

where 3 , and is a positive real constant.

Proof. Let x = x x . The observer error dynamics are obtained as
x = (A LC)x + 0 (x, u) 0 (x, u) + b( T (y, u) T (x, u)T )
= (A LC)x + 0 (x, u) 0 (x, u) + b((x, u) (x, u))T

+ b T (x, u),
where = . Consider a Lyapunov function candidate
V = x T P x + T  1 .
Its derivative along the dynamics in (8.65) is obtained as
V = x T ((A LC)T P + P(A LC))x + 2xT P(0 (x, u) 0 (x, u))

+ 2xT Pb((x, u) (x, u))T + 2xT Pb T (x, u) + 2 T  1 .

(8.65)

140

Nonlinear and adaptive control systems

From the Lipschitz constants of 0 and , we have

2xT P(0 (x, u) 0 (x, u)) 1 x T (PP + I )x,
2xT Pb((x, u) (x, u))T 2 3 bxT (PP + I )x.
Then from the adaptive law and (8.64), we have
V x T ((A LC)T P + P(A LC) + (1 + 2 3 b)(PP + I ))xT
+ 2xT C T T (x, u) 2 T  1
x T x .
This implies that the variables x and are bounded, and furthermore x L2 L .
Since all the variables are bounded, x is bounded, from x L2 L and the
boundedness of x , we conclude limt x (t) = 0 by invoking Babalats Lemma. 2

Chapter 9

Backstepping design

For a nonlinear system, the stability around an equilibrium point can be established
if one can nd a Lyapunov function. Nonlinear control design can be carried out by
exploring the possibility of making a Lyapunov function candidate as a Lyapunov
function through control design. In Chapter 7, parameter adaptive laws are designed
in this way by setting the parameter adaptive laws to make the derivative of a Lyapunov
function candidate negative semi-denite. Backstepping is a nonlinear control design
method based on Lyapunov functions. It enables a designed control to be extended to
an augmented system, provided that the system is augmented in some specic way.
One scheme is so-called adding an integrator in the sense that if a control input is
designed for a nonlinear system, then one can design a control input for the augmented
system of which an integrator is added between the original system input and the
input to be designed. This design strategy can be applied iteratively. There are a few
systematic control design methods for nonlinear systems, and backstepping is one
of them. In this chapter, we start with the fundamental form of adding an integrator,
and then introduce the method for iterative backstepping with state feedback. We also
introduce backstepping using output feedback, and adaptive backstepping for certain
nonlinear systems with unknown parameters.

9.1 Integrator backstepping

Consider
x = f (x) + g(x)
= u,

(9.1)

where x Rn and R are the state variables; u R is the control input; and f :
Rn Rn with f (0) = 0 and g : Rn Rn are continuous functions. Viewing the
rst equation of (9.1) as the original system with x as the state and as the input,
the integration of u gives , which means that an integrator is added to the original
system.
We consider the control design problem under the assumption that a known
control input exists for the original system. Furthermore, we assume that the Lyapunov
function is also known, associated with the known control for x-subsystem. Suppose
that control input for the x-subsystem is (x) with differentiable and (0) = 0, and
the associated Lyapunov function is V (x). We assume that

142

Nonlinear and adaptive control systems

V
(9.2)
(f (x) + g(x)(x)) W (x),
x
where W (x) is positive denite.
Condition (9.2) implies that the system x = f (x) + g(x)(x) is asymptotically
stable. Consider
x = f (x) + g(x)(x) + g(x)( (x)).
Intuitively, if we can design a control input u = u(x, ) to force to converge to (x),
we have a good chance to ensure the stability of the entire system. Let us dene
z = (x).
It is easy to obtain the dynamics under the coordinates (x, z) as
x = f (x) + g(x)(x) + g(x)z

z = u = u
(f (x) + g(x) ).
x
Consider a Lyapunov function candidate
1
Vc (x, z) = V (x) + z 2 .
2
Its derivative is given by

(9.3)

V
V
V c =
g(x)z
(f (x) + g(x)(x)) +
x
x



+z u
(f (x) + g(x) )
x


V

= W (x) + z u +
g(x)
(f (x) + g(x) ) .
x
x
Let
V

g(x) +
(f (x) + g(x) )
x
x
with c > 0 which results in
u = cz

V c = W (x) cz 2 .

(9.4)

(9.5)

It is clear that W (x) cz is negative denite with respect to variables (x, z). Hence,
we can conclude that Vc (x, z) is a Lyapunov function, and that (0, 0) in the coordinates
(x, z) is a globally asymptotic equilibrium. From (0) = 0, we can conclude that (0, 0)
in the coordinates (x, ) is also a globally asymptotic equilibrium, which means that
the system (9.1) is globally asymptotically stable under the control input (9.4). We
summarise the above result in the following lemma.
2

Lemma 9.1. For a system described in (9.1), if there exist differentiable function (x)
and a positive-denite function V (x) such that (9.2) holds, the control design given
in (9.4) ensures the global asymptotic stability of the closed-loop system.

Backstepping design

143

Remark 9.1. Considering the structure of (9.1), if is viewed as the control input for
the x-subsystem, = (x) is the desired control, ignoring the -system. This is why
can be referred to as a virtual control input for the x-subsystem. The control input
u for the overall system is designed with the consideration of the dynamics back to
the control design for the x-subsystem, and it may suggest the name of this particular
design method as backstepping.

Example 9.1. Consider
x 1 = x12 + x2
x 2 = u.
We design a control input using backstepping. Comparing with (9.1), we can identify
x x1 ,

x2 ,

f (x) x12 ,

g(x) 1.

First, we need to design (x1 ) to stabilise

x 1 = x12 + (x1 ).
An obvious choice is
(x1 ) = c1 x1 x12
with c1 > 0 a constant. This design leads to
x 1 = x12 + (x1 ) = c1 x1 .
Hence, we take
V (x1 ) =

1 2
x
2 1

with
V (x1 ) = c1 x12 .
Therefore, the condition specied in Lemma 9.1 is satised with (x1 ) = c1 x1
x12 , V (x1 ) = 12 x12 and W (x1 ) = c1 x12 . The control input u can then be obtained
from (9.4) by substituting proper functions in the equation. Alternatively, we can
obtain the control input by directly following the backstepping method. Indeed, let
z = x2 (x1 ). The dynamics of the system in coordinate (x1 , z) are obtained as
x 1 = c1 x1 + z
(x1 ) 2
z = u
(x1 + x2 ),
x1
where
(x1 )
= c1 2x1 .
x1
Let
1
1
Vc (x1 , z) = x12 + z 2 .
2
2

144

Nonlinear and adaptive control systems

Its derivative along the system dynamics is obtained as



(x1 ) 2
2

Vc (x1 , z) = c1 x1 + x1 z + z u
(x1 + x2 ) .
x1
Designing the control u as
u = x1 c2 z +

(x1 ) 2
(x1 + x2 )
x1

results in
V c (x1 , z) = c1 x12 c2 z 2 .
Hence, the system is asymptotically stable with (x1 , z). As (0) = 0, we conclude limt x2 (t) = limt (z(t) + (x1 (t))) = 0, which implies that the system
is asymptotically stable in the equilibrium (0, 0) in (x1 , x2 ). Note that the closed-loop
system in (x1 , z) is written as
  
 
x 1
1
x1
c1
=
.
1 c2
z
z

In Example 9.1, backstepping design has been used to design a control input for a
nonlinear system with unmatched nonlinearities. When a nonlinear function appears
in the same line as the control input, it is referred as a matched nonlinear function,
and it can be cancelled by adding the same term in u with an opposite sign. From the
system considered in Example 9.1, the nonlinear function x12 does not appear in the
same line as the control input u, and therefore it is unmatched. However, it is in the
same line as x2 , which is viewed as a virtual control. As a consequence, (x1 ), which
is often referred to as a stabilising function, can be designed to cancel the nonlinearity
which matches with the virtual control, and backstepping method enables the control
in the next line to be designed to stabilise the entire system. This process can be
repeated by identifying a virtual control, designing a stabilising function and using
backstepping to design control input for more complicated nonlinear systems.

9.2 Iterative backstepping

Consider a nonlinear system
x 1 = x2 + 1 (x1 )
x 2 = x3 + 2 (x1 , x2 )
...
x n1 = xn + n1 (x1 , x2 , . . . , xn1 )
x n = u + n (x1 , x2 , . . . , xn )

(9.6)

Backstepping design

145


where xi R for i = 1, . . . , n are state variables; i : R R R for
i = 1, . . . , n are differentiable functions up to the order n i with i (0, . . . , 0) = 0;
and u R is the control input.
When xi+1 in the ith equation of (9.6) is identied as the virtual control, the nonlinear function i is then matched with respect to the virtual control, and backstepping
method can be applied to move down the control design to (i + 1)th equation with
xi+2 as the next virtual control. This process starts from i = 1 and can be repeated
until i = n 1 when u is reached. We will present this iterative backstepping design
for the system (9.6) in n steps. In each step, we could show the Lyapunov function
and other details for which Lemma 9.1 can be applied to. Although, for the simplicity
of the control design, we leave the stability analysis to the end, stability is considered
in designing the stabilising function at each step. The control design will be shown
in n steps.
Let
z1 = x1 ,
zi = xi i1 (x1 , . . . , xi1 ),

for i = 2, . . . , n,

where i1 for i = 2, . . . , n are stabilising functions obtained in the iterative

backstepping design.
Step 1. For the design of the stabilising function, we arrange the dynamics of z1 as
z1 = (x2 1 ) + 1 + 1 (x1 )
= z2 + 1 + 1 (x1 ).
Let
1 = c1 z1 1 (x1 ).

(9.7)

The resultant dynamics of z1 are

z1 = c1 z1 + z2 .

(9.8)

Step 2. The dynamics of z2 are obtained as

z2 = x 2 1
1
(x2 + 1 (x1 ))
x1
1
(x2 + 1 (x1 )).
= z3 + 2 + 2 (x1 , x2 )
x1

= x3 + 2 (x1 , x2 )

Design 2 as
2 = z1 c2 z2 2 (x1 , x2 ) +

1
(x2 + 1 (x1 )).
x1

(9.9)

The resultant dynamics of z2 are given by

z2 = z1 c2 z2 + z3 .

(9.10)

146

Nonlinear and adaptive control systems

Note that the term z1 in 2 is used to tackle a cross-term caused by z2 in the dynamics
of z1 in the stability analysis. Other terms in 2 are taken to cancel the nonlinear terms
and stabilise the dynamics of z2 .
Step i. For 2 < i < n, the dynamics of zi are given by
zi = x i i1 (x1 , . . . , xi1 )
= xi+1 + i (x1 , . . . , xi )
i1

i1

j=1

xj

(xj+1 + j (x1 , . . . , xj ))

= zi+1 + i + i (x1 , . . . , xi )
i1

i1

j=1

xj

(xj+1 + j (x1 , . . . , xj )).

Design i as
i = zi1 ci zi i (x1 , . . . , xi )
+

i1

i1
j=1

xj

(xj+1 + j (x1 , . . . , xj )).

(9.11)

The resultant dynamics of zi are given by

zi = zi1 ci zi + zi+1 .

(9.12)

Note that similar to the design of 2 , the term zi1 is used to tackle a cross-term
caused by zi in the dynamics of zi1 in the stability analysis, and the other terms in i
are used to stabilise the dynamics of zi .
Step n. At the nal step, we have
zn = x n n1 (x1 , . . . , xn1 )
= u + n (x1 , . . . , xn )

n1

n1
j=1

xj

(xj+1 + j (x1 , . . . , xj )).

Design the control input as

u = zn1 cn zn n (x1 , . . . , xn )
+

n1

n1
j=1

xj

(xj+1 + j (x1 , . . . , xj )).

(9.13)

The resultant dynamics of zn are given by

zn = zn1 cn zn .

(9.14)

Backstepping design

147

Note that we can write u = n by setting i = n in the expression of i in (9.11).

Let us establish the stability of the closed-loop system under the proposed
control. The closed-loop dynamics of the system in coordinate z = [z1 , . . . , zn ]T can
be written as

z =

c1

c2

0
..
.

1
..
.

c3
..
.

...
0

..
.
0

..
.
0
z := Az z.

..
.
1

..
. cn

Notice that all the off-diagonal Az are skew symmetric. Let

V =

1 T
z z.
2

(9.15)

Its derivative along the dynamics of z is obtained as

V =

ci zi2 2 min ci V .

i=1

i=1

(9.16)

Therefore, we can conclude that the system is exponentially stable in z-coordinate.

From the property that i (0, . . . , 0) = 0 for i = 1, . . . , n, we can establish that
i (0, . . . , 0) = 0 for i = 1, . . . , n 1 and u(0, . . . , 0) = 0, which implies that
limt xi (t) = 0 for i = 1, . . . , n. Hence, we have established the following result.
Theorem 9.2. For a system in the form of (9.6), the control input (9.13) renders the
closed-loop system asymptotically stable.

9.3 Observer backstepping

We have presented integrator backstepping and iterative backstepping based on state
feedback. In this section, we present a control design for nonlinear systems using
output feedback. An observer is designed, and the observer state, or estimate of the
state variables of the original system, is then used for backstepping design.
Consider a nonlinear system which can be transformed to
x = Ac x + bu + (y)
y = Cx
with

(9.17)

148

Nonlinear and adaptive control systems

0 1 0 ...
0 0 1 ...

Ac = ... ... ... . . .

0 0 0 ...
0 0 0 ...

0
0

.. ,
.

1
0

0
..
.

0

b=
b ,

.
..
bn

T
1
0

C=. ,
..
0

where x Rn is the state vector; u R is the control; : R Rn with (0) = 0 is

a nonlinear function with element i being differentiable up to the (n i)th order;
and b Rn is a known constant Hurwitz vector with b  = 0, which implies that the
relative degree of the system is . This form of the system is often referred to as the
output feedback form. Since b is Hurwitz, the linear system characterised by (Ac , b, C)
is minimum phase. Note that a vector is said Hurwitz if its corresponding polynomial
is Hurwitz.
Remark 9.2. For a system in the output feedback form, if the input is zero, the
system is in exactly the same form as (8.10). We have shown the geometric conditions
in Chapter 8 for systems to be transformed to the output injection form (8.10), and
similar geometric conditions can be specied for nonlinear systems to be transformed
to the output feedback form. Clearly we can see that for the system (9.17) with any
observable pair (A, C), there exists a linear transformation to put the system in the

form of (9.17) with the specic (Ac , C).
Since the system (9.17) is in the output injection form, we design an observer as
x = Ac x + bu + (y) + L(y C x ),

(9.18)

where x Rn is the state estimate and L Rn is an observer gain designed such that
(A LC) is Hurwitz. Let x = x x , and it is easy to see
x = (Ac LC)x.

(9.19)

The backstepping design can be carried out with the state estimate x in steps.
From the structure of the system (9.17) we have y = x1 . The backstepping design will
start with the dynamics of y. In the following design, we assume > 1. In the case
of = 1, control input can be designed directly without using backstepping.
To apply the observer backstepping through x in (9.18), we dene
z1 = y,
zi = x i i1 , i = 2, . . . , ,

(9.20)

z+1 = x +1 + b u ,
where i , i = 1, . . . , , are stabilising functions decided in the control design.
Consider the dynamics of z1
z1 = x2 + 1 (y).

(9.21)

Backstepping design

149

We use x 2 to replace the unmeasurable x2 in (9.21), resulting at

z1 = x 2 + x 2 + 1 (y)
= z2 + 1 + x 2 + 1 (y).

(9.22)

We design 1 as
1 = c1 z1 k1 z1 1 (y),

(9.23)

where ci and ki for i = 1, . . . , are positive real design parameters. Comparing the
backstepping design using the output feedback with the one using state feedback, we
have one additional term, k1 z1 , which is used to tackle the observer error x 2 in the
closed-loop system dynamics. Then from (9.22) and (9.23), we have
z1 = z2 c1 z1 k1 z1 + x 2 .

(9.24)

Note that 1 is a function of y, i.e., 1 = 1 (y).

For the dynamics of z2 , we have
z2 = x 2 1
1
(x2 + 1 (y))
y
1
= z3 + 2 + 2 (y) + l2 (y x 1 )
(x2 + x 2 + 1 (y)).
y
= x 3 + 2 (y) + l2 (y x 1 )

where l2 is the second element of the observer gain L, and in the subsequent design,
li is the ith element of L. We design 2 as

2 = z1 c2 z2 k2
+

1
y

2
z2 2 (y) l2 (y x 1 )

1 (y)
(x2 + 1 (y)).
y

(9.25)

Note that 2 = 2 (y, x 1 , x 2 ). The resultant dynamics of z2 are given by


z2 = z1 c2 z2 k2

1
y

2
z2 + z 3

1
x 2 .
y

For the dynamics of zi , 2 < i , we have

zi = x i i1
= zi+1 + i + i (y) + li (y x 1 )
 i1
i1
x j .
(x2 + x 2 + 1 (y))

j
j=1
i1

150

Nonlinear and adaptive control systems

We design i , 2 < i , as
i = zi1 ci zi ki

i1
y

2
zi i (y) li (y x 1 )

 i1
i1
(x2 + 1 (y)) +
x j .
+
y
x j
j=1
i1

(9.26)

Note that i = i (y, x 1 , . . . , x i ). The resultant dynamics of zi , 2 < i , are given by



i1 2
i1
zi + zi+1
(9.27)
x 2 .
zi = zi1 ci zi ki
y
y
When i = , the control input appears in the dynamics of zi , and it is included in z+1 ,
as shown in the denition (9.20). We design the control input by setting z+1 = 0,
which gives
u=

(y, x 1 , . . . , x ) x +1
.
b

(9.28)

The stability result of the above control design is given in the following theorem.
Theorem 9.3. For a system in the form of (9.17), the dynamic output feedback control
with the input (9.28) and the observer (9.18) asymptotically stabilise the system.
Proof. From the observer error dynamics, we know that the error exponentially converges to zero. Since (Ac LC) is Hurwitz, there exists a positive denite matrix
P Rnn such that
(Ac LC)T P + P(Ac LC) = I .
This implies that for
Ve = x T P x ,
we have
V e = x 2 .

(9.29)

Let
Vz =

zi2 .

i=1

From the dynamics of z1 , z2 and zi , we can obtain that



2


i1
i1
2
2
zi x 2 ,
ci zi ki
zi
V z =
y
y
i=1
where we dene 0 = y for notational convenience. Note that if we ignore the two
terms concerning with ki and x 2 in the dynamics of zi , the evaluation of the derivative

Backstepping design

151

of Vz will be exactly the same as the stability analysis that leads to Theorem 9.2. For
the cross-term concerning with x 2 , we have, from Youngs inequality,





 i1
i1 2 2
1 2



k
z
x

zi +
x .
i
 y i 2 
y
4ki 2
Hence, we obtain that



1 2
2

ci zi +
x .
Vz
4ki 2
i=1

(9.30)

Let



1
Ve ,
V = Vz + 1 +
4d

where d = mini=1 ki . From (9.29) and (9.30), we have

 



1
1 2
2

ci zi +
x 2 1 +
x 2
V
4k
4d
i
i=1

ci zi2 x 2 .

i=1

Therefore, we can conclude that zi for i = 1, . . . , and x exponentially converge to

zero. Noticing that y = z1 , and 1 (0) = 0, we can conclude that limt x 2 (t) = 0,
which further implies that limt x2 (t) = 0. Following the same process, we can
show that limt xi (t) = 0 for i = 1, . . . , .
We still need to establish the stability property for xi , with i = + 1, . . . , n. Let

b+1
x+1

x

= ... ... ,
b
xn
bn
and from the system dynamics (9.17), it can be shown that

b+1
b+1
+1 (y)
x

.. (y)

..
+ B ... ,
= B +
.
.
b
b
n (y)
bn
bn
where

b+1 /b

b+2 /b

..
B=

bn1 /b
bn /b

1
0
..
.
0
0

... 0

..
. 0

..
..
.
.

... 1
... 0

(9.31)

152

Nonlinear and adaptive control systems

is a companion matrix associated with vector b. Since b is Hurwitz, the matrix B is

Hurwitz. All the terms other than B in the right-hand side of (9.31) converge to zero,
which implies that limt (t) = 0. Therefore, we can conclude limt xi (t) = 0 for
i = + 1, . . . , n. This concludes the proof.
2

9.4 Backstepping with ltered transformation

We have presented backstepping design for a class of dynamic systems using output
feedback in the previous section. The control design starts from the system output
and the subsequent steps in the backstepping design are carried out with estimates of
state variables provided by an observer. The backstepping design, often referred to
as observer backstepping, completes in steps, with being the relative degree of
the system, and other state estimates of xi for i > are not used in the design. Those
estimates are redundant, and make the dynamic order of the controller higher. There
is an alternative design method to observer backstepping for nonlinear systems in the
output feedback form, of which the resultant order of the controller is exactly 1.
In this section, we present backstepping design with ltered transformation for the
system (9.17), of which the main equations are shown here again for the convenience
of presentation,
x = Ac x + bu + (y)

(9.32)

y = Cx.
Dene an input lter
1 = 1 1 + 2
..
.

(9.33)

1 = 1 1 + u,
where i > 0 for i = 1, . . . , 1 are the design parameters. Dene the ltered
transformation
= x

d i i ,

(9.34)

i=1

where d i Rn for i = 1, . . . , 1 and they are generated recursively by

d 1 = b,
d i = (Ac + i+1 I )d i+1
We also denote
d = (Ac + 1 I )d 1 .

for i = 2, . . . , 1.

Backstepping design

153

From the ltered transformation, we have

= Ac x + bu + (y)

d i (i i + i+1 ) d 1 (i 1 + u)

i=1

= Ac +

Ad i i + (y) +

i=1

= Ac +

d i i i

i=1

(A + i I )d i i + (y)

d i i+1

i=1
2

i=1

d i i+1

i=1

= Ac + (y) + d1 .
The output under the coordinate is given by
y = C +

C d i i

i=1

= C
because C d i = 0 for i = 1, . . . , 1, from the fact that d i,j = 0 for i = 1, . . . ,
1, 1 j i. Hence, under the ltered transformation, the system (9.32) is then
transformed to
= Ac + (y) + d1 ,

(9.35)

y = C .
Let us nd a bit more information of d. From the denition
d 2 = (Ac + 1 I )b
we have
n


d 2,i sni = (s + 1 )

bi sni .

i=1

Repeating the process iteratively, we can obtain

n

i=1

di sni =

1

i=1

(s + i )

bi sni

(9.36)

which implies that d1 = b and that d is Hurwitz if b Hurwitz. In the special form
of Ac and C used here, b and d decide the zeros of the linear systems characterised
by (Ac , b, C) and (Ac , d, C) respectively as the solutions to the following polynomial
equations:

154

Nonlinear and adaptive control systems

n


bi sni = 0,

di sni = 0.

i=1

Hence, the invariant zeros of (Ac , d, C) are the invariant zeros of (Ac , b, C) plus i for
i = 1, . . . , 1. For the transformed system, 1 can be viewed as the new input. In
this case, the relative degree with 1 as the input is 1. The ltered transformation lifts
the relative degree from to 1.
As the ltered transformation may have its use independent of backstepping
design shown here, we summarise the property of the ltered transformation in the
following lemma.
Lemma 9.4. For a system in the form of (9.32) with relative degree , the ltered
transformation dened in (9.34) transforms the system to (9.35) of relative degree 1,
with the same high frequency gain. Furthermore, the zeros of (9.35) consist of the
zeros of the original system (9.32) and i for i = 1, . . . , 1.
We introduce another state transform to extract the internal dynamics of (9.35)
with Rn1 given by
d2:n
= 2:n
y,
(9.37)
d1
where Rn1 forms the state variable of the transformed system together with y,
the notation ()2:n refers to the vector or matrix formed by the 2nd row to the nth row.
With the coordinates ( , y), (9.35) is rewritten as
= D + (y)
y = 1 + y (y) + b 1 ,
where D is the companion matrix of d given by

d2 /d1 1 . . . 0

d3 /d1 0 . . . 0

..
..
..
.. ,
D=

.
.
.
.

dn1 /d1 0 . . . 1
dn /d1 0 . . . 0
and
(y) = D
y (y) =

d2:n
d2:n
y + 2:n (y)
1 (y),
d1
d1

d2
y + 1 (y).
d1

(9.38)

Backstepping design

155

If we view 1 as the input, the system (9.35) is of relative degree 1 with the stable
zero dynamics. For such a system, there exists an output feedback law to globally and
exponentially stabilise the system.
For this, we have the following lemma, stating in a more stand alone manner.
Lemma 9.5. For a nonlinear system (9.32), if the relative degree is 1, there exist
a continuous function : R R with (0) = 0 and a positive real constant c such
that the control input in the form of
u = cy (y)

(9.39)

globally and asymptotically stabilises the system.

Proof. Introducing the same transformation as (9.37), i.e.,
b2:n
y,
b1
we can obtain exactly the same transformed system as in (9.38) with d being replaced
by b and 1 by u. Since D is Hurwitz, there exists a positive denite matrix P such
that
= x2:n

DT P + PD = 3I .

(9.40)

We set
(y) = y (y) +

P 2 (y) 2
.
y

(9.41)

Note that (0) = 0, and therefore (y) 2 /y is well dened. The closed-loop system
is then obtained as
= D + (y),
y = 1 cy

P 2 (y) 2
.
y

Let
1
V = T P + y2 .
2
We have
V = cy2 3 2 + 2 T P(y) + y1 P 2 (y) 2 .
With the inequalities of the cross terms
|2 T P(y)| 2 + P 2 (y) 2 ,
1
|y1 | y2 + 2 ,
4
we have


1 2

y 2 .
V c
4

156

Nonlinear and adaptive control systems

Therefore, the proposed control design with c > 14 and (9.41) exponentially stabilises
the system in the coordinate ( , y), which implies the exponential stability of the
closed-loop system in x coordinate, because the transformation from ( , y) to x is
linear.
2
From Lemma 9.5, we know the desired value of 1 . But we cannot directly assign
a function to 1 as it is not the actual control input. Here backstepping can be applied
to design control input based on the desired function of 1 . Together with the ltered
transformation, the overall system is given by
= D + (y)
y = 1 + y (y) + b 1
1 = 1 1 + 2

(9.42)

...
1 = 1 1 + u,
to which the backstepping design is then applied. Indeed, in the backstepping design,
i for i = 1, . . . , 1 can be viewed as virtual controls.
Let
z1 = y,
zi = i1 i1 ,

for i = 2, . . . ,

z+1 = u ,
where i for i = 2, . . . , are stabilising functions to be designed. We also use the
positive real design parameters ci and ki for i = 1, . . . , and > 0.
Based on the result shown in Lemma 9.5, we have
1 = c1 z1 k1 z1 + y (y)

P 2 (y) 2
y

(9.43)

and
z1 = z2 c1 z1 k1 z1 P 2 (y) 2 .

(9.44)

For the dynamics of z2 , we have

1
y
y
1
= z 3 + 2 1 1
(1 + (y)).
y

z2 = 1 1 + 2

The design of 2 is then given by



1 2
1
2 = z1 c2 z2 k2
z2 +
(y) + 1 1 .
y
y

(9.45)

Backstepping design
The resultant dynamics of z2 is obtained as


1 2
1
z2 = z1 c2 z2 k2
z2 + z 3
1 .
y
y

157

(9.46)

Note that 2 = 2 (y, 1 ).

For the subsequent steps for i = 3, . . . , , we have
 i1
i1
y
j
y
j
j=1
i2

zi = i1 i1 + i

= zi+1 + i i1 i1

i2

i1
j=1

i1
(1 + (y))
y

(j i1 + j+1 ).

The design of i is given by


i = zi1 ci zi ki
+

i2

i1
j=1

i1
y

2
zi +

i1
(y)
y

(j j + j+1 ) + i1 i1 .

The resultant dynamics of zi is obtained as



i1 2
i1
zi = zi1 ci zi ki
1 .
zi + zi+1
y
y

(9.47)

(9.48)

Note that i = i (y, 1 , . . . , i1 ).

When i = , the control input appears in the dynamics of zi , through the term
z+1 . We design the control input by setting z+1 = 0, which gives u = , that is


1 2
1
u = z1 c z k
(y)
z +
y
y
+

i2

1
j=1

(j j + j+1 ) + 1 1 .

(9.49)

For the control design parameters, ci , i = 1, . . . , and can be any positive, and for
di , the following condition must be satised:


1
.
4ki
1=1

The stability result of the above control design is given in the following
theorem.

158

Nonlinear and adaptive control systems

Theorem 9.6. For a system in the form of (9.32), the dynamic output feedback
control (9.49) obtained through backstepping with the input ltered transformation
asymptotically stabilises the system.
Proof. Let
Vz =

zi2 .

i=1

From the dynamics for zi shown in (9.44), (9.46) and (9.48) we can obtain that



2

i1
i1
2
2
ci zi ki
zi
zi 1 P 2 (y) 2 ,
V z =
y
y
i=1
where we dene 0 = y for notational convenience. For the cross-term concerning
with 1 , we have





 i1
i1 2 2
1 2


zi +
.
 y zi 1  ki
y
4ki 1
Hence, we obtain that



1 2
ci zi2 +
1 P 2 (y) 2 .
V z
4ki
i=1

(9.50)

Let
V = T P
and we can obtain, similar to the proof of Lemma 9.5,
V 2 2 + P 2 (y) 2 .
Let
V = Vz + V .
and we have
V

ci zi2 +

i=1


1 2
1 2 2
4ki

ci zi2 2 .

(9.51)

(9.52)

i=1

Therefore, we have shown that the system (9.42) is exponentially stable under the
coordinate ( , z1 , . . . , z1 ). With y = z1 , we can conclude limt (t) = 0. From
y = z1 , and 1 (0) = 0, we can conclude that limt 1 (t) = 0. Following the same
process, we can show that limt (t) = 0 for i = 1, . . . , 1. Finally from the
ltered transformation (9.34), we can establish limt x(t) = 0.
2

Backstepping design

159

9.5 Adaptive backstepping

We have shown backstepping design for two classes of systems with state feedback
and output feedback respectively. In this section, we will show the nonlinear adaptive
control design for a class of system of which there are unknown parameters.
Consider a rst-order nonlinear system described by
y = u + T (y)

(9.53)

where : R Rp is a smooth nonlinear function, and Rp is an unknown vector

of constant parameters. For this system, adaptive control law can be designed as
u = cy T (y)
= y(y)

(9.54)
(9.55)

where c is a positive real constant, and Rpp is a positive denite gain matrix.
The closed-loop dynamics is given by
y = cy + T (y)
with the usual notation = .
For stability analysis, let
V =

1 2 1 T 1
y +
2
2

and its derivative is obtained as

V = cy2 ,
which ensures the boundedness of y and . We can show limt y(t) = 0 in the same
way by invoking Babalats Lemma as in the stability analysis of adaptive control
systems shown in Chapter 7.
Remark 9.3. The system considered above is nonlinear with unknown parameters.
However, the unknown parameters are linearly parameterised, i.e., the terms relating
to the unknown parameters, T (y) are linear with the unknown parameters, instead of
some nonlinear functions (y, ), which are referred to as nonlinearly parameterised.
Obviously, nonlinear parameterised unknown parameters are much more difcult to
deal with in adaptive control. In this book, we only consider linearly parameterised
unknown parameters.

For the rst-order system, the control input is matched with the uncertainty and
the nonlinear function. Backstepping can also be used with adaptive control to deal
with nonlinear and unknown parameters which are not in line with the input, or,
unmatched.

160

Nonlinear and adaptive control systems

Consider a nonlinear system
x 1 = x2 + 1 (x1 )T
x 1 = x3 + 2 (x1 , x2 )T
...

(9.56)

x n1 = xn + n1 (x1 , x1 , . . . , xn1 )
T

x n = u + n (x1 , x2 , . . . , xn )T ,
where xi R for i = 1, . . . , n are state variables; Rp is an unknown vector of
i



constant parameters; i : R R Rp for i = 1, . . . , n are differentiable
functions up to the order n i with i (0, . . . , 0) = 0; and u R is the control input.
Note that this system is exactly the same as (9.6) if the parameter vector is
known.
The backstepping design method will be applied iteratively in a similar way to
the control design for (9.6), with only difference of including adaptive parameters in
the control design.
Let
z1 = x1 ,
zi = xi i1 (xi , . . . , xi1 , ),

for i = 2, . . . , n,

zn+1 = u n (xi , . . . , xn , ),
1 = 1 ,
i = i

i1

i1
j=1

xj

j ,

for i = 2, . . . , n,

where i1 , for i = 2, . . . , n, are stabilising functions obtained in the adaptive

beackstepping design, and denotes an estimate of .
We start the adaptive backstepping from the dynamics of z1
z1 = z2 + 1 + 1T .
Design 1 as

1 = c1 z1 1T ,

(9.57)

where ci , for i = 1, . . . , n, are set of positive design parameters. The closed-loop

dynamics are obtained as

z1 = c1 z1 + z2 + 1T ,
where = .

Backstepping design

161

For the dynamics of z2 , we have

z2 = x 2 1
1
1
x 1

x1

1
1

.
= z3 + 2 + 2T
x2
x1

= x3 + 2T

The dynamics of z2 involve the adaptive law . Even though the adaptive law has
not been designed, it is surely known to the control design, and can be used in the
control input. However, the dynamics of zi with i > 2 will also affect the design of the
adaptive law. For this reason, we would like to leave the design of the adaptive law to
the end. Inevitably, the adaptive law will include zi for i > 2. This causes a problem,
if we use 2 to cancel at this step, because z3 depends on 2 . Instead, we only deal
with the part of the adaptive law that depends on z1 and z2 at this step, and we denote
that as 2 , which is a function of z1 , z2 and . In the subsequent steps, we use notations
i = i (z1 , . . . , zi , ) for i = 3, . . . , n, which are often referred to as tuning functions.
Based on the above discussion, we design 2 as
1
1
x2 +
2 .
x1

The closed-loop dynamics are obtained as
2 = z1 c2 z2 2T +

z2 = z1 c2 z2 + z3 + 2T

(9.58)

1
( 2 ).

Then the adaptive backstepping can be carried on for zi with 2 < i n. The
dynamics of zi can be written as
zi = x i i1
= xi+1 + iT

i1

i1
j=1

= zi+1 + i +

iT

xj

x j

i1

i1
j=1

xj

i1

xj+1

i1

The stabilising function i , for 2 < i n, are designed as

i = zi1 ci zi iT +

i1

i1
j=1

xj

xj+1 +

i1
i + i ,

(9.59)

where i = i (z1 , . . . , zi , ), for i = 3, . . . , n, are functions to be designed later to

tackle the terms ( i ) in stability analysis. The closed-loop dynamics are obtained
as, for 2 < i n,
zi = zi1 ci zi + zi+1 + iT

i1
( i ) + i .

162

Nonlinear and adaptive control systems

The control input u appears in the dynamics zn , in the term zn+1 . When i = n,
we have i = by denition. We obtain the control input by setting zn+1 = 0, which
gives
u = n
= zn1 cn zn nT
+

n1

n1

xj

j=1

xj+1 +

n1
+ n .

(9.60)

We need to design the adaptive law, tuning functions and i to complete the
control design. We will do it based on Lyapunov analysis. For notational convenience,
we set 1 = 2 = 0.
Let
n
1  2 1 T 1

V =
z + ,
(9.61)
2 i=1 i
2
where Rpp is a positive denite matrix. From the closed-loop dynamics of zi ,
for i = 1, . . . , n, we obtain
n

i1
zi
( i ) + T 1

i=1
i=2

n
n 


i1
2

zi i zi
ci zi +
( i )
=

i=1
i=2
 n
T

1
+
z .

V =

(ci zi2 + zi iT + zi i )

i i

i=1

We set the adaptive law as

=

zi i .

(9.62)

i=1

We can conclude the design by setting the tuning functions to satisfy


n 

i1

zi i z i
( i ) = 0.

i=2
Substituting the adaptive law in the above equation, and with some manipulation of
index, we obtain that

n
n



zi i zi i1
0=
zj j i

i=2

n

i=2

j=1

z i i

n
n 

i=2 j=2

 i1
i1
j +
(i z1 1 )
zi zj
zi


n

i=2

Backstepping design
=

zi i

n
n 


i=2

i=2 j=i+1

i
n 


zi i

i1
j

 i1
i1
zi
j +
(i z1 1 )


n

zi zj

i=2 j=2

zi zj

163

i=2

j1
n 


i=2

j=3 i=2

zi zj

i1
j

n
i


i1
+
zi
i z1 1
zj j

i=2

j=2

i1

j1
=
i
zj

i=3
j=2

n
i


i1
+
i z1 1
zi
zj j .

zi i

i=2

j=2

Hence, we obtain
i1

j1
i =
i ,
zj

j=2

i =

zj j ,

for i = 3, . . . , n,

for i = 2, . . . , n.

(9.63)

(9.64)

j=1

we nally obtain that

With the complete design of u and ,
V =

ci zi2 .

i=1

from which we can deduce that zi L2 L , i = 1, . . . , n, and is bounded. Since

all the variables are bounded, we have zi bounded. From Barbalats Lemma, we have
limt zi (t) = 0, i = 1, . . . , n. Noticing that x1 = z1 and 1 (0) = 0, we can conclude
that 1 converges to zero, which further implies that limt x2 (t) = 0. Repeating the
same process, we can show that limt xi (t) = 0 for i = 1, . . . , n.
Theorem 9.7. For a system in the form of (9.57), the control input (9.60) and adaptive
law (9.62) designed by adaptive backstepping ensure the boundedness of all the
variables and limt xi (t) = 0 for i = 1, . . . , n.

164

Nonlinear and adaptive control systems

Example 9.2. Consider a second-order system

x 1 = x2 + (ex1 1)
x 2 = u,
where R is the only unknown parameter. We will follow the presented design
procedure to design an adaptive control input to stabilise the system.
Let z1 = x1 and z2 = x2 1 . We design the stabilising function 1 as
1 = c1 z1 (ex1 1) .
The resultant dynamics of z1 are given by
z1 = c1 z1 + z2 + (ex1 1) .
The dynamics of z2 are obtained as
z2 = u

1
1

(x2 + (ex1 1) )
,
x1

where
1

= c1 ex1 ,
x1

1
= (ex1 1).

Therefore, we design the control input u as

1
1

(x2 + (ex1 1) ) +
.
x1

Note that the control input u contains , which is to be designed later, as a function
of the state variables and . The resultant dynamics of z2 are obtained as
u = z1 c2 z2 +

z2 = z1 c2 z2
Let

1 x1
(e 1) .
x1



1 2
2
2
V =
z + z2 +
.
2 1

We obtain that



1 x1
1
2
2
x1

V = c1 z1 c2 z2 + z1 (e 1) z2
(e 1) .
x1

Therefore, an obvious choice of the adaptive law is

1 x1
= z1 (ex1 1) z2
(e 1)
x1
which gives
V = c1 z12 c2 z22 .
The rest part of the stability analysis follows Theorem 9.7.

Backstepping design

165

Example 9.3. In Example 9.2, the nonlinear system is in the standard format as
shown in (9.56). In this example, we show adaptive control design for a system which
is slightly different from the standard form (9.56), but the same design procedure can
be applied with some modications.
Consider a second-order nonlinear system
x 1 = x2 + x13 + x12
x 2 = (1 + x12 )u + x12 ,
where R is the only unknown parameter.
Let z1 = x1 and z2 = x2 1 . The stabilising function 1 is designed as
1 = c1 z1 x13 x12 ,
which results in the dynamics of z1 as
z1 = c1 z1 + z2 + x13 .
The dynamics of z2 are obtained as
z2 = (1 + x12 )u + x12

1
1

(x2 + x13 + x12 )

,
x1

where
1
1
= x13 .
= c1 3x12 2x1 ,
x1

Therefore, we design the control input u as


1
1
1
2
3
2

u=
z1 c2 z2 x1 +
(x2 + x1 + x1 ) +
.
x1
1 + x12

The resultant dynamics of z2 are obtained as
z2 = z1 c2 z2 + x12
Let

1 3
x .
x1 1



1 2
2
2
V =
z + z2 +
.
2 1

We obtain that




1
1 3

V = c1 z12 c2 z22 + z1 x1 3 + z2 x12

.
x1
x1

We can set the adaptive law as



= z x 3 + z x2 1 x3
1 1
2
1
x1 1
to obtain
V = c1 z12 c2 z22 .

166

Nonlinear and adaptive control systems

2
x1

x2

State variables

0
1
2
3
4
5
6
0

10

10

t (s)

Figure 9.1 State variables

4
3.5

Estimated parameter

3
2.5
2
1.5
1
0.5
0

6
t (s)

Figure 9.2 Estimated parameter

Backstepping design

167

The rest part of the stability analysis follows Theorem 9.7. Simulation results are
shown in Figures 9.1 and 9.2 with x(0) = [1, 1]T , c1 = c2 = = = 1. The state
variables converge to zero as expected, and the estimated parameter converges to
a constant, but not to the correct value = 1. In general, estimated parameters in
adaptive control are not guaranteed to converge to their actual values.


9.6 Adaptive observer backstepping

Backstepping can also be used to design control input for a class of nonlinear systems
with unknown parameters using output feedback. We consider a system which can be
transformed to the output feedback form with unknown parameters
x = Ac x + b (y)u + 0 (y) +

i (y)ai

i=1

:= Ac x + b (y)u + 0 (y) + (y)a

(9.65)

y = Cx
with

T
0 1 0 ... 0
1
0 0 1 ... 0

Ac = ... ... ... . . . ... , C = . ,

..
0 0 0 ... 1
0
0 0 0 ... 0

0

..
a1
.



a2
0

:= 0(1)1 ,
a = . , b =
b
b
..

.
ap
..
bn

where x Rn is the state vector, u R is the control and b Rn is an unknown

constant Hurwitz vector with b  = 0, which implies that the relative degree of the
system is , a denotes unknown system parameters, i : R Rn for 0 i p
are nonlinear functions with their elements being differentiable up to the th order,
: R R is a continuous function and (y)  = 0, y R.
For the system (9.65), we assume that only the system output is available for
control design. Both observer backstepping and backstepping with ltered transformation design methods presented earlier can be applied for the control design of
(9.65) if there are no unknown parameters. Even with the unknown parameters, the
same ltered transformation introduced earlier in (9.34) can still be applied to the
system (9.65), and adaptive backstepping can then be applied to design the control
input. Here we take an approach similar to observer backstepping. Because there

168

Nonlinear and adaptive control systems

are unknown parameters, we will not be able to design an observer in the same way
as in the observer backstepping. Instead, we can design lters similar to observers,
and obtain an expression of state estimation which contains unknown parameters. The
unknown parameters in the state estimation are then tackled by adaptive backstepping.
For the state estimation, we re-arrange the system as
x = Ac x + 0 (y) + F T (y, u) ,

(9.66)

where the vector Rq , with q = n + 1 + p, is dened by

 
b
=
a
and


F(y, u) =
T



0(1)(n+1)
(y)u, (y) .
In+1

Similar to observer design, we design the following lters:

= A0 + Ly + 0 (y),

(9.67)

= A0  + F(y, u) ,


(9.68)

where Rn , T Rnq and

L = [l1 , . . . , ln ]T , A0 = Ac LC
with L being chosen so that A0 is Hurwitz. An estimate of the state is then given by
x = + T .

(9.69)

Let
 = x x
and from direct evaluation, we have
 = A0 .

(9.70)

Therefore, the state estimate shown in (9.69) is an exponentially convergent one.

Notice that this estimate contains unknown parameter vector , and therefore x cannot
be directly used in control design. The relationship between an convergent estimate
x and the unknown parameter vector can be used in adaptive backstepping control
design.
We can reduce the order of the lters. Let us partition T as T = [v, ] with
v Rn(n+1) and  Rnp . We then obtain, from (9.68), that
= A0  + (y),

vj = A0 vj + ej (y)u,

for j = , . . . , n,

where ej denotes jth column of identity matrix I in Rn . For 1 < j < n, we have
A0 ej = (Ac LC)ej = Ac ej = ej+1 .

Backstepping design

169

This implies that

nj

vj = A0 vn .
Finally, we summarise the lters for T as
T

vj

=
=
=
=

[v , . . . , vn , ],
A0  + (y),
A0 + en (y)u,
nj
A0 , for j = , . . . , n.

(9.71)

With the lters being designed, control design can be carried out using adaptive
backstepping in a similar way as the observer backstepping, by combining the design
of tuning functions in parameter adaptation. During the control design, the state
variable x will be replaced by
x = + T + ,
and in particular, whenever we encounter x2 , we will replace it by
x2 = 2 + T(2) + 2 ,
where the subscript (i) denotes the ith row of a matrix. In the following control design,
we will consider the tracking control instead of stabilisation. The output y is designed
to track a trajectory yr with its derivatives available for the control design.
Let us dene a number of notations:
z1 = y yr ,
zi = v,i y
r(i1) i1 , i = 2, . . . , ,
z+1 = (y)u + v,+1 y
r() ,
0 = 2 + 0,1 ,
= [v,2 , . . . , vn,2 , (1) + (2) ]T ,
= [0, v+1,2 , . . . , vn,2 , (1) + (2) ]T ,
i = [1 , . . . , i ]T ,
y i = [yr , y r , . . . , yr(i) ]T ,
Xi = [ T , vec()T , ,
Ti , y iT ]T ,
j,i =

j1 i1
,

y

where ,
an estimate of  = 1/b , and i , i = 1, . . . , , are the stabilising functions
to be designed.
Consider the dynamics of z1
z1 = x2 + 0,1 (y) + (1) y r
= 2 + T(2) + 2 + 0,1 (y) + (1) y r
= b v,2 + 0 + T + 2 y r .

(9.72)

170

Nonlinear and adaptive control systems

Note that b is unknown. To deal with unknown control coefcient, we often estimate
its reciprocal, instead of itself, to avoid using the reciprocal of an estimate. This is the
reason why we dene
yr
z2 = v,2 1 
:= v,2  1 
yr .
From b  = 1, we have

b  = 1 b ,
where  =  .
Then from (9.72), we have
yr ) + 0 + T + 2 y r
z1 = b (z2 +  1 + 
= b z2 b (
1 + y r ) + 1 + 0 + T + 2 .

(9.73)

Hence, we design

1 = c1 z1 k1 z1 0 T ,

(9.74)

where ci and ki for i = 1, . . . , are positive real design parameters. Note that with
=  1 , we have 1 = 1 (y, X1 , ).
The resultant closed-loop dynamics are obtained as
1 + y r ) + T + 2
z1 = c1 z1 k1 z1 + b z2 b (
= c1 z1 k1 z1 + b z2 + b z2 b (
1 + y r ) + T + 2
= c1 z1 k1 z1 + b z2 b (
1 + y r )
+ ( (
1 + y r )e1 )T + 2 ,

(9.75)

Note that b = T e1 .
where = .
As shown in the observer backstepping, the term k1 z1 is used to tackle the
error term 2 . In the subsequent steps, the terms headed by ki are used to deal with
the terms caused by 2 in stability analysis. Indeed, the method of tackling observer
errors is exactly the same as in the observer backstepping when all the parameters are
known. The adaptive law for  can be designed in this step, as it will not appear in
the subsequent steps,
 = sgn(b )( 1 + y r ))z1 ,

(9.76)

where is a positive real constant.

Similar to adaptive backstepping shown in the previous section, the unknown
parameter vector will appear in the subsequent steps, and tuning functions can be
introduced in a similar way. We dene the tuning functions i as
yr + 1 )e1 )z1 ,
1 = ( (
i = i1 yi1 zi ,

i = 2, . . . , ,

(9.77)

where i are the stabilising functions to be designed, and Rqq is a positive

denite matrix, as the adaptive gain.

Backstepping design

171

For the dynamics of z2 , we have

yr 
yr
z2 = v ,2 1 
= v,3 l2 v,1

1
1
1
X1

yr 
y
yr .
X1
y

With
y = 0 + T + 2 ,
z3 = v,3 2 
yr ,
we obtain that
z2 = z3 + 2 l2 v,1

1
1
1
X1

yr ,
(0 + T + 2 )
X1
y

from which the stabilising function 2 is dened as



1 2
yr

z2 + l2 v,1 + 
2 = b z1 c2 z2 k2
y
+

1
1

+ 1 2 .
X1 +
(0 + T )
X1
y

(9.78)

The resultant dynamics of z2 are obtained as



1 2
z2 + z 3
z2 = b z1 c2 z2 k2
y

1 T
1
1

2
( 2 ).
y
y

(9.79)

For the dynamics of zi , 2 < i , we have

zi = zi+1 + i li v,1

i1
i1
Xi1
(0 + T + 2 )
Xi1
y

i1 (i1)
y
r .

We design i , 2 < i , as
i = zi1 ci zi ki
(i1) +
+ y
r

i1
y

2
zi + li v,1

i1
i1

Xi1 +
(0 + T )
Xi1
y


i1
i
j,i zj ,

i1

i = 3, . . . , ,

(9.80)

j=2


where the last term i1
j=2 j,i zj is similar to the term i in the adaptive backstepping
with tuning functions. The resultant dynamics of zi are obtained as

172

Nonlinear and adaptive control systems


zi = zi1 ci zi ki

i1
y

2
zi + zi+1


i1 T
i1

j,i zj
2
y
y
j=2

i1
( i )

i1

i = 3, . . . , .

(9.81)

Now we design the adaptive law for as

= .

(9.82)

The control input is obtained by setting z+1 = 0 as

u=

1
r() ).
( v,+1 + y
(y)

(9.83)

For the adaptive observer backstepping, we have the following stability result.
Theorem 9.8. For a system in the form of (9.65), the control input (9.83) and adaptive laws (9.76) and (9.82) designed by adaptive observer backstepping ensure the
boundedness of all the variables and limt (y(t) yr (t)) = 0.
Proof. With the control design and adaptive laws presented earlier for the case > 1,
the dynamics of zi , for i = 1, . . . , can be written as
z1 = c1 z1 k1 z1 + 2 + ( (
yr + 1 )e1 )T
b (yr + 1 ) + b z2 ,
z2 = b z1 c2 z2 k2

1
y

(9.84)
2
z2 + z 3


1
1 T

2 +
2,j zj ,
y
y
j=3


zi = zi1 ci zi ki

i1
y

(9.85)

2
zi + zi+1


i1 T
i1

2 +
i,j zj
y
y
j=i+1
i1

j=2

j,i zj

i = 3, . . . , .

(9.86)

Backstepping design

173

Let

1  2 1 T 1
|b | 2  1 T
V =
zi + +
 P,
 +
2 i=1
2
2
4ki
i=1

(9.87)

where P is a positive denite matrix that satises

AT0 P + PA0 = I .
From (9.84)(9.86), (9.82) and (9.76), it can be shown that






i1 2 2  i1
1

V =
ci + k i
zi
zi
2
 2 ,
y
y
4k
i
i=1
i=1
i=1
0
= 1. Noting
where we set
y



2
 i1 
zi
 1 2 2 + ki i1 z 2 ,

2
i
 y
 4k
y
i

(9.88)

(9.89)

we have
V

ci zi2 .

(9.90)

i=1

 and  are bounded. The boundedness of

This implies that zi , i = 1, . . . , , ,
further implies that is bounded. The boundedness of  follows from the fact that
 = 1/b is a constant. Since y = z1 + yr , the boundedness of and  follows the
boundedness of y. The boundedness of can be established from the minimum phase
property of the system. Therefore, u is bounded and we can conclude that all the
variables of the feedback control system are bounded. From the above analysis, we
can deduce that zi L2 L , i = 1, . . . , and are bounded. Since all the variables
are bounded, we have zi bounded. From Barbalats Lemma, we have limt zi (t) =
0, i = 1, . . . , , of which the result for z1 means the asymptotic output tracking,
limt (y(t) yr (t)) = 0.
2

Chapter 10

Disturbance rejection and output regulation

Disturbances are often inevitable in control system design. A well-performed controller is expected to suppress undesirable effects of the disturbances in the system.
There are various types of disturbances in physical systems, from random disturbances, wide-band, narrow-band disturbances, in terms of disturbance power spectra,
to deterministic disturbances that include harmonic disturbances, i.e., sinusoidal functions, general periodic disturbances and other deterministic signals generated from
nonlinear dynamic systems such as limit cycles. The spectral information of random
disturbances may be considered in loop-shaping and other conventional design methods. In this chapter, we concentrate on suppression and rejection of deterministic
periodic disturbances. One control design objective is to track a specic signal. If we
take the tracking error as the state variable, the tracking problem could be converted
to a stabilisation problem. Indeed, we can formulate both the disturbance rejection
and output tracing problems in terms of output regulation. This will be discussed later
in this chapter.
Disturbance rejection and output regulation are big topics in control systems,
design. We have to be selective to limit the contents in one chapter. In this chapter, we
will concentrate on rejection of deterministic disturbances in a class of nonlinear output feedback systems. As for disturbances, we will start from sinusoidal disturbances
with unknown frequencies, then disturbances generated from nonlinear exosystems
and then general periodical disturbances, etc. Adaptive control techniques are used
to deal with the unknown disturbance frequencies and unknown parameters in the
system. The presented design concepts can be applied to other classes of nonlinear
systems.

10.1 Asymptotic rejection of sinusoidal disturbances

In this section, we consider asymptotic rejection of sinusoidal disturbances with
unknown frequencies for dynamic systems in the output feedback form.
We consider a SISO nonlinear system which can be transformed into the output
feedback form
= Ac + bu + (y) + Ew
y = C ,

(10.1)

176

Nonlinear and adaptive control systems

with

0 1 0 ...
0 0 1 ...

Ac = ... ... ... . . .

0 0 0 ...
0 0 0 ...

0
0

.. ,
.

1
0

0
..
.

0

b=
b ,

.
..
bn

T
1
0

C=. ,
..
0

where Rn is the state vector; u R is the control; : R Rn with (0) = 0 is

a nonlinear function with element i being differentiable up to the (n i)th order;
b Rn is a known constant Hurwitz vector with b  = 0, which implies the relative
degree of the system is ; E Rnm is a constant matrix; and w Rm are disturbances,
and they are generated from an unknown exosystem
w = Sw
of which, S is a constant matrix with distinct eigenvalues of zero real parts.
Remark 10.1. For the system (10.1), if the disturbance w = 0, it is exactly same
as (9.17), of which observer backstepping can be used to design a control input.
Due to the unknown disturbance, although the observer backstepping presented in
Chapter 9 cannot be applied directly for control design, a similar technique can be
developed using an observer and an adaptive internal model. Also note that the
linear system characterised by (Ac , b, C) is minimum phase.

Remark 10.2. The dynamic model w = Sw is referred to as an exosystem, because
w is a disturbance, not a part of the system state. This is a convention adopted for
disturbance rejection and output regulation. Of course, one could argue that w could
be considered as a part of the system state, or at least, of an augmented system. In
such a case, w is not controllable.

Remark 10.3. With the assumption that S has distinct eigenvalues with zero real
parts, w is restrict to sinusoidal signals (sinusoidal disturbances) with a possible
constant bias. This is a common assumption for disturbance rejection. Roughly
speaking, all the periodic signal can be approximated by nite number of sinusoidal
functions.

The disturbance rejection problem to be solved here is to design a control input
that ensures the boundedness of the variables in the closed-loop system, and the
convergence to zero of the system output.
To solve the problem, we start from state transformation, based on an invariant
manifold. The basic idea for disturbance suppression or output regulation is the internal model principle. A controller for disturbance rejection or output regulation should
generate a feedforward input term to cancel the inuence caused by the disturbance or
to track a desired trajectory. This feedforward term is also referred to as the equivalent

Disturbance rejection and output regulation

177

input disturbance. For nonlinear systems, asymptotic disturbance rejection depends

on the existence of an invariant manifold. We shall show that there exists an invariant
manifold for any exosystem specied in (10.1). This invariant manifold is then used
in the state transformation. The following lemma summarises the results.
Lemma 10.1. For the system (10.1) with an exosystem whose eigenvalues are distinct
and with zero real parts, there exist (w) Rn with 1 (w) = 0 and (w) R such
that
(w)
(10.2)
Sw = Ac (w) + Ew + b(w).
w
Proof. Asymptotic disturbance rejection aims at y = 0, which implies that 1 = 0.
The manifold is invariant and it should satisfy the system equation (10.1) with
y 0. From the rst equation of (10.1), we have
2 = E1 w

(10.3)

where E1 denotes the rst row of E. Furthermore, we have, for 2 i ,

d
i1 Ei1 w.
dt
From equations to n of (10.1), we obtain
i =

(10.4)

n
n


d ni
d n+1
d ni
b
(w)
=

Ei w.
i

dt ni
dt n+1
dt ni
i=
i=

(10.5)

A solution of (w) can always be found from (10.5). With (w), we can write, for
< i n,
i =

d
i1 Ei1 w bi1 (w).
dt

With the invariant manifold (w), we dene a transformation of state as

x = (w).

(10.6)
2
(10.7)

It can be easily shown from (10.1) and (10.2) that

x = Ac x + b(u ) + (y)
y = Cx.

(10.8)

The stabilisation and disturbance suppression problem of (10.1) degenerates to the

stabilisation problem of (10.8).
For dynamic output feedback control, state variables need to be estimated for
the control design directly or indirectly. The difculty for designing a state estimator
for (10.8) is that (w), the feedforward control input for disturbance suppression, is
unknown. Let us consider the observers
p = (Ac LC)p + (y) + bu + Ly
q = (Ac LC)q + b(w),

(10.9)
(10.10)

178

Nonlinear and adaptive control systems

where L Rn is chosen so that Ac LC is Hurwitz, and q denotes the steady-state

contribution of (w) to the state variable. Notice that the observer (10.10) cannot be
implemented because (w) is unknown, due to the unknown exosystem. Nevertheless,
if we dene
x = p q

(10.11)

then the error of the observer dened by  = x x is an exponentially decaying signal

with the dynamics
 = (Ac LC),

(10.12)

which can be obtained by a direct evaluation from (10.8) to (10.10).

Observe from (10.5) that (w) is a linear combination of w. Since the lter
(10.10) is just a stable linear system, the steady-state solution of state variables is
linear combinations of w as well. Therefore, there exists an l Rm for q2 , and we can
write
w = Sw
(10.13)
q2 = l T w.
We re-parameterise (10.13) for state estimation here for q2 . For any known controllable
pair (F, G) with F Rmm being Hurwitz and G Rm , there exists a Rm so that
= (F + G T )

(10.14)

q2 = T ,
with the initial value (0) dependent on exogenerous variables.

Remark 10.4. The importance of (10.14) compared with (10.13) is the re-formulation
of the uncertainty caused by the unknown exosystem. The uncertainty in (10.13)
parameterised by unknown S and l is represented by a single vector in (10.14). The
relation between the two parameterisations is discussed here. Suppose that M Rmm
is the unique solution of
MS FM = Gl T .

(10.15)

The existence of a non-singular M is ensured by the fact that S and F have exclusively different eigenvalues, and (S, l) and (F, G) are observable and controllable
respectively. From (10.15), we have
MSM 1 = F + Gl T M 1 ,

(10.16)
1

which implies = Mw and = l M

To achieve asymptotic rejection with global stability using output feedback, an
asymptotic state estimator, probably depending on unknown parameters, is instrumental. In the control design shown later for the problem considered, the estimate of
state variable x2 is crucial, and therefore we introduce the internal model (10.14) to
describe the inuence of disturbance w on this state.

T

The following lemma summarises the results on state observation.

Disturbance rejection and output regulation

179

Lemma 10.2. The state variable x can be expressed as

x = p q + ,

(10.17)

where p is generated from (10.9) with q and  satisfying (10.10) and (10.12)
respectively. In particular,
x2 = p2 T + 2 ,

(10.18)

where satises (10.14).

Remark 10.5. The expression (10.18) cannot be directly implemented, because
and are not available. The relation shown in (10.18) is very useful in the control
design, and it allows adaptive control technique to be introduced to deal with the
unknown parameter later.

Based on the parameterisation (10.14) of the internal model, we design an
estimator of as
= (F + G T ) + (y)
q 2 = T ,

(10.19)

where (y) is an interlace function to be designed later.

With the state estimation introduced in previous chapter, in particular, the lter
(10.9), the relation (10.18) and estimator (10.19), the control design can be carried out
using adaptive observer backstepping technique described in previous chapter. In the
backstepping design, ci , for i = 1, . . . , , denote constant design parameters which
can be set in the design, while ki , for i = 1, . . . , , denote the unknown parameters
depending on an upper bound of . The estimates, ki , for i = 1, . . . , , are then
used. To apply the observer backstepping through p in (10.9), we dene
z1 = y,
zi = pi i1 ,

(10.20)
i = 2, . . . , ,

z+1 = b u + p+1 ,

(10.21)
(10.22)

where i , for i = 1, . . . , , are stabilising functions decided in the control design.

Consider the dynamics of z1
z1 = x2 + 1 (y).

(10.23)

We use (10.18) to replace the unmeasurable x2 in (10.23), resulting at

z1 = p2 T + 2 + 1 (y)
= z2 + 1 T + 2 + 1 (y).

(10.24)

We design 1 as
1 = c1 z1 k1 z1 1 (y) + T .

(10.25)

180

Nonlinear and adaptive control systems

Then from (10.25) and (10.24), we have

z1 = z2 c1 z1 k1 z1 + 2 + ( T T ).

(10.26)

After the design of 1 , the remaining stabilising functions can be designed in a similar
way to the standard adaptive backstepping with tuning functions shown in the previous
chapter. Omitting the deriving procedures, we briey describe the nal results

i1 2

zi li (y p1 ) i (y)
i = zi1 ci zi ki
y
+

 i1
 i1
i1
i1
(p2 T + 1 ) +
p j +
kj +

j
y
p

k
j=1
j=1

 j1 i1
i1
i +

zj ,
y

i1

i1

i1

i = 2, . . . , ,

(10.27)

j=2

where li are the ith element of the observer gain L in (10.9),  is a positive denite
matrix, i , i = 2, . . . , , are the tuning functions dened by
i =

i

j1

zj ,
y
j=1

where we set

0
y

i = 2, . . . , ,

(10.28)

= 1. The adaptive law for is set as

= .
The adaptive laws for ki , i = 1, . . . , , are given by

i1 2 2

zi ,
ki = i
y

(10.29)

(10.30)

where i is a positive real design parameter. The control input is obtained by setting
z+1 = 0 as
u=

1
( p+1 ).
b

(10.31)

Considering the stability, we set a restriction on c2 by

c2 > PG2 ,

(10.32)

where P is a positive denite matrix, satisfying

F T P + PF = 2I .

(10.33)

To end the control design, we set the interlace function in (10.19) by

(y) = (FG + c1 G + k1 G)y.

(10.34)

Remark 10.6. The adaptive coefcients ki , i = 1, . . . , , are introduced to allow the

exosystem to be truly unknown. It implies that the proposed control design can reject

Disturbance rejection and output regulation

181

disturbances completely at any frequency. If an upper bound of  is known, we

can replace ki , i = 1, . . . , , by constant positive real parameters.

Remark 10.7. The nal control u in (10.31) does not explicitly contain (w), the
feedforward control term for disturbance rejection. Instead, the proposed control
design considers q2 , the contribution of the inuence of (w) to x2 , from the rst step
in 1 throughout to the nal step in .

Theorem 10.3. For the system (10.1), the control input (10.31) ensures the boundedness of all the variables, and asymptotically rejects the unknown disturbances in
the sense that limt y(t) = 0. Furthermore, if w(0) Rm is such that w(t) contains
= .
the components at m/2 distinct frequencies, then limt (t)
Proof. We start from the analysis of the internal model. Dene
e = Gy.

(10.35)

From (10.14), (10.19), (10.34) and (10.26), it can be obtained that

e = Fe G(z2 + 2 ).

(10.36)

Based on the stabilising functions shown in (10.27), the dynamics of zi , for

i = 2, . . . , , can be written as
zi = zi1 ci zi ki

i1
y

2
zi + zi+1


i1
i1 j
2
zj


y
y
j=i+1

i1

j1 i1

zj ,

y
j=2

i = 2, . . . , ,

i1 T
( T )
y

(10.37)

where the term T T can be rephrased by

T T = T e T + T Gz1

(10.38)

with = .
In the following analysis, we denote 0 , i,j , i = 1, . . . , 3, and j = 1, . . . , , as
constant positive reals, which satisfy
0 +


j=1

3,j <

1
.
2

(10.39)

182

Nonlinear and adaptive control systems

Furthermore, we set constant positive reals ki , i = 1, . . . , , satisfying the following

conditions:
k1 > | T G| + 1,1 +

2,i | T G|2 +

j=2

ki > 1,i +

1
2
+
42,i
43,i

2
,
43,1

i = 2, . . . , .

(10.40)

(10.41)

Dene a Lyapunov function candidate






1 T
2
1 2
T
1
T
V =
e Pe +
zi +
i ki +  +  P  ,
2
i=1
i=1

(10.42)

where ki = ki ki , i = 1, . . . , , P is a positive denite matrix satisfying

(Ac LC)T P + P (Ac LC) = 2I ,
and is a constant positive real satisfying

>

PG2  1
+
.
40
41,i
j=1

(10.43)

Evaluating the derivative of V along the dynamics in (10.12), (10.26), (10.36)

and (10.37) together with adaptive laws in (10.29) and (10.30), we have




i1

i1
2

ci + k i
zi
zi
2
V =
y
y
i=1
i=1


i1 T
zi
( e + T Gz1 )
y
i=1

eT e eT PG(z2 + 2 )  T .

(10.44)

For the cross-terms in (10.44), we have

T

e PG2 < 0 eT e + 1 PG2  2 ,
2
40

2
i1
i1
1 2
zi

zi2 +
 ,
i = 1, . . . , ,
y 2 < 1,i
y
41,i 2

i1 T
i1 2 2
1
T
2 2


zi ,
i = 2, . . . , ,
y Gz1 zi < 2,i | G| z1 + 4
y
2,i

i1 T
2 i1 2 2
T


zi ,
i = 1, . . . , ,
y ezi < 3,i e e + 4
y
3,i
e
|eT PGz2 | < eT + PG2 z22 /2.
2

Disturbance rejection and output regulation

183

Then based on the conditions specied in (10.32), (10.39), (10.40), (10.41) and
(10.43), we can conclude that there exist positive real constants i , for i = 1, 2, 3,
such that


V 1
zi2 2 eT e 3  T .
(10.45)
i=1

We conclude zi L2 L , i = 1, . . . , , and e L2 L and the boundedness

and ki , i = 1, . . . , . With y = z1 L , the boundedness of p
of the variables ,
can be established from the minimum phase property of the system and the neutral
stability of w. Therefore, we can conclude that all the variables in the proposed
closed-loop system are bounded. Since the derivatives of zi and e are bounded, from
zi L2 L , i = 1, . . . , , e L2 L and Babalats lemma, we further conclude
that limt zi = 0, i = 1, . . . , , and limt e = 0.
From (10.35), we have limt  (t) (t) = 0, which means that asymptotically converges to . From the boundedness of all the variables and limt zi = 0,

i = 1, . . . , , we can conclude that limt = 0 and limt ki = 0, i = 1, . . . , ,

which implies that the adaptive controller will converge to a xed-parameter type.
Since limt = 0, there exists a
We now establish the convergence of .

= limt (t).
From (10.14), (10.19) and (10.35), we obtain
d
( ) = F( ) + G( )T + (t),
dt

(10.46)

where
T
T + G
(t) = G( )
(e + Gy) .

From limt ( ) = 0 and limt (t) = 0, we can conclude that

( )T = 0,

(10.47)

because, if otherwise, ( )T would be a persistently excited signal, and we

could never have limt ( ) = 0. If the disturbance w(t) contains components
at m/2 distinct frequencies, the signal (t) is persistently excited, and from (10.47)
= .
2
we have = 0, i.e., limt (t)
Remark 10.8. The condition imposed on w(0) does not really affect the convergence
It is added to avoid the situation that w(t) degenerates to have less independent
of .
frequency components. In that situation, we can reform the exosystem and E in (10.1)
with a smaller dimension m
such that the reformed w(t) is persistently excited in
together
reduced space Rm and accordingly we have , Rm . With the estimate ,
with (F, G), the disturbance frequencies can be estimated.

Example 10.1. Consider the nonlinear system
1 = 2 + (ey 1) + w1
2 = u + w1
y = 1 ,

184

Nonlinear and adaptive control systems

where w1 R is a disturbance with two sinusoidal components at unknown frequencies 1 and 2 . An augmented disturbance vector w = [w1 , w2 , w3 , w4 ]T satises
w = Sw with eigenvalues of S at {j1 , j2 }. It is easy to obtain that = [0, w1 ]T
and = w1 w 1 . Using the state transform x = , we have
x 1 = x2 + (ey 1)
x 2 = u

(10.48)

y = x1 .
A transform w = T w can also be introduced to the disturbance model so that the
disturbance is generated by



0
1
022
1 0


 w := S w
w =

0
2
022
2 0
w1 = [t11 , t12 , t13 , t14 ]w,

where T , 1 and 2 are unknown. With the coordinate w,

we have
(w) = [t11 + 1 t12 , t12 1 t11 , t13 + 2 t14 , t14 2 t13 , ]w.

The steady-state contribution of w to x2 is given by q2 = Q2 w,

i.e., l T = Q2 T 1 , where
Q2 is the second row of Q which satises
QS = (Ac + kC)Q + [0, 1]T [t11 + 1 t12 , t12 1 t11 , t13 + 2 t14 , t14 2 t13 ].
The system (10.48) is of relative degree 2, and the control input and parameter
estimator can be designed by following the steps presented earlier.
In the simulation study, we chose the pair

0
1
0
0
0
0

0
1
0
, G = 0 ,
F =
0

0
0
1
0
1 4 6 4
1/10
which is controllable with the eigenvalues of F at {1, 1, 1, 1} and PG < 1.
We set c1 = c2 = 10, with the condition (10.32) being satised. Other parameters
in the control design were set as 1 = 2 = 1,  = 1000I , L1 = 3 and L2 = 2. The
disturbance was set as
w1 = 4 sin 1 t + 4 sin 2 t
with 1 = 1 and

2,
250 > t 0,
2 =
1.5, t 250.
A set of simulation results are presented here. Figure 10.1 shows the system output
together with the control input. Figure 10.2 shows the estimates of . The ideal values
of are [30, 40, 10, 40]T and [12.5, 40, 27.5, 40]T for 1 = 1, 2 = 2 and 1 = 1,

Disturbance rejection and output regulation

0.6

0.4
0.2
0
0.2

50

100

150

200

250

300

350

400

450

500

300

350

400

450

500

450

500

Time (s)
30
20

10
0
10
20
0

50

100

150

200

250
Time (s)

Figure 10.1 System output and control input

80

60

Estimated parameters

40

20

20

40

60

80

50

100

150

200

250
Time (s)

300

350

400

Figure 10.2 Estimated parameters 1 (dashed), 2 (dotted), 3 (dashdot) and

4 (solid)

185

186

Nonlinear and adaptive control systems

2 = 1.5 respectively. Under both sets of the frequencies, converged to the ideal
values. It can be seen through this example that the disturbance of two unknown
frequencies has been rejected completely.


10.2 Adaptive output regulation

In the previous section, we presented a control design for asymptotic rejection of
harmonic disturbances in nonlinear systems when the disturbance frequencies are
unknown. When the measurement, or the output, does not explicitly contain the
disturbances, the control design is referred to as asymptotic rejection problem for
the boundedness of the signals and the asymptotic convergence to zero. When the
measurement contains the disturbance or exogenous signals, the control design to
ensure the measurement converge to zero asymptotically is often referred to as the
output regulation problem. When the measurement contains the exogenous signal, the
output is normally different from the measurement, and in such a case, the output can
be viewed as to track an exogenous signal. In this section, we will present a control
design for output regulation when the measurement is different from the controlled
output.
We consider a SISO nonlinear system which can be transformed into the output
feedback form
x = Ac x + (y, w, a) + bu
y = Cx

(10.49)

e = y q(w),
with

0 1 0 ...
0 0 1 ...

Ac = ... ... ... . . .

0 0 0 ...
0 0 0 ...

0
0

.. ,
.

1
0

0
..
.

0

b=
b ,

.
..
bn

T
1
0

C=. ,
..
0

where x Rn is the state vector; u R is the control; y R is the output; e is

the measurement output; a Rq and b Rn are vectors of unknown parameters,
with b being a Hurwitz vector with b  = 0, which implies the relative degree of the
system is ; : R Rm Rq Rn is a smooth vector eld with each element being
polynomials with a known upper order of its variables and satisfying (0, w, a) = 0;
q is an unknown polynomial of w; and w Rm are disturbances, and they are generated
from an unknown exosystem
w = S( )w

Disturbance rejection and output regulation

187

with unknown Rs , of which, S Rmm is a constant matrix with distinct

eigenvalues of zero real parts.
The control design problem considered in this section is to design a control input
using the measurement feedback to ensure that the regulated measurement converges
to zero asymptotically while keeping all other variables bounded.
The system (10.1) has unknown parameters in the system and in the exosystem,
and therefore adaptive control techniques are used to tackle the uncertainties. For this
reason, the problem under consideration is referred to as adaptive output regulation.
Remark 10.9. Different from (10.1), the system (10.49) has a measurement e that is
perturbed by an unknown polynomial of the unknown disturbance. This is the reason
why the problem to be solved is an output regulation problem, rather than a disturbance
rejection problem.

Remark 10.10. The nonlinear functions in (10.49) are restricted to be polynomials
of its variables, to guarantee the existence of invariant manifold for the solution
of output regulation problem. Comparing with (10.1), the nonlinear term (y, w, a)
has a more complicated structure, containing the output, disturbance, and unknown
parameters all together. Assuming the nonlinear functions to be polynomials also
allows that the nonlinear terms in the control design can be bounded by polynomials
of the measurement.

Remark 10.11. In the system, all the parameters are assumed unknown, including
the sign of the high-frequency gain, b , and the parameters of the exosystem. A
Nussbaum gain is used to deal with the unknown sign of the high-frequency gain.
The adaptive control techniques presented in this control design can be easily applied
to other adaptive control schemes introduced in this book. The nonlinear functions
(y, w, a) and q(w) are only assumed to have known upper orders. This class of
nonlinear systems perhaps remains as the largest class of uncertain nonlinear systems
of which global output regulation problem can be solved with unknown disturbance
frequencies.

When we set the unknown disturbance w and unknown parameter a to zero,
the system (10.49) is in exactly the same format as (9.32), to which backstepping
with ltered transformation has been applied. With unknown parameters, adaptive
backstepping with ltered transformation can be applied to solve the problem here.
One reason for us to use backstepping with ltered transformation is due to the
uncertainty in the nonlinear function (y, w, a), which prevents the application of
adaptive observer backstepping. In the following design, we only consider the case
for relative degree greater than 1.
For the system (10.49) with relative degree > 1, we introduce the same ltered
transformation as in Section 9.3 with the lter
1 = 1 1 + 2
...
(10.50)
1 = 1 1 + u,

188

Nonlinear and adaptive control systems

where i > 0, for i = 1, . . . , 1, are the design parameters, and the ltered
transformation
z = x [d 1 , . . . , d 1 ] ,

(10.51)

where = [1 , . . . , 1 ]T and d i Rn for i = 1, . . . , 1, and they are generated

recursively by d 1 = b and d i = (Ac + i+1 I )d i+1 for i = 2, . . . , 1. The system
(10.49) is then transformed to
z = Ac z + (y, w, a) + d1

(10.52)

y = C z ,
where d = (Ac + 1 I )d 1 . It has been shown in (9.36) that d1 = b and
n


di s

ni

i=1

1


(s + i )

i=1

n


bi sni .

(10.53)

With 1 as the input, the system (10.52) is with relative degree 1 and minimum phase.
We introduce another state transform to extract the internal dynamics of (10.52) with
z Rn1 given by
z = z2:n

d2:n
y,
d1

(10.54)

where ()2:n refers to the vector or matrix formed by the 2nd row to the nth row. With
the coordinates (z, y), (10.52) is rewritten as
z = Dz + (y, w, )
y = z1 + y (y, w, ) + b 1 ,

(10.55)

where the unknown parameter vector = [aT , bT ]T , and D is the left companion
matrix of d given by

d2 /d1 1 . . . 0

d3 /d1 0 . . . 0

..
..
..
.. ,
(10.56)
D=

.
.
.
.

dn1 /d1 0 . . . 1
dn /d1 0 . . . 0
and
d2:n
d2:n
y + 2:n (y, w, a)
1 (y, w, a),
d1
d1
d2
d2:n
y (y, w, ) =
y+
1 (y, w, a).
d1
d1
(y, w, ) = D

Notice that D is Hurwitz, from (9.36), and that the dependence of d on b is reected in
the parameter in (y, w, ) and y (y, w, ), and it is easy to check that (0, w, ) = 0
and y (0, w, ) = 0.

Disturbance rejection and output regulation

189

The solution of the output regulation problem depends on the existence of certain
invariant manifold and feedforward input. For this problem, we have the following
result.
Proposition 10.4. Suppose that an invariant manifold (w) Rn1 satises
(w)
S( )w = D (w) + (q(w), w, ).
w

(10.57)

Then there exists an immersion for the feedforward control input

(w, , )
S( )w = ( ) (w, , )
w
(w, , ) =  (w, , ),
where

(w, , ) =

b1

q(w)
S( )w 1 (w) y (q(w), w, ) .
w

Furthermore, this immersion can be re-parameterised as

= (F + Gl T )
= l T ,

(10.58)

where (F, G) is a controllable pair with compatible dimensions, = M and

l = M 1 with M satisfying
M ( )( ) FM ( ) = G.

(10.59)

Proof. With 1 being viewed as the input, is the feedforward term used for output
regulation to tackle the disturbances, and from the second equation of (10.55), we
have

q(w)
(w, , ) = b1
(w)

(q(w),
w,
)
.
S(
)w

1
y

w
From the structure of the exosystem, the disturbances are sinusoidal functions.
Polynomials of sinusoidal functions are still sinusoidal functions, but with some highfrequency terms. Since all the nonlinear functions involved in the system (10.49) are
polynomials of their variables, the immersion in (10.58) always exists. For a controllable pair (F, G), M is an invertible solution of (10.59) if (, ) is observable, which
is guaranteed by the immersion.
2

We now introduce the last transformation based on the invariant manifold with
z = z

(10.60)

190

Nonlinear and adaptive control systems

Finally we have the model for the control design

z = Dz +
e = z1 + y + b ( l T )
1 = 1 1 + 2

(10.61)

...
1 = 1 1 + u,
where
= (y, w, ) (q(w), w, )
and
y = y (y, w, ) y (q(w), w, ).
Since the state in the internal model is unknown, we design the adaptive internal
model
= F + G1 .

(10.62)

If we dene the auxiliary error

= + b1
Ge,

(10.63)

it can be shown that

1

z1 + b1
= F FGb1
e + b G
G y .

(10.64)

If the system (10.49) is of relative degree 1, then 1 in (10.61) is the control

input. For the systems with higher relative degrees, adaptive backstepping will be
used to nd the nal control input u from the desirable value of 1 . Supposing that 1
is desirable value for 1 , we introduce a Nussbaum gain N () such that
1 = N ()1
= e1 ,

(10.65)

where the Nussbaum gain N is a function (e.g. N () = 2 cos ) which satises the
two-sided Nussbaum properties

1
N (s)ds = +,
(10.66)
lim sup

0

1
lim inf
N (s)ds = ,
(10.67)

0
where denotes + and respectively. From (10.61)
and the denition of the Nussbaum gain, we have
e = z1 + (b N 1)1 + 1 + b 1 + b 1 lbT + y ,

Disturbance rejection and output regulation

191

where lb = b l, b is an estimate of b and b = b b , and 1 = 1 1 . Sine the

w, , ) = 0 and
nonlinear functions involved in and y are polynomials with (0,
y (0, w, , ) = 0, w is bounded, and the unknown parameters are constants, it can
be shown that
< rz (|e| + |e|p ),
||
| y | < ry (|e| + |e|p ),
where p is a known positive integer, depending on the polynomials in and y , and
rz and ry are unknown positive real constants. We now design the virtual control 1
as, with c0 > 0,
1 = c0 e k0 (e + e2p1 ) + lbT .

Using (10.63), we have the resultant error dynamics

e = c0 e k0 (e + e2p1 ) + z1 + (b N 1)1 + b 1 + b 1
lbT lbT + l T Ge + y .

(10.68)

The adaptive laws are given by

k0 = e2 + e2p ,
b,0 = 1 e,

(10.69)

l,0 = e,

where b,0 and l,0 denote the rst tuning functions in adaptive backstepping design
for the nal adaptive laws for b and lb . If the relative degree = 1, we set u = 1 .
For > 1, adaptive backstepping can be used to obtain the following results:

2 = b e c1 1 k1

1
e

2
1

1
1
(b 1 lbT )

+
e

1
k0 +
k0

1
l,1 ,
lb

(10.70)

192

Nonlinear and adaptive control systems


2
i1
i = i2 ci1 i1 ki1
i1
e
+

i1
i1
(b 1 lbT )
+

i1
i1
i1
b,i1 +
k0 +
l,1
k0
b
lb

i

i1 j2
j=4

i

i1 j2
j=3

lb

1 j2
j2

for i = 2, . . . , ,

(10.71)

where i = i i for i = 1, . . . , 2, ci and ki , i = 2, . . . , 1, are positive real

design parameters, and b,i and l,i , for i = 1, . . . , 2, are tuning functions. The
adaptive law and tuning functions are given by
i
1 i ,
e
i
= l,i1 +
i ,
e

b,i = b,i1
l,i

for i = 1, . . . , 1,
for i = 1, . . . , 1,

b = b,1 ,

(10.72)

lb = l,1 .

(10.73)

Finally we design the control input as

u = .

(10.74)

For the proposed control design, we have the following result for stability.
Theorem 10.5. For a system (10.49) satisfying the invariant manifold condition
(10.57), the adaptive output regulation problem is globally solved by the feedback
control system consisting the -lters (10.50), the adaptive internal model (10.62),
Nussbaum gain parameter (10.65), the parameter adaptive laws (10.69), (10.72),
(10.73) and the feedback control (10.74), which ensures the convergence to zero of
the regulated measurement, and the boundedness of all the variables in the closed-loop
system.
Proof. Dene a Lyapunov function candidate
V = 1 T P + 2 z T Pz z


1
1 2  2
2
2
T

+
e +
i + (k0 k0 ) + b + lb lb ,
2
i=1

Disturbance rejection and output regulation

193

where 1 and 2 are two positive reals and Pz and P are positive denite matrices
satisfying
Pz D + DT Pz = I ,
P F + F T P = I .
With the design of i , for i = 1, . . . , , the dynamics of i can be easily evaluated.
From the dynamics of z in (10.61) and the dynamics of in (10.64), virtual controls
and adaptive laws designed earlier, we have the derivative of V as

V = 1 ( T 2 T P Fb1
T P b1
z1 + 2 T P b1
Ge + 2
G
G y )
+ (k0 k0 )(e2 + e2p )
+ 2 (z T z + 2z T Pz )
c0 e2 k0 (e2 + e2p ) + (b N 1)e1
+ ez1 + e y elbT + l T Ge2


2
1

i

A
i
2
ci i ki
+
i2 i
z1
e
e
i=1

i
i T
i T
y + i
l i
l Ge .
i
e
e b
e
The stability analysis can be proceeded by using the inequalities 2xy < rx2 + y2 /r
or xy < rx2 + y2 /(4r) for x > 0, y > 0 and r being any positive real, to tackle the
cross-terms between the variables z , ,
e, i , for i = 1, . . . , 1. It can be shown
that there exist sufciently big positive real 1 , and then sufciently big positive real
2 , and nally the sufcient big k0 such that the following result holds:
1


1
1
ci i2 .
V (b N () 1) 1 T 2 z T z c0 e2
3
4
i=1

(10.75)

The boundedness of V can be established based on the Nussbaum gain properties

(10.66) and (10.67) via an argument of contradiction. In fact, integrating (10.75) gives

 t
1

1
1 T
T
2
2
1 + 2 z z + c0 e +
ci i dt
V (t) +
3
4
0
i=1

b

(t)

N (s)ds (t) + V (0).

(10.76)

If (t), t R+ , is not bounded from above or below, then from (10.66) and (10.67)
it can be shown that the right-hand side of (10.76) will be negative at some instances
of time, which is a contradiction, since the left-hand side of (10.76) is non-negative.

194

Nonlinear and adaptive control systems

Therefore, is bounded, which implies the boundedness of V . The boundedness of

V further implies ,
z , e, i L2 L for i = 1, . . . , 1, and the boundedness

of k0 , b and lb . Since the disturbance w is bounded, e, z , L implies the

boundedness of y, z and ,
which further implies the boundedness of 1 and then
the boundedness of 1 . The boundedness of 1 and 1 , together with the boundedness
of e, ,
k0 , b and lb , implies the boundedness of 2 , and then the boundedness of
2 follows the boundedness of 2 . Applying the above reasoning recursively, we can
establish the boundedness of i for i > 2 to i = 1. We then conclude that all the
variables are bounded.
z , e and
The boundedness of all the variables implies the boundedness of ,
, which further implies, together with ,

z , e, i L2 L and Barbalats
i
lemma, limt = 0, limt z = 0, limt e(t) = 0 and limt i = 0 for
i = 1, . . . , 1.
2

10.3 Output regulation with nonlinear exosystems

In the previous sections, the disturbances are sinusoidal functions which are generated
from linear exosystems with the restriction that the eigenvalues of the exosystem
matrix are distinct and with zero real parts. Sinusoidal disturbances are important as
disturbances in practical systems can often be approximated by a nite number of
sinusoidal functions. However, there are situations where disturbances are generated
from nonlinear exosystems, such as nonlinear vibration, etc. Such disturbances can
still be approximated by a nite number of sinusoidal functions, possibly with a big
number of sinusoidal functions for a good accuracy. The more sinusoidal functions
involved in the approximation of a periodic signal, the higher order will be the
corresponding system matrix for the linear exosystem. If an internal model can
be designed directly based on the nonlinear exosystem, it is possible to achieve
asymptotic rejection of the disturbances, which cannot be achieved by approximation
using sinusoidal functions, and the order of the internal model can also remain much
lower. Of course, it is expected to be a very difcult problem to directly design
an internal model to deal with nonlinear exosystems, even though there exists an
invariant manifold for output regulation. In this section, we will show nonlinear
internal model design for a class of nonlinear exosystem to achieve output regulation.
For the internal model design, we exploit a technique for nonlinear observer design
based on conditions similar to circle criteria. The dynamic model considered for
output regulation is still the class of output feedback form.
We consider a SISO nonlinear system
x = Ac x + (y)a + E(w) + bu
y = Cx
e = y q(w),

(10.77)

Disturbance rejection and output regulation

with

0 1 0 ...
0 0 1 ...

Ac = ... ... ... . . .

0 0 0 ...
0 0 0 ...

0
0

.. ,
.

1
0

195

0
..
.

0

b=
b ,

.
..

T
1
0

C=. ,
..
0

bn

where x R is the state vector; u R is the control; y R is the output; e is the

measurement output; a Rq and b Rn are vectors of unknown parameters, with b
being a Hurwitz vector with b  = 0, which implies the relative degree of the system
is , and with known sign of b , E : Rm Rn , : R Rnq with (0) = 0 and
|(y1 ) (y2 )| 1 (|y1 |)1 (|y1 y2 |) and 1 () K and 1 () is non-decreasing
and the function 1 () is a known smooth function; and w Rm are disturbances, and
they are generated from a nonlinear exosystem
n

w = s(w)

(10.78)

of which the ows are bounded and converge to periodic solutions.

Remark 10.12. The assumption about the function is satised for many kinds of
functions, for example polynomial functions.

Remark 10.13. The nonlinear exosystem (10.78) includes nonlinear systems that
have limit cycles.

Remark 10.14. The system (10.78) is very similar to the system (10.49) considered
in the previous section. The main difference is that the exosystem is nonlinear.

The system (10.78) has the same structure of Ac , bu and C as in (10.49), and
therefore the same ltered transformation as in the previous section can be used here.
We can use the same ltered transformation, and the transformation for extracting
the zero dynamics of the system as in the previous section. Using the transformations
(10.51) and (10.54), the system (10.78) is put in the coordinate (z, y) as

di+1
di+2
di+1 d2
di+1
zi =
y + (i+1 (y)
z1 + zi+1 +

1 (y))a
d1
d1
d1
d12
di+1
E1 (w), i = 1, . . . , n 2,
(10.79)
d1

dn
d n d2
dn
dn
= z1 2 y + n (y) 1 (y) a + En (w) E1 (w),
d1
d1
d1
d1
+ Ei+1 (w)

zn1

y = z1 +

d2
y + 1 (y)a + E1 (w) + b 1
d1

where di are dened in (10.53).

196

Nonlinear and adaptive control systems

It is necessary to have the existence of certain invariant manifolds for a solution

to the output regulation problem. When the exosystem is nonlinear, it is even more
challenging to have the necessary conditions for invariant manifold to exist.
Proposition 10.6. Suppose that there exist  (w) Rn and (w) with 1 (w) = q(w)
for each a, b such that

s(w) = Ac  + (q(w))a + E(w) + b(w).
(10.80)
w
Then there exists (w) Rn1 along the trajectories of exosystem satisfying

i (w)
di+2
di+1 d2
di+1
1 (w) + i+1 (w) + q(w)

s(w) =
w
d1
d1
d12
+ Ei+1 (w) E1 (w)

di+1
+ (i+1 (q(w))
d1

di+1
1 (q(w)))a, i = 1, . . . , n 2,
d1
n1 (w)
dn
dn d2
s(w) = 1 (w) 2 q(w) + (n (q(w))
w
d1
d1

dn
dn
1 (q(w)))a + En (w) E1 (w).
d1
d1

Proof. Since the last equation of input lter (10.50) used for the ltered transformation
is an asymptotically stable linear system, there is a static response for every external
input u(w), i.e., there exists a function 1 (w) such that
1 (w)
s(w) = 1 1 (w) + (w).
w
Recursively, if there exists i (w) such that
i (w)
s(w) = i i (w) + i+1 (w),
w
then there exists i1 (w) such that
i1 (w)
s(w) = i1 i1 (w) + i (w).
w
Dene


(w)
= Da ( (w) [d 1 , . . . , d 1 ] ),
q(w)
where = [1 , . . . , 1 ]T

d2 /d1 1

..
..

.
.
Da =
dn /d1 0
1
0

and

... 0
.
..
. ..
.
... 1
... 0

Disturbance rejection and output regulation

197

It can be seen that (w) satises the dynamics of z along the trajectories of (10.78)
as shown in (10.80), and hence the proposition is proved.
2
Based on the above lemma, we have
d2
q(w)
s(w) = 1 (w) + q(w) + 1 (q(w))a + E1 (w) + b (w),
w
d1
where (w) = 1 (w). With 1 being viewed as the input, (w) is the feedforward term
used for output regulation to tackle the disturbances, and it is given by

q(w)
d2
(w)

q(w)

(q(w))a

E
(w)
.
s(w)

= b1
1
1
1

w
d1
We now introduce the last transformation based on the invariant manifold with
z = z (w(t)).
Finally we have the model for the control design

zi = di+1 z1 + zi+1 + di+2 di+1 d2 e
d1
d1
d12
+ (i+1 (y) i+1 (q(w)))a
di+1
(1 (y) 1 (q(w)))a, i = 1, . . . , n 2
d1
dn
d n d2
= z1 2 e + (n (y) n (q(w)))a
d1
d1

zn1

dn
(1 (y) 1 (q(w)))a
d1
d2
e = z1 + e + (1 (y) 1 (q(w)))a + b (1 (w)),
d1

i.e., the system can be represented as

z = Dz + e + (y, w, d)a
e = z1 +

d2
e + (1 (y) 1 (q(w)))a + b (1 (w)),
d1

where D is a companion matrix of d shown in (10.56), and

T
d3
d2
dn
dn1 d2 dn d2
=
22 , . . . ,

,
d1
d1
d1
d12
d12

d2
2 (y) 2 (q(w)) (1 (y) 1 (q(w)))

d1

.
.

..
(y, w, d) =

dn
n (y) n (q(w)) (1 (y) 1 (q(w)))
d1

(10.81)

198

Nonlinear and adaptive control systems

Lemma 10.7. There exists a known function () which is non-decreasing and an

unknown constant , which is dependent on the initial state w0 of exosystem, such that
|(y, w, d)| |e| (|e|),
|1 (y) 1 (q(w))| |e| (|e|).
Proof. From the assumption of we can see that
|(y) (q(w))| 1 (|q(w)|)1 (|e|).
Since the trajectories of exosystem are bounded and 1 () is smooth there exist smooth
nondecreasing known function () and a nondecreasing known function 2 (|w0 |),
such that
1 (|e|) |e| (|e|),
1 (|q(w)|) 2 (|w0 |).
From previous discussion the result of the lemma is obtained.

Let
Vz = z T Pd z ,
where
Pd D + DT Pd = I .
Then using 2ab ca2 + c1 b2 and 2 (|e|) 2 (1 + e2 ), there exist unknown positive
real constants 1 and 2 such that
V z = z T z + 2z T Pd (e + (y, w, d)a)
3
z T z + 1 e2 + 2 e2 2 (1 + e2 ),
4
noting that
1
2z T Pd e z T z + 8eT T Pd2 e
8
1
z T z + 1 e2 ,
8
and
1
2z T Pd (y, w, d)a z T z + 8aT T Pd2 a
8
1
z T z + 12 ||2
8
1
z T z + 12 2 |e|2 2 (|e|)
8
1 T
z z + 2 e2 2 (1 + e2 ),
8
where 12 is an unknown positive real constant.

(10.82)

Disturbance rejection and output regulation

199

Now let us consider the internal model design. We need an internal model to
produce a feedforward input that converges to the ideal feedforward control term
(w), which can be viewed as the output of the exosystem as
w = s(w)
= (w).
Suppose that there exists an immersion of the exosystem
= F + G (J )
= H ,

(10.83)

where Rr , H = [1, 0, . . . , 0], (H , F) is observable

(v1 v2 )T ( (v1 ) (v2 )) 0,
and G and J are some appropriate dimensional matrices. We then design an internal
model as
= (F KH )( b1
b1
Ke) + G (J (
Ke)) + K1 ,

(10.84)

where K Rr is chosen such that F0 = F KH is Hurwitz and there exist a positive

denite matrix Pf and a semi-positive denite matrix Q satisfying

Pf F0 + F0T Pf = Q

Pf G + J T = 0
(10.85)

T Q 0 |1 |2 , 0 > 0, Rr

span(P K) span(Q).
F
Remark 10.15. It reminds us a challenging problem to design internal models for
output regulation with nonlinear exosystems. It is not clear at the moment what
general conditions are needed to guarantee the existence of an internal model for
output regulation of nonlinear systems with nonlinear internal models. Here we use
the condition of the existence of an immersion (10.83) for the internal model design.
Also note that even when the exosystem is linear, an internal model can be nonlinear
for a nonlinear dynamic system.

Remark 10.16. Note the condition specied in (10.85) is weaker than the condition
that there exist Pf > 0 and Q > 0 satisfying

PF F0 + F0T PF = Q
(10.86)
PF G + J T = 0,
which can be checked by LMI. This will be seen in the example later in this section. In
particular, if G and J T are two column vectors, (F0 , G) controllable, (J , F0 ) observable
and Re[J (jI F0 )1 G] > 0, R, then there exists a solution of (10.86) from
KalmanYacubovich lemma.


200

Nonlinear and adaptive control systems

If we dene the auxiliary error
= + b1
Ke,

it can be shown that

= F0 + G( (J ) (J ( b1
Ke)))

d2
1
+ b K z1 + e + (1 (y) 1 (q(w)))a .
d1
Let
F .

V = P
Then following the spirit of (10.82), there exist unknown positive real constants 1
and 2 such that

d2
T
T
1

V = Q + 2 PF b K z1 + e
d1
+ 2 T PF b1
K(1 (y) 1 (q(w)))a
+ 2 T PF G( (J ) (J ( b1
Ke))
3
12
0 | 1 |2 + b2
z 2 + 1 e2 + 2 e2 2 (1 + e2 ).
4
0 1

(10.87)

Let us proceed with the control design. From (10.81) and

= 1 = 1 + 1 b1
K1 e,
we have
e = z1 +

d2
e + (1 (y) 1 (q(w)))a + 1 + b (1 1 1 + b1
K1 e),
d1

where 1 = 1 1 and

1 = b1
1 .

(10.88)

For the virtual control 1 , we design 1 as, with c0 > 0,

1 = c0 e + b 1 K1 e le(1 + 2 (1 + e2 )),

(10.89)

where l is an adaptive coefcient. Then we have the resultant error dynamics

e = z1 c0 e +
Then for
Ve =

1 2
e,
2

d2
e le(1 + 2 (1 + e2 )) + (1 (y) 1 (q(w)))a + b (1 1 ).
d1

Disturbance rejection and output regulation

there exist unknown positive real constants
unknown positive constant such that

and

2,

201

and a sufciently large

d2
V e = c0 e2 + ez1 + e2 + eb (1 1 )
d1
+ e(1 (y) 1 (q(w)))a le2 (1 + 2 (1 + e2 ))
1
1
c0 e2 + z12 + 0 12 + 1 e2 +
8
4
2
2

le (1 + (1 + e2 )) + b e1 .

2e

(1 + e2 )
(10.90)

Let
1
V0 = Vz + V + Ve + 1 (l l)2 ,
2
where 96
b2 is chosen and l =
0
unknown constant. Let

+ 1 + 2 + (1 + 2 ) is an

l = e2 (1 + 2 (1 + e2 )).
Then, it can be obtained that
1
1
V 0 z T z 0 | 1 |2 c0 e2 + b e1 .
2
2
If the system (10.78) has relative degree 1, the virtual control 1 shown in (10.88)
together with 1 in (10.88) gives the input, i.e., u = 1 . For the system with higher
relative degrees, the control design can be proceeded with backstepping using (10.50)
in the same way as the adaptive backstepping with ltered transformation shown in the
previous section for adaptive output regulation with linear exosystems. We summarise
the stability result in the following theorem.
Theorem 10.8. For the system (10.78) with the nonlinear exosystem (10.78), if there
exists an invariant manifold (10.80) and an immersion (10.83), then there exists
K Rr such that F0 = F KH is Hurwitz and there exist a positive denite matrix PF
and a semi-positive denite matrix Q satisfying (10.85), and there exists a controller
to solve the output regulation in the sense the regulated measurement converges to
zero asymptotically while other variables remain bounded.
We use an example to illustrate the proposed control design, concentrating on
the design of nonlinear internal model.
Example 10.2. Consider a rst-order system
y = 2y + sin y y3 sin w1 + w2 + u
e = y w1 ,

202

Nonlinear and adaptive control systems

where is an unknown parameter, and the disturbance w is generated by

w 1 = w1 + w2 w13
w 2 = w1 w23 .
It is easy to see that V (w) = 12 w12 + 12 w22 satises
dV
= w12 w14 w24 0,
dt

when |w1 | 1,

and that
q(w) = w1 ,
= w1 ,
(w) = w1 .
From the exosystem and the desired feedforward input , it can be seen that the
condition specied in (10.85) is satised with = w and





1 1
1 0

F = 1 0 , G =
0 1



1
0

.
1 (s) = 2 (s) = s , J =
0 1
Let K = [2, 0]T . Then with

1 1
F0 =
, PF = I ,
1 0

Q = diag(2, 0),

the internal model is designed as the following:

1 = ( 1 2e) + 2 ( 1 2e)3 + 2u
2 = ( 1 2e) 23 .
The control input and the adaptive law are given by
u = ce + 1 le(1 + (e2 + 1)2 ),

l = e2 (1 + (e2 + 1)2 ).
For simulation study, we set c = 1, = 1, = 1, and the initial states are
y(0) = 1, w1 (0) = 2 and w2 (0) = 2. The initial state of dynamic controller is zero.
The system output and input are shown in Figure 10.3, while the feedforward term
and its estimation are shown in Figure 10.4 and the portrait of the exosystem is
shown in Figure 10.5. As shown in the gures, the internal model successfully reproduces the feedforward control needed after a transient period, and the system output
measurement is regulated to zero, as required.


Disturbance rejection and output regulation

203

0.4
Tracking error e

0.2
0
0.2
0.4
0.6
0.8
1

10

15

10

15

Time t (s)
5
4
2
1
0
1
2

5
Time t (s)

Figure 10.3 The systems output e and input u

0.5

0
h1 and h1

Control u

0.5

1.5

10

15

Time t (s)

Figure 10.4 The systemss feedforward control 1 and its estimation 1

204

Nonlinear and adaptive control systems

2

1.5

w2

0.5

0.5

1
1

0.5

0.5
w1

1.5

Figure 10.5 The portrait of exosystems

10.4 Asymptotic rejection of general periodic disturbances

We have presented design methods for asymptotic rejection and output regulation
of disturbances generated from linear exosystems, i.e., sinusoidal functions and disturbances from a specic class of nonlinear exosystems, which generally produce
non-harmonic but still periodic disturbances. For disturbances from linear exosystem, internal models can normally be designed under some mild assumptions, while
for nonlinear exosystems, the conditions are more restrictive for designing an internal model for asymptotic rejection and output regulation. The difculty lies in the
guaranteed existence of the invariant manifold, and then the nonlinear internal model
design, which is often more involved even than nonlinear observers for nonlinear
systems.
We consider some more general periodic disturbances than harmonic disturbances in this section. These general periodic disturbances can be modelled as outputs
of nonlinear systems, and in particular, as the outputs of linear dynamic systems with
nonlinear output functions. For the systems with Lipschitz nonlinearities, nonlinear
observers can be designed as shown in Section 8.4 and other results in literature. Of
course, the problem addressed in this chapter cannot be directly solved by nonlinear
observer design, not even the state estimation of the disturbances system, as the disturbance is not measured. However, there is an intrinsic relationship between observer
design and internal model design, as evidenced in the previous results of disturbance
rejection and output regulation in the earlier sections in this chapter.

Disturbance rejection and output regulation

205

With the formulation of general periodic disturbances as the nonlinear outputs

of a linear dynamic system, the information of the phase and amplitude of a general
periodic disturbance is then embedded in the state variables, and the information of
the wave prole in the nonlinear output function. The nonlinear output functions are
assumed to be Lipschitz. By formulating the general periodic disturbances in this way,
we are able to explore nonlinear observer design of nonlinear systems with output
Lipschitz nonlinearities shown in Section 8.4. We will show that general periodic
disturbances can be modelled as nonlinear outputs of a second-order linear system
with a pair of pure imaginary poles which depend on the frequencies. For this specic
system with Lipschitz nonlinear output, a rened condition on the Lipschitz constant
will be given by applying the proposed method in this section, and observer gain will
be explicitly expressed in terms of the Lipschitz constant and the period or frequency
of the disturbance.
An internal model design is then introduced based on the proposed Lipschitz
output observer for a class of nonlinear systems. Conditions are identied for the
nonlinear system, and control design is carried out using the proposed internal model.
Two examples are included to demonstrate the proposed internal model and control
design procedures. These examples also demonstrate that some other problems can
be converted to the problem addressed in this section.
We consider a nonlinear system
y = a(z) + 0 (y) + (y, v) + b(u (v))
z = f (z, v, y),

(10.91)

where y R is the output; a and 0 : Rn R are continuous functions; v Rm

denotes general periodic disturbances; : Rm R is a continuous function; :
R Rm R is a continuous function and satises the condition that |(y, v)|2

y(y)
with being a continuous function; b is a known constant; u R is the input;
n
z R is the internal state variable; and f : Rn Rm R Rn is a continuous
function.
Remark 10.17. For the convenience of presentation, we only consider the system
with relative degree 1 as in (10.91). The systems with higher relative degrees
can be dealt with similarly by invoking backstepping. The second equation in
(10.91) describes the internal dynamics of the system states, and if we set v = 0
and y = 0, z = f (z, 0, 0) denotes the zero dynamics of this system.

Remark 10.18. The system in (10.91) species a kind of standard form for
asymptotic rejection of general periodic disturbances. For example, consider
x = Ax + (y, v) + bu
y = cT x,

206

Nonlinear and adaptive control systems

with b, c Rn and

a1

a2
A=
.
..
an

... 0

..
. 0
0

.. . .
.. ,
. .
.
... ... 0
1

1
b1
0
b2


b = . , c = . ,
..
..

bn

where x Rn is the state vector; y and u R are the output and input respectively
of the system; v Rm denotes general periodic disturbances; and : R Rm Rn
is a nonlinear smooth vector eld in Rn with (0, 0) = 0. This system is similar to
the system (10.49) with q(w) = 0. For this class of nonlinear systems, the asymptotic
disturbance rejection depends on the existence of state transform to put the systems
in the form shown in (10.91), and it has been shown in Section 10.2 that such a
transformation exists under some mild assumptions.

The wave prole information of a general periodic disturbance is used to construct
a nonlinear function as the output function for a linear exosystem, for the generation
of the desired feedforward input. By doing that, an observer with nonlinear output
function can be designed, viewing the feedforward input as the output, and an internal
model can then be designed based on the nonlinear observer. The problem to solve in
this section is to design a control scheme, using a nonlinear observer-based internal
model, to asymptotically reject a class of general periodic disturbances for the system
in (10.91).
We start with modelling general periodic disturbances as the outputs of a linear
oscillator with nonlinear output functions, and then propose nonlinear observer design
for such a system, for the preparation of internal model design.
Many periodic functions with period T can be modelled as outputs of a secondorder system


0
w = Aw, with A =
0
(v) = h(w),
where = 2
. Here, the desired feedforward input (v) is modelled as the nonlinear
T
output h(w) of the second-order system. With


cos t sin t
eAt =
,
sin t cos t
the linear part of the output h(w), Hw, is always in the form of a sin (t + ) where
a and denote the amplitude and phase respectively. Hence, we can set H = [1 0]
without loss of generality, as the amplitude and the phase can be decided by the initial
value with
w(0) = [a sin () a cos ()]T .

Disturbance rejection and output regulation

207

Based on the above discussion, the dynamic model for general periodic disturbance
is described by
w 1 = w2
w 2 = w1

(10.92)

= w1 + h1 (w1 , w2 ),
where h1 (w1 , w2 ) is a Lipschitz nonlinear function with Lipschitz constant .
Remark 10.19. General periodic disturbances can be modelled as af (t + ) with a
and for the amplitude and phase of a disturbance, and the wave prole is specied by
a periodic function f . In the model shown in (10.93), the amplitude and phase of the
disturbance are determined by the system state variables w1 and w2 , and the prole is
determined by the nonlinear output function. In some results shown in literature, the
phase and amplitude are obtained by delay and half-period integral operations. Here,
we use nonlinear observers for the estimation of phases and amplitudes of general
periodic disturbances.

For the model shown in (10.93), the dynamics are linear, but the output function
is nonlinear. Many results in literature on observer design for nonlinear Lipschitz
systems are for the system with nonlinearities in the system dynamics while the output
functions are linear. Here we need the results for observer design with nonlinear output
functions. Similar techniques to the observer design of nonlinearities in dynamics can
be applied to the case when the output functions are nonlinear.
We have shown the observer design for a linear dynamic system with a nonlinear
Lipschitz output function in Section 8.4 with the observer format in (8.38) and gain
in Theorem 8.11. Now we can apply this result to observer design for the model of
general periodic disturbances. Forthe model shown in (10.93), the observer shown
0
in (8.38) can be applied with A =
and H = [1 0]. We have the following
0
lemma for the stability of this observer.
Lemma 10.9. An observer in the form of (8.38) can be designed to provide an exponentially convergent state estimate for the general periodic disturbance model (10.93)
if the Lipschitz constant for h1 satises < 12 .
Proof. Our proof is constructive. Let

1
p

4 2
,
P=

1
2
p
4
where p >

1
.
4 2

It is easy to see that P is positive denite. A direct evaluation gives

PA + AT P

HTH
1
= 2 I.
2

208

Nonlinear and adaptive control systems

Therefore, the second condition in (8.40) is satised. Following the rst condition
specied in (8.39), we set

4(4p 2 )
(4p 2 )2 1
.
L=
(10.93)

4
(4p 2 )2 1
The rest part of the proof can be completed by invoking Theorem 8.11.

Hence from the above lemma, we design the observer for the general disturbance
model as
x = Ax + L(y h(x)),
(10.94)


0
where A =
and H = [1 0] with the observer gain L as shown in (10.93).
0
Before introducing the control design, we need to examine the stability issues of
the zsubsystem, and hence introduce a number of functions that are needed later
for the control design and stability analysis of the entire system.
Lemma 10.10. Assuming that the subsystem
z = f (z, v, y)
is ISS with state z and input y, characterised by an ISS pair (, ), and furthermore,
(s) = O(a2 (s)) as s 0, there exist a differentiable positive denite function V (z)
and a K function satisfying (z) a2 (z) such that
V (z) (z) + (y).

(10.95)

where is a continuous function.

Proof. From Corollary 5.10, there exists a Lyapunov function Vz (z) that satises that
1 (z) Vz (z) 2 (z)
V z (z) (z) + (|y|),

(10.96)

where , 1 and 2 are class K functions, and is a class K function. Let

be a K function such that (z) a2 (z) and (s) = O(a2 (s)) as s 0. Since
(s) = O(a2 (s)) = O((s)) as s 0, there exists a smooth nondecreasing (SN )
function q such that, r R+
1
q (r)(r) (r).
2
Let us dene two functions
q(r) := q (11 (r)),
 r
(r) :=
q(t)dt.
0

Disturbance rejection and output regulation

209

Dene
V (z) := (V (z)),
and it can be obtained that
V (z) q(V (z))(z) + q(V (z)) (|y|)
1
q(V (z))(z) + q( (|y|)) (|y|)
2
1
q(1 (z))(z) + q( (|y|)) (|y|)
2
1
= q (z)(z) + q((|y|)) (|y|),
2
where is dened as
(r) := 2 ( 1 (2 (r))
for r R+ . Let us dene a smooth function such that
(r) q( (|r|)) (|r|)
for r R and (0) = 0, and then we have established (10.95).

Based on observer design presented in (10.94), we design the following internal

model:
= A + b1 L0 (y) + Lu b1 ALy Lh( b1 Ly),

(10.97)

where L is designed as in (10.93).

The control input is then designed as

(y)

u = b1 0 (y) + k0 y + k1 y + k2
+ h( b1 Ly), (10.98)
+ k3 (y)
y
where k0 is a positive real constant, and
3
k1 = 1 b2 ( + H )2 + ,
4
k2 = 4 1 ||b1 PL2 + 2,
1
k3 = 4 1 ||b1 PL2 + .
2
For the stability of the closed-loop system, we have the following theorem.
Theorem 10.11. For a system in the form shown in (10.91), if

feedforward term (v) can be modelled as the output of a system in the format
shown in (10.93) and the Lipschitz constant of the output nonlinear function
satises < 12

210

Nonlinear and adaptive control systems

the subsystem z = f (z, v, y) is ISS with state z and input y, characterized by ISS
pair (, ), and furthermore, (s) = O(a2 (s)) as s 0

the output feedback control design with the internal model (10.97) and the control
input (10.98) ensures the boundedness of all the variables of the closed-loop system
and the asymptotic convergence to zero of the state variables z and y and the estimation
error (w + b1 Ly).
Proof. Let
= w + b1 Ly.
It can be obtained from (10.97) that
= (A LH ) + b1 L(h1 (w) h1 (w )) + b1 La(z) + b1 L(y, v).
Let Vw = T P . It can be obtained that
V w ( )  2 + 2| T b1 PLa(z)| + 2| T b1 PL(y, v)|
1
 2 + 2 1 ||b1 PL2 (a2 (z) + (y, v)|2 )
2

1
1
2

y(y)
  + (k2 2)(z) + k3
2
2

(10.99)

where = 21 2 1.
Based on the control input (10.98), we have
y = k0 y k1 y k2

(y)

k3 (y)
+ a(z) + (y, v) + b(h(w ) h(w)).
y

Let Vy = 12 y2 . It follows from the previous equation that

V y = (k0 + k1 )y2 k2 (y) k3 y(y)

+ ya(z) + y(y, v) + yb(h(w ) h(w))

1
1
2

k0 y k2 (y) k3
y(y)
+ (z) +  2 . (10.100)
2
4
Let us dene a Lyapunov function candidate for the entire closed-loop system as
V = Vy + Vw + k2 V z .
Following the results shown in (10.96), (10.99) and (10.100), we have
1
V k0 y2  2 (z).
4
Therefore, we can conclude that closed-loop system is asymptotically stable with
respect to the state variables y, z and the estimation error .
Several types of disturbance rejection and output regulation problems can be
converted to the form (10.91). In this section, we show two examples. The rst
example deals with rejection of general periodic disturbances, and the second example
demonstrates how the proposed method can be used for output regulation.

Disturbance rejection and output regulation

211

Example 10.3. Consider

x 1 = x2 + 1 (x1 ) + b1 u
x 2 = 2 (x1 ) + (w) + b2 u
w = Aw
y = x1 ,

(10.101)

where y R is the measurement output; i : R R, for i = 1, 2, are continuous

nonlinear functions; : R2 R is a nonlinear function which produces a periodic
disturbance from the exosystem state w; and b1 and b2 are known constants with the
same sign, which ensures that stability of the zero dynamics. The control objective
is to design an output feedback control input to ensure the overall stability of the
entire system, and the asymptotic convergence to zero of the measurement output.
The system shown in (10.101) is not in the form of (10.91) and the disturbance is not
matched. We will show that the problem can be transformed to the problem considered
in the previous section.
Let
z = x2

b2
x1 .
b1

In the coordinates (y, z ), we have

y = z +

b2
y + 1 (y) + b1 u
b1

b2
b2
z = z + 2 (y) 1 (y)
b1
b1

b2
b1

2
y + (w).

Consider
z =

b2
z + (w).
b1

It can be shown that there exists a steady-state solution, and furthermore, we can
express the solution as a nonlinear function of w, denoted by z (w). Let us introduce
another state transformation with z = z z (w). We then have
b2
y + 1 (y) + b1 (u + b1
1 z (w))
b1

2
b2
b2
b2
z = z + 2 (y) 1 (y)
y.
b1
b1
b1

y = z +

(10.102)

212

Nonlinear and adaptive control systems

Comparing (10.102) with (10.91), we have
a(z) = z,
b2
y + 1 (y),
b1
b = b1 ,

(y) =

h(w) = b1
1 z (w),
f (z, v, y) = 2 (y)

b2
1 (y)
b1

b2
b1

2
y.

From a(z) = z, we can set (z) = z2 = z 2 .

It can be shown that the second condition of Theorem 10.11 is satised by
(10.102). Indeed, let V (z) = 12 z 2 , and we have


2 
b
b
b2
2 2
2
y
V z = z + z 2 (y) 1 (y)
b1
b1
b1


2 2
1 b2 2 1 b1
b2
b2

2 (y) 1 (y)
z +
y .
2 b1
2 b2
b1
b1
Let
b1
V z = 2 Vz
b2
and nally we have
V z (|z|) +

b1
b2

b2
2 (y) 1 (y)
b1

b2
b1

2 2
y .

(10.103)

It can be seen that there exists a class K function (|y|) to dominate the second term
on the right-hand side of (10.103), and the z-subsystem is ISS. For the control design,
we can take

2 2

2 
b2
b2
b1
2 (y) 1 (y)
y .
(y) =
b2
b1
b1
The rest part of the control design follows the steps shown earlier.
For the simulation study, we set the periodic disturbance as a square wave. For
convenience, we abuse the notations of (w(t)) and h(w(t)) as (t) and h(t). For
with t in one period, we have

d, 0 t < ,
2
(10.104)
=
T

d,
t < T,
2

Disturbance rejection and output regulation

213

where d is an unknown positive constant, denoting the amplitude. It can be obtained

that

h = d h(t),
where

b
1
T
1 bb2 t
T b2
b2 t

1
1
, 0t< ,
)+ e
tanh

b (1 e
b2
4 b1
2
2
=
h(t)
(10.105)

b
b
b

T
2
2
2
1
T
1 t
T b2

t
( t)

,
t < T.
(1 + e b1 2e b1 2 ) + e b1 tanh
b2
b2
4 b1
2
Eventually we have the matched periodic disturbance h(w) given by


w2
2
2
h(w) = w1 + w2 h arctan
.
w1

Note that w12 + w22 decides the amplitude, which can be determined by the initial
state of w.
In the simulation study, we set T = 1, d = 10, 1 = y3 , 2 = y2 and b1 = b2 = 1.
The simulation results are shown in Figures 10.610.9. It can be seen from Figure 10.6
that the measurement output converges to zero and the control input converges to a
periodic function. In fact, the control input converges to h(w) as shown in Figure 10.7.
0.2
0.15

0.1
0.05
0
0.05

10

15

10

15

Time (s)
5
0

5
10
15
20
0

5
Time (s)

Figure 10.6 The system input and output

214

Nonlinear and adaptive control systems

5
u
h

u and h

10

15

20

10

15

Time (s)

Figure 10.7 Control input and the equivalent input disturbance

8
h
Estimate of h
6

h and estimate

4
0

10
Time (s)

Figure 10.8 The equivalent input disturbance and its estimate

15

Disturbance rejection and output regulation

215

40
w1
w2
h1
h2

30

w and estimates

20

10

10

20

10

15

Time (s)

Figure 10.9 The exosystem states and the internal model states
As for the internal model and state estimation, it is clear from Figure 10.8 that the
estimated equivalent input disturbance converges to h(w), and converges to w.

Example 10.4. In this example, we briey show that an output regulation problem
can also be converted to the form in (10.91). Consider
x 1 = x2 + (ey 1) + u
x 2 = (ey 1) + 2w1 + u
w = Aw
y = x1 w1 ,

(10.106)

where y R is the measurement output and w1 = [1 0]w. In this example, the measured output contains the unknown disturbance, unlike Example 10.3. The control
objective remains the same, to design an output feedback control law to ensure the
overall stability of the system and the convergence to zero of the measured output.
The key step in the control design is to show that the system shown in (10.106) can
be converted to the form as shown in (10.91).
Let
z =

1
[1 ]w,
1 + 2

216

Nonlinear and adaptive control systems

and it is easy to check that z satises

z = z + [1 0]w.
Let z = x2 z x1 . It can be obtained that
y = z + y + ew1 (ey 1) + (u h(w))
z = z y + 2w1 ,
where
h(w) =

2 + 2
[1 ]w (ew1 1).
1 + 2

It can be seen that we have transformed the system to the format as shown in
(10.91) with (y, v) = ew1 (ey 1).
To make H = [1 0], we introduce a state transform for the disturbance model as
2 + 2
=
1 + 2

w.
1

It can be easily checked that = A . The inverse transformation is given as

w=

1
2 + 2


1
.
1

With as the disturbance state, we can write the transformed system as

2

y = z + y + e1/(2+ )[1] (ey 1) + (u h( ))

2
z = z y +
[1 ]
2 + 2
= A ,

(10.107)

where
h( ) = 1 (e1/(2+

2 )[1]

1).

y
y

Note that e y1 is a continuous function, and we can take (y)

= d0 y( e y1 )2 where
d0 is a positive real constant depending on the frequency and the knowledge of an
upper limit of the disturbance amplitude. The control design presented in the previous
section can then be applied to (10.108). Simulation studies were carried out with the
results shown in Figures 10.1010.13.


Disturbance rejection and output regulation

0.2
0.15

0.1
0.05
0
0.05
0

0.5

1.5

2.5
Time (s)

3.5

4.5

0.5

1.5

2.5
Time (s)

3.5

4.5

50
0
50
u

100
150
200
250
300

Figure 10.10 The system input and output

50
u
h
0

u and h

50

100

150

200

250

300
0

0.5

1.5

2.5

3.5

4.5

Time (s)

Figure 10.11 Control input and the equivalent input disturbance

217

Nonlinear and adaptive control systems

20
h
Estimate of h
15
10

h and estimate

5
0
5
10
15
20
25

0.5

1.5

2.5
Time (s)

3.5

4.5

Figure 10.12 The equivalent input disturbance and its estimate

40
m1
m2
h1
h2

30

20
m and estimates

218

10

10

20

30

0.5

1.5

2.5

3.5

4.5

Time (s)

Figure 10.13 The exosystem states and the internal model states

Chapter 11

Control applications

In this chapter, we will address a few issues about control applications. Several
methods of disturbance rejection are presented in Chapter 10, including rejection
of general periodic disturbances. A potential application can be the estimation and
rejection of undesirable harmonics in power systems. Harmonics, often referred to
high-order harmonics in power systems, are caused by nonlinearities in power systems,
and the successful rejection depends on accurate estimation of amplitudes and phase
of harmonics. We will show an iterative estimation method based on a new observer
design method.
There are tremendous nonlinearities in biological systems, and there have been
some signicant applications of nonlinear system analysis and control methods in
system biology. We will show a case that nonlinear observer and control are applied
to circadian rhythms. A Lipschitz observer is used to estimate unknown states, and
backstepping control design is then applied to restore circadian rhythms.
Most of the control systems are implemented in computers or other digital devices
which are in discrete-time in nature. Control implementation using digital devices
inevitably ends with sample-data control. For linear systems, the sampled systems are
still linear, and the stability of the sampled-date system can be resolved in stability
analysis using standard tools of linear systems in discrete-time. However, when a
nonlinear system is sampled, the system description may not have a closed form, and
the structure cannot be preserved. The stability cannot be assumed for a sampled-data
implementation of nonlinear control strategy. We will show that for certain nonlinear
control schemes, the stability can be preserved by fast sampling in the last section.

11.1 Harmonics estimation and rejection in power

distribution systems
There is a steady increase in nonlinear loading in power distribution networks due to
the increase in the use of electrical cars, solar panels for electricity generation, etc.
Nonlinear loading distorts the sinusoidal waveforms of voltage and current in networks. The distorted waveforms are normally still periodic, and they can be viewed as
general periodic disturbances. Based on Fourier series, a general periodic signal can
be decomposed into sinusoidal functions with multiple frequencies of the base frequency. The components with high frequencies are referred to as harmonics. In other
words, they are individual frequency modes at frequencies which are multiples of the
base frequency.

220

Nonlinear and adaptive control systems

Harmonics are undesirable in power distribution networks for various reasons.

They occupy the limited power capacity, and can be harmful to electrical and electronic devices. Often in power distribution networks, active and passive power lters
are used to reduce the undesired harmonics. Effective rejection of harmonics depends
on accurate phase and amplitude estimation of individual frequency modes. In a power
distribution network, harmonics with certain frequencies are more critical for rejection
than others. For example the double frequency harmonics normally disappear in the
system due to a particular connection in power distribution networks, and third-order
harmonics would be most important to reject, perhaps due to high-attenuation high frequencies of distribution networks. Based on this discussion, estimation and rejection
of specic frequency modes will be of interest to power distribution networks.
As explained above, harmonics appear as general periodic signals. Rejection of
general periodic disturbances is discussed in Chapter 10, for matched cases, i.e., the
input is in exactly the same location as the disturbances, which is a restriction to
certain applications, even though we may be able to convert an unmatched case to a
matched one for a certain class of nonlinear systems. The method in Chapter 10 does
not apply to rejection of individual frequency modes. We will show the conversion
of an unmatched disturbance to a matched one, and show how individual frequency
modes can be estimated and rejected.
We will consider a class of nonlinear systems that has a similar structure to
the systems considered in the earlier chapters, but with unmatched general periodic
disturbances. We will show how an equivalent input, also a periodic disturbance, can
be obtained for this system, and then propose an estimation and rejection method
for the system. The presentation at this stage is not exactly based on power systems,
but the proposed methods can be directly applied to power distribution systems. We
will show an example of estimation of harmonics using the proposed method for
individual frequency modes.

11.1.1 System model

Consider a SISO nonlinear system which can be transformed into the output feedback
form
= Ac + (y) + bu + dw
(11.1)
y = C ,
with

0 1 0 ...
0 0 1 ...

Ac = ... ... ... . . .

0 0 0 ...
0 0 0 ...

0
0

.. ,
.

1
0

0
..
.

0

b=
b ,

.
..
bn

T
1
0

C=. ,
..
0

d1

d = ... ,

dn

where Rn is the state vector; u R is the control input; : R Rn with (0) = 0

is a nonlinear function with element i being differentiable up to the (n i)th order;

Control applications

221

b Rn is a known constant Hurwitz vector, with b  = 0, which implies the relative

of the system is ; d is an unknown constant vector; and w R is a bounded periodic
disturbance, which has continuous derivative up to the order of max{ , 0} with
being the index of the rst non-zero element of vector d.
Remark 11.1. The disturbance-free system of (11.1) is in the output feedback form
discussed in the previous chapters. With the disturbance, it is similar to the system
(10.1), with the only difference that w is a general periodic disturbance. It is also
different from the system (10.91) of which the disturbance is in the matched form. 
Remark 11.2. The continuity requirement specied for the general periodic disturbance w in (11.1) is for the existence of a continuous input equivalent disturbance
and a continuous invariant manifold in the state space. For the case of < , we may
allow disturbance to have nite discontinuous points within each period, and for each
of the discontinuous points, the left and right derivatives exist.

Remark 11.3. The minimum phase assumption is needed for the convenience of
presentation of the equivalent input disturbance and the control design based on
backstepping. It is not essential for control design, disturbance rejection or disturbance
estimation. We could allow the system to be non-minimum phase, provided that there
exists a control design for the disturbance-free system which renders the closed-loop
system exponentially stable.

The zero dynamics of (11.1) is linear. To obtain the equivalent input disturbance,
we need a result for steady-state response for stable linear systems.
Lemma 11.1. For a linear system
x = Ax + bw,

(11.2)

where x Rn is the system state, A is Hurwitz, b Rn is a constant vector and w is the

periodic disturbance with period T , the steady state under the input of the periodic
disturbance w is given by
 t
eA(t ) bw( )d + eAt (I eAT )1 eAT WT ,
(11.3)
xs (t) =
0

where WT is a constant vector in Rn given by

 T
WT =
eA bw( )d.
0

Proof. The state response to the input w is given by

 t
eA(t ) bw( )d.
x(t) = eAt x(0) +
0

222

Nonlinear and adaptive control systems

Considering the response after a number of periods, we have

x(NT + t) = eA(NT +t) x(0) + eA(NT +t)

NT +t

eA bw( )d

= eA(NT +t) x(0) + eA(NT +t)

+eA(NT +t)

N 1 

eA bw( )d

iT

i=0
NT +t

(i+1)T

eA bw( )d ,

NT

where N is a positive integer. Since w(t) is a periodic function, we have



(i+1)T

eA bw( )d =

iT

(i+1)T

eA bw( iT )d

iT


=

eA(iT + ) bw( )d

= eiAT

eA bw( )d.

Therefore, we have
x(NT + t) = eA(NT +t) x(0) + eAt

+

N 1

eA(N i)T

eA bw( )d

i=0
t

eA(t ) bw( )d.

(11.4)

The steady-state response in (11.3) is obtained by taking the limit of (11.4) for
t .
2
To obtain the equivalent input disturbance, we need to introduce state transformation to (11.1). To extract the zero dynamics, we introduce a partial state
transformation for system (11.1) as

+1

B bi ,
z = ...
n

i=1

where

b+1 /b

..

.
B=

bn1 /b
bn /b

... 0
.
..
. ..
,

... 1

... 0

1
..
.

b+1 /b

..
b =
.
.

bn /b

Control applications

223

The dynamics with the coordinates (1 , . . . , , z) can be obtained as

i = i+1 + i (y) + di w,

i = 1, . . . , 1

= z1 +

ri i + (y) + d w + b u

(11.5)

i=1

z = Bz + z (y) + dz w,
where
i , for i = 1, . . . , ,
ri = (Bi b)

+1

z (y) = ...
B bi + B by,
n

and

i=1

d+1

i
dz = ...
B bdi .
dn

i=1

The periodic trajectory and the equivalent input disturbance can be found using
the system in the coordinate (1 , . . . , , z). Since the system output y does not
contain the periodic disturbance, we have the invariant manifold for 1 = 0. From
Lemma 11.1 which is used for the result of the steady-state response of linear systems
to the periodic input, we have, for 0 t < T ,

z (t) =

eB(t ) dz w( )d + eBt (I eBT )1 eBT WT

T
with WT = 0 eB dz w( )d . From the rst equation of (11.6), we have, for
i = 1, . . . , 1
i+1 (t) =

di (t)
di w.
dt

Based on the state transformation introduced earlier, we can use its inverse
transformation to obtain

+1

..
i.
Bi b
. = z +
n

i=1

Therefore, the periodic trajectory in the state space is obtained as

= [, . . . , n ]T .

(11.6)

224

Nonlinear and adaptive control systems

Finally, the equivalent input disturbance is given by

1
=
b

d (t)
z,1
ri z,i d w .
dt
i=1

(11.7)

Let x = denote the difference between the state variable and the periodic
trajectory.
The periodic trajectory, , plays a similar role as the invariant manifold in the
set-up for the rejection of disturbances generated from linear exosystems. For this,
we have the following result.
Theorem 11.2. For the general periodic disturbance w in (11.1), the periodic trajectory given in (11.6) and the equivalent input disturbance given in (11.7) are well
dened and continuous, and the difference between the state variable (11.1) and the
periodic trajectory, denoted by x = , satises the following equation:
x = Ac x + (y) + b(u )
y = Cx.

(11.8)

The control design and disturbance rejection will be based on (11.8) instead
of (11.1).
Remark 11.4. The control design and disturbance rejection only use the output y,
with no reference to any of other state of the system. Therefore, there is no difference
whether we refer to (11.1) or (11.8) for the system, because they have the same output.
The format in (11.8) shows that there exist an invariant manifold and an equivalent
input disturbance. However, the proposed control design does not depend on any
information of , other than its period, which is the same as the period of w. In other
words, control design only relies on the form shown in (11.1). The form shown in
(11.8) is useful for the analysis of the performance of the proposed control design,
including the stability. In this section, we start our presentation from (11.1) rather
than (11.8) in order to clearly indicate the class of the systems to which the proposed
control design can be applied, without the restriction to the rejection of matched
disturbances.


11.1.2 Iterative observer design for estimating frequency

modes in input
We will propose an iterative observer design method to estimate specic frequency
modes in the input to a linear system from its output. For the convenience of discussion,
we have the following denitions.

Control applications

225

Denition 11.1. A T -periodic function f is said to be orthogonal to a frequency mode

with frequency k if
 T
f ( ) sin k d = 0,
(11.9)


0
T

f ( ) cos k d = 0.

(11.10)

Denition 11.2. A function f is said to be asymptotically orthogonal to a frequency

mode with frequency k if
 t+T
lim
f ( ) sin k d = 0,
(11.11)
t

lim

t+T

f ( ) cos k d = 0.

(11.12)

We consider a stable linear system

x = Ax + b
y = Cx,

(11.13)

where x Rn is the state variable; and y and u R are output and input, the matrices
A, b and C are with proper dimensions; and the system is stable and observable with
the transfer function Q(s) = C(sI A)1 b. The problem considered in this subsection
is to create a signal such that specic frequency modes can then be removed from the
input , which is a general periodic disturbance described by
(t) =

ak sin (k t + k ),

k=1

and ak and k are the amplitude and phase angle of the mode for
where k = 2k
T
frequency k . The dynamics for a single frequency mode can be described as
w k = Sk wk ,
where
Sk =

0
k

(11.14)
k
0


.

For a single frequency mode with the frequency k as the input denoted by
k = ak sin (k t + k ), its output in y, denoted by yk , is a sinusoidal function with
the same frequency, if we only consider the steady-state response. In fact, based on
the frequency response, we have the following result.
Lemma 11.3. Consider a stable linear system (A, b, C) with no zero at jk for any
integer k. For the output yk of a single frequency mode with the frequency k , there
exists an initial state wk (0) such that
yk = g T wk ,

(11.15)

226

Nonlinear and adaptive control systems

where g = [1 0]T and wk is the state of (11.14). Furthermore, the input k for this
frequency mode can be expressed by
k = gkT wk ,
where
1
gk =
mk

(11.16)
cos k

sin k

sin k cos k

g := Qk g

(11.17)

with k = Q(jk ) and mk = |Q(jk )|.

Proof. For the state model (11.14), we have

1

 j t
1 1
1 1
e k
Sk t
e =
j j
j j
ejk t


cos (k t) sin (k t)
=
.
sin (k t) cos (k t)
For the single frequency mode k , the output is given by
yk = mk ak sin (k t + k + k ).
With g = [1 0]T , we have


mk ak sin (k + k )
wk (0) =
mk ak cos (k + k )
and

wk (t) = eSk t wk (0) =

mk ak sin (k t + k + k )
mk ak cos (k t + k + k )

such that
yk = g T eSk t wk (0) = g T wk (t).
Considering the gain and phase shift of Q(s), we have
1 T Sk (t(k /k ))
g e
wk (0)
mk
1 T (k /k )Sk Sk t
=
g e
e wk (0).
mk

k =

Hence, we have
gk =

1 (k /k )S T
k g
e
mk

and therefore
1 (k /k )S T
1 (k /k )Sk
1
k =
e
e
=
Qk =
mk
mk
mk

cos k
sin k

sin k
cos k


.

Control applications

227

For a stable single-input linear system, if the input is a T -periodic signal that is
orthogonal to a frequency k , the steady state, as shown earlier, is also T -periodic.
Furthermore, we have the following results.
Lemma 11.4. If the input to a stable single-input linear system (A, b) is T -periodic
signal that is orthogonal to a frequency mode k , for any positive integer k, the steady
state is orthogonal to the frequency mode and the state variable is asymptotically
orthogonal to the frequency mode. Furthermore, if the linear system (A, b, C) has
no zero at jk , the steady-state output is orthogonal to the frequency mode k if and
only if the input to the system is orthogonal to the frequency mode.
Proof. We denote the input as , and the state variable x satises
x = Ax + b.
Since is T -periodic, the steady-state solution of the above state equation, denoted
by xs , is also T -periodic and
x s = Axs + b.
Let


Jk =

xs ( ) sin k d.

Using integration by part, we have

 T
xs d cos k
Jk = k1
= k1
= k1
=

k1 A

= k2 A

x s cos k d

0
T

(Axs + b(t)) cos k d

xs cos k d

xs d sin k

= k2 A

(Axs + b(t)) sin k d

= k2 A2k Jk .
Hence, we have
(k2 I + A2 )Jk = 0.

228

Nonlinear and adaptive control systems

Since A is a Hurwitz matrix which cannot have k j as its eigenvalues, we conclude

Jk = 0. Similarly we can establish


xs ( ) cos k d = 0

and therefore xs is orthogonal to the frequency mode k .

If we denote ex = x xs , we have e x = Aex . It is clear that ex exponentially
converges to zero, and therefore we can conclude that x is asymptotically orthogonal
to the frequency mode k . This completes the proof of the rst part.
For the second part of the lemma, the if part follows directly from the rst part
of the lemma, and we now establish the only if part by seeking a contradiction.
Suppose that the output y is orthogonal to the frequency mode k , and the input is
not. In this case, it can be shown in a similar way as in the proof Lemma 11.3, that
there exists a proper initial condition for wk (0) such that gkT wk is orthogonal to
k . Since the system is linear, we can write
y = y + yw ,
where y denotes the steady-state output generated by gkT wk and yw generated
by gkT wk . We have y orthogonal to the frequency mode k . However, yw would be
a sinusoidal function with frequency k and it is denitely not orthogonal. Thus we
conclude y is not orthogonal to k , which is a contradiction. Therefore, must be
orthogonal to k if y is. This completes the proof.
2
To remove frequency modes in the input is to nd an estimate such that
does not contain those frequency modes asymptotically. For a single frequency mode
with frequency k , the task to obtain a is accomplished by
w k = Sk w k + lk (y(t) g T w k )
= (Qk g)T w k = gkT w k ,

(11.18)

where lk is chosen such that Sk lk g T is Hurwitz. For this observer, we have a useful
result stated in the following lemma.
Lemma 11.5. For any positive integer k, with the observer as designed in (11.18),
(( ) gkT w k ) is asymptotically orthogonal to the frequency mode k , i.e.,


t+T

lim


lim

t+T

(( ) gkT w k ) sin k d = 0,

(11.19)

(( ) gkT w k ) cos k d = 0.

(11.20)

Control applications

229

Proof. Consider the steady-state output y of input . From Lemma 11.3, there exists
an initial state wk (0) such that
 T
(y( ) g T wk ( )) sin k d = 0,


0
T

(y( ) g T wk ( )) cos k d = 0,

which implies that gkT wk is orthogonal to the frequency mode k , again based on
Lemma 11.3.
Let w k = wk w k . The dynamics of w k can be obtained from (11.14) and
(11.18) as
w k = S k w k lk (y g T wk ),

(11.21)

where S k = Sk lk g T . Note that S k is a Hurwitz matrix and (y g T wk ) is a

T -periodic signal. There exists a periodic steady-state solution of (11.21) such that
k = S k k lk (y g T wk ).
From Lemma 11.4, k is orthogonal to the frequency mode k because (y g T wk )
is. Let ek = w k k . We have
e k = S k ek ,
which implies that ek exponentially converges to zero. The observer state w k can be
expressed as
w k = wk k ek .
Therefore, (11.19) and (11.20) can be established, and this completes the proof.

The result in Lemma 11.5 shows how an individual frequency mode can be
removed with the observer designed in the way as if the output would not contain
other frequency modes. From the proof of Lemma 11.5, it can be seen that there
is an asymptotic error, k , between the observer state and the actual state variables
associated with the frequency mode k . Although k is orthogonal to the frequency
mode k , it does in general contain components generated from all the other frequency
modes. Because of this, a set of observers of the same form as shown in (11.18) would
not be able to extract multiple frequency modes simultaneously. To remove multiple
frequency modes, it is essential to nd an estimate which is asymptotically orthogonal
to the multiple frequency modes. For this, the interactions between the observers must
be dealt with.
Suppose that we need to remove a number of frequency modes k for all the k
in a nite set of positive integers K = {ki }, for = 1, . . . , m. To estimate the frequency
modes for k,i , i = 1, . . . , m, we propose a sequence of observers,
w k,1 = Sk,1 w k,1 + lk,1 (y g T w k,1 )

(11.22)

230

Nonlinear and adaptive control systems

and, for i = 2, . . . , m,
T
w k,i1
k,i1 = Ak,i1 + bgk,i1

i1

Ck,j g T w k,i ,
w k,i = Sk,i w k,i + lk,i y

(11.23)
(11.24)

j=1

where lk,i , for i = 1, . . . , m, are designed such that S k,i := Sk,i lk,i g T are Hurwitz,
and


1
cos k,i sin k,i
g := Qk,i g
gk,i =
mk,i sin k,i cos k,i
with mk,i = |C(jwk,i A)1 b| and k,i = C(jwk,i A)1 b.
The estimate for the input disturbance which contains the required frequency
modes for asymptotic rejection is given by
m =

T
gk,i
w k,i .

(11.25)

i=1

The estimate m contains all the frequency modes k,i , for i = 1, . . . , m. The useful
property of the estimate is given in the following theorem.
Theorem 11.6. For the estimate m given in (11.25), m is asymptotically
orthogonal to the frequency modes k,i for i = 1, . . . , m.
Proof. In the proof, we will show how to establish the asymptotic orthogonality in
detail by induction.
We introduce the notations w k,i = wk,i w k,i . We use k,i to denote the steadystate solutions of w k,i and ek,i = w k,i k,i , for i = 1, . . . , m.
Lemma 11.5 shows that the results hold for m = 1. Let
T
(wk,1 k,1 )
1 = gk,1

and 1 is orthogonal to the frequency mode k,1 .

We now establish the result for m = 2. From Lemma 11.3, there exists an initial
state variable wk,2 (0) for the dynamic system
w k,2 = Sk,2 wk,2

(11.26)

such that y Cqk,1 g wk,2 is orthogonal to the frequency mode k,2 where qk,i , for
i = 1, . . . , m 1, denote the steady-state solution of
T

T
(wk,i k,i ).
q k,i = Aqk,i + bgk,i
T
wk,2 to the system
Note that g T wk,2 can be viewed as the output for the input gk,2
T
(A, b, C), based on Lemma 11.3. Hence, y Cqk,1 g wk,2 is the output for the input
T
T
gk,1
(wk,1 k,1 ) gk,2
wk,2 to the system (A, b, C). Therefore, from Lemma 11.4,
T
T
gk,1 (wk,1 k,1 ) gk,2
wk,2 is orthogonal to k,2 . Furthermore, from the previous

Control applications

231

T
T
(wk,1 k,1 ) gk,2
wk,2 is orthogonal
step k,1 is orthogonal to k,1 . Hence, gk,1
to the frequency modes k,j for j = 1, 2.
The dynamics of w k,2 and k,2 are obtained as

w k,2 = S k,2 w k,2 lk,2 (y Ck,1 g T wk,2 ),

k,2 = S k,2 k,2 lk,2 (y Cqk,1 g T wk,2 ),

(11.27)
(11.28)

and the error ek,2 = w k,2 k,2 satises

e k,2 = S k,2 ek,2 lk,2 C(qk,1 k,1 ).
Since qk,1 k,1 exponentially converges to zero, so does ek,2 . From (11.28) and
Lemma 11.4, k,2 is orthogonal to the frequency modes k,j for j = 1, 2. Therefore,
by letting
T
T
(wk,1 k,1 ) + gk,2
(wk,2 k,2 ),
2 = gk,1

2 is orthogonal to the frequency modes k,j for j = 1, 2.

Notice that
2 2 =

T
gk,j
(wk,j k,j w k,j )

j=1

T
gk,1
ek,j .

j=1

Hence 2 2 converges to zero exponentially. Therefore, we conclude that 2

is asymptotically orthogonal to the frequency modes k,j for j = 1, 2.
Now suppose that the result holds for m = i and therefore i is orthogonal
to the frequency modes k,j for j = 1, . . . , i, with
i =

T
gk,j
(wk,j k,j ).

j=1

We need to establish that the result holds for m = i + 1.

For the frequency mode k,i+1 , there exists an initial state variable wk,i (0) for the
dynamic system
w k,i+1 = Sk,i wk,i+1

such that y C ij=1 qk,j g T wk,i+1 is orthogonal to the frequency mode k,i+1 . Note
i
T
that y C j=1 qk,j g T wk,i+1 is the output for the input i gk,i+1
wk,i+1 to the
T
system (A, b, C). Therefore from Lemma 11.4, i gk,i+1 wk,i+1 is orthogonal
to k,i+1 . Since i is orthogonal to the frequency modes k,j for j = 1, . . . , i,
T
wk,i+1 is orthogonal to the frequency modes k,j for j = 1, . . . , i + 1
i gk,i+1

and so is (y C ij=1 qk,j g T wk,i+1 ).

232

Nonlinear and adaptive control systems

The dynamics of w k,i+1 and k,i+1 are obtained as
w k,i+1 = S k,i+1 w k,i+1

lk,i+1 y C

k,j g T wk,i+1 ,

(11.29)

j=1

k,i+1 = S k,i+1 k,i+1

lk,i+1 y C

qk,j g T wk,i+1 .

(11.30)

j=1

Since the input is orthogonal to the frequency modes k,j for j = 1, . . . , i + 1, we

conclude from Lemma 11.4 that k+i+1 is orthogonal to k,j for j = 1, . . . , i + 1, and
therefore i+1 is orthogonal to the frequency modes k,j for j = 1, . . . , i + 1 with
T
i+1 = i + gk,i+1
(wk,i+1 k+i+1 ).

The error ek,i+1 = w k,i+1 k,i+1 satises

e k,i+1

= S k,i+1 ek,i+1 lk,i+1 C

(qk,j k,j ).

j=1

Since qk,j k,j , for j = 1, . . . , i, exponentially converge to zero, so does ek,i+1 . With
i+1 i+1 =

i+1

T
gk,j
(wk,j k,j w k,j )

j=1

T
gk,j
ek,j ,

j=1

i+1 i+1 converges to zero exponentially. Therefore, we conclude that i+1

is asymptotically orthogonal to the frequency modes k,j for j = 1, . . . , i + 1.
Since i is an arbitrary integer between 1 and m, the result holds for i = m 1,
and this completes the proof.
2

11.1.3 Estimation of specic frequency modes in input

From the previous analysis on the equivalent input disturbance, we can convert the
unmatched disturbance case to the matched one. Therefore, we can now carry on the
disturbance rejection based on the matched case:
x = Ac x + (y) + b(u )
y = Cx.

(11.31)

Control applications

233

If the disturbance does not exist in (11.31), the system (11.31) is in the linear
observer error with output injection that is shown in Chapter 8. In that case, we can
design a state observer as
p = (Ac LC)p + (y) + bu + Ly,

(11.32)

where p Rn , L Rn is chosen so that Ac LC is Hurwitz. The difculty in the state

estimation is due to the unknown disturbance . Consider
q = (Ac LC)q + b,

(11.33)

where q denotes the steady-state solution. Such a solution exists, and an explicit
solution is given earlier. Each element of q is a periodic function, as is periodic.
We have an important property for p and q stated in the following lemma.
Lemma 11.7. The state variable x can be expressed as
x = p q + ,

(11.34)

where p is generated from (11.32) with q satisfying (11.33) and  satisfying

 = (Ac LC).

(11.35)

From (11.33), we have Cq as the steady-state output of the system ((Ac

kC), b, C) for the equivalent input disturbance . If Cq is available, we are ready to use
the results presented in the previous section for disturbance estimation of the specied
frequency modes. The result shown in Lemma 11.7 indicates that Cq = Cp y C,
and hence Cp y exponentially converges to Cq. Therefore, we now propose an
observer for frequency modes in the input disturbances using Cp y. To estimate the
frequency modes for k,i i = 1, . . . , m, we have
w k,1 = Sk,1 w k,1 + lk,1 (Cp y g T w k,1 )

(11.36)

and, for i = 2, . . . , m,
T
k,i1 = (Ac kC)k,i1 + bgk,i1
w k,i1 ,

i1

Ck,j g T w k,i ,
w k,i = Sk,i w k,i + lk,i Cp y

(11.37)
(11.38)

j=1

where lk,i , for i = 1, . . . , m, are designed such that S k,i := Sk,i lk,i g T are Hurwitz,
and


cos k,i sin k,i
1
gk,i =
g := Qk,i g
(11.39)
mk,i sin k,i cos k,i
with mk,i = |C(jwk,i (A kC))1 b| and k,i = C(jwk,i (A kC))1 b.

234

Nonlinear and adaptive control systems

The estimate for the input disturbance which contains the required frequency
modes for asymptotic rejection is given by
m

m =

T
gk,i
w k,i .

(11.40)

i=1

The estimate m contains all the frequency modes k,i , for i = 1, . . . , m. The useful
property of the estimate is given in the following theorem.
Theorem 11.8. The estimate m given in (11.40) is bounded and satises the
following:
 t+T
[( ) m ( )] sin k,i d = 0,
(11.41)
lim
t

lim

t+T

[( ) m ( )] cos k,i d = 0,

(11.42)

for i = 1, . . . , m.
Proof. From Theorem 11.6, the results in (11.41) and (11.42) hold if the input
Cp y in (11.36), (11.37) and (11.38) equals Cq. Since Cp y Cq converges
exponentially to 0, and (11.36)(11.38) are stable linear systems, the errors in
the estimation of caused by replacing Cq by Cp y in (11.36)(11.38) are also
exponentially convergent to zero. Therefore, (11.41) and (11.42) can be established
from Theorem 11.6.
2

11.1.4 Rejection of frequency modes

With the estimated frequency modes m , we set the control input as
u = v + m ,
where v is designed based on backstepping design which is shown in Chapter 9. Due
to the involvements of disturbance in the system, we need certain details of stability
analysis with explicit expression of Lyapunov functions. For the convenience of stability analysis, we adopt an approach using backstepping with ltered transformation
shown in Section 9.4. The control input v can be designed as a function
v = v(y, ),
where is the state variable of the input lter for the ltered transformation. The nal
control input is designed as
u = v(y, ) + .

(11.43)

For the stability of the closed-loop system with disturbance rejection, we have the
following theorem.

Control applications

235

Theorem 11.9. The closed-loop system of (11.1) under the control input (11.43)
ensures the boundedness of all the state variables, and the disturbance modes of
specied frequencies k,i , for i = 1, . . . , m, are asymptotically rejected from the
system in the sense that all the variables of the closed-loop system is bounded and
the system is asymptotically driven by the frequency modes other than the specied
modes.
Proof. For the standard backstepping design for the nonlinear systems, we establish
the stability by considering the Lyapunov function
1

1
1 2
,
V = z T Pz + y2 +
2
2 i=1 i

(11.44)

where is a constant and i = i i and P is a positive denite matrix satisfying DT P + PD = I . The standard backstepping design ensures that there exists a
positive real constant 1 such that
V 1 V [2z T Pbz + yby ]

1 V 2 V
3 V + 4 2

(11.45)

for some positive real constants i , i = 2, 3, 4. Hence, the boundedness of ensures

the boundedness of all the state variables. It is clear from Theorem 11.8 that all the
specied frequency modes are asymptotically removed from .
This completes the
proof.
Remark 11.5. In the control design, we have used the control input obtained with the
backstepping design for the proof of Theorem 11.9. In case that there exists a known
control design for the disturbance-free system which ensures exponential stability of
the disturbance-free case, we can pursue the proof of Theorem 11.9 in a similar way
to obtain the result shown in (11.45). This is very useful in particular for the case of
non-minimum phase systems, to which the standard backstepping presented earlier
does not apply.


11.1.5 Example
In this section, we use a simple example to demonstrate estimation of harmonics using
the method proposed earlier in this section.
We consider a voltage signal with diados as shown in Figure 11.1. We are
going to use iterative observers to estimate harmonics in this signal. To simplify
the presentation, we assume that the signal is directly measurable. In this case,
there are no dynamics between the measurement and the point of estimation.
Hence, we can simplify the algorithms presented in Subsection 11.1.2 ((11.22),

236

Nonlinear and adaptive control systems

1.5

Voltage

0.5

0.5

1
0.7

0.72

0.74

0.76

0.78

0.8

t (s)

Figure 11.1 Voltage signal of a nonlinear load

(11.23) and (11.24)). In this case, there is no need to use the observer model in
(11.23). The simplied algorithm for direct estimation of harmonics is presented
below.
To estimate the frequency modes for k,i , i = 1, . . . , m, the iterative observers
are given by
w k,1 = Sk,1 w k,1 + lk,1 (y g T w k,1 )

(11.46)

and, for i = 2, . . . , m,

w k,i = Sk,i w k,i + lk,i y

g T w k,j ,

(11.47)

j=1

where lk,i , for i = 1, . . . , m, are designed such that S k,i := Sk,i lk,i g T are Hurwitz.
For the electricity, the base frequency is 50 Hz. We will show the estimates
of harmonics from the base frequency, that is, we set k,1 = 100 , k,2 = 200 ,
k,3 = 300, etc. For these frequency modes, we have


0
i100
.
Sk,i =
i100
0

Control applications

237

The observer gain lk,i can be easily designed to place the close-loop poles at any
specied positions. For the simulation study, we place the poles at {200, 200}. For
such pole positions, we have

lk,i =

400
40000
200
i100


.

The harmonics of second and third orders are plotted with the original signal and
the estimated base frequency component in Figure 11.2. The approximations of
accumulative harmonics to the original signal are shown in Figure 11.3.
In this example, we have shown the estimation of harmonics from direct
measurements. The advantage of this estimation is that the estimation is on-line
and can be implemented in real time. Also, the estimation using the proposed
method can provide the phase information, which is very important for harmonic
rejection. We did not show the rejection of harmonics in this example. However,
it is not difcult to see that rejection can be simulated easily with the proposed
method.

1.5
Original
1st order
2nd order
3rd order

Signal & harmonics

0.5

0.5

1
0.74

0.745

0.75

0.755

0.76

0.765

t (s)

Figure 11.2 Estimated harmonics

0.77

0.775

0.78

238

Nonlinear and adaptive control systems

1.5
Original
1st order
Up to 2nd order
Up to 3rd order

Signal & approximation

0.5

0.5

1
0.74

0.745

0.75

0.755

0.76
t (s)

0.765

0.77

0.775

0.78

Figure 11.3 Approximation using harmonic components

11.2 Observer and control design for circadian rhythms

Circadian rhythms play an important role in daily biological activities of living
species. Circadian rhythms, or biological clocks, normally have a period of 24 h, and
changes to the rhythms can be referred to as changes in phases. Circadian disorder is
a phenomenon of circadian rhythms which occurs when internal rhythms cannot keep
up with the changes of external environmental rhythms. Changes of environmental
rhythms, presented by the change of light/dark cycles or by irregular rhythms, result
in phase shifts between internal and external rhythms. For example, jet lags are
caused when people travel from one time zone to another. The existence of these
phase shifts in longer term has negative effect to health. Therefore, in biological
study of circadian rhythms, it is important to nd methods to recover the shifted
phases, such as jet lags, to their normal rhythms. These methods could then lead to
treatments of circadian disorder, like sleep disorder and jet lag. Circadian rhythms can
be accurately modelled as limit cycles of nonlinear dynamic systems. We will apply
the observer and control design shown in earlier chapters for analysis and control
of circadian rhythms. The phase restoration is carried out by the synchronisation of
trajectories generated from a controlled model with the trajectories of a reference
system via nonlinear control design. Both reference and controlled systems are based

Control applications

239

on a given third-order model of Neurospora circadian rhythms. An observer is also

designed to estimate two unknown states from the measurement of a single state.

11.2.1 Circadian model

For almost every living organism on the Earth, their daily biological activities are
governed by rhythms. These biological rhythms are called circadian rhythms. The
circadian rhythms exist as self-sustained and periodic oscillations, and they are also
known with their entrainment to 24-h day/night cycle. The entrainment is based on
the activity of a circadian clock gene. For mammals, the circadian clock gene, also
known as the circadian pacemaker, is found in the suprachiasmatic nuclei (SCN) of
the anterior hypothalamus. The circadian pacemaker captures information sent from
an external environment cue such as light, and then coordinates the timing of other
slave clocks or slave oscillators in other parts of the body. Any changes of environment
cues which cause the mismatch between external and internal rhythms can lead to
disruption of circadian rhythms. This phenomenon is known as circadian disorders.
Jet lags due to trans-continent ights and sleeping disorders due to irregular sleep
wake cycles are two typical examples of circadian disorders. In practice, one of the
known medical treatments for circadian disorders is the application of light. Light is
major external environmental cue, and with light input, the circadian phase can be
adjusted to the light/dark rhythms at destination.
In this section, we aim for the restoration of circadian phase using nonlinear
control. In order to achieve this objective, we propose an alternative control design
method which synchronises trajectories generated from a controlled model with the
trajectories generated from a reference model via backstepping approach. Both reference system and controlled system are based on a third-order mathematical model
of Neurospora circadian rhythms. The trajectories generated by controlled system
represent the altered rhythms. Meanwhile, the reference trajectories represent the
desired rhythms which the trajectories of controlled system are adjusted to match. We
also present an observer design for this circadian rhythm based on observer design
presented in Chapter 8 for nonlinear systems with Lipschitz nonlinearities.
A third-order mathematical model is developed to describe molecular mechanism
of circadian rhythms in Neurospora. Its molecular mechanism is based on the negative
feedback exerted by FRQ proteins on the expression of frq gene. Transcription of frq
gene yields messenger RNA (mRNA), and the translation of which synthesises FRQ
protein. These synthesised FRQ proteins are then transferred back into nucleus where
they inhibit the transcription of frq gene. A new activation of frq gene transcription
will restart the cycle. Dynamics of these variables, frq mRNA, FRQ protein and
nuclear FRQ protein, are by the following dynamic model:
Kin
x1
vm
n
n
K i + x3
KM + x 1
x2
x 2 = ks x1 vd
k 1 x2 + k 2 x 3
Kd + x 2
x 3 = k1 x2 k2 x3 ,

x 1 = vs

(11.48)

240

Nonlinear and adaptive control systems

where x1 , x2 and x3 denote concentration of frq mRNA, concentration of FRQ protein

outside nucleus and concentration of nucleus FRQ protein respectively. Values of
three state variables x1 , x2 and x3 are assumed to be positive values. In system
(11.49), the parameter vs denotes the transcription rate of frq gene. The other
parameters involved in system (11.49) are Ki , n, vm , KM , ks , k1 , k2 , vd and Kd . The
parameters Ki , n, vm , KM represent the threshold constant beyond which nuclear FRQ
protein inhibits the transcription of frq, the Hill coefcient showing the degree of
co-operativity of the inhibition process, the maximum rate of frq mRNA degradation
and the Michaelis constant related to the latter process respectively. The parameters
ks , k1 and k2 denote the rate constant measuring the rate of FRQ synthesis, the rate constants of the transport of FRQ into and out of the nucleus respectively. The parameter
vd denotes maximum rate of FRQ degradation and Kd is the Michaelis constant related
to this process. For the third-order Neurospora model, the parameters have their typical values as vs = 1.6 nM h1 , Ki = 1 nM, n = 4, vm = 0.7 nM h1 , KM = 0.4 nM,
ks = 1 h1 , vd = 4 nMh1 , Kd = 1.4 nM, k1 = 0.3 h1 , k2 = 0.15 h1 .
Circadian rhythms are self-sustained and periodic oscillations. With the set
parameters as above, dynamics of state variables of (11.48) can sustain periodic oscillations, and in fact, a limit cycle. Figure 11.4 shows the plots of three state variables
obtained in simulation with the model (11.48). For simulation study, we choose the

T
initial values x(0) = 5 1 1 . This particular initial value is actual a point on the
limit cycle. If the initial value starts from a point outside the limit cycle, then there
trajectory will converge to the limit cycle. There can be other selections for values of
initial conditions.

State variables of Neurospora model

6
x1
x2
x3

0
0

10

20

30

40

50

60

70

80

90

100

Time (h)

Figure 11.4 Circadian rhythm of Neurospora

110

120

Control applications

241

The model (11.48) is not in the right format for control and observer design. We
introduce a state transformation as
z = Tx
with

0
T = 0
k1 ks

(11.49)

1
0,
0

0
k1
0

and the transformed system is described by

z1 = z2 k2 z1
k 1 z2
k 1 Kd + z 2
k1 ks Kin
k1 k s z 3
z3 = vs n
vm
.
Ki + z1n
K M k1 ks + z 3

z2 = z3 k1 z2 + k1 k2 z1 vd

(11.50)

Now the transformed model (11.50) is in the lower-triangular format.

11.2.2 Lipschitz observer design

We have presented observer design for nonlinear dynamic systems with Lipschitz
nonlinearity in Section 8.4. Basically, for a nonlinear system described by (8.29)
x = Ax + (x, u)
y = Cx,
a nonlinear observer can be designed as in (8.30)
x = Ax + L(y C x ) + (x, u),

(11.51)

provided that the conditions specied in Theorem 8.10 are satised. That suggests
that we need nd the Lipschitz constant together with a positive denite matrix P
for nonlinearities in the circadian model (11.48) such that the inequality (8.36)
AT P + PA + 2I 2 C T C < 0
is satised, and then we obtain the observer gain as
L = P 1 C T .
The one-sided Lipschitz condition shown in (8.36) can be checked using the
Lipschitz constant for individual elements in the vector . Let us denote the Lipschitz
constants of i by i . The condition shown in (8.36) can be guaranteed by
AT P + PA + 2n

n

i=1

i i I 2 C T C < 0,

(11.52)

242

Nonlinear and adaptive control systems

where i are positive real constants, and P ni=1 P n (i, i ). The denition of
P ni=1 P n (i, i ) is given by
P n (i, i ) = {P : |pji | < i , for j = 1, 2, . . . , n}.
For the system (11.50), we set


C1 = 1 0 0
and therefore state variables z2 and z3 are unknown. Comparing with the structure in
(8.29), we have

0
,
2
(z, u) =
3a + 3b
where clearly 1 = 0, and
2 = vd
3a = vs

k 1 z2
,
k1 K d + z 2

k1 ks Kin
,
Kin + z1n

3b = vm

k 1 k s z3
.
K M k1 k s + z 3

From the denition of Lipschitz constant in Denition 8.2, we have



 vd k1 z2

vd 
vd k1 z2 




 k K + z + k K + z  K z2 z2 ,
1 d
2
1 d
2
d


 vm k1 ks z3

vm 
vm k1 ks z3 

z3 z3  ,

 K k k + z + K k k + z  K
M 1 s
3
M 1 s
M
3
and therefore we obtain the Lipschitz constants as 2 = Kvd =10.7962 for 2 (z2 ), and
d
3b = KvmM =1.01 for 3b (z3 ). For nonlinear function 3a (z1 ), its Lipschitz constant can
be computed by using mean value theorem which is described by


    f (x) f (x) 
f ( ) = 

 x x  ,
where [x, x ]. Setting f = 3a , we have
 


    nvs k1 ks Kin n1   3a (z1 ) 3a (z1 ) 
f ( ) =  

,
=

2

 

z1 z1
Kin + n
where [min (z1 , z1 ), max (z1 , z1 )]. We nd maximum value of |f  ( )| by solving
|f  ( )| = 0, and obtain the result as 0.325. Since the value of Lipschitz constant is
equivalent to maximum value of f  ( ), the Lipschitz constant is given by 3a = 0.325.
The Lipschitz constant 3 is given by 3 = 3a + 3b = 1.326.

Control applications

243

After the Lipschitz constants are found, we solve (11.52) to obtain

3.9887
= 67.8032, L = 2.6067
1.3469
for the observer (11.51). Simulation study for the observer has been carried out with
the results shown in Figures 11.5 and 11.6.
4.5
z2
Estimate of z2

Transformed state z2 and its estimate

4
3.5
3
2.5
2
1.5
1
0.5
0
0

10

20

30

40

50

60
70
Time (h)

80

90

100

110

120

Figure 11.5 Observer state for circadian model

11.2.3 Phase control of circadian rhythms

The objective of phase control is to reset a distorted phase. Our strategy is to control
the distorted circadian rhythm to follow a target rhythm. In order to design a control
strategy, we need to dene a sensible control input for the circadian model. Among
the parameters appeared in (11.48), parameter vs , which denotes the rate of frq mRNA
transcription, is sensitive to light input. Therefore, for Neurospora circadian rhythms,
this parameter is usually used as control input in many results which have been
presented in literature. If vs is taken as the control input, the circadian model (11.48)
is rewritten as
x1
Kn
x 1 = (vs + u) n i n vm
K i + x3
KM + x 1
x2
x 2 = ks x1 vd
k 1 x2 + k 2 x 3
Kd + x 2
x 3 = k1 x2 k2 x3 ,

244

Nonlinear and adaptive control systems

1.8
z3
Estimate of z3

Transformed state z3 and its estimate

1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0

10

20

30

40

50

60
70
Time (h)

80

90

100

110

120

Figure 11.6 Observer state for circadian model

where u is the control input. The system model after the state transformation (11.49)
is then obtained as
z1 = z2 k2 z1
k 1 z2
k1 K d + z 2
k1 ks Kin
k1 ks z 3
k1 ks Kin
z3 = vs n

v
+
u
.
m
Ki + z1n
K M k1 k s + z 3
Kin + z1n
z2 = z3 k1 z2 + k1 k2 z1 vd

(11.53)

This model (11.53) is then used for phase control design. We use q to denote the state
variable for the target circadian model
q 1 = q2 k2 q1
q 2 = q3 k1 q2 + k1 k2 q1 vd
q 3 = vs

k1 q2
k1 K d + q 2

k1 ks Kin
k1 k s q 3
vm
,
Kin + q1n
K M k1 ks + q 3

(11.54)

to which the variable z is controlled to follow.

The transformed dynamic models (11.53) and (11.54) are in the triangular form,
and therefore the iterative backstepping method shown in Section 9.2 can be applied.
There is a slight difference in the control objective here from the convergence to zero

Control applications

245

in Section 9.2. For this, we dene

e1 = z1 q1 ,
e2 = z2 q2 ,

(11.55)

e3 = z3 q3 ,
where ei , for i = 1, 2 and 3, are the tracking errors. The error dynamics are obtained as
e 1 = e2 k2 e1
e 2 = e3 k1 e2 + k1 k2 e1
e 3 = vs

vd k12 Kd e2
(k1 Kd + z2 ) (k1 Kd + q2 )

k1 ks Kin
k1 ks Kin
vm KM (k1 ks )2 e3
k1 ks Kin

.
+
u
s
Kin + z1n
Kin + q1n
Kin + z1n
(KM k1 ks + z3 ) (KM k1 ks + q3 )
(11.56)

It is easy to see that the model (11.56) is still in the triangular form. The phase resetting
is achieved if we can ensure that the errors ei , i = 1, 2, 3, converge to zero. To the
model (11.56), iterative backstepping can be applied. Following the procedures shown
in Section 9.2, we dene
w1 = e1 ,
w2 = e2 1 ,

(11.57)

w 3 = e3 2 ,
where 1 and 2 are stabilising functions to be designed.
From the rst equation of (11.56), we obtain dynamics of w1 as
w 1 = e2 k2 e1
= w 2 + 1 k 2 e1 .
The stabilising function 1 is then designed as
1 = c1 w1 + k2 e1 ,

(11.58)

where c1 is a positive real constant. The resultant dynamics of w1 are given by

w 1 = c1 w1 + w2 .

(11.59)

The dynamics of w2 are obtained as

w 2 = e3 k1 e2 + k1 k2 e1

vd k12 Kd e2
1
e 1

e1
(k1 Kd + z2 ) (k1 Kd + q2 )

= w 3 + 2 k 1 e2 + k 1 k 2 e 1

vd k12 Kd e2
1
e 1 .

e1
(k1 Kd + z2 ) (k1 Kd + q2 )

The stabilising function 2 can be designed as

2 = w1 c2 w2 + k1 e2 k1 k2 e1 + vd k1

k1 K d e 2
1
e 1
+
e1
(k1 Kd + z2 ) (k1 Kd + q2 )

246

Nonlinear and adaptive control systems

with c2 being a positive real constant, which results in the dynamics of w2

w 2 = w1 c2 w2 + w3 .

(11.60)

From the dynamics of w3

w 3 = vs

k1 ks Kin
k1 ks Kin
k1 ks Kin

v
+
u
s
Kin + z1n
Kin + q1n
Kin + z1n

vm KM (k1 ks )2 e3
2
2
e 1
e 2 ,

k
k
+
z
k
k
+
q
e
e2
(KM 1 s
3 ) (KM 1 s
3)
1

we design the control input u as

k1 ks Kin
k1 ks Kin
2
Kin + z1n
w

c
w

v
+
v
+
u=
(e2 k2 e1 )
2
3
3
s
s
n
n
n
n
n
k1 k s K i
Ki + z1
Ki + q1
e1

vd k12 Kd e2
2
2
(e3 k1 e2 + k1 k2 e1 )
+
,
(11.61)
e2
e2 (k1 Kd + z2 ) (k1 Kd + q2 )
with c3 as a positive real constant. The resultant dynamics of w3 are then obtained as
w 3 = w2 c3 w3 .

(11.62)

The stability analysis of the proposed control design can be established in the
same way as the proof shown in Section 9.2 for iterative backstepping control design.
Indeed, consider a Lyapunov function candidate
V =


1 2
w1 + w22 + w32 .
2

(11.63)

Using the dynamics of wi , i = 1, 2, 3, in (11.59), (11.60) and (11.62), we obtain

V = w1 w 1 + w2 w 2 + w3 w 3
= w1 (c1 w1 + w2 ) + w2 (w1 c2 w2 + w3 ) + w3 (w2 c3 w3 )
= c1 w12 c2 w22 c3 w32 .
Therefore, we can conclude that the closed-loop system under the control of u in
(11.61) is exponentially stable with respect to the variables wi , i = 1, 2, 3, which
suggests that the controlled circadian rhythm asymptotically tracks the targeted one.
Simulation study of the control design proposed above has been carried with the
control parameters c1 = c2 = c3 = 0.1. The states z2 and z3 are shown in Figures 11.7
and 11.8 when the control input is applied at t = 50 h, and the control input is shown
in Figure 11.9. The plot of z1 is very similar to the plot of z2 , and it is omitted. It can
be seen from the gures that there is a phase difference before the control input is
applied, and the control input resets the phase of the control circadian rhythm to the
targeted one.

Control applications

247

4.5
q2
z2

4
Control input applied

3.5

q2 and z2

3
2.5
2
1.5
1
0.5
0
0

10

20

30

40

50

60
70
Time (h)

80

90

100

110

120

Figure 11.7 Circadian phase reset for state 2

1.8
q3
z3

1.6
Control input applied
1.4

q3 and z3

1.2
1
0.8
0.6
0.4
0.2
0

10

20

30

40

50

60
70
Time (h)

80

90

100

110

120

Figure 11.8 Circadian phase reset for state 3

11.3 Sampled-data control of nonlinear systems

A practical issue of applying nonlinear control strategies is to implement them in
computers or microprocessors which are discrete-time in nature. For a continuoustime system with a discrete-time controller, inevitably we end with a sampled-data

248

Nonlinear and adaptive control systems

5
Control input u
0

Control input

5
10
15
20
25
30
0

10

20

30

40

50

60
70
Time (h)

80

90

100

110

120

Figure 11.9 Control input for circadian phase reset

control system. It is well known for linear systems, we can use emulation method,
that is, to design a controller in continuous-time and then implement its discrete-time
version. Alternative to emulation method is the direct design in discrete-time. Stability
analysis for sampled-data control of linear dynamic systems can then be carried out in
the framework of linear systems in discrete-time. Sampled-data control of nonlinear
systems is a much more challenging task.
For a nonlinear system, it is difcult or even impossible to obtain a nonlinear
discrete-time model after sampling. Even with a sampled-data model in discretetime, the nonlinear model structure cannot be preserved in general. It is well known
that nonlinear control design methods do require certain structures. For example
backstepping can be applied to lower-triangular systems, but a lower-triangular system
in continuous-time will not have its discrete-time model in the triangular structure in
general after sampling. Hence, it is difcult to carry out directly design in discrete-time
for continuous-time nonlinear systems.
For a control input designed based on the continuous-time model, it can be sampled
and implemented in discrete-time. However, the stability cannot be automatically
guaranteed for the sampled-data system resulted from emulation method. The stability
analysis is challenging because there is no corresponding method in discrete-time for
nonlinear systems after sampling, unlike linear systems. One would expect that if the
sampling is fast enough, the stability of the sampled-data system might be guaranteed
by the stability of the continuous-time system. In fact, there is a counter example to
this claim. Therefore, it is important to analyse the stability of sampled-data system
in addition to the stability of the continuous-time control, for which the stability is
guaranteed in control design.
In this section, we will show the stability analysis of sampled-data control for a
class of nonlinear systems in the output feedback form. A link between the sampling

Control applications

249

period and initial condition is established, which suggests that for a given domain
of initial conditions, there always exists a sampling time such that if the sampling is
faster than that time, the stability of the sampled-data system is guaranteed.

11.3.1 System model and sampled-data control

In this section, we consider sampled-data control for a class of nonlinear systems in
the output feedback form. This class of nonlinear systems (9.17) has been considered
for control design in continuous-time in Chapter 9, using observer backstepping in
Section 9.3 and using backstepping with ltered transformation in Section 9.4. We will
present sampled-data control based on the control input in continuous-time obtained
using backstepping with ltered transformation. For the convenience of presentation,
we will show the system again and outline the key steps in the continuous-time here.
The system, as shown in (9.17) and (9.32), is described as
x = Ac x + bu + (y)

(11.64)

y = Cx,
with

0 1 0 ...
0 0 1 ...

Ac = ... ... ... . . .

0 0 0 ...
0 0 0 ...

0
0

.. ,
.

1
0

0
..
.

0

b=
b ,

.
..
bn

T
1
0

C=. ,
..
0

where x Rn is the state vector; u R is the control input; : R Rn with (0) = 0

is a nonlinear function with element i being differentiable up to the (n i)th order;
and b Rn is a known constant Hurwitz vector with b  = 0, which implies the relative
of the system is .
We design an output feedback control for the system in (11.64). As shown in
Chapter 9, the continuous-time control input can be designed using the backstepping with ltered transformation in Section 9.4. In this section, we carry out the
stability analysis based on the control input obtained using backstepping with ltered
transformation.
In this section, we need to deal with control in continuous-time and discrete-time.
For notation, we use uc for the control input obtained in continuous-time and ud for
the resultant sampled-data control input. For notational convenience, we re-write the
lter (9.33) as
=  + bf u

(11.65)

u = uc (y, ),

(11.66)

250

Nonlinear and adaptive control systems

where

1
0

= .
..

1
2
..
.

0 ...
0
1 ...
0

..
..
.. ,
.
.
.
0 . . . 1

0
..

bf = . .
0
1

For a given uc , the sampled-data controller based on emulation method is given as

ud (t) = uc (y(mT ), (mT )),

t [mT , mT + T ),

(11.67)

T


(mT ) = eT ((m 1)T ) + bf uc (y((m 1)T ), ((m 1)T )) e d ,

(11.68)

where y(mT ) is obtained by sampling y(t) at each sampling instant; (mT ) is the
discrete-time implementation of the lter shown in (11.65); T is the xed sampling
period; and m is the discrete-time index, starting from 0.
Before the analysis of the sampled-data control, we briey review the control
design in continuous-time.
As shown in Section 9.4, through the state transformations, we obtain
= D + (y)
y = 1 + y (y) + b 1 .

(11.69)

For the system with relative degree = 1, as shown in Lemma 9.5 and its proof,
the continuous-time control uc1 can be designed as


1
P 2 (y) 2
y y (y)
,
(11.70)
uc1 = c0 +
4
y
where c0 is a positive real constant and P is a positive real constant that satises
DT P + PD = 3I .
For the control (11.70), the stability of the continuous-time system can be established
with the Lyapunov function
1
V = T P + y2
2
and its derivative
V c0 y2 2 .
For the case of > 1, backstepping is used to obtain the nal control input uc2 .
We introduce the same notations zi , for i = 1, . . . , , as in Section 9.4 for backstepping
z1 = y,
zi = i1 i1 ,
z+1 = u ,

(11.71)
for i = 2, . . . , ,

(11.72)
(11.73)

where i for i = 2, . . . , are stabilising functions to be designed. We also use the

positive real design parameters ci and ki for i = 1, . . . , and > 0.

Control applications

251

The control design has been shown in Section 9.4. We list a few key steps here
for the convenience of the stability analysis. The stabilising functions are designed as
1 = c1 z1 k1 z1 + y (y)


P 2 (y) 2
,
y


1 2
1
2 = z1 c2 z2 k2
z2 +
(y) + 1 1
y
y


i1 2
i1
i = zi1 ci zi ki
zi +
(y) + i1 i1
y
y
 i1
+ i2
(j j + j+1 ) for i = 3, . . . , .
j=1
j

(11.74)

The continuous-time control input is given by uc2 = , that is


uc2 = z1 c z k
+

i2

1
j=1

1
y

2
z +

1
(y)
y

(j j + j+1 ) + 1 1 .

(11.75)

For the control input (11.75) in the continuous-time, the stability result has been
shown in Theorem 9.6. In the stability analysis, the Lyapunov function candidate is
chosen as
V =

zi2 + T P

i=1

and its derivative is shown to satisfy

V

ci zi2 2 .

(11.76)

i=1

11.3.2 Stability analysis of sampled-data systems

The following lemma is needed for stability analysis in sampled-data case.
Lemma 11.10. Let V : Rn R+ be a continuously differentiable, radially
unbounded, positive denite function. Dene D := { Rn |V ( ) r} with r > 0.
Suppose
V V + Vm ,

t (mT , (m + 1)T ],

(11.77)

252

Nonlinear and adaptive control systems

hold for all (mT ) D, where , are any given positive reals with > , T > 0 the
xed sampling period and Vm := V ( (mT )). If (0) D, then the following holds:
lim (t) = 0.

Proof. Since (0) D, then (11.77) holds for t (0, T ] with the following form:
V V + V ( (0)).
Using the comparison lemma (Lemma 4.5), it is easy to obtain from the above that
for t (0, T ]
V ( (t)) et V0 +

1 et
V0 = q(t)V0 ,

(11.78)

where q(t) := (et + (1 et )). Since > > 0, then q(t) (0, 1), t
(0, T ]. Then we have
V ((t)) < V0 ,

t (0, T ].

(11.79)

Particularly, setting t = T in (11.78) leads to

V1 q(T )V0 ,

(11.80)

which means that (T ) D. Therefore, (11.77) holds for t (T , 2T ]. By induction,

we have
V ( (t)) < Vm ,

t (mT , (m + 1)T ].

(11.81)

which states inter-sample behaviour of the sampled-data system concerned, and in

particular
Vm+1 q(T )Vm

(11.82)

indicating that V decreases at two consecutive sampling points with a xed ratio.
From (11.82),
Vm q(T )Vm1 qm (T )V0 ,

(11.83)

which implies that limm Vm = 0. The conclusion then follows from (11.81), which
completes the proof.
2
Remark 11.6. Lemma 11.10 plays an important role in stability analysis for the
sampled-data system considered in this section. When sampled-data systems are
analysed in discrete-time in literature, only the performances at sampling instances are
considered. In this section, Lemma 11.10 provides an integrated analysis framework
where the behaviour of the sampled-data system both at and between sampling instants
can be characterised. In particular, the systems behaviour at sampling instants is
portrayed in (11.82), and (11.81) shows that the inter-sample behaviour of the system
is well bounded by a compact set dened by Vm .


Control applications

253

For the case of relative degree 1, the sampled-data system takes the following
form:
= D + (y)
(11.84)
y = 1 + y (y) + ud1 ,
where ud1 is the sampled-data controller and can be simply implemented via a zeroorder hold device as the following:
ud1 (t) = uc1 (y(m)),

t [mT , mT + T ).

(11.85)

Dene := [ T , yT ]T and we have the following result.

Theorem 11.11. For system (11.84) and the sampled-data controller ud1 shown in
(11.85), and a given neighbourhood of the origin Br := { Rn | r} with r
any given positive real, there exists a constant T1 > 0 such that, for all 0 < T < T1
and for all (0) Br , the system is asymptotically stable.
Proof. We choose
1
V ( ) = T P + y2
2
as the Lyapunov function candidate for the sampled-data system. We start with some
sets used throughout the proof. Dene
c := max V ( )
Br

and the level set

c := {x Rn |V ( ) c}.
There exist two class K functions 1 and 2 such that
1 ( ) V ( ) 2 ( ).
Let l > 11 (c), and dene Bl := { Rn | l}. Then we have
Br c Bl .
The constants Lu1 , L1 and L2 are Lipschitz constants of the functions uc1 and y and
with respect to Bl .
These local Lipschitz conditions establish that for the overall sampled-data system
with (0) c , there exists a unique solution (t) over some interval [0, t1 ). Notice
that t1 might be nite. However, later analysis shows that the solution can be extended
one sampling interval after another, and thus exists for all t 0 with the property
that limt (t) = 0. Particularly, we intend to formulate the time derivative of the
Lyapunov function V into the form shown in (11.77) or (11.93), which is shown below.
Consider the case when t = 0, (0) Br c . First, choose a sufciently large
l such that there exists a T1 > 0, and for all T (0, T1 ), the following holds:
(t) Bl ,

t [0, T ], (0) c .

(11.86)

254

Nonlinear and adaptive control systems

The existence of T1 is ensured by continuous dependency of the solution (t) on the

initial conditions.
Next, calculate the estimate of |y(t) y(0)| forced by the sampled-data control
ud1 during the interval [0, T ], provided that (0) Br c and T (0, T1 ). From
the second equation of (11.84), the dynamics of y are given by
y = 1 + y (y) + ud1 .
It follows that

y(t) = y(0) +

+

ud1 ( )d

0
t

1 ( )d +
0


y (y) y (y(0)) d +

y (y(0)) d.

Then we have

|y(t) y(0)|

( ) d +
0


!

+

Lu1 |y(0)|d

1

L1 |y( ) y(0)|d +

L1 |y(0)|d.

(11.87)

We rst calculate the integral 1 . From the rst equation of system (11.69), we obtain
 t
Dt
(t) = e (0) +
eDt (y( ))d.
(11.88)
0

Since D is a Hurwitz matrix, there exist positive reals 1 and 1 such that eDt
1 e1 t . Thus, from (11.88)
 t
1 t
(t) 1 e (0) +
1 e1 (t ) (y( )) (y(0)) d

+

0
t

1 e1 (t ) (y(0)) d

1 e

1 t


(0) + L2


+ L2

1 e1 (t ) |y( ) y(0)|d

0
t

1 e1 (t ) |y(0)|d.

Then the following inequality holds:

1

1 L2
1 (0)
|y(0)|t
(1 e1 t ) +
1
1

1 L2 t
+
|y( ) y(0)|d.
1 0

(11.89)

Control applications

255

Now we are ready to compute |y(t) y(0)|. In fact, we have from (11.87) and (11.89)
|y(t) y(0)| A1 (1 e1 t ) + B1 t + H

|y( ) y(0)|d ,

(11.90)

where
A1 = 11 1 (0) ,
B1 = Lu1 |y(0)| + L1 |y(0)| + 11 1 L2 |y(0)|,
H = 11 1 L2 + L1 .
Applying GronwallBellman inequality to (11.90) produces
|y(t) y(0)| A1 (1 e1 t ) +

B1 Ht
(e 1)
H

+ A1 (1 eHt + He1 t (H + 1 ))(H + 1 )1

(11.91)

Setting t = T on the right side of (11.91) leads to

|y(t) y(0)| 1 (T )|y(0)| + 2 (T ) (0) ,
where
1 (T ) = H 1 (Lu1 + L1 + 11 1 L2 )(eHT 1)
2 (T ) = 11 1 (1 eHT + He1 T (H + 1 ))(H + 1 )1
+ 11 1 (1 e1 T ).

(11.92)

Note that 1 (T ) and 2 (T ) only depend on the sampling period T once the Lipschitz
constants and the control parameters are chosen.
Next we shall study the behaviour of the sampled-data system during each interval using a Lyapunov function candidate V (y, ) = T P + 12 y2 . Consider (0) =
[ (0), y(0)]T Br . When t (0, T ], its time derivative satises
V = 3 2 + 2 T P + y(1 + y + ud1 )
c0 y2 2 + |y||ud1 (y(0)) uc1 |
c0 y2 2 + Lu1 |y y(0)||y|


Lu1
(1 (T ) + 2 (T )) y2 2
c0
2
Lu1
Lu1
1 (T )|y(0)|2 +
2 (T ) (0) 2
2
2
1 (T )V + 1 (T )V ( (0), y(0)),
+

(11.93)

256

Nonlinear and adaptive control systems

where

"

#
1
1 = min 2c0 Lu1 1 Lu1 2 (T ),
,
max (P)
"
#
Lu1 2 (T )
1 = max Lu1 1 (T ),
,
2min (P)

(11.94)

with max ( ) and min ( ) denoting the maximum and minimum eigenvalues of a matrix
respectively.
Next we shall show that there exists a constant T2 > 0 such that the condition 1 (T ) > 1 (T ) > 0 is satised for all 0 < T < T2 . Note from (11.92) that both
1 (T ) and 2 (T ) are actually the continuous functions of T with 1 (0) = 2 (0). Dene
e1 (T ) := 1 (T ) 1 (T ) and we have e1 (0) > 0. It can also be established from
(11.94) that e1 (T ) is a decreasing and continuous function of T , which asserts by
the continuity of e1 (T ) the existence of T2 so that for 0 < T < T2 , e1 (T ) > 0, that is
0 < 1 (T ) < 1 (T ).
Finally, set T1 = min (T1 , T2 ), and from Lemma 11.10, it is known that V1 c,
which means (T ) c , and subsequently, all the above analysis can be repeated for
every interval [mT , mT + T ]. Applying Lemma 11.10 completes the proof.
2
For systems with relative degree > 1, the implementation of the sampled-data
controller ud2 is given in (11.67) and (11.68). It is easy to see from (11.68) that (mT )
is the exact, discrete-time model of the lter
=  + bf ud2

(11.95)

due to the fact that ud remains constant during each interval and the dynamics of
shown in (11.95) is linear. Then (11.68) and (11.95) are virtually equivalent at each
sampling instant. This indicates that we can use (11.95) instead of (11.68) for stability
analysis of the sampled-data system.
Let := [ T , z T ]T and we have the following result.
Theorem 11.12. For the extended system consisting (11.65), (11.69) and the
sampled-data controller ud2 shown in (11.67) and (11.68), and a given neighbourhood of the origin Br := { Rn | r} with r any given positive real, there exists
a constant T2 > 0 such that, for all 0 < T < T2 and for all (0) Br , the system is
asymptotically stable.
Proof. The proof can be carried out in a similar way to that for the case = 1,
except that the effect of the dynamic lter of has to be dealt with.
For the overall sampled-data system, a Lyapunov function candidate is chosen
the same as for continuous-time case, that is

V = T P +

1 2
z .
2 i=1 i

Control applications

257

Similar to the case of = 1, the sets Br , c and Bl can also be dened such that
Br c Bl , and there exists a T3 > 0 such that for all T (0, T3 ), the following
holds:
(t) Bl ,

t (0, T ],

(0) c .

As in the proof for Theorem 11.11, we also aim to formulate the time derivative
of V ( ) into form (11.77). Next we shall derive the bounds for (t) (0) and
|y(t) y(0)| during t [0, T ] with 0 < T < T3 . Consider the case where (0) Br .
We have from (11.95)
 t
e(t ) bf ud2 d.
(11.96)
(t) = et (0) +
0

Since  is a Hurwitz matrix, there exist positive reals 2 , 3 and 2 such, that
et 2 e2 t ,
et I 3 (1 e2 t ),
where I is the identity matrix. Then, using the Lipschitz property of uc2 with respect
to the set Bl and the fact that ud2 (0, 0) = 0, it can be obtained from (11.96) that
 t
 2 Lu2
2 (0) 
(11.97)
( ) d
1 e2 t +
(|y(0)| + (0) )t
2
2
0
and
(t) (0) 3 (0) (1 e2 t ) + ud2 (y(0), (0))

2 e2 (t ) d

3 (T )|y(0)| + 4 (T ) (0) ,

(11.98)

where
3 (T ) = 21 2 Lu2 (1 e2 T ),
4 (T ) = (3 + 21 2 Lu2 )(1 e2 T ),
and Lu2 is a Lipschitz constant of uc2 . As for |y y(0)|, we have from (11.69)
 t
 t
1 ( )d +
( )d
y(t) = y(0) +

+

0
t


y (y) y (y(0)) d +

y (y(0))d.

It can then be shown that

 t
 t
|y(t) y(0)|
z( ) d +
( ) d
0 
! 0 
!

+
0

1

2

L1 |y( ) y(0)|d +
0

L1 |y(0)|d ,

(11.99)

258

Nonlinear and adaptive control systems

where 1 is already shown in (11.89) and 2 in (11.97). With (11.89), (11.97) and
(11.99), it follows that
|y(t) y(0)| A1 (1 e1 t ) + A2 (1 e2 t ) + B2 t
 t
+H
|y( ) y(0)|d

(11.100)

where A1 and H are dened in (11.90)

A2 = 21 2 (0) ,
B2 = L1 |y(0)| + 11 1 L2 |y(0)| + 21 2 Lu2 |y(0)| + 21 2 Lu2 (0) .
Dening A3 := A1 + A2 and 0 := max (1 , 2 ), we have
 t
|y( ) y(0)|d.
|y(t) y(0)| A3 (1 e0 t ) + B2 t + H
0

Applying the GronwallBellman lemma produces

|y(t) y(0)| 5 (T )|y(0)| + 6 (T ) z(0) + 7 (T ) (0) ,

(11.101)

where
5 (T ) = H 1 (L1 + 11 1 L2 + 2 21 Lu2 )(eHT 1),
6 (T ) = 11 1 (0 eHT + He0 T (H + 0 ))(H + 0 )1 + 11 1 (1 e0 T ),
7 (T ) = 21 2 (1 e0 T ) + 21 2 Lu2 (eHT 1)
+ 21 2 Lu2 (0 eHT + He0 T (H + 0 ))(H + 0 )1 .
Note that (0) appears in (11.98) and (11.101) while for the analysis using Lyapunov
function to carry on, z(0) is needed. Therefore, it is necessary to nd out an
expression of (0) that exclusively involves z(0) , which is shown below.
Notice that due to the special structure of the lter (11.65) and the backstepping
technique, each stabilising function has the property that 1 = 1 (y), 1 (0) = 0, and
i = i (y, 1 , . . . , i1 ) and i (0, . . . , 0) = 0, i = 2, . . . , 1. From (11.74) we have
|1 (0)| |z2 (0)| + |1 (0)| z2 (0)| + L1 |y(0)|
|2 (0)| |z3 (0)| + |2 (0)| |z3 (0)| + L2 |y(0)| + L2 |1 (0)|
..
.
|1 (0)| |z (0)| + |1 (0)| |z (0)| + L1 |y(0)| + L1

|i (0)|

i=1

where with a bit abuse of notation, Li is the Lipschitz constant of i with respect to
the set Bl . Thus, a constant L0 can be found such that the following holds:
(0) L0 ( z (0) + |y(0)|).

(11.102)

Control applications

259

with z = [z2 , . . . , z ]T , which implies that if [ (0)T , y(0)T , z (0)T ]T Bl , then

[ (0)T , y(0)T , (0)T ]T will be conned in a bounded set, denoted by Bl .
Then the time derivative of

1
1 2
Vd2 = T P + y2 +
z
2
2 i=2 i
during the interval (0, T ] satises
V d2 2

ci zi2 + z (ud2 uc2 )

i=1

c1 y2 2 0

zi2 + z |ud2 uc2 |,

(11.103)

i=2

where 0 = min{c2 , . . . , c }. In addition, we have

|ud2 uc2 | Lu2 (|y y(0)| + (0) )
Lu2 (3 (T ) + 5 (T ))|y(0)| + Lu2 6 (T ) (0)
+ Lu2 (4 (T ) + 7 (T )) (0)
1 (T )|y(0)|2 + 2 (T ) (0) 2 + 3 (T ) (0) 2
+ 4 (T ) z 2 ,

(11.104)

where 1 (T ) = L2u2 (3 (T ) + 5 (T )), 2 = L2u2 6 (T ), 3 = L2u2 (4 (T ) + 7 (T )), 4 =

Lu2 7

i=3 i (T ) and Lu2 is a Lipschitz constant of uc2 with respect to the set Bl
2
dependent on Bl .
From (11.102) to (11.103), we then have
V d2 c1 y2 2 (0 4 )z 2
+ (1 (T ) + 2L20 3 (T ))|y(0)|2 + 2 (T ) (0) 2 + 2L20 3 (T ) z (0) 2
= 2 (T )Vd2 + 2 (T )Vd2 ( (0), y(0), z (0)),
where
"
2 (T ) = min 2c1 ,

max (P)

#
, 2(0 4 (T )) ,

"
#
2 (T )
2 (T ) = max 2(1 (T ) + 2L20 3 (T )),
, 4L20 3 (T ) .
min (P)

260

Nonlinear and adaptive control systems

Note from (11.98), (11.101) and (11.104) that each i (1 i 4) is a continuous

function of T with i (0) = 0. Then, as shown in the analysis for the case = 1, it
can be claimed that given arbitrary positive reals c1 , and , there exists a constant
T4 > 0 such that for all 0 < T < T4 , 2 (T ) > 2 (T ) > 0.
Finally, letting T2 := min (T3 , T4 ) proves the theorem.
2
Remark 11.7. In the results shown in Theorems 11.11 and 11.12, the radius r of Br
can be any positive value. It means that we can set the stability region as large as we
like, and the stability can still be guaranteed by determining a fast enough sampling
time. Therefore, Theorems 11.11 and 11.12 establish the semi-global stability of the
sampled-data control of nonlinear systems in the output feedback form.


11.3.3 Simulation
The following example is a simplied model of a jet engine without stall:
3
1
m = + m2 m3
2
2
= u,
where m is the mass ow and the pressure rise. Take m as the output and then the
above system is in the output feedback form. The lter = + u is introduced so
that the ltered transformation y = x1 and = x2 , and the state transformation
= y can render the system into the following form:


3 2 1 3
=
y y 2 y
2
2

y = + y +


3 2 1 3
y y + .
2
2

Finally, the stabilising function




2
3 2 1 3
3
1 2
2
y y y
y y +
1 = y
2
2
2
2
with P = 1, and the control uc can be obtained using 2 as shown in (11.74). For
simulation, we choose = 0.5.
Simulations are carried out with the initial values set as x1 (0) = 1 and
x2 (0) = 1, which are for simulation purpose only and have no physical meaning.
Results shown in Figures 11.10 and 11.11 indicate that the sampled-data system
is asymptotically stable when T = 0.01 s, which is conrmed by a closer look
at the convergence of V shown in Figure 11.12. Further simulations show that
the overall system is unstable if T = 0.5 s. In summary, the example illustrates
that for a range of sampling period T , the sampled-data control design presented
earlier in this section can asymptotically stabilise the sampled-data system.

Control applications

261

1.2

0.8

x1

0.6

0.4

0.2

0.2

10

t (s)

Figure 11.10 The time response of x1 for T = 0.01 s

0.5

x2

0.5

1.5

2
0.2

0.2

0.4

0.6

0.8

x1

Figure 11.11 Phase portrait of the system for T = 0.01 s

1.2

262

Nonlinear and adaptive control systems

4.5
4
3.5

3
2.5
2
1.5
1
0.5

0.01

0.01

0.02

0.03

0.04

0.05

t (s)

Figure 11.12 The convergence of V for T = 0.01 s

0.06

Bibliographical Notes

Chapter 1. The basic concepts of systems and states of nonlinear systems discussed in this chapter can be found in many books on nonlinear systems and
nonlinear control systems, for example Cook [1], Slotine and Li [2], Vidyasagar
[3], Khalil [4] and Verhulst [5]. For the background knowledge of linear systems,
readers can consult Ogata [6], Kailth [7]. Antsaklis and Michel [8], Zheng [9],
and Chen, Lin and Shamash [10]. The existence of unique solutions of differential equations is discussed in the references such as Arnold [11] and Borrelli and
Coleman [12]. Many concepts such as stability and backstepping control design are
covered in detail in later chapters of the book. For system with saturation, readers may consult Hu and Lin [13]. Sliding mode control is covered in detail in
Edwards and Spurgeon [14] and feedforward control in Isidori [15]. Limit cycles
and chaos are further discussed in Chapter 2. Semi-global stability is often related
to systems with saturation, as discussed in Hu and Lin [13], and Isidori [15].
Chapter 2. Lipschitz conditions and the existence of solutions for differential
equations are covered in detail in Arnold [11] and Borrelli and Coleman [12]. Classications of singular points of second-order nonlinear systems are covered in almost
all the introductory texts on nonlinear control systems, Cook [1], Slotine and Li
[2], Vidyasagar [3] and Khalil [4]. More phase portraits of systems similar to the
one in Example 2.1 can be found in Slotine and Li [2], Khalil [4] and Verhulst [5].
The swing machine equation in Example 2.2 was a simplied model of the system
shown in Elgerd [16] and a similar treatment is shown in Cook [1]. More examples
of similar systems can also be found in Verhulst [5]. The van der Pol oscillator was
rst discussed in his publication van der Pol [17], and perhaps it is one of the best
known systems with limit cycles. Detailed coverage of van der Pol oscillator can be
found in several books, including Verhulst [5] and Borrelli and Coleman [12]. Lorenz
attractor was rst shown in Lorenz [18] for his research on heat transfer. The boundedness of the trajectory shown in the book was taken from Borrelli and Coleman [12].
Chapter 3. Describing function analysis is covered in many classic texts on nonlinear control, such as Cook [1], Slotine and Li [2], Vidyasagar [3], and Khalil [4].
The stability criterion of limit cycles with describing functions is inuenced by the
treatment in Slotine and Li [2]. The describing function analysis for van der Pol systems is adapted from Cook [1]. More results on describing functions are to be found
in Atherton [19] and Gelb and Velde [20].

264

Nonlinear and adaptive control systems

Chapter 4. Lyapunovs work on stability was published in Russian in 1892 and

was translated in French in 1907 [21]. Lyapunov stability is covered in almost all the
books on nonlinear systems and control. One of the early books on motion stability is
published by Hahn [22]. Example 4.4 is adapted from Slotine and Li [2], and a more
general form is shown in Reference [5]. More discussions on radially unbounded
functions and global stability can be found in Khalil [4] and in Slotine and Li [2].
Further results on comparison lemma and its applications are to be found in Khalil
[4] and Vidyasagar [3].
Chapter 5. The denition of positive real systems are common in many books
such as Slotine and Li [2], Khalil [4] and Vidyasagar [3]. For strictly positive real
systems, there may have some variations in denitions. A discussion on the variations can be found in Narendra and Annaswamy [23]. The denition in this book
follows Slotine and Li [2], Khalil [4]. Lemma 5.3 was adapted from Khalil [4],
where a poof of necessity can also be found. KalmanYakubovich Lemma can appear
in different variations. A proof of it can be found in Khalil [4] and Marino and
Tomei [24]. A book on absolute stability was published by Narendra and Taylor [25].
Circle criterion is covered in a number of books such as Cook [1], Slotine and Li
[2], Vidyasagar [3] and Khalil [4]. The treatment of loop-transformation to obtain
the condition of circle criterion is similar to Cook [1]. The interpretation based on
Figure 5.3 is adapted from Khalil [4]. Complex bilinear mapping is also used to
justify the circle interpretation. Basic concepts of complex bilinear mapping can
be found from text books on complex analysis, such as Brown and Churchill [26].
Input-to-state stability was rst introduced by Sontag [27]. For the use of comparison
functions of classes K and K , more are to be found in Khalil [4]. The denitions for ISS are adapted from Isidori [15]. Theorem 5.8 is a simplied version, by
using powers of the state norm, rather than K to characterise the stability for the
convenience of presentation. The original result of Theorem 5.8 was shown by Sontag and Wang [28]. Using ISS pair (, ) for interconnected systems is based on
a technique, changing supply functions, which is shown in Sontag and Teel [29].
The small gain theorem for ISS was shown in Jiang, Teel and Praly [30]. A systematic presentation of ISS stability issues can be found in Isidori [15]. Differential
stability describes the stability issue of nonlinear systems for observer design. It is
introduced in Ding [31], and the presentation in Section 5.4 is adapted from that
paper.
Chapter 6. Fundamentals of Lie derivatives and differential manifolds can be
found in Boothby [32]. A brief description of these concepts, sufcient for the
concepts used in this book, is presented in Isidori [33]. Early results on exact linearisation in state space started from Brockett [34], and the presentation in this chapter
was greatly inuenced by the work of Isidori [33] and Marino and Tomei [24]. A
proof of Frobenius Theorem can be found in Isidori [33].
Chapter 7. For self-tuning control, readers can refer to the textbooks such as
Astrom and Wittenmark [35] and Wellstead and Zarrop [36]. The approach to obtain

Bibliographical Notes

265

the model reference control is based on Ding [37]. Early work of using Lyapunov
function for stability analysis can be found in Parks [38]. A complete treatment of
MRAC of linear systems with high relative degrees can be found in Narendra and
Annaswamy [23] and Ioannou and Sun [39]. The example to show the divergence
of parameter estimation under a bounded disturbance is adapted from Ioannou and
Sun [39], where more robust adaptive laws for linear systems are to be found. Results
on robust adaptive control and on relaxing assumptions of the sign of high-frequency
gain, minimum-phase, etc., for nonlinear systems can be found in Ding [37, 4045].
Chapter 8. Linear observer design is covered in many books on linear systems.
A proof of Lemma 8.2 can be found in Zheng [9]. Early results on nonlinear
observer design can be traced back to 1970s, for example in Thau [46], and Kou,
Ellitt and Tarn [47]. The result on the output injection form was shown in Krener
and Isidori [48]. The geometric conditions for the existence of a state transformation to the output injection form can also be found in Isidori [33] and Marino and
Tomei [24]. Basic knowledge of differential manifolds can be found in Boothby
[32]. Linear observer error dynamics via direct state transformation was initially
shown by Kazantzis and Kravaris [49]. Further results on this topic can be found
in Xiao [50] and Ding [51]. In the latter one, polynomial approximation to nonlinear functions for a solution to a nonlinear state transformation is discussed and
an explicit region of convergence of such an approximation is given. The basic
idea of observer design for systems with Lipschitz nonlinearities is to use the linear part of the observer dynamics to dominate the nonlinear terms. This idea was
shown in some of the early results on nonlinear observers, but the systematic introduction to the topic was due to Rajamani [52], in which the condition (8.31) was
shown. Some later results on this topic are shown in Zhu and Han [53]. Onesided Lipschtiz condition for nonlinear observer design was shown in Zhao, Tao
and Shi [54]. The result on observer design for systems with Lipschitz output nonlinearities is adapted from Ding [55]. The result on reduced-order observer of linear
systems can be traced back to the very early stage of observer design in Luenberger
[56]. Early results on this topic were discussed in Kailath [7]. The reduced-order
observer design is closely related to observer design for systems with unknown
inputs, of which some results may be found in Hou and Muller [57]. For nonlinear systems, a reduced-order observer was used for estimation of unknown states
for control design of a class of nonlinear systems with nonlinearities of unmeasured states by Ding [58]. The result shown here is based on Ding [31]. An
early result on adaptive observer for nonlinear systems is reported in Bastin and
Gevers [59]. Results on adaptive observers with output nonlinearities are covered
in Marino and Tomei [24]. The results shown in Section 8.6 is mainly based on Cho
and Rajamani [60]. The method to construct a rst-order positive real system in
Remark 8.9 is based on Ding [40].
Chapter 9. Early ideas of backstepping appeared in a number of papers, such
as Tsinias [61]. Iterative backstepping for state feedback is shown in Kanellakopoulos, Kokotovic and Morse [62] with adaptive laws for unknown parameters.

266

Nonlinear and adaptive control systems

Backstepping designs without adaptive laws shown in this chapter are basically simplied versions of their adaptive counterparts. Filtered transformations were used for
backstepping in Marino and Tomei [63], and the presentation in this chapter is different from the forms used in [63]. Initially, multiple estimation parameters are used
for adaptive control with backstepping, for example the adaptive laws in [63]. Tuning function method was rst introduced in Krstic, Kanellakopoulos and Kokotovic
[64] to remove multiple adaptive parameters for one unknown parameter vector with
backstepping. Adaptive backstepping with ltered transformation was based on Ding
[65], without the disturbance rejection part. Adaptive observer backstepping is largely
based on the presentation in Krstic, Kanellakopoulos and Kokotovic [66]. Nonlinear
adaptive control with backstepping is also discussed in details in Marino and Tomei
[24]. Some further developments of nonlinear adaptive control using backstepping can
be found in Ding [40] for robust adaptive control with dead-zone modication, robust
adaptive control with -modication in Ding [67], with unknown control directions
and unknown high-frequency gains in Ding [41, 42]. Adaptive backstepping was also
applied to nonlinear systems with nonlinear parameterisation in Ding [68].
Chapter 10. Early results on rejection of sinusoidal disturbances can be found
in Bodson, Sacks and Khosla [69] and Bodson and Douglas [70]. Asymptotic
rejection of sinusoidal disturbances for nonlinear systems in the output feedback
form is reported in Ding [71] when the frequencies are known. The result shown
in Section 10.1 is adapted mainly from Ding [72]. Asymptotic rejection of periodic disturbances can be formulated as an output regulation problem when the
periodic disturbances are generated from a dynamic system, which is known as
an exosystem. Output regulation for linear systems is shown in Davison [73]
and Francis [74], and for nonlinear systems with local stability in Huang and
Rugh [75], and Isidori and Byrnes [76]. More results on output regulation with
local stability can be found in the books by Isidori [33] and Huang [77]. The
result shown in Section 10.2 is based on Ding [65]. The format of the exosystem for internal model design was inspired by Nikiforov [78]. The Nussbaum
gain was rst used by Nussbaum [79], and the treatment here is similar to Ye
and Ding [80] used in the control design to tackle the unknown sign of high frequency gain. Early results on output regulation with nonlinear exosystems can
be found in Ding [81, 82], and Chen and Huang [83]. The result shown in
Section 10.3 is based on Xi and Ding [84]. Asymptotic rejection of general disturbances was studied as an extension of output regulation of nonlinear exosystems.
Waveform proles of the general periodic disturbances were used to estimate the
equivalent input disturbance in some early results in Ding [8587]. The approach
for asymptotic rejection with an observer-like internal model is adapted from
Ding [55, 88].
Chapter 11. Harmonic estimation is treated as a special case of state estimation.
The recursive estimation and rejection method in Section 11.1 is based on Ding
[89]. Practical issues on harmonics estimation and rejections are covered in Arrillaga,
Watson and Chen [90] and Wakileh [91].

Bibliographical Notes

267

Clock gene was reported in 1994 by Takahashi et al. [92]. Reports on key genes
in circadian oscillators are in Albrecht et al. [93], van der Horst et al. [94] and Bunger
et al. [95]. The effects of jet lags and sleep disorders to circadian rhythms are reported
in Sack et al. [96] and Sack et al. [97]. The effects of light and light treatments
to disorders of circadian rhythms are shown by Boulos et al. [98], Kurosawa and
Goldbeter [99] and Geier et al. [100]. The circadian model (11.48) is adapted from
Gonze, Leloup and Goldbeter [101]. The condition (11.52) is shown by Zhao, Tao
and Shi [54]. The observer design is adapted from That and Ding [102], and the control design is different from the version shown in that paper. More results on observer
design for a seventh-order circadian model are published by That and Ding [103].
There are many results in literature on sampled-data control of linear systems,
for example see Chen and Francis [104] and the references therein. The difculty
of obtaining exact discrete-time model for nonlinear systems is shown by Nesic and
Teel [105]. Structures in nonlinear systems cannot be preserved as shown in Grizzle
and Kokotovic [106]. Approximation to nonlinear discrete-time models is shown
by several results, for example Nesic, Teel and Kokotovic [107] and Nesic and Laila
[108]. The effect on fast-sampling of nonlinear static controllers is reported in Owens,
Zheng and Billings [109]. Several other results on emulation method for nonlinear
systems are shown in Laila, Nesic and Teel [110], Shim and Teel [111] and Bian
and French [112]. The results presented in Section 10.3 are mainly based on Wu and
Ding [113]. Sampled-data control for disturbance rejection of nonlinear systems is
shown by Wu and Ding [114]. A result on sampled-data adaptive control of a class of
nonlinear systems is to be found in Wu and Ding [115].

References

[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]

Cook, P. A. Nonlinear Dynamic Systems. London: Prentice-Hall, 1986.

Slotine, J.-J. E. and Li, W. Applied Nonlinear Control. London: PrenticeHall, 1991.
Vidyasagar, M. Nonlinear Systems Analysis, 2nd ed. Englewood Cliffs,
New Jersey: Prentice-Hall International, 1993.
Khalil, H. K. Nonlinear Systems, 3rd ed. Upper Saddle River, New Jersey:
Prentice Hall, 2002.
Verhulst, F. Nonlinear Differential Equations and Dynamic Systems. Berlin:
Springer-Verlag, 1990.
Ogata, K. Modern Control Engineering, 3rd ed. London: Prentice-Hall,
1997.
Kailath, T. Linear Systems. London: Prentice-Hall, 1980.
Antsaklis, P. J. and Michel, A. N. Linear Systems. New York: McGraw-Hill,
1997.
Zheng, D. Z. Linear Systems Theory (in Chinese). Beijing: Tsinghua
University Press, 1990.
Chen, B. M., Lin, Z. and Shamash, Y. Linear Systems Theory. Boston:
Birkhauser, 2004.
Arnold, V. I. Ordinary Differential Equations. Berlin: Springer International,
1994.
Borrelli, R. L. and Coleman, C. S. Differential Equations. New York: John
Wiley & Sons, 1998.
Hu, T. and Lin, Z. Control Systems With Actuator Saturation: Analysis And
Design. Boston, MA: Birhuser, 2001.
Edwards, C. and Spurgeon, S. Sliding Mode Control. London: Taylor &
Francis, 1998.
Isidori, A. Nonlinear Control Systems II. London: Springer-Verlag, 1999.
Elgerd, O. I. Electric Energy Systems Theory. New York: McGraw-Hill,
1971.
van der Pol, B. On relaxation oscillations, Phil. Mag., vol. 2, pp. 978992,
1926.
Lorenz, E. N. Deterministic nonperiodic ow, J. Atmos. Sci., vol. 20,
pp. 130141, 1963.
Atherton, D. P. Nonlinear Control Engineering. Workingham: Van Nostraind
Reinhold, 1975.

References
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]

[30]
[31]

[32]
[33]
[34]
[35]
[36]
[37]

[38]
[39]
[40]

269

Gelb, A. andVelde, W. E.V. Multi-Input Describing Functions and Nonlinear

System Design. New York: McGraw-Hill, 1963.
Lyapunov, A. M. The general problem of motion stability (translated in
french), Ann. Fac. Sci. Toulouse, vol. 9, pp. 203474, 1907.
Hahn, W. Stability of Motion. Berlin: Springer, 1967.
Narendra, K. S. and Annaswamy, A. M. Stable Adaptive Systems. Englewood
Cliffs, New Jersey: Prentice-Hall, 1989.
Marino, R. and Tomei, P. Nonlinear Control Design: Geometric, Adaptive,
and Robust. London: Prentice-Hall, 1995.
Narendra, K. S. and Taylor, J. H. Frequency Domain Stability for Absolute
Stability. New York: Academic Press, 1973.
Brown, J.W. and Churchill, R.V. ComplexVariables and Applications, 7th ed.
New York: McGraw-Hill, 2004.
Sontag, E. D. Smooth stabilization implies coprime factorization, IEEE
Transaction on Automatic Control, vol. 34, pp. 435443, 1989.
Sontag, E.D. and Wang, Y. On characterization of the input-to-state stability
properties, Syst. Contr. Lett., vol. 24, pp. 351359, 1995.
Sontag, E. and Teel, A. Changing supply functions input/state stable
systems, IEEE Trans. Automat. Control, vol. 40, no. 8, pp. 14761478,
1995.
Jiang, Z. -P., Teel, A. R. and Praly, L. Small-gain theorem for iss systems
and applications, Math. Contr. Sign. Syst., vol. 7, pp. 95120, 1994.
Ding, Z. Differential stability and design of reduced-order observers for
nonlinear systems, IET Control Theory and Applications, vol. 5, no. 2,
pp. 315322, 2011.
Boothby, W. A. An Introduction to Differential Manifolds and Riemaniann
Geometry. London: Academic Press, 1975.
Isidori, A. Nonlinear Control Systems, 3rd ed. Berlin: Springer-Verlag,
1995.
Brockett, R. W. Feedback invariants for non-linear systems, in
The Proceedings of 7th IFAC Congress, vol. 6, Helsinki, Finland.
Astrom, K. J. and Wittenmark, B. Adaptive Control, 2nd ed. Reading, MA:
Addison-Wesley, 1995.
Wellstead, P. E. and Zarrop, M. B. Self-Tuning Systems: Control and Signal
Processing. New York: John Wiley & Sons, 1991.
Ding, Z. Model reference adaptive control of dynamic feedback linearisable systems with unknown high frequency gain, IEE Proceedings Control
Theory and Applications, vol. 144, pp. 427434, 1997.
Parks, P. C. Lyapunov redesign of model reference adaptive control
systems, IEEE Trans. Automat. Contr., vol. 11, pp. 362367, 1966.
Ioannou, P. A. and Sun, J. Robust Adaptive Control. Upper Saddle River,
New Jersey: Prentice Hall, 1996.
Ding, Z. Robust adaptive control of nonlinear output-feedback systems
under bounded disturbances, IEE Proc. Control Theory Appl., vol. 145,
pp. 323329, 1998.

270

Nonlinear and adaptive control systems

[41]

[42]

[43]

[44]

[45]

[46]
[47]
[48]
[49]
[50]

[51]

[52]
[53]

[54]
[55]

[56]
[57]
[58]

Ding, Z. Global adaptive output feedback stabilization of nonlinear systems

of any relative degree with unknown high frequency gain, IEEE Trans.
Automat. Control, vol. 43, pp. 14421446, 1998.
Ding, Z. Adaptive control of nonlinear systems with unknown virtual control coefcients, Int. J. Adaptive Control Signal Proc., vol. 14, no. 5,
pp. 505517, 2000.
Ding, Z. Analysis and design of robust adaptive control for nonlinear output
feedback systems under disturbances with unknown bounds, IEE Proc.
Control Theory Appl., vol. 147, no. 6, pp. 655663, 2000.
Ding, Z. and Ye, X. A at-zone modication for robust adaptive control
of nonlinear output feedback systems with unknown high frequency gains,
IEEE Trans. Automat. Control, vol. 47, no. 2, pp. 358363, 2002.
Ding, Z. Adaptive stabilization of a class of nonlinear systems with unstable internal dynamics, IEEE Trans. Automat. Control, vol. 48, no. 10,
pp. 17881792, 2003.
Thau, F. Observing the states of nonlinear dynamical systems, Int. J.
Control, vol. 18, pp. 471479, 1973.
Kou, S., Ellitt, D. and Tarn, T. Exponential observers for nonlinear
dynamical systems, Inf. Control, vol. 29, pp. 204216, 1975.
Krener, A. J. and Isidori, A. Linearization by output injection and nonlinear
observers, Syst. Control Lett., vol. 3, pp. 4752, 1983.
Kazantzis, M. and Kravaris, C. Nonlinear observer design using lyapunovs
auxiliary theorem, Syst. Control Lett., vol. 34, pp. 241247, 1998.
Xiao, M. The global existence of nonlinear observers with linear error
dynamics: A topological point of view, Syst. Control Lett., vol. 55, no. 10,
pp. 849858, 2006.
Ding, Z. Observer design in convergent series for a class of nonlinear
systems, IEEE Trans. Automat. Control, vol. 57, no. 7, pp. 18491854,
2012.
Rajamani, R. Observers for Lipschitz nonlinear systems, IEEE Trans.
Automat. Control, vol. 43, no. 3, pp. 397401, 1998.
Zhu, F. and Han, Z. A note on observer design for Lipschitz nonlinear
systems, IEEE Trans. Automat. Control, vol. 47, no. 10, pp. 17511754,
2002.
Zhao, Y., Tao, J. and Shi, N. A note on observer design for one-sided
Lipschitz nonlinear systems, Syst. Control Lett., vol. 59, pp. 6671, 2010.
Ding, Z. Asymptotic rejection of unmatched general periodic disturbances
with nonlinear Lipschitz internal model, Int. J. Control, vol. 86, no. 2,
pp. 210221, 2013.
Luenberger, D. G. Observing the state of a linear system, IEEE Trans. Mil.
Electron., vol. 8, pp. 7480, 1964.
Hou, M. and Muller, P. Design of observers for linear systems with unknown
inputs, IEEE Trans. Automat. Control, vol. 37, no. 6, pp. 871875, 1992.
Ding, Z. Global output feedback stabilization of nonlinear systems with
nonlinearity of unmeasured states, IEEE Trans. Automat. Control, vol. 54,
no. 5, pp. 11171122, 2009.

References
[59]

[60]

[61]
[62]

[63]

[64]

[65]

[66]
[67]
[68]

[69]

[70]

[71]
[72]

[73]

[74]
[75]

271

Bastin, G. and Gevers, M. Stable adaptive observers for nonlinear time

varying systems, IEEE Trans. Automat. Control, vol. 33, no. 7, pp. 650658,
1988.
Cho, Y. M. and Rajamani, R. A systematic approach to adaptive observer
synthesis for nonlinear systems, IEEE Trans. Automat. Control, vol. 42,
no. 4, pp. 534537, 1997.
Tsinias, J., Sufcient lyapunov-like conditions for stabilization, Math.
Control Signals Systems, vol. 2, pp. 343357, 1989.
Kanellakopoulos, I., Kokotovic, P. V. and Morse, A. S. Systematic design
of adaptive controllers for feedback linearizable systems, IEEE Trans.
Automat. Control, vol. 36, pp. 12411253, 1991.
Marino, R. and Tomei, P. Global adaptive output feedback control of nonlinear systems, part i: Linear parameterization, IEEE Trans. Automat. Control,
vol. 38, pp. 1732, 1993.
Krstic, M., Kanellakopoulos, I. and Kokotovic, P. V. Adaptive nonlinear
control without overparametrization, Syst. Control Lett., vol. 19, pp. 177
185, 1992.
Ding, Z. Adaptive output regulation of class of nonlinear systems with
completely unknown parameters, in Proceedings of 2003American Control
Conference, Denver, CO, 2003, pp. 15661571.
Krstic, M., Kanellakopoulos, I. and Kokotovic, P. V. Nonlinear and Adaptive
Control Design. New York: John Wiley & Sons, 1995.
Ding, Z. Almost disturbance decoupling of uncertain output feedback
systems, IEE Proc. Control Theory Appl., vol. 146, pp. 220226, 1999.
Ding, Z. Adaptive control of triangular systems with nonlinear parameterization, IEEE Trans. Automat. Control, vol. 46, no. 12, pp. 19631968,
2001.
Bodson, M., Sacks, A. and Khosla, P. Harmonic generation in adaptive
feedforward cancellation schemes, IEEE Trans. Automat. Control, vol. 39,
no. 9, pp. 19391944, 1994.
Bodson, M. and Douglas, S. C. Adaptive algorithms for the rejection of
sinusoidal disturbances with unknown frequencies, Automatica, vol. 33,
no. 10, pp. 22132221, 1997.
Ding, Z. Global output regulation of uncertain nonlinear systems with
exogenous signals, Automatica, vol. 37, pp. 113119, 2001.
Ding, Z. Global stabilization and disturbance suppression of a class of
nonlinear systems with uncertain internal model,Automatica, vol. 39, no. 3,
pp. 471479, 2003.
Davison, E. J. The robust control of a servomechanism problem for linear time-invariant multivariable systems, IEEE Trans. Automat. Control,
vol. 21, no. 1, pp. 2534, 1976.
Francis, B. A. The linear multivariable regulator problem, SIAM J. Control
Optimiz., vol. 15, pp. 486505, 1977.
Huang, J. and Rugh, W. J. On a nonlinear multivariable servomechanism
problem, Automatica, vol. 26, no. 6, pp. 963972, 1990.

272

Nonlinear and adaptive control systems

[76]
[77]
[78]
[79]
[80]

[81]

[82]

[83]
[84]

[85]

[86]

[87]

[88]

[89]

[90]
[91]
[92]

Isidori, A. and Byrnes, C. I. Output regulation of nonlinear systems, IEEE

Trans. Automat. Control, vol. 35, no. 2, pp. 131140, 1990.
Huang, J. Nonlinear Output Regulation Theory and Applications. Philadelphia, PA: SIAM, 2004.
Nikiforov, V. O. Adaptive non-linear tracking with complete compensation
of unknown disturbances, Eur. J. Control, vol. 4, pp. 132139, 1998.
Nussbaum, R. D. Some remarks on a conjecture in parameter adaptive
control, Syst. Control Lett., vol. 3, pp. 243246, 1983.
Ye, X. and Ding, Z. Robust tracking control of uncertain nonlinear systems
with unknown control directions, Syst. Control Lett., vol. 42, pp. 110,
2001.
Ding, Z. Output regulation of uncertain nonlinear systems with nonlinear exosystems, in Proceeding of the 2004 American Control Conference,
Boston, MA, 2004, pp. 36773682.
Ding, Z. Output regulation of uncertain nonlinear systems with nonlinear
exosystems, IEEE Trans. Automat. Control, vol. 51, no. 3, pp. 498503,
2006.
Chen, Z. and Huang, J. Robust output regulation with nonlinear exosystems, Automatica, vol. 41, pp. 14471454, 2005.
Xi, Z. and Ding, Z. Global adaptive output regulation of a class of
nonlinear systems with nonlinear exosystems, Automatica, vol. 43, no. 1,
pp. 143149, 2007.
Ding, Z. Asymptotic rejection of general periodic disturbances in outputfeedback nonlinear systems, IEEE Trans. Automat. Control, vol. 51, no. 2,
pp. 303308, 2006.
Ding, Z. Asymptotic rejection of asymmetric periodic disturbances
in output-feedback nonlinear systems, Automatica, vol. 43, no. 3,
pp. 555561, 2007.
Ding, Z. Asymptotic rejection of unmatched general periodic disturbances
in a class of nonlinear systems, IET Control Theory Appl., vol. 2, no. 4,
pp. 269276, 2008.
Ding, Z. Observer design of Lipschitz output nonlinearity with application
to asymptotic rejection of general periodic disturbances, in The Proceedings
of The 18th IFAC Congress, vol. 8, Milan, Italy, 2011.
Ding, Z. Asymptotic rejection of nite frequency modes of general periodic
disturbances in output-feedback nonlinear systems, Automatica, vol. 44,
no. 9, pp. 23172325, 2008.
Arrillaga, J., Watson, N. R. and Chen, S. Power System Quality Assessment.
Chichester: John Wiley & Sons, 2000.
Wakileh, G. J. Power Systems Harmonics. Berlin: Springer-Verlag,
2010.
Takahashi, J., Vitaterna, M., King, D., Chang, A., Kornhauser, J., Lowrey, P.,
McDonald, J., Dove, W., Pinto, L., and Turek, F. Mutagenesis and mapping
of a mouse gene, clock, essential for circadian behaviour, Science, vol. 264,
pp. 719725, 1994.

References

273

[93] Albrecht, U., Sun, Z., Eichele, G. and Lee, C. A differential response of two
putative mammalian circadian regulators, mper1 and mper2 to light, Cell
Press, vol. 91, pp. 10551064, 1997.
[94] van der Horst, G., Muijtjens, M., Kobayashi, K., Takano, R., Kanno,
S., Takao, M., de Wit, J., Verkerk, A., Eker, A., van Leenen, D., Buijs,
R., Bootsma, D., Hoeijmakers, J., and Yasui, A. Mammalian cry1 and
cry2 are essential for maintenance of circadian rhythms, Nature, vol. 398,
pp. 627630, 1999.
[95] Bunger, M., Wilsbacher, L., Moran, S., Clendenin, C., Radcliffe, L.,
Hogenesch, J., Simon, M., Takahashi, J., and Bradeld, C., Mop3 is an
essential component of the master circadian pacemaker in mammals, Cell,
vol. 103, pp. 10091017, 2000.
[96] Sack, R., Auckley, D., Auger, R., Carskadon, M., Wright, K., Vitiello, M.,
and Zhdanova, I. Circadian rhythm sleep disorders: part i, basic principles,
shift work and jet lag disorders, Sleep, vol. 30, pp. 14601483, 2007.
[97] Sack, R., Auckley, D., Auger, R., Carskadon, M., Wright, K., Vitiello, M.,
and Zhdanova, I. Circadian rhythm sleep disorders: Part ii, advanced sleep
phase disorder, delayed sleep phase disorder, free-running disorder, and
irregular sleep-wake rhythm, Sleep, vol. 30, pp. 14841501, 2007.
[98] Boulos, Z., Macchi, M., Sturchler, M., Stewart, K., Brainard, G., Suhner, A.,
Wallace, G., and Steffen, R., Light visor treatment for jet lag after west
ward travel across six time zones, Aviat. Space Environ. Med., vol. 73,
pp. 953963, 2002.
[99] Kurosawa, G. and Goldbeter, A. Amplitude of circadian oscillations
entrained by 24-h light-dark cycles, J. Theor. Biol., vol. 242, pp. 478488,
2006.
[100] Geier, F., Becker-Weimann, S., Kramer, K., and Herzel, H. Entrainment in
a model of the mammalian circadian oscillator, J. Biol. Rhythms, vol. 20,
pp. 8393, 2005.
[101] Gonze, D., Leloup, J., and Goldbeter, A., Theoretical models for circadian
rhythms in neurospora and drosophila, Comptes Rendus de lAcademie des
Sciences -Series III -Sciences de la Vie, vol. 323, pp. 5767, 2000.
[102] That, L. T. and Ding, Z. Circadian phase re-setting using nonlinear outputfeedback control, J. Biol. Syst., vol. 20, no. 1, pp. 119, 2012.
[103] That, L. T. and Ding, Z. Reduced-order observer design of multi-output
nonlinear systems with application to a circadian model, Trans. Inst. Meas.
Control, to appear, 2013.
[104] Chen, T. and Francis, B. Optimal Sampled-Data Control Systems. London:
Springer-Verlag, 1995.
[105] Nesic, D. and Teel, A. A framework for stabilization of nonlinear sampleddata systems based on their approximate discrete-time models, IEEE Trans.
Automat. Control, vol. 49, pp. 11031122, 2004.
[106] Grizzle, J. W. and Kokotovic, P. V. Feedback linearization of sampled-data
systems, IEEE Trans. Automat. Control, vol. 33, pp. 857859, 1988.

274

Nonlinear and adaptive control systems

[107]

[108]

[109]

[110]

[111]

[112]
[113]

[114]

[115]

Nesic, D., Teel, A., and Kokotovic, P.V. Sufcient conditions for stabilization of sampled-data nonlinear systems via discrete-time approximations,
Syst. Control Lett., vol. 38, pp. 259270, 1999.
Nesic, D. and Laila, D. A note on input-to-state stabilization for nonlinear
sampled-data systems, IEEE Trans. Automat. Control, vol. 47, pp. 1153
1158, 2002.
Owens, D. H., Zheng, Y. and Billings, S. A. Fast sampling and stability of
nonlinear sampled-data systems: Part 1. Existing theorems, IMA J. Math.
Control Inf., vol. 7, pp. 111, 1990.
Laila, D. S., Nesic, D. and Teel, A. R. Open-and closed-loop dissipation inequalities under sampling and controller emulation, Eur. J. Control,
vol. 18, pp. 109125, 2002.
Shim, H. and Teel, A. R. Asymptotic controllability and observability
imply semiglobal practical asymptotic stabilizability by sampled-data output
feedback, Automatica, vol. 39, pp. 441454, 2003.
Bian, W. and French, M. General fast sampling theorems for nonlinear
systems, Syst. Control Lett., vol. 54, pp. 10371050, 2005.
Wu, B. and Ding, Z. Asymptotic stablilisation of a class of nonlinear systems
via sampled-data output feedback control, Int. J. Control, vol. 82, no. 9,
pp. 17381746, 2009.
Wu, B. and Ding, Z. Practical disturbance rejection of a class of nonlinear systems via sampled output, J. Control Theory Appl., vol. 8, no. 3,
pp. 382389, 2010.
Wu, B. and Ding, Z. Sampled-data adaptive control of a class of nonlinear
systems, Int. J. Adapt. Control Signal Process., vol. 25, pp. 10501060,
2011.

Index

-modication, 106
absolute stability, 60
adaptive control
direct, 89
indirect, 89
model reference, 89
nonlinear, 159
adaptive law
linear system, 101
nonlinear, 162, 172
robust, 107
asymptotic stability, 43
autonomous system, 11
backstepping
adaptive, 159
adaptive observer, 167
ltered transformation, 152
integrator, 141
iterative, 144
observer, 147
Barbalats lemma, 92
centre, 14
certainty equivalence principle, 90
chaos, 4, 21
circadian disorders, 238
circadian rhythm, 239
Neurospora control input, 246
Neurospora model, 239
circle criterion, 62
comparison functions
class K 66
class KL 66
class K , 66

comparison lemma, 50
dead zone, 31
dead-zone modication, 105
describing function, 27
dead zone, 32
ideal relay, 30
relay with hysteresis, 32
saturation, 30
stability of limit cycle, 35
detectability
linear system, 110
differential stability, 71
distribution, 84
involutive, 84
disturbance
bounded, 104, 107
equivalent input, 17677
general periodic, 204, 205
dynamic model, 207
generated from nonlinear
exosystem, 194
sinusoidal, 175
disturbance rejection
general periodic, 204
sinusoidal, 175
eigenvalue
resonant, 121
equivalent input disturbance, 17677
exosystem, 175, 176
linear, 176, 186
nonlinear, 195
exponential stability, 43
feedback linearisation, 76
feedforward

276

Nonlinear and adaptive control systems

control, 181, 189, 199

input, 176, 189, 206
term, 189, 197, 209
nite time convergence, 4
nite time escape, 4
frequency mode, 219, 224
asymptotically orthogonal, 225
asymptotically rejected, 235
multiple, 229
Frobenius theorem, 85
full state feedback linearisable, 85
global stability properties, 45
globally asymptotic stability, 45
globally exponential stability, 45
harmonics, 4
estimation, 235
power system, 219
high-frequency gain, 94, 100, 102, 187
Hurwitz
matrix, 52
polynomial, 94
transfer function, 57
vector, 148
ideal relay, 30
immersion, 189, 199
inter-sample behaviour, 252
internal model, 199
adaptive, 179, 190
nonlinear, 199, 201
observer based, 209
internal model principle, 176
invariant manifold, 177, 189, 196
invariant zero, 133, 134
ISS pair (, ), 69, 208
ISS stability, 66
ISS-Lyapunov function, 69

Lie derivative, 77
limit cycle, 4, 18, 34, 175, 195, 238, 240
linear observer error dynamics, 112,121
linearisation
around operation point, 10
feedback, 75, 76, 80
full state, 83, 85
linearised model, 10
Lipschitz constant, 9, 123, 207, 241
one-sided, 124, 241
Lorenz attractor, 21
Lyapunov function, 46
adaptive control, 101
adaptive observer, 139
ISS, 6770
linear systems, 52
Lyapunov stability, 42
Lyapunov theorem
global stability, 49
local stability, 47
Lyapunovs direct method, 46
Lyapunovs linearisation method, 45
model reference control
relative degree > 1, 9798
relative degree one, 9496
monic polynomial, 94
MRAC, 89
multiple equilibrium points, 4

Kalman Yakubovich lemma, 59

nonlinearity, 1, 3
dead zone, 31
hard, 3
ideal relay, 30
Lipschitz, 7, 122
multi-valued, 3
relay with hysteresis, 32
saturation, 29
sector-bounded, 59
normal form, 81
Nussbaum gain, 190

level set, 47, 49, 253

Lie bracket, 79

observability
linear system, 109

jet engine model, 260

Index
observer
general periodic disturbance, 207
observer canonical form
linear system, 118
observer design
adaptive, 137
direct state transformation, 121
Lipschitz nonlinearity, 122
output injection, 112
output nonlinearity, 126
reduced-order, 127
output feedback form, 148, 152
with disturbance, 175, 186, 194
with unknown parameter, 167
output feedback stabilisation, 158
output injection form, 112
state transformation, 114
output regulation, 186
adaptive, 186
nonlinear exosystem, 194
output-complement transformation, 127
pendulum model, 48
phase portrait, 12
Poincare-Bendixson theorem, 19
positive denite function, 46
positive limit set, 21
positive real transfer function, 55
strictly, 56
radially unbounded function, 49
relative degree, 77
relay with hysteresis, 32
robust adaptive law, 107
-modication, 106
dead-zone modication, 105

saddle point, 13
sampled-data
emulation method, 248
sampled-data controller, 250
saturation, 2, 29
sector-bounded nonlinearity, 59
self-tuning control, 89
semi-global stability, 3, 260
singular point, 12
small gain theorem
ISS, 70
stability
absolute, 60
asymptotic, 43
differential, 71
exponential, 43
globally asymptotic, 45
globally exponential, 45
ISS, 66
Lyapunov, 42
semi-global, 3, 260
stable focus, 1314
stable node, 13
STC, 89
transformation
ltered, 152, 187
output-complement, 127
tuning function, 161
unstable focus, 14
unstable node, 13
van der Pol oscillator, 18, 38
Youngs inequality, 128
zero dynamics, 81, 155

277

Control Engineering Series 84

An adaptive system for linear systems with unknown parameters is a

nonlinear system. The analysis of such adaptive systems requires similar
techniques to analyse nonlinear systems. Therefore it is natural to treat
adaptive control as a part of nonlinear control systems.
Nonlinear and Adaptive Control Systems treats nonlinear control
and adaptive control in a unified framework, presenting the major
results at a moderate mathematical level, suitable for MSc students
and engineers with undergraduate degrees. Topics covered include
introduction to nonlinear systems; state space models; describing
functions for common nonlinear components; stability theory; feedback
linearization; adaptive control; nonlinear observer design; backstepping
design; disturbance rejection and output regulation; and control
applications, including harmonic estimation and rejection in power
distribution systems, observer and control design for circadian rhythms,
and discrete-time implementation of continuous-time nonlinear control
laws.

Zhengtao Ding is a Senior Lecturer in Control

Engineering and Director for MSc in Advanced Control
and Systems Engineering at the Control Systems
Centre, School of Electrical and Electronic Engineering,
The University of Manchester, UK. His research interests
focus on nonlinear and adaptive control design. He
pioneered research in asymptotic rejection of general
periodic disturbances in nonlinear systems and
produced a series of results to systematically solve this
problem in various situations. He also made significant
contributions in output regulation and adaptive control
of nonlinear systems with some more recent results on
observer design and output feedback control as well.
Dr Ding has been teaching Nonlinear and Adaptive
Control Systems to MSc students for 9 years, and he
has accumulated tremendous experiences in explaining
difficult control concepts to students.

Nonlinear and Adaptive

Control Systems

Nonlinear and Adaptive

Control Systems

Nonlinear and
Adaptive Control
Systems

Ding

Zhengtao Ding

The Institution of Engineering and Technology

www.theiet.org
978-1-84919-574-4

Nonlinear and Adaptive Control.indd 1

04/03/2013 09:32:20