See discussions, stats, and author profiles for this publication at: [Link]

net/publication/393003170

Modern Control Systems Lecture Note

Book · June 2025

CITATIONS READS

0 135

1 author:

Daniel Abebe Beyene

Addis Ababa Science and Technology University
8 PUBLICATIONS 3 CITATIONS

SEE PROFILE

All content following this page was uploaded by Daniel Abebe Beyene on 07 August 2025.

The user has requested enhancement of the downloaded file.

Preface

This lecture note is a valuable resource for anyone seeking to understand the State Space
Representation of Linear Time-Invariant (LTI) Systems. It covers a wide range of topics,
including Mathematical Formulation of Control System Problems, Mathematical Model-
ing and State Space representation of Dynamic Systems, and State Space Representation
From Transfer Function.

The lecture note also provides a comprehensive review of Eigenvalues and Eigenvectors,
Solution of State Space Equations, Similarity Transformation, Phase Plane, and Stability
Analysis of LTI systems. These topics are crucial for understanding the behavior of LTI
systems.

In addition, the lecture note covers Controllability and Observability of LTI Systems,
Modality Tests, Kalman Decomposition, and Weaker Conditions for LTI Systems. It
also discusses Controllability and Observability in the Sense of Transfer Function. These
topics are essential for designing control systems that can achieve desired performance.

The lecture notes also include Full-state Feedback Controller Design for LTI Systems, Pole
Placement State Feedback Controller Design, Desired State Tracking, Servo-mechanism
Controller Design, and Determining Poles of Second Order LTI Systems From Time Do-
main Specification. These topics are crucial for designing controllers that can achieve
desired performance.

Furthermore, the lecture note covers State Estimators/Observers Design for LTI systems,
Full Order State Observer Design, Reduced Order State Observer Design, Continuous
Time Kalman Filtering, Closed Loop State Feedback Controller Design with Full State
Observer, and Closed Loop State Feedback Controller Design with Reduced Order State
Observer. These topics are essential for designing state estimators/observers that can
estimate the system’s state accurately.

The lecture note also provides an Introduction to Optimal Control, Performance Indices,
Elements of Calculus of Variation, Solution to Optimal Linear Quadratic Regulator Prob-
lem, Optimal Steady State Reference Tracking, and Optimal State Estimation. These
topics are crucial for designing optimal control systems that can achieve desired perfor-
mance.

In addition, the lecture notes cover Introduction to MIMO Systems and Nonlinear Sys-
tems, Introduction to Multiple Input Multiple Output Systems, and Introduction to
Nonlinear Systems. These topics are essential for understanding the behavior of complex
systems.

i
Finally, the lecture note includes Analog Simulation, Analog Computers and Simulation,
Solving Differential Equations with an Analog Computer, Physical Realization of Analog
Computer, Programming Differential Equation Using Analog Computer, Magnitude and
Time Scaling, Analog Simulation of Transfer Functions, and State Space representation
of Systems. These topics are crucial for understanding how analog computers can be
used to simulate complex systems.

Overall, this lecture note serves as an essential guide for anyone interested in understand-
ing the State Space Representation of linear time-invariant (LTI) Systems. It provides
a comprehensive overview of various topics related to LTI systems and is an excellent
resource for students and professionals alike.

In these lectures, the author has taken into consideration students who may not have
a strong background in state space representation and analysis for different systems.
Therefore, the author has deliberately avoided introducing basic concepts in state space
representation and state feedback control for linear systems whenever possible. While
some may view this as a drawback, the author hopes that users will not be too critical.

It is important to note that these are only introductory lectures and examples on the
topic. Due to time constraints and the availability of other books on the subject, the
author has chosen to focus on the most critical aspects and applications of the topic.
This decision was made to make the discussion more interesting from both an analytic
point of view.

The author hopes that this brief introduction will encourage interested students to ex-
plore more about this fascinating mathematical tool for analyzing and designing state
feedback controllers for different systems using state space representation. It is impor-
tant to note that there are many other books available on this topic for further reading,
such as ”Modern Control Engineering” by Katsuhiko Ogata [1] and ”Modern Control
Systems” by Richard C. Dorf and Robert H. Bishop [2].

Daniel Abebe, July 30, 2025

ii
Acronym’s (Abbreviation)

Abbreviation Description
DC Direct Current
BIBO Bounded Input Bounded Output
FBD Free Body Diagram
2DOF 2 Degree of Freedom
IPC Inverted Pendulum on Cart
LQR Linear Quadratic Regulator
LTI Linear Time Invariant
MATLAB Matrix Laboratory
MIMO Multiple Inputs and Multiple Outputs
SIMO Single Input and Multiple Outputs

iii
Contents

Preface i

Acronym’s (Abbreviation) iii

1 State Space Representation of LTI Systems 1

1.1 Mathematical Modeling of Dynamic Systems . . . . . . . . . . . . . . . . 1
1.2 Control System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Mathematical Formulation of Control System Problems . . . . . . . . . . 2
1.4 Mathematical Modeling of Electrical Systems . . . . . . . . . . . . . . . 4
1.4.1 Basic Laws Used in Modeling Electrical Systems . . . . . . . . . . 4
1.4.2 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.5 Mathematical Modeling of Magnetic Circuit . . . . . . . . . . . . . . . . 6
1.5.1 Magnetic Flux . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.5.2 Magnetomotive Force (mmf) . . . . . . . . . . . . . . . . . . . . . 6
1.5.3 Magnetic Reluctance . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.5.4 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.6 Mathematical Modeling of Mechanical Systems . . . . . . . . . . . . . . 7
1.6.1 Basic Laws Used in Modeling Mechanical Systems . . . . . . . . . 7
1.6.2 Components in Mechanical Systems Modeling . . . . . . . . . . . 8
1.6.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.7 Mathematical Modeling of Electro-mechanical Systems . . . . . . . . . . 23
1.7.1 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.8 Mathematical Modeling of Fluid Systems . . . . . . . . . . . . . . . . . . 27
1.8.1 Basic Laws Used in Fluid Systems Modeling . . . . . . . . . . . . 27
1.8.2 Key Fluid Properties . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.8.3 Flow Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.8.4 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.9 Mathematical Modeling of Chemical Process . . . . . . . . . . . . . . . . 32
1.9.1 Fundamental Laws for modeling chemical processes . . . . . . . . 33
1.9.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.10 State Space Representation From Transfer Function . . . . . . . . . . . . 42
1.11 Transfer Function From State Space Representation . . . . . . . . . . . . 46
1.12 Review on Eigenvalues and Eigenvectors . . . . . . . . . . . . . . . . . . 46
1.12.1 Relationships Between Right and Left Eigenvectors . . . . . . . . 47
1.12.2 Application of Eigen structure in Control Systems . . . . . . . . . 48
1.13 Solution of State Space Equations . . . . . . . . . . . . . . . . . . . . . . 48
1.14 Similarity Transformation . . . . . . . . . . . . . . . . . . . . . . . . . . 50
1.15 Phase Plane and Stability Analysis of LTI Systems . . . . . . . . . . . . 52

iv
 1.15.1 Stability analysis of LTI systems . . . . . . . . . . . . . . . . . . 52
1.15.2 Phase plane analysis of LTI systems . . . . . . . . . . . . . . . . . 53
1.16 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
1.17 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

2 Controllability and Observability of LTI Systems 76

2.1 Controllability of LTI Systems . . . . . . . . . . . . . . . . . . . . . . . . 76
2.2 Observability of LTI Systems . . . . . . . . . . . . . . . . . . . . . . . . 78
2.3 Modality Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
2.4 Kalman Decomposition of LTI Systems . . . . . . . . . . . . . . . . . . . 80
2.4.1 Weaker Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . 82
2.4.2 Transformation Matrix for Kalman Decomposition . . . . . . . . . 82
2.5 Controllability and Observability in a Sense of Transfer Function . . . . . 86
2.6 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
2.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

3 Full-state Feedback Controller Design for LTI Systems 97

3.1 Pole Placement State Feedback Controller Design . . . . . . . . . . . . . 98
3.2 Desired Input Tracking . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
3.3 Servo-mechanism Controller Design . . . . . . . . . . . . . . . . . . . . . 100
3.4 Determining Poles of Second Order LTI Systems From Time Domain Spec-
ification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
3.5 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
3.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

4 State Estimators/Observers Design for LTI systems 111

4.1 Full Order State Observer Design . . . . . . . . . . . . . . . . . . . . . . 111
4.2 Reduced Order State Observer Design . . . . . . . . . . . . . . . . . . . 113
4.3 Continuous Time Kalman Filtering . . . . . . . . . . . . . . . . . . . . . 116
4.4 Closed Loop State Feedback Controller Design with Full State Observer . 117
4.5 Closed Loop State Feedback Controller Design with Reduced Order State
Observer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
4.6 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
4.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138

5 Introduction to Optimal Control 140

5.1 Performance Indices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
5.1.1 Minimum time control problem . . . . . . . . . . . . . . . . . . . 140
5.1.2 Minimum terminal error control problem . . . . . . . . . . . . . . 141
5.1.3 Minimum effort/energy control problem . . . . . . . . . . . . . . 141
5.1.4 Optimal servo-mechanism or tracking problem . . . . . . . . . . . 141
5.1.5 Optimal regulator problem . . . . . . . . . . . . . . . . . . . . . . 142
5.2 Review on Elements of Calculus of Variation . . . . . . . . . . . . . . . . 142
5.2.1 Increment of a function and a functional . . . . . . . . . . . . . . 142
5.2.2 Differential and variation . . . . . . . . . . . . . . . . . . . . . . . 142
5.2.3 Optimization of a function and unconstrained functional . . . . . 143
5.2.4 Optimum of constrained functional . . . . . . . . . . . . . . . . . 146
5.3 Solution to Optimal Linear Quadratic Regulator Problem . . . . . . . . . 147
5.4 Optimal Steady State Reference Tracking . . . . . . . . . . . . . . . . . . 150

v
 5.5 Optimal State Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . 153
5.6 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
5.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164

6 Introduction to Nonlinear Systems and MIMO Systems 167

6.1 Introduction to Nonlinear Systems . . . . . . . . . . . . . . . . . . . . . 167
6.1.1 Determining equilibrium points for nonlinear dynamic systems . . 167
6.1.2 Linearization of nonlinear systems . . . . . . . . . . . . . . . . . . 168
6.2 Introduction to Multiple Input Multiple Output Systems . . . . . . . . . 176
6.2.1 Poles and zeros of MIMO Systems . . . . . . . . . . . . . . . . . . 176
6.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
6.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184

7 Analog Simulation 185

7.1 Analog Computers and Simulation . . . . . . . . . . . . . . . . . . . . . 185
7.2 Solving Differential Equations with an Analog Computer . . . . . . . . . 186
7.3 Physical Realization of Analog Computer . . . . . . . . . . . . . . . . . . 188
7.4 Programming Differential Equation Using Analog Computer . . . . . . . 190
7.5 Magnitude and Time Scaling . . . . . . . . . . . . . . . . . . . . . . . . . 192
7.5.1 Magnitude scaling . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
7.5.2 Time scaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
7.6 Analog Simulation of Transfer Functions and State Space Representation
of Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
7.6.1 Initial conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
7.7 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
7.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202

A Hamilton’s Principle and Lagrangian Mechanics i

A.1 Hamilton’s Principle (Generalized Classical Mechanics) . . . . . . . . . . i
A.1.1 What is the Lagrangian? . . . . . . . . . . . . . . . . . . . . . . . i
A.1.2 Properties of the Lagrangian . . . . . . . . . . . . . . . . . . . . . i
A.1.3 Connection to Newton’s Second Law . . . . . . . . . . . . . . . . ii
A.1.4 Why Use the Euler-Lagrange Method? . . . . . . . . . . . . . . . ii
A.2 Derivation of Euler-Lagrange Equations . . . . . . . . . . . . . . . . . . . ii
A.2.1 Functional vs Function . . . . . . . . . . . . . . . . . . . . . . . . ii
A.2.2 Increment of a Function and a Functional . . . . . . . . . . . . . ii
A.2.3 Differential of a Function . . . . . . . . . . . . . . . . . . . . . . . iii
A.3 Optimization of Functions and Functionals . . . . . . . . . . . . . . . . . iii
A.3.1 Optimum of a Function . . . . . . . . . . . . . . . . . . . . . . . iii
A.3.2 Variation of a Functional . . . . . . . . . . . . . . . . . . . . . . . iii
A.4 Optimization of an Unconstrained Functional . . . . . . . . . . . . . . . iv
A.4.1 Typical Functional in Control Systems . . . . . . . . . . . . . . . iv
A.4.2 Necessary Condition for Optimum . . . . . . . . . . . . . . . . . . iv
A.4.3 Sufficient Condition for Optimum . . . . . . . . . . . . . . . . . . v
A.5 Typical Functional in Control Systems . . . . . . . . . . . . . . . . . . . vi
A.6 Optimization of Constrained Functional . . . . . . . . . . . . . . . . . . vi
A.6.1 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . vi
A.6.2 General Function Definition . . . . . . . . . . . . . . . . . . . . . vii
A.6.3 Euler-Lagrangian Equation . . . . . . . . . . . . . . . . . . . . . . vii

vi
 A.7 Euler-Lagrangian Equation for systems with Generalized Forces . . . . . vii
A.8 Example:-1 Inverted Pendulum on a Cart System Dynamics . . . . . . . viii
A.8.1 System Description . . . . . . . . . . . . . . . . . . . . . . . . . . viii
A.8.2 Dynamic Model of the System . . . . . . . . . . . . . . . . . . . . viii
A.9 Example:-2 Ball and Beam System . . . . . . . . . . . . . . . . . . . . . xi
A.9.1 System Description . . . . . . . . . . . . . . . . . . . . . . . . . . xi
A.9.2 Dynamic Model of the System . . . . . . . . . . . . . . . . . . . . xi

B Modern Control System xiv

B.1 Modern Control System Tests . . . . . . . . . . . . . . . . . . . . . . . . xiv
B.1.1 Test 1 (2024 G.C.) . . . . . . . . . . . . . . . . . . . . . . . . . . xiv
B.1.2 Test 2 (2024 G.C.) . . . . . . . . . . . . . . . . . . . . . . . . . . xv
B.1.3 Test 3 (2020 G.C.) . . . . . . . . . . . . . . . . . . . . . . . . . . xv
B.1.4 Test 4 (2020 G.C.) . . . . . . . . . . . . . . . . . . . . . . . . . . xvi
B.2 Modern Control Systems Final Exams . . . . . . . . . . . . . . . . . . . xvii
B.2.1 Final Exam 1 (2020 G.C.) . . . . . . . . . . . . . . . . . . . . . . xvii
B.2.2 Final Exam 2 (2024 G.C.) . . . . . . . . . . . . . . . . . . . . . . xix
B.2.3 Final Exam 3 (2020 G.C.) . . . . . . . . . . . . . . . . . . . . . . xxi
B.2.4 Final Exam 4 (2021 G.C.) . . . . . . . . . . . . . . . . . . . . . . xxii

C Control Systems Exams xxiv

C.1 Control Systems Exams Test . . . . . . . . . . . . . . . . . . . . . . . . . xxiv
C.1.1 Test 1 (2025 G.C) . . . . . . . . . . . . . . . . . . . . . . . . . . . xxiv
C.1.2 Test 2 (2025 G.C) . . . . . . . . . . . . . . . . . . . . . . . . . . . xxvi
C.1.3 Test 3 (2025 G.C) . . . . . . . . . . . . . . . . . . . . . . . . . . . xxvii
C.2 Control Systems Exams Final . . . . . . . . . . . . . . . . . . . . . . . . xxviii
C.2.1 Final Exam 1 (2025 G.C.) . . . . . . . . . . . . . . . . . . . . . . xxviii
C.2.2 Final Exam 2 (2025 G.C.) . . . . . . . . . . . . . . . . . . . . . . xxix
C.2.3 Final Exam 3 (2025 G.C.) . . . . . . . . . . . . . . . . . . . . . . xxxi

D Digital Control Systems xxxii

D.1 Digital Control Systems Exams . . . . . . . . . . . . . . . . . . . . . . . xxxii
D.1.1 Final Exam 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxxii
D.1.2 Final Exam 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxxiv
D.1.3 Final Exam 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxxiv

vii
Chapter 1

State Space Representation of LTI

Systems

1.1 Mathematical Modeling of Dynamic Systems

Mathematical modeling is a technique for representing dynamic systems through math-
ematical expressions, simplifying complex systems to facilitate analysis, prediction, and
control of their behavior. These models describe time-dependent processes, aiding in un-
derstanding transient responses, such as changes from one steady state to another. They
provide insights into system dynamics by characterizing time constants and behaviors.
Additionally, mathematical models are instrumental in developing training simulators
for operator education and support ”what-if” analyses, allowing exploration of scenarios,
such as the effects of closing a valve, using tools like MATLAB or LabVIEW. They are
essential for optimization and control tasks, including control structure design, controller
tuning, and model-based control strategies. Ultimately, models serve as abstractions that
encapsulate the behavior of systems and can manifest in various forms, including

• Empirical Models: These models are constructed based on observed data rather
than derived from first principles. They often employ statistical methods to fit
data, making them effective for predicting system behavior within the range of
observed conditions.

• Mathematical Models: These models utilize mathematical language to describe the

relationships among system variables. They can be further categorized into:

– Differential Equations: Applicable to continuous systems, where the state of

the system evolves.
– Difference Equations: Relevant for discrete systems, where changes occur at
specific intervals.

• Hybrid Models: These models incorporate both continuous and discrete elements,
making them suitable for systems that exhibit both behaviors.

In this course, we will concentrate on continuous-time systems, primarily modeling them

through differential equations. The mathematical representation may take the form of
a transfer function, which establishes the relationship between the output and input in
the Laplace domain as the ratio of their respective Laplace transforms, assuming zero
initial conditions. Alternatively, we will explore state space representation, which focuses

1
Daniel Abebe Modern Control Systems Lecture Note

on the internal state of the system and employs a set of first-order differential equations.
While you have previously covered analysis and controller design using transfer function
models in classical control systems, this course will specifically emphasize analysis and
controller design within the state space framework.

1.2 Control System

A control system can be viewed informally as a dynamical object (e.g., expressed by
an ordinary differential equation) containing a parameter that can be manipulated to
influence the behavior of the system to achieve the desired goal.

A system is a thing having parts that can be considered as a single entity. The system
must be defined clearly with a particular set of relationships that completely characterize
the conceivable states of the system.

The state of the system describes the condition(status) of the system at any time t ≥ 0
and provides information which (together with the knowledge of the equation describing
the system) enables us to calculate/analyze the future behavior of the system under
certain control input.

In practice, it is not possible to get all state variables; instead, we get controlled variables
(outputs) which depend on someway on measured states.

1.3 Mathematical Formulation of Control System Problems

We assume that the dynamics of a system are an evolution of its states over time under
a given input. The evolution of the state of the system can be viewed as a function of
the state, control input, and time. Hence, in the evolution of a system, we have system
states, a state mapping function, control inputs, control input mapping function. The
state mapping function gives us the effect of the status of the states on the dynamics of
a system, while the input mapping function gives us the effect of control input on the
dynamics of a system.
Mathematically

ẋ = F (x(t), u(t), t) (1.3.1)

Equation 1.3.1 is called the state equation of a dynamic system. The function F (x(t), u(t), t)
can be linear or nonlinear, it can be separable or not(i.e it may or may not written as
sum of input mapping function and state mapping function explicitly as F (x(t), u(t), t) =
f (x(t), t) + g(u(t), t)).

Since all states of the system may not be accessible directly, we define the output (acces-
sible states) of a dynamic system as

y(t) = h(x(t), u(t), t) (1.3.2)

Equation 1.3.2 is called the output equation of a dynamic system. y(t) is an output
(accessible states/controlled states) of a dynamic system, and h(x(t), u(t), t) is an output
mapping function. The function h(x(t), u(t), t) defines the relation between the output

Chapter 1 2
Daniel Abebe Modern Control Systems Lecture Note

of the dynamic system, the state of the system, and the control input.

Thus the control system problem is finding admissible (inputs that we can apply to the
dynamic system) control input u(t) that derive the evolution of the dynamic system in
the desired way or toward the desired state, by using the relation between the state
dynamics and control input as well as the relation between the controlled states (output)
of the system and the control input.

In this course, we only deal with linear time-invariant (LTI) dynamic systems. LTI
dynamic systems have linear time-invariant state mapping and input mapping functions.
This implies that the evolution of LTI dynamic systems is linear concerning the state of
the system and control input. Which simplifies the state equation 1.3.1

ẋ(t) = Ax(t) + Bu(t) (1.3.3)

Matrices A, B are proportionality constant matrices. Matrix B is called the system

matrix, and matrix B is called the input matrix. Their size is defined based on the
number of states and inputs a dynamic system has. For a dynamic system with n states
and m control inputs, matrix A is n × n matrix, while matrix B is an n × m matrix.
The output equation 1.3.2 is simplified to

y(t) = Cx(t) + Du(t) (1.3.4)

for LTI dynamic systems. Matrices C, D are proportionality constant matrices. Matrix
C is called an output matrix, and matrix D is called an input transmission matrix. Their
size is defined based on the number of states, outputs, and inputs a dynamic system
has. For a dynamic system with n states, p number of outputs, and m number of control
inputs, matrix C is a p × n matrix, while matrix D is a p × m matrix.

Any LTI dynamic system can be represented by a state equation and output equation,
which is called the state-space representation of LTI dynamic systems.

ẋ(t) =Ax(t) + Bu(t) (1.3.5)

y(t) =Cx(t) + Du(t) (1.3.6)

For simplicity of expressions and convenience in the rest of the document, any function
of time f (t) is written as f , unless expressed explicitly. With this, the state space
representation of LTI dynamic systems is written as:

ẋ =Ax + Bu (1.3.7)
y =Cx + Du (1.3.8)

The block diagram of state space representation is shown in Figure 1.1.

Chapter 1 3
Daniel Abebe Modern Control Systems Lecture Note

D
u x + y
+
B ʃ C +
+

A
Figure 1.1: State Space Representation of a Dynamic System in Block Diagram

1.4 Mathematical Modeling of Electrical Systems

Mathematical modeling of electrical systems often relies on fundamental laws such as
Ohm’s Law, Kirchhoff’s Laws, and the principles of energy conservation.

1.4.1 Basic Laws Used in Modeling Electrical Systems

• Ohm’s Law: Relates voltage (vR ), current (iR ), and resistance (R) in a linear
circuit:

vR = iR R (1.4.1)

• Kirchhoff ’s Current Law (KCL): States that the total current entering a junc-
tion equals the total current leaving the junction.

• Kirchhoff ’s Voltage Law (KVL): States that the sum of the electrical potential
differences (voltage) around any closed network is zero..

• Inductor Voltage-Current Relation: The voltage across an inductor (vL ) is

proportional to the rate of change of current (iL ) through it:

diL
vL = L (1.4.2)
dt
where L is the inductance of an inductor.

• Capacitor Voltage-Current Relation: The current through a capacitor (iC ) is

proportional to the rate of change of voltage (vc ) across it:

dvC
iC = C (1.4.3)
dt
where C is the capacitance of a capacitor.

Chapter 1 4
Daniel Abebe Modern Control Systems Lecture Note

1.4.2 Example
RLC Circuit
Consider the series RLC circuit in Figure 1.2, if the concern of the designer is the voltage
across the capacitor, From Kirchhoff’s current law

R L

vs C vo

Figure 1.2: Series RLC Circuit

d
vs = iR + L i + vc
dt
d R 1 1
i = − i − vc + vs (1.4.4)
dt L L L
d
C vc = i
dt
d 1
vc = i (1.4.5)
dt C
The state space representation will be
   
d R 1   1
 dt i  − L − L   i   L 
 d = 1  + vs (1.4.6)
vc 0 vc 0
dt C  
h i i
y= 0 1   (1.4.7)
vc

Chapter 1 5
Daniel Abebe Modern Control Systems Lecture Note

1.5 Mathematical Modeling of Magnetic Circuit

Magnetic circuits are analogous to electrical circuits and are used to analyze the behavior
of magnetic fields in various materials. The fundamental concepts of magnetic circuit
modeling include magnetic flux, magnetomotive force (MMF), and magnetic reluctance.

1.5.1 Magnetic Flux

Magnetic flux (Φ) is defined as the total magnetic field density (Bm ) passing through a
given area (A). It can be expressed mathematically as:

Φ = Bm · A (1.5.1)

where Bm is measured in teslas (T) and A in square meters (m2 ). The direction of the
magnetic flux is determined by the orientation of the magnetic field lines.

1.5.2 Magnetomotive Force (mmf )

The magnetomotive force (F ) is the driving force that produces magnetic flux in a mag-
netic circuit. It is analogous to voltage in an electrical circuit and is calculated as:

Fmmf = N · i (1.5.2)

where N is the number of turns in the coil and I is the current flowing through the coil,
measured in amperes (A).

1.5.3 Magnetic Reluctance

Magnetic reluctance (Rm ) is the opposition to the flow of magnetic flux in a magnetic
circuit, similar to resistance in an electrical circuit. It is defined as:
Fmmf
Rm = (1.5.3)
Φ
The unit of magnetic reluctance is ampere-turns per weber (A/Wb). The relationship
between MMF, magnetic flux, and reluctance can be expressed as:

Fmmf = Rm · Φ (1.5.4)

1.5.4 Example
Hinged Armature Type Relay
The hinged armature type relay in Figure 1.3 operates as an electromagnetic circuit
consisting of a coil (electromagnet), an armature (movable part), and a core. The strength
of this magnetic field is proportional to the current flowing through the coil (I) and the
number of turns in the coil (N ):
N ·i
Bm = µ ·
l
where Bm is the magnetic flux density, µ is the permeability of the core material, and l
is the length of the magnetic path.

Chapter 1 6
Daniel Abebe Modern Control Systems Lecture Note

Figure 1.3: Hinged armature type relay

The force (Fem ) exerted on the armature can be calculated using the magnetic flux and
the area of the armature:
B2 · A
Fem =
2µ0
where A is the area of the armature and µ0 is the permeability of free space.

The electromagnetic force is balanced by the spring force. When voltage is applied to
the coil, it generates a magnetic field that attracts the pivoted armature, which is held
away from the core by a spring. As the armature moves towards the core, it closes the
relay contacts, allowing current to flow through the connected circuit and activating the
device. When the current is removed, the magnetic field collapses, and the armature
returns to its original position by the spring, reopening the contacts and interrupting the
current flow.

1.6 Mathematical Modeling of Mechanical Systems

Mathematical modeling of mechanical systems involves the use of mathematical expres-
sions to represent the behavior and dynamics of mechanical components and systems.
These models are crucial for understanding, analyzing, and predicting the performance
of mechanical systems across various applications in engineering.

1.6.1 Basic Laws Used in Modeling Mechanical Systems

The modeling of mechanical systems is grounded in fundamental physical laws, which
provide the framework for understanding how forces and motions interact. The key laws
include:
• Newton’s Laws of Motion: These laws describe the relationship between the
motion of an object and the forces acting on it.
• Conservation of Energy: This principle states that energy cannot be created or
destroyed, only transformed from one form to another. In mechanical systems, this
often involves converting potential energy to kinetic energy and vice versa.
• Principles of Dynamics: These principles govern the motion of objects under the
influence of forces, including concepts such as inertia, momentum, and equilibrium.

Chapter 1 7
Daniel Abebe Modern Control Systems Lecture Note

1.6.2 Components in Mechanical Systems Modeling

In mechanical modeling, several key components are commonly used to represent the
behavior of systems:

Stiffness Elements (Springs)

Springs are mechanical components that store and release energy. The behavior of springs
is defined by Hooke’s Law, which states that the force or torque exerted by a spring is pro-
portional to its linear displacement (x) or angular displacement (θ) from its equilibrium
position:

F = −kx (1.6.1)
τ = −kθ (1.6.2)

where F and τ are the restoring force and restoring torque, respectively, k is the spring
constant (a measure of stiffness), and x and θ are the linear and angular displacements,
respectively. This relationship indicates that the force exerted by the spring opposes the

x 
k
k
F

(a) (b)

Figure 1.4: Stiffness elements

direction of displacement, which is critical in applications such as vibration isolation and

energy storage. For example, in automotive suspensions, springs absorb shocks from the
road, enhancing ride comfort.

If both ends are free to move or rotate, the equations describing the motion are:

F = −k(x1 − x2 ) (1.6.3)
τ = −k(θ1 − θ2 ) (1.6.4)

These equations account for the relative displacements between two points, which is

1
1
k
x1 k x2
F F 


(a) (b)

Figure 1.5: Bi-directional motion stiffness elements

important in systems where springs may be subjected to forces from multiple directions.

Chapter 1 8
Daniel Abebe Modern Control Systems Lecture Note

Friction Elements (Dampers)

Dampers are devices that dissipate energy, usually in the form of heat, and are used to
reduce oscillations in mechanical systems. The force exerted by a damper is proportional
to the velocity (v) of the moving component:

Fd = −dẋ (1.6.5)
τd = −dθ̇ (1.6.6)

where Fd and τd are the damping force and damping torque, respectively, b is the damping
coefficient, and ẋ and θ̇ are the linear and angular velocities. This relationship indicates
that the damping force opposes the motion, providing a stabilizing effect in dynamic
systems. Dampers are essential for controlling vibrations and ensuring system stability.
If both ends are free to move or rotate, the equations describing the motion are:

Fd = −d(ẋ1 − ẋ2 ) (1.6.7)

τd = −d(θ̇1 − θ̇2 ) (1.6.8)

x1 x2 1 2
b b

(a) (b)

Figure 1.6: Dampers: A schematic representation of damping mechanisms in mechanical sys-

tems.

These equations illustrate how dampers respond to relative velocities, which is crucial in
applications such as automotive shock absorbers, where they mitigate the effects of road
irregularities on vehicle stability and comfort.

Mechanical Transformers
Mechanical transformers, such as levers and gears, are devices that change the direction
or magnitude of forces and motion. These components are fundamental in mechanical
systems, allowing for the efficient transfer and modification of energy.

• Levers: A lever, as shown in Figure 1.7, amplifies an input force to lift a load. The
relationship between the input force (F1 ), output force (F2 ), input distance (l1 ),
and output distance (l2 ) can be described by the law of the lever:

F1 l1 = F2 l2 (1.6.9)

This equation indicates that the moments about the fulcrum are equal, allowing
for force amplification depending on the distances from the fulcrum. The longer the
input arm (l1 ), the smaller the input force needed to lift a given load. This principle
is utilized in various applications, such as crowbars and seesaws, demonstrating the
mechanical advantage achieved through lever systems.

Chapter 1 9
Daniel Abebe Modern Control Systems Lecture Note

l1 F1
F2 l2

Figure 1.7: Lever system

• Gears: Gears, as shown in Figure 1.8, transmit torque and change the speed of
rotation. The relationship between the input and output angular velocities (ω1 , ω2 )
and torques (τ1 , τ2 ) can be expressed as:
ω1 r2
= and τ1 ω1 = τ2 ω2 (1.6.10)
ω2 r1
where r1 and r2 are the number of teeth on the input and output gears (gear radius
ratio), respectively.

r1

 2  2
1 1

r2

Figure 1.8: Gear system

This relationship shows how gears can change speed and torque, allowing for me-
chanical advantage in various applications. For instance, in a bicycle, the gear
system enables different pedaling speeds and forces, optimizing performance based
on terrain.

Chapter 1 10
Daniel Abebe Modern Control Systems Lecture Note

1.6.3 Examples
Two Degree of Freedom Quarter Car Suspension System
A vehicle suspension system is a complex vibration system having multiple degrees of
freedom. The purpose of the suspension system is to isolate the vehicle body from the
road imperfections. Suspension systems serve a dual purpose

• Contributing to the vehicle’s road holding/handling and braking for good active
safety and driving pleasure, and

• Keeping vehicle occupants comfortable and reasonably well isolated from road noise,
bumps, and vibrations.

The system may be

• Passive suspension system, uses springs and dampers only, or

• Semi-active suspension systems include devices such as air springs and switchable
shock absorbers, various self-leveling solutions, as well as systems like hydro pneu-
matic, hydro elastic, and hydro-gas suspensions or

• Fully Active suspension system uses electronic monitoring of vehicle conditions,

coupled with the means to impact vehicle suspension and behavior in real time to
directly control the motion of the car.

Consider a fully active suspension system shown in Figure 1.9, the free body diagram of
the system is shown in Figure 1.10.

y
mc
y1
k2 d
mw
y2
k1
z
x
Figure 1.9: 2-DOF Quarter Car Suspension Model

Using Newton’s second law of motion, the force on the body of the car is

ΣFy = mc ÿ1
fk2 + fd + u = mc ÿ1
k2 (y2 − y1 ) + d(ẏ2 − ẏ1 ) + u = mc ÿ1
k2 k2 d d 1
ÿ1 = − y1 + y2 − ẏ1 + ẏ2 + u (1.6.11)
mc mc mc mc mc

Chapter 1 11
Daniel Abebe Modern Control Systems Lecture Note

fk2 u fd

mc mw
y1 y2

fk2 u fd fk1

Figure 1.10: FBD of 2-DOF Quarter Car Suspension Model

In the same way, the force on the wheel is

ΣFy = mw ÿ2
− fk2 − fd − u + fk1 = mw ÿ2
− k2 (y2 − y1 ) − d(ẏ2 − ẏ1 ) − u + k1 (z − y2 ) = mw ÿ2
k2 k1 + k2 d d k1 1
ÿ2 = y1 − y2 + ẏ1 − ẏ2 + z− u (1.6.12)
mw mw mw mw mw mw
Let y1 = x1 , ẏ1 = x2 , y2 = x3 and ẏ2 = x4 , the equations 1.6.11 and 1.6.12 simplify

ẋ1 =x2
k2 k2 d d 1
ẋ2 = − x1 + x3 − x2 + x4 + u
mc mc mc mc mc
ẋ3 =x4
k2 k1 + k2 d d k1 1
ẋ4 = x1 − x3 + x2 − x4 + z− u (1.6.13)
mw mw mw mw mw mw
The state space equation of the system is then
        
0 1 0 0 0
ẋ
 1   x1    0 
k d k d 1

  − 2 −
 2 


   
 0 

ẋ2   m mc mc mc  x 2

 mc  u + 
 
 = c  +  
z

(1.6.14)
  
ẋ3   0 0 0 1 
 
x3   0   0 
    
  
k2 d k1 + k2 d 
  
1   k1 
ẋ4 − − x4 −
mw mw mw mw mw mw

If the output h is the position of the body of the car concerning the road x1 , the output
equation is
 
x
 1
h i 
x2 
h= 1 0 0 0   
 (1.6.15)
x3 
 
x4

The goal of the designer is to design a controller that keeps the car body in a predefined
range, irrespective of the roughness of the road.

Chapter 1 12
Daniel Abebe Modern Control Systems Lecture Note

Wind Turbine Generator System

Wind turbine generator systems convert kinetic energy from the wind into electrical en-
ergy through a series of mechanical and electrical processes. Understanding the dynamics
of these systems is crucial for optimizing performance, predicting energy output, and en-
suring reliable operation. The mathematical modeling of a wind turbine involves several
key components, including the aerodynamic characteristics of the turbine blades, the me-
chanical dynamics of the rotor, and the electrical characteristics of the generator.

The relationship between the angular speeds and torques of the turbine and generator is
essential for assessing the efficiency and overall performance of the system.

In a typical configuration, the turbine’s angular speed (θ̇1 ) is associated with the aero-
dynamic torque (τaero ) generated by the wind acting on the blades. This torque can be
expressed as a function of wind speed, blade characteristics, and the turbine’s design
parameters. The turbine is generally coupled to the generator through a gearbox, which
adjusts the speed and torque to enhance generator performance.

The generator, in turn, possesses its angular speed (θ̇2 ) and produces a generator torque
r2
(τg ). The gearbox ratio (N = ) is critical in relating these parameters, where r1 and
r1
r2 are the radii of the turbine and generator gears, respectively.

1

wind
 aero r1
b

2

 g
fc

r2

Figure 1.11: Wind Turbine Generator System

The mathematical model of the system can be divided into two main components: the
turbine system and the generator system. The mathematical model for the turbine system
is given by:

τaero − bθ̇1 − fc r1 = J1 θ̈1 (1.6.16)

where fc represents the contact force between the gears, b1 is a damping coefficient, and
J1 is the moment of inertia of the turbine system.

Similarly, the mathematical model for the generator system is expressed as:

fc r2 − b2 θ̇2 − τg = J2 θ̈2 (1.6.17)

Chapter 1 13
Daniel Abebe Modern Control Systems Lecture Note

where J2 is the moment of inertia of the generator system, b2 is a damping coefficient

and τg is the torque produced by the generator.

To solve for fc from equation 1.6.17, we can substitute fc into equation 1.6.16:
1  
τaero − b1 θ̇1 − τg + b2 θ̇2 + J2 θ̈2 = J1 θ̈1 (1.6.18)
N

Using the relationship between the angular speeds, we have θ̇1 = N θ̇2 , where N is the
gearbox ratio. The final expression for τaero in terms of θ̇2 is:
   
b2 1 J2
τaero = b1 N + θ̇2 + τg + + J1 N θ̈2
N N N
b2 + b1 N 2
     
N 1
θ̈2 = τaero − τg − θ̇2 (1.6.19)
J2 + J1 N 2 J2 + J1 N 2 J2 + J1 N 2
b2 + b1 N 2
     
N 1
Let γ1 = , γ2 = , γ3 = and the states of system
J2 + J1 N 2 J2 + J1 N 2 J2 + J1 N 2
be x1 = θ2 and x2 = θ̇2 the position and velocity of the mass of the spool, then the sate
space representation will be
ẋ1 = x2
ẋ2 = −γ3 x2 + γ1 τaero − γ2 τg (1.6.20)
       
ẋ1 0 1 x1 0 0 τaero
 =   +    (1.6.21)
ẋ2 0 −γ3 x2 γ1 −γ2 τg
 
h i x1
y= 0 1   (1.6.22)
x2

Inverted Pendulum on a Cart (IPC) System

An inverted pendulum on a cart is a mechanical system having a freely rotating inverted
pendulum with mass mp and length l mounted on a cart with mass mc . The main goal
of the control engineer is to determine the control signal u that drives the cart to the
desired location and, at the same time, keeps the inverted pendulum from falling. The
system is shown in Figure 1.12a.
Let us denote the distance of the cart from the origin by (x, y) and its velocity by
(ẋ, ẏ). For simplicity, assume the cart moves along the x axis only; then the velocity and
position along the y axis are zero. If the pendulum deviation from the vertical axis is θ,
then the position of the pendulum (xp , yp ) with respect to the origin is given as
   
x x + l sin θ
 p =   (1.6.23)
yp l cos θ
With velocity
   
ẋp ẋ + θ̇ cos θ
 =  (1.6.24)
ẏp −lθ̇ sin θ

Chapter 1 14
Daniel Abebe Modern Control Systems Lecture Note

lsinθ lsinθ

y Jα

lcosθ
θ
θ

lcosθ
P N
mpg
N
P
x
u
x ff

(a) Inverted Pendulum on a Cart System (b) Free Body Diagram

Figure 1.12: Inverted Pendulum on a Cart (IPC) System

The free-body diagram of the IPC is shown in Figure 1.12b. If we denote the coefficient
of friction between the cart and the rail by b, the normal force exerted by the cart on the
pendulum by P , and the horizontal force exerted by the cart on the pendulum by N .

Using Newton’s second law of motion on a pendulum free body diagram, we have,
ΣFpx = mp ẍp
N = mp ẍp = mp (ẍ + θ̈l cos θ − θ̇2 l sin θ) (1.6.25)
ΣFpy = mp ÿp
P = mp g − mp θ̈l sin θ − mp lθ̇2 cosθ (1.6.26)
Using Newton’s second law of motion forces exerted on a cart are
ΣFcx = mc ẍ
u = mc ẍ + bẋ + N (1.6.27)
ΣFcy = mc ÿ = 0 (1.6.28)
Substituting for N in equation 1.6.27 we have,
u = (mc + mp )ẍ + bẋ + mp lθ̈ cos θ − mp lθ̇2 sin θ (1.6.29)
Assuming the friction between the pendulum and the cart is negligible, the momentum
of the pendulum rod about the pivot is described by,
Jα = J θ̈ = P l sin θ − N l cos θ (1.6.30)
Combining equation 1.6.25 , equation 1.6.26 and equation 1.6.30 we have
(J + mp l2 )θ̈ − mp lg sin θ = −mp lẍ cos θ (1.6.31)

Solving for ẍ and θ̈ in-terms of ẋ, x, θ̇, θ, we get the two governing dynamic equations for
IPC ,
(mp l2 + I)mp lθ̇2 sin θ − (mp l)2 g sin θ cos θ − b(mp l2 + J)ẋ + (mp l2 + I)u
ẍ = (1.6.32)
(mp + mc )(mp l2 + I) − (mp l cos θ)2
−(mp l)2 θ̇2 g sin θ cos θ + bẋmp l cos θ + (mp + mc )mp lg sin θ − mp l cos θu
θ̈ = (1.6.33)
(mp + mc )(mp l2 + I) − (mp l cos θ)2

Chapter 1 15
Daniel Abebe Modern Control Systems Lecture Note

Using small angle approximation about θ = 0 deg, we know that for small angle cos θ ≈ 1,
sin θ ≈ θ and θ̇ ≈ 0. Replacing cos θ = 1, sin θ = θ and θ̇ = 0 in equation 1.6.32 and
equation 1.6.33 we have
mp l 2 + J b(mp l2 + J) (mp l)2 g
     
ẍ = u− ẋ − θ (1.6.34)
q q q
     
bmp l mp lg(mp + mc ) mp l
θ̈ = ẋ + θ̇ − u (1.6.35)
q q q
where q = (mc + mp )(mp l2 + J) − (mp l)2 .

To represent equation 1.6.34 and equation 1.6.35 in state space let x1 = x, x2 = ẋ, x3 = θ
and x4 = θ̇. The state space representation of IPC is;
   
0 1 0 0 0
   
ẋ  x1
 1  2
(mp l)2 g     mp l2 + J 
      
  0 − b(mp l + J) − 0 x  
    
ẋ2   q q 2 q 
 =   +  u
      
ẋ3  0 0 0 1 x3   0 
             
ẋ4 bmp l mp lg(mp + mc ) mp l
0 x4
   
0 −
q q q
(1.6.36)
 
x
     1
 
y1 1 0 0 0 x2 
y=   =    
  (1.6.37)
y2 0 0 1 0 x3 
 
x4

Planar Drone
The planar quadrotor is a simplified model of a quadrotor drone that operates within
a two-dimensional plane, making it a valuable tool in control theory and robotics for
studying dynamics, control strategies, and navigation algorithms. This model features a
frame with four rotors positioned at the corners, each generating lift and capable of in-
dependent control. The quadrotor possesses two degrees of freedom in position, allowing
movement along the x and y axes, and one degree of freedom in orientation, enabling
rotation about the vertical axis (yaw).

Control inputs consist of thrust produced by each rotor, with differential thrust utilized
to manage rotation and directional movement. The dynamics of the planar quadrotor
can be described by equations derived from Newton’s laws or Lagrangian mechanics (see
appendix A), incorporating forces from thrust and moments from rotor speeds. These
equations govern its position, velocity, orientation, and angular velocity, accounting for
factors such as mass, acceleration due to gravity, and moment of inertia.

The planar quadrotor finds applications in various fields, including aerial robotics re-
search, flight dynamics simulation, and the development of control algorithms. Its sim-
plicity allows for easier analysis and the creation of control strategies that can be adapted

Chapter 1 16
Daniel Abebe Modern Control Systems Lecture Note

(a) Planar Drone (b) Planar drone FBD

Figure 1.13: Planar Drone

for more complex three-dimensional systems, making it an essential model for understand-
ing the principles of flight control in aerial vehicles.

The equations of motion are almost trivial, since it is only a single rigid body, and cer-
tainly fit into our standard manipulator equations:
mẍ = −mg sin θ (1.6.38)
mÿ = (u1 + u2 ) − mg cos θ (1.6.39)
I θ̈ = r(u1 − u2 ) (1.6.40)
To express the given equations in state-space representation let x1 = x, x2 = ẋ, x3 =
y, x4 = ẏ, x5 = θ, x6 = θ̇ and let F = u1 + u2 the lift force and τ = u1 − u2 the torque,
that cause the drone to pitch.
ẋ1 = x2
ẋ2 = −g sin x5
ẋ3 = x4
F
ẋ4 = − g cos x5
m
ẋ5 = x6
r
ẋ6 = τ (1.6.41)
I
Linearizing the given state-space representation using the small angle approximation, we
assume that the angle θ or x5 is small enough such that: cos x5 ≈ 1 and sin x5 ≈ x5 .
Using these approximations, we can rewrite the equations as follows:
ẋ1 = x2
ẋ2 = −gx5
ẋ3 = x4
F
ẋ4 = −g
m
ẋ5 = x6
r
ẋ6 = τ (1.6.42)
I

Chapter 1 17
Daniel Abebe Modern Control Systems Lecture Note

The state-space equation is:

     
0 1 0 0 0 0 0 0 0
     
0 0
0 0 0 0 −g 0 0
   
 
     
  0 0  
0 0 0 1 0 0  0
ẋ =  x +  1 u −   (1.6.43)
0
     
0 0 0 0 0 0 g 
m

    
     
0 0 0 0 0 1 0 0 0
r
     
0 0 0 0 0 0 0 0
I
The output equation is given by:
   
1 0 0 0 0 0 0 0
y= x +  u (1.6.44)
0 0 0 1 0 0 0 0

Ball and Beam System

The ball and beam system is a classic control theory problem that involves balancing a
ball on a beam, which can tilt to control the ball’s position. This system is widely used
in control engineering to study dynamics, stability, and control strategies. The system
consists of a beam that can rotate about its center and a ball that rolls along the beam.
The goal is to control the beam’s angle to regulate the ball’s position. The system has
two primary degrees of freedom: the position of the ball along the beam and the angular
position of the beam.

I,m
x




Figure 1.14: Ball and Beam System

Let I1 , m, and I2 denote the inertia of the beam, the mass of the ball, and the inertia
of the ball, respectively. The position and velocity of the center of mass of the ball are

xc = (x cos θ, x sin θ) (1.6.45)

ẋc = (ẋ cos θ − xθ̇ sin θ, ẋ sin θ + xθ̇ cos θ) (1.6.46)

Chapter 1 18
Daniel Abebe Modern Control Systems Lecture Note

The translational part of the kinetic energy is given by

1 1
Ktrans = mẋ2c = m(ẋ2 + x2 θ̇2 ) (1.6.47)
2 2
The rotational kinetic energy of the beam-and-ball system is

 2
1 1 ẋ
Krot = I1 θ̇2 + I2 (1.6.48)
2 2 r
The total kinetic energy of the system is
 
1 2 2 1 I2
K = Ktrans + Krot = (I1 + mx )θ̇ + m 1 + ẋ2 (1.6.49)
2 2 mr2
I2
Let λ = 1 + is a constant.
mr2
The potential energy of the system is
V (x, θ) = mgx sin θ (1.6.50)
The Lagrangian of the system is
L = K(x, θ) − V (x, θ)
 
1 2 2 1 I2
= (I1 + mx )θ̇ + m 1 + ẋ2 − mgx sin θ (1.6.51)
2 2 mr2
The only variable that can be actuated is the angle of the beam; the position of the ball
is an unactuated variable. Thus the Euler-Lagrangian equation (see appendix A) of the
systems is
 
d ∂L ∂L
− = Qi (1.6.52)
dt ∂ q̇ ∂q
where Qi represents the generalized forces or torques, including friction and external
inputs. Assuming that the pivot friction is negligible, the only externalforce
 applied to
τ
the system is the torque at the pivot joint τ . Therefore, we have Qi =  . By solving
0
the Euler-Lagrange equation, we obtain:
(I1 + mx2 )θ̈ + 2mxẋθ̇ + mgx cos θ = τ (1.6.53)
λẍ − xθ̇2 + g sin θ = 0 (1.6.54)

With state variables x1 = x, x2 = ẋ, x3 = θ, x4 = θ̇, the dynamics of the ball and beam
system can be described by the following set of nonlinear differential equations:
ẋ1 = x2
mx24 x1 − g sin x3
ẋ2 =
λ
ẋ3 = x4
τ − 2mx1 x2 x3 − mgx1 cos x3
ẋ4 = (1.6.55)
I + mx21

Chapter 1 19
Daniel Abebe Modern Control Systems Lecture Note

Using small angle approximation(x3 = 0) and the ball is at the center of the beam
(x1 = 0), which means (sin x3 ≈ x3 and cos x3 ≈ 1), the linearized state-space equations
become:

ẋ1 = x2
−gx3
ẋ2 =
λ
ẋ3 = x4
τ − mgx1
ẋ4 = (1.6.56)
I
The state-space representation of the system is:
   
0 1 0 0 0
 −g    
 0 0 0 0
  
ẋ =  λ x +  τ (1.6.57)
0
   
 0 0 0 1
−mg 1
   
0 0 0
I I
The control objective is to regulate the position of the ball; thus, the output must be x1 ,
which results in the output equation
h i
y= 1 0 0 0 x (1.6.58)

The Acrobot
The Acrobot is representative of the primary challenge in underactuated robots. In
order to swing up and balance the entire system, the controller must reason about and
exploit the state-dependent coupling between the actuated degree of freedom and the
unactuated degree of freedom. It is also an important system because, as we will see, it
closely resembles one of the simplest models of a walking robot.

Figure 1.15: Acrobot

Figure 1.15 illustrates the model parameters used in our analysis. θ1 is the shoulder
joint angle, θ2 is the elbow (relative) joint angle, and we will use q = [θ1 , θ2 ]⊤ , x = [q, q̇]⊤ .

Chapter 1 20
Daniel Abebe Modern Control Systems Lecture Note

The zero configuration is with both links pointed directly down. The moments of iner-
tia, I1 , I2 , are taken about the pivots. The task is to stabilize the unstable fixed point
x = [π, 0, 0, 0]⊤ .

We will derive the equations of motion for the Acrobot using the method of Lagrange.
The locations of the center of mass of each link, pc1 , pc2 , are given by the kinematics:
   
lc1 s1 l1 s1 + lc2 s1+2
pc1 =   , pc2 =  , (1.6.59)
−lc1 c1 −l1 c1 − lc2 c1+2

where s1 is shorthand for sin(θ1 ), c1+2 is shorthand for cos(θ1 + θ2 ), etc. The energy
is given by:

1
T = T1 + T2 , T1 = I1 q̇12 (1.6.60)
2

1 1
T2 = (m2 l12 + I2 + 2m2 l1 lc2 c2 )q̇12 + I2 q̇22 + (I2 + m2 l1 lc2 c2 )q̇1 q̇2 (1.6.61)
2 2

u = −m1 glc1 c1 − m2 g(l1 c1 + lc2 c1+2 ) (1.6.62)

Entering these quantities into the Lagrangian yields the equations of motion:

(I1 + I2 + m2 l12 + 2m2 l1 lc2 c2 )q̈1 + (I2 + m2 l1 lc2 c2 )q̈2 − 2m2 l1 lc2 s2 q̇1 q̇2 (1.6.63)

−m2 l1 lc2 s2 q̇22 + m1 glc1 s1 + m2 g(l1 s1 + lc2 s1+2 ) = τ1 (1.6.64)

(I2 + m2 l1 lc2 c2 )q̈1 + I2 q̈2 + m2 l1 lc2 s2 q̇12 + m2 glc2 s1+2 = τ2 (1.6.65)

In standard manipulator equation form:

M(q)q̈ + C(q, q̇)q̇ = τg (q) + Bu, (1.6.66)

using q = [θ1 , θ2 ]⊤ , u = τ , the mass matrix M(q) is given by:

 
2
I1 + I2 + m2 l1 + 2m2 l1 lc2 c2 I2 + m2 l1 lc2 c2
M(q) =  , (1.6.67)
I2 + m2 l1 lc2 c2 I2

the Coriolis and centrifugal matrix C(q, q̇) is given by:

 
−2m2 l1 lc2 s2 q̇2 −m2 l1 lc2 s2 q̇2
C(q, q̇) =  , (1.6.68)
m2 l1 lc2 s2 q̇1 0

Chapter 1 21
Daniel Abebe Modern Control Systems Lecture Note

the gravitational torque vector τg (q) is given by:

 
−m1 glc1 s1 − m2 g(l1 s1 + lc2 s1+2 )
τg (q) =  , (1.6.69)
−m2 glc2 s1+2

The input matrix B is given by:

 
0
B =  . (1.6.70)
1

The nonlinear equations of motion for the Acrobot are linearized around the equilib-
rium point q = [π, 0]⊤ and q̇ = [0, 0]⊤ . The linearization is performed using small-angle
approximations. Assume small deviations from the equilibrium point:
θ1 = π + ∆θ1 , θ2 = 0 + ∆θ2 , (1.6.71)
where ∆θ1 and ∆θ2 are small angles. For small angles, we use the following approxima-
tions:
sin(θ1 ) ≈ −∆θ1 , cos(θ1 ) ≈ −1, (1.6.72)

sin(θ1 + θ2 ) ≈ −∆θ1 + ∆θ2 , cos(θ1 + θ2 ) ≈ −1. (1.6.73)

The mass matrix M(q) is linearized as:
 
2
I1 + I2 + m2 l1 + 2m2 l1 lc2 I2 + m2 l1 lc2
M= . (1.6.74)
I2 + m2 l1 lc2 I2

The Coriolis matrix C(q, q̇) is linearized as:

 
0 0
C≈ . (1.6.75)
0 0

The linearized gravitational torque vector τg (q) is linearized as:

 
(m1 glc1 + m2 gl1 − m2 glc2 )∆θ1 + m2 glc2 ∆θ2
τg =  . (1.6.76)
−m2 glc2 ∆θ1 + m2 glc2 ∆θ2

The linearized equations of motion can be written in state-space form:

ẋ = Ax + Bu, (1.6.77)
Where:
 
∆θ1
   
 
∆θ2  τ1
x=
 ,
 u =  . (1.6.78)
∆θ̇1  τ2
 
∆θ̇2

Chapter 1 22
Daniel Abebe Modern Control Systems Lecture Note

The state matrix A and input matrix B are:

   
0 0 1 0 0 0
   
   
0 0 0 1 0 0
A= 
, B = 
 
,
 (1.6.79)
a31 a32 0 0 b b 
 31 32 
 
 
a41 a42 0 0 b41 b42

Where:
m1 glc1 + m2 gl1 − m2 glc2
a31 = ,
M11
m2 glc2 −m2 glc2 m2 glc2
a32 = , a41 = , a42 = ,
M11 M22 M22
1 1
b31 = , b32 = 0, b41 = 0, b42 = .
M11 M22

1.7 Mathematical Modeling of Electro-mechanical Systems

Electro-mechanical systems can be represented using differential equations that describe
the dynamic behavior of both electrical and mechanical components. A common example
of an electro-mechanical system is a solenoid actuator valve and a DC motor.

1.7.1 Example
Solenoid Actuator-Valve System
The 3/2 solenoid actuator valve, depicted in Figure 1.16a, plays a crucial role in regulating
fluid flow through three ports: one inlet and two outlets.

Spool Valve

Air gap sping

Seated Position Coil Supply Drain

- Fs
Fem m
Plunger
Fd
Drain

(a) Solenoid Actuator-Valve System (b) Free Body Diagram

Figure 1.16: Solenoid Actuator-Valve System

When an electrical current is applied to the solenoid, it generates a magnetic field

that actuates the internal mechanism of the valve. This movement facilitates fluid flow
from the inlet to one outlet while simultaneously blocking the other. Upon deactivation
of the current, a spring mechanism restores the valve to its original position, thereby
redirecting the flow to the second outlet. This efficient operation enables precise control

Chapter 1 23
Daniel Abebe Modern Control Systems Lecture Note

over actuators and cylinders in various pneumatic and hydraulic applications, establish-
ing the solenoid actuator valve as a vital component in automation and fluid management
systems.

Since the spool valve operates within a fluid medium, which introduces viscous friction
that influences its dynamic behavior. The free body diagram of the system, illustrated
in Figure 1.16b, elucidates the forces acting on the valve and provides insights into its
dynamics, particularly regarding the impact of fluid resistance on its operation.

Applying Newton’s second law of motion, the force acting on the spool valve can be
expressed as:

ΣF = Fem − Fs − Fd = mẍ
N i − kx − dẋ = mẍ (1.7.1)

where m is mass of the spool, x is displacement of the spool from its equilibrium position,
N is number of turns in the solenoid, i is the current flowing through the solenoid, k is
spring constant and d is damping coefficient.

Let the states of the system be x1 = x and x2 = ẋ, the position and velocity of the mass
of the spool. The state space representation will be

ẋ1 = x2
k d N
ẋ2 = − x1 − x2 + i (1.7.2)
m m m

      
ẋ 0 1 x 0
 1 =    1 +   i (1.7.3)
k d N
ẋ2 − − x2
m  m  m
h i x1
y= 1 0   (1.7.4)
x2

DC Motor System
Consider an armature voltage-controlled DC motor with armature resistance Ra and
armature inductance La used to drive a mechanical load with load torque τL having
rotational inertia J through a coupling gear with DC motor shaft gear to load side gear
rm
ratio at a speed ωL . Let the mechanical friction coefficient between the rotor and the
rL
load be b, as shown by the equivalent circuit in Figure 1.17. The control input to the DC
motor is armature voltage Va , and the output is the speed of the load.
The desired goal is to find Va that can derive the load at the required speed. To determine
Va , we have to know the relation between the speed of the DC motor ωm and armature
voltage Va , which, of course, needs the mathematical model of the DC motor.

From the electrical characteristics of the DC motor, the torque produced by the motor τm
is proportional to armature current ia , while the back emf Eb of the motor is proportional

Chapter 1 24
Daniel Abebe Modern Control Systems Lecture Note

Ra ia La
+ τm , ωm rm
Va M

- τL , ωL
Load

rL

Figure 1.17: Armature Voltage Controlled DC Motor with Mechanical Load

to the speed of the motor ωm , so we have

τm =kt ia (1.7.5)
Eb =kb ωm (1.7.6)

where kt and kb are torque and back emf proportionality constants, respectively.
Using Kirchhoff’s current law
d
Va = La ia + Ra ia + Eb (1.7.7)
dt
and from the mechanical analysis of the load side
d
τm = J ωL + bωL + τL (1.7.8)
dt
Replacing for Eb and τm in equation 1.7.7 and equation 1.7.8 respectively and rearranging;
d Ra kb 1
ia = − ia − ωm + Va (1.7.9)
dt La La La
d kt b 1
ωL = ia − ωL − τL (1.7.10)
dt J J J
From the relation between gear radius ratio and angular speed of the gears, we know that
rL
ωm = ωL (1.7.11)
rm
replacing for ωm in equation 1.7.9, we have
d Ra kb rL 1
ia = − ia − ωL + Va (1.7.12)
dt La La rm La
d kt b 1
ωL = ia − ωL − τL (1.7.13)
dt J J J
If we consider load torque τL as an uncontrolled (disturbance) input, the state space
representation of DC motor dynamics is
   Ra kb rL    1 
  
i̇a − − i a 0
 =  kLa La rm 
b    +  La  Va +  1  τL (1.7.14)
t −
ω̇ − ωL 0
J J J

Chapter 1 25
Daniel Abebe Modern Control Systems Lecture Note
 
h i ia
y= 0 1   (1.7.15)
ωL

As we can see from equation 1.7.14 and equation 1.7.15, the dc motor dynamics has two
states, namely ia and ωL .

If the position of the DC motor is required for control purposes, like in servo motors, we
add one state θ̇ = ωm . In this case, the state space representation becomes
 R kb rL    
 
−
a
− 0 1  
i̇a La i
La rm   a   La  0
  
k b  1
 
   t
   
ω̇L  =  − 0 ωL + V + −  τL (1.7.16)
  0
a
     
    J J  J
rL 
θ̇L 0 0 θL 0 0
rm

 
i
h i a 
y = 0 0 1 ωL  (1.7.17)
 
 
θL

Projector Screen Positioning System

The DC motor position control system for the remote-controlled projector screen incor-
porates a DC motor paired with a non-inverting pre-amplifier to accurately manage the
screen’s position. This configuration boosts the input armature voltage supplied to the
DC motor, enhancing its performance and responsiveness.

The system setup incorporates a DC motor paired with a pre-amplifier to accurately

+ ia
Ra La
+
− +

Rf ωm
Vs Va M Load

Ri
-
-
(a) Armature Voltage Controlled DC Mo- (b) Remote Controlled
tor With Pre-amplifier Projector Screen

Figure 1.18: Projector Screen Positioning System

control the position of a projector screen. The pre-amplifier boosts the input armature
voltage supplied to the DC motor, enhancing its performance and responsiveness. The

Chapter 1 26
Daniel Abebe Modern Control Systems Lecture Note

position of the leading edge of the projector screen is proportional to the angular position
of the coupling gear. Mathematically, this relationship is expressed as:

y = θrL (1.7.18)

where: y is the position of the leading edge of the screen, theta is the angular position
of the coupling gear, rL is the radius of the coupling gear.

The state-space model describes the dynamics of the system, including how the armature
current, angular velocity, and screen position evolve in response to the input voltage and
load torque. The output equation specifies that the position of the leading edge of the
screen y(t) is the measured output of the system.

 R kb rL    
 
−
a
− 0 Av  
i̇ La La rm   ia   La  0
 a 
   kt b  1
     
ω̇L  =  − + V + − τL (1.7.19)
   
0 ωL
   0  a  J

   J J    
rL   
ẏ 0 rL 0 y 0 0
rm

 
i
h i a 
y(t) = 0 0 1 ωL  (1.7.20)
 
 
y

1.8 Mathematical Modeling of Fluid Systems

Mathematical modeling of fluid systems involves using mathematical expressions to de-
scribe the behavior and dynamics of fluids, whether they are in motion or at rest. These
models are essential for understanding and predicting fluid behavior in various engineering
applications, such as hydraulics, aerodynamics, and process engineering.
Fluid systems typically consist of a flow source or pressure source, energy-storing
elements, and energy-dissipating components, similar to electrical systems.

1.8.1 Basic Laws Used in Fluid Systems Modeling

The modeling of fluid systems is based on fundamental physical laws, including the con-
servation of mass, momentum, and energy. Key concepts include:

• Continuity Equation: This principle states that mass cannot be created or de-
stroyed in a closed system. For incompressible fluids, the continuity equation can
be expressed as:

∇·v =0 (1.8.1)

where v is the fluid velocity vector. This equation ensures that the mass flow rate
remains constant throughout a streamline.

Chapter 1 27
Daniel Abebe Modern Control Systems Lecture Note

• Bernoulli’s Equation: This equation relates the pressure, velocity, and elevation
in a moving fluid and is derived from the conservation of energy. For an incom-
pressible, non-viscous fluid, it can be expressed as:
1
P + ρv 2 + ρgh = K (1.8.2)
2
where P is the pressure, ρ is the fluid density, v is the flow velocity, g is the
acceleration due to gravity, and h is the height above a reference level.

• Navier-Stokes Equations: These equations describe the motion of fluid sub-

stances and are derived from Newton’s second law. They account for the forces
acting on the fluid, including viscous forces, pressure forces, and external forces:
 
∂v
ρ + v · ∇v = −∇P + µ∇2 v + f (1.8.3)
∂t

where ρ is the fluid density, p is the pressure, µ is the dynamic viscosity, and f
represents body forces (e.g., gravity).

• Law of Conservation of Mass: This law states that the mass of a closed system
must remain constant over time. In fluid dynamics, this is often represented by the
continuity equation mentioned above.

1.8.2 Key Fluid Properties

Several key properties are used to characterize fluid behavior in mathematical models:

Fluid Inertia
Fluid inertia refers to the tendency of a fluid to resist changes in its state of motion. It
is quantified by the mass of the fluid and can be represented in terms of fluid pressure
(P ) and volumetric flow rate (Q̇):

∆P
Lf = (1.8.4)
∆Q̇
where Lf is the fluid inertia. This property is crucial in analyzing transient flows and
dynamic responses in fluid systems.

Fluid Capacitance
Fluid capacitance describes the ability of a fluid system to store volume in response to
pressure changes. It is defined as the ratio of the change in volume (∆V ) to the change
in pressure (∆P ):

∆V
C= (1.8.5)
∆P
where C is the fluid capacitance. This property is important in systems like hydraulic
accumulators, where fluid storage plays a significant role in performance.

Chapter 1 28
Daniel Abebe Modern Control Systems Lecture Note

Fluid Bulk Modulus

A fluid is classified as incompressible if its density remains constant, whereas it is con-
sidered compressible if its density varies with pressure. Liquids can typically be treated
as incompressible under relatively low pressures, while hydraulic fluids at high pressures
are regarded as compressible. The bulk modulus (β) is a measure of a fluid’s resistance
to compression. It is defined as the ratio of the change in pressure to the relative change
in volume:
∆P
β = ρo (1.8.6)
∆ρ
where ρo is a reference fluid density at a nominal pressure and temperature. A high bulk
modulus indicates that a fluid is incompressible and can resist changes in volume under
pressure, which is a key assumption in many fluid dynamics problems.

1.8.3 Flow Characteristics

Fluid flow can be categorized into laminar and turbulent flow, each exhibiting different
characteristics and resistances.

Laminar Flow
Laminar flow occurs when a fluid flows in parallel layers, with minimal disruption between
them. This type of flow is characterized by smooth, orderly motion and is typically
observed when the source pressure and outlet pressure are the same, or when the diameter
of the pipe is larger than the length of the pipe. The laminar flow resistance is proportional
to the ratio of pressure difference to volumetric flow rate:
∆P
RL = (1.8.7)
Q

Turbulent Flow
Turbulent flow is characterized by chaotic, irregular fluid motion. This type of flow
involves mixing and eddies, leading to increased energy dissipation and resistance. The
transition from laminar to turbulent flow is influenced by factors such as flow velocity,
fluid viscosity, and pipe diameter. The Turbulent flow resistance is proportional to the
ratio of the pressure difference to the square of the volumetric flow rate:
∆P
RT = (1.8.8)
Q2

1.8.4 Examples
Hydraulic Piston and Cylinder System
Hydraulic systems are essential in various industrial applications, facilitating the efficient
transfer of power through the controlled movement of fluids. Central to many hydraulic
systems is the hydraulic piston and cylinder arrangement, which converts hydraulic energy
into mechanical work. This system operates on the principles of fluid mechanics and
adheres to fundamental laws, including Pascal’s principle, which asserts that pressure

Chapter 1 29
Daniel Abebe Modern Control Systems Lecture Note

Piston Head (A)

Qin CV

Figure 1.19: Hydraulic Piston and Cylinder

applied to a confined fluid is transmitted undiminished in all directions. As illustrated

in Figure 1.19, a hydraulic piston is positioned within a cylindrical chamber filled with
hydraulic fluid. When pressure is applied to the fluid, it generates a force on the piston,
prompting its movement. This movement can be utilized to perform a range of tasks,
including lifting heavy loads, actuating machinery, and controlling the motion of other
components within a system.

Using the law of conservation of mass, we can express the rate of accumulation of mass
as the difference between the mass inflow rate and the mass outflow rate. In this case,
since there is no mass leaving the system (no outflow), the dynamic equation governing
the motion of the piston can be formulated as:

ṁ = ρQin (1.8.9)

Here, ṁ represents the rate of change of mass, ρ is the fluid density, and Qin is the
volumetric flow rate into the system.

Next, we can substitute the mass m with the product of density and volume, m = ρV .
This leads to the following relationship:

ρV̇ + V ρ̇ = ρQin (1.8.10)

In this equation, V̇ denotes the rate of change of volume, and V is the volume of the fluid
within the cylinder.

To further refine our model, we can replace the density ρ using the relationship derived
from the bulk modulus of elasticity, which is given by:

ρo dP
ρ= (1.8.11)
β dt

dP
where ρo is the reference density, β is the bulk modulus, and represents the rate of
dt
change of pressure. Additionally, we can express the rate of change of volume V̇ in terms
of the piston head area A and the displacement of the piston head x:

V̇ = Aẋ (1.8.12)

Chapter 1 30
Daniel Abebe Modern Control Systems Lecture Note

Substituting these expressions into our earlier equation, we arrive at:

ρ
Ṗ Ax + ρAẋ = ρQin
β
βAx
Ṗ = (Qin − Aẋ) (1.8.13)
ρ

In this final equation, Ṗ represents the rate of change of pressure, A is the cross-sectional
area of the piston, and ẋ is the velocity of the piston head. This equation captures
the dynamic relationship between pressure changes, fluid density, and the movement of
the piston, allowing us to analyze the behavior of the hydraulic system under varying
conditions.

By integrating these principles, we can effectively model the hydraulic piston and cylinder
system, facilitating predictions of its performance in practical applications.

Hydraulic Actuator System

The simple hydraulic actuator shown in Figure 1.20a is connected to a load and accom-
panied by a restoring spring. As the piston moves, it drives the load while compressing
the spring. This interaction is crucial for understanding the dynamics of the actuator.

Piston head (A)

k Fatm
Fs
Qin Fh FL
FL Fv

Viscous friction (b)

(a) Hydraulic Actuator (b) Free Body Diagram of Hydraulic Actuator

Figure 1.20: Hydraulic Actuator System

The free body diagram of the system is illustrated in Figure 1.20b. This diagram helps
visualize the forces acting on the actuator, which include the hydraulic force, atmospheric
pressure force, viscous force, spring force, and the load force.

The mathematical model describing the motion of the actuator is derived using New-
ton’s first law of motion, which states that the sum of forces acting on an object is equal
to the mass of the object multiplied by its acceleration. This can be expressed as:

ΣF = Fh − Fatm − Fv − Fs − FL = mẍ (1.8.14)

Where Fh is the hydraulic force acting on the piston, Fatm is the force due to atmospheric
pressure, Fv is the viscous force, Fs is the spring force, FL is the load force, m is the mass
of the piston, and ẍ is the acceleration of the piston.

By replacing Fh with P A (where P is the pressure and A is the piston head area), and
substituting Fatm = Patm A for the atmospheric pressure force, the equation becomes:

(F − Fatm )A − FL = mẍ + bẋ + kx (1.8.15)

Chapter 1 31
Daniel Abebe Modern Control Systems Lecture Note

Selecting the states of the system x1 = x, x2 = ẋ, the sate space representation will be

ẋ1 = x2
k b PA 1 Patm A
ẋ2 = − x1 − x2 + − FL − (1.8.16)
m m m m m

         
ẋ1 0 1 x1 0 0 P 0
 =   +     −   Patm (1.8.17)
k b A 1 A
ẋ2 − − x2 − FL
m  m  m m m
h i x1
y= 1 0   (1.8.18)
x2

To find the pressure required to hold the piston at its nominal position, we can set the
velocity ẋ = 0 and assume x = x0 (the equilibrium position). This simplifies our equation
to:
FL k
P = Patm + + x0 (1.8.19)
A A
This equation indicates that the pressure P must balance the load force per unit area,
plus the atmospheric pressure acting on the piston. Understanding this relationship
is essential for designing hydraulic systems that effectively manage loads and maintain
desired positions.

1.9 Mathematical Modeling of Chemical Process

The dynamic model of chemical systems is derived from the balance equations for total
mass, energy, and component mass (mole). These equations are crucial for understanding
and predicting the behavior of chemical processes. Let’s delve into the concepts and
equations involved in the mathematical modeling of chemical processes.
The dynamic model of chemical systems is derived from the balance equations for total
mass, energy, and component mass (mole). For a system with a well-defined boundary,

Change Inventory = (IN − OU T )T ZB + (Generated − Loss)IZB (1.9.1)

Note:- TZB (through the boundary of the system), IZB (In the boundary of the system)
To formulate the dynamic equation, we have to decide the control boundary and which
balance we have to use. Once the boundary is set, if the interest is,

• Mass, volume or pressure → mass balance,

• Concentration→component balance,

• Temperature→energy balance, and

• Interaction between flow and pressure→ Mechanical energy balance(= momentum

balance = Bernoulli = Newton’s second law)

Chapter 1 32
Daniel Abebe Modern Control Systems Lecture Note

1.9.1 Fundamental Laws for modeling chemical processes

Mass balance
The rate of change of mass accumulation in a closed boundary is,
dm
= win − wout (1.9.2)
dt
win and wout are mass flow rates for the entering and exiting streams (bulk flow).
Introducing density,
dV
ρ = ρin qin − ρout qout (1.9.3)
dt
qin and qout are volumetric1 flow rates.
For constant density
dV
= qin − qout (1.9.4)
dt
In most dynamic processes, volume is not conserved; often the interest is height of the
liquid, thus the equation becomes
dV dh dA dh
=A +A = qin − qout (1.9.5)
dt dt dh dt
In most cases, the area is conserved. hence
dh
A = qin − qout (1.9.6)
dt

Dynamic Component Balance

The dynamic component balance can, for an arbitrary component A, be written
dnA
= FA,in − FA,out + GA (1.9.7)
dt
Here, nA [mol A] is the inventory (amount) of component A inside the system’s boundary,
FA,in − FA,out [mol A/s] are the molar flow rates of A in the streams (bulk flow) and GA
[mol A/s] is net generated in the chemical reactions.

With concentrations c̄A = nA /V , cA,in = FA,in /qin , cA,out = FA,out /qout and the generation
of component as average reaction rate times the volume GA = r̄A V , the equation will be

d(c̄A V )
= cA,in qin − cA,out qout + r̄A V [molA/s] (1.9.8)
dt
1
Volumetric flow rate is the product of flow rate and cross-sectional area perpendicular to the flow
rate

Chapter 1 33
Daniel Abebe Modern Control Systems Lecture Note

Dynamic Energy Balance

The general energy balance over a small duration of time is
 
dH dV dp
= Hin − Hout + Q + Ws − (pex − p) +V [J/s] (1.9.9)
dt dt dt

where H[J] is enthalpy of the system (H = U(internal energy) + pV ), Hin − Hout is the
sum of internal energy in the streams plus the flow work that the streams perform on the
system as they are ”pushed” in or out of the system. Q [J/s] is supplied heat (through
the system’s wall) and Ws [J/s] is supplied useful mechanical work (usually shaft work,
for example, from a compressor, pump, or turbine).
 
dV dp
The term −(pex − p) +V occurs if the volume and pressure in the boundary of
dt dt
control changes, which for may process it is negligible compared to the other parameters.
H
Using specific enthalpy ℏ = , we can combine mass balance and energy balance equa-
m
tions, since
dH dℏ dm
=m +ℏ (1.9.10)
dt dt dt
dm
by replacing for = win − wout
dt
 
dℏ dV dp
m = win (ℏin − ℏ) − wout (ℏout − ℏ) + Q + Ws − (pex − p) +V (1.9.11)
dt dt dt

Note:- Enthalpy ℏ(T, p, f, nj ) [J/kg] is generally a function of temperature T , pressure p,

phase distribution f (where f is fraction of light phase) and composition (nj ). The time
derivative of the enthalpy should be treated correctly (ı.e, apply chain rule carefully).

1.9.2 Examples
Stirred-Tank Heating Process: Constant Holdup process
The liquid inlet stream consists of a single component with a mass flow rate wi and an
inlet temperature Ti . The tank contents are agitated and heated using an electrical heater
that provides a heating rate,Q. The modeling assumptions

• Perfect mixing; thus, the exit temperature T is also the temperature of the tank
contents.

• The inlet and outlet flow rates are equal; thus, wi = w and the liquid holdup V is
constant.

• The density ρ and heat capacity C of the liquid are assumed to be constant. Thus,
their temperature dependence is neglected.

• Heat losses are negligible

Chapter 1 34
Daniel Abebe Modern Control Systems Lecture Note

Ti
wi

T
w

Figure 1.21: Stirred-Tank Heating Process with Constant Hold-up

The enthalpy (total energy of the system) is the sum of internal energy and the product of
volume and pressure ℏ = Ūint + pV , but for low pressure system, it can be approximated
to ℏ ≈ Ūint , internal energy per unit mass.

The system enthalphy can be considered as a function of only temperature and pressure,
ℏ(T, p), hence

dℏ ∂ℏ dT ∂ℏ dp
= f (p) + f (T ) (1.9.12)
dt ∂T dt ∂p dt
∂ℏ ∂ℏ
For the fluid and solids ≈ 0, and f (p) = Cp (heat capacity at constant pressure),
∂p ∂T
For the total volume we have
dH dℏ dT
=m = ρV Cp (1.9.13)
dt dt dt
Solving for H by integrating we have H − Href = ρV Cp (T − Tref ), in specific enthalpy,
ℏ−ℏref = Cp (T −Tref ), for inlet stream ℏi −href = Cp (Ti −Tref ) and ℏ−href = Cp (T −Tref )
outlet stream.

Chapter 1 35
Daniel Abebe Modern Control Systems Lecture Note

Replacing the terms in equation 1.9.11,

dℏ
m = win (ℏin − ℏ) − wout (ℏout − ℏ) + Q
dt
dℏ
m = wCp (Ti − Tref ) + wCp (T − Tref ) + Q
dt
dℏ dℏ
m = wCp (Ti − T ) + Q =⇒ ρV = wCp (Ti − T ) + Q
dt      dt
w w 1
Ṫ = Ti − T+ Q (1.9.14)
ρV ρV ρV Cp

Stirred-Tank Heating Process: Variable Holdup

In this case, the only difference is the volume changes; hence, there will be a rate of
change of mass, and we should consider mass balance.
dm dV
= wi − w ⇒ ρ = wi − w (1.9.15)
dt dt
We know that the rate of change of the enthalpy of the system is
dH
= Hin − Hout + Q
dt
d(mℏ)
= wi (ℏi − ℏref ) − wout (hout − href ) + Q
dt
d(V ℏ)
ρ = wi (ℏi − ℏref ) − wout (hout − href ) + Q (1.9.16)
dt
d(V ℏ)
Since the volume is variable, we have to apply the chain rule to , hence
dt
d(V ℏ) dV dℏ
ρ = ρℏ + ρV (1.9.17)
dt dt dt

We know that ℏ − ℏref = Cp (T − Tref ), without loss of generality, we can set ℏref = 0,
dℏ dT
hence ℏ = Cp (T − Tref ), replacing this values in equation 1.9.11 and = Cp ,
dt dt
dV dT
ρCp (T − Tref ) + ρV Cp = wi Cp (Ti − Tref ) − wCp (T − Tref ) + Q (1.9.18)
dt dt
dV
replacing for ρ with wi − w from equation 1.9.15 and rearranging
dt
1 1 1
Ṫ = wi Ti − T+ Q (1.9.19)
ρV Cp ρV Cp ρV Cp
1 1
V̇ = wi − w (1.9.20)
ρ ρ

Electrically Heated Stirred Tank

In a stirred-tank heating system with constant holdup, the metal heating element has
significant thermal capacitance. The electrical heating rate directly impacts the element’s

Chapter 1 36
Daniel Abebe Modern Control Systems Lecture Note

temperature, not the liquid contents. Temperature gradients in the heating element are
neglected, assuming it has a uniform temperature (Te), which represents the average tem-
perature for the heating. Based on this new assumption, and the previous assumptions,
the unsteady-state energy balances for the tank and the heating element can be written
as
dT
mCp = wCp (Ti − T ) + he Ae (Te − T ) (1.9.21)
dt
dTe
me Ce = Q − he Ae (Te − T ) (1.9.22)
dt

Where m = ρV and me Ce is the product of the mass of metal in the heating element and
its specific heat. The term he Ae is the product of the heat transfer coefficient and area
available for heat transfer.

Note that mC and me Ce are the thermal capacitance of the tank contents and the heating
element, respectively. Q is an input variable, the thermal equivalent of the instantaneous
electrical power dissipation in the heating element.

The Continuous Stirred-Tank Reactor (CSTR)

The Continuous Stirred-Tank Reactor (CSTR) is a type of reactor used for continuous
mixing of reactants within a vessel. The reactants are continuously fed into the reactor,
while the products are continuously removed. The contents of the reactor are well mixed,
and the operating conditions, such as temperature, pressure, and concentration, are uni-
form throughout the reactor. It is the commonly used reactor in industrial processes.

To simplify the analysis, some common assumptions must be made. These assumptions
include:

• It is assumed that the contents of the reactor are perfectly mixed, meaning that
the concentration of reactants and products is uniform throughout the reactor at
any given time.

• The reactor operates under steady-state conditions, where the operating parameters
such as flow rates, temperatures, and concentrations do not change with time.

• The volume of the reactor remains constant during the reaction, and there are no
significant volume changes due to the reaction.

• Instantaneous reaction: It is assumed that the reaction occurs instantaneously once

the reactants are mixed, without any time delay or reaction kinetics effects.

• The reactor is assumed to behave ideally, with no heat losses, no mass transfer
limitations, and no non-ideal mixing effects.

Consider a simple liquid-phase, irreversible chemical reaction where chemical species A

reacts to form species B. The reaction can be written as A → B. We assume that the
rate of reaction is first-order concerning component A,

r = kCA (1.9.23)

Chapter 1 37
Daniel Abebe Modern Control Systems Lecture Note

Pure A
q, c Ai , Ti

Mixture of A and B
q, c A , T

Tc

Figure 1.22: Non-isothermal Continuous Stirred-Tank Reactor

where r is the reaction rate of A per unit volume, k is the reaction constant, and CA is
the concentration of A. For single-phase reactions, the rate constant is typically a strong
function of reaction temperature given by the Arrhenius relation,
−E
k = ko e RT (1.9.24)

where ko is the frequency factor and E is the activation energy, usually determined by
fitting experimental data, and R is the gas constant.

The inlet stream consists of pure component A with molar concentration, CAi . A cooling
coil is used to maintain the reaction mixture at the desired operating temperature by
removing heat that is released in the exothermic reaction.

Chapter 1 38
Daniel Abebe Modern Control Systems Lecture Note

Our initial CSTR model development is based on three assumptions:

• The CSTR is perfectly mixed.
• The mass densities of the feed and product streams are equal and constant. They
are denoted by ρ.
• The liquid volume V in the reactor is kept constant by an overflow line.
For these assumptions, the unsteady-state mass balance for the CSTR is
d(ρV )
= ρqi − ρq ⇒ q = qi (Since V is constant) (1.9.25)
dt
Even though the inlet and outlet flow rates may change due to upstream or downstream
conditions, this equation must be satisfied at all times.

Note:- both flow rates are denoted by the symbol q in the Figure.

For the stated assumptions, the unsteady-state component balances for species A (in
molar concentration units) are
dCA
V = q(CAi − CA ) − V kCA (1.9.26)
dt
This balance is a special case of the general component balance.

For the unsteady-state energy balance for the CSTR, we need an additional five assump-
tions:
• The thermal capacitance of the coolant and the cooling coil wall is negligible com-
pared to the thermal capacitance of the liquid in the tank.
• All of the coolant is at a uniform temperature, Tc . (That is, the increase in coolant
temperature as the coolant passes through the coil is neglected.)
• The rate of heat transfer from the reactor contents to the coolant is given by

Q = U A(Tc − T )

where U is the overall heat transfer coefficient and A is the heat transfer area. Both
of these model parameters are assumed to be constant.
• The enthalpy change associated with the mixing of the feed and the liquid in the
tank is negligible compared with the enthalpy change for the chemical reaction.
• Shaft work and heat losses to the ambient can be neglected.
With the use of the new assumptions and inclusion of energy balance, the dynamic model
of CSTR is given by
dT
V ρC = wC(Ti − T ) + (−∆HR )V kCA + U A(Tc − T ) (1.9.27)
dt
This model is nonlinear as a result of the many product terms and the exponential
temperature dependence of k; hence, it should be solved by numerical integration. The
model will be more complex if

Chapter 1 39
Daniel Abebe Modern Control Systems Lecture Note

• More complicated rate expressions are considered

• Additional species or chemical reactions are involved
Reactions involving multiple species are described by high-order, highly coupled, nonlin-
ear reaction models, because several component balances must be written.

Quadruple Tank System

Consider a quadruple tank process shown in Figure 1.23. The objective is to control the
level of the liquid in the tanks by using pump 1 and pump 2. The pumps pump the liquid
from the reservoir. The volumetric flow rate2 q produced by each pump is proportional
to the voltage applied to each pump. Let these proportionality constants be k1 and k2 for
pump 1 and pump 2, respectively. The position of valve 1 controls the volumetric flow
rate to tank 1 and tank 4, while the position of valve 2 controls the volumetric flow rate
to tank 2 and tank 3. Let the position of valve 1 be γ1 and the position of valve 2 be γ2 .

Figure 1.23: Quadruple Tank Process

So we have
q1in = γ1 k1 V1 q2in = γ2 k2 V2
q3in = (1 − γ2 )k2 V2 q4in = (1 − γ1 )k1 V1

From fluid dynamics, the rate of mass accumulation in each tank is equal to the difference
between mass inflow and mass outflow.
ṁ = ρ(qin − qout ) (1.9.28)
where qin is volumetric inflow rate and qout is volumetric outflow rate. The volumetric
outflow rate is determined by considering each tank as an isolated system. Using the
Bernoulli equation
1 1
P1 + gh1 + v12 = P2 + gh2 + v22 (1.9.29)
2 2
2
Volumetric flow rate q is equal to the product of the flow rate of the fluid (ν) times the area (a)
perpendicular to the flow. (q = ν × a)

Chapter 1 40
Daniel Abebe Modern Control Systems Lecture Note

If we take the top as point 2 and the bottom of the tank as point 1 (h1 = 0), both of
them are exposed to atmospheric pressure, so P1 = P2 . Similarly, since there is no inflow
from the top, v2 = 0. The fluid flows out of each tank due to gravity only.
with these assumption’s equation 1.9.29 simplifies to
p
v1 = 2gh2 (1.9.30)

Let the outlet have a cross-sectional area a, the outflow rate will be
p
qout = av = 2gh (1.9.31)
m m
from basic physics ρ = = , rate of accumulation of mass ṁ is
V Ah
dh
ṁ = ρA (1.9.32)
dt
Combining equation 1.9.28 and equation 1.9.32, the dynamic equation for each tank will
be

Aḣ = qin − qout (1.9.33)

The dynamic equation of the quadruple tank system is then

p p
A1 ḣ1 = q1in + q3out − q1out = γ1 k1 V1 + a3 2gh3 − a1 2gh1
p p
A2 ḣ2 = q2in + q4out − q2out = γ2 k2 V2 + a4 2gh4 − a2 2gh2
p
A3 ḣ3 = q3in − q3out = (1 − γ2 )k2 V2 − a3 2gh3
p
A4 ḣ4 = q4in − q4out = (1 − γ1 )k1 V1 − a4 2gh4

 √ √
a3 2gh3 a1 2gh1
  
γ1 k1
  − 0
ḣ
 1 
 √A1 √A1    A1 
a 2gh a 2gh γ2 k2   

4 4 2 2
− 0
     
ḣ2   A A2 
 
A2
 V
  1
 = 2 √ +

(1.9.34)
   a3 2gh3
  (1 − γ2 )k2 
ḣ3   −
  0  V
2
A3
  
   A
√ 3   
ḣ4  a4 2gh4   (1 − γ1 )k1 
− 0
A4 A4

If the levels of the liquid in tank 1 and tank 2 are required to be controlled, the output
equation will be
 
h
     1
 
y1 1 0 0 0 h2 
y= =  
  (1.9.35)
y2 0 1 0 0 h3 
 
h4

Chapter 1 41
Daniel Abebe Modern Control Systems Lecture Note

1.10 State Space Representation From Transfer Function

When a transfer function is converted to state space representation, the state space com-
monly has four normal forms (canonical forms). The forms depend on the characteristics
polynomial of the transfer function or some other conditions (conditions like controlla-
bility and observability, which are explained later).
1. Controllable canonical form
Consider a transfer function
Y (s) N (s) bn sn + bn−1 sn−1 + bn−2 sn−2 . . . + b1 s + b0
= = n (1.10.1)
U (s) D(s) s + an−1 sn−1 + an−2 sn−2 . . . + a1 s + a0
If we take an intermediate function X(s) (state of the system) then we have,
Y (s) X(s) bn sn + bn−1 sn−1 + bn−2 sn−2 . . . + b1 s + b0
· = n (1.10.2)
X(s) U (s) s + an−1 sn−1 + an−2 sn−2 . . . + a1 s + a0
which can be written as
X(s) 1
= n (1.10.3)
U (s) s + an−1 sn−1 + an−2 sn−2 . . . + a1 s + a0
Y (s)
= bn sn + bn−1 sn−1 + bn−2 sn−2 . . . + b1 s + b0 (1.10.4)
X(s)
From equation 1.10.3
U (s) = (sn + an−1 sn−1 + an−2 sn−2 . . . + a1 s + a0 )X(s) (1.10.5)
Letting X1 = X(s), X2 = sX(s), X3 = s2 X(s), . . . , Xn = sn−1 X(s), we have
ẋ1 = x2
ẋ2 = x3
..
.
ẋn−1 = xn
ẋn = −an−1 xn − an−2 xn−1 − . . . − a1 x2 − a0 x1 + u (1.10.6)
which leads to a state equation
    
 
ẋ 0 1 ... 0 x 0
 1    1   
 ẋ2   0 0 ... 0   x2  0
      
 ..   .. .. .. ..   ..   .. 
      
 . = . . . .   .  + . u (1.10.7)
      
      
ẋn−1   0 0 ... 1  xn−1  0
      
ẋn −a0 −a1 ... −an−1 xn 1

From equation 1.10.4

Y (s) = (bn sn + bn−1 sn−1 + bn−2 sn−2 + . . . + b1 s + b0 )X(s)
y = bn ẋn + bn−1 xn + bn−2 xn−1 + . . . + b1 x2 + b0 x1
y = (b0 − bn a0 )x1 + (b1 − bn a1 )x2 + . . . + (bn−1 − bn an−1 )xn−1 + bn u (1.10.8)

Chapter 1 42
Daniel Abebe Modern Control Systems Lecture Note

Which gives us an output equation

 
x
 1 
 x2 
 
 . 
h i 
y = (b0 − bn a0 ) (b1 − bn a1 ) . . . (bn−1 − bn an−1 )  ..  + bn u (1.10.9)
 
 
xn−1 
 
xn

2. Observable canonical form

Factoring sn and rearranging equation 1.10.1

Y (s) = bn U (s) + s−1 (bn−1 U (s) − an−1 Y (s) + s−1 (bn−2 U (s) − an−2 Y (s) + . . .
+s−1 (b1 U (s) − a1 Y (s) + s−1 (b0 U (s) − a0 Y (s)))))
(1.10.10)

Let

X1 (s) = s−1 (b0 U (s) − a0 Y (s))

X2 (s) = s−1 (b1 U (s) − a1 Y (s) + X1 (s))
..
.
Xn (s) = s−1 (bn−1 U (s) − an−1 U (s) + Xn−1 (s)) (1.10.11)

With this equation 1.10.10 becomes,

Y (s) = bn U (s) + Xn (s) (1.10.12)

Solving for Y (s) from equation 1.10.12 and replace for Y (s) in equation 1.10.11 and
rearranging we have

sX1 (s) = ((b0 − bn a0 )U (s) − a0 Xn (s))

sX2 (s) = ((b1 − bn a1 )U (s) − a1 Xn (s) + X1 (s))
..
.
sXn (s) = ((bn−1 − bn an−1 )U (s) − an−1 Xn (s) + Xn−1 (s)) (1.10.13)

Which leads to a state equation

      
ẋ1 0 0 . . . 0 −a0 x1 b 0 − b n a0
      
 ẋ2  1 0 . . . 0 −a1   x2   b1 − bn a1 
      
 ..   .. .. .. .. ..   ..   ..
      
 .  = . . .  .  +  u (1.10.14)

. . .
      
ẋn−1  0 0 0 . . . 0 −an−2  xn−1  bn−2 − bn an−2 
      
      
ẋn 0 0 . . . 1 −an−1 xn bn−1 − bn an−1

Chapter 1 43
Daniel Abebe Modern Control Systems Lecture Note

and output equation

 
x
 1 
 x2 
 
 . 
h i 
y = 0 0 . . . 0 1  ..  + bn u (1.10.15)
 
 
xn−1 
 
xn
As we can see from equation 1.10.7, equation 1.10.9, equation 1.10.14 and equation
1.10.15, the system matrix of observable canonical form is a transpose of the sys-
tem matrix of controllable canonical form (Ao = A⊤ c ). Input matrix of observable
canonical form is equal to the transpose of the output matrix of controllable canoni-
cal form (Bo = C⊤ c ). The output matrix of observable canonical form is equal to the
transpose of the controllable canonical form input matrix (Co = B⊤ c ). State trans-
mission matrix of observable canonical form is a transpose of state transmission
matrix of controllable canonical form (Do = D⊤ c ).

3. Diagonal canonical Form

Suppose the transfer function has distinct poles p1 , p2 , . . . , pn and can be factorized
to
c1 c2 cn
Y (s) = U (s) + U (s) + . . . + U (s) (1.10.16)
s + p1 s + p2 s + pn
1 1 1
Letting X1 (s) = U (s), X2 (s) = U (s), . . . , Xn (s) = U (s)
s + p1 s + p2 s + pn
we have
ẋ1 = −p1 x1 + u
ẋ2 = −p2 x2 + u
..
.
ẋn = −pn xn + u (1.10.17)
Which leads to a state equation
 
ẋ1     
  −p1 0 . . . 0 x 1
 ẋ2  
    1  
.
0 −p2 . . . 0   x2   .. 
    
 ..  
 
 . = .. .. .. ..   ..  +   u
    (1.10.18)
 
  . . . .   .  1


ẋn−1      
  0 0 . . . −pn xn 1
ẋn
and output equation
 
x
 1
h i  x2 


y = c1 c2 . . . cn  .. 
 (1.10.19)
.
 
xn

Chapter 1 44
Daniel Abebe Modern Control Systems Lecture Note

4. Jordan Canonical Form

If the transfer function can be factorized and has r multiplicities of pole p1 , and the
rest poles are distinct, the transfer function can be written as
c1 c2 cr
Y (s) = U (s) + 2
U (s) + . . . + U (s)+
s + p1 (s + p1 ) (s + p1 )r
cr+1 cr+2 cn
U (s) + U (s) + . . . + U (s) (1.10.20)
s + p2 (s + p3 ) (s + p(n−r) )

Let
1
X1 (s) = U (s)
s + p1
1
X2 (s) = U (s)
(s + p1 )2
which can be written as
1
X2 (s) = X1 (s)
s + p1
in similar way
1 1
X3 (s) = X2 (s), . . . , Xr (s) = X(r−1) (s)
s + p1 s + p1

Since the rest poles are distinct, the derivation follows the way we drive for diagonal
normal form. In the time domain, the states will be

ẋ1 = −p1 x1 + u
ẋ2 = −p1 x2 + x1
..
.
ẋr = −p1 xr + xr−1
ẋr+1 = −p2 xr+1 + u
..
.
ẋn = −pn−r xn−r + u (1.10.21)

Which led to a state equation

     
ẋ −p 0 0 ... 0 0... 0 x 1
 1   1  1   
 ẋ2   1 −p1 0 ... 0 0... 0   x2  0
      
      
 ẋ3   0 1 −p1 ... 0 0... 0   x3  0
      
 ..   .. .. .. .. .. .. ..   ..   .. 
      
 .   . . . . . . .   .  .
 =   +   u (1.10.22)
−p1 . . .
      
 ẋr   0 0 0 ... 1 0  xr−1  0
      
−p2 . . .
      
ẋr+1   0 0 0 ... 0 0   xr  1
 ..   .. .. .. .. .. .. ..   ..   .. 
      
 .   . . . . . . .   .  .
      
ẋn 0 0 0 ... 0 0 . . . −pn−r xn 1

Chapter 1 45
Daniel Abebe Modern Control Systems Lecture Note

The output matrix is similar to the diagonal normal form

 
x
 1
h i  x2 


y = c1 c2 . . . cn   .. 
 (1.10.23)
.
 
xn

1.11 Transfer Function From State Space Representation

Consider a LTI system represented by state space representation on equation 1.3.7 and
equation 1.3.8, with initial condition x(0) = x0 , taking the Laplace transform of the
equations, we have

sX(s) − x0 = AX(s) + BU(s)

(sI − A)X(s) = x0 + BU(s)
X(s) = (sI − A)−1 (x0 + BU(s)) (1.11.1)

Y(s) = CX(s) + DU(s)

= C(sI − A)−1 (x0 + BU(s)) + DU(s)
= C(sI − A)−1 x0 + (C(sI − A)−1 B + D)U(s) (1.11.2)

If the initial condition is zero(x0 = 0, which is the case for all transfer functions), the
transfer function from input to output is

Y(s)
= C(sI − A)−1 B + D
U(s)
Cadj(sI − A)B + det(sI − A)D
= (1.11.3)
det(sI − A)

The determinant of (sI − A) is a characteristic polynomial of the transfer function.

1.12 Review on Eigenvalues and Eigenvectors

In linear algebra, given a square matrix A of size n × n, an eigenvalue λ is a scalar for
which there exists a non-zero vector v (called an eigenvector) such that:

Av = λv (1.12.1)

This equation can be rewritten as:

(A − λI)v = 0 (1.12.2)

where I is the n × n identity matrix. For a non-trivial solution (v ̸= 0) to exist, the

matrix (A − λI) must be singular. This leads to the characteristic equation:

Chapter 1 46
Daniel Abebe Modern Control Systems Lecture Note

det(A − λI) = 0 (1.12.3)

Eigenvalues characterize how a linear transformation (represented by a matrix) acts on its
eigenvectors. Each eigenvalue λ specifies the scaling factor applied to the corresponding
eigenvector while preserving its direction (for real eigenvalues). Eigenvalues may be real
numbers, indicating pure scaling, or complex numbers, indicating rotational transforma-
tions. In the special case of diagonal matrices, the eigenvalues are simply the entries
along the main diagonal.

Any nonzero vectors ν i , qi which satisfies the matrix equation

(λi I − A)qi = 0
ν i (λi I − A) = 0
where λi are eigenvalues of the matrix, are called left eigenvectors and right eigenvectors
of matrix A corresponding to eigenvalue λi respectively. The matrix formed from right
eigenvectors (qi ) of the matrix A is called the modality matrix of A and is denoted by
M.
h i
M = q1 | q2 | . . . |qn (1.12.4)
Eigenvectors possess a remarkable geometric property: when transformed by the matrix
A, they remain within their original span, meaning they don’t get rotated but merely get
scaled by their corresponding eigenvalue λ. This characteristic makes them fundamental
indicators of the matrix’s intrinsic stretching behavior. Unlike ordinary vectors that
may change direction under transformation, eigenvectors precisely reveal the primary
directions along which the matrix:
• Eigenvector stays on its original line (span(qi ))
• Transformation only changes the vector’s magnitude by a factor λ
• They form natural axes for understanding A’s action
The eigenvalues λi quantify exactly how much stretching or compression occurs along
each eigenvector direction. Together, the eigenvector-eigenvalue pairs (qi , λi ) provide a
complete description of the matrix’s most fundamental geometric action on the vector
space.

Left eigenvectors describe invariant directions under the row transformation induced by
A (as opposed to column transformation for right eigenvectors).

1.12.1 Relationships Between Right and Left Eigenvectors

1. For eigenvectors corresponding to different eigenvalues:
qi ν j = 0 if λi ̸= λj (1.12.5)
This means right and left eigenvectors from distinct eigen spaces are orthogonal.
2. For the same eigenvalue λi :
qi ν i ̸= 0 (1.12.6)
The dot product is non-zero, and often normalized to 1.

Chapter 1 47
Daniel Abebe Modern Control Systems Lecture Note

1.12.2 Application of Eigen structure in Control Systems

The eigenvalues λi play a vital role in determining the fundamental dynamic characteris-
tics of the system. The real part Re(λi ) < 0 indicates stability, while the imaginary part
determines oscillation characteristics.

The right eigenvectors define mode shapes, which represent specific combinations of state
variables that evolve exponentially according to their corresponding eigenvalues. Physi-
cally, these manifest as observable vibration patterns.

Left eigenvectors determine the controllability properties of the system through their inner
products ν i B; when this product is zero, the corresponding mode becomes uncontrollable.

The bi-orthogonality relationship qi ν j = δij enables modal decomposition, which allows

us to express the transfer function as
X (Cqi )(ν i B)
G(s) = . (1.12.7)
i
s − λi

This eigen-structure underpins key control techniques, including pole placement (eigen-
value assignment), model reduction (mode truncation, approximation of the system with
dominant poles), and fault detection (since right eigenvectors project the states to the
observable subspace).

1.13 Solution of State Space Equations

Consider the time derivative of e−At x, using chain rule we have
d −At d dx
(e x) = (e−At )x + e−At
dt dt dt
= −Ae−At x + e−At (ẋ)
= −Ae−At x + e−At (Ax + Bu)
= e−At (−Ax + Ax) + eAt Bu
= e−At Bu (1.13.1)
d
From equation 1.13.1 we can see that (e−At x) = e−At Au = e−At (ẋ − Ax), using this
dt
the solution to state equation is
e−At (ẋ − Ax) = e−At Bu
Z t Z t
d −At
(e x)dt = e−Aτ Bu(τ )dτ
t=0 dt t=0
Z t
−At t=t
e x |t=0 = e−Aτ Bu(τ )dτ
Zt=0
t
e−xt (x(t) − x(0)) = e−Aτ Bu(τ )dτ
t=0
Z t
At
x(t) = e x(0) + e At
e−Aτ Bu(τ )dτ
Z t t=0
x(t) = eAt x(0) + eA(t−τ ) Bu(τ )dτ (1.13.2)
t=0

Chapter 1 48
Daniel Abebe Modern Control Systems Lecture Note

And the output solution will be

y(t) = Cx(t) + Du(t)

Z t
At
y(t) = Ce x(0) + CeA(t−τ ) Bu(τ )dτ + Du(t)
Zt=0
t
At
y(t) = Ce x(0) + (CeA(t−τ ) B + D)u(τ )dτ (1.13.3)
t=0

The CeAt x(0) inZ equation 1.13.3 is called the zero-input response of the dynamic system,
t
while the term (CeA(t−τ ) B + D)u(τ )dτ is called the zero initial condition response.
t=0
The frequency domain representation of equation 1.13.3 is equation 1.11.2.

The term eAt denoted as Φ(t) is called state transition matrix as it transfer the state of a
dynamic system from any initial condition x(t0 ) to a final state x(t) whether the dynamic
system is driven by an input or not. A few properties of the state transition matrix are
stated below (It is a property due to the exponential nature of the function).

• Φ(0) = I since eA∗0 = I

• Φ(t2 − t1 )Φ(t1 − t0 ) = Φ(t2 − t0 )

• Φ−1 (t) = Φ(−1) since (eAt )−1 = eA(−t) = Φ(−t)

• Φ(t2 − t1 )=Φ(t1 )Φ(−t2 )

• Φ(t)n =Φ(nt)

The state transition matrix is determined by one of the following methods,

1. Using Inverse Laplace transform method

Φ(t) = L−1 ((sI − A)−1 ) (1.13.4)

2. Using Series expansion method

i=n
At
X (At)i
Φ(t) = e = (1.13.5)
i=0
i!

where n is the dimension of matrix A.

3. Using Cayley Hamilton Theorem

Cayley Hamilton’s theorem states that any function of an n×n square matrix f (A)
can be represented by a finite sum of powers of the matrix not exceeding order
n − 1. Let the eigenvalues of A be λi ’s where i = 1, 2, . . . n and the characteristics
polynomial corresponding to each λi be pi (λ). So we have n simultaneous equations
to solve for the coefficient of the characteristic polynomial. As Cayley Hamilton’s
theorem states, matrix A satisfies its characteristic equation, which is why we say

Chapter 1 49
Daniel Abebe Modern Control Systems Lecture Note

f (A) can be represented by a finite sum of powers of the matrix not exceeding order
n − 1.

f (At) = α0 (t)I + α1 (t)A + α2 (t)A2 + . . . + αn−1 (t)An−1 (1.13.6)

The constants are determined from a simultaneous equation resulting from

p(λi t) = α0 (t) + α1 (t)λi + α2 (t)λ2i + . . . + αn−1 (t)λi = f (λi t) (1.13.7)

Where i = 1, 2, . . . n.
If the matrix A r multiplicity of eigenvalue, the polynomial with the multiplicity
eigenvalues is differentiated r times and used as an additional equation, to have an
equal number of equations as the number of variables.

4. Using Modality Matrix

Let the diagonal matrix formed from eλi t ordered from smallest value of λi to largest
value of λi (ı.e. λ1 < λ2 < · · · < λn ), where λi are eigenvalues of A be denoted as
eΛt , then the state transition matrix can be solved using modality matrix M and
eΛt as
 
λ1 t
e 0 ··· 0
 
λ2 t
··· 0 
 
−1 Λt
 0 e
−1 
Φ(t) = M e M = M  . .. . M
 (1.13.8)
 .. . · · · .. 
 
λn t
0 0 ··· e

If the matrix has r multiplicity of eigenvalues, the determinant of the modality

matrix is zero since the eigenvectors are linearly dependent. So, this method will
not work in finding a state transition matrix for a matrix with r multiplicity of
eigenvalues.

1.14 Similarity Transformation

If there exist a transformation matrix that transform a state x into new state ξ, in such a
way that x = Tξ which means ξ = T−1 x and ẋ = Tξ̇ then the state space representation
of LTI system on equation 1.3.7 and equation 1.3.8 will be transformed into a dynamics
equation with different state coordinates of the form

ξ̇ = T−1 ATξ + T−1 Bu

ξ̇ = Āξ + B̄u (1.14.1)

y = CTξ + Du
y = C̄ξ + D̄u (1.14.2)

where Ā = T−1 AT,B̄ = T−1 B, C̄ = CT and D̄ = D.

The main purpose of similarity transformation is to simplify the dynamic equation of
the system for analysis (converting any state representation of any dynamic system to
controllable or observable canonical form will simplify the analysis of the system). Some

Chapter 1 50
Daniel Abebe Modern Control Systems Lecture Note

dynamic systems are complex in some coordinate systems, but simpler in other coordinate
systems.

Similarity transformation only changes the coordinates in which the dynamic system is
expressed. It does not change

• Eigenvalues of the system matrix

Proof:- The characteristic equation of the transformed system matrix is

det(λI − Ā) = 0
det(λI − T−1 AT) = 0
det(T−1 λT − T−1 AT) = 0
det(T−1 )det(λI − A)det(T) = 0
det(λI − A) = 0

• Transfer function of the dynamic equation

Proof:- The transfer function of the new dynamic system is

Y(s) C̄adj(sI − Ā)B̄

= + D̄
U(s) det(sI − Ā)
CTadj(sI − T−1 AT)T−1 B
= +D
det(sI − T−1 AT)
CTadj(T−1 sT − T−1 AT)T−1 B
= +D
det(T−1 sT − T−1 AT)
Cadj(sI − A)B
= +D
det(sI − A)

• Solution of output equation

Proof:- Solution of the new dynamic system have initial condition ξ o = T−1 xo , and
transfer function same as original dynamic system.

Y(s) = C̄(sI − Ā)−1 Tx0 + (C̄(sI − Ā)−1 B̄ + D̄)U(s)

= CT(sI − T−1 AT)−1 T−1 x0 + (CT(sI − T−1 AT)−1 T−1 B + D)U(s)
= C(sI − A)−1 x0 + (C(sI − A)−1 B + D)U(s)

Note Since the state of the system has changed solution to the state equation is different.
But we can derive a solution for the original state equation from the solution of the new
state equation by x(t) = Tξ(t).

Chapter 1 51
Daniel Abebe Modern Control Systems Lecture Note

1.15 Phase Plane and Stability Analysis of LTI Systems

1.15.1 Stability analysis of LTI systems
When we say a dynamic system is stable, we mean that the system will stay where it is
until it is disturbed by internal or external factors. The point where the system stays
forever after it arrives is called a stable equilibrium point. No change in a dynamic system
doesn’t guarantee us the existence of a stable equilibrium point or even the existence
of an equilibrium point at all. But it is a necessary condition for the existence of an
equilibrium point. So when we deal with the stability of LTI systems, we try to find
whether an LTI system has a stable equilibrium point or whether it can be stabilized at
preferred operating points. Stability of the dynamic system may be internal or external.
In the next section, we will explain them briefly.

Internal Stability
Internal stability deals with the stability of the states of the dynamic system. It analyzes
the behavior of the system state under no external inputs(zero-input response of the
states). Consider a dynamic system ẋ = Ax evolving from an initial condition x0 ;
the equilibrium point xeq (t) satisfies the state equation of the dynamic system. This
equilibrium point is

• Laypunov stable if and only if given ϵ > 0 there exist σ > 0 such that ||x0 || < σ
implies that ||x(t)|| < ϵ for all time t > 0. This means for every initial condition in
a sphere of radius σ, the final state of the system should be in a sphere of radius
ϵ. LTI system is Laypunov stable if and only if trace3 (A) < 0, det(A) ≥ 0 and
rank(A) = rank(A2 ) (i.e. All eigenvalues of system matrix A should have zero or
negative real parts).

• Asymptotically stable if it is stable and if it is possible to choose σ > 0 such that

||x0 || < σ implies lim ||x(t)|| = 0. If σ can be any real number(anywhere), then
t→∞
the system is globally asymptotically stable. LTI system is asymptotically stable if
and only if trace(A) < 0 and det(A) > 0 (i.e., all eigenvalues of system matrix A
should have strictly negative real parts).

• Exponentially stable if it is stable and there exist σ > 0, k > 0, λ > 0 such that
||x0 || < σ implies ||x(t)|| < ke−λt ||x0 || for all t > 0. If σ can be any real number
(anywhere), then the system is globally exponentially stable.

To determine whether the LTI system is Lyapunov stable or not, Let us take a quadratic
function V = x⊤ Px, where P are symmetric4 positive definite5 matrix, then

V̇ = x⊤ Pẋ + ẋ⊤ Px
= x⊤ P(Ax) + (Ax)⊤ Px
= x⊤ (PA + A⊤ P)x (1.15.1)
3
trace(A) is the sum of all diagonal elements of A
4
A matrix P is called symmetric if and only if P = P⊤
5
A matrix P is positive definite if and only if all eigenvalues of P are positive or leading principal
minors of the are positive

Chapter 1 52
Daniel Abebe Modern Control Systems Lecture Note

If (PA + A⊤ P) < 0 (i.e. negative definite) then V̇ decreases monotonically. If we take

positive definite symmetric matrix R such that,

PA + A⊤ P + R = 0 (1.15.2)

Equation 1.15.2 is called the Lyapunov equation.

BIBO/External Stability
It is a stability analysis of the response of the dynamic system to bounded input.
Definition:- A dynamic system is bounded-input input bounded-output, and stable if
and only if for zero initial condition, if bounded input always invokes bounded output.
Consider an LTI dynamic system with zero initial condition response
Z t
y(t) = C eA(t−τ ) Bu(τ )dτ + Du(t) (1.15.3)
0

Since u(t) is bounded, let ||u(t)|| < α, where α > 0 then

Z t Z t
A(t−τ )
y(t) = Ce Bu(τ )dτ + Du(t) < CeA(t−τ ) Bαdτ + Dα
0 0
Z t
<α CeA(t−τ ) Bdτ + D (1.15.4)
0

Which implies that forZ an LTI system to be BIBO stable the weighting matrix pattern
t
must be bounded (ı.e CeA(t−τ ) Bdτ < ∞).
0
For the LTI system, asymptotic stability always implies bounded-input input bounded-
output stability. While minimal bounded input and bounded output imply asymptotic
stability.

1.15.2 Phase plane analysis of LTI systems

Phase plane analysis is a graphical analysis of a second-order system concerning initial
conditions. Consider the following second-order system

ẋ1 = f1 (x1 , x2 )
ẋ2 = f1 (x1 , x2 ) (1.15.5)

where x1 and x2 are states of the system and f1 and f2 are linear (for LTI systems) and
nonlinear for nonlinear autonomous systems.
Since there is no control input, the phase plane analysis is used to understand the prop-
erties of the system under various initial conditions. For linear time invariant systems
equation 1.15.5 simplifies to

ẋ1 = a11 x1 + a12 x2

ẋ2 = a21 x1 + a22 x2 (1.15.6)

Geometrically, the state space of this system is a plane having x1 , x2 as coordinates. This
plane is called a phase plane. The solution of equation 1.15.5 or equation 1.15.6 with
time varying from zero to infinity can be represented as a curve in the phase plane. Such

Chapter 1 53
Daniel Abebe Modern Control Systems Lecture Note

a curve is called a phase plane trajectory. A family of phase plane trajectories is called
a phase portrait of a system.
Equation 1.15.6 can be written as
ẍ1 =a11 ẋ1 + a12 ẋ2 = a11 ẋ1 + a12 (a21 x1 + a22 x2 )
ẋ1 − a11 x1
ẍ1 =a11 ẋ1 + a12 (a21 x1 + a22 )
a12
(a11 a12 + a12 a22 ) (a12 a21 − a12 a22 a11 )
ẍ1 = ẋ1 + x1
a12 a12
ẍ1 +ξ1 ẋ1 + ξ2 x1 = 0 (1.15.7)
(a11 a12 + a12 a22 ) (a12 a21 − a12 a22 a11 )
where ξ1 = and ξ2 = .
a12 a12
The solution of equation 1.15.7 has the form
x(t) = k1 eλ1 t + k2 eλ2 t f or λ1 ̸= λ2 (1.15.8)
λ1 t λ2 t
x(t) = k1 e + tk2 e f or λ1 = λ2 (1.15.9)
A point in the phase plane where ẋ = 0 is called a singular point or equilibrium point of
the system. For a Linear system, we have only one singular/equilibrium point, namely
the origin. The trajectories in the vicinity of this singularity point can display quite
different characteristics, depending on the values of λ1 and λ2 . The following cases can
occur
• λ1 ,λ2 are both real and have the same sign (+ or -)
• λ1 ,λ2 are both real and have opposite sign
• λ1 ,λ2 are complex conjugates with non-zero real parts
• λ1 ,λ2 are complex conjugates with real parts equal to 0
The characteristics of the system under these cases are described below
1. The first condition corresponds to a node
(a) λ1 < 0,λ2 < 0 singularity point is called stable node. In this, every phase
portrait goes to the singularity point as shown in Figure 1.24, which means
the system is asymptotically stable.
(b) λ1 > 0,λ2 > 0 singularity point is called unstable node. In this, every phase
portrait goes away from the singularity point as shown in Figure 1.25, which
means the system is unstable.

Figure 1.24: Stable Node Figure 1.25: Unstable Node

2. λ1 < 0 < λ2 , corresponds to a saddle point. Trajectories tend to the equilibrium

point (converge to the singular point) at one end and move away from the equilib-
rium point (diverge to infinity) at the other end as shown in Figure 1.26.

Chapter 1 54
Daniel Abebe Modern Control Systems Lecture Note

Figure 1.26: Saddle Point

3. The third case corresponds to a focus.

(a) Re(λ1 < 0,λ2 < 0): stable focus, where trajectories circulate the point and
finally tend to the point as shown in Figure 1.27. The circulation shows that
there is an oscillation in the system response, and the convergence shows that
these oscillations will die out and the system becomes stable over time.
(b) Re(λ1 > 0,λ2 > 0): unstable focus, where trajectories circulate the point and
diverge to infinity as shown in Figure 1.28. The circulation shows that there
is an oscillation in the system response, and the divergence shows that these
oscillations grow large and the system becomes unstable over time.

Figure 1.27: Stable focus Figure 1.28: Unstable Focus

4. The last condition corresponds to a center point. All trajectories are ellipses, and
the singularity point is the center of these ellipses as shown in Figure 1.29.

Figure 1.29: Center Point

The stability characteristics of linear systems are uniquely determined by the nature of
their singularity points.

Chapter 1 55
Daniel Abebe Modern Control Systems Lecture Note

1.16 Examples
1. Find the controllable canonical form, observable canonical form, and diagonal nor-
mal form or Jordan canonical form of the following transfer function.
Y (s) 2s3 + 5s2 + 3s + 1
= 4
U (s) 4s + 2s2 + 3s + 6
0.5s3 + 1.25s2 + 0.75s + 0.25
=
s4 + 0.5s2 + 0.75s + 1.5
Solution:

(a) Controllable canonical form: since the transfer function is of order n = 4, we

have 4 states, 1 input, and 1 output. The constants are b4 = 0, b3 = 0.5, b2 =
1.25, b1 = 0.75, b0 = 0.25, a4 = 1, a3 = 0, a2 = 0.5, a1 = 0.75, a0 = 1.5

      
ẋ 0 1 0 0 x1 0
 1     
      
ẋ2   0 0 1 0 x2  0
 =   +  u
      
ẋ3   0 0 0 1 x3  0
      
ẋ4 −1.5 −0.75 −0.5 0 x4 1
 
x
 1
h i
x2 

y = 0.25 0.75 1.25 0.5 
 

x3 
 
x4

(b) Observable canonical form can be determined from controllable canonical form
by Ao = A⊤ ⊤ ⊤
c , Bo = Cc , Co = Bc and Do = Dc .
⊤

      
ẋ 0 0 0 −1.5 x 0.25
 1    1  
ẋ2  1 0 0 −0.75 x2  0.75
      
 =   +  u
ẋ3  0 1 0 −0.5  x3  1.25
      
      
ẋ4 0 0 1 0 x4 0.5

 
x
 1
h i 
x2 
y= 0 0 0 1  


x3 
 
x4

(c) Since the poles of the system are distinct, p1 = 0.71 + i,p2 = 0.71 − i, p3 =
−0.71 + 0.70i and p4 = −0.71 − 0.70i. We have a diagonal normal form. The

Chapter 1 56
Daniel Abebe Modern Control Systems Lecture Note

transfer function is factorized into

Y (s) 0.30 − 0.33i 0.30 + 0.33i −0.05 − 0.06i −0.05 + 0.06i
= + + +
U (s) s + 0.71 + i s + 0.71 − i s − 0.71 + 0.70i s − 0.71 − 0.70i

which has the Jordan canonical form of

      
ẋ −0.71 − i 0 0 0 x 1
 1    1  
−0.71 + i
      
ẋ2   0 0 0  x2  1
 =   +  u
0.71 − 0.70i
      
ẋ3   0 0 0  x3  1
      
ẋ4 0 0 0 0.71 + 0.70i x4 1
 
x
 1
h i  x2 


y = 0.30 − 0.33i 0.30 + 0.33i −0.05 − 0.06i −0.05 + 0.06i   

 x3 
 
x4

2. Find the state transition matrix for the following dynamic system
 
0 1
ẋ =  x
−2 −3

Solution

(a) Method 1:-Using Inverse Laplace transform

   −1 
s 0 0 1
Φ(t) = L−1 ((sI − A)−1 ) = L−1  −  
0 s −2 −3
 −1 
s −1
= L−1   
2 s+3
   
s −1
 adj   
2 s+3
 
−1 
 
=L    
 det s −1  
 

2 s+3

Chapter 1 57
Daniel Abebe Modern Control Systems Lecture Note

s+3 1 
 (s + 2)(s + 1) (s + 2)(s + 1) 
= L−1  −2 s 
(s + 2)(s + 1) (s + 2)(s + 1)
−1 2 −1 1
 
 + 2 + s + 1 s + 2 + s + 1 
= L−1  s−2 −2 2 −1 
+ +
 s + 2 s + 1 s +2 s + 1
2e−t − e−2t e−t − e−2t
Φ(t) =  
2e−2t − 2e−t 2e−2t − e−t

(b) Method 2:-Using Series expansion

     2
i=2
X (At) i 1 0 0 −1 0 −1
Φ(t) = = + t +   t2
i=0
i! 0 1 −2 −3 −2 −3
     
1 0 0 −1 −2 −3 2
= + t + 1  t
0 1 −2 −3 2 6 7
 
2 3 2
 1−t −t − t 
2
= 7 
−2t + 3t 1 − 3t + t2
2
2
i=2
at
X (at)i
From Taylor expansion of e = ,
i=0
i!
1 4
e−t = 1 − t + t2 , e−2t = 1 − 2t + t2
2 2
−t −2t −3 2
2
1 − t = 2e − e , −t + t = e−t − e−2t
2
−2t −t 7
2
− 2t + 3t = 2e − 2e , 1 − 3t + t2 = 2e−2t − e−t
2
So,
−3 2
 
2
1−t −t + t
 
−t −2t −t −2t
 2  2e − e e − e
Φ(t) = 
 = 
7 2 −2t −t −2t −t
−2t + 3t2 1 − 3t + t 2e − 2e 2e − e
2
(c) Method 3:-Using Cayley Hamilton Theorem
Determining the eigenvalues
det (λI − A) = (λ + 2)(λ + 1) = 0
Which result in eigenvalues of matrix A, λ1 = −2 and λ2 = −1. Hence the
simultaneous equations from eλi t = α0 + α1 λi are
e−2t = α0 − 2α1
e−t = α0 − α1

Chapter 1 58
Daniel Abebe Modern Control Systems Lecture Note

Solving for α0 and α1 , gives a result

α0 = 2e−t − e−2t , α1 = e−t − e−2t

So by Cayley Hamilton’s Theorem

Φ(t) = α0 I + α1 A
   
1 0 0 −1
= (2e−t − e−2t )   + (e−t − e−2t )  
0 1 −2 −3
 
−t −2t −t −2t
2e − e e −e
Φ(t) =  
−2t −t −2t −t
2e − 2e 2e − e

(d) Method 4:-Using Modality Matrix

We have already determined the eigenvalues; the next step is to determine the
eigenvector corresponding to each eigenvalue by using (λi I − A)ν i = 0. The
eigenvector corresponding to λ1 = −2, ν 1 is
     
−2 0 0 −1 ν
 −   11  = 0
0 −2 −2 −3 ν12
  
−2 1 ν
   11  = 0 ⇒ 2ν11 = ν12
−2 1 ν12
   
ν11 1
ν1 =   = m  
ν12 2

The eigenvector corresponding to λ2 = −1, ν 2 is

     
−1 0 0 −1 ν
 −   11  = 0
0 −1 −2 −3 ν12
  
−1 1 ν
   21  = 0 ⇒ ν21 = ν22
−2 2 ν22
   
ν21 1
⇒ ν2 =   = m  
ν22 1

Hence the modality M matrix, it’s inverse M−1 and eΛt are
     
−2t
1 1 −1 1 e 0
M=  , M−1 =  , eΛt =  ,
2 1 2 −1 0 e−t

Chapter 1 59
Daniel Abebe Modern Control Systems Lecture Note

Finally
   
−2t
−1 1 e 0 1 1
Φ(t) = M−1 eΛt M =    
−t
2 −1 0 e 2 1
 
2e−t − e−2t e−t − e−2t
=  
−2t −t −2t −t
2e − 2e 2e − e

3. Find the transfer function of the following state space representation

   
1 0 −1 0
   
ẋ = −2 2 −5 x + 1 u
   
   
0 1 −1 0
h i
y = 0 1 1 x + 2u

Solution

Y (s) Cadj(sI − A)B + det(sI − A)D

=
U (s) det(sI − A)

  
2
s − s + 3 −1 2−s 0
h i  
0 1 1  −2s − 2 s2 − 1 7 − 5s  1
  
  
2
−2 s − 1 s − 3s + 2 0
= +2
s3 − 2s2 + 4s − 5

s2 + s − 2
= +2
s3 − 2s2 + 4s − 5

Y (s) 2s3 − 3s2 + 9s − 12

= 3
U (s) s − 2s2 + 4s − 5

4. Find the eigenvalues and their corresponding eigenvectors for a matrix

 
1 −1
A= 
0 −1
Solution
The characteristics equation of the matrix A determined from det(λI − A) = 0
   
λ 0 1 −1
det  −  = 0
0 λ 0 −1
(λ1 − 1)(λ2 + 1) = 0
λ1 = 1, λ2 = 2

Chapter 1 60
Daniel Abebe Modern Control Systems Lecture Note

The eigenvector corresponding to λ1 = −1, ν 1 is

(λ1 I − A)ν 1 = 0
     
−1 0 1 −1 ν
 −   11  = 0
0 −1 0 −1 ν12
  
−2 1 ν
   11  = 0
0 0 ν12

Which results in

−2ν11 + ν12 = 0 ⇒ ν12 = 2ν11

Hence ν 1
   
ν11 1
ν1 =   = m  
ν12 2

where m is any real number.

The eigenvector corresponding to λ2 , ν 2 is

(λ2 I − A)ν 2 = 0
     
1 0 1 −1 ν
 −   21  = 0
0 1 0 −1 ν22
  
0 1 ν
   21  = 0
0 2 ν22

Which results in

ν22 = 0, 2ν22 = 0

ν21 can be any value and ν22 = 0. So ν 2 is

   
ν21 k
ν2 =   = m  
ν22 0

5. Determine the eigenvalues, right eigenvector, and left eigenvectors of the following
matrix
 
1 2
A= 
3 0

Chapter 1 61
Daniel Abebe Modern Control Systems Lecture Note

Solution

  
  
λ 0
1 2 λ − 1 −2
det( −  = det   = 0
0 λ 3 0 −3 λ
λ(λ − 1) + 5 = 0
λ2 − λ + 6 = 0 ⇒ (λ − 3)(λ + 2) = 0
λ1 = 3 λ2 = −2
the eigenvectors associated with λ1 = 3, ν 1 is
(λI − A) ν 1 = 0

        
3 0 1 2 ν11 2 −2 ν11
 −    = 0 ⇒    = 0
0 3 3 0 ν12 −3 3 ν12

which results in
2ν11 − 2ν12 = 0, −3ν11 + 3ν12 = 0
   
ν11 1
ν11 = ν12 = 1 ν 1 ⇒   = ν11  
ν11 1

the eigenvectors associated with λ2 = −2, ν 2 is

(λI − A) ν 2 = 0

       
−2 0 1 2 ν11 −3 −2 ν21
 −    = 0 ⇒    = 0
0 −2 3 0 ν12 −3 −2 ν22

which results in
− 3ν21 − 2ν22 = 0, −3ν21 − 2ν22 = 0
 −2   
−2 ν22 −2
ν21 = ν22 ν2 ⇒  3  = ν22  
3 ν22 3 3

the modality matrix M is

 
h i 1 −2
M = ν1 ν2 = 
1 3

6. For the following matrix determine the state transition matrix Φ(t), eigenvectors
 
−6 2
A= 
−6 1

Chapter 1 62
Daniel Abebe Modern Control Systems Lecture Note

Solution

     
λ 0 −6 2 λ + 6 −2
det( −  = det   = 0
0 λ −6 1 6 λ−1
(λ + 6)(λ − 1) + 12 = 0
λ2 5λ + 6 = 0 ⇒ (λ + 3)(λ + 2) = 0
λ1 = −3 λ2 = −2

the eigenvectors associated with λ1 = −3, ν 1 is

(λI − A) ν 1 = 0

    
ν11 −6 2 ν
−3   =    11 
ν12 −6 1 ν12

which results in

− 6ν11 + 2ν12 = −3ν11 , −3ν11 + 2ν12 = 0

− 6ν11 + ν12 = −3ν12 − 6ν11 + 4ν22 = 0
3ν11 = 2ν11 ⇒ ν11 = 2, ν12 = 3
 
2
⇒ ν1 =  
3

the eigenvectors associated with λ2 = −2, ν 2 is

(λI − A) ν 2 = 0

[λI − A] ν 1 = 0

    
ν11 −6 2 ν
−2   =    21 
ν12 −6 1 ν22

which results in

− 6ν21 + 2ν22 = −2ν11 , −3ν11 + 2ν12 = 0

− 6ν21 + ν22 = −2ν12 − 6ν11 + 4ν22 = 0
3ν22 = 2ν21 ⇒ ν21 = 1, ν22 = 2
 
1
⇒ ν2 =  
2

Chapter 1 63
Daniel Abebe Modern Control Systems Lecture Note

the modality matrix M and it inverse is

   −1  
h i 2 1 2 1 2 −1
M = ν1 ν2 =   M−1 =   = 
3 2 3 2 −3 2

The diagonal of matrix Ā = M−1 AM

     
2 −1 −6 2 2 1 −3 0
Ā =    = 
−3 2 −6 1 3 2 0 −2

the state transition matrix Φ(t) is given as

   
λ1 t
2 1 e 0 2 −1
Φ(t) = MeΛt M−1 =    
3 2 0 eλ2 t −3 2
   
−3t
2 1 e 0 2 −1
Φ(t) =    
3 2 0 e−2t −3 2
 
−3t −2 −3t −2t
4e − 3e −2e + 2e
Φ(t) =  
−3t −2 −3t −2t
6e − 6e −3e + 4e

7. Given matrix A find Φ(t)

 
−1 0 0
 
A =  0 −1 0 
 
 
1 0 −1

Solution:
The characteristic equation of the matrix is
 
λ+1 0 0
 
det  0 0  = 0
 
λ+1
 
−1 0 λ+1
(λ + 1)(λ + 1)(λ + 1) = 0
⇒ λ1 = λ2 = λ3 = −1

When any matrix has a multiplicity of eigenvalues, the method we don’t rely on is
using the modality matrix since the determinant becomes zero. So let us use the

Chapter 1 64
Daniel Abebe Modern Control Systems Lecture Note

Laplace inverse transform.

 −1 
s+1 0 0
  
Φ(t) =L−1 (sI − A)−1 = L−1  0

  
s+1 0  
  
−1 0 s+1
 
1
 s + 1 0 0 
 1 
= L−1  0 0 
 
s+1

 
 1 1 
− 0
(s + 1)2 s+1
 
e−t 0 0
 
=  0 e−t 0 
 
 
−t −t
te 0 e

8. For the following dynamic system, find eigenvalues, zero-input response, and zero-
state response to step input for t > 0.
   
−1 −3 5 h i
ẋ =   x +   u, y = 3 4 x
0 −2 6
h i⊤
with initial condition x(0) = 4 5 .
Solution:
The characteristic equation of the system is,

||λI − A|| = 0 ⇒ (λ + 1)(λ + 2) = 0 ⇒ λ1 = −1, λ2 = −2

zero input response yzinput is

 
−1 adj(sI − A)
yzinput =L C x0
det(sI − A)
1 −3 3
    
+  4 
= L−1 C  s + 1 s + 1 1 s + 2   
 
0 5
 s + 2  
h i e−t −3e−t + 3e−2t 4
= 3 4   
0 e−2t 5
= −33e−t + 65−2t

Chapter 1 65
Daniel Abebe Modern Control Systems Lecture Note

zero state response yzs will be

 
−1 adj(sI − A)
yzs = L C BU (s)
det(sI − A)
1 −3 3
    
h i +  5 1
= L−1  3 4  s + 1 s + 1 1 s + 2    
 
0 6 s
 s + 2   
5
 
3 −9 9 4   1
= L−1  +
s+1 s+1 s+2s+2 6 s
 
−1 15 78 54
=L + −
s(s + 1) s(s + 2) s(s + 1)
 
−1 39
=L
(s + 1)(s + 2)
 
−1 39 39
=L −
s+1 s+2
= 39e−t − 39e−2t
the total response of the system is y = yzinput + yzs ,
y = 6e−t − 26e−2t

9. Check the stability of the following system matrix

 
3 −2
A= 
1 −1

Solution:
using Lyapunov stability test: use R = I
AP + PA⊤ = −R
       
3 −2 p11 p12 p p12 3 1 −1 0
   +  11  = 
1 −1 p21 p22 p21 p22 −2 −1 0 −1

using the following command on MATLAB:

syms p11 p12 p21 p22
A = [3 − 1; 1 − 1]; R = eye(2); P = [p11 p12 ; p21 p22 ];
′
eqn = AP + PA + R
[p11 p12 ; p21 p22 ] = solve(eqn == 0);
P = [p11 p12 ; p21 p22 ];
The result is
 
1
7/4
P= 
7/4 9/4

Chapter 1 66
Daniel Abebe Modern Control Systems Lecture Note

−13
P is symmetric but not positive definite because det(P) = < 0. Hence, the A
16
is not stable. Checking the eigenvalues
√ √
||λI − A|| = 0 ⇒ (λ − 3)(λ + 1) + 2 = 0 ⇒ λ1 = 1 − 2, λ2 = 1 + 2
both eigenvalues has strictly positive real parts, so A is unstable.
10. For dc motor model derive on 1.7.1, check stability with Ra = 0.25, J = 0.0025, La =
0.003, kb = 0.0067, kt = kb .

Solution:
using MATLAB command [ν, D] = eig(A), give us the eigenvaluesλ1 = −83.25, λ2 =
−8.08 and corresponding eigenvectors of the dc-motor dynamics
   
−1 0.03
ν1 =   ν2 =  
0.034 −1
Since all eigenvalues have strictly negative real parts, the dynamic system is stable.
11. Check stability of IPC dynamic system derived on 1.6.3, with parameters mp =
0.5, mc = 0.2, b = 0.1, l = 0.3, j = 0.5mp l2 , g = 9.81.
Solution:
using MATLAB command [ν, D] = eig(A), give us the eigenvalues λ1 = 0, λ2 =
−0.14 λ3 = −5.21, λ4 = 5.17 and corresponding eigenvectors of the IPC dynamic
system
       
1 −0.99 −0.01661 0.0158
       
       
0  0.1414   0.0862  0.0816
ν1 =   ν2 = 
  
 ν3 = 
 
 ν4 = 
  
 −0.1879 

0  0.0021  0.1891
       
0 −0.0003 0.9782 0.9784
Since not all eigenvalues doesn’t have strictly negative real parts, the dynamic
system is unstable.
12. Consider the following simple mass measuring system, with system parameters k =
20N/m, ms = 1kg, and d = 9N s/m shown in Figure C.4b. (Mid-Exam 2021)

ms x Scale
k d

Figure 1.30: Simple Mass Measuring System

The system shows the mass of the object to be measured on the scale. The reading
on the scale is proportional to the mass of the object(y = αmo ). Where α is a
proportionality constant (for this problem, use α = 1).

Chapter 1 67
Daniel Abebe Modern Control Systems Lecture Note

(a) Derive a state-space representation of the system.

(b) Find the eigenvalues and eigenvectors of the system.
(c) Determine whether the system is stable or not.
(d) Determine the state transition matrix of the system.
(e) Determine zero input response of the system for initial condition x(0) =
h i⊤
1 0 .
(f) Determine the zero state response of the system under a step input.
(g) If the maximum compression of the spring is 10cm, determine the maximum
mass the system can handle.
(h) If you are the designer and you want to increase the maximum mass the system
can handle 5 times,
• What is your best engineering solution?
• Why do you select the solution?
• Determine the eigenvalues for your new system.
• What do you conclude from the result?

Solution:

(a) From the free body diagram of the system

ms x
fk fd

Figure 1.31: FBD of Simple Mass Measuring System

u − kx − dẋ = ms ẍ
−k d u
ẍ = x− ẋ +
ms ms m
Let x1 = x, x2 = ẋ = ẋ1 , then the state space representation of the system is:
      
ẋ 0 1  x1 0
 1 =   −k −d    +  1  u
ẋ2 x2
ms ms m
u
The output is y = αmo , but mo = (since u is the weight of the measured
g
αu
mass), which means y = . The scale shows the result after the transient
g
response dies out (under steady-state conditions). At steady-state conditions,

Chapter 1 68
Daniel Abebe Modern Control Systems Lecture Note

the only forces on the system are the spring force and the weight of the system.
Hence,

u = mg = kx
 
αkx αkx1
y= ⇒ y=
g g
 
x1
 
αk
y= 0  
g x2

with the given parameters

      
ẋ1 0 1 x 0
 =   1 +   u
ẋ2 −20 −9 x2 1
 
h i x1
y= 2 0  
x2

(b) Eigenvalues and vectors of the system are

   
λ 0 0 1
||λI − A|| = 0 ⇒  −  = 0
0 λ −20 −9
λ(λ + 9) + 20 = 0
(λ + 4)(λ + 5) = 0
λ1 = −4 λ2 = −5

The eigenvector corresponding to λ2 = −5, (ν 1 ) is

         
ν11 0 1 ν ν 0 1 ν
λ1   =    11  ⇒ −5  11  =    11 
ν12 −20 −9 ν12 ν12 −20 −9 ν12
5ν11 = −ν12 20ν11 = −4ν12
 
1
ν 1 = ν11  
−5

The eigenvector corresponding to λ2 = −4, (ν 2 ) is

         
ν21 0 1 ν ν 0 1 ν
λ2   =    21  ⇒ −4  21  =    21 
ν22 −20 −9 ν22 ν22 −20 −9 ν22
− 4ν21 = ν22 20ν21 = −5ν52
 
1
ν 2 = ν21  
−4

Chapter 1 69
Daniel Abebe Modern Control Systems Lecture Note

(c) Stability analysis

Since both eigenvalues are strictly negative, the system is stable.
(d) state transition matrix

Φ(t) = L−1 (sI − A)−1

   
s+9 1
   
−20 s
   
−1 adj(sI − A) −1 
 
=L =L  
det(sI − A)  (s + 5)(s + 4) 

 

 s+9 1 
 (s + 5)(s + 4) (s + 5)(s + 4) 
= L−1  −20 s 
(s + 5)(s + 4) (s + 5)(s + 4)

20 5 −1 1
 
+ +
= L−1  s−20
 + 5 s + 4 s + 5 s + 4 
5 5 −4 
+ +
 s+5 s+4 s + 5 s + 4
−4e−5t + 5e−4t −e−5t + e−4t
=  
−20e−5t + 20e−4t −5t
5e − 4e −4t

(e) Zero input response

yzi (t) = CΦ(t)xo

  
h i −4e−5t + 5e−4t −e−5t + e−4t 1
= 2 0   
−20e−5t + 20e−4t 5e−5t − 4e−4t 0
= −8e−5t + 10e−4t

(f) Zero state response for step input

yzs (t) = L−1 C(sI − A)−1 BU (s)

 
−1 Cadj(sI − A)BU (s)
=L
det(sI − A)
  s+9 1   

= L−1  2 0  (s + −20 5)(s + 4) (s + 5)(s + 4)  0 1 

h i
 
s
1 s
 
(s + 5)(s + 4) (s + 5)(s + 4)
   
−1 2 −1 0.1 0.4 0.5
=L =L + −
s(s + 5)(s + 4) s s+5 s+4
= 0.1 + 0.4e−5t − 0.5e−4t

Chapter 1 70
Daniel Abebe Modern Control Systems Lecture Note

(g) The maximum mass of an object that the system can handle is

kxmax 20N/m ∗ 0.1m

momax = = = 0.2kg
g 10m/s2

(h) If we want to increase the maximum mass the system can measure, we have to
replace the spring with another spring having a higher spring constant, because
the spring force is the only force that supports the weight of the object at a
steady state. But this will affect the sensitivity of the system, which has to be
calibrated again. In this case, we have to replace the spring with a constant
kn = 5k = 100N/m.
The eigenvalues of the new system are

Figure 1.32: System response with k = 20, d = 9 and k = 100, d = 20

   
λ 0 0 1
||λI − A|| = 0 ⇒  −  =0
0 λ −100 −9
λ(λ + 9) + 100 = 0
λ1 = −4.5 + j8.93 λ2 = −4.5 − j8.93

The system is stable, but there will be an oscillation since the poles are complex
conjugates. To solve this problem, we have to use a damper with a higher
damping coefficient. If we take a damper with a damping coefficient d =
20(the minimum value for this case). The eigenvalues of the system will be
λ1 = λ2 = −10. This damper will remove the oscillation.

Chapter 1 71
Daniel Abebe Modern Control Systems Lecture Note

Figure 1.33: System Response with k = 100 d = 9 and d = 20

1.17 Exercises
1. Find the characteristic polynomial, eigenvalues, and the corresponding eigenvectors
for each of the following matrices.
   
1 4 2 2 0
(a) A =    
2 3 (e) F = 1 2 1
 
   
2 −2 1 2 1
(b) C =    
2 2 1 1 1
   
(f) H =
 
2 2  1 1 1 
(c) D =    
−1 −1 1 1 1
   
0 −1 cos θ sin θ
(d) E =   (g) I =  
ab a + b − sin θ cos θ

2. For the following matrix find state transition matrix Φ(t)

   
1 0 1 1
(a) A =   (e) E =  
0 2 1 1
   
0 1 0 1 1
(b) B =    
0 −0 (f) F = 0
 
0 1
   
0 1 0 0 1
(c) C =    
−1 0 2 0 0
   
(g) G = 0 −3 0
 
−1 1
(d) D =    
0 −1 0 0 7

3. Find the general solution to the following initialized systems with step input.

Chapter 1 72
Daniel Abebe Modern Control Systems Lecture Note

(a)
     
−1 −1 1 h i 0.5
ẋ =   x +   u, y = 0 1 x, x0 =  
2 −4 0 0.5

(b)
     
2 −1 0 h i 0
ẋ =   x +   u, y = 1 1 x, x0 =  
−1 2 1 0.5

(c)
     
0 1 1 h i 1
ẋ =   x +   u, y = 1 0 x, x0 =  
1 0 1 1

(d)
     
1 0 −1 1 0
    h i  
ẋ = 1 2 1  x + 0 u, y = 0 0 1 x, x0 = 1
     
     
2 2 3 0 0

4. For a simple harmonic system described by

z̈ + ωz = 0

determine the state transition matrix.

5. For a dynamic system described by

ẋ1 = x1 + 4x2 + u
ẋ2 = 3x1 + 2x2

determine the state transition matrix.

6. For LTI system

     
−1 −4 1 1
ẋ =   x +   u, x0 =   & u = e2t
−1 −1 1 2

Find the state transition matrix and the solution of the state equation

7. Determine the range of values of k for which the polynomial

(3 − k)λ3 + 2λ2 + (5 − 2k)λ + 2 = 0

is asymptotically stable.

Chapter 1 73
Daniel Abebe Modern Control Systems Lecture Note

8. Determine the range of values of k, which makes the dynamic system

ẋ1 = −kx1 − 3x2

ẋ2 = kx1 − 2x2

is asymptotically stable.

9. Investigate the stability of

(a)

ẋ1 = −x1 + x2
ẋ2 = (x1 + x2 ) sin k − 3x2

(b)

ẋ1 = −kx1 + x2
ẋ2 = −ax1 − bx2

10. Show that the system described by

   
0 −1 0 0
   
ẋ = 1 0 0  x + 0 u y = x1
   
   
0 0 −1 1

Is BIBO stable but not asymptotically stable?

11. Consider an RLC Circuit in Figure B.3, with system parameters L = 0.1H, C =
0.1F and R = 1Ω. If the designer aims to regulate the voltage across the capacitor.

R L

vs C vo

Figure 1.34: Series RLC circuit

(a) Derive a state space representation of the system .

(b) Find the eigenvalues and eigenvectors of the system.
(c) Determine whether the system is stable or not.
(d) Determine the state transition matrix of the system.
(e) Determine zero input response of the system for initial condition x(0) =
h i⊤
0 1 .

Chapter 1 74
Daniel Abebe Modern Control Systems Lecture Note

(f) Determine the zero state response of the system under step u = e−2t .
(g) What is the maximum value of voltage the capacitor can be charged to?
(h) If the output is the voltage across the inductor, determine the state space
representation of the system .
(i) Compare the result on 11h with the result on 11a and state their difference.
(j) Determine the zero state response of the state space representation derived in
part 11a and part 11h for u = cos t.

Chapter 1 75
Chapter 2

Controllability and Observability of

LTI Systems

2.1 Controllability of LTI Systems

In order to do whatever we want with the given dynamic system under control input, the
system must be controllable.

Definition:-A linear time invariant system with state space representation on equation
1.3.6 and equation 1.3.7 is completely controllable if for any t0 any initial state x(t0 ) = x0
can be transferred to a final state x(tf ) at tf using control input u (i.e. if there exist a
control signal u that transfer a state of a dynamic system from any initial condition x0
to a final state x(tf )).

Consider A linear time-invariant system on equation 1.3.7 with initial condition x0 at

time t0 = 0. We want to bring the system to the zero final state x(t1 ) = 0 at time t1 .
From the solution to the state equation
Z t1
At
x(t1 ) = e x0 + eA(t−τ ) Bu(τ )dτ
Z0 t1
0 = eAt x0 + eA(t−τ ) Bu(τ )dτ
0
Z t1
x0 = − e−Aτ Bu(τ )dτ (2.1.1)
0

and from Cayley-Hamilton theorem

k=n−1
X
−Aτ
e = α(τ )Ak (2.1.2)
k=0

Replacing equation 2.1.2 in equation 2.1.1 an rearranging we have

k=n−1
X Z t1
k
x0 = − A B α(τ )u(τ )dτ (2.1.3)
k=0 0

76
Daniel Abebe Modern Control Systems Lecture Note
Z t1
Letting α(τ )u(τ )dτ = βk .
0
we have
k=n−1
X
x0 = − Ak Bβk (2.1.4)
k=0

 
β
 1
i
β2 

.. .. .. ..
h
2 n−1
x0 = − B . AB . A B . · · · . A B  . 
 (2.1.5)
 .. 

 
βk

If the system is completely controllable, given any initial state x0 , equation 2.1.4 must
be satisfied, which requires the matrix
. . . .
h i
C o = B .. AB .. A2 B .. · · · .. An−1 B (2.1.6)

to have full rank. This imply that the vectors B, AB, A2 B, · · · , An−1 B should be linearly
independent. The matrix in equation 2.1.6 is called the controllability matrix. If the
controllability matrix has full rank, then the system defined by (A, B) is completely
controllable.
Or if a matrix Q is right Lyapunov and the solution for Q of equation 2.1.7

AQ + QA⊤ + BB⊤ = 0 (2.1.7)

is positive definite then the system defined by (A, B) is completely controllable. Q

is determine from co-variance of x(t) which is called controllability grammian matrix,
defined as
Z ∞ Z ∞
⊤ ⊤
Q= x (t)x(t)dt = (eA t B⊤ BeAt )dt (2.1.8)
0 0

Chapter 2 77
Daniel Abebe Modern Control Systems Lecture Note

2.2 Observability of LTI Systems

To see what is going on inside a dynamic system under observation, the system must be
completely observable.

Definition:- A linear time invariant system with state space representation on equation
1.3.6 and equation 1.3.7 is completely observable if for any t0 and any initial state x0 ,
there exist finite time t1 > t0 such that the knowledge of u(·) and y(·) for t ∈ [t0 , t1 ]
suffices to determine x0 properly.
From the solution to the output equation
Z t1
At
y(t) = Ce x0 + CeA(t−τ ) Bu(τ )dτ + Du(t) (2.2.1)
0

Knowing u(t) and y(t) will give information about x0 , if the dynamic system defined by
(A, C) is completely observable. Let the input be zero to derive the required condition.
This means the system response is only the response to an initial condition. So,
y(t) = CeAt x0 (2.2.2)
k=n−1
X
At
Replacing e = α(t)Ak
k=0

k=n−1
X
y(t) = C α(t)Ak x0
k=0
y(t) = α0 (t)Cx0 + α0 (t)Cx0 + α0 (t)CA2 x0 + · · · + α0 (t)CAn−1 x0

 
C
 
 CA 
 
h i  
 2
y(t) = α0 (t) α1 (t) · · · αn−1 (t)  CA  x0

(2.2.3)
 .. 
 
 . 
 
n−1
CA

Thus, complete observability requires

 
C
 
 CA 
 
 
 2
O v =  CA 

(2.2.4)
 .. 
 
 . 
 
n−1
CA

to have full rank. This implies that the co-vectors1 C, CA, CA2 . . . CAn−1 should be
linearly independent. The matrix in equation 2.2.4 is called the observability matrix. If
1
A co-vector (or dual vector ) is an element of the dual space V ∗ of a vector space V . It is a linear
functional that maps vectors to scalars.

Chapter 2 78
Daniel Abebe Modern Control Systems Lecture Note

the observability matrix has full rank, then the system defined by (A, C) is completely
observable.
Or if a matrix P is left Lyapunov and the solution for P of the equation 2.2.5

A⊤ P + QA + C⊤ C = 0 (2.2.5)

is positive definite then the system defined by (A, C) is completely observable. P is de-
termined from co-variance of y(t) which is called observability grammian matrix, defined
as
Z ∞ Z ∞
⊤ ⊤
Q= y (t)y(t)dt = (eA t C⊤ CeAt )dt (2.2.6)
0 0

2.3 Modality Tests

A mode of matrix A is a transformation that converts the matrix into a diagonal form.
This transformation is composed of eigenvectors corresponding to the eigenvalues of the
matrix. Hence modality test uses this concept to analyze the controllability and observ-
ability of a linear time-invariant system.

Consider a linear time invariant system on equation 1.3.6 and equation 1.3.7, if there
is a transformation V that converts a matrix A to a diagonal matrix Λ. Where

V−1 ΛV = A (2.3.1)

Where Λ is a diagonal matrix composed of eigenvalues of matrix A placed either in

decreasing or increasing order.
 
λ1 0 0 . . . 0
 
 
 0 λ1 0 . . . 0 
Λ=
 ..

 (2.3.2)
. 
 
0 0 0 . . . λ1

The system is completely controllable if and only if no left (row) eigenvectors of A are
orthogonal to B. If ν is left eigenvector of matrix A corresponding to any eigenvalue λ,
for the system to be completely controllable

ν · B ̸= 0 (2.3.3)

The system is completely observable if and only if no right (column) eigenvectors of A are
orthogonal to C. If q is a right eigenvector of matrix A corresponding to an eigenvalue
λ, for the system to be completely observable

C · q ̸= 0 (2.3.4)

These tests are also called PHB (Popov Belvitch Hautus) tests.

Chapter 2 79
Daniel Abebe Modern Control Systems Lecture Note

2.4 Kalman Decomposition of LTI Systems

If the system is not completely controllable, the system can be decomposed into its con-
trollable states and uncontrollable states using a similarity transformation. Consider LTI
system having r controlled states xc , and n − r uncontrolled states xc̄ , with a transfor-
mation matrix T, the system can be transformed into
      
ẋ A A12 x B
 c =  c   c +  c u (2.4.1)
ẋc̄ 0 Ac̄ xc̄ 0
 
h i xc
y = Cc Cc̄   + Du (2.4.2)
xc̄
The Block diagram representation of controllable and uncontrollable modes decomposi-
tion is shown in Figure 2.1.

u + xc y
Bc ʃ Cc
+

+ Ac
+
A12

Ac

+ xc y
+
0 ʃ Cc

Figure 2.1: Block Diagram Representation of Controllable and Uncontrollable Modes of a System

Similarly, if a system is not completely observable, the system can be decomposed into
its observable states and unobservable states using a similarity transformation. Con-
sider LTI system having r observable states xo , and n − r unobservable states xō , with a
transformation matrix T, the system can be transformed into
      
ẋo Ao 0 x B
 =   o +  o u (2.4.3)
ẋō A21 Aō xō Bō
 
h i xo
y = Co 0   + Du (2.4.4)
xō
The Block diagram representation of observable and unobservable modes decomposition
is shown in Figure 2.2.

Chapter 2 80
Daniel Abebe Modern Control Systems Lecture Note

u + xo y
Bo ʃ Co
+

Ao

A21
+ +
Ao

+ + xo
Bo ʃ 0

Figure 2.2: Block Diagram Representation of Observable and Unobservable Modes of a System

Kalman Decomposition via Eigen-structure

The eigen-structure is utilized to decompose the state space into controllable and uncon-
trollable subspaces, as well as observable and unobservable subspaces:

1. Controllable and Uncontrollable Subspaces:

• The controllable subspace is spanned by the right eigenvectors qi whose cor-

responding left eigenvectors ν i satisfy ν i B ̸= 0.
• The uncontrollable subspace consists of the modes where ν i B = 0. These
modes cannot be controlled through any input.

2. Observable and Unobservable Subspaces:

• The observable subspace is defined by the right eigenvectors qi for which Cqi ̸=
0. These modes can be detected through the output matrix C.
• The unobservable subspace comprises right eigenvectors qi where Cqi = 0,
indicating that these states cannot be inferred from output measurements.

Kalman decomposition is essential for system analysis and design, as it enables us to

focus on the controllable and observable parts of the system while understanding the
limitations imposed by uncontrollable and unobservable dynamics.

Chapter 2 81
Daniel Abebe Modern Control Systems Lecture Note

2.4.1 Weaker Conditions

If the uncontrollable mode of the system is stable, or conversely, if the unstable mode is
controllable, then the system is said to be stabilizable.

If the unobservable mode of the system is stable, or conversely if the unstable mode is
observable, then the system is said to be detectable.

Stabilizability and detectability are called weak conditions since it doesn’t impose tight
conditions like controllability and observability on system dynamics. If the system state
is uncontrollable and stable, there is no need to put additional restrictions since the rest
of the states are controllable. Similarly, as long as the unstable states are observable,
there is no need to add a restriction on unobservable stable states.

Stability of uncontrollable states can be deduced from the eigenvalues of the uncontrol-
lable subsystem matrix (Ac̄ , while the stability of unobservable dynamics is deduced
from the eigenvalues of the unobservable subsystem matrix Aō ). If the matrices have
eigenvalues with a negative real part, then the dynamics are stable.

2.4.2 Transformation Matrix for Kalman Decomposition

To decompose a system into controllable and uncontrollable or observable and unobserv-
able subspaces, we need a transformation matrix T. The transformation matrix T is
composed of the basis vectors that span the controllable and observable space of the
system their orthogonal complements. Thus, we need to recap basic concepts in matrix
space.

Recap on the Four Fundamental Subspaces of a Matrix

An m×n matrix operates between two vector spaces: it takes inputs from Rn (all possible
column vectors with n components) and produces outputs in Rm . The matrix’s behavior
can be completely understood by examining four special subspaces (vector spaces within
these larger spaces) that reveal how the matrix transforms vectors.

1. Column Space (or Range): The column space, denoted Col(A), consists of all
possible linear combinations of the columns of A. It lives in Rm and represents all
possible outputs the matrix can produce when multiplied by a vector. Its dimension
is equal to the rank of A, which tells us how many linearly independent columns
A has. Essentially, it shows us where the matrix can ”reach” in the output space.

Col(A) = {y ∈ Rm | y = Ax for some x ∈ Rn } .

2. Null Space (or Kernel): The null space, denoted Null(A), contains all vectors
x in Rn that get mapped to the zero vector when multiplied by A. It shows us
which inputs completely disappear when transformed by A. The dimension of
the null space is called the nullity and equals n − rank(A). This tells us how much
information gets ”lost” when the matrix transforms vectors. The null space consists
of all eigenvectors for λ = 0 (plus the zero vector) for a square matrix.

Ax = 0.

Chapter 2 82
Daniel Abebe Modern Control Systems Lecture Note

3. Row Space: The row space is the span of all row vectors of A and lives in Rn .
Surprisingly, it always has the same dimension as the column space - both equal to
the rank of A. This space represents all possible linear combinations of the matrix’s
rows, which corresponds to all possible ”measurements” or ”constraints” that the
matrix’s rows can represent.

Row(A) = x ∈ Rn | x = A⊤ y for some y ∈ Rm .



4. Left Null Space: This less obvious space, denoted Null(A⊤ ), consists of all vectors
y in Rm that get mapped to zero when multiplied by A⊤ (the transpose of A).
These vectors are orthogonal to all columns of A, and the space has dimension
m − rank(A). It represents all possible ”constraints” or ”tests” that vectors in the
output space must satisfy to be in the column space of A.

A⊤ y = 0.

These spaces are connected:

• The row space and null space are perpendicular complements in Rn (Null(A) ⊥
Row(A) ), while the column space and left null space are perpendicular comple-
ments in Rm (Col(A) ⊥ Null(A⊤ )).
• Their dimensions add up perfectly following the Rank-Nullity Theorem: dim(Row(A))+
dim(Null(A)) = n and dim(Col(A)) + dim(Null(A⊤ )) = m. These relationships
form the Fundamental Theorem of Linear Algebra, showing how a matrix connects
its input and output spaces through these four fundamental subspaces.

Example
For matrix:
 
1 0 1
B=  (2.4.5)
0 1 1

• Null Space Null(B) is determined by by solving:

 
  x1  
1 0 1    0
  x2  =   (2.4.6)
0 1 1   0
x3

This gives the system:

x1 + x3 = 0
x2 + x3 = 0

The solution is:

x1 = −x3 , x2 = −x3 , x3 is a free variable while x1 , x2 are basic variables.

(2.4.7)

Chapter 2 83
Daniel Abebe Modern Control Systems Lecture Note

Note: In linear algebra, when a matrix is transformed into its reduced row echelon
form (RREF), the variables corresponding to pivot columns are called basic vari-
ables, while those corresponding to non-pivot columns are free variables. The basic
variables can be expressed uniquely in terms of the free variables, which represent
the independent degrees of freedom in the solution space.
     
 −1 −1

 
 
 
 
  

Null(B) = x3 −1 | x3 ∈ R → span −1 (2.4.8)
   

   
 
   
  
 1 
 1  

• Column Space The column space Col(B) is the span of the column vectors of B.
The columns are:
     
1 0 1
 ,  ,   (2.4.9)
0 1 1

Since the first two columns are linearly independent and the third is their sum, the
column space is:
   
 1 0 
Col(B) = span   ,  = R2
 (2.4.10)
 0 1 

• Row Space The row space Row(B) is the span of the row vectors of B.
The rows are:
h i h i
1 0 1 , 0 1 1 (2.4.11)

These rows are linearly independent, so:

nh i h io
Row(B) = span 1 0 1 , 0 1 1 (2.4.12)

• Left Null Space The left null space Null(B⊤ ) is the set of solutions to B⊤ y = 0.
Compute B⊤ :
 
1 0
 
⊤
B = 0 1 (2.4.13)
 
 
1 1

Solve:
   
1 0   0
  y  
  1  
= 0 (2.4.14)

0 1
  y2  
1 1 0

Chapter 2 84
Daniel Abebe Modern Control Systems Lecture Note

This gives:
y1 = 0
y2 = 0
y1 + y2 = 0

The only solution is y1 = y2 = 0, so:

 
 0 
⊤
Null(B ) =   (2.4.15)
 0 

Application of Matrix Subspace in Control Systems

For a dynamic system described by a state equation:

ẋ = Ax + Bu (2.4.16)
y = Cx + Du (2.4.17)
1. Controllability: The controllability matrix Co is defined as:
h i
C o = B AB A2 B · · · An−1 B (2.4.18)

• If Col(C o ) = Rn or rank(C o ) = n, the system is fully controllable.

• If rank(C o ) < n, there exist n − rank(C o ) uncontrollable subspaces.
• The uncontrollable subspaces correspond to Null(C ⊤
o ), as they are orthogonal to
Col(C o ).

2. Observability: The observability matrix O v is defined as:

h i⊤
2
O v = C CA CA . . . CA n−1 (2.4.19)

• If rank(O v ) = n, the system is completely observable.

• If rank(O v ) < n, there exist n − rank(O v ) unobservable subspaces.
• The unobservable subspaces correspond to Null(O v ).

Controllability Decomposition
The concept of controllability fundamentally concerns whether a system can be driven
from any initial state x(0) to any desired final state x(tf ) in finite time through appro-
priate control inputs u(t). This property is characterized by the reachable subspace of
the system by control inputs u(t), which corresponds precisely to the column space of
the controllability matrix C o .

The state space decomposition into controllable and uncontrollable subspaces can be ob-
tained through a similarity transformation matrix Tc constructed from the basis vectors
spanning Col(C o ) (the controllable subspace) and the basis vectors spanning Null(C ⊤
o)
(the uncontrollable subspace, being the orthogonal complement).
h i
⊤
Tc = basis(Col(C o )) basis(Null(C o )) (2.4.20)

Chapter 2 85
Daniel Abebe Modern Control Systems Lecture Note

Observability Decomposition
The concept of observability fundamentally addresses whether a system’s initial state
x(0) can be uniquely determined from knowledge of the input u(t) and output y(t) over
a finite time interval [0, tf ]. This property characterizes which state information is visible
through the output measurements, defining the observable subspace of the system.

The state space decomposition into observable and unobservable components can be
obtained through a similarity transformation matrix To constructed from the basis vec-
tors spanning Col(O ⊤v ) (the observable subspace, equivalent to the row space of O v ) and
the basis vectors spanning Null(O v ) (the unobservable subspace).
h i
To = basis(Col(O ⊤ v )) basis(Null(O v )) (2.4.21)

2.5 Controllability and Observability in a Sense of Transfer

Function
Consider a transfer function from U (s) to X(s)

X(s) adj(sI − A)B

= (2.5.1)
U (s) det(sI − A)

For a system to be completely controllable, adj(sI − A)B and det(sI − A) shouldn’t have
common factors (i.e, there shouldn’t be a pole-zero cancellation in a transfer function
from input to state of the system).
Consider a transfer function from U (s) to Y (s)

Y (s) adj(sI − A)B

=C + DU (s) (2.5.2)
U (s) det(sI − A)

For a system to be completely observable, Cadj(sI − A)B and det(sI − A) shouldn’t have
common factors (i.e, there shouldn’t be a pole-zero cancellation in the transfer function
from input to output of the system).

2.6 Examples
1. Consider the following dynamic system, check whether the system is completely
controllable and completely observable.
   
−1 −3 5 h i
ẋ =   x +   u, y = 3 4 x
0 −2 6

Solution:
the controllability matrix C o is


h 5 −23
i
C o = B AB =  
6 −12

Chapter 2 86
Daniel Abebe Modern Control Systems Lecture Note

checking for linear independence of the columns of C o

   
5 −23
α1   + α2   = 0 ⇒ α1 = α2 = 0
6 −12
They are linearly independent, hence the controllability matrix has full rank, which
means the system is completely controllable.

The observability matrix O v is

   
C 3 4
Ov =  = 
CA −3 −17
checking for linear independence of the columns of O v

   
3 4
α1   + α2   = 0 ⇒ α1 = α2 = 0
−3 −17
They are linearly independent, hence the observability matrix has full rank, which
means the system is completely observable.
2. Check controllability and observability of the following dynamic system.
   
−2 −2 −2 1
    h i
ẋ =  1 x + u y = 1 1 0 x
   
0 0 0
   
1 1 −1 0
The controllability matrix is,
 
1 −2 0
 
C o = 0 1 −2
 
 
0 1 −2

The rank rank(C o ) = 2, hence the dynamic system has an uncontrollable state.
The observability matrix is,
 
1 1 0
 
O v = −1 −2 −2
 
 
−2 0 4

having rank rank(O v ) = 2, hence the dynamic system have unobservable state.

The transformation matrix Tc that decomposes the system into controllable and
uncontrollable subspaces is
h i
Tc = basis(Col(C o )) basis(Null(C ⊤
o ))

Chapter 2 87
Daniel Abebe Modern Control Systems Lecture Note

The null space of (C ⊤

o ) is determined as
  
1 0 0 ν
   1
C⊤o ν = 0 → −2 1  ν2  = 0
   
1
  
0 −2 −2 ν3

Transforming to Reduced row echelon form,

  
1 0 0 ν1
  
0 1 1 ν2  = 0 → ν3 is the free variable, leading to
  
  
0 0 0 ν3
   
0 0
   
ν = −ν3  The Null space of C ⊤o is spanned by −1
   
   
ν3 1

The basis of (Col(C o )) is any linearly independent columns of C o , taking the first
two columns and the basis vector that span the null space of C ⊤
o , the transformation
matrix Tc becomes,
 
1 −2 0
 
Tc = 0 1 −1
 
 
0 1 1

To solve for T−1c , transform the augmented matrix [Tc | I] into [I | F ] using the
Gauss-Jordan elimination method, where I is the identity matrix of the same size
as Tc . The resulting matrix F will be the inverse of Tc . Using this method, the
inverse of Tc is
 
2 2 2
 
T−1 = 0.5
 
c  0 1 1 
 
0 −1 1

With this transformation matrix, the decomposed system system matrix Ā, input
matrix B̄ and C̄ is,
..
   
0 −4 . −2 1
..
   
 1 −2 −1
   
. 0
Āc = T−1 AT = 

,
 B̄c = T−1 B = 
 

. . . . . . . . . . . . . . .
..
   
0 0 . −1 0
.
h i
C̄c = CT = 1 −1 .. −1

Chapter 2 88
Daniel Abebe Modern Control Systems Lecture Note

from the decomposition, we see that the uncontrollable state is stable (with eigen-
value −1), hence the system is stabilizable.

The transformation matrix To that decomposes the system into controllable and
uncontrollable subspaces is
h i
To = basis(Col(O ⊤
v )) basis(Null(O v ))

The null space of (O v ) is determined as

  
1 1 0 ν
   1
O v ν = 0 → −1 −2 −2 ν2  = 0
  
  
−2 0 4 ν3

Transforming to Reduced row echelon form,

  
1 0 −2 ν1
  
−1 −2 ν2  = 0 → ν3 is the free variable, leading to
  
0
  
0 0 0 ν3
   
2ν 2
 3   
ν = −2ν3  The Null space of O ⊤v is spanned by −2
   
   
ν3 1

The basis of (Col(O ⊤ ⊤

o )) is any linearly independent columns of O v , taking the first
two columns and the basis vector that span the null space of O v , the transformation
matrix To becomes,
 
1 1 2
 
To = 1 −2 −2
 
 
0 −2 1

Solving for T−1

o result in,
 
6 3 −6
1 
T−1 = 1 −1 −4
 
o
9 
2 −2 1

With this transformation matrix, the decomposed system system matrix Ā, input

Chapter 2 89
Daniel Abebe Modern Control Systems Lecture Note

matrix B̄ and C̄ is,

..
   
−33 63 . 0 6
..
   
−13
   
−1 1 15 . 0 −1 1 1
Āo = T AT =  , B̄o = T B =  
9  ...
 
. . . . . . . . . 9 
. . .
.
   
−8 21 .. −9 2
..
h i
C̄o = CT = 2 −3 . 0
From the decomposition, we see that the unobservable state is stable(with eigen-
value −1), hence the system is detectable.
3. Check the stabilizability and detectability of the dynamic system
   
1 0 0 1
   
ẋ = 1 −1 1  x + 1 u
   
   
0 0 −2 0
h i
y= 1 0 0 x
Checking the controllability of the system
 
1 1 1
h i  
C o = B AB A2 B = 1 0 1
 
 
0 0 0
The rank of the controllability matrix is 2, hence the system has one uncontrollable
dynamics.

Checking the observability of the system

   
C 1 0 0
   
O v =  CA  = 1 0 0
   
   
2
CA 1 0 0
The rank of the observability matrix is 1; hence, the system has two unobservable
dynamics.

The transformation matrix Tc that decomposes the system into controllable and
uncontrollable subspaces is
h i
Tc = basis(Col(C o )) basis(Null(C ⊤
o ))

The null space of (C ⊤

o ) is spanned by
 
0
 
 
0
 
1

Chapter 2 90
Daniel Abebe Modern Control Systems Lecture Note

The basis of (Col(C o )) is any linearly independent columns of C o , taking the first
two column and the basis vector that span the null space of C ⊤
o , the transformation
matrix Tc and T−1
c becomes,
   
1 1 0 0 −1 0
   
Tc = 1 0 0 T−1 = −1 1
   
c  0
   
0 0 1 0 0 −1

With this transformation matrix, the decomposed system system matrix Ā, input
matrix B̄ and C̄ is,
..
   
0 1 . 1 1
..
   
. −1
   
1 0 0
Āc = T−1 AT = 

 , B̄c = T−1 B =  
  
. . . . . . . . . . . .  . . .
..
   
0 0 . −2 0
.
h i
C̄c = CT = 1 1 .. 0

From the decomposition, we see that the uncontrollable state is stable (with eigen-
value −2), hence the system is stabilizable.

The transformation matrix To that decomposes the system into controllable and
uncontrollable subspaces is
h i
To = basis(Col(O ⊤
v )) basis(Null(O v ))

The null space of (O v ) is spanned by,

 
0 0
 
 
1 0
 
0 1

The basis of (Col(O ⊤ ⊤

o )) is any linearly independent columns of O v , taking the first
column and the basis vector that span the null space of O v , the transformation
matrix To and T−1o becomes,
   
1 0 0 1 0 0
   
To = 0 1 0 T−1 =
   
o  0 1 0
   
0 0 1 0 0 1

With this transformation matrix, the decomposed system system matrix Ā, input

Chapter 2 91
Daniel Abebe Modern Control Systems Lecture Note

matrix B̄ and C̄ is,

..
   
1 . 0 0 1
   
   
−1 1. . . . . . . . . . . . −1
. . .
Āo = T AT =  
..
, B̄o = T B =  
91 . −1 1 
  
1
..
   
0 . 0 −2 0
.
h i
C̄o = CT = 1 .. 0 0

from the decomposition we see that the unobservable states are stable(with eigen-
values −1 and −2 ), hence the system is detectable.
4. Check the detectability and stabilizability of the dynamic system
   
−2 0 −3 −1
   
ẋ =  0 0 0  x +  0  u
   
3 5
   
0 −2
2 2  
21
y= 1 0 x
2
The controllability matrix of the system is
 
−1 8 3.5
 
Co =  0
 
0 0 
 
−2 −6.5 −4.25
The rank of the controllability matrix is 2, hence the system has one uncontrollable
dynamics.

checking the observability of the dynamic system, the observability matrix O v ,

 
1 10.5 0
 
O v =  −2 −3 
 
0
 
−0.5 0 −1.5
has rank full, hence the dynamic system is completely observable.

The transformation matrix Tc that decomposes the system into controllable and
uncontrollable subspaces is
h i
Tc = basis(Col(C o )) basis(Null(C ⊤
o ))

The null space of (C ⊤

o ) is spanned by
 
0
 
 
1
 
0

Chapter 2 92
Daniel Abebe Modern Control Systems Lecture Note

The basis of (Col(C o )) is any linearly independent columns of C o , taking the first
two column and the basis vector that span the null space of C ⊤
o , the transformation
matrix Tc and T−1
c becomes,
   
−1 8 0 6.5 0 8
 
−1 −1  
Tc =  0 T = −2
   
0 0 c 0 1
22.5 
 
  
−2 −6.5 1 0 −22.5 0

With this transformation matrix, the decomposed system system matrix Ā, input
matrix B̄ and C̄ is,
..
   
0 −11.25 . 0 −22.5
..
   
−22.5 −11.25 .
   
−1 −1  0 −1 −1  0 
Āc = T AT =   , B̄c = T B =  
22.5 
 ... ...

. . . . . . 22.5  
 ... 
..
   
0 0 . 0 0
.
h i
C̄c = CT = −1 8 .. 10.5

From the decomposition, we see that the uncontrollable state is marginally (with
eigenvalue 0), hence the system is stabilizable.

5. Check the detectability and stabilizability of the dynamic system

   
−6 −11 −6 1
    h i
ẋ =  1 0  x + 0 u y = 0 1 2 x
   
0
   
0 1 0 0

The controllability matrix of the system is

 
1 −6 25
 
C o = 0 −6
 
1
 
0 0 1

C o has full rank, hence the dynamic system is completely controllable.

The observability matrix of the system is

 
0 1 2
 
Ov =  1
 
2 0
 
−4 −11 −6

which has rank 2, hence the dynamic system has one unobservable mode.

Chapter 2 93
Daniel Abebe Modern Control Systems Lecture Note

The transformation matrix To that decomposes the system into controllable and
uncontrollable subspaces is
h i
To = basis(Col(O ⊤
v )) basis(Null(O v ))

The null space of (O v ) is spanned by,

 
4
 
−2
 
 
1

The basis of (Col(O ⊤ ⊤

o )) is any linearly independent columns of O v , taking the first
column and the basis vector that span the null space of O v , the transformation
matrix To and T−1o becomes,
   
0 1 4 2 −1 −10
 
−1 −1  
To = 1 2 −2 To = −5 −8 4 
   
  21  
2 0 4 −4 2 −1

With this tranformation matrix, the decomposed system system matrix Ā, input
matrix B̄ and C̄ is,
..
   
−56 −77 . 0 2
..
   
−5
   
−1  119 140 . 0
Āo = T−1 AT =   , B̄o = T−1 B = −1  
21  
 . . . . . . . . . . . . 21  
. . .
.
   
91 112 .. 42 −4
.
h i
C̄o = CT = 5 2 .. 0

From the decomposition, we see that the unobservable state is stable(with eigen-
value −2), hence the system is detectable.

6. Given the following LTI system, determine the value of α for which the system is
not c.c.
   
2 α−1 1
ẋ =  x +  u
0 2 α

Solution:
The controllability matrix of the system is
 
1 2 + α(α − 1)
Co =  
α 2α

Chapter 2 94
Daniel Abebe Modern Control Systems Lecture Note

The determinant of the controllability matrix must be non-zero if the system is c.c.
So, to find the value of α that makes the system not c.c.
det(C o ) = 0 ⇒ 2α − α3 + α2 − 2α = 0
⇒ α2 (1 − α) = 0 ⇒ α = 0, or α = 1
The same will hold for observability of the system det(O v ) ̸= 0 if the system is c.o.

2.7 Exercises
1. Verify that the following control system is completely controllable
   
0 1 0 0
ẋ =  x +  u
0 0 1 −1

2. For what value of b is the dynamic system below not c.c.

   
−1 −1 b
ẋ =  x +  u
2 −4 0

3. For what value of α is the dynamic system below not c.c.

(a)
   
−1 −1 −1 0
   
ẋ =  0 −1 α  x + 2 u
   
   
0 1 3 1

(b)
   
2 α−3 1 1
ẋ =  x +  u
2
0 2 0 α −2

4. For a system with output

 
−1 −1 h i
ẋ =   x, y= 1 2 x
2 −4

find x0 if the output y = −20e−3t + 21e−2t .

5. Show that the control system described by
ẋ1 = x2
ẋ2 = −2x1 − 3x2 + u
y = x1 + x 2
is not completely observable. Determine the initial state x0 such that if u = 0 for
t > 0, then y = 0, for t > 0.

Chapter 2 95
Daniel Abebe Modern Control Systems Lecture Note

6. Check the detectability and stabilizability of the dynamic system

   
0 0 1 1
    h i
ẋ =  0 1 0 x + 0 u y = 3 1 0 x
   
   
−6 0 6 0

7. Check the detectability and stabilizability of the dynamic system

   
2 0 −3 −1
    h i
ẋ =  0 0 0  x +  0  u y = 1 1 −1 x
   
3 5
   
0 −2
2 2
8. Investigate the stabilizability and detectability of the following control systems
(a)
   
2 0 1 h i
ẋ =   x+   u y= 0 1 x
0 −1 1

(b)
   
0 0 0 h i
ẋ =   x+   u y= 1 1 x
0 −1 1

(c)
 
4 3 5  
  −2 1 h i
ẋ = 1 −2 −3 x +  u y = 0 1 1 x
 
  −1 0
2 1 8

(d)
 
0 1 5  
  0 h i
ẋ =  0 1 x +  u y = 1 0 0 x
 
0
  1
−18 −15 −2

9. Given the LTI system described by

ẋ1 =αx1 + 2x2 + u
ẋ2 =x1 − x2
y =x1 + αx2
determine for what value of the real parameters α the system is
(a) completely controllable and completely observable
(b) completely controllable but not completely observable
(c) completely observable but not completely controllable

Chapter 2 96
Chapter 3

Full-state Feedback Controller

Design for LTI Systems

We know that we can affect the performance and behavior of a dynamic system by
changing the location of the closed-loop poles of the system. We can cancel the unstable
poles by multiplying the system transfer function by another transfer function having a
zero at the location of the unstable pole of the system. This other system is called a
controller or compensator.

Consider a completely controllable system on equation 1.3.6 and equation 1.3.7, suppose
we apply a linear feedback control law, which means that the control law is a linear
combination of state variables,

u = −Kf x (3.0.1)

where Kf is m × n feedback matrix. The closed-loop system under this control law
becomes,

ẋ = Ax + B(−Kf x)
ẋ = (A − BKf )x (3.0.2)

The block diagram representation of the system with a full state feedback controller is
shown in Figure 3.1.

ud + u + x y
- B ʃ C
+

Kf

Figure 3.1: Block Diagram Representation of System with Full State Feedback Controller

97
Daniel Abebe Modern Control Systems Lecture Note

Pole shifting theorem:- for any controllable system, it is possible to obtain arbi-
trary eigenvalues for matrix A − BKf , using suitable feedback matrix Kf , subjected to
the only obvious constraint that complex eigenvalues must appear in pair.

The state feedback controller design can be done either by selecting the location of the
poles and determine the feedback gain Kf that locate the closed-loop poles of the system
at the desired location, or determining the gain Kf that determine the optimal location
of the closed-loop poles and that minimize u at the same time.

3.1 Pole Placement State Feedback Controller Design

In pole placement controller design, given the desired location of the closed-loop system
poles (the eigenvalues of A − BKf ), we are required to determine a feedback gain matrix
Kf such that with a control law u = −Kf x the system has poles at the desired location.

The pole placement or eigenvalue assignment problem can be define as: let λ1 , λ2 , . . . , λn
be the eigenvalues of the matrix A of the open loop LTI system and λ̃1 , λ̃2 , . . . , λ̃n be
desired eigenvalues of the matrix A − BKf of the closed loop system. Where all complex
eigenvalues exist in complex conjugate pairs. Let p(s) and p̂(s) be respective character-
istics polynomial of matrix A and A − BKf respectively.
n
Y
p(s) = (s − λi ) = ∥sI − A∥ = sn + a1 sn−1 + . . . + an−1 s + an (3.1.1)
i=1
Yn
p̂(s) = (s − λ̃i ) = ∥sI − A + BKf ∥ = sn + â1 sn−1 + . . . + ân−1 s + ân (3.1.2)
i=1

Kf is determined if and only if the open loop system is controllable (i.e., the controllability
matrix C o has full rank).
The feedback matrix Kf can be determined by one of the following methods.

1. The Base-Gura method:

Kf = (W⊤ C ⊤ −1
o ) (â − a) (3.1.3)

Where
     
1 a1 a2 a3 . . . an−1 â1 a
     1
0 1 a1 a2 . . . an−2   â2   a2 
     
     
W = 0 0 1 a1 . . . an−3  , â =  â3  , a =  a3 
     
 .. .. .. .. .. ..   ..   .. 
     
. . . . . .  . .
     
0 0 0 0 ... 1 ân an

Chapter 3 98
Daniel Abebe Modern Control Systems Lecture Note

2. The phase canonical form formula: Consider the case where A is in controllable
canonical form Ac , bc .
   
0 1 0 ... 0 0
   
 0 0 1 ... 0  0
   
 .. .. .. .. ..  .
   
Ac =  . . . . .  , bc =  .. 
   
   
 0 0 0 ... 1  0
   
−a0 −a1 −a2 . . . −an1 1

Which means:

Kf = (W⊤ C ⊤ −1 ˜
o ) (â − a) = I = (â − a)
Kf = (â − a) (3.1.4)

3. Ackerman’s formula:

⊤ −1
K⊤
f = e C o p̂(A) (3.1.5)

Where

e⊤ = [0, 0, . . . , 1],

and p̂(A) is obtained from polynomial on equation 3.1.1 by replacing s with matrix
A, which means

p̂(A) = An + â1 An−1 + . . . + ân−1 A + ân

An LTI system defined by (A, B) is stabilizable if there exists a matrix Kf such that
(A − BKf ) has a strictly negative real part.

Chapter 3 99
Daniel Abebe Modern Control Systems Lecture Note

3.2 Desired Input Tracking

If the system is required to track a required reference input trajectory (desired input
trajectory (ud (t)) other than a constant value (which doesn’t change with time), such a
control problem is called a desired input trajectory tracking problem. If the desired input
trajectory is constant, the control problem is called the input regulator problem. Let the
desired input trajectory be ud (t), hence under state feedback control law
u(t) = ud (t) − Kf x(t) (3.2.1)
Then with this control law, the state equation of the LTI becomes,
ẋ = Ax + B(ud − Kf x)
ẋ = (A − BKf )x + Bud (3.2.2)
The block diagram representation of a state feedback controller with reference time vari-
ant input is shown in Figure 3.2.

ud + u + x y
- B ʃ C
+

Kf

Figure 3.2: Block Diagram Representation of Full State Feedback with Reference State

3.3 Servo-mechanism Controller Design

Suppose we need our system to follow certain input (reference tracking), but due to
mathematical approximation, the matrices of the nominal system deviate from the actual
system (model uncertainty), and/or there may be parameter variation. In that case, the
steady-state performance of the system may be poor. To enhance the performance of the
system at a steady state, an integrator error term is added to the dynamics so that the
steady-state error tends to zero in finite time. The error dynamics ξ˙ is defined as,
ξ̇(t) = r(t) − y(t) (3.3.1)
with the control law,
u(t) = −Kf x(t) + k1 ξ(t) (3.3.2)
The closed-loop dynamic system becomes
ẋ = Ax + B(−Kf x + k1 ξ)
ẋ = (A − BKf )x + Bk1 ξ (3.3.3)
y = Cx (3.3.4)

Chapter 3 100
Daniel Abebe Modern Control Systems Lecture Note

Which leads to an augmented system

      
ẋ A − BKf Bk1 x 0
 =   +  r (3.3.5)
ξ̇ −C 0 ξ 1

The block diagram representation of a servomechanism is shown in Figure 3.3.

r + + u + x y
ʃ k1 - B ʃ C
- +

Kf

Figure 3.3: Block Diagram Representation of Servomechanism Controller Design

To determine the feedback gain matrix Kf the matrix

     
A − BKf Bk1 A 0 B h i
  ⇒   −   Kf −k1 (3.3.6)
−C 0 −C 0 0

which means
   
A 0 B
  ,   (3.3.7)
−C 0 0

must be completely controllable.

To find the value of k1 , we can analyze the steady state response of the system. At steady
state, the output of the system should follow the reference input y = r, which means
that ξ̇ = 0. This implies that the output of the integrator is a constant value equal to
r. Therefore, the states of the system should also follow the desired states of the system
xd , given by xd = k1 r.

However, at this point, the output of the system is given by y = r = Cxd + Du, where
u = xd − Kf xd . Thus, we can write:

r = Cxd + D(I − Kf )xd → r = (C + D(I − K))k1 r (3.3.8)

Solving for k1 , we get:

k1 = (C + D(I − Kf ))−1 (3.3.9)

Chapter 3 101
Daniel Abebe Modern Control Systems Lecture Note

3.4 Determining Poles of Second Order LTI Systems From Time

Domain Specification
Any second-order system can be written as

ωn2
G(s) = (3.4.1)
s2 + 2ξωn s + ωn2

where ξ is damping p factor and ωn is undamped natural frequency. Damped natural

frequency is ωd = ωn 1 − ξ 2 . The time domain specification of the second-order system
is shown in Table B.1. where θ = cos−1 ξ.

Table 3.1: Time Domain Specification of Second Order Systems

Time domain specification Formula

1 + 0.7ξ
Delay time td =
ωn
π−θ
Rise time tr =
ωd
π
Peak time tp =
ω
 d 
− √ ξπ
1−ξ2
Percentage Peak overshoot Mp = e
4
Settling time with 2% tolerance ts =
ξωn

Since any second-order system can be written in this form, we can determine the desired
poles of the system from the time domain specification of the system.

The controllable canonical form of equation 3.4.1 is

      
ẋ1 0 1 x 0
  =   1 +   u (3.4.2)
ẋ2 −ωn2 −2ξωn x2 1
 
h i x1
2
y = ωn 0   (3.4.3)
x2

Note:- The goal of the designer is to find Kf such that a closed loop time domain
specification of the given second order system meets the requirement.

Chapter 3 102
Daniel Abebe Modern Control Systems Lecture Note

3.5 Examples
1. Given a continuous-time system
   
1 2 0
ẋ =   +  u
3 1 1
h i
y= 1 3 x

discuss controllability and observability of the system.

Solution
Check for controllability
 
h 0 2
i
C o = B AB =  
1 1

which has rank 2, hence it is completely controllable.

Check for observability
   
C 1 2
Ov =  = 
CA 7 10

which has rank 2, hence it is completely observable.

If we use control law u = [3 1]x + ν, check controllability and observability of the
closed loop system.
   
1 2 0
ẋ = (A − BK)x + Bν ⇒ ẋ =  x +  ν
0 0 1

heck for controllability

 
h i 0 2
C o = B AB =  
1 0

which has rank 2, hence it is completely controllable.

Check for observability
   
C 1 2
Ov =  = 
CA 1 2

the rows of O v are are not linearly independent,hence rank of O v is 1, hence it is

not completely observable.

Chapter 3 103
Daniel Abebe Modern Control Systems Lecture Note

2. Given a continuous-time system

(s − 1)(s + 2)
H(s) =
(s + 1)(s − 2)(s + 3)
is that possible to convert H(s) to
s−1
G(s) =
(s + 2)(s + 3)
if possible determine state feedback gain matrix Kf .

Solution
We have to cancel zero at z = −2 by a pole at p = −2, cancel the pole at p = −1
by a zero at z = −1, cancel the pole at p = 2 by a zero at z = 2, and add one pole
at p = −2 to get G(s) from H(s).
 
(s + 1)(s − 2) (s + 1)(s − 2) (s − 1)(s + 2)
G(s) = H(s) =
(s + 2)2 (s + 2)2 (s + 1)(s − 2)(s + 3)
(s − 1)(s + 2)
=
(s + 2)2 (s + 3)

Since the degree of the numerator is less than the degree of the denominator, it is
possible to have such a kind of compensator. Hence it is possible to get G(s) from
H(s).
The state-space representation of H(s) is
   
−2 5 6 1
   
ẋ =  1 0 0 x + 0 u
   
   
0 1 0 0
h i
y = 1 1 −2 x

We need a state feedback control law u = −Kf x that converts the system H(s) to
G(s), hence the desired location of poles of the closed-loop system is at p = −2, p =
−2, p = −3. Using the Base-Gura method,
det(sI − A + BKf ) = (s + 2)2 (s + 3)
 
    
s 0 0 −2 5 6 1
     h i
det 0 s 0 −  1 0 0 + 0 k1 k2 k3  = s3 + 7s2 + 16s + 12
      
      
0 0 s 0 1 0 0
 
s + 2 + k1 −5 + k2 −6 + k3
 
3 2
det  −1 0  = s + 7s + 16s + 12
 
s
 
0 −1 s
s3 + (2 + k1 )s2 + (−5 + 6)s + (−6 + k3 ) = s3 + 7s2 + 16s + 12

Chapter 3 104
Daniel Abebe Modern Control Systems Lecture Note

2 + k1 = 7 ⇒ k1 = 5, −5 + 6 = 16 ⇒ k2 = 21, −6 + k3 = 12 ⇒ k3 = 18
h i h i
Kf = k1 k2 k3 = 5 21 18

The closed loop dynamics of the system with full state feedback gain matrix Kf
are

ẋ = (A − BKf )x
       
ẋ −2 5 6 5 21 8 x
 1       1 
ẋ2  =  1 0 0 − 0 0 0 x2 
       
       
ẋ3 0 1 0 0 0 0 x3
    
ẋ −7 −17 −2 x1
 1   
ẋ2  =  1
    
0 0  x2 
    
ẋ3 0 1 0 x3

3. For IPC dynamic system on 1.6.3 with parameters mc = 0.5, mp = 0.2, b = 0.1, l =
0.3, I = 0.5mp l2 , g = 9.81, Find a feedback gain matrix that place the poles of the
closed loop system at p1 = −1, p2 = −2, p3 = −1 − i, p4 = −1 + i.

Solution
To check for complete controllability of IPC dynamic system, use C o = ctrb(A, B),
and rank(C o ) commands on MATLAB to check controllability and the rank of
controllability matrix, respectively. With the given parameters
   
0 1.00 0 0 0
   
0 −0.17 2.31
   
0  1.75
ẋ = 

x + 
  u

0 0 0 1.00  0 
   
0 −0.39 26.93 0 3.92
h i
y= 1 0 1 0 x

which result in controllability matrix C o , with rank 4.

 
0 1.76 −0.31 9.11
 
1.76 −0.31 9.11 −3.2 
 
Co =  
3.92 −0.69 105.73 
 
 0
 
3.92 −0.69 105.73 −22.21

To design Khf at the desired locations use

i MATLAB command Kf = place(A, B, p),
where p = −1 −2 −1 − i −1 + i . Which results in a feedback gain matrix
h i
Kf = −0.10 −0.36 9.46 1.39

Chapter 3 105
Daniel Abebe Modern Control Systems Lecture Note

The closed loop dynamics of the system with full state feedback gain matrix Kf
are

ẋ = (A − BKf )x
       
ẋ1 0 1.00 0 0 0 0 0 0 x
       1 
ẋ2  0 −0.17 2.31 0  −0.18 −0.64 16.7 2.46 x2 
       
  =  −   
       
ẋ3  0 0 0 1.00  0 0 0 0  x3 
       
ẋ4 0 −0.39 26.93 0 −0.41 −1.41 37.11 5.46 x4
    
ẋ 0 1 0 0 x
 1    1
ẋ2  0.18 0.46 −14.39 −2.46 x2 
    
 =  
    
ẋ3   0 0 0 1  x3 
    
ẋ4 0.41 1.02 −10.18 −5.46 x4

Figure 3.4: Closed Loop Response of IPC for Impulse Input with Kf

4. For the following system, determine a feedback gain matrix that places the closed-
loop poles of the system at p = −2 ± 2i.
   
1 0 0 1
   
ẋ = 1 −1 1  x + 1 u
   
   
0 0 −2 0
h i
y= 1 0 0 x

Solution
As shown in chapter two example 3, the system is not completely controllable. So
we have to decompose the system into controllable and uncontrollable modes as
shown in chapter two example 3. So we can design a state feedback controller for

Chapter 3 106
Daniel Abebe Modern Control Systems Lecture Note

controllable mode (A11 , B1 ).

∥sI − A11 + B1 Kf ∥ = (s + 2 + 2i)(s + 2 − 2i)

   
s−1 0 1 h i
  +   k1 k2 = s2 + 4s + 8
−1 s + 1 1
 
s − 1 + k1 k2
  = s2 + 4s + 8
−1 + k1 s + 1 + k2

(s − 1 + k1 )(s + 1 + k2 ) − k2(−1 + k1 ) = s2 + 4s + 8
s2 + (k1 + k2 )s − 1 + k1 = s2 + 4s + 8
k1 − 1 = 8 ⇒ k1 = 9 k1 + k2 = 4 ⇒ k2 = −5
h i
Kf = 9 −5

What happens if we want to design a full state feedback controller for as a whole?
Let us assume the third pole is at p = α, then

||sI − A + BKf || = (s + α)(s2 + 4s + 8)

     
s 0 0 1 0 0 1
     h i
0 s 0 − 1 −1 1  + 1 k1 k2 k3 = (s + α)(s2 + 4s + 8)
     
     
0 0 s 0 0 −2 0
   
s−1 0 0 k k k
   1 2 3
 −1 s + 1 −1  + k1 k2 k3  = (s + α)(s2 + 4s + 8)
   
   
0 0 s+2 0 0 0

which results in

(s + 2)(s − 1 + k1 )(s + 1 + k2 ) − k2 (1 + k1 ) = (s + α)(s2 + 4s + 8)

Hence, the system requires the stability of the uncontrollable mode only. k3 does
not affect the system. If the uncontrollable mode is stable, the feedback gain corre-
sponding to the uncontrollable mode does not affect the system closed closed-loop
dynamics. This example proves this scenario.

From Kalman decomposition, we have

 
A − B1 k1 A12 − B1 k2
 11 
0 A22

The system only requires the stability of uncontrollable mode; no need to design
feedback gain Kf for the uncontrollable mode, even if the uncontrollable mode is
not stable. We cannot impose any effect on the uncontrollable mode (that is why

Chapter 3 107
Daniel Abebe Modern Control Systems Lecture Note

it is said uncontrollable mode).

The closed loop dynamics of the system with full state feedback gain matrix Kf
are
ẋ = (A − BKf )x
 
       x1
ẋ 1 0 0 9 −5 k3  
 1       x2 

 
ẋ2  = 1 −1 1  − 9 −5 k3   
     
      . . .

ẋ3 0 0 −2 0 0 0  
x3
.
  
  −8 5 .. −k3 x
ẋ1  1
.

  −8 4 .. 1 − k   x 
  
3
 2

ẋ2  = 
  
   . . . . . . ...  
. . .  . . .
ẋ3 
..
 
0 0 . −2 x3

5. Determine a full state feedback matrix for the following second-order system such
that the closed loop system has settling time ts = 4sec and peak time tp = 1.813sec.
   
1 0 1
ẋ =  x +  u
1 1 0
h i
y= 1 0 x
Solution
From given time domain specifications
4 4
ξωn = = = 1
ts 4
π 3.14
ωd = = = 1.732
tp 1.813
p
ωn 1 − ξ 2 = 1.732 ⇒ 1 − ξ 2 = 3ξ 2
⇒ ξ = 0.5 ⇒ ωn = 2
Thus the closed loop √ system has characteristics
√ equation of s2 + 2ξωn + ωn2 =
s2 + 2s + 4 = (s + 1 + 3i)(s + 1 − 3i). The state feedback controller gain must
locate the closed-loop system poles at s = −1 ± i. We determine Kf as follows
∥sI − A + BKf ∥ = s2 + 2s + 4
   
s−1 0 k1 k2
 +  = s2 + 2s + 4
−1 s − 1 0 0
 
s − 1 + k1 k2
  = s2 + 2s + 4
−1 s−1
s2 + (−2 + k1 )s + 1 − k1 + k2 = s2 + 2s + 4
− 2 + k1 = 2 ⇒ k1 = 4 1 − k1 + k2 = 4 ⇒ k2 = 7

Chapter 3 108
Daniel Abebe Modern Control Systems Lecture Note
h i
Kf = 4 7

The closed loop dynamics of the system with full state feedback gain matrix Kf
are

ẋ = (A − BKf )x
       
ẋ 1 0 4 7 x
 1  =  −   1 
ẋ2 1 1 0 0 x2
    
ẋ −3 −7 x1
 1 =   
ẋ2 1 1 x2

3.6 Exercises
1. Consider a system
   
−1 −2 1
ẋ =  x +  u
2 −4 3

find a feedback gain matrix Kf such that the closed loop system has eigenvalues at
−4 and −5.
2. For LTI system
   
1 0 −1 1
   
ẋ = 1 2 1  x + 0 u
   
   
2 2 3 1

find state feedback gain matrix Kf such that the eigenvalues of closed loop system
at −1, −1 ± 2i.
3. For the dynamic system
   
0 1 0 1 0
   
ẋ = 0 0 1 x + 0 1 u
   
   
6 −11 6 1 1

Find the state feedback gain matrix Kf such that the eigenvalues of the closed loop
system are at 1,2, and 3, then check for stability of the system with the control law.
4. Determine whether the system described by
   
−1 0 3 1
   
ẋ =  0 −3 0  x +  1  u
   
   
1 0 −3 −1

Chapter 3 109
Daniel Abebe Modern Control Systems Lecture Note

is c.c. and show that under linear feedback of the form u = αx1 + βx3 , the closed
loop system has two fixed eigenvalues, one of which is equal to −3. Determine the
second fixed eigenvalues and also values of α and β such that the third closed loop
eigenvalue is −4.
5. Determine the range of values of k, the linear dynamic system
 
0 1 0
 
ẋ =  0 1 x
 
0
 
−k −1 −2
h i⊤
is asymptotically stable. If k = −1 and control term 0 1 0 u is added, find
linear feedback gain Kf that makes the closed loop system to have poles at −1.
6. For the control system described by
ẋ1 = x1 − x3 + u
ẋ2 = x1 + 2x2 + x3
ẋ3 = 2x1 + 2x2 + 3x3 + u
find a suitable feedback matrix Kf such that the closed loop system has eigenvalues
−1 and −1 ± 2i.
7. For the control system described by
ẋ1 = x2 + u1
ẋ2 = x3 + u2
ẋ3 = 6x1 − 11x2 + 6x3 + u1 + u2
find a suitable feedback matrix Kf such that the closed loop system has eigenvalues
−1 and −2 ± i.
8. For the following system, find a state feedback controller gain matrix so that the
closed-loop system has maximum percentage overshoot mp = 14% and rise time
tr = 1.25sec.
   
0 −1 0
ẋ =  x +  u
−3 −2 1

9. Let A is n × n and B is n × 1, given any arbitrary set Λ = {θ1 , θ2 , . . . , θm } of

complex numbers(appearing i conjugate pairs), there exist a feedback matrix Kf
such that the eigenvalues of A − BKf are the set Λ.
10. Find linear feedback control u = −Kf x for
   
1 0 −1 1
   
ẋ = 1 2 1  x + 0 u
   
   
2 2 3 1
that place closed loop poles of the system at −1, −1 ± 2i. Does this verify problem
number 9?

Chapter 3 110
Chapter 4

State Estimators/Observers Design

for LTI systems

To design a full-state feedback control for a given system, all the states of the system must
be available as output (must be measured directly). But in many real-world problems, all
states of the system are not available as output. If the system is completely observable,
the states of the system are related to the output of the system in some way. So we can
recover the unavailable states from the output of the system. The method of recovering
these states from the output of the system is called the observer or state estimator design
problem. If all states are required to be observed or estimated, the problem is a full state
observer problem; if we can measure some states directly and the others are to be observed
or estimated, then the problem becomes a reduced-order state observer/estimator.

If the system is partially observable, we can design an observer for the observable mode
of the system. Since we cannot say anything about the relation between the output of
the system and the unobservable mode of the system, we cannot design an observer for
the unobservable mode of the system.

4.1 Full Order State Observer Design

Consider a completely observable dynamic system

ẋ = Ax + Bu (4.1.1)
y = Cx (4.1.2)

Assume the state vector x is given by its estimate x̂

x̂˙ = Âx̂ + B̂u + Lc y (4.1.3)

The matrices Â, B̂ are unknown.

In the full state observer design problem, x and x̂ have equal dimension. The deviation
(estimation error) x̃ = x − x̂, will be

x̃˙ = ẋ − x̂˙
x̃˙ = Ax + Bu − Âx̂ − B̂u − Lc y
x̃˙ = Ax + Bu − Âx̂ − B̂u − Lc Cx (4.1.4)

111
Daniel Abebe Modern Control Systems Lecture Note

replacing x̂ by x − x̃, which leads to

x̃˙ = Ax + Bu − Â(x − x̃) − B̂u − Lc Cx

x̃˙ = (A − Â − Lc C)x + Âx̃ + (B − B̂)u (4.1.5)

If we need x̃ to become zero independent of x and u, is only possible if and only if B̂ = B,

 = (A−Lc C) and  is stable. If this condition is fulfilled, the error dynamics x̃ becomes

x̃˙ = Âx̃ = (A − Lc C)x̃ (4.1.6)

Thus, the design problem is finding a suitable Lc such that the estimation error dynamics
go to zero as t → ∞, or determining a suitable Lc that places the closed loop eigenvalues
of estimation error dynamics x̃˙ = (A − Lc C)x at the desired location.
Replacing  with A − Lc C in the full state observer equation on 4.1.3 we have

x̂˙ = Ax̂ + Bu + Lc (Cx − x̂)

x̂˙ = Ax̂ + Bu + Lc Cx̃ (4.1.7)

which is called Luenberger observer.

The full state observer state equation (full-order state observer dynamics ) is
      
˙x̂ A Lc C x̂ B
 =   +  u
x̃˙ 0 A − Lc C x̃ 0

If the system is represented in observable canonical form,

x̃˙ ⊤ = (Ao − Lc Co )⊤ x⊤ = (A⊤ ⊤ ⊤ ⊤ ⊤

o − Co Lc )x ⇔ ẋ = (Ac − Bc K)x
⊤
(4.1.8)

Hence, we can use the methods we used to determine Kf ,

Lc = K⊤ (4.1.9)

This implies that if estimation error dynamics is expressed in an observable canonical

form, the feedback matrix Lc can be determined from full state feedback matrix Kf for
the corresponding controllable canonical form representation of estimation error dynamics
just by transposing Kf and vise-versa.

If the desired poles of x̃˙ = (A − Lc C)x and ẋ = (A − BKf )x are the same. For
a system expressed in observable canonical form Lc equal to the transpose of the full
state feedback matrix K⊤ f , that places the closed-loop eigenvalues of the corresponding
controllable canonical form representation of the same system at the desired location.

An LTI system defined by (A, C) is detectable, if there exists a matrix Lc such that
(A − Lc C) has a strictly negative real part.

Chapter 4 112
Daniel Abebe Modern Control Systems Lecture Note

The block diagram representation of a full-order state observer is shown in Figure 4.1.

u + x +
B + ʃ C
y
+

A
+
Lc
--
x̂
+ ʃ C
ŷ
A
Figure 4.1: Block Diagram Representation of Full Order State Observer

4.2 Reduced Order State Observer Design

Consider a completely observable system in which m states are measured directly from
the output and n − m states are those that cannot be measured directly. Let η 1 represent
the states that are measured directly, and η 2 represent the states that cannot be measured
directly from outputs.
   
x x
 1  m+1 
 x2   xm+2 
   
   
η 1 =  x3  , η 2 =  xm+3 
   
 ..   .. 
   
 .   . 
   
xm xn−m

The states to be estimated are related to output so, the output of the state estimator
is η̂ 2 = Lc y + ω, where ω is the state of reduced order dynamics. In terms of η 1 and
η 2 , the dynamic system can be decomposed into directly measured state dynamics and
directly unmeasured state dynamics as follows,

η̇ 1 = A11 η 1 + A12 η 2 + B1 u
η̇ 2 = A21 η 1 + A22 η 2 + B2 u
y = η1 (4.2.1)

Since η 1 states are directly measured, the estimated state is equal to the output, which
means

η̂ 1 = η 1 = y (4.2.2)

Chapter 4 113
Daniel Abebe Modern Control Systems Lecture Note

But the remaining required states are determined from y

η̂ 2 = Lc y + ω (4.2.3)
η̂˙ 2 = Lc ẏ + ω̇ (4.2.4)
Equation 4.2.3 is called state observer output equation.

The estimation error is

     
η̃ η − η̂ 1 0
 1 =  1 =  (4.2.5)
η̃ 2 η 2 − η̂ 2 η 2 − Lc η 1 − ω

Let ω̇ = Jω + Ny + Mu, which is called minimum order state observer dynamics. The
estimation error dynamics become
η̃˙ 2 = η̇ 2 − η̂˙ 2
= (A21 η 1 + A22 η 2 + B2 u) − (Lc ẏ + ω̇)
= (A21 η 1 + A22 η 2 + B2 u) − (Lc A11 η 1 + Lc A12 η 2 + Lc Bu + Jω + Nη 1 + Mu)
(4.2.6)
replacing for ω = η 2 − Lc η 1 − η̃ 2 , then we have
η̃˙ 2 =(A21 η 1 + A22 η 2 + B2 u) − (Lc A11 η 1 + Lc A12 η 12 + Lc B1 u)
− (Jη 2 − JLc η 1 − Jη̃ 2 + Nη 1 + Mu)
η̃˙ 2 =(A12 − Lc A + JLc − N)η 1 + (A22 − Lc A12 − J)η 2
+ (B2 − Lc B1 − M)u + Jη̃ 2 (4.2.7)
For estimation error dynamics to go to zero independent of x and u, the following
conditions should be full-filled
A12 − Lc A11 + JLc − N = 0 ⇒ N = A21 − Lc A11 + JLc
B2 − Lc B1 − M = 0 ⇒ M = B2 − Lc B1
A22 − J − Lc A12 = 0 ⇒ J = A22 − Lc A12
If these conditions are fulfilled, the estimation error dynamics become,
η̃˙ 2 = Jη̃ 2
η̃˙ 2 = (A22 − Lc A12 )η̃ 2 (4.2.8)
As we can see, Lc is the state feedback gain matrix that locates the estimation error
dynamics closed-loop poles at the desired location. The reduced order observer state
equation will be,
ω̇ = Jω + Nη 1 + Mu
ω̇ = (A22 − Lc A12 )ω + (A21 + JLc − Lc A11 )η 1 + (B2 − Lc B1 )u (4.2.9)
If J has eigenvalues with a negative real part, the error dynamics go to zero asymptot-
ically. So we determine Lc such that the closed-loop error dynamics A22 − Lc A12 has
poles at the desired location (i.e., the poles of the reduced-order observer location). In
determining Lc , A22 and A12 take a role of A and C is full order observer respectively.

Chapter 4 114
Daniel Abebe Modern Control Systems Lecture Note

Finally, the combined system state equation (state dynamics) with reduced-order
observer dynamic equation (state equation) is given as
      
η̇ A11 A12 0 η B1
 1    1  
η̇ 2  =   η 2  +  u
      
A21 A22 0 B2
      
ω̇ [A12 + JLc − Lc A11 ] 0 [A22 − Lc A12 ] ω [B2 − Lc B1 ]
(4.2.10)

The block diagram representation of the reduced-order state observer is shown in Figure
4.2.

u + 1 y
B1 ʃ
+

+ A11
+

A12

A21
+ +
A22

+
x2
+
B2 ʃ

N
Lc

ω ̂ 2
M + ʃ +

Figure 4.2: Block Diagram Representation of Reduced Order State Observer

Chapter 4 115
Daniel Abebe Modern Control Systems Lecture Note

4.3 Continuous Time Kalman Filtering

Consider a dynamical system
ẋ =Ax + Bu + Gω(t)
y =Cx + ν(t) (4.3.1)
ω(t) and ν(t) are stochastic disturbances, which are estimated to be white noises with
zero means and covariance with x(0) = N ∼ (0, X0 ), given as
E{ω(t)ω(t)⊤ } = Qδ(t − τ )
E{ν(t)ν(t)⊤ } = Rδ(t − τ ) (4.3.2)
White noise is an abstraction of the derivation of the Brownian motion process b(t),
which means b(t) is the integral of ω(t) white noise. The dynamic system on equation
4.3.1 becomes,
ẋ = Ax + Bu + Gḃ(t) (4.3.3)
Using conditional expectation:
x̂ = E{(x(t) − x̂(t))⊤ (x(t) − x̂(t))} (4.3.4)
This coincides with the optimal estimation, which minimizes the sum of the error vari-
ances:
J = E{(x(t) − x̂(t))⊤ (x(t) − x̂(t))} (4.3.5)
and the error covariance in the sense of positive definiteness of
E{(x(t) − x̂(t))⊤ (x(t) − x̂(t))} (4.3.6)
The Kalman filter equation dictate the dynamical evolution of x̂ are
x̂˙ = Ax̂ + Lk (y − Cx̂) + Bu
= (A − Lk C)x̂ + Lk y + Bu (4.3.7)
where Lk = PC⊤ R−1 which result in riccati equation
Ṗ = AP + PA⊤ + G⊤ QG − PC⊤ R−1 CP (4.3.8)
The steady state solution, in case A,G, Q and R are constants for LTI system and
the noise
p statistics are constant under the assumption that (A, C) is observable and
(A, G Q) is controllable the steady state solution Pss = lim ∞ϑ of the differential
t→
riccati equation is unique. Since P is constant for LTI systems Ṗ = 0 which leads to:
0 = AP + PA⊤ + G⊤ QG − PC⊤ R−1 CP (4.3.9)
further, the estimation error x̃ = x − x̂ dynamics is:
x̃˙ = (A − Lk C)x̃ (4.3.10)
are stable (if the eigenvalues of A − Lk C has strictly negative real parts).
Note that ẋ = A doesn’t need to be stable, and in the filter above
x̂˙ = (A − Lk C)x̂ + Lk y + G1 u (4.3.11)
If A has eigenvalues with a positive real part, the state evolution can still be combined
and dictated by input. If the input is chosen according to u = −Kf x̂, with Kf chosen so
that (A − BKf ) is stable. The closed-loop system is stable with eigenvalues of the union
of (A − BKf ) and (A − Lk C).

Chapter 4 116
Daniel Abebe Modern Control Systems Lecture Note

4.4 Closed Loop State Feedback Controller Design with Full

State Observer
Consider a completely observable and completely controllable system

ẋ = Ax + Bu
y = Cx (4.4.1)

with state observer

x̂˙ = Ax̂ + Bu + Lc (y − Cx̂) (4.4.2)

and control law

u = −Kf x̂ (4.4.3)

Replacing x̂ with x − x̃ the system becomes

ẋ = Ax − BKf x̂
ẋ = (A − BKf )x + BKf x̃ (4.4.4)

and

x̃˙ = (A − Lc C)x̃ (4.4.5)

This shows that Kf and Lc can be designed independently of each other. If (A, B) is
controllable, Kf can be chosen in a way that the poles of the closed-loop system are at the
desired location, and if (A, C) is observable, Lc can be chosen so that the error dynamics
go to zero rapidly. This property, where the two design problems (observer design problem
and state feedback design problem of a system) can be handled independently, is called
the separation principle. The dynamics of the closed-loop system of full state feedback
with full order state observer is then,
    
ẋ A − BKf BKf x
 =  
x̃˙ 0 A − Lc C x̃

Chapter 4 117
Daniel Abebe Modern Control Systems Lecture Note

The block diagram representation of full state feedback with full order state observer
is shown in Figure 4.3.

+ u + x
xud
-
B + ʃ C
y

A
+
Lc
-
+ ʃ
x̂ C
ŷ
A

Kf

Figure 4.3: Block Diagram Representation of Full State Feedback Controller with Full Order
State Observer

4.5 Closed Loop State Feedback Controller Design with Re-

duced Order State Observer
In this case, we use the same control law as full state observer controller design u = −Kx̂;
the only difference is x̂, which is the reduced-order state observer. The compensator for
the states that cannot be measured directly (minimum order state observer dynamics
(state equation)) is ω̇ = Jω + Ny + Mu. This means the control law becomes

u = −K(S1 η 1 + S2 η̂ 2 )
= −K(S2 η 1 + S2 (η 2 − η̃ 2 )) (4.5.1)

where S1 and S2 are weighing matrices.

With this control law, the state dynamics

η̇ 1 = A11 η 1 + A12 η 2 + B1 (−KS1 η 1 − KS2 η̂ 2 )

= A11 η 1 + A12 η 2 − B1 KS1 η 1 − B1 KS2 (η 2 − η̃ 2 )
= (A11 − B1 KS1 )η 1 + (A12 − B1 KS2 )η 2 + B1 KS2 η̃ 2 (4.5.2)
η̇ 2 = A21 η 1 + A22 η 2 + B2 (−KS1 η 1 − KS2 η̂ 2 )
= A21 η 1 + A22 η 2 − B2 KS1 η 1 − B2 KS2 (η 2 − η̃ 2 )
= (A21 − B2 KS1 )η 1 + (A22 − B2 KS2 )η 2 + B2 KS2 η̃ 2 (4.5.3)

and

η̃˙ 2 = (A22 − Lc A12 )η̃ 2 (4.5.4)

Chapter 4 118
Daniel Abebe Modern Control Systems Lecture Note

The closed loop dynamics (closed loop state equation) of the system with reduced order
observer (reduced order observer state equation) is
    
η̇ A − B1 KS1 A12 − B1 KS2 0 B1 KS2 η
 1   11   1
η̇ 2  A21 − B2 KS1 A22 − B2 KS2 0
    
B2 KS2  η 2 
 =   (4.5.5)
˙    
η̃ 1   0 0 0 0  η̃ 1 
    
˙η̃ 2 0 0 0 A22 − Lc A12 η̃ 2

The block diagram representation of full state feedback with reduced order state observer
is shown in Figure 4.4.

uxd + u + 1 y
B1 ʃ
- +

+ A11
+
A12
+
Kf S1
+ A21
+ +
A22

+ + x2
B2 ʃ

Lc

 ̂2
M + ʃ +

S2

Figure 4.4: Block Diagram Representation of Full State Feedback Controller with Reduced Order
State Observer

Chapter 4 119
Daniel Abebe Modern Control Systems Lecture Note

4.6 Examples
1. For a dynamic system
   
0 1 0
ẋ =  x +  u
−1 −2 1
h i
y= 0 1 x

Design a full state observer with poles at p1 = −0.25, p2 = −0.5.

Solution
Check for observability of the system
   
C 0 1
Ov =  = 
CA −1 −2

This has a rank 2, hence the system is observable.

The characteristic of the observer is (s + 0.25)(s + 0.5) = s2 + 0.75s + 0.125. The
poles of (A − Lc C) must be at the desired poles. so

det(sI − A + Lc C) = s2 + 0.75s + 0.125

      
s 0 0 1 l1 h i
det  −  +   0 1  = s2 + 0.75s + 0.125
0 s −1 −2 l2
 
s −1 + l1
det   = s2 + 0.75s + 0.125
1 s + 2 + l2
s2 + (2 + l2 )s + (1 − l1 ) = s2 + 0.75s + 0.125

The state observer gain matrix that puts the poles of the state observer at the
desired location is then,

2 + l2 = 0.75 ⇒ l2 = −1.25, 1 − l1 = 0.125 ⇒ l1 = 0.875

   
l1 0.875
Lc =   =  
l2 −1.25

Chapter 4 120
Daniel Abebe Modern Control Systems Lecture Note

The full state observer (Luenberger state observer) dynamics is

      
˙x̂ A Lc C x̂ B
 =   +  u
x̃˙ 0 A − Lc C x̃ 0
..
      
˙x̂1 0 1 . 0 0.875 x̂ 0
 1  
..
  
˙  
 x̂2  −1 −2 . 0 −1.25  x̂2   1 
   
      
. . . =  . . . . . . . . . . . . . . .  . . . + . . . u
      
..
      
˙      
 x̃1   0 0 . 0 0.125   x̃1   0 
..
      
x̃˙ 2 0 0 . −1 −0.75 x̃2 0

2. For dynamic system

 
 
0 1 0 0
   
ẋ =  0 1  x + 0 u
   
0
   
−4 −4 −1 1
h i
y= 1 0 0 x

Determine the gain Lc that result in estimation error dynamics with poles at p1 =
−1, p2 = −2, p3 = −3 and closed loop dynamics of the system with full state
feedback using full state observer, if the feedback gain Kf required to place the
closed loop poles of the system at s = −2, s = −2 ± 2i.

Solution
Check for observability of the system
   
C 1 0 0
   
O v =  CA  = 0 1 0
   
   
2
CA 0 0 1

which has rank 3, hence the system is observable.

The characteristic of the observer is (s + 1)(s + 2)(s + 3) = s3 + 6s2 + 11s + 6. The

poles of (A − Lc C) must be at the desired poles. so

det(sI − A + Lc C) = s3 + 6s2 + 11s + 6

Chapter 4 121
Daniel Abebe Modern Control Systems Lecture Note
      
s 0 0 0 1 0 l
     1 h i
3 2
det 0 s 0 −  0 1  + l2  1 0 0  = s + 6s + 11s + 6
      
0
      
0 0 s −4 −4 −1 l3
 
s + l1 −1 0
 
3 2
det  l2 −1  = s + 6s + 11s + 6
 
s
 
l3 + 4 4 s+1

s3 + (1 + l1 )s2 + (l1 + l2 + 4)s + (4l1 + l2 + l3 + 4) = s3 + 6s2 + 11s + 6

The state observer gain matrix that puts the poles of the state observer at the
desired location is then,

1 + l1 = 6 ⇒ l1 = 5,
l1 + l2 + 4 = 11 ⇒ l2 = 2
4l1 + l2 + l3 + 4 = 6 ⇒ l3 = −20
   
l 5
 1  
Lc = l2  =  2 
   
   
l3 −20

The full state observer (Luenberger state observer) dynamics is

      
˙x̂ A Lc C x̂ B
 =   +  u
x̃˙ 0 A − Lc C x̃ 0
..
      
˙x̂1 0 1 0 . 5 0 0 x̂ 0
 1  
..
  
˙  
 x̂2   0 0 1 . 2 0 0   x̂2   0 
   
..
      
 x̂3  −4 −4 −1 . −20 0 0   x̂3   1 
      
      
. . .  . . . . . . . . . . . . . . . . . . . . .  . . . + . . . u
=
      
..
      
˙  
−5 1
   
 x̃1   0 0 0 . 0   x̃1   0 
..
      
˙  
−2 0
   
 x̃2   0 0 0 . 1   x̃2   0 
..
      
x̃3 0 0 0 . 16 −4 −1 x̃3 0

Chapter 4 122
Daniel Abebe Modern Control Systems Lecture Note

The state feedback gain matrix Kf can be determined as

∥sI − A + BKf ∥ = (s + 2)(s + 2 + 2i)(s + 2 − 2i)

     
s 0 0 0 1 0 0 0 0
     
2
0 s 0 −  0 1  +  0 0 0  = (s + 2)(s + 4s + 8)
     
0
     
0 0 s −4 −4 −1 k1 k2 k3
 
s −1 0
 
−1  = s3 + 6s2 + 16s + 16
 
 0 s
 
4 + k1 4 + k2 s + 1 + k3
s(s(s + 1 + k3 ) + 4 + k2 ) + 4 + k1 = s3 + 6s2 + 16s + 16
s3 + (1 + k3 )s2 + (4 + k2 )s + 4 + k1 = s3 + 6s2 + 16s + 16
1 + k3 = 6 ⇒ k3 = 5 4 + k2 = 16 ⇒ k2 = 12 4 + k1 = 16 ⇒ k1 = 12
h i h i
Kf = k1 k2 k3 = 12 12 5

The closed-loop dynamics with a full-order state observer based on state feedback
control are
    
ẋ A − BK BKf x
 =  
x̃˙ 0 A − Lc C x̃

Replacing the matrices, we have

..
    
ẋ1 0 1 0 . 0 0 0 x1
..
    
 ẋ2   0 0 1 . 0 0 0   x2 
    
.
    
 ẋ3  −16 −16 −6 .. 12 12 5   x3 
    
    
. . . =  . . . . . . . . . . . .
    
. . . . . . . . .  . . .
..
    
˙  
−5 1
 
 x̃1   0 0 0 . 0   x̃1 
..
    
˙  
−2 0
 
 x̃2   0 0 0 . 1   x̃2 
..
    
x̃˙ 3 0 0 0 . 16 −4 −1 x̃3

Chapter 4 123
Daniel Abebe Modern Control Systems Lecture Note

3. For a dynamic system

     
ẋ 0 1 0 0
 1    
ẋ2  =  0 1  x + 0 u
     
0
     
ẋ3 −4 −4 −1 1
 
x
h i  1
y = 1 0 0 x2 
 
 
x3

Design a reduce order state observer for the system that has poles at p1 = −5, p2 =
−5 and closed loop dynamics of the system with full state feedback using reduced
order state observer, if the feedback gain Kf required to place the closed loop poles
of the system at s = −2, s = −2 ± 2i.

Solution
Decomposing the system into a state that can be measured directly from the output
and that can’t be measured directly from the output,
..
   
0 . 1 0 1
   
   
. . . . . . . . . . . . . . .
A= 
..

 , B =  
 
0 . 0 1 0
..
   
−4 . −4 −1 0
.
h i
C = 1 .. 0 0

We have
h i
A11 = 0, A12 = 1 0 B1 = 1 C1 = 1
     
0 0 1 0 h i
A21 =   A22 =   B2 =   C2 = 0 0
−4 −4 −1 0

As we can see from the decomposition of the system, x2 and x3 can’t be measured
directly, while x1 can be measured directly. So the observer is designed for x2 and
x3 .

The characteristic of the observer is (s + 5)(s + 5) = s2 + 10s + 25. The poles of

(A22 − Lc A12 ) must be at the desired poles. Checking for the observability of the
pair (A22 , A12 )
 
1 0
Ov =  
0 1

Chapter 4 124
Daniel Abebe Modern Control Systems Lecture Note

which have rank 2, hence it is completely observable.

det(sI − A22 A + Lc A12 ) = s2 + 10s + 25

      
s 0 0 1 l1 h i
det  −  +   1 0  = s2 + 10s + 25
0 s −4 −1 l2
 
s + L1 −1
det   = s2 + 10s + 25
4 + L2 s + 1

s2 + (1 + l1 )s + (l1 + l2 + 4) = s2 + 10s + 25

The state observer gain matrix that puts the poles of the state observer at the
desired location is then,

1 + l1 = 10 ⇒ L1 = 9, l1 + l2 + 4 = 25 ⇒ l2 = 12
   
l1 9
Lc =   =  
l2 12

The matrices
 
−9 1
J = A22 − Lc A12 =  
−16 −1
 
−69
N = A21 + JLc + Lc A11 =  
−160
 
0
M = B2 − Lc B1 =  
10

The minimum order state observer dynamics is

        
ω̇ −9 1 ω −69 0
 1 =    1 +   x1 +   u
ω̇2 −16 −1 ω2 −160 10

The reduced-order state observer output equation is then

 
x
 1
x 
 
 2
..
    
 
x̂2 9y + ω1 9 0 0 . 1 0  x3 
 = =
..
 
 
x̂3 12y + ω2 12 0 0 . 0 1 . . .
 
 
 ω1 
 
ω2

Chapter 4 125
Daniel Abebe Modern Control Systems Lecture Note

The combined open-loop system dynamics of the system and reduced-order observer
dynamics are given as,
..
      
ẋ 0 1 0 . 0 0 x 0
 1   1  
..
 ẋ2   0 0 1 . 0 0   x2   0 
      
..
      
 ẋ3   −4 −4 −1 .
      
0 0   x3   1 
 =   +  u
      
. . .  . . . . . . . . . . . . . . . . . .  . . . . . .
..
      
 ω̇1   −69 0 −9 1   ω1   0 
      
0 .
..
      
ω̇2 −160 0 0 . −16 −1 ω2 1

Since we have already calculated the feedback gain Kf in example 2, to determine

the closed-loop dynamics of the system with reduced-order observer,
h i
A11 − B1 KS1 = 0 − 12 12 5 S1
h i⊤
which imply S1 is 3 × 1 matrix, Let S1 = 1 1 1 , with this
 
1
h i 
A11 − B1 Kf S1 = 0 − 12 12 5 1 = −29
 
 
1

h h i i
A12 − B1 Kf S2 = 1 0 − 12 12 5 S2
 ⊤
1 1 1
which imply S2 is 3 × 2 matrix, let S2 =   we have,
1 1 1
 
1 1
h i h i  h i h i h i
A12 − B1 KS2 = 1 0 − 12 12 5 1 1 = 1 0 − 29 29 = −28 −29
 
 
1 1

 
1 1
h i  h i
B1 KS2 = 12 12 5 1 1 = 29 29
 
 
1 1

 
 
  1  
0 0 h i   0
A21 − B2 KS1 =   −   12 12 5 1 =  
 
−4 0   −4
1

Chapter 4 126
Daniel Abebe Modern Control Systems Lecture Note
 
 
  1 1  
0 1 0 h i   0 1
A22 − B2 KS2 =   −   12 12 5 1 1 = 
 
−4 −1 0   −4 −1
1 1

 
  1 1  
0 h i   0 0
B2 KS2 =   12 12 5 1 1 = 
  
0   0 0
1 1

  
  
0 1 9 h i −9 1
A22 − Lc A12 = −  1 0 = 
−4 −1 12 −16 −1

Resulting in the closed-loop system dynamics with reduced state observer after
h i⊤
replacing η1 = x1 (directly measured states)and η 2 = x2 x3 (estimated states)
we have,
.. .. ..
 
−29 . −28 −29 . 0 . 29 29
    
ẋ1  . . . . . . . . . . . . . . . . . . . . . . . . . . .  x1
 
.. .. ..
    
 ẋ2   0 . 0 1 . 0 . 0 0   x2 
    
.. .. ..
    
 ẋ3   −4 . −4 −1 . 0 . 0 0   x3 
    
    
. . . =  . . . . . . . . . . . . . . . . . . . . . . . . . . .  . . .
    
.. .. ..
    
˙    
 x̃1   0 . 0 0 . 0 . 0 0   x̃1 
    
˙    
 x̃2   . . . . . . . . . . . . . . . . . . . . . . . . . . .   x̃2 
.. .. ..
    
˙x̃3 
 0 . 0 0 . 0 . −9 1  x̃3

.. .. ..
 
0 . 0 0 . 0 . −16 −1

4. For a dynamic system

     
ẋ 0 1 0 0
 1    
ẋ2  =  0 1  x + 0 u
     
0
     
ẋ3 −4 −4 −1 1
 
x
h i  1
y = 0 1 0 x2 
 
 
x3

Design a reduce order state observer for the system that has poles at p1 = −5, p2 =
−5 and closed loop dynamics of the system with full state feedback using reduced

Chapter 4 127
Daniel Abebe Modern Control Systems Lecture Note

order state observer, if the feedback gain Kf required to place the closed loop poles
of the system at s = −2, s = −2 ± 2i.

Solution
Check the controllability of the system
 
0 0 1
 
0 1 −1
 
 
1 −1 −3

having rank 3 hence the system is completely controllable.

Check the observability of the system

 
0 1 0
 
 
0 0 1
 
−4 −4 −1

having rank 3 hence the system is completely observable.

Let us arrange the dynamic equation as follows to simplify the problem

      
ẋ 0 0 1 x 0
 2    2  
ẋ1  =  1 0  x1  + 0 u
      
0
      
ẋ3 −4 −4 −1 x3 1
 
x
h i  2
y = 1 0 0 x1 
 
 
x3

Thus the measurable state is η1 = x2 and unmeasurable state (to be estimated) is

h i⊤
η 2 = x1 x3 , which is straightforward (similar way as above)

..
   
0 . 0 1 0
..
      
   
η̇  . . . . . . . . . .  η1 . . .
 1 =    +  u
.
 1 .. 0
   
η̇ 2 0  η2 0
..
   
−4 . −4 −1 1
 
i η1
.
h
y = 1 .. 0 0  
η2

Chapter 4 128
Daniel Abebe Modern Control Systems Lecture Note
h i
A11 = 0 A12 = 0 1 B1 = 0 C 1 = 1
     
1 0 0 0 h i
A21 =   A22 =   B2 =   C2 = 0 0
−4 −4 −1 1

we have

A22 − Lc A12 − J = 0 ⇒ J = A22 − Lc A12

A21 − Lc A11 + JLc − N = 0 ⇒ N = A21 − Lc A11 + JLc
B2 − Lc B1 − M = 0 ⇒ M = B2 − Lc B1

The value of the reduced observer matrix gain Lc that places the poles of the
estimation reduced-order state observer error dynamics at the desired location is,

∥sI − J∥ = s2 + 10s + 25
∥sI − A22 + Lc A12 ∥ = s2 + 10s + 25
     
s 0 0 0 l h i
 −  +  1  0 1 = s2 + 10s + 25
0 s −4 −1 l2
 
s l1
  = s2 + 10s + 25
4 s + 1 + l2
s2 + (l2 + 1)s − 4l1 = s2 + 10s + 25
 
−6.25
l2 + 1 = 10 ⇒ l2 = 9 − 4l1 = 25 ⇒ l1 = −6.25 Lc =  
9

then
 
0 6.25
J = A22 − Lc A12 =  
−4 −10
 
57.25
N = A21 − Lc A11 + JLc =  
−69
 
0
M = B2 − Lc B1 =  
1

The reduced-order state observer output equation is

 
x
  1
.

0 −6.25 0 .. 1 0 x2 
       
x̂ −6.25y + ω1 −6.25x2 + ω1 
..
 
 1 =  = =
0 1
 
0 . 0 0 x3 
x̂3 9y + ω2 9x2 + ω2 
..
 
 
0 9 0 . 0 1 ω1 
 
ω2

Chapter 4 129
Daniel Abebe Modern Control Systems Lecture Note

The minimum order state observer dynamics is

        
ω̇1 0 6.25 ω1 57.25 0
 =   +   x2 +   u
ω̇2 −4 −10 ω2 −69 1

The open-loop system dynamics combined with minimum state observer dynamics
are
..
      
ẋ 0 1 0 . 0 0 x 0
 1   1  
..
 ẋ2   0 0 1 . 0 0   x2   1 
      
..
      
 ẋ3  −4 −4 −1
      
. 0 0   x3   0 
 =   +  u
      
. . .  . . . . . . . . . . . . . . . . . .  . . . . . .
.
      
0 57.25 ..
      
 ω̇1   0 0 6.25  ω1   0 
..
      
ω̇2 0 0 −69 . −4 −10 ω2 1
h i
Since matrix B is not changed Kf = 12 12 5 , the closed loop dynamics of the
system with reduced order observer are determined as,
h i
A11 − B1 Kf S1 = 0 − 12 12 5 S1
h i⊤
which imply S1 is 3 × 1 matrix, let S1 = 1 1 1 , with this
 
1
h i 
A11 − B1 Kf S1 = 0 − [0] 12 12 5 1 = 0
 
 
1

h i h i
A12 − B1 Kf S2 = 0 1 − [0] 12 12 5 S2
 ⊤
1 1 1
which imply S2 is 3 × 2 matrix, let S2 =   we have,
1 1 1
 
1 1
h i h i  h i
A12 − B1 Kf S2 = 0 1 − [0] 12 12 5 1 1 = 0 1
 
 
1 1

 
1 1
h i  h i
B1 Kf S2 = [0] 12 12 5 1 1 = 0 0
 
 
1 1

Chapter 4 130
Daniel Abebe Modern Control Systems Lecture Note
 
    1  
1 0 h i   1
A21 − B2 Kf S1 =   −   12 12 5 1 = 
  
−4 1   −33
1
 
 
  1 1  
0 0 0 h i   0 0
A22 − B2 Kf S2 =   −   12 12 5 1 1 = 
 
−4 −1 1   −33 −30
1 1
 
  1 1  
0 h i  0 0
B2 Kf S2 =   12 12 5 1 1 = 
  
1   29 29
1 1
     
0 0 −6.25 h i 0 6.25
A22 − Lc A12 = −  0 1 = 
−4 −1 9 −4 −10
Finally, the closed-loop system dynamics with reduced state observer after replacing
h i⊤
η1 = x2 (directly measured states)and η 2 = x1 x3 (unmeasured states) is,
.. .. ..
 
0 . 0 1 . 0 . 0 0
    
ẋ2  . . . . . . . . . . . . . . . . . . . . . . . . . . .  x2
 
.. .. ..
    
 ẋ1   1 . 0 0 . 0 . 0 0   x1 
    
.. .. ..
    
 ẋ3  −33 . −33 −30 . 0 . 29 29   x3 
    
    
. . . =  . . . . . . . . . . . . . . . . . . . . . . . . . . .  . . .
    
.. .. ..
    
˙    
 x̃2   0 . 0 0 . 0 . 0 0   x̃2 
    
˙    
 x̃1   . . . . . . . . . . . . . . . . . . . . . . . . . . .   x̃1 
.. .. ..
    
˙x̃3 
 0 . 0 0 . 0 .

0 6.25 x̃3
.. .. ..
 
0 . 0 0 . 0 . −4 −10

5. For a dynamic system

     
ẋ 0 1 0 0
 1    
ẋ2  =  0 1  x + 0 u
     
0
     
ẋ3 −4 −4 −1 1
 
x
h i  1
y = 0 0 1 x2 
 
 
x3

Chapter 4 131
Daniel Abebe Modern Control Systems Lecture Note

Design a reduce order state observer for the system that has poles at p1 = −5, p2 =
−5 and closed loop dynamics of the system with full state feedback using reduced
order state observer, if the feedback gain Kf required to place the closed loop poles
of the system at s = −2, s = −2 ± 2i.

Solution
Check the controllability of the system
 
0 0 1
 
0 1 −1
 
 
1 −1 −3

having rank 3, hence the system is completely controllable.

Check the observability of the system
 
0 0 1
 
−4 −4 −1
 
 
4 0 −3

having rank 3 hence the system is completely observable.

Let us arrange the dynamic equation as follows to simplify the problem

      
ẋ −1 −4 −4 x3 1
 3     
ẋ1  =  0 1  x1  + 0 u
      
0
      
ẋ2 1 0 0 x2 0
 
x
h i  3
y = 1 0 0 x1 
 
 
x1
h i⊤
Thus the measurable state is η1 = x3 and unmeasurable state is η 2 = x1 x2 ,
which is straightforward (similar way as above)
..
   
−1 . −4 −4 1
.
      
 . . . .. . . . . . .  η1
   
η̇ . . .
 1 =    +  u
.
 0 .. 1
   
η̇ 2 0  η2 0
..
   
1 . 0 0 0
 
i η1
.
h
y = 1 .. 0 0  
η2

Chapter 4 132
Daniel Abebe Modern Control Systems Lecture Note
h i
A11 = −1 A12 = −4 −4 B1 = 1 C1 = 1
     
0 1 0 0 h i
A21 =   A22 =   B2 =   C2 = 0 0
1 0 0 0

So we have

A22 − Lc A12 − J = 0 ⇒ J = A22 − Lc A12

A21 − Lc A11 + JLc − N = 0 ⇒ N = A21 − Lc A11 + JLc
B2 − Lc B1 − M = 0 ⇒ M = B2 − Lc B1

The value of the reduced observer matrix gain Lc that place the poles of estimation
reduced order state observer error dynamics at the desired location is

∥sI − J∥ = s2 + 10s + 25
∥sI − A22 + Lc A12 ∥ = s2 + 10s + 25
     
s 0 1 0 l h i
 −  +  1  −4 −4 = s2 + 10s + 25
0 s 0 0 l2
 
s + 1 − 4l1 −4l1
  = s2 + 10s + 25
−4l2 s − 4l2
s2 + (1 − 4l1 − 4l2 )s − 4l2 = s2 + 10s + 25
 
4
1 − 4l1 − 4l2 = 10 ⇒ l2 = 4 − 4l2 = 25 ⇒ l2 = −6.25 Lc =  
−6.25

then
 
17 16
J = A22 − Lc A12 =  
−25 −25
 
−28
N = A21 − Lc A11 + JLc =  
49
 
−4
M = B2 − Lc B1 =  
6.25

The reduced-order state observer output equation is

 
x
  1
.

      0 0 4 .. 1 0 x2 


x̂1 4y + ω1 4x3 + ω1 
.
 
 = = 0 0 −6.25 .. 0 1 x3 
=  
x̂2 −6.25y + ω2 −6.25x3 + ω2 
..
 
 
0 0 1 . 0 0 ω1 
 
ω2

Chapter 4 133
Daniel Abebe Modern Control Systems Lecture Note

The minimum order state observer dynamics is

        
ω̇ 17 16 ω −28 −4
 1 =    1 +   x3 +  u
ω̇2 −25 −25 ω2 49 6.25

The open-loop system dynamics combined with minimum state observer dynamics
are
..
      
ẋ1 0 1 0 . 0 0 x 0
 1  
..
  
 ẋ2   0 0 1 . 0 0   x2   1 
      
..
      
 ẋ3  −4 −4 −1
      
. 0 0   x3   0 
 =   +  u
      
. . .  . . . . . . . . . . . . . . . . . .  . . . . . .
.
      
0 −28 ..
      
 ω̇1   0 17 16   ω1   0 
..
      
ω̇2 0 0 49 . −25 −25 ω2 1
h i
Since matrix B is not changed Kf = 12 12 5 , the closed loop dynamics of the
system with reduced order observer are
h i⊤
by letting S1 = 1 1 1 , with this

A11 − B1 KS1 = −30

 ⊤
1 1 1
Let S2 =   we have,
1 1 1
h i
A12 − B1 KS2 = −33 −33

h i
B1 KS2 = 29 29

 
0
A21 − B2 KS1 =  
−1

 
1 0
A22 − B2 KS2 =  
0 0

 
0 0
B2 KS2 =  
0 0

Chapter 4 134
Daniel Abebe Modern Control Systems Lecture Note
 
17 16
A22 − Lc A12 =  
−25 −25
h i⊤
Replacing η1 = x3 (directly measured states)and η 2 = x1 x2 (unmeasured
states),, the closed-d-loop system with reduced state observer is,
.. .. ..
 
−30 . −33 −33 . 0 . 29 29
    
ẋ3  . . . . . . . . . . . . . . . . . . . . . . . . . . .  x3
 
.. .. ..
    
 ẋ1   0 . 1 0 . 0 . 0 0   x1 
    
.. .. ..
    
 ẋ2   −1 . 0 0 . 0 . 0 0   x2 
    
    
. . . =  . . . . . . . . . . . . . . . . . . . . . . . . . . .  . . .
    
.. .. ..
    
˙    
 x̃3   0 . 0 0 . 0 . 0 0   x̃3 
    
˙    
 x̃1   . . . . . . . . . . . . . . . . . . . . . . . . . . .   x̃1 
.. .. ..
    
˙x̃2 
 0 . 0 0 . 0 . 17

16  x̃2
.. .. ..
 
0 . 0 0 . 0 . −25 −25

6. For the system below, design a minimum order state observer.

       
ẋ 0 1 0 0 x
 1     h i  1
ẋ2  =  0 1  x +  1  u y = 1 0 0 x2 
       
0
       
ẋ3 −6 −11 −6 −3 x3

Solution
The observability matrix of the system is
 
1 0 0
 
 
0 1 0
 
0 0 1

which has full rank, implying that the system is completely observable. But from
the system output matrix, only one state can be directly measured. The two states
are not available. Hence, the reduced order observer has order 2. If the observer is
required to give a pole at p = −2 ± 2i, then we have
..
   
0 . 1 0 0
   
   
. . . . . . . . . . . . . . .
ẋ = 
 ..
x +  u
  
0 . 0 1 1
..
   
−6 . −11 −6 −3
.
h i
y = 1 .. 0 0

Chapter 4 135
Daniel Abebe Modern Control Systems Lecture Note

Implying that
   
0 1 0
h i
A22 =   A11 = 0 A12 = 1 0
A21 =  
−11 −6 −6
 
1 h i
B1 = 0 B2 =   C1 = 1 C2 = 0 0
−3

So J = A22 − LA12 is the error dynamics matrix which needs to be stabilized.

   
s −1 l h i
  +  1  1 0 = s2 + 4s + 8
11 s + 6 l2
 
s + l1 −1
  = s2 + 4s + 8
11 + l2 s + 6
s2 + (6 + l1 )s + 6l1 + l2 + 11 = s2 + 4s + 8
6 + l1 = 4 ⇒ l1 = −2 6l1 + l2 + 11 = 8 ⇒ l2 = 9
   
−2 2 1
Lc =   ⇒ J =  
9 −20 −6

Finally
 
   
0 13 13
N = A21 + JLc + Lc A11 =   +   =  
−6 4 −2
 
1
M = B2 − Lc B1 =  
−3

The minimum order state observer dynamic equation is

        
ω̇ −2 1 ω 13 1
 1 =    1  +   x1 +   u
ω̇2 −20 −6 ω2 −2 −3

The reduced-order state observer output equation is then

 
x
  1
.

1 0 0 .. 0 0 x2 
     
x̂ −2y + ω1 
..
 
 2 =  =
−2 0 0 . 1 0 x3 
 
x̂3 9y + ω2 
..
 
 
9 0 0 . 0 1 ω1 
 
ω2

Chapter 4 136
Daniel Abebe Modern Control Systems Lecture Note

The open-loop dynamics of the system, combined with reduced-order state observer
dynamics, are
..
      
ẋ1 0 1 0 . 0 0 x 0
 1  
..
  
 ẋ2   0 0 1 . 0 0   x2   1 
      
..
      
 ẋ3  −6 −11 −6 . 0   x3  −3
      
0
 =   +  u
      
. . .  . . . . . . . . . . . . . . . . . .  . . .  . . . 
..
      
−2 1   ω1   1 
      
 ω̇1   13 0 0 .
..
      
ω̇2 −2 0 0 . −20 −6 ω2 −3

7. For the system in chapter 2 on example 2 design a minimum order observer.

Solution
As shown in chapter 2 on example 2, the system has one unobservable mode. So
we cannot design an observer for the unobservable mode of the system, but we can
design for the observable modes of the system. As we can see from the output
matrix of the decomposed system, one state can be directly measured, while one
cannot. Thus, we design a state observer of order 1. The dynamics look of the
observable mode of the system is
..
   
2 . 1 1
   
ẋ =  . . . . . . . . . x + . . . u
   
..
   
−4 . 0 1
.
h i
y = 1 .. 0 x

From this A11 = 2, A12 = 1, A21 = −4 A22 = 0. If the observer is required to have
a pole at p = −1, then

sI − A22 + lA12 = s + 1
s − 0 + l = s + 1 ⇒ l = 1 ⇒ Lc = l = 1
J = A22 − Ac A12 = 0 − 1 = −1
N = A21 + JLc + Lc A11 = −4 − 1 + 2 = −3
M = B2 − Lc B1 = 1 − 1 = 0

The minimum order state observer dynamic equation is

ω̇ = −ω − 3x1

The reduced-order state observer output equation is then

 
x
i  1
..
h i h i h
x̂2 = Lc y + ω = 1 0 . 1 x2 
 
 
ω

Chapter 4 137
Daniel Abebe Modern Control Systems Lecture Note

The open-loop system state equation, combined with the reduced-order state ob-
server equation, is
..
      
ẋ1 2 1 . 0 x 1
 1  
..
  
 ẋ2  −4 0
      
. 0   x2   1 
 =   +  u
      
. . .  . . . . . . . . . . . .  . . . . . .
..
      
ω̇ −3 0 . 1 ω 0

4.7 Exercises
1. Consider a system
  

0.16 2.16 −1 h i
ẋ =   x+   u y= 1 1 x
−0.16 −1.16 1

Find it open-loop eigenvalues, corresponding eigenvectors, and check the stability,

controllability, and observability of the system. Determine feedback gain matrix Kf
such that both eigenvalues of the closed-loop system are at −1, and finally design
a minimum order observer for this system that has an eigenvalue at −3.

2. Design a full state observer for IPC dynamics on 1.6.3 that has poles at −1, −1, −1±
i. Use any valid parameter.

3. Design a full state observer for DC motor dynamics having poles at −1 ± 2i. Use
any valid parameters.

4. Design a reduced-order state observer for the dynamics of the DC motor on 1.7.1,
where the speed is directly measured and the armature current ia is required to be
estimated. The pole of the observer must be at −1, use any valid parameters.

5. Design a reduced state observer for IPC dynamics on 1.6.3, where the position of the
cart and position of the pendulum are directly measured and the angular velocity
of the pendulum and linear velocity of the cart are not. The poles of the observer
are at−1 ± 2i.

6. Design a Kalman-Bucy filter for IPC dynamics and DC motor dynamics. Use any
valid parameters.

7. For a dynamic system

     
0 0 1 1 0
ẋ =  x +  u +  ν
1 −1 0 0 1
h i
y = 0 1 x+e

Design a Kalman filter.

Chapter 4 138
Daniel Abebe Modern Control Systems Lecture Note

8. For a system described by

   
−1 −2 −2 2
    h i
ẋ =  0 −1 1  x + 0 u y= 1 1 0 x
   
   
1 0 −1 1
Check controllability, observability, stabilizability, and detectability. Finally, design
an observer-based feedback control system that places all closed-loop eigenvalues of
the system at −2. The observer has all poles at −4.
9. Design a full state observer having poles at −1, −1 ± 2i, for a system described by
   
0 3 0 2
    h i
ẋ = 0 2 1 x +  1  u y = 0 2 1 x
   
   
1 1 0 −1

10. For a dynamic system

 
 
0 1 0 0
   
ẋ =  0 1  x + 1 u
   
0
   
−4 −4 −1 1
h i
y= 0 1 0 x
determine minimum order state observer having pole at −2, −2 ± 2i.
11. For a dynamic system
  

0 1 0 1
   
ẋ =  0 1  x + 0 u
   
0
   
−4 −4 −1 1
h i
y= 0 1 0 x
determine minimum order state observer having pole at −2, −2 ± 2i.
12. For a dynamic system
 
 
0 1 0 0
   
ẋ =  0 1  x + 1 u
   
0
   
−4 −4 −1 0
h i
y= 1 0 1 x
determine minimum order state observer having pole at −2, −2 ± 2i.
13. For examples 6 and 7 determine the closed loop dynamics of the system with full
state feedback using reduced order state observer, if the feedback gain Kf required
to place the closed loop poles of the system at s = −1, s = −1 ± 1i

Chapter 4 139
Chapter 5

Introduction to Optimal Control

Optimal control deals with the most celebrated problem of modern control theory, ”deter-
mination of best control strategy, selecting optimum control strategy, selecting optimum
control vector u(t), that minimizes cost function or performance index”. Consider an
LTI system with initial condition x0 :

ẋ = Ax + Bu
y = Cx + Du (5.0.1)

The objective of optimal control is to determine the control vector u that forces the
behavior of the dynamic system to the required final state by maximizing the required
cost function and at the same time fulfilling the physical constraints.
The cost function has a general representation of:
Z tf
t=tf
J = θ[x(t), t] |t=t0 + φ(x(t), u(t), t)dt (5.0.2)
t0

t=t
The first term θ[x(t), t] |t=tf0 is called terminal error optimization problem (terminal cost
Z tf
function) and the second term φ(x(t), u(t), t)dt is called optimal cost function over
t0
the entire trajectory (the function to be optimized over the entire trajectory).

5.1 Performance Indices

The cost function (performance index) may be a function that optimizes energy con-
sumed, optimizes the time required to reach the desired state, or minimizes the tracking
or terminal value error. Depending on the desired optimization problem, we have five
performance indices.

5.1.1 Minimum time control problem

The aim is to determine a control vector that derives the state of the dynamic system
from its initial value x0 to its final value xf in minimum time. The best example is car
racing, where finishing time is the only required parameter to be minimized. The cost
function for minimum time is given as
Z tf
J= dt (5.1.1)
t0

140
Daniel Abebe Modern Control Systems Lecture Note

5.1.2 Minimum terminal error control problem

The aim is to determine a control vector u that minimizes terminal error. Consider a
system with state x, we want to reach xd after a certain time t. In a terminal control
problem, we need to determine the control signal u that minimizes ξ = xf − xd . The
best example is rocket launching, where the emphasis is put on the target only. The cost
function of this control problem is:

J = (xf − xd )⊤ S(xf − xd ) (5.1.2)

where S is n × n, symmetric positive definite matrix.

5.1.3 Minimum effort/energy control problem

The aim is to determine the minimum amount of energy that is required to derive the
system from x0 to xf . The best example is car braking, where stopping time is required
to be achieved with minimum energy. The cost function is:
Z tf
J= u⊤ Ru dt (5.1.3)
t0

where R is m × m, symmetric positive definite matrix.

5.1.4 Optimal servo-mechanism or tracking problem

This control problem aims to determine a control vector u that minimizes the tracking
error when driving the system from any initial state x0 to the desired state (final state).
If the desired trajectory is xd over the course from t0 to tf , the control vector u is required
to minimize x̃ = x − xd . The best example is space shuttle tracking and control, where
tracking error, energy consumption, and the difference between the desired terminal point
and the shuttle’s final point are all required to be minimized. The cost function for this
control problem is:
Z tf
J= ((x − xd )⊤ Q(x − xd ))dt (5.1.4)
t0

where Q is n × n, symmetric positive definite matrix. If we want to minimize the en-

ergy required by the system, in addition to minimizing tracking error, the cost function
becomes the sum of tracking error and the minimum effort cost function.
Z tf Z tf
⊤
J= ((x − xd ) Q(x − xd ))dt + u⊤ Ru dt
t t0
Z 0tf
= ((x − xd )⊤ Q(x − xd ) + u⊤ Ru) dt (5.1.5)
t0

If we add terminal error minimization to the problem in addition to minimizing tracking

error and energy optimization, the cost function takes a general form,
Z tf
⊤
J = (xf − xd ) S(xf − xd ) + ((x − xd )⊤ Q(x − xd ) + u⊤ Ru) dt (5.1.6)
t0

Chapter 5 141
Daniel Abebe Modern Control Systems Lecture Note

5.1.5 Optimal regulator problem

This control problem is a special case of the optimal tracking problem where the desired
final state is a point in space, particularly an origin xd = 0. The best example is restoring
the equilibrium point of any dynamic system. The tracking optimal problem cost function
is:
Z tf
⊤
J = xf Sxf + (x⊤ Qx + u⊤ Ru) dt (5.1.7)
t0

5.2 Review on Elements of Calculus of Variation

5.2.1 Increment of a function and a functional
Consider a function f (t), an increment of function f (t) is defined as

∆f = f (t + ∆t) − f (t) (5.2.1)

∆f depends both on t and ∆t, hence it is written as ∆f (t, ∆t). In similar way for a
functional J(x(t), t) the incremental of a functional is defined as:

∆J = J((x(t) + δx(t)), t) − J(x(t), t) (5.2.2)

δx(t) is called variation of the function x(t). Since the increment of a functional depends
on x(t) and δx(t) it is written as J((x(t), δx(t)).

5.2.2 Differential and variation

Differential of a function:- For a function f (t), the Taylor series expansion about a
point t∗ is given as
1 ∂ 2f
   
∂f
∆f = ∆t + ∆t2 + . . . − f (t∗ ) (5.2.3)
∂t t∗ 2! ∂t2 t∗

neglecting higher-order terms in ∆t,

 
∗ ∂f
∆f = f (t ) + ∆t − f (t∗ )
∂t t∗
∆f = f˙(t∗ )∆t = df (5.2.4)

df is called the differential of f at the point t∗ , and f˙(t) is the slope of f at t∗ .

Variation of a functional: Expanding a functional J(x(t), t) using the Taylor series on

a function x∗ (t), gives us:

1 ∂ 2J 2
   
∗ ∂J
∆J = J(x (t)) + δx + 2
δx + . . . − J(x∗ (t)) (5.2.5)
∂x ∗
x (t) 2! ∂x ∗
x (t)

∂J 1 ∂ 2J 2
δJ = δx is called the first variation, and δ 2 J = δx is called the second variation
∂x ∗
2! ∂x2
of the functional about a function x (t).

Chapter 5 142
Daniel Abebe Modern Control Systems Lecture Note

Neglecting higher-order terms in δx, linear approximation of variation of J(x(t), t) about

a function x∗ (t) is,

∂J
∆J ≈ δJ = δx (5.2.6)
∂x

5.2.3 Optimization of a function and unconstrained functional

For a function to have an optimum value at a point, the following two conditions should
be met.

• Necessary Condition: For a function to have an optimum value at a point t∗ , its

first derivative must vanish at the point(i.e. df = 0 at t∗ ).

• Sufficient Condition: For a function to have an optimum value at a point t∗ ,

either d2 f > 0 for the minimum or d2 f < 0 for maximum, for all t, such that
(t − t∗ < ϵ, for ϵ > 0).

• If d2 f = 0, the point is called an inflection point or a stationary point.

Similar conditions hold for functional to have optimum value, the necessary condition
is that its first variation must vanish (∆J = 0) at x∗ (t) and the sufficient condition is
that ∆2 J < 0 for local maximum or ∆2 J > 0 for local minimum, for all x(t), such that
(x(t) − x∗ (t) < ϵ, for ϵ > 0).

A functional may be a function of one or more functions, like a function of several variables
V (x, y, z). In a similar manner, we have a functional which can be J(x(t), y(t), z(t)). The
only difference is in function; the variables x, y, and z are independent variables, while
in functional, they are a function of another independent variable.

In a control system, we deal with functions of the form

J(x(t), ẋ(t), u(t), t) (5.2.7)

If the system is time-invariant, the functional will be

J(x(t), ẋ(t), u(t)) (5.2.8)

Consider a functional J(x, ẋ) defined as in equation A.5.1, if we take δx and δ ẋ as

deviation from x and ẋ respectively,
Z tf
J(x, ẋ) = φ(x, ẋ)dt (5.2.9)
t0

The increment of J(x, ẋ) will be

Z tf
∆J = (φ(x + δx, ẋ + δ ẋ) − φ(x, ẋ)) dt (5.2.10)
t0

Chapter 5 143
Daniel Abebe Modern Control Systems Lecture Note

Using Taylor series expansion:

Z tf
∆J = (φ(x + δx, ẋ + δ ẋ) − φ(x, ẋ)) dt
t0
Z tf  
∂φ ∂φ
= φ(x, ẋ) + δx + δ ẋ + . . . + hot − φ(x, ẋ) dt
t0 ∂x ∂ ẋ
Z tf  
∂φ ∂φ
= δx + δ ẋ + . . . + hot dt
t0 ∂x ∂ ẋ
Z tf  Z tf  2
∂ 2φ ∂ 2φ
 
∂φ ∂φ 1 ∂ φ 2 2
= δx + δ ẋ dt + (δx) + (δ ẋ) + 2 δ ẋδx dt + . . . + hot
t0 ∂x ∂ ẋ t0 2! ∂x2 ∂ ẋ ∂ ẋ∂x
= δJ + δ 2 J + . . . + hot (5.2.11)

The linear part of ∆J with respect to δx and δ ẋ, is

Z tf  
∂φ ∂φ
δJ = δx + δ ẋ dt (5.2.12)
t0 ∂x ∂ ẋ

For the functional to have optimum value ∆J = 0 (Necessary Condition), so

Z tf  
∂φ ∂φ
δx + δ ẋ dt = 0
t0 ∂x ∂ ẋ
Z tf   Z tf   Z tf   Z tf  
∂φ ∂φ ∂φ ∂φ dδx
δx dt + δ ẋ dt = 0 =⇒ δx dt + dt = 0
t0 ∂x t0 ∂ ẋ t0 ∂x t0 ∂ ẋ dt
Z tf   Z tf  
∂φ ∂φ
δx dt + d (δx) = 0 (5.2.13)
t0 ∂x t0 ∂ ẋ

Solving the second term in equation 5.2.13 using integration by part

Z b Z b
b
udv = uv|a − vdu
a a

∂φ
By letting u = and dv = d(δx),
∂ ẋ
Z tf     Z tf   
∂φ ∂φ tf d ∂φ
d (δx) = δx |t0 − δx dt
t0 ∂ ẋ ∂ ẋ t0 dt ∂ ẋ
Z tf     Z tf    
∂φ ∂φ tf d ∂φ
d (δx) = δx |t0 − δx dt (5.2.14)
t0 ∂ ẋ ∂ ẋ t0 dt ∂ ẋ

Replacing equation 5.2.14 in equation 5.2.13 for the second part and rearranging we have,
Z tf      
∂φ d ∂φ ∂φ t
− δx dt + δx |tf0 = 0 (5.2.15)
t0 ∂x dt ∂ ẋ ∂ ẋ

Thus, for the functional to have optimum value, both

 
∂φ d ∂φ
− =0 (5.2.16)
∂x dt ∂ ẋ

Chapter 5 144
Daniel Abebe Modern Control Systems Lecture Note

over the path and

 
∂φ
δx =0 (5.2.17)
∂ ẋ
at the boundaries.

Equation A.4.5 is called the Euler-Lagrange equation, while equation A.4.6 is called the
∂φ
boundary condition equation. For the boundary condition to be zero, either must be
∂ ẋ
zero or δx must be zero at the boundary. This leads to four cases:
1. Case-I: x is fixed at t0 and tf (fixed end trajectory).
Here, x(t0 ) = C1 and x(tf ) = C2 , where C1 and C2 are constants. Since δx = 0 at
∂φ
both t0 and tf , there are no restrictions on .
∂ ẋ
2. Case-II: x is fixed at t0 and free at tf .
∂φ
Here, x(t0 ) = C1 , and = 0 at tf because δx is unknown at tf . No restrictions
∂ ẋ
∂φ
apply to at t0 .
∂ ẋ
3. Case-III: x is fixed at tf and free at t0 .
∂φ
Here, x(tf ) = C2 , and = 0 at t0 because δx is unknown at t0 . No restrictions
∂ ẋ
∂φ
apply to at tf .
∂ ẋ
4. Case-IV: x is free at t0 and tf .
∂φ
Here, = 0 at both t0 and tf because δx is unknown at both times.
∂ ẋ
The sufficient condition for the functional to have optimum value is derived using δ 2 J
(second variation),
Z tf  2
∂ 2φ ∂ 2φ

2 1 ∂ φ 2 2
δ J= (δx) + (δ ẋ) + 2 δ ẋδx dt (5.2.18)
t0 2 ∂x2 ∂ ẋ ∂ ẋ∂x
∂ 2φ
Solving the last term using integration by part u = δ ẋδx and dv = δ ẋ and rearrang-
∂x
ing,

tf
∂ 2φ ∂ 2φ ∂ 2φ
Z      
2 1 d 2 2
δ J= 2
− (δx) + (δ ẋ) dt (5.2.19)
2 t0 ∂x dt ∂ ẋ∂x ∂ ẋ2
which means
∂ 2φ ∂ 2φ ∂ 2φ
 
d
2
− ≥0 and ≥0
∂x dt ∂x∂ ẋ ∂ ẋ2
for local minimum and
∂ 2φ ∂ 2φ ∂ 2φ
 
d
2
− ≤ 0 and ≤0
∂x dt ∂x∂ ẋ ∂ ẋ2
for local maximum.

Chapter 5 145
Daniel Abebe Modern Control Systems Lecture Note

5.2.4 Optimum of constrained functional

Real-world problems, especially control problems, are restricted by different natural laws
or the limitations of technology. The optimization of such kinds of problems under
different constraints is the mainstream of optimal control. For example, the tracking
control of a satellite is restricted by the trajectory selected and the maximum input
allowed. The objective of such a control problem is to determine a control signal u that
optimizes a cost function subjected to the constraints. If we want to optimize a cost
function J at the same time required to fulfill the state dynamics of the system, then the
optimal control problem formulation is to determine a control signal u that minimizes.
Z tf
Υ(x, ẋ, u)dt (5.2.20)
t0

subjected to

ẋ = Ax + Bu (5.2.21)

Let us define a general function that includes the constraint and the function to be
optimized as

φ(x, ẋ, u) = Υ(x, ẋ, u) + λ⊤ (Ax + Bu − ẋ) (5.2.22)

The optimization problem is reduced to unconstrained optimization of the form

Z tf
Υ(x, ẋ, u) + λ⊤ (Ax + Bu − ẋ) dt


t
Z 0tf
φ(x, ẋ, u)dt (5.2.23)
t0

By the Euler-Lagrangian equation for the function to have an optimum value

 
∂φ d ∂φ
− =0 (5.2.24)
∂x dt ∂ ẋ
 
∂φ d ∂φ
− =0 (5.2.25)
∂u dt ∂ u̇
 
∂φ d ∂φ
− =0 (5.2.26)
∂λ⊤ dt ∂ λ̇⊤

Chapter 5 146
Daniel Abebe Modern Control Systems Lecture Note

5.3 Solution to Optimal Linear Quadratic Regulator Problem

The problem is determining control law u that minimize cost function on equation 5.1.7
subjected to ẋ = Ax + Bu. Assume we need to bring the final state to zero (xf = 0).
Which means we need to find u that minimize
Z tf
x⊤ Qu + u⊤ Ru + λ⊤ (Au + Bu − ẋ) dt


J= (5.3.1)
t0

Matrix Q is a weighing matrix for state error, and R is a weighing matrix for input.
The Choice depended on the optimization problem and design specifications. If state
error is required to be minimized matrix Q should be high, and if input to the system is
required to be minimized more than the state tracking error matrix R should be high.
In state feedback controller design, penalizing input is the same as penalizing state, since
the control signal is a linear combination of the state of the system.

The possible starting of guessing the values of the matrices can be one of the following
• Simplest choice: Q = I, R = ρI ⇒ L = x⊤ · x + ρu⊤ u then vary ρ to get something
that has good response.
• Diagonal weights
 
q 0 0 0
 1 
 
 0 q2 0 0 
Q=



 0 0 q3 0 
 
0 0 0 q4
1
Choose each qi = , where x̃i is the tolerable error of each state of the system.
x̃2i
• Output weighting
Let z = Hx be the output you want to keep small. Assume (A, H) observable. Use
Q = x⊤ x, R = ρI ⇒ trade off z⊤ · z vs ρu2 .
Where H = x⊤ Qx + u⊤ Ru + λ⊤ and λ(t) = P(t)x(t)
• Trial and error (on weights)
Most of the time Q = C⊤ C and the weight of the matrix is adjusted based on the per-
formance requirement of the system.

Our objective is to determine a control law u(t) that optimizes the optimal linear quadratic
regulator cost function subjected to ẋ = Ax + Bu. To do so, we use Euler-Lagrangian
equations:
 
∂φ d ∂φ ⊤
− = 0 ⇒ x⊤ Q + x⊤ Q⊤ + λ⊤ A − λ̇ = 0 (5.3.2)
∂x dt ∂ ẋ
 
∂φ d ∂φ
− = 0 ⇒ u⊤ Q + u⊤ Q⊤ + λ⊤ B = 0 (5.3.3)
∂u dt ∂ u̇
 
∂φ d ∂φ
− = 0 ⇒ ẋ = Ax + Bu (5.3.4)
∂λ⊤ dt ∂ λ̇⊤

Chapter 5 147
Daniel Abebe Modern Control Systems Lecture Note

⊤
Let λ = Px, where P is positive definite symmetric matrix, So λ⊤ = x⊤ P and λ̇ =
Q + Q⊤ R + R⊤
ẋ⊤ P = (Ax + Bu)⊤ P = (x⊤ A⊤ + u⊤ B⊤ )P, replacing for Q = ,R= ,
2 2
⊤
λ⊤ and λ̇ , the Euler Lagrangian equation become:

x⊤ Q + x⊤ PA + (x⊤ A⊤ + u⊤ B⊤ )P = 0
x⊤ (Q + pA + A⊤ P) + u⊤ B⊤ P = 0 (5.3.5)
⊤ ⊤
u R + x PB = 0 (5.3.6)
ẋ = Ax + Bu (5.3.7)

If the control law is design based on state feedback u = −Kf x, optimizing u(t) is sames
as determining the optimal state feedback gain matrix Kf , hence with state feedback
control law, equation 5.3.5 becomes

x⊤ (Q + PA + A⊤ P)x⊤ K⊤ ⊤
fB P = 0

x⊤ (Q + PA + A⊤ P − K⊤ B⊤ P) = 0
Q + PA + A⊤ P − K⊤ ⊤
fB P = 0 (5.3.8)

and from equation 5.3.6.

− x⊤ K⊤ ⊤
f R + x PB = 0

x⊤ (−K⊤
f R + PB) = 0

K = R−1 B⊤ P (5.3.9)

The control law u(t) that optimizes the functional (cost function) is,

u = −Kf x(t) = −R−1 B⊤ Px (5.3.10)

K is the optimal feedback gain matrix that optimizes u(t), which in turn optimizes our
optimal linear quadratic regulator cost function, subjected to ẋ = Ax + Bu.

Replacing Kf in equation 5.3.8 results in a Riccati equation in which we solve for the
matrix P.

Q + PA + A⊤ P − PBR−1 B⊤ P = 0 (5.3.11)

given that (A, B) is stabilizable.

The other method of finding solution for linear quadratic optimal regulator without using
d ⊤
Euler-Lagrangian equations achieved is by letting x⊤ (Q+K⊤ f RKf )x = − (x Px) after
dt
replacing for u in equation 5.1.7 and final state xf = 0 (same control objective bring the
system to zero final state). which results in

x⊤ (Q + K⊤ ⊤ ⊤
f RKf )x = −ẋ Px − x Pẋ

x⊤ (Q + K⊤ ⊤ ⊤ ⊤ ⊤ ⊤
f RKf )x = −x (A − Kf B )Px − x P(A − BKf )x

A⊤ P − K⊤ ⊤ ⊤
f B P + PA − PBKf = −(Q + Kf RKf ) (5.3.12)

Chapter 5 148
Daniel Abebe Modern Control Systems Lecture Note

Finding u that minimize J is same as finding Kf that minimize J, which in-turn minimizes
equation 5.3.11. Derivative of equation 5.3.11 with respect to Kf to get Kf that minimize
the equation result in:

(B⊤ P)⊤ + PB − (RKf )⊤ + K⊤

fR = 0

Kf = R−1 B⊤ P (5.3.13)

Since R = R⊤ are P = P⊤ symmetric positive definite matrices.

So the control signal that minimizes the optimal quadratic regulator function is

u = −Kf x = −R−1 B⊤ Px (5.3.14)

With this control law, the closed-loop system becomes

ẋ = Ax + Bu ⇒ ẋ = (A − BKf )x

P is determined from Riccati equation after replacing for Kf = R−1 B⊤ P in equation

5.3.12 which is the same as equation 5.3.11. The Block diagram representation of a
system with an optimal state feedback controller same as a full state feedback controller,
is shown in Figure 5.1.

ud + u + x y
- B ʃ C
+

Kf

Figure 5.1: Block Diagram Representation of LTI System with Optimal State Feedback Controller

Chapter 5 149
Daniel Abebe Modern Control Systems Lecture Note

5.4 Optimal Steady State Reference Tracking

In control systems, it is often desired to achieve fast convergence of the controlled output
y to a specific nonzero constant set-point value r. The set-point values (xd , ud ), xd ,
represents the desired state of the system and while ud represent the desired control
signal that the output of system require to follow the reference input.

ẋ = Ax + Bu
y = Cx + Du (5.4.1)

At this point, the rate of change of the state variables will be zero, and the output is
expected to be equal to the reference values. Hence for a pair (xd ,ud ) equation 5.4.1 will
be

Axd + Bxd = 0
r = Cxd + Dud (5.4.2)
(5.4.3)

    
−A B −x 0
   d =   (5.4.4)
−C D ud r

This corresponds to an LQR criterion of the form

Z tf
J= (ỹ⊤ Qỹ + ũ⊤ Rũ) dt (5.4.5)
t0

where ỹ = y − r and ũ = u − ud . To understand when this is possible, three distinct

cases should be considered:

• Case-I: When the number of inputs m, is strictly smaller than the number of
controlled outputs p, we have an under-actuated system. In this case, the system of
equations 5.4.4 generally does not have a solution because it presents more equations
than unknowns.

• Case-II: When the number of inputs m is equal to the number of controlled outputs
p, equation 5.4.4 always has a solution as long as the Rosenbrock’s 1 system matrix
 
sI − A B
P (s) =   (5.4.6)
−C D

is non-singular for s = 0. This means that s = 0 should not be an invariant zero

of the system and therefore it cannot also be a transmission zero of the transfer
function C(sI − A)−1 B + D.

1
Rosenbrock’s matrix is obtained by regarding the controller output as the only output of the system.

Chapter 5 150
Daniel Abebe Modern Control Systems Lecture Note

Intuitively, one should expect problems when s = 0 is an invariant zero of the

system because when the state converges to an equilibrium point, the control input
u(t) must converge to a constant. By the zero-blocking property, one should then
expect the controlled output y(t) to converge to zero and not to r.

• Case-III: When the number of inputs m is strictly larger than the number of
controlled outputs p, we have an over-actuated system and equation 5.4.4 generally
has multiple solutions.
When P(0) is full row-rank, i.e., when it has n + p linearly independent rows, the
(n + p) × (n + p) matrix P(0)P(0)⊤ is non-singular and one solution to equation
5.4.4 is given by
   
x 0
 d  = P(0)⊤ (P(0)P(0)⊤ )−1   (5.4.7)
ud r

Also in this case, s = 0 should not be an invariant zero of the system because
otherwise P(0) cannot be full rank.
The optimal set-point problem can be reduced to that of optimal regulation by considering
an auxiliary system with state x̃ = x − xd , ũ = u − ud whose dynamics are:

x̃˙ = A(x − xd ) + B(u − ud )

x̃˙ = Ax̃ + Bũ (5.4.8)

We can then regard equation 5.4.5 and equation 5.4.8 as an optimal regulation problem
for which the optimal solution is given by:

ũ = −Kf x̃ (5.4.9)

Going back to the original input and state variables u and x, we conclude that the optimal
control for the set-point problem defined by equation 5.4.1 and equation 5.4.5 is given by

u = −Kf (x − xd ) + ud (5.4.10)

Since the solution to equation 5.4.4 can be written in the form

xd = Fr (5.4.11)
ud = Nr (5.4.12)

The values of F and N are determined using equation 5.4.2, by replacing for xe = Fr and
ud = Nr:

Axd + Bud = 0 ⇒ AFr + BNr = 0 ⇒ (AF + BN) = 0 (5.4.13)

Cxd + Dud = r ⇒ CFr + DNr = u ⇒ (CF + DN) = I (5.4.14)

Solving for F from equation 5.4.13 and replacing in equation 5.4.14, the values for matrices
N and F is:

N = (D − CA−1 B)−1 (5.4.15)

F = −A−1 B(D − CA−1 B)−1 (5.4.16)

Chapter 5 151
Daniel Abebe Modern Control Systems Lecture Note

ud
N

r xd + u + x y
F + -
Kf + B ʃ C
+

Figure 5.2: Steady State Reference Tracking

The other method is at steady state

u = −Kf x + Ko r (5.4.17)

replacing for u in the state space equation;

ẋ = Ax − Bu ⇒ ẋ = (A − BKf )x + BKo r (5.4.18)

as time goes to infinity (t → ∞) the state x, goes to the desired state xd and the input
u, tends toward desired input ud , having dynamic equation:

ẋd = Axd + Bud ⇒ ẋd = (A − BKf )xd + BKo r (5.4.19)

ud = −Kxd + Ko r (5.4.20)

the tracking error x̃ = x − xd is then

x̃˙ = (A − BKf )x̃ (5.4.21)

since ẋd is zero at t → ∞, then the dynamic equation 5.4.19, can be written as

xd = −(A − BKf )−1 BKo r (5.4.22)

replacing equation 5.4.22 in equation 5.4.20, we have

ud = Kf (A − BKf )−1 BKo r + Ko r

= (Kf (A − BKf )−1 BKo + Ko )r (5.4.23)

and the output of the dynamic system y, goes to r as t → ∞,

y = Cxd + Dud = r (5.4.24)

Form equation 5.4.24, by replacing for ud and xd ,the value of matrix Ko is:

Ko = −(C(A − BKf )−1 B − D((Kf (A − BKf )−1 B + I))−1 (5.4.25)

In cases where the states of the dynamic system are not directly accessible, a state
observer can be utilized to estimate the state of the system, thereby allowing the use of
the estimated state for control purposes.

Chapter 5 152
Daniel Abebe Modern Control Systems Lecture Note

r u + x y
Ko
+ B ʃ C
- +

Kf

Figure 5.3: Steady State Reference Tracking

5.5 Optimal State Estimation

The state feedback controller gain matrix design problem involves finding a gain matrix
that can stabilize a given linear system by controlling its states. On the other hand,
the state estimator gain matrix design problem involves finding a gain matrix that can
estimate the states of a given linear system based on its input and output signals. Inter-
estingly, these two problems obey a duality property. This means that the solution to one
problem can be used to solve the other problem. In this context, the problem of finding
the state feedback controller gain matrix using a linear quadratic regulator (LQR) can be
used to design an optimal state estimator gain matrix for linear time-invariant systems.

The LQR approach is a widely used method for designing state feedback controllers. It
involves minimizing a quadratic cost function that captures the system’s performance and
control effort. The optimal state feedback controller gain matrix is obtained by solving
the associated algebraic Riccati equation. Once the controller gain matrix is obtained, it
can be used to stabilize the system and achieve the desired performance.

On the other hand, the problem of designing an optimal state estimator gain matrix in-
volves minimizing a similar quadratic cost function that captures the difference between
the actual state and the estimated state. The optimal estimator gain matrix is obtained
by solving another algebraic Riccati equation. Once the estimator gain matrix is ob-
tained, it can be used to estimate the system’s state accurately.

The duality property between these two problems arises because the same cost function
is used in both cases, but with different weighting matrices. In other words, the solution
to one problem can be used to solve the other problem by simply swapping the weighting
matrices.

Consider a linear time-invariant system described by the following state-space model:

ẋ = Ax + Bu
y = Cx + Du (5.5.1)

where x is the state vector, u is the control input, y is the output, and A, B, C, and D
are matrices that describe the system’s dynamics.

Chapter 5 153
Daniel Abebe Modern Control Systems Lecture Note

To design an optimal state feedback controller gain matrix using LQR, we first define
the cost function as follows:
Z tf
J= ((x̂ − x)⊤ Q(x̂ − x) + u⊤ Ru) dt (5.5.2)
t0

where Q and R are weighting matrices that capture the relative importance of the state
and control effort, respectively. The optimal controller gain matrix Kf is obtained by
solving the following algebraic Riccati equation:

Q + PA + A⊤ P − PBR−1 B⊤ P = 0 (5.5.3)

where P is any symmetric positive definite matrix.

Once Kf = R−1 B⊤ P is obtained, it can be used to stabilize the system and achieve the
desired performance.

To design an optimal state estimator gain matrix using LQR, we define a similar cost
function as follows:
Z tf
J= ((x̂ − x)⊤ Q(x̂ − x) + (y⊤ Ry y)) dt (5.5.4)
t0

where Q and Ry are weighting matrices that capture the relative importance of the
output and state estimation error, respectively. The optimal estimator gain matrix Lc is
obtained by solving the following algebraic Riccati equation:

Q + PA + A⊤ P − PC⊤ R−1
y CP = 0 (5.5.5)

where P is any symmetric positive definite matrix.

Once Lc = PC⊤ R−1y is obtained, it can be used to estimate the system’s state accurately.

The duality property between the two problems arises from the fact that they share
the same mathematical structure. Specifically, the state feedback controller gain matrix
design problem can be formulated as a quadratic optimization problem using the linear
quadratic regulator (LQR) technique. Similarly, the state estimator gain matrix design
problem can be formulated as a quadratic optimization problem using the linear quadratic
estimator (LQE) technique.

The LQR and LQE techniques are based on minimizing a cost function that measures
the deviation of the system’s states from a desired trajectory and the deviation of the
estimated states from the actual states, respectively. The cost function is formulated as
a quadratic function of the system’s states and inputs, and its coefficients are determined
by the system’s dynamics and performance specifications.

The duality property between the two problems implies that the optimal gain matrices
obtained from the LQR and LQE techniques are related to each other. Specifically, the
optimal state feedback controller gain matrix is equal to the transpose of the optimal
state estimator gain matrix, up to a scaling factor. This means that if we solve the state
feedback controller gain matrix design problem using LQR, we can obtain the optimal
state estimator gain matrix by transposing the solution and scaling it appropriately.

Chapter 5 154
Daniel Abebe Modern Control Systems Lecture Note

5.6 Examples
1. Determine the increment of the following functional

Z tf
J= (2x2 (t) + 1)dt (5.6.1)
t0

Solution

∆J = J(x(t) + δx(t)) − J(x(t))

Z tf Z tf
2
= (2(x(t) + δx(t)) + 1) dt − (2x2 (t) + 1) dt
t t0
Z 0tf
= (2x2 (t) + 4x(t)δx(t) + 2(δx(t))2 + 1 − 2x2 (t) − 1)dt
t
Z 0tf
= (4x(t)δx(t) + 2(δx(t))2 )dt
t0

2. Find the first variation of the following function

Z tf
J= (2x2 (t) + 1) dt
t0

Solution

∂J
δJ = δx(t)
Z∂xtf
∂
= (4x(t)δx(t) + 2(δx(t))2 )δx(t) dt
t ∂x
Z 0tf
= (4δx(t) + 4δx(t))δx(t) dt
t0
Z tf
= 8(δx(t))2 dt
t0

3. Find x that minimizes the following cost function

Z π
2
J= (x2 − ẋ2 )dt (5.6.2)
0

π
with boundary condition x(0) = 0, x( ) = 1
2
Solution:
From the cost function we know that φ(x, ẋ) = x2 − ẋ2 , from Euler Lagrangian
equations
 
∂φ d ∂φ
− =0
∂x dt ∂ ẋ

Chapter 5 155
Daniel Abebe Modern Control Systems Lecture Note

which results in a homogeneous second-order system

ẍ + x = 0

The characteristic equation of this system is

λ2 + λ = 0 ⇒ λ1 = −i, λ2 = i

The solution to this system has a form

x = C1 eλ1 t + C2 eλ2 t
= C1 e−it + C2 eit
= (C1 + C2 ) cos t + i(C2 − C1 ) sin t

Using boundary condition to solve for the constants C1 and C2 , at t = 0, x = 0

(C1 + C2 ) cos 0 + i(C2 − C1 ) sin 0 = 0

(C1 + C2 ) = 0
π
at t = ,x = 1
2
π π
(C1 + C2 ) cos + i(C2 − C1 ) sin = 1
2 2
(C2 − C1 ) = 1

Solving the two equations simultaneously results in C1 = 0.5 and C2 = −0.5. Hence,
the value of x that minimizes the cost function is

x(t) = sin(t)

4. Find u that minimizes the following cost function

Z 1
1 2
J= u dt (5.6.3)
0 2

subjected to
   
0 1 0
ẋ =  x +  u
0 0 1

with initial and final states x(0) = [1 1]⊤ and x(1) = [0 0]⊤ .

Chapter 5 156
Daniel Abebe Modern Control Systems Lecture Note

Solution:
1
From constrained optimization we know that φ(x, ẋ, u) = u2 + λ⊤ (Ax + Bu − ẋ).
2
The control law we determine should minimize
Z 1 
1 2
J= u + λ1 (x2 − ẋ1 ) + λ2 (u − ẋ2 ) dt
0 2
Using the Euler-Lagrange equations
 
∂φ d ∂φ
− = 0 ⇒ 0 + λ̇1 = 0 ⇒ λ̇1 = 0
∂x1 dt ∂ ẋ1

 
∂φ d ∂φ
− = 0 ⇒ x2 − ẋ1 = 0 ⇒ ẋ1 = x2
∂λ1 dt ∂ λ̇1
result in the state equation.
 
∂φ d ∂φ
− = 0 ⇒ λ1 + λ̇2 = 0 ⇒ λ̇2 = −λ1
∂x2 dt ∂ ẋ2

 
∂φ d ∂φ
− = 0 ⇒ u − ẋ2 = 0 ⇒ ẋ2 = u
∂λ2 dt ∂ λ̇2
result in the state equation.
 
∂φ d ∂φ
− = 0 ⇒ u − λ2 = 0 ⇒ λ2 = −u
∂u dt ∂ u̇

Solving these equations

λ̇1 = 0 ⇒ λ1 = C1 ,
λ̇2 = −λ1 ⇒ λ2 = −C1 t + C2
u(t) = −λ2 ⇒ u(t) = C1 t + C2

from state equations

C1 2
ẋ2 = u ⇒ ẋ2 = C1 t − C2 ⇒ x2 (t) = t − C2 t + C3
2
C1 2 C1 3 C2 2
ẋ1 = x2 ⇒ ẋ1 = t − C2 t + C3 ⇒ x1 (t) = t − t + C3 t + C4
2 6 2
Using the boundary conditions to solve for the constants C1 , C2 , C3 , C4 , we have
the following equations

x2 (0) ⇒ C3 = 1, x1 (0) ⇒ C1 − 2C2 + 2 = 0

x2 (0) ⇒ C3 = 1, x1 (0) ⇒ C1 − 3C2 + 12 = 0

solving this equations C1 = 18 and C2 = 10. Hence, the control law that optimizes
the constrained cost function is

u(t) = 18t − 10

Chapter 5 157
Daniel Abebe Modern Control Systems Lecture Note

5. For the following system, determine an optimal full state feedback gain matrix Kf .
   
−1 0 0
ẋ =  x +  u
0 −2 1
h i
y= 1 0 x

Solution
From Riccati equation
Q + PA + A⊤ P − PBR−1 B⊤ P = 0

Let
   
p11 p12 1 0
P=  Q = C⊤
1 C1 =
  R = 0.5;
p21 p22 0 0

       
1 0 p p −1 0 −1 0 p p
  +  11 12   +   11 12 
0 0 p21 p22 0 −2 0 −2 p21 p22
    
p11 p12 0 h i p p
−    2 0 1  11 12  = 0
p21 p22 1 p21 p22
       
1 0 −p −2p12 −p11 −p12 p p p p
  +  11 +  − 2  12 21 12 22  = 0
0 0 −p21 −2p22 −2p21 −2p22 p22 p21 p22 p22
   
1 − 2p11 − 2p12 p21 −3p12 − 2p12 p22 0 0
 = 
2
−3p12 − 2p12 p22 −4p22 − 2p22 0 0

Since P must be symmetric p12 = p21 , which imply

   
2
1 − 2p11 − 2p12 −3p12 − 2p12 p22 0 0
 = 
−3p12 − 2p12 p22 −4p22 − 2p222 0 0

Solving for P
− 4p22 − 2p222 = 0 ⇒ p22 = 0 or p22 = −2
− 3p12 − 2p12 p22 = 0
1 − 2p11 − 2p212 = 0
Solving for p12 and p11
(a) Case I p22 = 0
− 3p12 − 2p12 p22 = 0 ⇒ p12 (3 + 2p22 ) = 0 ⇒ p12 = 0
1 − 2p11 − 2p212 = 0 ⇒ 1 − 2p11 = 0 ⇒ p11 = 0.5

Chapter 5 158
Daniel Abebe Modern Control Systems Lecture Note

(b) Case II p22 = −2

− 3p12 − 2p12 p22 = 0 ⇒ p12 (3 + 2p22 ) = 0 ⇒ p12 = 0

1 − 2p11 − 2p212 = 0 ⇒ 1 − 2p11 = 0 ⇒ p11 = 0.5

So we have
   
0.5 0 0.5 0
P1 =   P2 =  
0 0 0 −0.8

Since P has to be positive definite (all eigenvalues of P must be greater than or

equal to zero), the solution for P is P1 . The value of feedback gain matrix Kf is
h i
Kf = R−1 B⊤ P = 0 0

what does Kf = 0 imply?

The control law becomes u = −Kf x = 0, with this control law the closed loop
system becomes

ẋ = Ax + Bu ⇒ ẋ = (A − BKf )x

Eigenvalues of closed loop system ẋ = (A − BKf ) x is

∥λI − (A − BKf )∥ = 0
   
λ 0 −1 0
 −  =0
0 λ 0 −2
 
λ+1 0
  =0
0 λ+2
(λ + 1)(λ + 2) = 0 ⇒ λ1 = −1 λ2 = −2

If the error x̃1 = 0.01 and x̃2 = 0.1 is tolerable for the system, then Q will be
 
1000 0
Q= 
0 100

and R = 0.5 with these parameters the value of P and Kf is

 
500 0 h i
P=  Kf = 0 12.2829
0 6.1414

u = −Kf x ⇒ u = −12.2829x2 , with this control law the eigenvalues of closed loop

Chapter 5 159
Daniel Abebe Modern Control Systems Lecture Note

system ẋ = (A − BKf ) x is
∥λI − (A − BKf )∥ = 0
     
λ 0 −1 0 0 0
 − +  =0
0 λ 0 −2 0 12.2829
 
λ+1 0
  =0
0 λ + 14.2829
(λ + 1)(λ + 14.2829) = 0 ⇒ λ1 = −1 λ2 = −14.2829
If  
2 0
Q= 
0 2
and R = 20 the value of P and Kf is
 
500 0 h i
P=  K= 0 1
0 20
u = −Kf x ⇒ u = −x2 , with this control law the eigenvalues of closed loop system
ẋ = (A − BKf ) x is
∥λI − (A − BKf )∥ = 0
     
λ 0 −1 0 0 0
 − +  =0
0 λ 0 −2 0 1
 
λ+1 0
  =0
0 λ+3
(λ + 1)(λ + 3) = 0 ⇒ λ1 = −1 λ2 = −3

6. For the following system, determine an optimal full state feedback gain matrix Kf .
   
1 1 0
ẋ =  x +  u
0 2 1
h i
y= 1 0 x

Solution
From Riccati equation
Q + PA + A⊤ P − PBR−1 B⊤ P = 0

Let
   
p11 p12 1 0
P=  Q = C⊤
1 C1 =
  R = 1;
p21 p22 0 0

Chapter 5 160
Daniel Abebe Modern Control Systems Lecture Note
       
1 0 p p 1 1 1 0 p p
  +  11 12   +   11 12 
0 0 p21 p22 0 2 1 2 p21 p22
    
p11 p12 0 h i p p
−    0 1  11 12  = 0
p21 p22 1 p21 p22
       
1 0 p p + 2p12 p11 p12 p p p p
  +  11 11 +  −  12 21 12 22  = 0
0 0 p21 p21 + 2p22 p11 + 2p21 p21 + 2p22 p22 p21 p22 p22
   
1 + 2p11 − p12 p21 p11 + 3p12 − p12 p22 0 0
 = 
2
p11 + 3p12 − p12 p22 2(p21 + 2p22 ) − p22 0 0
Solving for P results in two values
   
−0.3702 −0.5095 19.4317 6.3138
P1 =   P2 =  
−0.5095 3.7265 6.3138 6.0777
   
1.4259 −1.9626 −0.4875 0.1584
P3 =   P4 =  
−1.9626 2.2735 0.1584 −0.0777
Since P has to be positive definite (all eigenvalues of P must be greater than or
equal to zero). Checking for the positive definiteness
   
−0.4326 3.5653
eig(P1 ) =   eig(P2 ) =  
3.7890 21.9442
   
−0.1582 −0.5415
eig(P3 ) =   eig(P4 ) =  
3.8575 −0.0236
 
19.4317 6.3138
Hence P = P2 =   with this value of P, the value of feedback
6.3138 6.0777
gain matrix Kf is
h i
Kf = R−1 B⊤ P = 6.3138 6.0777
The control law becomes u = −Kf x = −6.3138x1 − 6.0777x2 , with this control law
the eigenvalues of closed loop system ẋ = (A − BKf ) x is
∥λI − (A − BKf )∥ = 0
     
λ 0 1 1 0 0
 − +  =0
0 λ 0 2 6.3138 6.0777
 
λ−1 −1
  =0
6.3138 λ + 4.0777
(λ − 1)(λ + 4.0777) + 6.3138 = 0 ⇒ λ1 = −2.9990 λ2 = −0.0787

Chapter 5 161
Daniel Abebe Modern Control Systems Lecture Note

7. For the following system, determine an optimal full state feedback gain matrix K.
   
1 0 0 1
   
ẋ = 1 −1 1  x + 1 u
   
   
0 0 −2 0
h i
y= 1 0 0 x

Solution
As shown in chapter two example 3, the system is not completely controllable. So,
we have to decompose the system into controllable and uncontrollable modes as
shown in chapter two example 3. So we can design a state feedback controller for
controllable mode (A11 , B1 ). From Riccati equation

Q + PA + A⊤ P − PBR−1 B⊤ P = 0

Let
   
p11 p12 1 1
P=  Q = C⊤
1 C1 =
  R = 2;
p21 p22 1 1

       
1 1 p p 2 −4 2 1 p p
  +  11 12   +   11 12 
1 1 p21 p22 1 0 −4 0 p21 p22
    
p11 p12 1 h i p p
−    2 1 0  11 12  = 0
p21 p22 0 p21 p22
p211
 
(p11 p12 )
− + 4p11 + 2p12 + 1 2p12 − 4p11 + p22 − + 1
 2 2 =0
 (p11 p12 ) p212
2p12 − 4p11 + p22 − +1 − − 8p12 + 1
2 2

Chapter 5 162
Daniel Abebe Modern Control Systems Lecture Note

using MATLAB command

syms p11 p12 p22

A = [2 − 4; 1 0]; B = [1; 0]; C = [1 1]; Q = C′ ∗ C; R = 0.5; p21 = p12;
P = [p11 p12; p21 p22];
eqn = Q + P ∗ A + (A′ ) ∗ P − P ∗ B ∗ (1/R) ∗ (B′ ) ∗ P
eqn1 = eqn(1, 1); eqn2 = eqn(2, 1); eqn3 = eqn(2, 2);
vars = [p11, p12, p22]
[p11, p12, p22] = solve(eqn1 == 0, eqn2 == 0, eqn3 == 0, vars);
P 1 = double([p11(1, 1)p12(1, 1); p12(1, 1)p22(1, 1)])
P 2 = double([p11(2, 1)p12(2, 1); p12(2, 1)p22(2, 1)])
P 3 = double([p11(3, 1)p12(3, 1); p12(3, 1)p22(3, 1)])
P 4 = double([p11(4, 1)p12(4, 1); p12(4, 1)p22(4, 1)])
p1 = eig(P 1) p2 = eig(P 2) p3 = eig(P 3) p4 = eig(P 4)
K = (1/R) ∗ (B′ ) ∗ P2

P has four the following possible values,

   
4.0000 − 6.8188i −16.1240 + 0.0000i 8.3007 0.1240
P1 =   P2 =  
−16.1240 + 0.0000i 15.0000 + 27.6981i 0.1240 32.4696
   
4.0000 + 6.8188i −16.1240 + 0.0000i −0.3007 0.1240
P3 =   P4 =  
−16.1240 + 0.0000i 15.0000 − 27.6981i 0.1240 −2.4696

Since P has to be a positive definite matrix, the valid answer is (check using all
eigenvalues of P (eig(P)))
 
8.3007 0.1240
P2 =  
0.1240 32.4696

The corresponding full state feedback gain matrix Kf = R−1 B⊤ P is

h i
Kf = 4.1504 0.0620

We can also use MATLAB command Kf = lqr(A, B, C, D) to determine the opti-

mal feedback gain matrix, which gives the same result.
If we change the input weighting matrix R = 0.5, the value of P and Kf becomes
 
2.2733 0.1213 h i
P=  Kf = 4.5466 0.2426
0.1213 8.4022

If we change the state weighting matrix

 
10 0
Q= 
0 10

Chapter 5 163
Daniel Abebe Modern Control Systems Lecture Note

with R = 0.5 the value of P and Kf becomes

 
3.6458 1.0000 h i
P=  K = 7.2915 2.0000
1.0000 19.8745

u = −Kf x ⇒ u = −7.2915x1 − 2x2 , with this control law the eigenvalues of closed
loop system ẋ = (A − BKf ) x is

∥λI − (A − BKf )∥ = 0
     
λ 0 2 −4 7.2915 2
 − +  =0
0 λ 1 0 0 0
 
λ + 5.2915 6
  =0
−1 λ
λ2 + 5.2915λ + 6 = 0 ⇒ λ1 = −3.6457 λ2 = −1.6458

5.7 Exercises
1. A system described by

ẋ = −2x + u

and control u is required to minimize a cost function

Z 1
J= u2 dt
0

find such u for x(0) = 1 and x(1) = 0.

2. For a system described by

ẋ = −2x + 2u

find a control law u that minimize the performance index

Z 1
J= (3x2 + u2 ) dt
0

with x(0) = 1 and x(1) = 0.

3. Find a control law that minimizes
Z 2
1 1 1
J = (x1 (2) − 5)2 + (x2 (2) − 2)2 + u2 dt
2 2 2 0

for a dynamic system

ẋ1 = x2
ẋ2 = −x2 + u

with x(0) = 0 and x(2) is on a line x1 + 5x2 = 15.

Chapter 5 164
Daniel Abebe Modern Control Systems Lecture Note

4. Find a feedback control law that minimizes

Z ∞
1
J= (x22 + u2 ) dt
0 10
subjected to

ẋ1 = −x1 + u
ẋ2 = x1

5. Use Riccati equation to determine the feedback control law for a system

ẋ = −x + u

which minimize
Z 1
1
J= (3x2 + u2 ) dt
2 0

6. Find an optimal feedback gain Kf for the following systems with the given initial
conditions

(a)
    
0 1 0 1
ẋ =   x +   u, x0 =  
−4 0 1 1

that minimize
Z ∞
1
J= (x⊤ Qx + u⊤ Ru) dt
2 0

(b)
     
0 1 0 1
ẋ =   x +   u, x0 =  
2 −1 1 1

that minimize
Z ∞
1
J= (x⊤ Qx + Ru2 ) dt
2 0

(c)
     
−1 0 1 0
ẋ =   x +   u, x0 =  
0 −4 1 1

that minimize
Z ∞
1
J= (x⊤ Qx + u⊤ Ru) dt
2 0

Chapter 5 165
Daniel Abebe Modern Control Systems Lecture Note

(d)
     
0 1 0 1
ẋ =   x +   u, x0 =  
−6 −8 1 1

that minimize
Z ∞
1
J= (x⊤ Qx + u⊤ Ru) dt
2 0

(e)
    
0 1 0 1
ẋ =   x +   u, x0 =  
−6 0 1 1

that minimize
Z ∞
1
J= (x⊤ Qx + u⊤ Ru) dt
2 0

7. Design an optimal state feedback matrix Kf that minimize the cost function

1 ∞ ⊤
Z
J= (x Qx + u⊤ Ru) dt
2 0
for

(a) Inverted pendulum dynamics (refer 1.6.3)

(b) DC motor dynamics refer (1.7.1)
(c) Two degree of freedom car suspension system (1.6.3)

using MATLAB.

Chapter 5 166
Chapter 6

Introduction to Nonlinear Systems

and MIMO Systems

6.1 Introduction to Nonlinear Systems

A nonlinear system is a system having an output change that is not proportional to the
input. Every nonlinear time-invariant system is represented as
ẋ = f (x, u)
y = h(x, u) (6.1.1)
where h(x, u) and f (x, u) are nonlinear functions of state and inputs. The non-linearities
may be hard non-linearity and/or soft non-linearity. Hard non-linearity includes back-
lash, Coulomb’s friction, etc., while soft nonlinearities are non-linearities due to state
and/or input.

6.1.1 Determining equilibrium points for nonlinear dynamic systems

Nonlinear systems may have a point or set of points or in state space where the dynamics
of the system stay there forever after it arrives there. These points are called equilibrium
points or limit cycles. These points may be single, multiple, or may not exist for nonlinear
systems. At the equilibrium point or limit cycles, ẋ = 0 always.

Equilibrium points for IPC dynamics

For IPC dynamics, we derived from 1.6.3, the equilibrium points are
ẋ1 = x2 (6.1.2)
(mp l2 + I)mp lx24 sin x3 − (mp l)2 g sin x3 cos x3 − b(mp l2 + J)x2
ẋ2 = (6.1.3)
(mp + mc )(mp l2 + I) − (mp l cos x3 )2
ẋ3 = x4 (6.1.4)
−(mp l)2 x24 g sin x3 cos x3 + bx2 mp l cos x3 + (mp + mc )mp lg sin x3
ẋ4 = (6.1.5)
(mp + mc )(mp l2 + I) − (mp l cos x3 )2
The equilibrium points are determined as
ẋ1 = x2 = 0

167
Daniel Abebe Modern Control Systems Lecture Note

(mp l2 + I)mp lx24 sin x3 − (mp l)2 g sin x3 cos x3

2 2
= 0 ⇒ (mp l)2 g sin x3 cos x3 = 0
(mp + mc )(mp l + I) − (mp l cos x3 )

implying that sin x3 = 0 or cos x3 = 0,

ẋ3 = x4 = 0

−(mp l)2 g sin x3 cos x3 + (mp + mc )mp lg sin x3

= 0 ⇒ (mp + mc )mp lg sin x3 = 0
(mp + mc )(mp l2 + I) − (mp l cos x3 )2
sin x3 = 0
x3 = nπ

Hence, the equilibrium point for the IPC dynamic system is (x1 , 0, nπ, 0). We can’t draw
any conclusion about x1 since it vanishes during the equilibrium point calculation.

Equilibrium points for quadruple tank process dynamics

Since ẋ = 0 at equilibrium points for a quadruple tank process dynamics in equation
1.9.34, the equilibrium points are,
p p
γ1 k1 V1e + a3 2gh3e = a1 2gh1e
p p
γ2 k2 V2e + a4 2gh4e = a2 2gh2e
p
p a3 2gh3e
(1 − γ2 )k2 V2e − a3 2gh3e = 0 =⇒ V2e =
(1 − γ2 )k2
p
p a4 2gh4e
(1 − γ1 )k1 V1e − a4 2gh4e = 0 =⇒ V1e =
(1 − γ1 )k1

Since the tanks are coupled, the equilibrium point of each tank cannot be set individually,
for instance, to set the tank 4 to a height h4e , a voltage V1e is required which in turn
affect the level of tank 1 and tank 2.
 
p 1 γ1 k1 V1e + (1 − γ2 )k2 V2e
γ1 k1 V1e + (1 − γ2 )k2 V2e = a1 2gh1e =⇒ h1e =
2g a1
 
p 1 γ2 k2 V2e + (1 − γ1 )k1 V1e
γ2 k2 V2e + (1 − γ1 )k1 V1e = a2 2gh2e =⇒ h2e =
2g a2

6.1.2 Linearization of nonlinear systems

A linearization of a nonlinear system is an approximation of the system about some
desired point (operating point). Consider a system

ẋ = f (x) (6.1.6)

with operating point xe .

Assume a small deviation ∆x from operating point xe is given as,

x = xe + ∆x (6.1.7)

Chapter 6 168
Daniel Abebe Modern Control Systems Lecture Note

The Taylor series expansion of the function about this point xe is given as
′ 1 ′′
ẋe + ∆ẋ = f (x + ∆x) = f (xe ) + f (xe )∆x + f xe (∆x)2 + HOT (6.1.8)
2!
Neglecting higher-order terms, the function can be approximated by a linear function as
′
ẋe + ∆ẋ = f (xe ) + f (xe )∆x (6.1.9)

since xe satisfies a differential equation ẋe = f (xe )

′
f (xe ) + ∆ẋ =f (xe ) + f (xe )∆x
′
∆ẋ = f (xe )∆x (6.1.10)

Let the deviation ∆x be x then we have

′
ẋ = f (xe )x (6.1.11)

Similarly for multi-variable function f (x, u) the approximation of the function about
operating point xe and ue given as
∂f ∂f
f (x, u) = f (xe , ue ) + (x − xe ) + (u − ue ) (6.1.12)
∂x ∂u
If the operating point is the equilibrium point ẋ = f (xe , ue ) = 0, which means

∂f ∂f
ẋ = f (x, u) = x+ u (6.1.13)
∂x ∂u
∂f ∂f
Which implies that A = and B = evaluated at the equilibrium point (xe , ue ).
∂x ∂u
We have a nonlinear system with n number of states dynamics ẋi = fi (xi , uj ) i =
1, 2, . . . n, j = 1, 2, 3, . . . , m
   
ẋ f (x , x , . . . , xn , u1 , u2 , . . . , um )
 1  1 1 2 
   
 ẋ2  f2 (x1 , x2 , . . . , xn , u1 , u2 , . . . , um )
 = (6.1.14)
 ..   ..


.  . 
   
ẋn f3 (x1 , x2 , . . . , xn , u1 , u2 , . . . , um )

Then the approximation of the nonlinear system about an equilibrium point is

∂f1 ∂f1 ∂f1 ∂f1 ∂f1 ∂f1
 
 
ẋ x 1 x 2 . . . xn u1 u2 . . . um
 1   ∂x1 ∂x2 ∂xn ∂u1 ∂u2 ∂um 


   
   
   ∂f ∂f2 ∂f2 ∂f2 ∂f2 ∂f2 
   2
 ẋ2  =  x1 x2 . . . xn u1 u2 . . . um  (6.1.15)

   ∂x1 ∂x2 ∂xn ∂u1 ∂u2 ∂um 
 ..   . .. .. .. .. ..
 .   ..

   . . . . . 

 ∂fn ∂fn ∂fn ∂fn ∂fn ∂fn
ẋn

x1 x2 . . . xn u1 u2 . . . um
∂x1 ∂x2 ∂xn ∂u1 ∂u2 ∂um |(xeq ,ueq )

Chapter 6 169
Daniel Abebe Modern Control Systems Lecture Note

∂f1 ∂f1 ∂f1 ∂f1 ∂f1 ∂f1

   
     
ẋ ... x ... u
 1   ∂x1 ∂x2 ∂xn   1   ∂u1 ∂u2 ∂um   1
 
 
         
         
   ∂f ∂f2 ∂f2     ∂f ∂f2 ∂f2   
   2    2
 ẋ2  =  ...  x2  +  ...
   
 u2 
   ∂x1 ∂x2 ∂xn 

   ∂u1 ∂u2 ∂um 

 ..   .  ..   .  .. 
 
 .   .. .. ..   .   .. .. .. 
 . 
   . .      . . 
  
 ∂fn ∂fn ∂fn   ∂fn ∂fn ∂fn 
ẋn ... xn ... um
∂x1 ∂x2 ∂xn |(xe ,ue ) ∂u1 ∂u2 ∂um |(xe ,ue )
(6.1.16)

Comparing with ẋ = Ax + Bu,

∂f1 ∂f1 ∂f1
 
 ∂x1 ...
 ∂x2 ∂xn 
 
 
 ∂f ∂f2 ∂f2 
 2
A= ... (6.1.17)

 ∂x1 ∂x2 ∂xn 

 . .. .. 
 .. . . 
 
 ∂fn ∂fn ∂fn 
...
∂x1 ∂x2 ∂xn |(xeq ,ueq )

While
∂f1 ∂f1 ∂f1
 
 ∂u1 ...
 ∂u2 ∂ 

 
 
 ∂f ∂f2 ∂f2 
 2
B= ... (6.1.18)

 ∂u1 ∂u2 ∂um 

 . .. ..
 ..

 . . 

 ∂fn ∂fn ∂fn 
...
∂u1 ∂ ∂ |(xeq ,ueq )

In similar way the nonlinear system has p outputs yi = hi (x, u),i = 1, 2, . . . p the approx-
imation about equilibrium point will be
∂h1 ∂h1 ∂h1 ∂h1 ∂h1 ∂h1
 
 
y x1 x2 . . . xn u1 u2 . . . um
 1   ∂x1 ∂x2 ∂xn ∂h1 ∂u2 ∂um 


   
   
   ∂h ∂h2 ∂h2 ∂h2 ∂h2 ∂h2 
   2
y2  =  x1 x2 . . . xn u1 u2 . . . um  (6.1.19)

   ∂x1 ∂x2 ∂xn ∂u1 ∂u2 ∂um 
 ..   . .. .. .. .. ..
 .   ..

   . . . . . 

 ∂hp ∂h ∂h ∂h ∂h ∂h
yp p p p p p

x1 x2 . . . xn u1 u2 . . . um
∂x1 ∂x2 ∂xn ∂u1 ∂u2 ∂um |(xeq ,ueq )

Chapter 6 170
Daniel Abebe Modern Control Systems Lecture Note

∂h1 ∂h1 ∂h1 ∂h1 ∂h1 ∂h1

   
     
y ... x ... u
 1   ∂x1 ∂x2 ∂xn   1   ∂u1 ∂u2 ∂um   1
 
 
         
         
   ∂h ∂h2 ∂h2     ∂h ∂h2 ∂h2   
   2    2
 y2  =  ...  x2  +  ...
   
 u2 
   ∂x1 ∂x2 ∂xn 

   ∂u1 ∂u2 ∂um 

 ..   .  ..   .  .. 
 
 .   .. .. ..   .   .. .. .. 
 . 
   . .      . . 
  
 ∂hp ∂hp ∂hp   ∂hp ∂hp ∂hp 
yp ... xn ... um
∂x1 ∂x2 ∂xn |(xeq ,ueq ) ∂u1 ∂u2 ∂um |(xeq ,ueq )
(6.1.20)
Comparing with y = Cx + Du,
∂h1 ∂h1 ∂h1
 
 ∂x1 ...
 ∂x2 ∂xn 
 
 
 ∂h ∂h2 ∂h2 
 2
C= ... (6.1.21)

 ∂x1 ∂x2 ∂xn 

 . .. .. 
 .. . . 
 
 ∂hp ∂hp ∂hp 
...
∂x1 ∂x2 ∂xn |(xeq ,ueq )
While
∂f1 ∂f1 ∂f1
 
 ∂u1 ...
 ∂u2 ∂ 

 
 
 ∂f ∂f2 ∂f2 
 2
D= ... (6.1.22)

 ∂u1 ∂u2 ∂um 

 . .. ..
 ..

 . . 

 ∂fn ∂fn ∂fn 
...
∂u1 ∂ ∂ |(xeq ,ueq )

Linearizing Inverted Pendulum on a Cart Dynamics

For IPC dynamics derived in Chapter One
   
ẋ1 f1 (x, u)
   
   
ẋ2  f2 (x, u)
 = 
   
ẋ3  f3 (x, u)
   
ẋ4 f4 (x, u)
 
x2
 (mp l2 + I)mp lx24 sin x3 − (mp l)2 g sin x3 cos x3 − b(mp l2 + J)x2 + (mp l2 + I)u 
 
 
 (m + m )(m l 2 + I) − (m l cos x )2 
p c p p 3
=



 x4 
 2 2

 −(mp l) x4 g sin x3 cos x3 + bx2 mp l cos x3 + (mp + mc )mp lg sin x3 − mp l cos x3 u 
(mp + mc )(mp l2 + I) − (mp l cos x3 )2
(6.1.23)

Chapter 6 171
Daniel Abebe Modern Control Systems Lecture Note

To simplify the equation let

η1 + η2 ξ1 + ξ2
f2 = f4 =
η3 η3
where
η1 = (mp l2 + I)mp lx24 sin x3 − (mp l)2 g sin x3 cos x3 − b(mp l2 + J)x2
η2 = (mp l2 + I)u
ξ1 = −(mp l)2 x24 g sin x3 cos x3 + bx2 mp l cos x3 + (mp + mc )mp lg sin x3
ξ2 = −mp l cos x3 u
η3 = (mp + mc )(mp l2 + I) − (mp l cos x3 )2
The linearization of the functions fi (x, u), i = 1, 2, 3, 4 about the equilibrium point
(xeq , ueq ) = (x1 , 0, 0, 0, 0) is determined using equation 6.1.15.
∂f1 ∂x2 ∂f1 ∂x2 ∂f1 ∂x2 ∂f1 ∂x2
= = 0, = = 1, = = 0, = =0 (6.1.24)
∂x1 ∂x1 ∂x2 ∂x2 ∂x3 ∂x3 ∂x4 ∂x4
 
∂f2 ∂ η1 − η2
= =0 (6.1.25)
∂x1 ∂x1 η3
−b(mp l2 + J)
 
∂f2 ∂ η1 − η2
= = (6.1.26)
∂x2 ∂x2 η3 |(xe ,ue ) (mp + mc )(mp l2 + I) − (mp l)2
−(mp l)2 g
 
∂f2 ∂ η1 − η2
= = (6.1.27)
∂x3 ∂x3 η3 |(xe ,ue ) (mp + mc )(mp l2 + I) − (mp l)2
 
∂f2 ∂ η1 − η2
= =0 (6.1.28)
∂x4 ∂x3 η4

∂f3 ∂x3 ∂f3 ∂x3 ∂f3 ∂x3 ∂f3 ∂x3

= = 0, = = 0, = = 1, = =0 (6.1.29)
∂x1 ∂x1 ∂x2 ∂x2 ∂x3 ∂x3 ∂x4 ∂x4
 
∂f4 ∂ ξ1 − η2
= =0 (6.1.30)
∂x1 ∂x1 η3
 
∂f4 ∂ ξ1 − ξ2 bmp l
= = (6.1.31)
∂x2 ∂x2 η3 |(xe ,ue ) (mp + mc )(mp l2 + I) − (mp l)2
 
∂f4 ∂ ξ1 − ξ2 mp lg(mc + mp )
= = (6.1.32)
∂x3 ∂x3 η3 |(xe ,ue ) (mp + mc )(mp l2 + I) − (mp l)2
 
∂f4 ∂ ξ1 − ξ2
= =0 (6.1.33)
∂x4 ∂x3 η3

∂f1 ∂x2 ∂f3 ∂x3

= = 0, = =0 (6.1.34)
∂u ∂u  ∂u ∂u
mp l2 + I

∂f2 ∂ ξ1 − ξ2
= = (6.1.35)
∂u ∂u η3 | (mp + mc )(mp l2 + I) − (mp l)2
  (xe ,ue )
∂f4 ∂ η1 − η2 −mp lg
= = (6.1.36)
∂u ∂u η3 |(x ,u ) (mp + mc )(mp l2 + I) − (mp l)2
e e

Chapter 6 172
Daniel Abebe Modern Control Systems Lecture Note

From this, the system matrix is

 
0 1 0 0
2
(mp l)2 g
     
0 − b(mp l + J)
 
− 0
∂fi  q q 
Aij = = 
∂xj xe ,ue 
0 0 0

1
     
 bmp l mp lg(mp + mc ) 
0 0
q q

and the input matrix is

 
 0 
 mp l2 + J 
 
∂fi  q 
Bij = = 
∂uj xe ,ue

 0


 
 mp l 
−
q

where q = (mp + mc )(mp l2 + I) − (mp l)2 .

with system outputs y1 = h1 (x, u) = x1 and y2 = h2 (x, u) = x3

∂h1 ∂h1 ∂h1 ∂h1

= 1, = 0, = 0, =0 (6.1.37)
∂x1 ∂x2 ∂x3 ∂x4
∂h2 ∂h2 ∂h2 ∂h2
= 0, = 0, = 1, =0 (6.1.38)
∂x1 ∂x2 ∂x3 ∂x4
and
∂h1 ∂h2
= 0, =0 (6.1.39)
∂u ∂u
The output matrix becomes
 
∂hi 1 0 0 0
Cij = = 
∂xj xe ,ue
 
0 0 1 0

and input transmission matrix

 
∂hi 0
Dij = = 
∂uj xe ,ue
 
0

Chapter 6 173
Daniel Abebe Modern Control Systems Lecture Note

Thus the linearized model of IPC about equilibrium point (xe , ue ) = (x1 , 0, 0, 0, 0) is given
as
   
0 1 0 0 0
   
ẋ  x1
 1  2 2 2
      
  0 − b(mp l + J) (mp l) g mp l + J 
   
ẋ2   − 0 x  
    
 = q q 2
  +  q 
u
      
ẋ3  0 0 0 1 x3   0 
            
ẋ4 bmp l mp lg(mp + mc ) mp l
0 x4
   
0 −
q q q
(6.1.40)
 
x
     1  
 
y1 1 0 0 0 x2  0
y= =   +  u
  (6.1.41)
y2 0 0 1 0 x3  0
 
x4

Which is same as equation 1.6.36 and equation 1.6.37

Linearizing Quadruple Tank Process Dynamics

For quadruple tank process on equation 1.9.34 let (xe , ue ) = (h1e , h2e , h3e , h4e , V1e , V2e )
be an equilibrium point. We define small deviations:
   
h̃ h − h1e
 1  1     
h̃2  h2 − h2e  V − V1e
   
Ṽ
x̃ =  =  , ũ =  1  =  1 
h̃3  h3 − h3e  V2 − V2e
   
Ṽ2
   
h̃4 h4 − h4e

The linearized system takes the form:

x̃˙ = Ax̃ + Bũ

The matrix A ∈ R4×4 is obtained by taking partial derivatives of each equation concerning
the states:
∂fi
Aij =
∂xj xe ,ue

p p
∂ ḣ1 a1 g a1 g/2h1e ∂ ḣ1 a3 g a3 g/2h3e
=− ·p =− , = ·p =
∂h1 A1 2gh1e A1 ∂h3 A1 2gh3e A1
p p
∂ ḣ2 a2 g a2 g/2h2e ∂ ḣ2 a4 g a4 g/2h4e
=− ·p =− , = ·p =
∂h2 A2 2gh2e A2 ∂h4 A2 2gh4e A2
p p
∂ ḣ3 a3 g a3 g/2h3e ∂ ḣ4 a4 g a4 g/2h4e
=− ·p =− , =− ·p =−
∂h3 A3 2gh3e A3 ∂h4 A4 2gh4e A4

Chapter 6 174
Daniel Abebe Modern Control Systems Lecture Note

Let
p p p
a1
g/2h1e a2
g/2h2e a3 g/2h3e
α1 = , α2 = , α3 =
A1 A2 A
p p p 1
a4 g/2h4e a3 g/2h3e a4 g/2h4e
α4 = , α5 = , α6 =
A2 A3 A4
With this he system matrix A is then,
 
−α1 0 α3 0
 
 0 −α2
 
0 α4 
A=   (6.1.42)
0 −α5

 0 0 
 
0 0 0 −α6

The matrix B ∈ R4×2 is obtained by taking partial derivatives concerning the inputs:

∂fi
Bij =
∂uj xe ,ue

∂ ḣ1 γ1 k1 ∂ ḣ1 ∂ ḣ2 ∂ ḣ2 γ2 k2

= , = 0, = 0, =
∂V1 A1 ∂V2 ∂V1 ∂V2 A2
∂ ḣ3 ∂ ḣ3 (1 − γ2 )k2 ∂ ḣ4 (1 − γ1 )k1 ∂ ḣ4
= 0, = , = , =0
∂V1 ∂V2 A3 ∂V1 A4 ∂V2
Thus:

 
γ1 k1
0
 A1 
γ2 k2 
 
0

 
B=
 A2 
(6.1.43)
(1 − γ2 )k2 
 0 
A3
 
 
 (1 − γ1 )k1 
0
A4
with system outputs y1 = h1 (x, u) = h1 and y2 = h2 (x, u) = h2 , the output matrix
becomes
 
∂hi 1 0 0 0
Cij = =  (6.1.44)
∂xj xe ,ue  
0 1 0 0

and input transmission matrix

 
0 0
∂hi  
Dij = =  (6.1.45)
∂uj xe ,ue
 
0 0

Chapter 6 175
Daniel Abebe Modern Control Systems Lecture Note

6.2 Introduction to Multiple Input Multiple Output Systems

MIMO systems, as their name indicates they have multiple inputs and multiple outputs.
Consider a system with n inputs and m outputs; the transfer function of these systems
consists of transfer functions from each input to each output.

Yi (s)
= Gij (s) (6.2.1)
Ui (s)

where i1, 2, 3, . . . , n and j = 1, 2, 3, . . . , m.

If we expand equation 6.2.1, we have a set of transfer functions
    
Y (s) G (s) G21 (s) . . . Gn1 (s) U (s)
 1   11  1 
    
 Y2 (s)   G21 (s) G22 (s) . . . Gn2 (s)   U2 (s) 
 ..  =  .. .. .. ..   ..  (6.2.2)
    
 .   . . . .  . 
    
Ym (s) G1m (s) G2m (s) . . . Gnm (s) Un (s)

6.2.1 Poles and zeros of MIMO Systems

The poles of each transfer function are the poles of the MIMO system. If Gij has a pole
at s = p, then s = p is also a pole of the MIMO system. If one of the transfunctions has
poles at frequency r at s = p0 , then the MIMO system also has a pole of frequency r at
s = p0 .

To determine the largest multiplicity pole, determine the largest multiplicity pole among
1×1 minors of the transfer function, then the largest multiplicity pole among 2×2 minors
of the transfer function so on. Stop at minors of size k × k if k is the first size for which
the previous multiplicity of the pole is greater than the current minor’s pole multiplicity.
The largest of all the multiplicities of the minors is the multiplicity of the pole.

Determining zeros of the MIMO system is a tricky problem. To start with, we know
that zeros of the system prohibit the control signal from affecting the dynamics of the
system. Hence, from this knowledge, a MIMO system has zero at location z0 if the z0
causes a reduction in rank of the MIMO system transfer function (i.e., if the z0 causes
zero rows in the transfer function of the MIMO system). Otherwise, the zero of one
transfer function will not be considered as the zero of the MIMO system.

Consider IPC dynamics derived in section 1.6.3, the transfer function of IPC will be

(mp l)2 g (mp l2 + I)

 
Y1 (s) mp l
=− + (6.2.3)
U (s) qs (mp + mc )mp lg − qs sq

Y2 (s) mp l
= (6.2.4)
U (s) (mp + mc )mp lg − qs

Chapter 6 176
Daniel Abebe Modern Control Systems Lecture Note

(mp l)2 g (mp l2 + I)

   
  mp l
Y (s) − qs (mp + mc )mp lg − qs
+
sq
 1  


=  U (s) (6.2.5)
   

   
mp l
Y2 (s)
 
(mp + mc )mp lg − qs

As we can see, the system has only one input; hence, it is a single-input multiple-output
system (SIMO). The poles of the system are the poles of equation 6.2.3 and equation
6.2.4. The zero of the system is zero of equation 6.2.3, which will result in one zero row
in the system transfer function on equation 6.2.5.

(mp l)3 g mp l 2 + I
G11 (s) = − +
qs((mp + mc )mp lg − qs) sq
3 2
−(mp l) g + (mp l + I)((mp + mc )mp lg − qs)
=
sq((mp + mc )mp lg − qs)

The numerator becomes:

N (s) = mp lg[−(mp l)2 + (mp l2 + I)(mp + mc )] − (mp l2 + I)qs = 0

Since mp l ̸= 0, we only consider when N(s) = 0:

mp lg[−(mp l)2 + (mp l2 + I)(mp + mc )]

s= (6.2.6)
q(mp l2 + I)

For the linearized model of the quadruple tank level control system in section 6.1.2, if
height h1 and h2 are required to be controlled, we have two outputs and two [Link]
transfer function matrix G(s) relates inputs to outputs in the Laplace domain:

G(s) = C(sI − A)−1 B + D (6.2.7)

Given the structure of A:
 
s + α1 0 −α3 0
 
−α4 
 
 0 s + α2 0
sI − A = 
 
 (6.2.8)
 0 0 s + α5 0 
 
0 0 0 s + α6

The matrix is block diagonal, so its inverse is:

 
1 α3
 s + α1 0 0
(s + α1 )(s + α5 ) 
1 α4
 
 
 0 0 
−1
(sI − A) = 
 s + α 2 (s + α2 )(s + α6 )  (6.2.9)
1 
 0 0 0
 
s + α5

 
 1 
0 0 0
s + α6

Chapter 6 177
Daniel Abebe Modern Control Systems Lecture Note

 
γ1 k1 α3 (1 − γ2 )k2

 A1 (s + α1 ) A3 (s + α1 )(s + α5 ) 

G(s) = C(sI − A)−1 B =  (6.2.10)
 

 
 α4 (1 − γ1 )k1 γ2 k2 
A4 (s + α2 )(s + α6 ) A2 (s + α2 )

The transmission zeros are the values of s that make the transfer function matrix lose
rank. We find them by solving:

det(G(s)) = 0 =⇒ G11 (s)G22 (s) − G12 (s)G21 (s) = 0

     
γ1 k1 γ2 k2 α3 (1 − γ2 )k2 α4 (1 − γ1 )k1
= − =0
A1 (s + α1 ) A2 (s + α2 ) A3 (s + α1 )(s + α5 ) A4 (s + α2 )(s + α6 )

Factoring out common terms:

γ1 γ2 k1 k2 α3 α4 (1 − γ1 )(1 − γ2 )k1 k2
det G(s) = − (6.2.11)
A1 A2 (s + α1 )(s + α2 ) A3 A4 (s + α1 )(s + α2 )(s + α5 )(s + α6 )

Combine terms with a common denominator:

k1 k2 [γ1 γ2 A3 A4 (s + α5 )(s + α6 ) − α3 α4 (1 − γ1 )(1 − γ2 )]
det G(s) = (6.2.12)
A1 A2 A3 A4 (s + α1 )(s + α2 )(s + α5 )(s + α6 )

The zeros occur when the numerator polynomial vanishes:

γ1 γ2 A3 A4 (s + α5 )(s + α6 ) − α3 α4 (1 − γ1 )(1 − γ2 ) = 0 (6.2.13)

Expanding the quadratic:

 
2 α3 α4 (1 − γ1 )(1 − γ2 )
s + (α5 + α6 )s + α5 α6 − =0 (6.2.14)
γ1 γ2 A3 A4

The zeros are given by the roots:

r  
α3 α4 (1−γ1 )(1−γ2 )
−(α5 + α6 ) ± (α5 + α6 )2 − 4 α5 α6 − γ1 γ2 A3 A4
s= (6.2.15)
2
Consider a quadruple tank process with parameters in Table 6.1:
For desired equilibrium heights h1e = h2e = 10 cm, using
p p
(1 − γ2 )k2 V2e = a3 2gh3e , (1 − γ1 )k1 V1e = a4 2gh4e
p p p p
γ1 k1 V1e = a1 2gh1e − a3 2gh3e , γ2 k2 V2e = a2 2gh2e − a4 2gh4e

to solve for the rest equilibrium points yields:

h3e = 6.45 cm, h4e = 6.45 cm

V1e = 3.15 V, V2e = 3.15 V

Chapter 6 178
Daniel Abebe Modern Control Systems Lecture Note

Table 6.1: Quadruple tank process parameters

Parameter Value Description

A1 , A2 28 cm2 Cross-section of lower tanks
A3 , A4 32 cm2 Cross-section of upper tanks
a1 , a2 0.071 cm2 Outlet holes (lower tanks)
a3 , a4 0.057 cm2 Outlet holes (upper tanks)
k1 , k2 3.33 cm3 /Vs Pump constants
γ1 , γ2 0.7 Valve settings
g 981 cm/s2 Gravity

which yield the α coefficients:

p
0.071 981/(2 × 10)
α1 = = 0.028 s−1 , α2 = α1 = 0.028 s−1
p 28
0.057 981/(2 × 6.45)
α3 = = 0.023 s−1 , α4 = α3 = 0.023 s−1
p 28
0.057 981/(2 × 6.45)
α5 = = 0.020 s−1 , α6 = α5 = 0.020 s−1
32
Thus, the state and input matrices become:
   
−0.028 0 0.023 0 0.083 0
   
−0.028
   
 0 0 0.023   0 0.083
A=  , B=
 
−0.020
 
 0 0 0   0 0.062
   
0 0 0 −0.020 0.058 0

leading to the transfer function of the system,

 0.083 0.0014 
 s + 0.028 (s + 0.028)(s + 0.020) 
 
G(s) = 


 0.0013 0.083 
(s + 0.028)(s + 0.020) s + 0.028

solving for the transmission zeros and poles of the system by setting det G(s),

1.8200 × 10−06
   
0.0069
||G(s)|| = −
s + 0.028 (s + 0.028)(s + 0.020)
2
0.006889(s + 0.04s + 0.0001358)
=
(s + 0.028)2 (s + 0.020)2

Chapter 6 179
Daniel Abebe Modern Control Systems Lecture Note

The transmission zeros at

s2 + 0.04s + 0.0001358 = 0 =⇒ s = −0.020 ± 0.00208j
And the poles of the systems are at,
(s + 0.028)2 (s + 0.020)2 = 0 =⇒ double poles at s = −0.028 and s = −0.02

6.3 Examples
1. Find the poles and zeros of the following system
ẋ1 = x2 + u1
ẋ2 = x3 + u2
ẋ3 = 6x1 − 11x2 + 6x3 + u1 + u2
y1 = x1
y2 = x3

Solution
The state space representation of the system is
 
ẋ     
 1
  0 1 0 x1 1 0  
ẋ2       u
 1
  = 0 0 1 x2  + 0 1  
  
  
ẋ3      u2
  6 −11 6 x3 1 1
ẋ4
 
    x1   
y1 1 0 0   0 0 u
 = x2  + 
    1
y2 0 0 1   0 0 u2
x3
The transfer function of the system is
Cadj(sI − A)B
G(s) =
det(sI − A)
  
  s2 − 6s + 11 s + 6 1 1 0
1 0 0  2
 
 s − 6s s  0 1
 
  6
0 0 1 
2
  
6s s s 1 1
=
 s3 − 6s2+ 11s − 6
s2 − 6s + 12 s + 7
 
s2 + 6s s + s2
=
s3 − 6s2 + 11s − 6
The only number that can cause a zero row in the transfer function matrix is
s = 0. Hence s = 0 is the zero of the system. The poles of the system are zeros of
s3 − 6s2 + 11s − 6 = (s + 1)(s + 2)(s + 3) = 0, hence s = 1, s = 2 and s = 3 are the
poles of the system.

Chapter 6 180
Daniel Abebe Modern Control Systems Lecture Note

2. Find zeros as poles of the following transfer function

 
s s2 + 3
 s + 1 (s + 1)(s + 4) 
 
G(s) = 
 

 
 1 s+3 
s2 + 1 s(s + 5)

Solution
No number causes zero rows or zero columns in the transfer function matrix, hence
the system has no zeros. The poles of the system are
s s2 + 3 1 s+3
• poles of first minors, , , 2 and
s + 1 (s + 1)(s + 4) s + 1 s(s + 5)
• poles of the second minors

s2 + 3
     
s s+3 1
−
s+1 s(s + 5) s2 + 1 (s + 1)(s + 4)

Thus, the poles of the system are s = 0, s = −1, s = −4, s = −5, and s = ±i.

3. Find the equilibrium point/s and linearized model of the system at the equilibrium
point/s.

ẋ1 = 7x1 + 2 sin x2 − x42 + u3

ẋ2 = ex1 − 3x2 − 1 + 5x21 + sin u
y = ex1 +x2

Solution
At equilibrium point/s the the state dynamics is zero ẋ1 = 0, ẋ2 = 0, ueq = 0, thus

f1 (x1 , x2 , u) = 7x1 + 2 sin x2 − x42 + u3 = 0

f2 (x1 , x2 , u) = ex1 − 3x2 − 1 + 5x21 + sin u = 0

using the following MATLAB code

syms x1 x2 u
eqn1 = 7 ∗ x1 + 2 ∗ sin(x2 ) − x42 + u3 ;
eqn2 = exp(x1 ) − 3 ∗ x2 − 1 + 5 ∗ x21 + sin(u)
[x1 , x2 , u] = solve(eqn1 == 0, eqn1 == 0, [x1 , x2 , u])

the only equilibrium point is (x1eq , x2eq , ueq = 0) = (0, 0, 0). The linearized model

Chapter 6 181
Daniel Abebe Modern Control Systems Lecture Note

about (0, 0, 0) is
   ∂ ∂
   
∂

f (x , x , u) f (x , x , u)
ẋ
 1  ∂x1
1 1 2
∂x2
1 1 2
 x1   ∂u f1 (x1 , x2 , u)
     
 =   + u
   
       
∂ ∂   ∂ 
ẋ2 f2 (x1 , x2 , u) f2 (x1 , x2 , u) x2 f2 (x1 , x2 , u)
∂x1 ∂x2 |xeq ∂u |ueq=0
     
7 2 cos x2 − 4x32 x1 3u2
=     +   u
x1
e + 10x1 −3 x2 cos u
|(x1 =0,x2 =0) |ueq=0
    
7 1 x 0
=   1 +   u
1 −3 x2 1
 
x
 
∂ ∂  1
y= h(x1 , x2 , u) h(x1 , x2 , u)
∂x1 ∂x2 |(x1 =0,x2 =0) x2
 
h i x
= ex1 ex2  1
|(x1 =0,x2 =0) x
2
 
h i x1
= 1 1  
x2

4. Find the equilibrium point/s of the following system and the linearized model of
the system at the point/s.

ẋ1 = 3x1 + (x21 + x22 )

ẋ2 = 2x1 x2 (x21 + x22 )

Solution
At equilibrium point/s the the state dynamics is zero ẋ1 = 0, ẋ2 = 0, thus

f1 (x1 , x2 ) = 3x1 + (x21 − x22 ) = 0

f2 (x1 , x2 ) = 2x1 x2 (x21 − x22 ) = 0

The equilibrium points are

x1 = 0, x2 = 0, x21 − x22 = 0

Which means all of the points where x21 x22 = 0 ⇒ x1 = x2 , or x1 = −x2 are
equilibrium points of the system. This set of points is called a limit cycle of the
system.

Chapter 6 182
Daniel Abebe Modern Control Systems Lecture Note

The linearized model at (x1 = x2 ) is

   ∂ ∂
  
ẋ f (x
1 1 , x2 ) f (x
1 1 , x2 ) x
 1   ∂x1 ∂x2 
  1
 =
     
  
    
∂ ∂ 
ẋ2 f2 (x1 , x2 ) f2 (x1 , x2 ) x2
∂x1 ∂x2 |x
    eq
3 + 2x1 2x2 x
=   1
4x1 x2 4x2 x1 x2
|(x1 =x2 )
  
3 + 2x1 2x1 x
=   1
4x21 4x21 x2

or at (x1 = −x2 ) is
   ∂ ∂

ẋ f1 (x1 , x2 ) f1 (x1 , x2 )  
 1  ∂x1 ∂x2 
x
 1
 
 =
   

  
∂ ∂  x2
ẋ2 f2 (x1 , x2 ) f2 (x1 , x2 )
∂x1 ∂x2 |x
    eq
3 + 2x1 2x2 x
=   1
4x1 x2 4x2 x1 x2
|(x1 =−x2 )
  
3 − 2x2 −2x2 x
=   1
−4x22 −4x22 x2

Chapter 6 183
Daniel Abebe Modern Control Systems Lecture Note

6.4 Exercises
1. For the following nonlinear system, find the equilibrium point and the linear model
at the equilibrium point.
(a) ẋ1 = −x1 + x2
(f) ẋ1 = −3x2
ẋ2 = −x1 − x2 − x21
ẋ2 = x1 − α(x32 − x2 )
(b) ẋ1 = −3x2
(g) ẋ1 = −x1 + x2 − x3 + αx31
ẋ2 = x1 − x2 − 2x32
ẋ2 = −x1 − x2
(c) ẋ1 = −x31 − 3x1 + x2 ẋ3 = x1 − x2 + βx3 + u
ẋ2 = −2x1
(h) ẋ1 = −x1 + x2
(d) ẋ1 = x21 − x22 ẋ2 = (x1 + x2 ) sin x1 − 3x2
ẋ2 = −2x1 x2
(i) z̈ + ż + z 3 + u = 0
(e) ẋ1 = −x1 + x2
(j) ẍ − µ(1 − x2 )ẋ + x = 0
ẋ2 = −x1 − x2 − x21
2. For DC motor dynamics on 1.7.1, find the transfer function from input voltage to
output speed and input torque to output speed.

3. For the following systems, find a transfer function from each input to outputs, zeros,
and poles of the system.

(a)
   
0 1 0 0 0  
    1 0 0
ẋ =  0 x + 0 1 u, y= x
   
0 1
    0 1 0
−4 −4 −1 1 1

(b)
   
0 1 0 0 0  
    1 0 −2
ẋ =  0 1  x + 0 1 u, y= x
   
0
    0 2 1
−4 −4 −1 1 0

(c)
   
0 1 0 0  
    −3 1 0
ẋ =  0 1  x +  1  u, y= x
   
0
    4 0 1
−6 −11 −6 −3

(d)
     
1 1 0 0 1 1 0 −0.5
     
ẋ = 0 1 0 x + 1 0 u, y = −1 1 0 x
     
     
0 1 1 0 1 0 1 1

Chapter 6 184
Chapter 7

Analog Simulation

7.1 Analog Computers and Simulation

An analog computer is a type of computer that uses physical quantities, such as voltage,
to represent and solve mathematical problems. It works by creating an electrical circuit
that is analogous to the mathematical equation that represents the problem. The com-
puter then measures the voltage at different points in the circuit to obtain the solution
to the problem.

Analog computers are particularly well-suited for solving differential equations, which
are equations that describe how a quantity changes over time. This is because analog
computers can continuously measure and adjust the voltage in the circuit to represent
the changing quantity.

Analog computers were widely used in the 20th century to solve problems in engineering,
physics, and mathematics. However, they have been largely replaced by digital comput-
ers, which are more accurate and versatile. However, analog computers are still used in
some specialized applications, such as aircraft flight control systems.

Here is a simplified explanation of how an analog computer works:

• The mathematical model of the problem is converted into an analog block diagram.
The block diagram shows how the different parts of the problem are related to each
other.

• The analog block diagram is used to create an electrical circuit. The electrical
components are connected together in a way that represents the mathematical re-
lationships in the block diagram.

• The electrical circuit is powered up, and the voltages at different points in the
circuit are measured. The voltages represent the solutions to the problem.

• The voltages are converted back to the original problem variables. This gives the
final solution to the problem.

Analog computers have a number of advantages over digital computers for certain types
of problems. For example, analog computers can solve differential equations in real time.
This makes them ideal for applications such as aircraft flight control systems, where the
computer needs to be able to respond to changes in the environment very quickly.

185
Daniel Abebe Modern Control Systems Lecture Note

However, analog computers also have a number of disadvantages. For example, analog
computers are less accurate than digital computers. They are also more difficult to
program and maintain.

7.2 Solving Differential Equations with an Analog Computer

A typical simulation of a physical system involves a mathematical model consisting of a
set of one or more differential equations and initial conditions on the variables. If the
system is linear, the differential equations are linear, and the operations required are
1. Summation
2. Sign inversion
3. Multiplication by a constant
4. Integration
5. Differentiation
For practical reasons, the integration operation is easier to implement than the differen-
tiation operation. The reason lies in the fact that computer signals are real voltages and,
therefore, are corrupted by noise to some extent. Since integration tends to average out
the effects of noise (while differentiation will accentuate it), a more precise solution can
be obtained using integration techniques.

x2
+ x x
+ 
x1 - y  x1  x2  x3
x3

Summation Multiplication

x0
x x x
 1
t
x  x0   x( )d
Integration 0 Sign Inversion

Figure 7.1: Analog Computing Operations

Consider an unforced first-order system (zero input signal), with initial condition x0
ẋ = x x0 = 1 (7.2.1)
Integrating both sides
Z t
x = x0 + x(τ ) dτ (7.2.2)
0

Chapter 7 186
Daniel Abebe Modern Control Systems Lecture Note

x0

 t
x  x0   x ( )d
0

Figure 7.2: Unforced First Order System Realization

As we can understand from Figure 7.2, it is not necessary to know the input of the
integrator to solve equation 7.2.1. The only requirement is that the input to the integrator
must always be equal to the output. Hence, the concept of feeding the unknown output
back to the integrator is the solution to the differential equation in an analog computer.

This is because the solution to a differential equation is general form. An initial or

boundary condition is required to determine a specific solution for a particular differential
equation.

Since higher-order differential equations can be reduced to cascaded first-order differential

equations, the method of realization stays the same. For example, consider a second-order
unforced differential equation

ẍ + ẋ + x = 0 x0 = 1 ẋ0 = 0 (7.2.3)

This equation can be written as

ẋ1 x2
ẋ2 = −x1 − x2 (7.2.4)

Integration on both sides

Z t
x1 =x1 (0) + x2 (τ ) dτ
0
Z t
x2 =x2 (0) − [x1 (τ ) + x2 (τ )] dτ (7.2.5)
0

x2 (0) x1 (0)

x2 x1
+  

Figure 7.3: Unforced Second Order System Realization

Chapter 7 187
Daniel Abebe Modern Control Systems Lecture Note

7.3 Physical Realization of Analog Computer

1. Summation

R1 Rf
v1
R2
v2 −
.. vo
. +
Rn
vn

Figure 7.4: Summer Circuit Diagram Using Op-amp

From the inverting operational amplifier, the voltage transfer characteristics we

have,
Rf Rf Rf
vo = − v1 − v2 − . . . − vn (7.3.1)
R1 R2 Rn
Rf Rf Rf
Let k1 = , k2 = , . . ., and kn = we have,
R1 R2 Rn
vo = −k1 v1 − k1 v2 − . . . − k1 vn (7.3.2)

2. Integration

C
R1
v1
R2
v2 −
.. vo
. +
Rn
vn

Figure 7.5: Integrator Circuit Diagram Using Op-amp

For the integrator circuit, the voltage transfer function is,

Z t Z t Z t
1 1 1
vo = vo (0) − v1 (τ )dτ − v2 (τ )dτ − . . . − vn (τ )dτ
0 CR1 0 CR2 0 CRn
Z t 
1 1 1
vo = vo (0) − v1 (τ ) + v2 (τ ) + . . . + vn (τ ) dτ (7.3.3)
0 CR1 CR2 CRn
1 1 1
Let k1 = , k2 = , . . ., and kn = we have,
CR1 CR2 CRn
Z t
vo = vo (0) − [k1 v1 (τ ) + k2 v2 (τ ) + . . . + kn vn (τ )] dτ (7.3.4)
0

Chapter 7 188
Daniel Abebe Modern Control Systems Lecture Note

C1
Rf
v1

v2 −
vo
C2 +
..
.
vn

Cn

Figure 7.6: Differentiator Circuit Diagram Using Op-amp

3. Differentiation
For the differentiator circuit, the voltage transfer function is,
dv1 dv2 dvn
vo = −Rf C1 − Rf C2 − . . . − Rf Cn
Zdt t  dt dt 
1 1 1
vo = vo (0) − v1 (τ ) + v2 (τ ) + . . . + vn (τ ) dτ (7.3.5)
0 CR1 CR2 CRn

Let k1 = Rf C1 , k2 = Rf C2 , . . ., and kn = Rf Cn we have,

dv1 dv2 dvn

vo = −k1 − k2 − . . . − kn (7.3.6)
dt dt dt

4. Sign inversion

R1 Rf 1 R2 Rf 2
v1

− −
vo
+ +

Figure 7.7: Sign Invertor Circuit Diagram Using Op-amp

For the sign inversion circuit, the voltage transfer function is,
Rf 2 Rf 1
vo = v1 (7.3.7)
R1 R2
Rf 2 Rf 1
If = 1, we have vo = v1 .
R1 R2

Chapter 7 189
Daniel Abebe Modern Control Systems Lecture Note

7.4 Programming Differential Equation Using Analog Computer

To solve differential equations using an analog computer, we can combine the circuits for
an nth order integrator and nth order differentiator.

The circuit for an nth order integrator consists of n operational amplifiers (op-amps)
connected in series with capacitors and resistors. The output of each op-amp is connected
to the input of the next op-amp, and the input signal is applied to the first op-amp. The
output of the last op-amp is the integrated signal.

The circuit for an nth order differentiator is similar, but with capacitors and resistors
connected in parallel instead of in series. The input signal is applied to the first capacitor,
and the output of each op-amp is connected to the input of the next op-amp. The output
of the last op-amp is the differentiated signal.

By combining these circuits, we can solve differential equations by converting them into
an analog block diagram, creating an electrical circuit based on the block diagram, and
measuring the voltages at different points in the circuit to obtain the solution to the
problem.

Consider nth order differential equation

dn dn−1 d
n
x + n−1
x + ... + x + x = 0 (7.4.1)
dt dt dt
with n initial conditions.
dn dn
If x is known, then all the rest variables can be determined by integrating x, or
dtn dtn
dn−1 d dn
if n−1 x, . . ., x and x are known, we can determine n x. This implies that we are
dt dt dt
only expected to know the input and output of the integrator. The analog circuit that
dn
integrates n x is shown in Figure 7.8.
dt

dn x dn−1 x dn−2 x
− x
dtn dtn−1 dtn−2 . . .

Figure 7.8: Analog Circuit for a nth Order Integrator

Chapter 7 190
Daniel Abebe Modern Control Systems Lecture Note

dn−1 d dn
The analog circuit that can sum n−1 x, . . ., x and x to get n x the is shown in
dt dt dt
Figure 7.8.

R1 Rf
dn−1 x
dtn−1 R2
dn−2 x
−
dtn−2 dn x
.. dtn
. +
Rn
x

Figure 7.9: Summer Circuit

Combining the two analog circuits will give us an analog circuit used to solve nth order
ordinary differential equations shown in Figure 7.10.

d nx d n1 x dx
dt n dt n1 dt x

Figure 7.10: nth Order Ordinary Differential Solver

Chapter 7 191
Daniel Abebe Modern Control Systems Lecture Note

7.5 Magnitude and Time Scaling

In the field of analog computing, equipment constraints often limit the maximum voltage
output of each amplifier to 10 volts. This is because exceeding this voltage output can
result in amplifier overload and render the solution invalid. When faced with a problem
that requires a large voltage output, it becomes necessary to scale the problem down to
fit the computer’s limitations.

For instance, if we were to solve a problem involving rocket speed on an analog computer,
it would be impossible to relate 1 volt on the computer to 1cm/sec in the real problem.
However, we could relate 1 volt on the computer to 1km/sec in the problem. This illus-
trates the need to scale the problem so that it will fit the computer.

Similarly, in the same problem, an acceleration term might have a maximum value of
1cm/sec2 . Using the basic unit of miles for length as determined for scaling the velocity
would give a voltage of approximately 0.002 volts, which corresponds to the acceleration
term. However, this signal is in the noise level of the computer and would yield erroneous
results. Thus, it is necessary to scale up some parts of a problem and scale down other
parts.

Time scaling is also necessary in certain situations. For example, when studying a phys-
ical phenomenon that evolves much faster or slower than is convenient for the computer
operator, time scaling can be used. If we wish to study planetary motion, waiting for
a complete period of revolution about the sun would be impractical. Instead, an actual
time of one year may be represented by 10 seconds of computer time.

Equipment constraints and limitations require careful consideration when solving prob-
lems on analog computers. Scaling techniques must be employed to ensure that the
problem fits within the computer’s limitations. Time scaling can also be used to study
physical phenomena that evolve at different rates than is convenient for the computer
operator.

Chapter 7 192
Daniel Abebe Modern Control Systems Lecture Note

7.5.1 Magnitude scaling

Magnitude scaling is necessary when a problem variable, existing as an amplifier output,
becomes too large and causes amplifier overload or becomes too small, causing the am-
plifier output noise to dominate the signal.

Scaling is performed based on the maximum voltage supplied to the analog computer and
the maximum magnitude that appears in the solution of the differential equation. The
input and output of the analog computer are scaled down using the voltage divider rule,
while operational amplifiers are used to scale them up. Similarly, the initial conditions
can be scaled up or down.

vin

vin vo
vo   vin

Figure 7.11: Circuit Diagram of Scale Down Using Potentiometer

A Systematic Approach to Magnitude Scaling

1. Draw the unscaled analog diagram. Remember that + 10 volts is the maximum
reference available on the computer.
2. Estimate the maximum value of each variable appearing as an amplifier output.
Only the amplifier outputs need to be considered since all other elements have a
gain that is less than unity.
3. Define a scaling relationship between each problem variable and the corresponding
computer variable, and prepare a scaling table.
4. Write the scaled analog equations for each amplifier in terms of the computer vari-
ables (amplifier outputs) defined in step 3.
5. Draw the scaled analog diagram using the equations obtained in step 5. Label all
computer variables on the diagram. The final diagram should have the same form as
the unscaled diagram except for different potentiometer values and amplifier gains.
Suppose we need to find an analog simulation that converts m to cm, with the maximum
value of m converted being 5m. Let x be the number of m and y be the number of cm.
So we have
y = ax
Where a = 100.

We must scale the problem so that we can represent 500 centimeters, ymax , as an analog
voltage. We also know
V x = Lx x Vy = Ly y

Chapter 7 193
Daniel Abebe Modern Control Systems Lecture Note

The analog simulation diagram for the system is shown below.

Vx  Lx x Vy  Ly y

Figure 7.12: Scaling Diagram

A potentiometer is set to some value which we denote by 1. Electronically,

Vout =Overal Gain Vin

−Ly y = − kGLx x
kGLx
y= x
Ly

kGLx
Which means a = . So the gain of the potentiometer must have a setting
Ly

aLy
k=
Lx
In this example
10 10 aLy 100 ∗ 0.1
Lx = =2 Ly = = 0.1 ⇒ k = = =5
5 100 Lx 2
Generally
Level Out COEF
k= (7.5.1)
Level In gain
Where

• Level Out is the level of the variable immediately following the pot to be set,

• Level In the level of the variable preceding the pot,

• COEF relates to the equation to be simulated and,

• gain is the gain of a summer, inverter, or integrator following the pot.

Chapter 7 194
Daniel Abebe Modern Control Systems Lecture Note

7.5.2 Time scaling

Analog simulation time can be adjusted to run faster or slower than real time.

τc = βt (7.5.2)

where τc computer, t is real time and β scaling factor.

1
Time scaling can be achieved by multiplying the gains of every integrator by .
β
Therefore, we can adjust the previous pot setting formula to take care of magnitude and
time scaling. We simply use a single pot for each input to the integrators such that:
Level Out COEF 1
k= (7.5.3)
Level In gain β
Since there is no time variation for initial conditions and non-integrating operational
amplifiers, time scaling will not be applied to them.

7.6 Analog Simulation of Transfer Functions and State Space

Representation of Systems
To develop a scaled analog simulation diagram of transfer functions, use the following
procedures.

1. Convert the transfer function into differential equations using direct or indirect
programming or standard programming methods.

2. Write the differential equation to be solved in normal form.

3. Estimate maximum values for variables.

4. Draw integrators needed and put a pot between each pair of integrators.

5. Determine the scaling factors based on the maximum variable occurring in the
differential equation, as described in step 3.

To develop a scaled analog simulation diagram of state space representation, use the
following procedures.

1. Derive the differential equations from the state space equations.

2. Write the differential equation to be solved in normal form.

3. Estimate maximum values for variables.

4. Draw all integrators needed and put a pot between each pair of integrators.

5. Determine the scaling factors based on the maximum variable occurring in the
differential equation, as described in step 3.

Chapter 7 195
Daniel Abebe Modern Control Systems Lecture Note

7.6.1 Initial conditions

Suppose that we wish to simulate a transfer function, but with initial conditions included.
This is not a well-posed problem, as such, but we can work around that. Consider a
second-order transfer function
a2 s 2 + a1 s + a0
G(s) = 2 (7.6.1)
s + b1 s + b0
In order to incorporate these initial conditions, we need to “translate” them into the
initial conditions of the actual integrators, and possibly the input.

y(0− ) = a0 x(0− ) + a1 ẋ(0− ) + a2 ẍ(0− )

= a0 x(0− ) + a1 ẋ(0− ) + a2 (−u(0− ) − b1 ẋ(0− ) − b0 x(0))
= (a0 − a2 b0 )x(0− ) + (a1 − a2 b1 )x(0− ) − a2 u(0− ) (7.6.2)

and also

ẏ(0− ) = a0 ẋ(0− ) + a1 ẍ(0− ) + a2 x(3) (0− )

= a0 ẋ(0− ) + a1 (−u(0− ) − b1 ẋ(0− ) − b0 x(0)) + a2 (−u̇(0− ) − b1 ẍ(0− ) − b0 ẋ(0))
= −b0 (a1 − a2 b1 )x(0−1 ) + [a0 − a2 b0 − b1 (a1 − a2 b1 )]ẋ(0−1 ) − (a1 − a2 b1 )u(0−1 )
− a2 u̇(0−1 ) (7.6.3)

We note that the initial states of the integrators can be rewritten in terms of the initial
states of y and ẏ, and upon the input and its derivative. Note that when we take the
Laplace transform of a signal (like u(t)), we assume the signal to be zero and all its
derivatives to be zero. With is assumptions equations 7.6.2 and 7.6.3 simplifies to

y(0− ) = (a0 − a2 b0 )x(0− ) + (a1 − a2 b1 )ẋ(0− ) (7.6.4)

ẏ(0− ) = −b0 (a1 − a2 b1 )x(0−1 ) + [a0 − a2 b0 − b1 (a1 − a2 b1 )]ẋ(0−1 ) (7.6.5)

Let a11 = a0 −a2 b0 , a12 = a1 −a2 b1 , a21 = −b0 (a1 −a2 b1 ) and a22 = a0 −a2 b0 −b1 (a1 −a2 b1 ),

y(0− ) = a11 x(0− ) + a12 ẋ(0− ) (7.6.6)

ẏ(0− ) = a21 x(0−1 ) + a22 ẋ(0−1 ) (7.6.7)
        −1  
− − − −
y(0 ) a a x(0 ) x(0 ) a a y(0 )
  =  11 12   ⇔  =  11 12    (7.6.8)
ẏ(0− ) a21 a22 ẋ(0− ) ẋ(0− ) a21 a22 ẏ(0− )

thus we obtain
α22 y(0− ) − α12 ẏ(0− )
x(0− ) = (7.6.9)
α11 α22 − α12 α21
−α12 y(0− ) − α22 ẏ(0− )
ẋ(0− ) = (7.6.10)
α11 α22 − α12 α21

Chapter 7 196
Daniel Abebe Modern Control Systems Lecture Note

7.7 Examples
1. Determine the circuit to produce the output voltage given by

vo = −v1 − 10v2 + v3 + 10v4

Solution:
The analog circuit diagram that must be patched on the analog computer

50k 50k
v1
5k
v2 vo
50k 50k
v3
5k 
v4 - 50k

Figure 7.13: Circuit Diagram for Example 1

2. Determine the analog diagram and circuit to implement the equation

vo = −0.35v1 + 5.24v2 + 2.6

Solution:
The analog circuit diagram must be patched on the analog computer

v1 50k
.350
5k
50k
v2 5k
5k  
.524 - vo
-

10v 5k
.260

Figure 7.14: Circuit Diagram for Example 2

Chapter 7 197
Daniel Abebe Modern Control Systems Lecture Note

.350
v1 1
.524 vo
v2 10

-10v 1

Figure 7.15: Analog Diagram for Example 2

3. For the following second-order linear differential equation and initial conditions,
determine the analog circuit diagram.
ẍ + 0.5ẋ + x = 4 ẋ0 = 1 and x0 = 0 (7.7.1)
Solution:
The analog circuit diagram that must be patched on the analog computer

0.4 x0 x0
10v
x x x
  
0.5
1

Figure 7.16: Circuit Diagram for Example 3

4. For the following transfer function, determine an analog circuit diagram

a2 s 2 + a1 s + a0 N (s)
G(s) = 2 =
s + b1 s + b0 D(s)
Solution:
Using direct programming, we have
Y (s) a2 s2 + a1 s + a0
= 2 ;
U (s) s + b1 s + b0
With an intermediate term x(s) we have
Y (s)
= a2 s 2 + a1 s + a0
X(s)
Y (s) = a2 s2 X(s) + a1 sX(s) + a0 X(s)
and
X(s) 1
= 2
U (s) s + b1 s + b0
U (s) = s2 X(s) + b1 sX(s) + b0 X(s)
s2 X(s) − b1 sX(s) − b0 X(s) + U (s)

Chapter 7 198
Daniel Abebe Modern Control Systems Lecture Note

Y (s) X (s)
X (s) U (s)

Figure 7.17: Block diagram Representation of Transfer Function

This implies that the analog circuit diagram of the two differential equations can
be designed same way and combined as follows
The analog circuit diagram of the transfer function from the state to the output is

a2

x – 1 y
a1  1

–
a0

Figure 7.18: Analog Circuit Diagram of Transfer Function From State to Input

The analog circuit diagram of the transfer function from the input to the state is

u x x x
 –  – 
b1
b2
1

Figure 7.19: Analog Circuit Diagram of Transfer Function from Input to State

Combining the two analog circuit diagrams, the transfer function from input to
output is,

Chapter 7 199
Daniel Abebe Modern Control Systems Lecture Note

a2
u y
x x x
 –  –   1

b1 a0
1
b2
a1

Figure 7.20: Analog Circuit Diagram of Transfer Function from Input to Output

5. Simulate the following system and magnitude scale so that no overload occurs and
the maximum amplifier range is used

ẍ + 8ẋ + 25x = 500, x(0) = 40 ẋ(0) = 150

with |x|max = 50 and |ẋ|max = 250.

Solution:

(a) Step 1: Draw an unscaled analog diagram

10V 10V

5 4

10V G1
6
x x
G2
–  1 1 –  2
G3

0.8
2
0.25
10
3 5 -1

Figure 7.21: Unscaled Analog Circuit Diagram

(b) Step 2: This step is not necessary since the maximum values are already
specified in the problem statement.
(c) Step 3: Prepare a scaling table
(d) Step 4: Write the scaled equations for each amplifier.
1/5 1
k1 = , if we select G1 = 10, then k1 = 0.5. Similarly, computing for
1/25 G1

Chapter 7 200
Daniel Abebe Modern Control Systems Lecture Note

Table 7.1: Scaling table

Problem variable Estimated Maximum Level Computer variable

−ẋ 250 1/25 [-ẋ/25]
x 50 1/5 [x/5]
−2.5x 125 1/12.5 [−x/5]

x x
5

25


Figure 7.22: Establishing Integrator Outputs

the rest of the circuit, we have

1/25 8
k2 = = 0.8 G1 = 10
1/5 G1
1/25 25
k3 = = 0.5 G2 = 10
1/5 G2
1/5 40
k4 = = 0.8
10 1
1/25 150
k5 = = 0.6
10 1
1/25 500
k6 = = 0.2 G3 = 10
10 G3

(e) Step 5: Draw the scaled analog computer diagram

10V 10V

5 0.6 4 0.8
0.2
10V 10
6
x 10 x
10
–  1 1 –  2
10
0.5
0.8
2
0.25
1
3 5 -1
0.5

Figure 7.23: Fully Scaled Diagram

Chapter 7 201
Daniel Abebe Modern Control Systems Lecture Note

7.8 Exercises
1. For the following system, find an analog circuit diagram

s3 + 2s2 + 1 2s3 + 5s2 + 3s + 1

G1 (s) = G2 (s) =
s3 + 5s2 + 3s + 6 4s4 + 2s2 + 3s + 6

2. For the following system, find an analog circuit diagram

   
0 −1 2
ẋ =  x +  u
2 −4 0

3. For the following system, find an analog circuit diagram

   
0 −1 2 0
ẋ =  x +  u
2 −4 0 1

4. Magnitude-scale the following problem so that no overload occurs and maximum

amplifier range is used.

ẍ + ẋ + 25x = 0, x(0) = 20 ẋ(0) = 0

with |x|max ≤ 20 and |ẋ|max ≤ 100.

5. Magnitude-scale the following problem

ẍ + 10ẋ + 25x = 20, x(0) = 0 ẋ(0) = 0

with |x|max ≤ 1.6 and |ẋ|max ≤ 4.

6. Magnitude-scale and time scale the following problem

ẍ + 10ẋ + 50x = 20, x(0) = 10 ẋ(0) = 0

with |x|max ≤ 1.6, |ẋ|max ≤ 4 and β = 2.5.

Chapter 202
Appendix A

Hamilton’s Principle and Lagrangian

Mechanics

A.1 Hamilton’s Principle (Generalized Classical Mechanics)

Hamilton’s principle states that the motion of a system between two points in time t1
and t2 is such that the action S is stationary (usually optimized):
Z t2
J= L dt, (A.1.1)
t1

where L is the Lagrangian.

A.1.1 What is the Lagrangian?

The Lagrangian L is a function of generalized coordinates q, generalized velocities q̇, and
time t (in some cases). It is defined as:

L(q, q̇, t) = T (q, q̇, t) − V (q, t) (A.1.2)

where:

• T (q, q̇, t) is the kinetic energy of the system in generalized coordinates.

• V (q, t) is the potential energy of the system in generalized coordinates.

A.1.2 Properties of the Lagrangian

The Lagrangian of the system:

• Encodes the dynamics of the system in a single function.

• Is a scalar quantity, making it easier to work with than vector quantities like force.

• is particularly useful for systems with constraints or complex geometries.

i
A.1.3 Connection to Newton’s Second Law
Consider a particle with mass m moving in a generalized coordinate system. The kinetic
1
energy of the particle is T = mq̇2 , and its potential energy is V = V (q). The Lagrangian
2
of the system is:
1
L = mq̇2 − V (q) (A.1.3)
2
Applying the Euler-Lagrange equation:
∂V d
− − (mq̇) = 0 =⇒ F = mq̈ (A.1.4)
∂q dt
This recovers Newton’s second law.

A.1.4 Why Use the Euler-Lagrange Method?

The Euler-Lagrange method provides:
• A unified framework for mechanics:
– Generalizes classical mechanics to more complex systems (e.g., systems with
constraints).
– Works in any coordinate system (non-Cartesian coordinates).
• Extensions to other fields (e.g., electromagnetism, fluid dynamics).
• A foundation for quantum mechanics and modern physics.

A.2 Derivation of Euler-Lagrange Equations

A.2.1 Functional vs Function
A function depends on one or more independent variables:
• Function: V (x, y, z)
A functional can depend on one or more functions:
• Functional: J(x(t), y(t), z(t))
Note: The key difference is that in a functional, x, y, and z are functions of another
independent variable.

A.2.2 Increment of a Function and a Functional

Consider a function f (t). The increment of f (t) is defined as:
∆f = f (t + ∆t) − f (t) (A.2.1)
∆f depends on both t and ∆t, hence it is written as ∆f (t, ∆t). Similarly, for a functional
J(x(t), t), the increment is defined as:
∆J = J((x(t) + δx(t)), t) − J(x(t), t) (A.2.2)
Here, δx(t) is called the variation of the function x(t). Since the increment of a functional
depends on x(t) and δx(t), it is written as J(x(t), δx(t)).

ii
A.2.3 Differential of a Function
For a function f (t), the Taylor series expansion about a point t∗ is given as:

1 ∂ 2f
   
∗ ∂f
∆f = f (t ) + ∆t + ∆t2 + · · · − f (t∗ ) (A.2.3)
∂t t∗ 2! ∂t2 t∗

Neglecting higher-order terms in ∆t:

 
∗ ∂f
∆f = f (t ) + ∆t − f (t∗ )
∂t t∗
∆f = f˙(t∗ )∆t = df (A.2.4)

Here, df is called the differential of f at point t∗ , and f˙(t) is the slope of f at t∗ .

A.3 Optimization of Functions and Functionals

A.3.1 Optimum of a Function
For a function to have an optimum value at a point, the following conditions must be
met:

• Necessary Condition: The first derivative must vanish at the point (i.e., df = 0
at t∗ ).

• Sufficient Condition:

– d2 f > 0 for a minimum.

– d2 f < 0 for a maximum.

• If d2 f = 0, the point is called an inflection point or stationary point.

A.3.2 Variation of a Functional

Expanding a functional J(x(t), t) using Taylor series about a function x∗ (t), gives us:

1 ∂ 2J 2
   
∗ ∂J
∆J = J(x (t)) + δx + 2
δx + . . . − J(x∗ (t))
∂x x∗ (t) 2! ∂x x∗ (t)

∆J = δJ + δ 2 J + · · · + higher-order terms (A.3.1)

∂J 1 ∂ 2J 2
Here, δJ = δx is called the first variation, and δ 2 J = δx is called the second
∂x ∗
2! ∂x2
variation of the functional about a function x (t).

Neglecting higher-order terms in δx, the linear approximation of variation of J(x(t), t)

about a function x∗ (t) is:

∂J
∆J = δJ = δx (A.3.2)
∂x

iii
A.4 Optimization of an Unconstrained Functional
For a function to have an optimum value:

• Necessary Condition: The first variation must vanish (δJ = 0).

• Sufficient Condition:

– δ 2 J > 0 for a local minimum.

– δ 2 J < 0 for a local maximum.

A.4.1 Typical Functional in Control Systems

For linear time systems with functional J(x, ẋ):
Z tf
∆J = (φ(x + δx, ẋ + δ ẋ)) dt (A.4.1)
t0

Using Taylor Series Expansions:

Z tf  
∂φ ∂φ
∆J = δx + δ ẋ + HOT dt = 0 (A.4.2)
t0 ∂x ∂ ẋ

The first variation of the functional is:

Z tf  
∂φ ∂φ
δJ = δx + δ ẋ dt = 0 (A.4.3)
t0 ∂x ∂ ẋ

A.4.2 Necessary Condition for Optimum

For the functional to have an optimum value, δJ = 0. This leads to:
Z tf  
∂φ ∂φ
δx + δ ẋ dt = 0 (A.4.4)
t0 ∂x ∂ ẋ

Using integration by parts for the second term, we derive the Euler-Lagrange equation:
 
∂φ d ∂φ
− =0 (A.4.5)
∂x dt ∂ ẋ

And the boundary condition:

 
∂φ
δx =0 (A.4.6)
∂ ẋ

iv
Boundary Conditions
∂φ
For the boundary condition to be zero, either or δx must be zero at the boundary.
∂ ẋ
This leads to four cases:

1. Case-I: x is fixed at t0 and tf (fixed end trajectory). Here, x(t0 ) = C1 and x(tf ) =
C2 , where C1 and C2 are constants. Since δx = 0 at both t0 and tf , there are no
∂φ
restrictions on .
∂ ẋ
∂φ
2. Case-II: x is fixed at t0 and free at tf . Here, x(t0 ) = C1 , and = 0 at tf because
∂ ẋ
∂φ
δx is unknown at tf . No restrictions apply to at t0 .
∂ ẋ
∂φ
3. Case-III: x is fixed at tf and free at t0 . Here, x(tf ) = C2 , and = 0 at t0
∂ ẋ
∂φ
because δx is unknown at t0 . No restrictions apply to at tf .
∂ ẋ
∂φ
4. Case-IV: x is free at t0 and tf . Here, = 0 at both t0 and tf because δx is
∂ ẋ
unknown at both times.

A.4.3 Sufficient Condition for Optimum

The sufficient condition for an optimum is derived from the second variation of the func-
tional J, denoted as δ 2 J. The expression for δ 2 J is given by:

tf
∂ 2φ ∂ 2φ ∂ 2φ
Z      
2 1 d 2 2
δ J= 2
− (δx) + (δ ẋ) dt (A.4.7)
2 t0 ∂x dt ∂ ẋ∂x ∂ ẋ2

For a local minimum, the following conditions must be satisfied:

∂ 2φ ∂ 2φ ∂ 2φ
 
d
− ≥ 0 and ≥0 (A.4.8)
∂x2 dt ∂x∂ ẋ ∂ ẋ2

For a local maximum, the conditions are:

∂ 2φ ∂ 2φ ∂ 2φ
 
d
2
− ≤ 0 and ≤0 (A.4.9)
∂x dt ∂x∂ ẋ ∂ ẋ2

v
A.5 Typical Functional in Control Systems
In control systems, a typical functional has the form J(x, ẋ, u, u̇, t):

Z tf
J(x, ẋ, u, u̇, t) = φ(x, ẋ, u, u̇, t) dt (A.5.1)
t0

The increment of J(x, ẋ, u, u̇, t) is given by:

Z tf
∆J = (φ(x + δx, ẋ + δ ẋ, u + δu, u̇ + δ u̇, t + δt) − φ(x, ẋ, u, u̇, t)) dt (A.5.2)
t0

For linear time-invariant systems, the increment simplifies to:

Z tf
∆J = (φ(x + δx, ẋ + δ ẋ, u + δu, u̇ + δ u̇) − φ(x, ẋ, u, u̇)) dt (A.5.3)
t0

Using Taylor Series expansions, the first variation of the functional is:

Z tf  
∂φ ∂φ ∂φ ∂φ
∆J = δx + δ ẋ + δu + δ u̇ + hot dt = 0 (A.5.4)
t0 ∂x ∂ ẋ ∂u ∂ u̇
The first variation of the functional is:

Z tf   Z tf  
∂φ ∂φ ∂φ ∂φ
δJ = δx + δ ẋ dt + δu + δ u̇ dt = 0 (A.5.5)
t0 ∂x ∂ ẋ t0 ∂u ∂ u̇

A.6 Optimization of Constrained Functional

In real-world problems, especially in control systems, optimization is often subject to
constraints imposed by natural laws or technological limitations. For example, satellite
tracking control is restricted by the selected trajectory and the maximum input allowed.
The objective is to determine a control signal u that optimizes a cost function while
satisfying these constraints.

A.6.1 Problem Formulation

The problem is to optimize a cost function J while fulfilling the state dynamics of the
system. Specifically, we aim to determine a control signal u that minimizes:

Z tf
Υ(x, ẋ, u)dt (A.6.1)
t0

Subject to the state dynamics:

ẋ = Ax + Bu (A.6.2)

vi
A.6.2 General Function Definition
To incorporate the constraints into the optimization problem, we define a general function
that includes both the cost function and the constraints:

φ(x, ẋ, u) = Υ(x, ẋ, u) + λ⊤ (Ax + Bu − ẋ) (A.6.3)

This reduces the constrained optimization problem to an unconstrained one:

Z tf
Υ(x, ẋ, u) + λ⊤ (Ax + Bu − ẋ) dt


J=
t
Z 0tf
= φ(x, ẋ, u)dt (A.6.4)
t0

A.6.3 Euler-Lagrangian Equation

For the function to have an optimum value, the Euler-Lagrangian equations must be
satisfied:

 
∂φ d ∂φ
− =0 (A.6.5)
∂x dt ∂ ẋ
 
∂φ d ∂φ
− =0 (A.6.6)
∂u dt ∂ u̇
 
∂φ d ∂φ
− =0 (A.6.7)
∂λ⊤ dt ∂ λ̇⊤

A.7 Euler-Lagrangian Equation for systems with Generalized

Forces
The first variation of the functional is:
Z tf  
∂φ ∂φ
δJ = δx + δ ẋ dt = Fi (A.7.1)
t0 ∂x ∂ ẋ

where Fi are the generalized forces, including friction and external input.

Which leads to the Euler-Lagrangian equation:

 
∂L d ∂L
− = Fi , (A.7.2)
∂q dt ∂ q̇

vii
A.8 Example:-1 Inverted Pendulum on a Cart System Dynam-
ics
The objective of the inverted pendulum on a cart control system is to maintain the
pendulum in an upright position while simultaneously controlling the movement of the
cart along a track. This system is a classic example in control theory and is used to study
stability and control strategies.

lsinθ

y
θ

lcosθ
x

Figure A.1: Inverted Pendulum on a Cart (IPC) System

A.8.1 System Description

The system consists of:
• A cart moving along a rail.

• A pendulum attached to the cart.

The generalized coordinates are:

q1 = x (cart position), q2 = θ (pendulum angle). (A.8.1)

Friction forces in the system include:

• bc q̇1 : Friction between the cart and the rail (viscous damping).

• bp q̇2 : Friction at the pendulum joint (viscous damping).

A.8.2 Dynamic Model of the System

The linear velocity of the cart and the angular velocity of the pendulum are given by:
       
x q ẋ q̇
 c =  1 ,  c =  1 (A.8.2)
yc 0 ẏc 0

For the pendulum:

       
x q + l sin q2 ẋ q̇ + lq̇2 cos q2
 p =  1 ,  p =  1  (A.8.3)
yp l cos q2 ẏp −lq̇2 sin q2

viii
Kinetic and Potential Energy
The total kinetic energy of the system is:
1 1 1
Ke = mc ẋ2c + mp (ẋ2p + ẏp2 ) + I q̇22 (A.8.4)
2 2 2
Substituting the velocities:
1 1
Ke = (mc + mp )q̇12 + mp lq̇1 q̇2 cos q2 + (mp l2 + I)q̇22 (A.8.5)
2 2

The potential energy of the system is:

V = mp gl cos q2 (A.8.6)

Lagrangian Formulation
The Lagrangian is defined as:

L = Ke − V (A.8.7)

Substituting Ke and V :
1 1
L = (mc + mp )q̇12 + mp lq̇1 q̇2 cos q2 + (mp l2 + I)q̇22 − mp gl cos q2 (A.8.8)
2 2

Euler-Lagrange Equations with Friction

The Euler-Lagrange equations are:
 
d ∂L ∂L
− = Fi (A.8.9)
dt ∂ q̇i ∂qi

where Fi are the generalized forces, including friction and external input.

Generalized Forces
• For q1 (cart motion):

F1 = u − bc q̇1 (A.8.10)

where u is the external force applied to the cart, and bc q̇1 is the friction force.

• For q2 (pendulum motion):

F2 = −bp q̇2 (A.8.11)

where bp q̇2 is the friction torque at the pendulum joint.

ix
Equations of Motion
The equations of motion for the system are derived as follows:

Equation for q1 (Cart Motion)

(mc + mp )q̈1 + mp lq̈2 cos q2 − mp lq̇22 sin q2 = u − bc q̇1 (A.8.12)

Equation for q2 (Pendulum Motion)

mp lq̈1 cos q2 + (mp l2 + I)q̈2 − mp gl sin q2 = −bp q̇2 (A.8.13)

State-Space Representation
Let:

x1 = q 1 , x2 = q̇1 , x3 = q2 , x4 = q̇2 (A.8.14)

The state-space equations become:



ẋ1 = x2
ẋ2 = 1 (mp l2 + I)(u − bc x2 + mp lx24 sin x3 ) − mp l cos x3 (mp gl sin x3 − bp x4 )

  

λ

ẋ 3 = x 4
1


ẋ = −(m l cos x )(u − b x + m lx2 sin x ) + (m + m )(m gl sin x − b x )

4 p 3 c 2 p 4 3 c p p 3 p 4
λ
(A.8.15)
Where:

λ = (mc + mp )(mp l2 + I) − (mp l cos x3 )2 (A.8.16)

Linearization
Using the small angle approximation about θ = 0◦ , we assume cos θ ≈ 1, sin θ ≈ θ, and
θ̇ ≈ 0. Substituting these approximations, the linear model is:
   
0 1 0 0 0
   
ẋ  x1
 1  2 2 2
        
b c (m p l + I) (m p l) g b p m p l     m p l + I 
  0
ẋ2   − −   
 x2  

 = λ λ λ   +  λ 
u
      
ẋ3  0 0 0 1  x3   0 
              
ẋ4
 bc mp l mp lg(mp + mc ) bp (mp + mc )  x  mp l 
0 − 4 −
λ λ λ λ
(A.8.17)
 
x
     1
 
y1 1 0 0 0 x2 
y= =  
  (A.8.18)
y2 0 0 1 0 x3 
 
x4

x
A.9 Example:-2 Ball and Beam System
The ball and beam system is a classic control theory problem that involves balancing a
ball on a beam, which can tilt to control the ball’s position.

I,m
x




Figure A.2: Ball and Beam System

A.9.1 System Description

• The system consists of a beam that can rotate about its center and a ball that rolls
along the beam.
• The goal is to control the beam’s angle to regulate the ball’s position.
• The system has two primary degrees of freedom: the position of the ball along the
beam and the angular position of the beam.

A.9.2 Dynamic Model of the System

Let I1 , m, and I2 denote the inertia of the beam, the mass of the ball, and the inertia of
the ball, respectively.
• Position of the center of mass of the ball:

xc = (x cos θ, x sin θ) (A.9.1)

• Velocity of the center of mass of the ball:

ẋc = (ẋ cos θ − xθ̇ sin θ, ẋ sin θ + xθ̇ cos θ) (A.9.2)

• Translational kinetic energy:

1 1
Ktrans = mẋ2c = m(ẋ2 + x2 θ̇2 ) (A.9.3)
2 2

• Rotational kinetic energy:

 2
1 1 ẋ
Krot = I1 θ̇2 + I2 (A.9.4)
2 2 r

xi
Total Kinetic and Potential Energy
• Total kinetic energy:
 
1 2 2 1 I2
K = Ktrans + Krot = (I1 + mx )θ̇ + m 1 + ẋ2 (A.9.5)
2 2 mr2

I2
• Let λ = 1 + be a constant.
mr2
• Potential energy:

V (x, θ) = mgx sin θ (A.9.6)

• Lagrangian of the system:

 
1 2 2 1 I2
L = (I1 + mx )θ̇ + m 1 + ẋ2 − mgx sin θ (A.9.7)
2 2 mr2

Euler-Lagrange Equations
The Euler-Lagrange equation of the system is:
 
d ∂L ∂L
− = Fi (A.9.8)
dt ∂ q̇ ∂q

Where Fi are the generalized forces or torques, including friction and external input. In
this case, assuming the pivot friction is negligible, the only external force applied to the
system is the torque at the pivot joint τ .

• Resulting equations of motion:

(I1 + mx2 )θ̈ + 2mxẋθ̇ + mgx cos θ = τ (A.9.9)

λẍ − xθ̇2 + g sin θ = 0 (A.9.10)

State-Space Representation

With state variables x1 = x, x2 = ẋ, x3 = θ, and x4 = θ̇, the dynamics of the system are
described by:

ẋ1 = x2
mx24 x1 − g sin x3
ẋ2 =
λ
ẋ3 = x4
τ − 2mx1 x2 x3 − mgx1 cos x3
ẋ4 =
I + mx21

xii
Linearized Model State-Space Representation
Using small-angle approximations (sin x3 ≈ x3 and cos x3 ≈ 1) and assuming the ball is
at the center (x1 = 0), the linearized state-space equations become:

ẋ1 = x2
−gx3
ẋ2 =
λ
ẋ3 = x4
τ − mgx1
ẋ4 =
I
The state-space representation of the linearized system is:
   
0 1 0 0 0
 −g   
 0 0 0 0
   
ẋ = 
 λ  x +  τ (A.9.11)
0
  
 0 0 0 1
−mg 1
   
0 0 0
I I
The control objective is to regulate the position of the ball; thus, the output must be
x1 . The output equation is:
h i
y= 1 0 0 0 x (A.9.12)

xiii
Appendix B

Modern Control System

B.1 Modern Control System Tests

B.1.1 Test 1 (2024 G.C.)
Part I:-Conceptual Question(3 points)
1. Define stabilizability and detectability for a dynamic system and the requirements
that the system dynamics should have to be detectable and stabilizable. (3 points)
Part II:- Workout problem(12 points)
1. A cable reel dynamic system has the following overall transfer function,
2
G(s) = 2 (B.1.1)
s + βs + 21

Cable Reel System

Gdc ( s ) Gr ( s )
U(s) Y(s)

Figure B.1: Cable Reel System Block Diagram

For the cable reel system

(a) Derive a state space representation of the system.(3 points)
(b) Determine the range of β for which system is stable. (3 points)
(c) Select a single value of β from the possible β values you have determined in
part 1b and determine the zero initial condition step response of the resulting
system. (3 points)
(d) Is the system completely controllable and completely observable with the value
of β you have selected in part 1c? Why? Justify your answer mathematically.
(3 points)

xiv
B.1.2 Test 2 (2024 G.C.)
Part I:-Conceptual Question(3 points)

1. Define stabilizability and detectability for a dynamic system and the requirements
that the system dynamics should have to be detectable and stabilizable. (3 points)

Part II:- Workout problem(12 points)

1. Consider the following dynamic system,

s+α
G(s) = (B.1.2)
s2 + 5s + 6
(a) Derive diagonal normal form state space representation for the dynamic sys-
tem.(3 points)
(b) Determine the range of α for which system is completely observable. (3
points)
(c) Select single value of α from the possible α values you have determined in part
1b and determine state transition matrix of the system. (3 points)
(d) Does the value ofα affect the stability of the system? Justify your answer
mathematically. (3 points)

B.1.3 Test 3 (2020 G.C.)

Part I:-Conceptual Questions(5 points)

1. Show that the solution of the state equation will be changed with a similarity
transformation. (2.5 points)

2. Derive an expression for the solution of the state space equation.(2.5 points)

Part II:- Calculation(15 points)

Consider the RLC Circuit, with system parameters L = 0.1H, C = 0.1F , and R = 1Ω.

R L

vs C vo

Figure B.2: Series RLC Circuit

xv
 1. If the designer aims to regulate the voltage across the capacitor.

(a) Derive a state space representation of the system.(2.5 points)

(b) Find the eigenvalues and eigenvectors of the system.(1.5 points)
(c) Determine whether the system is stable or not.(0.5 points)
(d) Determine the state transition matrix of the system.(2.5 points)
(e) Determine zero input response of the system for initial condition x(0) =
h i⊤
0 1 .(2.5 points)
(f) Determine the zero state response of the system under step u = e−2t .(2.5
points)
(g) What is the maximum value of voltage the capacitor can be charged to? (0.5
points)

2. If the output is the voltage across the inductor, determine the output state space
representation of the system. (2 points)

3. Compare the result on 2 with the result on 1a and state their difference. (0.5 points)

B.1.4 Test 4 (2020 G.C.)

Part I:-Conceptual Question(5 points)
1. Show that the similarity transformation doesn’t change the transfer function of the
system and the solution of of output equation. (2.5 points each)
Part II:- Calculation(15 points)
1. Consider the following simple weight measuring system, with system parameters
k = 20N/m, ms = 1kg, and d = 9N s/m.

u = mo g

ms
x Scale
k d

If the output, which is shown on the scale, is the mass of the object to be measured
kx
(mo = ). (use g = 10m/s2 )
g
(a) Derive a state space representation of the system.(2.5 points)
(b) Find the eigenvalues and eigenvectors of the system.(1.5 points)
(c) Determine whether the system is stable or not.(0.5 points)
(d) Determine the state transition matrix of the system.(2.5 points)

xvi
 (e) Determine zero input response of the system for initial condition x(0) =
h i⊤
1 0 .(2.5 points)
(f) Determine the zero state response of the system under a step input.(2.5 points)

2. If the maximum compression of the spring is 10cm, determine the maximum mass
the system can handle.(0.5 points)

3. If you are the designer and you want to increase the maximum mass the system
can handle 5 times,

• What is your best engineering solution? (0.625 points)

• Why do you select the solution? (0.625 points)
• Determine the eigenvalues for your new system. (0.625 points)
• What do you conclude from the result? (0.625 points)

B.2 Modern Control Systems Final Exams

B.2.1 Final Exam 1 (2020 G.C.)
Part I:-Conceptual Questions(10 points)

1. What does it mean when we say a system is stabilizable or detectable? (5 points)

2. What is optimal control? Give an example of a terminal error minimization control

problem. (5 points)

Part II:- Calculation(40 points)

1. Consider the RLC Circuit, with system parameters L = 0.1H, C = 0.1F and
R = 1Ω. If the output of the system is the voltage across the capacitor.

R L

vs C vo

Figure B.3: Series RLC Circuit

(a) Derive a state space representation of the system.(3 points)

(b) Determine the range of the system parameters (R, L, C) for which the system
is completely controllable.(2 points)
(c) Determine the range of the system parameters (R, L, C) for which the system
is completely observable.(2 points)
(d) Is R = 0 a system detectable? If explain why.(2 points)

xvii
 (e) If C is very large, check for system observability and controllability and state
your conclusion? (2 points)

2. For the following completely controllable and completely observable system, (10
points)
   
0 1 0 0
   
ẋ =  0 1  x + 0 u
   
0
   
−5 −10 −5 1
h i
y= 0 0 1 x (B.2.1)

(a) Determine the feedback gain Kf so that the control law is u = −Kf x, assuming
the desired poles of the closed loop system are at s = −2±j4 and s = −2.(3.5
points)
(b) Design an full order observer gain Ke or Lc such that the eigenvalues of the
observer are located at s = −5 ± j4 and s = −10. (3.5 points)
(c) Draw the block diagram implementation of the full state feedback with full
state observer. (3 points)

3. For the system shown on equation B.2.1 (9 points)

(a) Determine the reduced observer gain matrix Ke or Lc assume that desired
observer poles to be at s = −5, s = −5. (4 points)
(b) Determine the reduced-order state observer dynamics and reduced-order state
observer output equations(open loop system dynamics).(3 points)
(c) Draw the overall block diagram for the above system (for question 3).(2
points)

4. For the dynamic system (10 points)

   
0 1 0
ẋ =  x +  u (B.2.2)
0 1 1
h i
y= 1 0 x (B.2.3)

The performance index J to be optimized is

1 ∞ ⊤
Z
J= (x Qx + u⊤ Ru) dt (B.2.4)
2 0
where
 
1 0
Q= , R = 0.25 (B.2.5)
0 1

(a) Determine the Riccati matrix P. (3 points)

xviii
 (b) Design the linear quadratic Regulator (LQR) feedback controller Kf . (2
points)
(c) The control signal u. (1 points)
(d) The closed loop eigenvalues of the system. (2 points)
(e) Draw the block diagram for the LQR controller. (2 points)

B.2.2 Final Exam 2 (2024 G.C.)

Part I:-Conceptual Questions (16 points)
1. Explain briefly the difference between state space representation and transfer func-
tion representation of dynamic systems.(4 points)
2. Explain briefly the basic difference between the pole placement controller design
technique and the optimal linear quadratic regulator (LQR) controller design tech-
nique. (4 points)
3. Explain briefly what optimal control is, and state the generalized cost function/performance
index and explain the terms in the cost function, giving an individual example for
each term.(8 points)
Part II:- Workout problems (34 points)
Consider an open-loop system with the transfer function:
s
G(s) = 2 (B.2.6)
s + 2s + 2
1. For the system in equation B.2.6. (6 points)
(a) Check the stability of the system.
(b) Check whether the system is completely controllable or not.
(c) Check whether the system is completely observable or not.
2. Identify the values of closed loop poles of the system p1 , and p2 for which the
system in equation B.2.6 meets the following specifications: Mp = 0.1630% and
ts (2%criterion) = 1s. (8 points)
3. Design a state feedback controller based on the poles p1 , p2 , identified in the previous
part for the system in equation B.2.6 . (6 points)
(a) Determine the state feedback gain matrix Kf .
(b) Determine the state feedback control law u.
(c) Draw a block diagram representation of the system with a full state feedback
controller.
4. Design a full state observer based on the poles p1 , p2 identified in the previous par
for the system in equation B.2.6. (7 points)
(a) Determine the state observer gain matrix Lc .
(b) Determine the dynamic equation of the observer.
(c) Determine the state feedback control law u with a full state observer.
(d) Draw a block diagram representation of the full state feedback controller with
full state observer.

xix
 5. Design a reduced state observer based on the poles p1 , p2 identified in the previous
part for the system in equation B.2.6. (7 points)

(a) Determine the reduced state observer gain matrix Lc , that locate the pole of
the reduced order observer at p = p1 + p2 .
(b) Determine the output equation of the observer.
(c) Determine the reduced-order state observer dynamic equation.
(d) Determine the state feedback control law u with full state observer.
(e) Draw a block diagram representation of the full state feedback system con-
troller with reduced state observer.

Use the following table and equations if necessary

ωn2
G2order (s) = (B.2.7)
s2 + 2ξωn s + ωn2

Table B.1: Time domain specification of second order systems

Time domain specification Formula

1 + 0.7ξ
Delay time td =
ωn
π−θ
Rise time tr =
ωd
π
Peak time tp =
ω
 d 
− √ ξπ
1−ξ2
Percentage Peak overshoot (%Mp ) Mp = e
4
Settling time with 2% tolerance ts =
ξωn

xx
B.2.3 Final Exam 3 (2020 G.C.)
Part I:-Conceptual Questions(10 points)

1. What does it mean when we say a system is controllable or observable? (5 points)

2. What is optimal control? Give an example of a terminal error minimization control

problem. (5 points)

Part II:- Calculation(40 points)

1. Consider the following simple mass spring damper system (11 points)

m u

b
x

If the output of the system is the velocity of the mass M .

(a) Derive a state space representation of the system.(3 points)

(b) Determine the range of the system parameters (k, b, m) for which the system
is completely controllable.(2 points)
(c) Determine the range of the system parameters (k, b, m) for which the system
is completely observable.(2 points)
(d) Assume k = 0 and b ̸= 0 is the system detectable? If explain why.(2 points)
(e) If m is very large, check for system observability and controllability and state
your conclusion? (2 points)

2. For the following completely controllable and completely observable system, (10
points)
   
0 1 0 0
   
ẋ =  0 1  x + 0 u
   
0
   
−5 −10 −5 1
h i
y= 0 1 0 x (B.2.8)

(a) Determine the feedback gain Kf so that the control law is u = −Kf x, as-
suming the desired pole of the closed loop system is at s = −2 ± j4 and
s = −2.(3.5 points)
(b) Design an full order observer gain Ke or Lc such that the eigenvalues of the
observer are located at s = −5 ± j4 and s = −10. (3.5 points)

xxi
 (c) Draw the block diagram implementation of the full state feedback with full
state observer. (3 points)

3. For the system shown on equation B.2.8 (9 points)

(a) Determine the reduced observer gain matrix Ke or Lc assume that desired
observer poles to be at s = −5, s = −5. (4 points)
(b) Determine the reduced-order state observer dynamics and reduced-order state
observer output equations(open loop system dynamics).(3 points)
(c) Draw the overall block diagram for the above system (for question 3).(2
points)

4. Consider the Dynamic System (10 points)

   
0 1 0
ẋ =  x +  u (B.2.9)
0 −1 1
h i
y= 1 0 x (B.2.10)

The performance index J to be optimized is

1 ∞ ⊤
Z
J= (x Qx + u⊤ Ru) dt (B.2.11)
2 0
where
 
1 0
Q= , R=1 (B.2.12)
0 1

(a) Determine the Riccati matrix P. (3 points)

(b) Design the LQR feedback controller Kf . (2 points)
(c) The control signal u. (1 points)
(d) The closed loop eigenvalues of the system. (2 points)
(e) Draw the block diagram for the LQR controller. (2 points)

B.2.4 Final Exam 4 (2021 G.C.)

Part I:-Conceptual Questions (16 points)

1. What does it mean when we say a system is stabilizable or detectable? (4 points)

2. What does it mean when we say a system is completely controllable or completely

observable? (4 points)

3. What is optimal control? List and give an example of the most popular optimal
control cost functions (performance indexes). (6 points)

4. What is a state observer/state estimator, and where do we use a state estima-

tor/observer? (2 points)

xxii
Part II:- Calculation (34 points)

Consider the following dynamic system

ẋ1 = x2
ẋ2 = x2 + u
y = x1 (B.2.13)

1. For the system shown on equation (B.2.13) (6 points)

(a) Check whether the system is completely controllable. (3 points)

(b) Check whether the system is completely observable. (3 points)

2. For the system shown on equation (B.2.13) (10 points)

(a) Determine the feedback gain Kf so that the control law is u = −Kf x, assuming
the desired poles of the closed loop system are at s = −4 and s = −1.(3.5
points)
(b) Design an full order observer gain Lc such that the eigenvalues of the observer
are located at s = −4 and s = −1. (3.5 points)
(c) Draw the block diagram implementation of the system with full state feedback
using estimated states. (3 points)

3. For the system shown on equation (B.2.13)(8 points)

(a) Determine the reduced observer gain matrix Lc assume that desired observer
pole is at s = −5. (4 points)
(b) Determine the reduced-order state observer dynamics and reduced-order state
observer output equation(open loop system dynamics).(3 points)
(c) Draw the overall block diagram for the system with reduced state observer.
(1 points)

4. For the system shown on equation (B.2.13) (10 points) If the designer wants to
optimize the performance index J,
1 ∞ ⊤
Z
J= (x Qx + u⊤ Ru) dt
2 0
where
 
1 0
Q= , R=1
0 1

(a) Determine the Riccati matrix P. (3 points)

(b) Design the LQR feedback controller Kf . (2 points)
(c) The control signal u. (1 points)
(d) The closed loop eigenvalues of the system. (2 points)
(e) Draw the block diagram for the system with LQR controller. (2 points)

xxiii
Appendix C

Control Systems Exams

C.1 Control Systems Exams Test

C.1.1 Test 1 (2025 G.C)
Solenoid Actuator Valve System

The 3/2 solenoid actuator valve, depicted in Figure C.1, plays a crucial role in regulating
fluid flow through three ports: one inlet and two outlets. When an electrical current is

Spool Valve
Air gap sping
Seated Position Coil Supply Drain

Plunger

Drain

Figure C.1: Solenoid actuator valve

applied to the solenoid, it generates a magnetic field that actuates the internal mechanism
of the valve. This movement facilitates fluid flow from the inlet to one outlet while
simultaneously blocking the other. Upon deactivation of the current, a spring mechanism
restores the valve to its original position, thereby redirecting the flow to the second
outlet. This efficient operation enables precise control over actuators and cylinders in
various pneumatic and hydraulic applications, establishing the solenoid actuator valve as
a vital component in automation and fluid management systems.

Since the spool valve operates within a fluid medium, which introduces viscous friction
that influences its dynamic behavior. The free body diagram of the system, illustrated
in Figure C.2, elucidates the forces acting on the valve and provides insights into its
dynamics, particularly regarding the impact of fluid resistance on its operation.

xxiv
 R
x
i k
Fem
vs L iL m

d
(a) Electrical Subsystem (b) Mechanical Subsystem

Figure C.2: Solenoid Actuator Valve System

1. For Solenoid Actuator Valve System: (12 points)

(a) Derive the transfer function of the system from the input (source voltage Vs (s))
to the output (Spool Valve position X(s)) in zero-pole-gain form (zpk form).
(3 points)
(b) Derive the transfer function, poles, and zeros of the system. (3 points)
(c) Determine the DC gain of the system. Which parameter affects the DC gain
of the system? What conclusions can you draw from your result? (2 points)
(d) Approximate the system to second order using dominant poles approximation.
(Note:- The DC gain of the original system transfer function (1a) and reduced
order (approximated) system should be equal.) (5 points)
(e) Which system parameter affects the steady-state error (ess ) for a step input?
(2 points)

2. If you want to reduce the settling time by 20%: (8 points)

(a) Which parameter do you adjust and why? (3 points)

(b) What would be the effect on the poles of the system with your suggestion? (3
points)
(c) What would be the effect on peak time (tp )? (2 points)

3. If we use unity negative feedback: (5 points)

(a) Does it affect the steady-state error of the system for a step input? Compare
your results with (1d) and draw a conclusion. (3 points)
(b) Does it affect the DC gain of the system? Compare your results with (1c) and
draw a conclusion. (2 points)

xxv
C.1.2 Test 2 (2025 G.C)
A shock absorber reduces unwanted oscillations in mechanical systems, enhancing sta-
bility and performance. It works by absorbing and dissipating energy from vibrations
through a combination of hydraulic fluid and mechanical components. When a force
is applied, the shock absorber compresses, allowing the fluid to flow through valves,
which dampens the motion. The stiffness and damping coefficients can vary based on
application, design, and specific requirements, enabling tailored performance for different
environments, such as vehicles or industrial machinery.

The 3D model and Mechanical System Model of a typical shock absorber are shown
in Figures C.3a and C.3b, respectively.

moving piston
fluid

SECTION A-A x
spring
d

k m u
A

(b) Mechanical System Model of Shock Ab-

(a) 3D Model of Shock Absorber System sorber

Figure C.3: Shock Absorber System

1. For the Shock absorber: (12 points)

(a) Derive the transfer function, poles, and zeros of the system. (3 points)
(b) Determine the DC gain of the system. Which parameter affects the DC gain
of the system? What conclusions can you draw from your result? (2 points)
(c) Determine the conditions for the system to be undamped, under-damped,
critically damped, and overdamped. How does each scenario influence the
system’s poles and its response to inputs?(5 points)
(d) Which system parameter affects the steady-state error (ess ) for a step input?
(2 points)

2. If you want to reduce the settling time by 20%: (8 points)

(a) Which parameter do you adjust and why? (3 points)

(b) What would be the effect on the poles of the system with your suggestion? (3
points)

xxvi
 (c) What would be the effect on peak time (tp )? (2 points)

3. If we use unity negative feedback: (5 points)

(a) Does it affect the steady-state error of the system for a step input? Compare
your results with (1d) and draw a conclusion. (3 points)
(b) Does it affect the DC gain of the system? Compare your results with (1b) and
draw a conclusion. (2 points)

C.1.3 Test 3 (2025 G.C)

1. Given a transfer function of a dynamic system,
60
G(s) = (C.1.1)
0.5s2 + 2.5s + 3
(a) Draw a block diagram representation of the system.
(b) Is the system Stable? Why? (2 points)
(c) Determine the steady-state error (ess ) of the system for a step input. (2 points)
(d) Determine the settling time, maximum percentage overshoot, and steady-state
error of the system with unity feedback for a step input.(2 points)
(e) For the closed-loop frequency response, determine the resonant frequency, the
magnitude of the system at the resonant frequency, and the phase angle at the
resonant frequency. What implications does your answer have for the closed-
loop time-domain analysis of the system? Relate your answer to part 1d. (4
points)
(f) Determine the response of the closed loop system for r(t) = 2 sin(4t) input. (4
points)
(g) Determine the gain crossover frequency (ωgc ), phase crossover frequency (ωpc ),
gain margin (GM ), and phase margin (P M ) of the open loop transfer function
of the system and interpret the implications of the results. (4 points)
(h) Design a compensator (any compensator, a lead or a lag or PI or PD, or PID)
with proper reasoning for your choice. The compensator should ensure that
the steady-state error for a step input to the specified system (referenced as
C.1.1) is less than 0.02, while achieving a desired phase margin of at least 45◦ .
(10 points)

xxvii
C.2 Control Systems Exams Final
C.2.1 Final Exam 1 (2025 G.C.)
Solenoid Actuator Valve System

The operational parameters of the solenoid actuator valve system in Figure C.1 are given
in Table C.2, including the parameters and voltage gain of the non-inverting operational
amplifier configuration.

Table C.1: System Parameters

Parameter Value
Stiffness Coefficient (K) 100N/m
Number of Turn (N ) 400
Winding Resistance (R) 1Ω
Winding Inductance (L) 0.5H
Spool valve mass (m) 1kg
Damping Friction Coefficient (b) 25N s/m
Electromagnetic Force Fem = N iL

1. For the solenoid actuator valve system shown in Figure C.1; (12 points)

(a) Derive the transfer function of the system from the input (source voltage Vs (s))
to the output (Spool Valve position X(s)) in zero-pole-gain form (zpk form).
(8 points)
(b) Draw a block diagram representation of the system.
(c) Approximate the system to second order using dominant poles approxima-
tion.(4 points)

2. For the reduced order system 1c, (18 points)

(a) Determine the time constants for the electrical and mechanical subsystems.
What conclusions can you draw from your results about part 1a? (5 points)
(b) Determine the DC gain of the system. (2 points)
(c) Determine the damping factor (ξ) and natural frequency (ωn ) of the system.
(5 points)
(d) Is the system Stable? Why? (3 points)
(e) Determine the steady-state error (ess ) of the system for a step input. (2 points)
(f) Determine the rise time, maximum percentage overshoot, and steady-state
error of the system with unity feedback for a step input.(5 points)

xxviii
 (g) For the closed-loop frequency response, determine the resonant frequency, the
magnitude of the system at the resonant frequency, and the phase angle at the
resonant frequency. What implications does your answer have for the closed-
loop time-domain analysis of the system? Relate your answer to part 2f. (6
points)
(h) Determine the response of the closed loop system for r(t) = 2 cos(4t) input.
(4 points)
(i) Determine the gain crossover frequency (ωgc ), phase crossover frequency (ωpc ),
gain margin (GM ), and phase margin (P M ) of the open loop transfer function
of the system and interpret the implications of the results. (7 points)
3. Design a lead compensator to ensure that the steady-state error for a step input is
less than 0.02, while achieving a desired phase margin of at least 45◦ at 20rad/s
frequency for the system you derived in part 1b. (20 points)

C.2.2 Final Exam 2 (2025 G.C.)

DC fan speed control system
The DC fan speed control system integrates a DC motor with a pre-amplifier configured
in a non-inverting manner to precisely regulate the fan speed. This design facilitates an
elevation of the input armature voltage supplied to the DC motor, thereby enhancing its
operational performance and responsiveness.

+ ia
Ra La
+
− +

Rf ωm
Vs Va M

Ri
-
-
(a) Armature Voltage Controlled DC Mo-
tor With Pre-amplifier (b) DC-Fan

Figure C.4: DC Fan Speed Control System

In this configuration, the weight of the fan is considered negligible (Zero load torque),
which contributes to optimal efficiency. The motor is directly coupled to the fan blades,
ensuring smooth and responsive operation. The pre-amplifier is instrumental in stepping
up the input voltage, allowing for refined control over the motor’s speed and improving
the overall effectiveness of the system.

The operational parameters of the DC motor are given in Table C.2, including the pa-
rameters and voltage gain of the non-inverting operational amplifier configuration.

xxix
 Table C.2: DC Motor Parameters

Parameter Value
Back EMF constant (Kb ) 0.001 V-s/rad
Torque constant (Kt ) 0.001 V-s/rad
Winding Resistance (Ra ) 1Ω
Winding Inductance (La ) 0.5 H
Motor Inertia (Jm ) 0.001kg/m2
Windage friction (b) 0.01 N-s
Input resistance (Ri ) 1 kΩ
Feedback Resistance (Rf ) 470 kΩ
 
Rf
Non-Inverting Op-amp voltage gain (Av ) 1+
Ri

1. For the DC fan speed control system shown in Figure C.4; (30 points)

(a) Derive the transfer function of the system from the input (source voltage Vs (s))
to the output (motor speed ωm (s)) in zero-pole-gain form (zpk form). (8
points)
(b) Draw a block diagram representation of the system.
(c) Determine the time constants for the electrical and mechanical subsystems.
What conclusions can you draw from your results about part 1a? (5 points)
(d) Determine the DC gain of the system. (2 points)
(e) Determine the damping factor (ξ) and natural frequency (ωn ) of the system.
(5 points)
(f) Is the system Stable? Why? (3 points)
(g) Determine the steady-state error (ess ) of the system for a step input. (2 points)
(h) Determine the settling time, maximum percentage overshoot, and steady-state
error of the system with unity feedback for a step input.(5 points)

2. Frequency domain analysis,

(a) For the closed-loop frequency response, determine the resonant frequency, the
magnitude of the system at the resonant frequency, and the phase angle at the
resonant frequency. What implications does your answer have for the closed-
loop time-domain analysis of the system? Relate your answer to part 2d. (4
points)
(b) Determine the response of the closed loop system for r(t) = 2 cos(4t) input.
(4 points)
(c) Determine the gain crossover frequency (ωgc ), phase crossover frequency (ωpc ),
gain margin (GM ), and phase margin (P M ) of the open loop transfer function
of the system and interpret the implications of the results. (8 points)

xxx
 3. Design a lead compensator to ensure that the steady-state error for a step input is
less than 0.02, while achieving a desired phase margin of at least 45◦ . (20 points)

C.2.3 Final Exam 3 (2025 G.C.)

Projector Screen Positioning System

Consider a projector screen positioning system in Figure 1.18, using the operational
parameters of the DC motor detailed in Table C.2, which includes specifications and the
voltage gain of the non-inverting operational amplifier configuration.

1. For the Projector Screen Positioning System shown in Figure C.4; (30 points)

(a) Derive the transfer function of the system from the input (source voltage Vs (s))
to the output (motor speed ωm (s)) in zero-pole-gain form (zpk form). (8
points)
(b) Approximate the system to second order using dominant poles approximation.
(4 points)
(c) Determine the DC gain of the system (1a) and the reduced order system (1b).
(2 points)
(d) Determine the steady-state error (ess ) of both systems for a step input. (2
points)

2. For the reduced order system 1b, (18 points)

(a) Determine the time constants for the electrical and mechanical subsystems.
What conclusions can you draw from your results about part 1b? (5 points)
(b) Determine the damping factor (ξ) and natural frequency (ωn ) of the reduced
order system. (5 points)
(c) Is the reduced order system Stable? Why? (3 points)
(d) Determine the settling time, maximum percentage overshoot, and steady-state
error of the reduced-order system with unity feedback for a step input.(5
points)
(e) For the closed-loop frequency response, determine the resonant frequency, the
magnitude of the system at the resonant frequency, and the phase angle at the
resonant frequency. What implications does your answer have for the closed-
loop time-domain analysis of the reduced-order system? Relate your answer
to part 2d. (4 points)
(f) Determine the gain crossover frequency (ωgc ), phase crossover frequency (ωpc ),
gain margin (GM ), and phase margin (P M ) of the open loop transfer function
of the reduced order system and interpret the implication of the results. (8
points)

3. Design a lead compensator to ensure that the steady-state error for a step input
is less than 0.02, while achieving a desired phase margin of at least 45◦ for the
reduced-order system in (1b). (20 points)

xxxi
Appendix D

Digital Control Systems

D.1 Digital Control Systems Exams

D.1.1 Final Exam 1
1. For the following analog transfer function
s+2
G(s) =
s2 − 5s + 6
(a) Determine its discrete equivalent for sampling period T=0.2sec. (2 point)
(b) If we use a microcontroller to implement the system, write C code only for
this system part (1 point)
(c) If the range of the input is −12V to 12V , determine the maximum quantization
error (assume that the microcontroller is capable of handling ±12V input) (1
point)
(d) Derive a discrete-time transfer function using a numerical approximation method.
(2 point)
(e) Realize the discrete transfer function using standard programming. (1 point)
2. Mr. Tomas has a plant having open open-loop forward transfer function of
3(s − 2)
G(s) =
s2 + s − 2
The plant operation is required to be controlled by a microcontroller. Assume that
the microcontroller ADC uses an ideal sampler to sample the analog input, DAC
uses zero-order hold to convert the processed digital signal to an Analog signal, and
the system has unity feedback gain. The plant should have a rise time of less than
20 seconds and a settling time of less than 50 seconds. Hint (use 10 samples per
rise time and choose tr and ts that simplify your calculation).
(a) Determine the discrete equivalent of the closed-loop transfer function using
the zero-order hold equivalent technique. (3 marks)
(b) Determine the natural frequency (ωn ), damping factor (ξ), maximum over-
shoot, and steady state error of the system for a unit step input. (2 marks)
(c) Determine whether the system closed-loop transfer function is stable or not
using the Jury stability test. (2 marks)

xxxii
 (d) If the system is not stable the design a controller using the direct Raggazzini
method with one pole of the closed loop transfer function at z = 0.5. (2 marks)

3. Given a sampling period T = 0.5 sec. determine the pulse transfer function for,

R(S) C(S)

Figure D.1: a

R(S) C(S)

Figure D.2: b

Compare and Contrast your result (8 points)

4. Given the discrete-time system represented by the transfer function

Kz
G(z) =
z2 − 1.5z + 0.5
(a) Determine the closed-loop transfer function with unity feedback. (4 points)
(b) Determine the range of values of K that will ensure the stability of the system
in equation 4a. (5 points)
(c) Determine the DC gain of G(z) and the closed-loop system in equation 4a.
Compare your results and provide an analysis. (5 points)

xxxiii
D.1.2 Final Exam 2
1. Batter making system has two inputs called water and flour. The system mixes
the two inputs and results in one output called batter. The discrete-time equation
governing the relation between the batter and flour in the discrete time domain is
as follows. y(k + 2) + 5y(k + 1) + 6y(k) = u1 (k + 1) + 2u1 (k)

(a) Derive a controllable canonical form state space representation for the system
(5 points)
(b) Derive diagonal canonical form state space representation for the system (5
points)
(c) Determine discrete time state space transition matrix. (7 points)
(d) For constant flour input 1kg/min determines the solution for the states vector
in the discrete time domain and the output equation in the discrete time
domain. (7 points)

2. For system in question 1, if the desired location of the poles is at z1 = −0.5 and
z2 = 0.4

(a) Determine state feedback gain Kf using control law. (5 points)

(b) Determine prediction estimator gain Lp for the system. (5 points)
(c) Determine current estimator Lc for the system. (5 points)
(d) Compare the results and state your conclusion. ( 2 points)

3. Derive an expression for diagonal controllable canonical form, assuming that the
system has distinct poles p1 , p2 , ... pn . (5 points)

4. Derive an expression for the state transition matrix using the z-transform method.
( 4 points)

D.1.3 Final Exam 3

1. Given a dynamic system,

y(k + 2) + 5y(k + 1) + 6y(k) = u(k + 1) + u(k) (D.1.1)

(a) Derive the discrete-time state-space representation of the system. (5 points)

(b) Is the system stable? Why? (2 points)
(c) Determine the discrete-time state transition matrix (Φ(k)). (6 points)
(d) Determine the zero input response of the system (yzinp (k)) for initial condition
h i⊤
xo = 1 0 . (6 points)
(e) Determine the zero initial condition response of the system (yzint (k)) for step
input. (7 points)

xxxiv
2. For the system derived in 1a,

(a) Determine the state feedback gain Kf that locates the poles of the discrete-
time system at s1 = −3 and s2 = −0.5. (5 points)
(b) Determine the predictive estimator gain Lp that locates the poles of the esti-
mator at s1 = −3 and s2 = −0.5. (5 points)
(c) Determine the current estimator gain Lc that locates the poles of the estimator
at s1 = −3 and s2 = −0.5. (6 points)
(d) Compare the results from 2b and 2c and derive a conclusion. (3 points)

3. If a PID controller is used to control the plant derived in 1a. Determine the
discrete-time equivalent of the controller. (5 points)

xxxv
"Strive for perfection in everything you do. Take the best that exists
and make it better. When it does not exist, design it."
-Sir Henry Royce, English engineer and car designer.

Remember always! We are yesterday, and you are tomorrow; work hard before you end
up being yesterday!
Good luck and be happy being an engineer!

I am very happy to address or clarify any confusing concepts or anything that feels
wrong. If you feel so, contact me using email: abebe.daniel7@[Link]
Daniel Abebe Beyene (Science Ambassador, MSc in Control Engineering)
Lecturer at AASTU, Electro-mechanical Engineering Department
Thank you very much!!!

xxxvi
 References

[1] K. Ogata, Modern control engineering. Prentice hall, 2010.

[2] R. C. Dorf and R. H. Bishop, Modern Control Systems. Addison-Wesley, 1998.
[3] C. C. Remsing, “Am3. 2-linear control,” Lecture Notes, South Africa, 2006.
[4] B. Kisacanin and G. C. Agarwal, Linear Control Systems: with solved problems and
MATLAB examples. Springer Science & Business Media, 2012.
[5] R. L. Williams, D. A. Lawrence, et al., Linear state-space control systems. Wiley
Online Library, 2007.
[6] D. Abebe, “Design and implementation of feedback linearization based adaptive
stabilizing controller coupled with fuzzy logic swing-up for pendulum on a cart,”
Addis Ababa Institute of Technology, 2019.
[7] S. Özkan, T. Kara, and M. Arıcı, “Modelling, simulation and control of quadruple
tank process,” in 2017 10th International Conference on Electrical and Electronics
Engineering (ELECO), IEEE, 2017, pp. 866–870.
[8] P. Paraskevopoulos, Modern Control Engineering. Marcel Dekker, Inc., 2004.
[9] R. E. Kalman, “A new approach to linear filtering and prediction problems,” 1960.
[10] T. Point. “Time domain specifications.” (2020), [Online]. Available: https : / /
[Link]/control_systemscontrol_systems_time_domain_
[Link].
[11] K. M. Mahala, P. Gadkari, and A. Deb, “Mathematical models for designing vehicles
for ride comfort,” in ICORD 09: Proceedings of the 2nd International Conference
on Research into Design, Bangalore, India, 2009.
[12] U. A. Bakshi and M. V. Bakshi, Modern control theory. Technical Publications,
2020.
[13] N. S. Nise, Control systems engineering. John Wiley & Sons, 2020.
[14] J.-J. E. Slotine, W. Li, et al., Applied nonlinear control. Prentice hall Englewood
Cliffs, NJ, 1991, vol. 199.
[15] R. Paz, Analog computing technique, University of Illinois, 2020.
[16] C. A. Kluever, Dynamic systems: modeling, simulation, and control. John Wiley &
Sons, 2020.

xxxvii

View publication stats