Automatic Flight Control System

Classical approach and modern control perspective

Said D. Jenie and Agus Budiyono

Department of Aeronautics and Astronautics, ITB
Jl. Ganesha 10
Bandung 40132
Indonesia
Phone: +62-22-250-4529
Fax: +62-22-253-4164
Email: agus.budiyono@ae.itb.ac.id
Copyright °c Bandung Institute of Technology

January 12, 2006

2

Abstract
The document is used as a lecture note for the graduate course on flight
control system at the Malaysian Institute of Aviation Technology (MIAT),
Kuala Lumpur, Malaysia. Some parts of the document represent an en-
hanced version of the material given as an elective undergraduate course
and a graduate course in optimal control engineering in the Department
of Aeronautics and Astronautics at ITB, Bandung, Indonesia. Singgih
S Wibowo helped prepare the typesetting of the formulas and check the
MATLAB programs. The constructive input and feedbacks from students
are also gratefully acknowledged.
Contents

I Classical Approach 9

1 Introduction 11
1.1 Types of Automatic Control System . . . . . . . . . . . . . . . . 11
1.1.1 AFCS as the trimmed flight holding system . . . . . . . . 11
1.1.2 AFCS as the stability augmentation system of the aircraft 13
1.1.3 AFCS as the command augmentation system of the aircraft 14
1.1.4 AFCS as the stability provider and command optimizer . 16
1.2 Elements of Automatic Flight Control System . . . . . . . . . . . 17
1.2.1 Front-end interface of flight control system . . . . . . . . 17
1.2.2 Back-end interface of flight control system . . . . . . . . . 19
1.2.3 Information processing system . . . . . . . . . . . . . . . 20
1.2.4 Control Mechanism System . . . . . . . . . . . . . . . . . 21

2 Autopilot System 27
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.2 Working Principle of Autopilot System . . . . . . . . . . . . . . . 27
2.3 Longitudinal Autopilot . . . . . . . . . . . . . . . . . . . . . . . . 34
2.3.1 Pitch Attitude Hold System . . . . . . . . . . . . . . . . . 34
2.3.2 Speed Hold System . . . . . . . . . . . . . . . . . . . . . . 44
2.3.3 Altitude Hold System . . . . . . . . . . . . . . . . . . . . 53
2.4 Lateral-Directional Autopilot . . . . . . . . . . . . . . . . . . . . 69
2.4.1 Bank Angle Hold (Wing Leveler System) . . . . . . . . . 69
2.4.2 Heading Hold System . . . . . . . . . . . . . . . . . . . . 77
2.4.3 VOR-Hold System . . . . . . . . . . . . . . . . . . . . . . 84

3 Stability Augmentation System 93

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
3.2 Working Principle of Stability Augmentation System . . . . . . . 93
3.3 Longitudinal Stability Augmentation System . . . . . . . . . . . 94
3.3.1 Pitch Damper System . . . . . . . . . . . . . . . . . . . . 95
3.3.2 Phugoid Damper . . . . . . . . . . . . . . . . . . . . . . . 99
3.4 Lateral-Directional Stability Augmentation System . . . . . . . . 102
3.4.1 The Dutch-roll stability augmentation: Yaw Damper . . . 102

3
4 CONTENTS

II Modern Approach 115

4 Introduction to optimal control 117

4.1 Some Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . 117

4.2 Linear Systems. . . . . . . . . . . . . . . . . . . . . . . . . . . 119
4.2.1 Controllability . . . . . . . . . . . . . . . . . . . . . . . . 120
4.2.2 Conventions for Derivatives . . . . . . . . . . . . . . . . . 121
4.2.3 Function Minimization . . . . . . . . . . . . . . . . . . . . 123
4.3 Constrained Problems . . . . . . . . . . . . . . . . . . . . . . 126
4.3.1 Elimination . . . . . . . . . . . . . . . . . . . . . . . . . 127
4.3.2 Method of Lagrange . . . . . . . . . . . . . . . . . . . . . 127
4.4 Inequality Constraints . . . . . . . . . . . . . . . . . . . . . . . . 129
4.5 Sensitivity of Cost to Constraint Variations . . . . . . . . . . . . 131
4.6 Dynamic Programming . . . . . . . . . . . . . . . . . . . . . . . . 132

5 Discrete time Optimal Control 137

5.1 Higher Dimension Control Problems . . . . . . . . . . . . . . . . 137

5.2 Discrete time optimal control problem . . . . . . . . . . . . . . . 142

List of Figures

1.1 Autopilot Control System Diagram: Example of Pitch Channel . 12

1.2 Stability Augmentation System Diagram. Example of Pitch Chan-
nel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.3 Command Augmentation System Diagram. Example of Pitch
Channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.4 The Stability Provider and Control Power Optimizer Control Sys-
tem Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.5 The automatic and manual flight control loop . . . . . . . . . . . 18
1.6 The aicraft cockpit as the interface between controller and con-
trolled systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.7 Mechanical Control System: example of longitudinal channel . . 22
1.8 Hydraulically power assisted control system: example of longitu-
dinal channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.9 Hydromechanical Control System . . . . . . . . . . . . . . . . . . 24
1.10 Electrohydromechanical Control System . . . . . . . . . . . . . . 25
1.11 Electrohydraulic Control System . . . . . . . . . . . . . . . . . . 26

2.1 SAS as an inner loop of aircraft autopilot system . . . . . . . . . 28

2.2 CN235-100 Autopilot control system APS-65 . . . . . . . . . . . 29
2.3 CN235-100 Autopilot control system APS-65 . . . . . . . . . . . 31
2.4 The location of auto pilot system components in the standard
control diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.5 Functional diagram of pitch attitude hold system . . . . . . . . . 35
2.6 Pitch attitude hold system . . . . . . . . . . . . . . . . . . . . . . 36
2.7 N250-100 aircraft— prototype 2 Krincing Wesi— manufactured by
the Indonesian Aerospace Inc. . . . . . . . . . . . . . . . . . . . . 38
2.8 Root locus of Pitch attitude hold system for N250-100 PA-2 aircraft 41
2.9 Root locus of Pitch attitude hold system for N250-100 PA-2
aircraft— enlarged to show the phugoid mode . . . . . . . . . . . 42
2.10 Time response of θ(t) due to step θref = 5o with and without
pitch attitude hold system. . . . . . . . . . . . . . . . . . . . . . 43
2.11 Time response for u(t) and α(t) for Kct = −8.9695 . . . . . . . . 44
2.12 Speed hold system functional diagram . . . . . . . . . . . . . . . 45
2.13 Mathematical diagram of speed hold system . . . . . . . . . . . . 46
2.14 Root locus diagram for the speed hold system of N250-100 PA-2 50

5
6 LIST OF FIGURES

2.15 Root locus diagram for the speed hold system of N250-100 PA-
2—zoomed around the pitch oscillation and phugoid modes . . . . 51
2.16 Time response of u(t) to maintain uref = 1 with and without the
speed hold system, K ∗ = 17.3838 . . . . . . . . . . . . . . . . . . 53
2.17 Time response of α(t) and θ(t) to maintain uref = 1, with the
speed hold gain K ∗ = 17.3838 . . . . . . . . . . . . . . . . . . . . 54
2.18 Tail air brake (Fokker F-100/70) . . . . . . . . . . . . . . . . . . 55
2.19 Outer-wing air brake (Airbus A-320/319/321) . . . . . . . . . . . 56
2.20 Functional diagram of altitude hold system . . . . . . . . . . . . 57
2.21 Kinematic diagram of aircraft rate of climb . . . . . . . . . . . . 58
2.22 Mathematical diagram of the altitude hold system . . . . . . . . 59
2.23 Root locus for the altitude hold system for the N250-100 with
h −→ δ e feedback with gain Kct < 0 . . . . . . . . . . . . . . . . 60
2.24 Root locus for the altitude hold system for the N250-100 with
h −→ δ e feedback with gain Kct < 0 —zoomed to show the
phugoid mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.25 Altitude hold system with an attitude hold system as the inner
loop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
∗
2.26 Altitude hold system with inner loop having gain Kct . . . . . . 63
2.27 Root locus diagram of the outer loop of the altitude hold system
for N250-100 PA-2 aircraft . . . . . . . . . . . . . . . . . . . . . . 64
2.28 Root locus diagram of the outer loop of the altitude hold system
for N250-100 PA-2 aircraft—zommed to show the phugoid mode . 65
2.29 Time response of h(t) for an input of href = 1 from a system
with and without altitude hold for N250-100 . . . . . . . . . . . . 66
2.30 Altitude hold system with forward acceleration ax as the inner
loop feedback . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
2.31 Altitude hold system with inner loop using forward acceleration
feedback ax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
2.32 Root locus diagram of inner control loop u̇ −→ δ e for N250-100
altitude hold system . . . . . . . . . . . . . . . . . . . . . . . . . 69
2.33 Root locus diagram of inner control loop u̇ −→ δ e for N250-100
altitude hold system—zoomed around the phugoid mode . . . . . 70
2.34 Root locus diagram of outer control loop u̇ −→ δ e for N250-100
altitude hold system . . . . . . . . . . . . . . . . . . . . . . . . . 71
2.35 Time response of h(t) for N250-100 altitude hold system with the
inner loop of ax −→ δ e feedback . . . . . . . . . . . . . . . . . . . 72
2.36 Functional diagram of Bank angle hold system . . . . . . . . . . 73
2.37 Mathematical diagram of bank hold system . . . . . . . . . . . . 74
2.38 Root locus diagram of the bank hold system of N250-100 aircraft
for a number of τ s values . . . . . . . . . . . . . . . . . . . . . . 76
2.39 Time response of ϕ(t) for a = 1/τ s = 10, 5, 2 due to an impulse
function input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
2.40 Time response of ϕ(t) for a = 1/τ s = 10, 5, 2 due to a step
function input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
2.41 Forces equilibrium during the turn maneuver . . . . . . . . . . . 79
LIST OF FIGURES 7

2.42 Functional diagram of heading hold system . . . . . . . . . . . . 80

2.43 Mathematical diagram of heading hold system . . . . . . . . . . 81
2.44 Heading hold system with bank angle hold as an inner loop . . . 82
2.45 Root locus diagram of the heading hold system of N250-100 air-
craft for a number of a = 1/τ s values . . . . . . . . . . . . . . . . 83
2.46 Time response of ψ(t) and ϕ(t) of the heading hold system of
N250-100 aircraft . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
2.47 Effect of wind to the aircraft flight path . . . . . . . . . . . . . . 85
2.48 VOR guidance path geometry . . . . . . . . . . . . . . . . . . . . 86
2.49 The functional diagram of VOR-hold guidance-control system . . 87
2.50 The mathematical diagram of the VOR-hold guidance-control
system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
2.51 Navigation and guidance system: VOR hold . . . . . . . . . . . . 89
2.52 Root locus diagram for the VOR offset λ(s) with respect to the
bearing Ψref of the N250-100 aircraft . . . . . . . . . . . . . . . 90
2.53 Time response of λ(t) of the VOR Hold system of the N250-100
aircraft with gains of: kλ∗ = −1.1152 (guidance loop), kψi ∗
=
4.5943 (outer loop control) and ki = 8.9344 (inner loop control) . 91

3.1 The functional diagram of the pitch damper system . . . . . . . 94

3.2 The mathematical diagram of the pitch damper system . . . . . 95
3.3 Root locus of the inner control loop: pitch damper system of
N250-100 PA-2 at cruise condition with V = 250 KIAS and h =
150000 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
3.4 Root locus of the inner control loop: pitch damper system of
N250-100 PA-2 at cruise condition with V = 250 KIAS and h =
150000 − enlarged around the phugoid mode . . . . . . . . . . . . 98
3.5 Time response of θ(t) and q(t) with and without pitch damper of
N250-100 aircraft . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
3.6 The mathematical diagram of the pitch attitude hold with the
pitch damper as the inner loop . . . . . . . . . . . . . . . . . . . 100
3.7 Root locus of outer control loop: phugoid damper of N250-100
PA2 at cruise condition . . . . . . . . . . . . . . . . . . . . . . . 101
3.8 Root locus of outer control loop: phugoid damper of N250-100
PA2 at cruise condition—enlarged around phugoid mode . . . . . 102
3.9 Time response of θ(t) of the phugoid damper with Kθ∗ = 3.7979
and Kq∗ = −0.2432 . . . . . . . . . . . . . . . . . . . . . . . . . . 104
3.10 The functional diagram of the yaw damper system with r −→ δr
feedback . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
3.11 The mathematical diagram of the yaw damper system with r −→
δr feedback . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
3.12 Root locus of the yaw damper system with r −→ δ r feedback of
N250-100 aircraft . . . . . . . . . . . . . . . . . . . . . . . . . . 108
3.13 Root locus of the yaw damper system with r −→ δ r feedback of
N250-100 aircraft—enlarged around the dutch-roll mode . . . . . . 109
3.14 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
8 LIST OF FIGURES

3.15 Time response of ψ̇(t), β(t) and ψ(t) of the N250-100 PA2 equipped
with the yaw damper . . . . . . . . . . . . . . . . . . . . . . . . . 113
3.16 Phase portrait of ϕ(t) vs p(t) for three cases: no yaw damper,
with yaw damper and yaw damper+wash-out . . . . . . . . . . . 114

4.1 Minimum of a cost function J(x) at a stationary point . . . . . . 123

4.2 Minimum of J(x) at the boundary . . . . . . . . . . . . . . . . . 124
4.3 Minimum of J(x) at a corner . . . . . . . . . . . . . . . . . . . . 124
4.4 Constrained minimum vs unconstrained minimum . . . . . . . . 128
4.5 Inequality constraints . . . . . . . . . . . . . . . . . . . . . . . . 130
4.6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
4.7 Principle of Optimality . . . . . . . . . . . . . . . . . . . . . . . . 132
4.8 Multistage decision example . . . . . . . . . . . . . . . . . . . . . 133
4.9 Flight Planning application . . . . . . . . . . . . . . . . . . . . . 134
4.10 One dimensional scalar state problem . . . . . . . . . . . . . . . 135

5.1 Discrete grid of x and t . . . . . . . . . . . . . . . . . . . . . . . 138

5.2 Two-dimensional array of states . . . . . . . . . . . . . . . . . . . 140
5.3 Double interpolation of Cost function . . . . . . . . . . . . . . . 141
5.4 Interpolation in the grid of u0 s . . . . . . . . . . . . . . . . . . . 143
 Part I

Classical Approach

9
Chapter 1

Introduction

The content of the book is centered on the discussion of automatic control

without the pilot in the control loop. The role of the automatic flight control
system is to support the pilot in performing his job as the steerer and the mission
executor so as to reduce the pilot’s load. As a steerer, the pilot has two main
tasks which are controlling and guiding the aircraft. The control is performed in
order to maintain the aircraft at the desired equilibrium flight attitude, whereas
the guidance is the task to bring the aircraft from a certain equilibrium state to
another equilibrium state. As a mission executor, the pilot’s task is dependent
on the type of the aircraft’s mission. For instance, for a fighter aircraft, the
pilot has the tasks to find and to investigate the target and then to aim and
shoot it after the process of search and investigation has been determined.
The Automatic Flight Control System (AFCS) is designed to ease the pi-
lot in performing the above tasks in such a way that his physical as well as
psychological load can be reduced.

1.1 Types of Automatic Control System

Based on the task’s level of difficulty, the Automatic Flight Control System can
be categorized into four different types:

1. AFCS as the trimmed flight holding system

2. AFCS as the stability augmentation system
3. AFCS as the command augmentation system
4. AFCS as the stability maker and command optimization

1.1.1 AFCS as the trimmed flight holding system

This type of automatic control system is commonly known as the auto-pilot
which is the abreviation of the automatic-pilot (AP). The AP system has the

11
12 CHAPTER 1. INTRODUCTION

task to take over some parts of the pilot’s routine tasks. The delegated tasks
to the AP system are typically easy and repeated tasks. The types of this AP
system, for example, are:

• Flight condition and configuration holding system, or also known as Hold-

Systems such as speed hold, altitude hold, attitude hold and directional
hold

• Flight trajectory holding system, or usually known as Guidance Systems

such as autoland system

This autopilot system does not work continuously but only at a certain pe-
riod of time. The AP system can be activated by turning the AP switch and
deactivated by over-riding the system through the movement of the control ma-
nipulator. Therefore the characteristic of the AP system is of limited authority.
Fig. 1.1 shows the functional diagram of the autopilot system used in the pitch
longitudinal channel of an aircraft.

Autopilot
Switch Basic Control System

AP motor

Autopilot
Loop

APC Sensor

Autopilot Computer Motion sensor Aircraft Motion

Figure 1.1: Autopilot Control System Diagram: Example of Pitch Channel

Note that by determining the desired attitude, the pilot can then press the
autopilot switch in order that the reference attitude can be obtained and main-
tained. During the period in which the AP is working the pilot does not need
to grasp the controller stick. To deactivate the autopilot, the pilot just need to
slightly move the control stick to cut the AP control circuit. The AP system
can work well if the aircraft has a good stability characteristic.
1.1. TYPES OF AUTOMATIC CONTROL SYSTEM 13

1.1.2 AFCS as the stability augmentation system of the

aircraft
The type of automatic flight control system that adds stability to the aircraft
is usually called the Stability Augmentation System or SAS. This type of auto-
matic flight control system improves the stability of an aircraft at certain flight
configurations and conditions within the flight envelope. For conventional air-
crafts, the stability augmentation will be needed during the flight at low speed
and low altitude for instance during landing or approach. The control opti-
mization of typical aircrafts are conducted only at a certain flight configuration
such as cruise configuration. This makes the aircraft stability at other flight
configurations namely approach, landing or other special configurations tend to
deteriorate. The stability augmentation is therefore necessary for those configu-
rations. The stability augmentation can be achieved by increasing the damping
ratio of the existing aerodynamic damping ratio through the application of feed-
back control system.

Basic Control System

SAS

SAS
Loop

Sensor

Motion sensor Aircraft Motion

Figure 1.2: Stability Augmentation System Diagram. Example of Pitch Channel

The types of the SAS, for example, are:

• Damping ratio augmentation system such as pitch damper, yaw damper

and roll damper

• Dynamic compensation supplier system such as wing leveler and turn co-
14 CHAPTER 1. INTRODUCTION

ordinator

Fig.1.2 shows the example of pitch damper SAS implemented for the air-
craft pitch longitudinal channel. Note that the SAS signal comes out of FCC
(Flight Control Computer) which processes the logic of stability augmentation.
This signal directly enters the ECU and is combined with the command signal
from the pilot to move the elevator. The SAS signal coupled with the aircraft
dynamics will improve the pitch damping ratio such that the aircraft dynamics
is more stable.
It can be observed that the SAS is different from AP in some ways. In the AP
system, the output from the AP computer is used to move the control stick in
lieu of the pilot input. In the SAS system, the output from the SAS computer is
entered into the ECU which forms a closed loop in order to increase the stability
of the aircraft. Thus, the SAS system will keep working even though there is an
input command from the pilot. Whereas in the AP system, the AP loop will be
automatically off once the pilot moves the controller stick. The SAS therefore
has higher level of authority compared to AP system. The SAS system is called
the flight control system with partial authority. To deactivate the SAS, the pilot
can turn the SAS switch.

1.1.3 AFCS as the command augmentation system of the

aircraft
The automatic flight control system with this type of task is commonly called
Command Augmentation Systems or CAS. This system adds the power of input
command of the pilot by processing the input command and the generated
aircraft motion to optimize the input command to the aerodynamic control
surface. The working principle of this system can be likened to that of the
power steering of the ground vehicle.
The types of this CAS system, for example, are:

• Pitch oriented flight control system

• Roll oriented flight control system
• Yaw oriented flight control system

Fig. 1.3 shows the example of CAS system for the pitch oriented column
steering. From the diagram, it is evident that input command from the con-
troller stick is processed to follow the pilot’s desired pitch angle. The command
signal is then corrected by the actual pitch angle and is processed and sent
through the ECU to the ECHP (electronically controlled hydraulically pow-
ered) actuator. It can also be inferred from the diagram that the pilot’s desired
pitch angle can effectively be achieved by an appropriate control stick input.
In summary, the differentiating features of the CAS, SAS and AP are:

• CAS — reacts due to control stick input and results in the desired orienta-
tion. If the pilot does not move the control stick, CAS is not operating
1.1. TYPES OF AUTOMATIC CONTROL SYSTEM 15

Basic Control System

Comparator
CAS
θ comm CAS
Loop

Sensor

Motion sensor Aircraft Motion

Figure 1.3: Command Augmentation System Diagram. Example of Pitch Chan-

nel

• SAS — reacts continuously regardless the motion of the controller stick.

When the SAS is operating, the stability of the aircraft is increased.

• AP — is operating in the condition that the control stick is not moved.

When AP is working, the aicraft will maintain its trimmed condition as
desired by the pilot.

From the perspective of control circuit, the following feature distinguishes

CAS, SAS and AP:

• CAS — the circuit is closed through the Flight Control Computer at the
junction point of controller stick and output from the aircraft motion
sensor

• SAS — the circuit is closed through the Flight Control Computer directly
to the actuator

• AP — the circuit is closed by the motion of the AP electromotor at the

controller stick

From the above comparison, it is clear that the CAS system has a higher
authority than the SAS does because it always reacts to follow the desired
attitude set by the pilot.
16 CHAPTER 1. INTRODUCTION

Basic Control System

Sensor FCC ACT

Actuator
Stick Motion
Computer
sensor

Sensor

Motion sensor Aircraft Motion

Figure 1.4: The Stability Provider and Control Power Optimizer Control System
Diagram

1.1.4 AFCS as the stability provider and command opti-

mizer
This kind of automatic flight control is commonly called Super Augmentation
Flight Control System. This control system is typically used to create an arti-
ficial stability for the class of aircrafts which are statically unstable. The same
system is simultaneously used to optimize the control power through the appli-
cation of control laws provided by the Flight Control Computer. The domain
of this type of control system is electronic and hydraulic. The super augmented
control system is often called electro (opto) — hydraulic flight control system or
Fly by Wire (Light) flight control system which is abbreviated as FbW or FbL.
Fig. 1.4 shows the example of the FbW control system diagram which is used
by the F-16 fighters of the Indonesian Air Forces.
From the diagram, it can be observed that the function of the FCC consists
of combination of three activities, namely:

• Superaugmentation: providing an artificial stability and optimizing the

control power of the aircraft. This subsystem works continuously and can
not be overridden by the pilot.

• Autopilot: taking over some parts of pilot’s routine tasks. If this system
is in operation, the pilot does not need to hold the control stick. This
1.2. ELEMENTS OF AUTOMATIC FLIGHT CONTROL SYSTEM 17

subsystem can be overruled by the pilot by moving the controller stick

• Control Law : governing the optimization of the aircraft motion output

following the desired mission. Using the control law, the aircraft motion
is optimized in such a way that it will not always be the same as the
motion due solely to the input command from the pilot. The control law
is also used for protection or limit of the state variables of the aircraft at
the a certain flight configuration.

The artificial stability provided by the superaugmentation system is the

longitudinal and/or lateral directional static stability. This static stability is
created through the continuos feedback process in such a way that the trimmed
condition of the aircraft is maintained.
It can be observed that this type of flight control system is a control system
with a full authority. Without the availability of this type of system, the aircrafts
that are statically unstable will not be able to fly. Thus, the characteristic of
this control system is flight critical.

1.2 Elements of Automatic Flight Control Sys-

tem
The basic elements in the control information loop are plant (the controlled
system) and the controller. For an aircraft, the controlled system consists of
control apparatus, control surface and the aircraft. Whereas the controller part
consists of three subsystems namely aicraft motion sensor, aircraft motion infor-
mation processor and control command generator. Fig.1.5 shows the functional
diagram of the manual and automatic control system for an aircraft. The di-
agram shows that the primary interface between the controlled and controller
systems can be divided into two parts: front-end interface which is the aircraft
sensory system and back-end interface which is the control command generator.

1.2.1 Front-end interface of flight control system

The front-end interface of the flight control system is the part where the motion
of the aircraft is observed, recorded and displayed in the presentation map or
transmitted in the form of information signal to the aicraft motion information
processor system. In the study of control engineering, the ability to observe the
aircraft motion and to reconstruct it as motion information signal is called ob-
servability. In the aircraft flight control system, the front-end interface element
is located inside the cockpit (flight deck) consisting of front window, side win-
dow, instrumentation display dashboard, pilot vision or aircraft motion sensors
such as pitot-static system, α − β vane and inertial platform to name a few.
18 CHAPTER 1. INTRODUCTION

Controller Aircraft
motion motion

Control apparatus Control surfaces Aircraft

HUMAN CONTROLLER
Command generator Information processor Motion sensor

Pilot’s hands, feet Pilot’s brain Cockpit display,

voices window view, pilot’s
eyes and ears

AUTOMATIC CONTROLLER
Actuator FCC RLG,Accel,Pitot

Information Fluids dynamics/

Command generator
Processor Inertial sensors

Figure 1.5: The automatic and manual flight control loop

1.2. ELEMENTS OF AUTOMATIC FLIGHT CONTROL SYSTEM 19

Front-end
interface

Back-end
interface

Figure 1.6: The aicraft cockpit as the interface between controller and controlled
systems

1.2.2 Back-end interface of flight control system

In the back-end interface, the control command generator is the part of the
system through which pilot command is inputted namely the controller manip-
ulator (stick, steering wheel or pedal) and propulsion controller manipulator
(power lever and condition lever). In the study of control engineering, the abil-
ity to move the control manipulator is called the controllability. In the aircraft
flight control system, the back interface element is located inside the cockpit
(flight deck) and is composed of among others the controller manipulator and
autopilot actuator. See Fig.1.6.

It can be concluded that the cockpit or the flight deck is the flight front-end
and is the most important part of the aircraft since it is in this part that the two
main elements of the control system namely controlled and controller parts are
connected. Particularly for unmanned aerial vehicles such as drones, missiles or
satellites, the two interfaces are combined into the ground control station sys-
tem. The ground control station is typically composed of display system which
comparable to conventional aircraft cockpit and control and navigation interface
where ground pilot can enter control commands or navigation waypoints.
20 CHAPTER 1. INTRODUCTION

1.2.3 Information processing system

Another very critical element of the controller is the information processor sys-
tem. In the manual controller (human controller), this system is represented
by the pilot’s brain supported by the basic information processing computers
to speed up the decision making process. In the automatic controller, the infor-
mation processing element is represented by a Flight Control Computer (FCC).
The FCC works continuously in real time depending on the authority level of
the implemented automatic control system. The software inside the FCC that
manipulates the input of the FCC to be converted to the desired control signal
by the control system is called the control law. The control law can be in the
form of simple instructions which typically used by the autopilot. Some of the
examples of control law are:

FCC
Constant Gain. The FCC repre-
sents a multiplier or an amplifier in out
only. KK

K - constant
yout = Kuin

FCC
Variable Gain (Gain scheduling). in
The FCC works as a modulated out
transformer. The gain can be reg- K
ulated as a function of one or more
η η
parameters.
K = f (ηi )
yout = f (ηi )uin

Robust Gain. The FCC gives the FCC

value of the gain K in the admis-
sible control region. The robust in y
property means that control law yR
will still work when there exists out
some level of uncertainty or param-
eter changes in the plant.
1.2. ELEMENTS OF AUTOMATIC FLIGHT CONTROL SYSTEM 21

FCC
Optimal Gain. The FCC calcu-
lates the optimal gain based on u out
x
a certain predetermined optimiza-
optimal
tion criterion such as minimum
control power, minimum time and
y gain
minimum fuel.
Reconstruction

Adaptive Gain. The FCC deter-

mines the varying gain that adjusts
to the most suitable model of a cer-
tain configuration.
Other than the above control laws, there are many other approaches that are
getting more applications in the automatic flight control design, namely: neural
networks, fuzzy logic, H2 and H∞ control, and passitivity-based control.

1.2.4 Control Mechanism System

The control mechanism is the element of control system as a whole that is also
important. This system allows the transmission of the control command to the
aircraft control surfaces. Based on the physical domain of the control command
transmission, the control mechanism is categorized into the following types:

1. Mechanical Control System

2. Hydro-mechanical Assisted Control System

3. Hydro-mechanical Powered Control System

4. Electro hydro-mechanical Control System

5. Electrohydraulic Control System (Fly-By-Wire)

6. Opto Hydraulic Control System (Fly-By-Light)

The following figures show the illustration of the physical diagram of each
control mechanism type.

Mechanical Control System

A mechanical control system is composed of physical object components with
translation and rotation mechanical domain. Fig.1.7 shows an example of a
mechanical control system in the longitudinal channel. The general components
of this type of control system are among others:

• push-pull rod
22 CHAPTER 1. INTRODUCTION

tension
regulator
cam

cable spring/
damper
mounting controller
stick
spring/ pulley
damper
elevator
pulley pulley
cam mounting

push-pull rod

Figure 1.7: Mechanical Control System: example of longitudinal channel

• pulleys for rotational transmission

• cable tension regulator, springs
• pulley roller cables
• component mountings
• mechanical damper

The dynamic of mechanical control system is influenced by their physical

dynamic properties such as mass, inertia, damping, friction and stiffness. To
some extend, these physical dynamic properties reduce the performance of the
control system as a control command transmission from the pilot to the aircraft.
On the other hand, the greatest advantage of the mechanical control system is
the fact that the pilot through the mechanical control linkage can retain the
feel of a direct connection with the aircraft that he is controlling. Another im-
portant property of the mechanical control system is the ability to imitate the
motion transmission from the controller stick to the control surface by the con-
trol surface motion back to the controller stick. Hence, the mechanical control
system is referred to as a reversible control system.

Hydro-mechanical Assisted Control System

A hydromechanically assisted control system is a control system with mechan-
ical domain where some parts of the subsystem are strengthened by hydraulic
actuators. See Fig.1.8. The typical components of this control system include:

¤ mechanical components (as in mechanical control system)

1.2. ELEMENTS OF AUTOMATIC FLIGHT CONTROL SYSTEM 23

tension
regulator
cam
hydraulic
actuator
controller
stick
cable spring/
damper
mounting
hydraulic pulley
actuator spring/
damper
elevator
pulley pulley
cam mounting

push-pull rod
hydraulic
actuator

Figure 1.8: Hydraulically power assisted control system: example of longitudinal

channel

¤ hydraulic actuators mounted on a number of locations to strenghten me-

chanical parts such as push-pull rod, cam and elevator

Note that the hydraulic actuators are employed to assist the performance of
the existing mechanical components. These actuators do not directly connect
the pilot and the aircraft. Thus if these actuators fail to work or are out of or-
der, the pilot can still control the aircraft through the mechanical linkage which
maintains a direct connection between the pilot and the aircraft. This type of
control system is commonly called Power Assisted Flight Control System. Simi-
lar to a fully mechanical control system, it is also reversible meaning the motion
transmission from the controller stick to the control surface can be reversed.

Hydro-mechanical Powered Control System

The hydromechanically powered control system consists of components with
the domain of mechanical and hydraulic. In this control system, the hydraulic
component directly connects the pilot and the aircraft. Consequently, the pilot
does not have the feel of directly controling the aircraft. The detachment of the
pilot from the aircraft increases the risk of aircraft operation. To overcome this
problem, an artificial feel system is introduced to provide the feel to pilot that
he is directly controlling the aircraft. As a result, the integration of the pilot
and the aircraft can be maintained.
The primary components of this type of control system are:
24 CHAPTER 1. INTRODUCTION

tension
regulator
cam
artificial hydraulic
controller actuator
feel system
stick
hydraulic
power supply

elevator
primary mounting
hydraulic
actuator

Figure 1.9: Hydromechanical Control System

• mechanical components

• hydraulic actuators

• artificial feel system

See Fig.1.9. From direct observation, it is clear that the system is irre-
versible. The motion from the controller stick to the control surface can not be
returned by the elevator motion to the controller stick.

Electro hydro-mechanical Control System

The electrohydromechanical control system consists of components with the
domain of electrical, mechanical and hydraulic. The automatic control systems
as discussed earlier are of this class of control system if the basic control system
domain is hydromechanical. A specific feature of this control system is the
existence of the closed-loop between the control system and the aircraft through
the aircraft motion sensor as shown in Fig.1.10. The electronic circuit closes
the control loop through the aircraft motion sensor, control stick motion sensor
and information processing computer for control strategy. Due to the hydraulic
actuator that directly links the controller stick to the control surface, an artificial
feel system is introduced to provide the feel to pilot that he is directly controlling
the aircraft. As a result, the system is irreversible. In the longitudinal channel,
for instance, the motion from the controller stick to the control surface can not
be returned by the elevator motion to the controller stick.
1.2. ELEMENTS OF AUTOMATIC FLIGHT CONTROL SYSTEM 25

tension
regulator
cam
artificial hydraulic
controller actuator
feel system
stick
primary
hydraulic
actuator

control sensor

elevator
mounting
electrical
command
signal
FCC
electrical
signal

Sensor

Motion sensor Aircraft Motion

Figure 1.10: Electrohydromechanical Control System

Electrohydraulic Control System (Fly-By-Wire)

The electrohydraulic control system consists of components with the domain

of electrical and hydraulic. In this control system, all mechanical components
are eliminated. The basic philosophy behind this type of control system is the
reduction of weight and space while simplifying the installation mechanism.
Fig.1.11 shows the schematic diagram of the electrohydraulic control system,
popularly known as Fly by Wire control system. A distinct feature of this
control system is related to the closed-loop through the aircraft motion sensor,
controller stick sensor and information processing computer. The Fly by Wire
control system is irreversible.
In this electrohydraulic control system, the primary components consist of
processor or intrument blocks namely computer block, sensor block, hydraulic
actuator block and power supply block. Each of the block is connected by elec-
trical transmission cables. The nature of the system consisting the blocks is
thus modular. This modular feature allows for the flexible and simple installa-
tion and repair which makes maintenance job easier and less time consuming.
The feature also enables the design of control system with minimum weight and
space requirement.
26 CHAPTER 1. INTRODUCTION

artificial
controller primary
feel system
stick hydraulic
actuator

information
processor

FCC elevator

control sensor

Motion sensor Aircraft Motion

Figure 1.11: Electrohydraulic Control System

Opto Hydro-Mechanical (Fly-By-Light)

The optohydraulic control system consists of components with the domain of
optronic (opto-electronic) and hydraulic. In this control system, all mechanical
components are eliminated. The working principle and basic philosophy behind
this type of control system is similar to that of Fly by Wire control system.
The distinction is on the data transmission which is done through the optical
domain.
Chapter 2

Autopilot System

2.1 Introduction
An autopilot is flight condition/attitude holding system of an aircraft. In a
number of textbooks, this flight condition/attitude holding system is referred
to as displacement autopilot due to its task to restore the state variable that
it maintains to the original desired value. The autopilot will work well for
the aircraft with good stability characteristic. For the aircraft with marginal
stability, the autopilot can have a better performance if a stability augmentation
system is installed onboard the aircraft as an inner loop of the autopilot system
as illustrated in Fig.2.1. Some types of autopilot system that is commonly
used for conventional transport aircraft are
1. Longitudinal mode

(a) Pitch attitude hold

(b) Speed/Mach number hold
(c) Altitude hold
(d) Glide-slope hold

2. Lateral Directional mode

(a) Heading hold

(b) Bank angle hold or wing leveler
(c) VOR-hold
(d) Turn coordinator

2.2 Working Principle of Autopilot System

Fig.2.2 shows an example of the AP system used for the Indonesian Aerospace
transport aircraft CN-235. The system is of type APS-65 produced by an avionic

27
28 CHAPTER 2. AUTOPILOT SYSTEM

Control linkages Aircraft motion Aircraft flight

Ref. attitude/condition
attitude

AFCS
computer
AP-SAS
SAS loop

SAS sensor

AP loop

AP sensor

Figure 2.1: SAS as an inner loop of aircraft autopilot system

manufacturer Collins. The figure specifically shows the panels inside the cockpit
in conjunction with the operation of an AP system. The associated panels,
indicators and control manipulator are listed in the following figures. Fig.2.3
further shows the components of an AP system mounted on a number of parts of
the aircraft. Refer to Fig.2.2 and 2.3 for the explanation of the working principle
of autopilot system. Fig.2.4 shows the location of the autopilot components in
the standard control diagram.
2.2. WORKING PRINCIPLE OF AUTOPILOT SYSTEM 29

(E) The status of autopilot and

(B) Circuit breaker automatic control system

(G) Attitude Direction

Indicator

(K) Control Manipulator (D) AP switch panel

Figure 2.2: CN235-100 Autopilot control system APS-65

30 CHAPTER 2. AUTOPILOT SYSTEM

(E) The status of autopilot and

automatic control system

(B) Circuit breaker

(D) AP switch panel

(K) Control Manipulator

(G) Attitude Direction

2.2. WORKING PRINCIPLE OF AUTOPILOT SYSTEM 31

(6) Aileron (9) Elevator

servo trim servo
(8) Rudder servo

(7) Elevator
servo

(3) Normal
accelerometer

(2) Slide slip

sensor

(4) Avionic rack

(1) Air data
sensor (5) Autopilot
computer

Figure 2.3: CN235-100 Autopilot control system APS-65

32 CHAPTER 2. AUTOPILOT SYSTEM
2.2. WORKING PRINCIPLE OF AUTOPILOT SYSTEM 33

Basically, an AP system is easy to operate. In this case, a pilot can select

the state variable to be maintained as the reference by bringing the aircraft
to the trimmed condition at the value of the the selected reference variable.
The pilot executes this process of trimming by the use the control manipulator
(K), wheels, control stick or pedal in consideration of the information from the
attitude direction indicator instrument (G). If the trim condition is achieved,
the pilot can then press the AP knob (D) so that the AP system is activated and
working to maintain the trim condition. The trim condition is maintained by the
AP system through the AP actuators (6−8) and AP data processing system (5)
which continuously keeping the aircraft trim condition at the value of the state
34 CHAPTER 2. AUTOPILOT SYSTEM

variable selected as the reference. On condition that the AP system is working,

the pilot can release his hand and feet from the control manipulators, wheels,
and controller stick or pedal. All he needs to do is occasionaly checking up
on the aircraft trim condition as displayed by the Attitude-Direction-Indicator
(ADI). To disengage the AP system, the pilot would just need to grasp controller
stick/wheel/pedal and to displace it a little bit. In this case, the AP control
loop will be automatically disengaged and the pilot regains the full control of
the aircraft. With this characteristic, the autopilot is often referred to as a low
authority system.

+ 4,5 6,7,8,9 CN235

-

1,2,3

Figure 2.4: The location of auto pilot system components in the standard control
diagram

2.3 Longitudinal Autopilot

The following subsections will be focused on the longitudinal autopilot design
for a transport aircraft. The discussion covers the design for pitch attitude
hold, speed hold and altitude hold systems. The elaboration of the altitude
hold system include a number of inner loop feedback designs including the pitch
attitude hold, forward acceleration and compensator integration.

2.3.1 Pitch Attitude Hold System

The block diagram of a pitch altitude hold system is shown in Fig.2.5. Note
that the aircraft is modeled by block (1) with the output of motion states in
the longitudinal mode x̄ = {u, α, θ, q} where the pitch angle θ is sensed by the
vertical gyro, represented by block (2). The output of the vertical gyro is the
signal θ̂m (t) which is then entered to and processed by the autopilot computer,
block(3). The computer receives an input from the pilot in the form of the
desired value of pitch angle as the reference angle to be maintained, θ̂ref (t).
In the autopilot computer (APC) the signal θ̂ref (t) is compared to θ̂m (t) by
2.3. LONGITUDINAL AUTOPILOT 35

using a comparator circuit and the result is amplified by an amplifier circuit

which yields the output signal ε̂θ . The signal ε̂θ is in turn sent to the autopilot
servo motor that moves the steering stick and afterward conveys the signal to
the elevator through the control mechanism, block(4). The elevator deflection
δ e changes the aircraft attitude and the new pitch angle θ(t) is sensed by the
vertical gyro. The whole process in the autopilot loop is repeated until the
desired value ε̂θ = 0 is reached. This condition means that θ̂m (t) = θ̂ref (t) or
the aircraft pitch angle θ(t) is the same as the reference pitch angle θref desired
by the pilot.

AP computer
δT
u (t )
θ ref εˆθ δe
+
Amplifier
α (t )
− θ (t )
q(t )
(4) AP servo to control stick
(3) AP comparator
Æelevator (1) aircraft

θ m (t ) θ (t )

(2) Vertical gyro

Figure 2.5: Functional diagram of pitch attitude hold system

Since the process from block(2) until block(4) is performed in the electrical
domain, it can be considered very fast. Thus the result is the maintenance of
the aircraft pitch angle θ(t) at the value of the reference pitch angle θref desired
by the pilot.
The above functional diagram can be described in the following mathemat-
ical diagram shown in Fig.2.6:

In the above model, the transfer functions of the pitch attitude hold closed
loop system consists of:
1. Aircraft transfer function matrix, for δ T = 0,
¯ Nlong (s)|δr =0 x̄(s)
GA/C (s)¯δ = = (2.1)
T =0 ∆long (s) δ e (s)
36 CHAPTER 2. AUTOPILOT SYSTEM

δT (s)
θ ref εˆθ u( s)
+ [ N long ( s )] α ( s)
Gct ( s )
−
δ e ( s) Δ long ( s ) θ (s)
q( s )
Servo AP aircraft

θ m (t ) θ (t )
Gvg ( s )

Vertical gyro

Figure 2.6: Pitch attitude hold system

2. Vertical gyro transfer function. The time response of the gyro in sensing
the pitch angle θ(t) is considered very fast, thus:

Gvg (s) = Svg ≡ constant (2.2)

3. Autopilot computer transfer function:

εθ (s) = b
b θref − b
θm (s) (2.3)

4. Autopilot servo and control mechanism transfer function. The block is

modeled as a system with time response of τ s representing the first order
lag factor in the transmission of signal ε̂θ to the aircraft elevator.

Kct
Gct (s) = (2.4)
s + 1/τ s
The characteristic polynomial of the pitch attitude hold closed loop system
is given by:

Nδθe (s)
∆cl (s) = ∆long (s) + kθδe (2.5a)
s + 1/τ s
2.3. LONGITUDINAL AUTOPILOT 37

where

kθδe ≡ Svg Kct (2.5b)

Therefore, the characteristic equation of the closed loop system can be writ-
ten as

Nδθe (s)
1 + kθδe =0 (2.6)
∆long (s) [s + 1/τ s ]

The above equation can be written in a simpler form as

1 + kθδe Gol (s) = 0 (2.7)

where

Nδθe (s)
Gol (s) = (2.8a)
∆long (s) [τ s s + 1]
kθδe = Svg Kct τ s (2.8b)

In the case of a conventional aircraft, GOL (s) will have two real zeros and five
poles consisting of two pairs of complex poles of the aircraft and one real pole
associated with the autopilot control mechanism servo. Observing the closed
loop characteristic polynomial, Eq.2.7, note that Kct (servo gain) is the only
gain that can still be changed or varied. The vertical gyro sensitivity Svg and
servo lag factor τ1s are typically of a constant value.
As a case study, the pitch attitude hold system of the N250-100 aircraft—
prototype 2 Krincing Wesi— manufactured by the Indonesian Aerospace Inc. is
presented. See Fig.2.7.
The longitudinal mode transfer function of the N250-100 aircraft during
cruise at the altitude of h = 15000 ft and the speed of Vc = 250 kts is given as
the following (for the elevator control channel):

³ s ´³ s ´
Nδue (s) = Suδe −1 +1 (2.9a)
37.6823 µ4.2634 ¶µ ¶
³ s ´ s s
α
Nδe (s) = Sαδe +1 +1 +1
91.1168 0.009 + j0.0831 0.009 − j0.0831
³ s ´³ s ´
Nδθe (s) = Sθδe +1 +1
1.2860 0.0222
Nδqe (s) = sNδθe (s)

with the static sensitivity coefficient of

38 CHAPTER 2. AUTOPILOT SYSTEM

Figure 2.7: N250-100 aircraft— prototype 2 Krincing Wesi— manufactured by

the Indonesian Aerospace Inc.

Suδe = 3598.817 (2.9b)

Sαδe = −1.992
Sθδe = −8.1443

and the characteristic polynomial of

∆long (s) = s4 + 3.3115s3 + 8.1448s2 + 0.1604s + 0.0554 (2.10a)

having the characteristic roots:

p1,2 = −1.6472 ± j2.317 (2.10b)

p3,4 = −0.0085 ± j0.0824

The N250-100 utilizes ring laser gyroscope (RLG) as its Inertial Navigation
System (INS), therefore the time response can be considered very fast. Hence,
the gain of the vertical gyro, Svg , in sensing the changes of the pitch angle θ(t),
can be taken as:
2.3. LONGITUDINAL AUTOPILOT 39

Svg = 1 (2.11)

Also, since the N250-100 employs Fly by Wire control system in its longi-
tudinal and lateral/directional channels, the domain of the autopilot servo is
therefore electromechanical with a relatively fast time response. As a result,
the time response of the AP servo is taken as

τ s = 0.1 sec (2.12)

Subtituting Eqs.2.9b until 2.12, to Eqs.2.7 and 2.8a, the following closed loop
characteristic polynomial of the pitch altitude hold system can be obtained:

1 + kθδe Gol (s) = 0

where

kθδe = Svg Kct τ s Sθδe (2.13a)

= −0.81443Kct

and

¡ s
¢¡ s
¢
+1 +1
Gol (s) = ³ ´ ³ 1.286´ ³ 0.0222
´³ ´ (2.13b)
s
p1 + 1 ps2 + 1 ps3 +1 s
p4 + 1 (τ s s + 1)

The control algorithm is implemented in MATLAB codes. The programs

for generating root locus diagram and time response for the design of the pitch
attitude hold is presented along with the commentary below.
% Pitch Attitude Hold -- Root locus drawing
close all
% Static sensitivity coefficient:
Sude = 3598.817;
Sade = -1.9920;
Stde = -8.1443;
% coefficients of characteristic polynomial:
D_long = [1 3.3115 8.1448 0.1604 0.0554];
% finding the roots:
p = roots(D_long) ;
% time response of Auto Pilot (AP) servo
ts = 0.1;
% open loop zeros
40 CHAPTER 2. AUTOPILOT SYSTEM

zero = [-1.286 -0.0222]’

% open loop poles
pole = [p(1) p(2) p(3) p(4) -1/ts]’
% open loop gain
Pipole = pole(1)*pole(2)*pole(3)*pole(4)*pole(5);
kol = real(Stde*Pipole/(zero(1)*zero(2)));
% Open loop transfer function
[N_ol,D_ol] = zp2tf(zero,pole,kol);
tf(N_ol,D_ol)
% Vertical gyro gain
Svg = 1;
%Drawing the root locus
sys_ol = zpk(zero,pole,kol);
figure(1);
set(1,’Name’,’Open Loop Root Locus’);
rlocus(sys_ol);
grid on

For a positive value of gain Kct , the feedback gain Kθδe will be negative.
Fig.2.8 shows the root locus diagram of the pitch attitude hold system of the
N250-100 aircraft.
From the root locus, it is evident that for a negative gain the natural fre-
quency of the pitch oscillation mode tends to increase and the damping ratio ζ D
also decreases. Nevertheless, it is still acceptable to the extent that the damping
ratio is still greater than 0.35. It is interesting to note that, the phugoid mode
tends to be more damped and for an even higher value of negative Kθδe this
mode breaks up into two phugoid subsidence modes. This condition is advan-
tageous for maintaining the pitch angle b θ to be at its reference value b
θref since
the aircraft speed is also maintained due to the ”overdamped” phugoid damping
ratio.
As an example, the pitch oscillation damping ratio ζ P O is taken to be 0.35.
The corresponding closed loop characteristics roots are given as follows

ς = 0.35 ⇒ Kct = −8.9695 (2.14)

the roots of the polynomial are:

psv = −11.2777 (2.15)

ppo1 = −0.5390 + j4.1205
ppo2 = −0.5390 − j4.1205

pph1 = −0.9305
pph2 = −0.0254
2.3. LONGITUDINAL AUTOPILOT 41

Root Locus

0.86 0.76 0.64 0.5 0.34 0.16

0.94
4

pitch oscillation phugoid

0.985
2
Imaginary Axis

0
12 10 8 6 4 2

-2
0.985

-4
AP servo
0.94

-6

0.76 0.64 0.5 0.34 0.16

-8 0.86
-12 -10 -8 -6 -4 -2 0 2
Real Axis

Figure 2.8: Root locus of Pitch attitude hold system for N250-100 PA-2 aircraft

The following figure shows the b

θref step response for two kind of cases:

Case 1 Without a feedback

Case 2 With a feedback of b

θ → δ e with Kθδe = −8.9695

% Pitch attitude hold -- Closed loop time response analysis

Kct = 8.9695; Ktde = ts*Svg*Kct;
% Closed loop
N_cl = Ktde*N_ol; D_cl = D_ol + Ktde*N_ol;
[zero_cl,pole_cl,Kcl] = tf2zp(N_cl,D_cl); pole_cl; tf(N_cl,D_cl);
sys_cl = zpk(zero_cl,pole_cl,Kcl);
% Time response for step input
t= 0:0.2:700; u = t*0; u(find(t>1)) = 5/57.3;
% Open loop time response
y_ol = lsim(sys_ol,u,t);
% Closed loop time response
42 CHAPTER 2. AUTOPILOT SYSTEM

Root Locus

0.86 0.76 0.64 0.5 0.34 0.16

0.15
0.94

0.1
phugoid
0.985
0.05
Imaginary Axis

0
35 0.3 0.25 0.2 0.15 0.1 0.05

-0.05
0.985

-0.1

0.94
-0.15

-0.2 0.76 0.64 0.5 0.34 0.16

0.86
-0.35 -0.3 -0.25 -0.2 -0.15 -0.1 -0.05 0 0.05
Real Axis

Figure 2.9: Root locus of Pitch attitude hold system for N250-100 PA-2 aircraft—
enlarged to show the phugoid mode

y_cl = lsim(sys_cl,u,t);
figure(’Name’,’Time Respons Theta’); subplot(211);
plot(t,u*57.3,t,y_ol*57.3,’-m’); ylabel(’open loop teta’); grid on
subplot(212); plot(t,u*57.3,t,y_cl*57.3,’-r’);axis([0 35 0 7]);
ylabel(’closed loop teta’);grid on
% u response
zero_u = [37.6823 -4.2634]’;
kcl_u = Kcl*Sude/Stde*zero(1)*zero(2)/(zero_u(1)*zero_u(2));
[Nu_cl,Du_cl] = zp2tf(zero_u,pole_cl,kcl_u);
[zero_ucl,pole_ucl,Kucl] = tf2zp(Nu_cl,Du_cl) ;
pole_ucl; tf(Nu_cl,Du_cl);sysu_cl = tf(Nu_cl,Du_cl);
yu_cl = lsim(sysu_cl,u,t);
% alpha response
zero_a = [ -91.1168 -0.009+0.0831j -0.009-0.0831j]’
kcl_a = -Kcl*Sade/Stde*zero(1)*zero(2)/real(zero_a(1)*zero_a(2)*zero_a(3))
[Na_cl,Da_cl] = zp2tf(zero_a,pole_cl,kcl_a);tf(Na_cl,Da_cl);
sysa_cl = tf(Na_cl,Da_cl);ya_cl = lsim(sysa_cl,u,t);
2.3. LONGITUDINAL AUTOPILOT 43

figure(’Name’,’Close Loop Time Respons’)

subplot(211);plot(t,yu_cl,’-g’);ylabel(’closed loop u [m/s]’);
axis([0 250 -35 0]);grid on
subplot(212);plot(t,ya_cl*57.3,’-b’);grid on
axis([0 25 -1 4]);ylabel(’closed loop \alpha [deg]’);grid on

It is apparent that for the case with the pitch attitude hold, the value of
θ can be quickly maintained in less than 15 time units, while when the pitch
attitude hold is not present, Kθδe = 0, the time it takes to reach the steady level
of b
θref = 0 is more than 500 time unit or 35 times as along as that of the system
with the pitch attitude hold. Also note that when the pitch attitude hold is off,
there exists an offset angle of about 15o whereas with the pitch attitude hold
the offset angle is only approximately 0.25o . Fig.2.11 shows the time response
of velocity, u(t), and angle of attack, α(t). The bleed-off speed to hold the pitch
angle θ = 5o is about −35 kts while the angle of attack increases by 0.5o .

200
open loop teta

100

-100
0 100 200 300 400 500 600 700

6
closed loop teta

0
0 5 10 15 20 25 30 35

Figure 2.10: Time response of θ(t) due to step θref = 5o with and without pitch
attitude hold system.

This pitch attitude hold system can further be improved by pitch oscillation
stability augmentation system (SAS) which will be discussed in more detail in
44 CHAPTER 2. AUTOPILOT SYSTEM

closed loop u [m/s]

-10

-20

-30

0 50 100 150 200 250

4
closed loop α [deg]

-1
0 5 10 15 20 25

Figure 2.11: Time response for u(t) and α(t) for Kct = −8.9695

the next chapter.

2.3.2 Speed Hold System

In the aircraft application, the speed hold system can be categorized into two
types:

• Speed Hold System: for a low speed aircraft (low subsonic)

• Mach Hold System: for a high speed aircraft (high subsonic to transonic)

Only the speed hold system will be covered in further detail in this course.
Refer to Fig.2.12. The figure illustrates the speed hold system that is typically
used for an aircraft. The aircraft’s airspeed is sensed by the pitot static (2)
and the result is sent to the autopilot computer (3) to be compared with the
reference flight speed (the speed which will be maintained by this speed hold
system). The speed difference b εu will be sent by the AP computer to the engine
control system (power lever to propulsion control mechanism) (4). The result is
a throttle deflection, δ th applied to the aircraft engine (5). The aircraft engine
in turn changes the thrust of the aircraft by δ T . The aircraft (1) will react to the
thrust input δ T and its velocity u(t) will change accordingly. The process then
continues as before and finally stops when the aircraft velocity u(t) has reached
2.3. LONGITUDINAL AUTOPILOT 45

bref (t). In this condition the signal of speed

the value of the reference speed u
difference bεu from the computer will be zero. When the autopilot is working,
the pilot can release the power lever and let the engine control manipulator
moves by itself following the closed loop process of the speed hold autopilot
system. Hence this type of flight speed holding system is commonly called an
Auto Throttle system.

(3) AP computer

u(t )
uˆ ref εˆu δ th δT
+ α (t )
− δe θ (t )
q(t )
(4) engine control (5) engine and (1) aircraft
system propeller

uˆm ( t ) u (t )

(2) Pitot static: velocity sensor

Figure 2.12: Speed hold system functional diagram

The following figure describes the mathematical diagram of the aircraft speed
hold system. This speed hold system has the following transfer functions
1. Aircraft transfer function matrix, for δ e = 0.

¯ Nlong (s)|δr =0 x̄(s)

GA/C (s)¯δe =0 = = (2.16)
∆long (s) δ T (s)
2. Pitot static transfer function. In measuring the flight speed, the pitot is
modeled by first order lag system which is called pitot-lag
1/τ ps
Gps (s) = (2.17)
s + 1/τ ps
3. Autopilot computer transfer function. The comparator computer is mod-
eled by:
b bref − u
εu (s) = u bm (s) (2.18)
46 CHAPTER 2. AUTOPILOT SYSTEM

uˆ ref εˆu δ th ( s ) δ T ( s ) [ N ( s )] u (t )
+
G PC ( s ) GENG ( s )
long α (t )
− Δ long ( s ) θ (t )
q (t )
propulsion control engine
δ e (s) aircraft

uˆm ( t ) u (t )
G PS ( s )

pitot static

Figure 2.13: Mathematical diagram of speed hold system

4. Propulsion control transfer function. For the current existing advanced

propulsion technology, the transfer function can be modeled by first order
lag as given by

1/τ pc
Gpc (s) = Kpc (2.19)
s + 1/τ pc

5. Engine (and propeller) transfer function. For the present technology, com-
pared to the aircraft dynamics, the corresponding transfer function can
also be modeled as a first order lag :

1/τ e
Geng (s) = Ke (2.20)
s + 1/τ e

The closed loop characteristic polynomial of the speed hold system can be
expressed as the following:

NδuT (s)
∆cl (s) = ∆long (s) + kuδT (2.21a)
[s + 1/τ ps ] [s + 1/τ pc ] [s + 1/τ e ]

where

kuδT ≡ Kpc Ke / [τ ps τ pc τ e ] (2.21b)

2.3. LONGITUDINAL AUTOPILOT 47

Thus, the closed loop characteristic equation is given by:

1 + KuδT Gol (s) = 0 (2.22)

where

KuδT = Kpc Ke (2.23a)

NδuT (s)
Gol (s) = (2.23b)
[τ ps s + 1] [τ pc s + 1] [τ e s + 1] ∆long (s)

The speed hold system is usually used during the approach and landing in
order to reduce the work load of the pilot who has been primarily occupied by
the aircraft guidance task.
As a case study, the speed hold system of the N250-100 aircraft manufactured
by the Indonesian Aerospace during the landing configuration will be analyzed.
In this configuration, the aircraft velocity is 125 kias at the altitude of h = 0
from sea level. The transfer functions associated with this configuration are
given by:
³ s ´µ s
¶µ
s
¶
u
NδT (s) = SuδT +1 +1 +1
0.3519 0.8248 + j0.809 0.8248 − j0.809

µ ¶µ ¶
s s
NδαT (s) = SαδT +1 +1
0.0342 + j0.1992 0.0342 − j0.1992

s ³´³ s ´
NδθT (s) = SθδT +1 −1 (2.24a)
0.9093 0.0706
NδqT (s) = sNδθT (s)

with the static sensitivity coefficients as follows,

SuδT = 25.9337 (2.24b)

SαδT = −0.1276
SθδT = 0.1912

and characteristic polynomial:

s4 s3 s2 s
∆long (s) = + + + +1 (2.25a)
0.0588 0.02876 0.03296 0.6164

with the characteristic roots:

p1,2 = −1.0147 ± j0.8304 (2.25b)

p3,4 = −0.0076 ± j0.1848
48 CHAPTER 2. AUTOPILOT SYSTEM

The aircraft velocity is sensed by the the pitot static system, with the transfer
function modeled by a first order lag having a time response of:

τ ps = 0.2 sec (2.26)

The N250-100 propulsion control system uses the Full Authority Digital
Engine Control (FADEC) technology and thus the time response is fast. The
value of the time response is assumed to be

τ pc = 0.1333 (2.27)

To generate thrust, the engine and propeller system of the N250-100 aircraft
employs advanced six bladed propeller yielding a relatively fast time response.
The value of the time response is taken to be

τe = 1 (2.28)

Substituting Eqs.2.24b—2.28 to Eqs.2.22—2.23a, the following equation is ob-

tained

1 + KuδT Gol (s) = 0 (2.29)

where

KuδT = Kpc Ke SuδT (2.30a)

¡ s
¢³ s
´³
s
´
0.3519 + 1 0.8348+j0.809 + 1 0.8348−j0.809 +1
Gol (s) = ¡s ¢¡ s ¢ ³ ´³ ´³ ´³ ´
5 + 1 7.5 + 1 (s + 1) ps1 + 1 ps2 + 1 ps3 + 1 ps4 + 1
(2.30b)

The following figure shows the root locus diagram of the speed hold system of
the N250-100 aircraft for positive KuδT gain. The root locus shown in the figure
is varied as a function of the gain K = KP C Ke . Notice that the pitch oscillation
mode does not change at all since it is constrained by the two complex zero of
NδuT (s) = 0. Nonetheless, the damping ratio of the phugoid mode increases up
to its maximum value before moving to the unstable region.
The MATLAB code implementation for the root locus drawing of the speed
hold system is presented below:
% Static sensitivity coefficient:
Sudt = 25.9337;
Sadt = -0.1276;
Stdt = 0.1912;
% coefficients of characteristic polynomial:
2.3. LONGITUDINAL AUTOPILOT 49

D_long = [1/0.0588 1/0.02876 1/0.03296 1/0.6164 1];

% roots:
p = roots(D_long) ;
% time constants
tps = 0.2; tpc = 0.13333; te = 1;
% open loop zeros
zero = [-0.3519 -0.8348-0.809i -0.8348+0.809i]’
% open loop poles
pole = [p(1) p(2) p(3) p(4) -1/tps -1/tpc -1/te]’
% open loop gain
Pipole = pole(1)*pole(2)*pole(3)*pole(4)*pole(5)*pole(6)*pole(7);
kol = real(Pipole/(zero(1)*zero(2)*zero(3)));
% Open loop transfer function
[N_ol,D_ol] = zp2tf(zero,pole,kol);tf(N_ol,D_ol)
% Drawing the root locus
sys_ol = zpk(zero,pole,kol);
figure(1)
set(1,’Name’,’Open Loop Root Locus’);
rlocus(sys_ol);zoom(3); grid on
To achieve the desired most favorable performance, a point in the root locus
associated with the highest phugoid damping ratio is chosen. Refer to the above
figure, this point is given by

ς max = 0.0898 ⇒ K ∗ = 17.3838

and its corresponding roots of the polynomial characteristics are

psv1,2 = −6.5126 ± j0.2513 (2.31)

ppo1,2 = −0.8657 ± j0.9481
pph1,2 = −0.1729 ± j0.9474
psv3 = −0.4422

Using the above values of poles and gain K, the closed loop characteristic
polynomial of this speed hold system can be written as

∆cl (s, K ∗ ) = s7 + 15.5445s6 + 79.3848s5 + 163.9051s4 (2.32)

+222.9647s3 + 185.1533s2 + 114.5985s + 28.7110

The MATLAB code implementation for the gain selection process and the as-
sociated time response analysis is given below:
% Selecting the gain
[Ku_dt,pcl] = rlocfind(sys_ol);D_cl = poly(pcl);
D_ps = [1 5];N_cl = Ku_dt/5*conv(N_ol,D_ps);
% Preserve the dim of N_cl
N_cl = N_cl(2:length(N_cl));D_cl = D_ol + Ku_dt*N_ol;
50 CHAPTER 2. AUTOPILOT SYSTEM

Root Locus

6 0.86 0.76 0.64 0.5 0.34 0.16

4 0.94
pitch oscillation
prop control system
engine and
2 0.985
propeller
phugoid
Imaginary Axis

10 8 6 4 2
0

-2 0.985
pitot static

-4 0.94

-6 0.86 0.76 0.64 0.5 0.34 0.16

-10 -8 -6 -4 -2 0
Real Axis

Figure 2.14: Root locus diagram for the speed hold system of N250-100 PA-2

[zero_cl,pcl,Kcl] = tf2zp(N_cl,D_cl) ;pcl

tf(N_cl,D_cl)
sys_cl = zpk(zero_cl,pcl,Kcl);
% Time response
t = 0:0.2:700;u = t*0;u(find(t>1)) = 1;
% open loop time response
y_ol = lsim(sys_ol,u,t);
% closed loop time response
V = 250*1.8*1000/(60*60);y_cl = lsim(sys_cl,u,t);
figure(’Name’,’Time Respons’)
subplot(211);plot(t,u,t,y_ol,’-m’);ylabel(’open loop u’);grid on
subplot(212);plot(t,u,t,y_cl,’-r’);axis([0 50 0 2]);
ylabel(’closed loop u’);grid on
% pitch angle response
zero_theta = [-0.9093 0.0706 -5]’;
Pipcl = pcl(1)*pcl(2)*pcl(3)*pcl(4)*pcl(5)*pcl(6)*pcl(7);
kcl_theta = real(Stdt*Pipcl/(zero_theta(1)*zero_theta(2)*zero_theta(3)));
[Ntheta_cl,Dtheta_cl] = zp2tf(zero_theta,pcl,kcl_theta);
2.3. LONGITUDINAL AUTOPILOT 51

Root Locus
1
0.76 0.64 0.5 0.34 0.16
0.86
0.8

engine and
0.6
0.94 propeller
pitch oscillation phugoid
0.4
0.985
0.2
Imaginary Axis

0
1.4 1.2 1 0.8 0.6 0.4 0.2

-0.2
0.985
-0.4
0.94
-0.6

-0.8
0.86
-1 0.76 0.64 0.5 0.34 0.16

-1.4 -1.2 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6

Real Axis

Figure 2.15: Root locus diagram for the speed hold system of N250-100 PA-2—
zoomed around the pitch oscillation and phugoid modes

tf(Ntheta_cl,Dtheta_cl)
systheta_cl = zpk(zero_theta,pcl,kcl_theta);
ytheta_cl = lsim(systheta_cl,u,t);
% angle of attack response
zero_a = [ -0.0342-0.1992j -0.0342+0.1992j -5]’;
kcl_a = real(Sadt*Pipcl/(zero_a(1)*zero_a(2)*zero_a(3)));
[Na_cl,Da_cl] = zp2tf(zero_a,pcl,kcl_a);
tf(Na_cl,Da_cl)
% Closed loop time response
sysa_cl = zpk(zero_a,pcl,kcl_a);
ya_cl = lsim(sysa_cl,u,t);
V = 250*1.8*1000/(60*60);
figure(’Name’,’Close Loop Time Respons’)
subplot(211);plot(t,ytheta_cl*57.3,’-g’);axis([0 50 -60 60]);
ylabel(’closed loop \theta [deg]’);grid on
52 CHAPTER 2. AUTOPILOT SYSTEM

subplot(212);plot(t,ya_cl*57.3,’-b’);grid on
axis([0 50 -40 40]);ylabel(’closed loop \alpha [deg]’);grid on

The time response due to a step input, u(t), can be obtained through the
inverse Laplace transformation of the closed loop transfer function as the fol-
lowing
u (t) = £−1 [Gcl (s)ūref (s)] (2.33a)

where
K ∗ (s + 1/τ ps ) NδαT (s)
Gcl (s) = (2.33b)
∆cl (s, K ∗ )

The following figures show the comparison of the time response u(t) due to
a unit step uref = 1, between the aircraft without speed hold system and one
with speed hold system operated with gain K ∗ . The effectivity of the speed hold
system can be observed from the comparison figures. The speed hold system
allows the achievement of the steady state value of u(t) at 35 time units. While
without the speed hold system, the steady state condition has not been reached
until 350 time units. An offset from the steady state value uref = 1 can be
compensated through the use of a gain adjuster which is governmed by the
Auto Pilot computer of the speed hold system.
The corresponding equations for α(t) and θ(t) can be derived as follows:
∙ ∗ ¸
−1
K (s + 1/τ pl ) NδαT (s)
α (t) = £ ūref (2.34)
∆cl (s, K ∗ )
" #
K ∗ (s + 1/τ pl ) NδθT (s)
θ (t) = £−1 ūref
∆cl (s, K ∗ )

The time responses of α(t) and θ(t) are shown by the following figures. In
the above analysis, the gain used for the variant in determining the root locus
is K = KP C Ke . Typically, for a throttle lever system, the gain KP C can be
altered. The only factor that can be varied therefore is the gain associated with
engine and propeller, Ke .
It is evident from the figure that the faster the time response of the engine
and propeller, the higher the phugoid damping ratio and the faster the time
response of the speed hold system.
For a number of jet transport aircrafts such as, among others, Fokker-100,
Airbus A-320/330/340 and Lockheed Tristar L-1011, an in-flight air brake is
provided in the aircraft to govern the speed during the approach and landing
phases. The following figure describes the speed brake system as the flight speed
regulator of the Fokker F-100 and Airbus A 319/321.
Through the use of the airspeed brake, the aircraft flight speed can be con-
trolled by modulating the brake relative to its open-close position.
2.3. LONGITUDINAL AUTOPILOT 53

1.5
open loop u

0.5

0
0 100 200 300 400 500 600 700

1.5
closed loop u

0.5

0
0 5 10 15 20 25 30 35 40 45 50

Figure 2.16: Time response of u(t) to maintain uref = 1 with and without the
speed hold system, K ∗ = 17.3838

2.3.3 Altitude Hold System

The altitude hold system is a standard system for medium and long range trans-
port aircrafts. This system maintains a cruise altitude that has been selected
by the pilot. The system clearly reduces the pilot’s work load significantly.
The basic principle of the altitude hold system is the use of a signal pro-
portional to the measured aircraft altitude as a feedback to the elevator in such
a way that the elevator motion enable the aircraft to maintain its prescribed
altitude. Fig.2.20 shows the functional diagram of the system. The flight alti-
tude is measured by a pitot static system and the elevator is moved by the basic
control mechanism through a servo motor. At glance this system looks similar
to the flight attitude holding system. The difference is on the fact that flight
altitude variable is not part of the aircraft state variables.
Since the flight altitude h(t) is not part of the motion state variable x, in
the mathematical analysis, firstly it has to be modeled. The variable h(t) is
54 CHAPTER 2. AUTOPILOT SYSTEM

60
40

closed loop θ [deg]

20
0
-20
-40
-60
0 5 10 15 20 25 30 35 40 45 50

40
closed loop α [deg]

20

-20

-40
0 5 10 15 20 25 30 35 40 45 50

Figure 2.17: Time response of α(t) and θ(t) to maintain uref = 1, with the
speed hold gain K ∗ = 17.3838

categorized as the output variable y(t) of the aircraft.

From the flight performance analysis, the model of rate of climb can be given
as the following. See Fig.2.21.

dH
= Vss sin γ (2.35)
dt

where, H —the aircraft altitude with respect to sea level

Vss —the aircraft steady state velocity
d
dt (.) —rate of change with respect to time
If this steady state condition is perturbed by a small disturbance, then the
following relations apply:

H = Ho + h̃ (2.36)
γ = γ o + γ̃
2.3. LONGITUDINAL AUTOPILOT 55

Air brake on the tail

Figure 2.18: Tail air brake (Fokker F-100/70)

The equation of the rate of change of flight altitude can then be obtained
as:

dh̃
= Vss γ (2.37)
dt
In the non-dimensional form, it can be rewritten as:

1 dh̃
≡ ḣ = γ (2.38)
Vss dt

From the kinematic diagram given in Fig.2.21, it can be deduced that:

γ =θ−α (2.39)

therefore,

Z t
dh
=θ−α ⇒ h= (θ − α) dτ (2.40)
dt
56 CHAPTER 2. AUTOPILOT SYSTEM

Air brake on the outer wing

Figure 2.19: Outer-wing air brake (Airbus A-320/319/321)

In the Laplace domain, the flight altitude can then be obtained as:
1
h (s) = [θ (s) − α (s)] (2.41)
s
This equation for h(s) is called the output equation of the altitude hold
system.
Using the available mathematical model of h(s), the mathematical diagram
of the altitude hold system can then be illustrated as the following figure.
Based on Eqs.2.4 and 2.17, the transfer function of the control mecha-
nism/servo and the pitot static can be given respectively as follows:

Kct
Gct (s) = (2.42)
s + 1/τ s
1/τ s
Gps (s) =
s + 1/τ ps
In this model, τ s and τ P S have been specified using the instrument data,
while the gain Kct can be varied to satisfy the control criterion of the altitude
hold system. Referring to Eq.2.41, the transfer function of the flight altitude
h(s) with respect to the elevator deflection δ e (s) can be obtained as follows:
2.3. LONGITUDINAL AUTOPILOT 57

AP computer
δT
u (t )
hˆref +
εˆu δe α (t )
− θ (t )
q (t )
control mechanism
and servo aircraft

hˆm ( t ) h (t )

Pitot static: altitude sensor

Figure 2.20: Functional diagram of altitude hold system

1
h (s) /δ e (s) = [θ (s) − α (s)] /δ e (s) (2.43)
s∙ ¸
1 θ (s) α (s)
= −
s δ e (s) δ e (s)
" #
1 Nδθe (s) − Nδαe (s)
=
s ∆long (s)
1 Nδhe (s)
= ≡ Ghδe (s)
s ∆long (s)

Therefore,

h (s) = Ghδe (s) δ e (s) (2.44)

where,

Nδhe (s)
Ghδe (s) = (2.45a)
s∆long (s)
Nδhe (s) = Nδθe (s) − Nδαe (s) (2.45b)
58 CHAPTER 2. AUTOPILOT SYSTEM

xs
α Vss
dH
γ θ dt
dV
dt
ys

Figure 2.21: Kinematic diagram of aircraft rate of climb

h(s)
Based on Fig.2.22, the closed loop transfer function href (s) can be written as
follows:
h (s)
= Gcl (s) (2.46)
href
where

Ncl (s)
Gcl (s) = (2.47a)
∆cl (s)
Nδhe (s)
Ncl (s) = Gct (s) Ghδe (s) = Kct (2.47b)
s (s + 1/τ s ) ∆long (s)
Nδhe (s)
∆cl (s) = 1 + (Kct /τ ps ) (2.47c)
s (s + 1/τ s ) (s + 1/τ ps ) ∆long (s)
To better understand the work principle of the altitude hold system, a case
study will be taken for the Indonesian Aerospace N250-100 PA2 ”Krincing Wesi”
at the cruise flight configuration with the speed of V = 250 kias and at the
altitude of h = 15000 ft. From Eqs.2.9b—2.10a, the data of the altitude hold
system of the N250-100 PA2 can be obtained as the following. For τ s = 0.1 and
τ P S = 0.2, then the numerator polynomial for h due to δ e can be given as
2.3. LONGITUDINAL AUTOPILOT 59

δT ( s)
u( s)
hˆref εˆh [ N long ( s )]
+
GCt ( s ) α (s)
−
δ e (s) Δ long ( s ) θ ( s)
q( s )
control mechanism aircraft
and servo

hˆm h( s) h
1 +
G PS ( s )
s −

pitot static

Figure 2.22: Mathematical diagram of the altitude hold system

³ s ´³ s ´³ s ´
Nδhe (s) = Shδe −1 +1 + 1 (2.48a)
10.8454 10.8389 0.0167

and the denumerator polynomial is:

1 1
∆h (s) = s(s + )(s + )∆long (s) (2.48b)
τs τPS
µ ¶
s6 + 18.3115s5 + 107.8173s4
∆h (s) = s (2.48c)
+287.9076s3 + 409.7023s2 + 8.8517s + 2.7708

where

Shδe = −0.1230 (2.48d)

The following figure illustrates the root locus of this altitude hold system. It
can be observed that for a negative gain, Kct < 0, the system quickly leads to
instability through its phugoid mode. The damping ratio for the pitch oscillation
mode is not substantially increased either.
.
For the positive gain, Kct > 0, the
condition is worse since the integrator h directly moves to the right half plane of
the root locus. Thus it can be concluded that the altitude hold system using the
h −→ δ e feedback fails to give a stable solution in maintaining the flight altitude.
60 CHAPTER 2. AUTOPILOT SYSTEM

Therefore a different strategy needs to be devised so that the instability of the

phugoid mode can be delayed and the damping ratio of the pitch oscillation can
be increased.

Root Locus
30

20 servo
pitch oscillation

10
Imaginary Axis

-10

pitot static
-20

-30
-40 -30 -20 -10 0 10 20 30
Real Axis

Figure 2.23: Root locus for the altitude hold system for the N250-100 with
h −→ δ e feedback with gain Kct < 0

To achieve the above objective a number of methods by adding inner feed-

back loop are in order, namely:

1. pitch attitude feedback, θ

2. forward acceleration feedback, ax

3. compensator addition, Gc (s)

Altitude hold system using an attitude hold as its inner loop

Since the first method has been discussed in Sec.2.3.1, the design of this pitch
attitude hold system is readily implementable for the inner loop of the altitude
hold system of the aircraft. See the mathematical diagram of the corresponding
in Fig.2.25.
2.3. LONGITUDINAL AUTOPILOT 61

Root Locus

0.1
phugoid

integrator
0.05
Imaginary Axis

-0.05

-0.1

-0.1 -0.05 0 0.05 0.1

Real Axis

Figure 2.24: Root locus for the altitude hold system for the N250-100 with
h −→ δ e feedback with gain Kct < 0 —zoomed to show the phugoid mode

Note that Shθ is the conversion gain from signal bεh to signal θref which is
a constant. From the elaboration given in Sec.2.3.1, a good pitch attitude hold
∗
control loop has been obtained with Kct = −8.9695. The corresponding roots
of the closed loop polynomial have been given in Eq.2.15. Thus the inner closed
loop characteristics polynomial can be expressed as:
∗
∆cli (s, Kct ) = s5 + 13.3115s4 + 41.2598s3 + 223.1724s2 + 186.8562s + 4.6024
(2.49)

By using this inner loop, the mathematical diagram of the altitude hold sys-
tem can be simplified as illustrated in Fig.2.26. In the diagram, the numerator
polynomial Nδhe can be calculated as follows:

¡ θ ¢
Nδh∗
e
∗
= Kct Nδe − Nδαe (2.50)

Thus, the result is:

Nδh∗
e
(s) = −1.5535s3 − 0.0159s2 + 182.6204s + 3.0580 (2.51)
62 CHAPTER 2. AUTOPILOT SYSTEM

δT ( s)
u(s)
hˆref εˆh θˆref εθ [ N long ( s )]
+
S hθ GCt ( s ) α (s)
−
δ e (s) Δ long ( s ) θ (s)
q( s )
θˆm
S vg

hˆm h( s) h
1 +
G PS ( s )
s −

Figure 2.25: Altitude hold system with an attitude hold system as the inner
loop

From the above diagram, the equation for h(s) can be expressed as:

h (s) = Gclo (s) h̄ref (2.52a)

where

Khθ τ ps (s + 1/τ ps ) Nδh∗

e
(s)
Gclo (s) ≡ (2.52b)
τ ps s (s + 1/τ ps ) ∆cli (s) + Khθ Nδh∗e
(s)
Nδhe
clo
=
∆clo

In this case, the root locus equation for the Gclo (s) is ∆clo (s) = 0 or

Khθ Nδh∗
e
(s)
1+ =0 (2.53)
τ ps s (s + 1/τ ps ) ∆cli (s)

Fig.2.27 shows the root locus of the altitude hold system using the attitude
hold inner loop. From the root locus diagram, it is manifest that the damping
ratio of the pitch oscillation mode increases for the low value of gain Khθ whereas
the phugoid mode moves from the subsidence to a long period oscillation before
tending to unstable oscillatory condition.
2.3. LONGITUDINAL AUTOPILOT 63

u( s)
hˆref εˆh εθ K ct * [ N long ( s )]
+
K hθ α (s)
− Δ cli ( s, K ct * ) θ (s)
q( s )
Outer gain
Inner loop with gain K ct *

θ ( s) α (s)
hˆm h( s) h ( s )
1 +
GPS ( s )
s −

∗
Figure 2.26: Altitude hold system with inner loop having gain Kct

If the gain is chosen such that:

Khθ
= 1.7372 ⇒ Khθ = 0.17372 (2.54)
τ ps

the following characteristic roots are obtained:

p1 = −11.2729 (2.55)
p2 = −5.0514
p3 = −0.0163
p4,5 = −0.5727 ± j4.1047
p6,7 = −0.4102 ± j0.4051

In this case, the outer closed loop gives the damping ratio of the pitch
oscillation and phugoid modes as follows:
Pitch oscillation ζ po = 0.098
Phugoid ζ ph = 0.356
If this damping can be accepted, then h(t) can be calculated by using
Eq.2.52a.
Fig.2.29 illustrates the time response of h(t) for href = 1, with and without
the altitude hold system.

Case 3 For the case without altitude hold system, the step input href = 1 can
not be maintained. As a result, the aircraft will keep losing the altitude unless
the pilot interferes with corrective actions
64 CHAPTER 2. AUTOPILOT SYSTEM

Root Locus

10

8 pitch oscillation

4
Imaginary Axis

-2

-4
pitot
-6
static
-8 phugoid

-10
-10 -5 0 5 10
Real Axis

Figure 2.27: Root locus diagram of the outer loop of the altitude hold system
for N250-100 PA-2 aircraft

Case 4 For the case with altitude hold system, where a pitch attitude hold sys-
tem is employed as an inner loop, the value of href = 1 can be quickly main-
tained in not more than 10 time unit. This system works perfectly and yields
no offset at all. The altitude hold system employing an attitude hold system as
the innner loop demonstrates reliable performance. A better performance of this
type of system can be further achieved if a pitch stability augmentation system
is employed.

Altitude hold system using a forward acceleration feedback as its

inner loop

A second method that can be used for the inner loop of the altitude hold system
is the feedback of forward acceleration du
dt . For illustration of the inner control
loop with ax feedback, see the following diagram given in Fig.2.30.

From the above diagram, the following expression for u(s) can be derived:
2.3. LONGITUDINAL AUTOPILOT 65

Root Locus

0.7 0.56 0.38 0.2

0.81
1.5

0.89
1 phugoid
0.95
0.5
Imaginary Axis

0.988
2 1.5 1 0.5
0 integrator
0.988
-0.5
0.95

-1
0.89

-1.5
0.81
0.7 0.56 0.38 0.2
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
Real Axis

Figure 2.28: Root locus diagram of the outer loop of the altitude hold system
for N250-100 PA-2 aircraft—zommed to show the phugoid mode

" #
Nδue (s)Gct (s)
u (s) = âref (2.56)
∆long (s) + sGct (s) Sacc Nδue (s)

Using the model of control mechanism and elevator servo as given in Eq.2.4
with the value of τ s = 0.1 sec., the polynomial of the inner control loop using
the forward acceleration feedback can be expressed as follows:

sNδue (s)
1 + Ki =0 (2.57)
(s + 1/τ s ) ∆long (s)
where,

Ki = Sacc /τ s
Fig.?? shows the root locus of the inner control loop. It is clear that the phugoid
mode gets more stable as the gain Ki goes up, due to the increasing damping
66 CHAPTER 2. AUTOPILOT SYSTEM

40

30

open loop h 20

10

0
0 20 40 60 80 100 120 140 160
closed loop h dengan pitch attitude hold

1.5

0.5

0
0 5 10 15 20 25 30 35

Figure 2.29: Time response of h(t) for an input of href = 1 from a system with
and without altitude hold for N250-100

ratio. However, the pitch oscillation mode becomes oscillatorily unstable when
the gain Ki increases. As an example the following design point is taken:

Ki∗ = 0.0409 (2.58)

The poles associated with this design point are:

p1 = −10.0185
p2,3 = −1.5856 ± j2.3045
p4,5 = −0.0609 ± j0.0580

The characteristic polynomial for the inner loop can then be obtained as:

∆i (s, Ki∗ ) = s5 + 13.3115s4 + 41.2090s3 + 83.3069s2 + 9.8237s + 0.5542

(2.59)

With the choice of gain Ki∗ for the inner control loop, the mathematical
diagram for the altitude hold system described in Fig.2.30 can be simplified as
the diagram shown in Fig.2.31.
2.3. LONGITUDINAL AUTOPILOT 67

δT ( s)
u( s)
hˆref εˆh aˆ ref εa [ N long ( s )]
+
S ha x
x
GCt ( s ) α ( s)
−
δ e ( s) Δ long ( s ) θ ( s)
q( s )
a xm aˆ x
S acc s

hˆm h( s) h
1 +
G PS ( s )
s −

Figure 2.30: Altitude hold system with forward acceleration ax as the inner loop
feedback

The outer closed loop transfer function relating h(s) and href (s) can then
be derived as follows:

h (s) Ko τ ps (s + 1/τ ps ) Nδhe (s)

= (2.60)
href (s) s (s + 1/τ ps ) ∆i (s, Ki∗ ) + Ko Nδhe (s)

where Nδhe (s) is given by Eq.2.45a and the model for the pitot static Gps (s) is
given by Eq.2.42. The outer loop gain is given by:

Shax
Ko = (2.61)
τ s τ ps

As a result, the outer loop characteristic polynomial of the altitude hold

system can be expressed as:

Nδhe (s)
1 + Ko =0 (2.62)
s (s + 1/τ ps ) ∆i (s, Ki∗ )

For the value of τ s = 0.2, the root locus of the outer control loop is given by
Fig.2.32. The root locus shows that the pitch oscillation mode remains oscilla-
torily stable but with rising natural frequency and slightly increasing damping
ratio. Whereas the stability of the phugoid mode decreases with the increase of
the gain Ko . If a working point of the outer root locus is selected as
68 CHAPTER 2. AUTOPILOT SYSTEM

u( s)
hˆref +
εˆh aˆ ref 1
[ N long ( s )] α (s)
−
S ha x τs θ (s)
Δ i ( s, K i * ) q( s )

Inner loop

hˆm h( s) h
GPS ( s )
1 +
s −

Figure 2.31: Altitude hold system with inner loop using forward acceleration
feedback ax

Ki∗ = −0.1752 (2.63)

The corresponding poles are given by

p1 = −10.0185
p2 = −0.0102
p3 = −5.0013

p4,5 = −1.5883 ± j2.3056

p6,7 = −0.0524 ± j0.1105
The outer loop characteristic polynomial can then be found as:

∆o (s, Ko∗ ) = s7 + 18.3115s6 + 107.77s5 + 289.35s4 (2.64)

+426.31s3 + 49.62s2 + 6.34s + 0.06
Hence, the time response h(t) due to the step input href = 1 for this altitude
hold system can be expressed by using Eq.2.60, which can be rewritten for
Ko = Ko∗ as follows:

" #
−1
Ko∗ τ ps (s + 1/τ ps ) NδhT (s)
h (t) = £ href (2.65)
∆o (s, Ko∗ )
2.4. LATERAL-DIRECTIONAL AUTOPILOT 69

Root Locus
25

20

15

10

5
Imaginary Axis

-5

-10

-15

-20

-25
-20 -10 0 10 20 30 40 50 60 70 80
Real Axis

Figure 2.32: Root locus diagram of inner control loop u̇ −→ δ e for N250-100
altitude hold system

The following figure shows the response of h(t) for the altitude hold system
using the forward acceleration ax feedback. Compared to the same system using
the pitch attitude hold system as an inner loop, this system exhibits a slower
response to reach a steady state value of href = 1.

2.4 Lateral-Directional Autopilot

The following subsections will be focused on the lateral directional autopilot
design for a transport aircraft. The discussion covers the design for bank hold,
heading hold and VOR-hold systems.

2.4.1 Bank Angle Hold (Wing Leveler System)

The bank angle hold system is designed to keep the magnitude of aircraft bank
angle. When the reference bank angle is zero, the system is also called a wing
leveler. The following diagram given in Fig.2.36, shows the functional diagram
70 CHAPTER 2. AUTOPILOT SYSTEM

Root Locus

0.08

0.06

0.04

0.02
Imaginary Axis

-0.02

-0.04

-0.06

-0.08

-0.1
-0.2 -0.15 -0.1 -0.05 0 0.05 0.1 0.15
Real Axis

Figure 2.33: Root locus diagram of inner control loop u̇ −→ δ e for N250-100
altitude hold system—zoomed around the phugoid mode

of the bank angle hold.

The step by step working procedure for this automatic pilot can be elabo-
rated as follows. First, the aircraft(1) motion is modeled by the state variable
x = {β, p, r, ϕ} and the control variable {δ a , δ r }. The rolling motion of the air-
craft ϕ(t) is sensed by the roll gyroscope(2) and the measured signal, in the
form of roll angle signal ϕ b m , is inputted to the autopilot computer(3). Within
the AP computer, the measured signal ϕ b m is compared to the reference bank
angle determined by the pilot. The pilot acquired the value of the reference
signal ϕ b ref by performing a maneuver in such a way that the aircraft plane
of symmetry coincides with the local vertical plane. In this attitude, the roll
angle of the aircraft ϕ(t) will be the same as the bank angle φ(t). The result of
the comparison between ϕ b m (t) and ϕb ref is amplified to be converted to control
command signal b εϕ . The signal bεϕ is entered to AP servo system(4) to move the
steering wheel and afterward the displacement of the steering wheel is trans-
mitted to the aileron control surface. The aileron will be deflected by an angle
of δ a (t) and in turn changes the aircraft attitude. This process is repeated in
the bank angle hold closed loop until the measured roll angle ϕ b m is the same as
the reference roll angle ϕb ref as desired by the pilot. In this situation, the value
of the command signal b εϕ = 0. Since the process from block(2) until block(4)
2.4. LATERAL-DIRECTIONAL AUTOPILOT 71

Root Locus

10

5
Imaginary Axis

-5

-10

-10 -5 0 5 10 15
Real Axis

Figure 2.34: Root locus diagram of outer control loop u̇ −→ δ e for N250-100
altitude hold system

is performed in the electrical domain, it can be considered very fast. While

the process from block(4) to block(2) is executed in the hydro-aero-mechanical,
therefore the time response is very dependent on the mass and inertia of the
associated system. Mathematically the functional diagram of such system can
be described in Fig.2.37.
In the above model, the transfer functions of the bank angle hold closed loop
system consists of:
1. Aircraft transfer function matrix, for δ r = 0,
¯ Nld (s)|δr =0 x̄(s)
GA/C (s)¯δ = = (2.66)
r =0 ∆ld (s) δ a (s)
2. Roll gyro transfer function. Due to the possible application of commer-
cially available laser and fiber optic gyroscopes, the time response of the
gyro in sensing the pitch angle ϕ(t) can be considered very fast, thus the
transfer function can be modeled as a constant:
Grg (s) = Srg ≡ constant (2.67)
72 CHAPTER 2. AUTOPILOT SYSTEM

1.5

1
closed loop teta

0.5

0
0 100 200 300 400 500 600

Figure 2.35: Time response of h(t) for N250-100 altitude hold system with the
inner loop of ax −→ δ e feedback

3. Comparator in the autopilot computer transfer function:

b b ref − ϕ
εϕ (s) = ϕ b m (s) (2.68)

4. Signal amplification system transfer function. The transfer function of

the amplification can be modeled as the constant gain due to its electrical
domain.

Gamp = Samp = constant (2.69)

5. Lateral autopilot servo and control mechanism transfer function. The sys-
tem can be considered to have a relatively faster time response compared
to the dynamics of the aircraft. The assumption can be especially jus-
tified if the domain of the control system is of electro-hydro-mechanical
2.4. LATERAL-DIRECTIONAL AUTOPILOT 73

(3) AP computer
δr
β (t )
ϕˆ ref +
εˆϕ εˆϕ δa
c p (t )
Amp r (t )
−
ϕ (t )
(4) AP servo to aileron steering
(1) aircraft

ϕˆm ϕ (t )

(2) Roll gyroscope

Figure 2.36: Functional diagram of Bank angle hold system

namely fly by wire or fly by light. The block of the AP servo and control
mechanism can then be modeled as a system with time response of τ s
representing the first order lag factor as follows:

1/τ s
Gact (s) = (2.70)
s + 1/τ s

The characteristic polynomial of the bank angle hold closed loop system is
given by:

Nδϕa (s)
∆cl (s) = ∆ld (s) + kqδa (2.71a)
s + 1/τ s

where:

kqδa ≡ Samp Srg /τ s (2.71b)

Therefore the closed loop characteristic equation of the bank angle hold system
can be given as:

1 + kqδa Gol (s) = 0 (2.72a)

74 CHAPTER 2. AUTOPILOT SYSTEM

δr
β (s)
ϕˆref +
εˆϕ εˆϕ [ N ld ( s )] p( s)
S amp c
Gact ( s )
−
δa Δ ld ( s ) r (s)
ϕ (s)

ϕˆm ϕ
Grg ( s )

Figure 2.37: Mathematical diagram of bank hold system

Nδϕa (s)
Gol (s) ≡ (2.72b)
(s + 1/τ s ) ∆ld (s)

For a conventional aircraft, GOL (s) typically has a pair of complex zeros and
five poles consisting of a pair of complex poles and three real poles where one
of the poles is associated with the control mechanism and servo system.
To have a more elaborate understanding of the work principle of the bank
angle hold system, a case study will be taken for the Indonesian Aerospace N250-
100 PA2 ”Krincing Wesi” at the cruise flight configuration with a low speed of
V = 180 KEAS and at the altitude of h = 15000 ft. The lateral directional
transfer function for aileron input for the aircraft is given as follows:

³ s ´µ s
¶µ
s
¶
Nδβa (s) = Sβδa +1 +1 +1
0.1232 7.7628 + j14.9198 7.7628 − j14.9198
(2.73)
µ ¶µ ¶
s s
Nδϕa (s) = Sϕδa +1 +1
0.2278 + j1.2565 0.2278 − j1.2565

³ s ´µ s
¶µ
s
¶
Nδra (s) = Srδa +1 −1 −1
0.6081 0.8112 + j1.2543 0.8112 − j1.2543
2.4. LATERAL-DIRECTIONAL AUTOPILOT 75

Nδpa (s) = sNδϕa (s)

with static sensitivity coefficients given by:

Sβδa = 3.01743 (2.74)

Sϕδa = 176.40281
Srδa = 14.54464
and the characteristic polynomial of:

s4 s3 s2 s
∆ld (s) = + + + +1 (2.75)
0.0381 0.01706 0.01544 0.01004
with the following characteristic roots:

p1 = −1.9577, associated with the roll subsidence (2.76)

p2 = −0.0101, associated with the spiral
p3,4 = −0.1325 ± j1.3817, associated with the dutch roll, where:
ω nDR = 1.39 rad/s and ζ DR = 0.0955
Due to the use of ring laser gyroscope, the value of the gain Srg can be taken
as unity. The root locus associated with the characteristic equation Eq.2.72a
is given for a number of time response, τ s = 0.1, 0.2 and 0.5, by the following
figure.
It is evident that for a sufficiently large value of the gain kϕδa , the dutch roll
mode moves to the right hand side of the imaginary axis leading to oscillatory
instability. The larger the value of τ s the quicker the dutch roll mode becomes
unstable. In contrast, the roll subsidence and spiral merges to become a stable
oscillatory mode. In each of the root locus shown, the following design point
can be taken:
i kϕδai τi characteristic roots
1 8.9344 0.1 −10.4101; −0.4014 ± j1.7353; −0.51 ± j12529
2 3.8243 0.2 −5.707; −0.2215 ± j1.6754; −0.5414 ± j1.1382
3 0.7711 0.5 −2.9695; −0.0969 ± j1.4637; −0.5348 ± j0.7328
For each of the above design point in the root locus, the associated closed
loop characteristic polynomial is given the following equations:

τ s1 = 0.1 =⇒ ∆cl1 = s5 + 12.2329s4 + 24.797s3 + 65.3037s2 + 54.7837s + 60.4316

(2.77)

τ s2 = 0.2 =⇒ ∆cl2 = s5 + 7.2329s4 + 13.6325s3 + 31.8998s2 + 26.2032s + 25.8943

76 CHAPTER 2. AUTOPILOT SYSTEM

Root Locus
3
0.5 0.38 0.28 0.17 0.08
2.5 0.2
0.64 0.5
0.1 2
2
1.5
0.8 Dutch roll
1
1
0.94 0.5 spiral
Imaginary Axis

0.94 0.5
-1
roll 1
0.8 subsidence
1.5
-2 0.1 2
0.64 0.5
2.5
0.5 0.38 0.28 0.17 0.08 0.2
-3
-2 -1.5 -1 -0.5 0 0.5 1
Real Axis

Figure 2.38: Root locus diagram of the bank hold system of N250-100 aircraft
for a number of τ s values

τ s3 = 0.1 =⇒ ∆cl3 = s5 + 4.2329s4 + 6.9338s3 + 11.9107s2 + 9.0791s + 5.2588

The time response of ϕ(t) for any case of τ si can be calculated as follows:

∙ ¸
−1
kϕδai Nδϕa
ϕi (t) = £ ϕ (2.78)
∆cli (s, kϕδai ) ref i=1,2,3

The following figure illustrates the time response of ϕ(t) for the three differ-
ent cases of τ s . It can be observed that for τ s = 0.1(a = 10), the roll response
will be quickly damped even though the overshoot peak is fairly large.
It is evident that the larger the value of τ s , the lower the damping of the
time response ϕ(t), even though the overshoot peak decreases. Fig.2.40 shows
the roll angle response due to unit step input. It is clear that the trend for
the damping and the overshoot peak is similar to that of the impulse input.
However note that by the selection of the gain for the case when τ = 0.5, the
time response shows no overshoot or oscillation.
2.4. LATERAL-DIRECTIONAL AUTOPILOT 77

a=10
0.8

0.6
a=5
roll angle phi (rad)

0.4
a=2

0.2

-0.2

-0.4
0 5 10 15 20 25 30 35

Figure 2.39: Time response of ϕ(t) for a = 1/τ s = 10, 5, 2 due to an impulse
function input

2.4.2 Heading Hold System

The majority of a heading hold system is designed as an extension of the bank

angle hold system.. This is so due the fact that there exists a relation between
the heading rate Ψ and the roll angle as given by Fig.2.41

The horizontal and vertical forces equilibrium during the turn maneuver at
the constant altitude as illustrated by the above diagram can be expressed as
follows:

L cos ϕ = W (2.79a)
W 2
L sin ϕ = Ψ̇ R (2.79b)
g
78 CHAPTER 2. AUTOPILOT SYSTEM

1.4
a=10
1.2

1
roll angle phi (rad)

0.8 a=5

0.6
a=2
0.4

0.2

0
0 5 10 15 20 25 30 35

Figure 2.40: Time response of ϕ(t) for a = 1/τ s = 10, 5, 2 due to a step function
input

where: L: lift
W : weight
g: gravitation acceleration
ϕ: roll angle
R: turning radius
.
Ψ : heading rate or turn angle rate
Substituting Eq.2.79a to Eq.2.79b, the following relation can be obtained:

Ψ̇2 R = g tan ϕ (2.80)

Meanwhile, an expression for heading rate can be written from the kinemat-
ics relation of the velocities:
V
Ψ̇ = (2.81)
R
Plugging Eq.2.81 to Eq.2.80, the heading rate can be expressed as:
2.4. LATERAL-DIRECTIONAL AUTOPILOT 79


Ψ
L
ϕ

W  2R
Ψ
g
R

W 
Ψ

V
R

Figure 2.41: Forces equilibrium during the turn maneuver

g tan ϕ g
Ψ̇ = ' ϕ (2.82)
V V
or in Laplace domain, it reads:

g ϕ(s)
Ψ(s) = (2.83)
V s
From the above relations, it is clear that the heading angle Ψ(s) can be
obtained from the integration of the roll rate angle ϕ(s). Therefore it can be
stated that a bank hold system represents an inner loop of a heading hold
system. The functional and mathematical diagram of the heading hold system
are illustrated in Fig.2.42 and Fig.2.43, respectively.
It is assumed that the heading gyro has a high performance, thus its time
response can be considered very fast. As a result, the transfer function of the
heading gyro can be represented by a constant gain:

Ghg (s) = Shg = constant (2.84)

In the following analysis, it will be assumed that a bank hold system has
been designed as the inner loop of the heading hold system. Its characteristic is
80 CHAPTER 2. AUTOPILOT SYSTEM

AP computer
servo and control mechanism aircraft
δr
β (t )
ψˆ ref + + δa p (t )
Amp r (t )
− −
ϕ (t )

Roll gyroscope

Heading gyroscope

Figure 2.42: Functional diagram of heading hold system

∗
specified by appropriately selecting the gain kϕδai
. Hence the transfer function
of the inner loop can be expressed as:

∗
x̄(s) kϕδa N x̄ ¡ ∗ ¢
= ³ i δa ´ = Ḡcli s, kϕδa (2.85)
ϕc (s) ∗
∆cli s, kϕδa
i

where,

x̄ = { β p r ϕ } (2.86a)
¡ ∗ ¢ ∗
Nδϕa (s)
and ∆cli s, kϕδa = ∆ld (s) + kϕδa (2.86b)
i i
s + τ1s

Using the above equation of inner loop transfer function, the mathematical
diagram of the heading hold system as described in Fig.2.43 can be simplified
as given by the following figure. See Fig.2.44.
The outer closed loop tranfer function, in this case, can be expressed as:

Ψ(s) = Gclo (s, kψ )Ψref (2.87)

where
2.4. LATERAL-DIRECTIONAL AUTOPILOT 81

δr
β (s)
ψˆ ref +
εˆψ εˆϕc [ N ld ( s )] p(s)
S amp Gact ( s )
−
δa Δ ld ( s ) r ( s)
ϕ (s)
servo Aircraft
Amplifier ϕˆm ϕ
Grg ( s )

Roll gyro
ψm ψ
g
Ghg ( s )
Vs

Heading gyro Model ψ

Figure 2.43: Mathematical diagram of heading hold system

ϕ
kψ Nδa (s)
Gclo (s, kψ ) = (2.88a)
Shg ∆clo (s, kψ )
∗ g
kψ = Samp Shg kϕδa (2.88b)
i
V
The characteristic polynomial of the outer closed loop ∆clo (s, kψ ) can be
expressed as:

¡ ∗ ¢
∆clo (s, kψ ) = s∆cli s, kϕδai
+ kψ Nδϕa (s) (2.89)

Thus the characteristic polynomial of the heading hold system can be given
as:

Nδϕa (s)
1 + kψ ³ ´ =0 (2.90)
∗
∆cli s, kϕδa i

The root locus of the heading hold system for the N250-100 is presented in
∗
Fig.??. In this case, the inner loop characteristic polynomial ∆cli (s, kϕδai
) is
taken from the example of bank angle hold system given in Sec.2.4.1; for three
different cases: τ s = 0.1, 0.2 and 0.5 or a = 10, 5 and 2. The three root locus
diagrams associated with Eq.2.90 is shown in the following figure for each value
82 CHAPTER 2. AUTOPILOT SYSTEM

β (s)
ψˆ ref +
εˆψ ϕc * p(s)
S amp Gcli ( s , k )
− ϕδ ai r (s)
ϕ (s)
Amplifier Bank angle hold

ψm ψ
g
S hg
Vs

Heading gyro Model ψ

Figure 2.44: Heading hold system with bank angle hold as an inner loop

of τ s (or a). From the above root locus diagrams, the following working points
can be taken:
i kψi ai = 1/τ i characteristic roots
1 4.5943 10 −10.3913; −0.1744 ± j1.8127; −0.3930 ± j1.0553; −0.7066
2 2.2746 5 −5.6489; −0.8521; −0.0429 ± j1.7072; −0.3230 ± j0.9923
3 0.1606 2 −2.9384; −0.3055; −0.0816 ± j1.4565; −0.4129 ± j0.6280
In this case, the characteristic polynomial for each value of a can be given
as the following:

a1 = 10 =⇒ ∆clo1 = s6 + 12.2329s5 + 24.7970s4 + 65.3037s3

+73.7215s2 + 69.0615s + 30.8796 (2.91)

a2 = 5 =⇒ ∆clo2 = s6 + 7.2329s5 + 13.6375s4 + 31.8998s3

+35.5789s2 + 30.1668s + 15.2879

a3 = 2 =⇒ ∆cl03 = s6 + 4.2329s5 + 6.9379s4 + 11.9107s3

+9.7410s2 + 5.5604s + 1.0793

Using the above data, the time response of ψ(t) and ϕ(t) can be calculated
from the following equations.
2.4. LATERAL-DIRECTIONAL AUTOPILOT 83

Root Locus
2

1.5 a=10
a=2
a=5
1

0.5
Imaginary Axis

-0.5

-1

-1.5

-1.5 -1 -0.5 0 0.5 1 1.5

Real Axis

Figure 2.45: Root locus diagram of the heading hold system of N250-100 aircraft
for a number of a = 1/τ s values

" #
kψi
−1
Nδϕa (s)
Ψi (t) = £ ¡ ¢ Ψref
Shg ∆clo s, kψi
∙ ¸
V
ϕi (t) = £−1 s Ψi (s) (2.92)
g
The time response is presented in Fig.2.46. It is clearly indicated that the
smaller the value of τ s (or the larger the value of a) the faster the time response
of the actuator which in turns causes the higher the damping ratio of the Dutch
roll mode. For a fairly large value of τ s (namely for a = 2), the damping ratio
can only be obtained for a small value of gain kψ . For a small value of τ s , the
high overshoot peak is due to the associated high gain.
The heading hold system is capable of maintaining the direction of flight
that has been selected by the pilot. However, this system would not be able to
maintain the flight path of the aircraft along a certain desired reference. See
the sketch presented in Fig.2.47. If the aircraft is induced by side wind, it will
84 CHAPTER 2. AUTOPILOT SYSTEM

0.6
a=10

heading angle psi(rad)

0.4
a=5

0.2 a=2

-0.2
0 5 10 15 20 25 30 35 40 45 50

0.6

0.4 a=10
roll angle phi(rad)

a=2
0.2 a=5

-0.2

-0.4
0 5 10 15 20 25 30 35 40 45 50

Figure 2.46: Time response of ψ(t) and ϕ(t) of the heading hold system of
N250-100 aircraft

be drifted from its desired flight path even though the aircraft heading angle
can be maintained. Considering this kind of situation, a system that is capable
of simultaneously maintaining the flight direction and flight path of the aircraft
is thus desired. This system is referred to as a VOR hold or lateral beam hold.

2.4.3 VOR-Hold System

The VOR-hold system has the role as an automatic guidance system which is
combined with the control system for flight direction hold system. The system
keeps the flight path to be aligned with the reference direction which is transmit-
ted by the VOR terrestrial guidance system. In principle, this system automates
the work of the pilot in order that the CDI (Course Deviation Indicator) needle
is kept centered. To understand the automation process associated with this
system see Fig.2.48. An aircraft is flying under the coverage of the transmitted
wave from a VOR station with a certain bearing reference wave and a certain
wave angle. The transmission bearing of the VOR main wave with respect to
the magnetic north is represented by ψ ref . The reference angle ψ ref has to be
followed by the aircraft flight angle. The width of the transmission wave is Φ.
2.4. LATERAL-DIRECTIONAL AUTOPILOT 85

To destination

wind

th ψ ref
t pa
lig h ψ ref
e df
sir
e de
Th
ψ ref
ψ ref

ψ ref

Figure 2.47: Effect of wind to the aircraft flight path

The actual position of the aircraft has an offset angle λ with respect to the main
radial line and a radial distance of R with respect of VOR station. The aircraft
position has a bearing ψ with respect to magnetic north and a distance d with
respect to the VOR main bearing. From the geometry of the VOR guidance
path the following relations can be readily obtained:
d
tan λ = (2.93)
R
For a small λ, the above relation can be approximated by:

57.3
λ (s) = d (s) (2.94)
R
˙ can be
The rate of change of distance with respect to the main radial, d(t)
obtained through the following relation:

d ¡ ¢
d˙ (t) ≡ (d) = Vp sin ψ − ψ ref (2.95)
dt
where Vp is the aircraft flight velocity. The above relation can be further
simplified for a small angle case as follows:

¡ ¢
d˙ (t) ∼
= Vp ψ − ψ ref (2.96)
¡ ¢
or sd (s) = Vp ψ (s) − ψ ref
86 CHAPTER 2. AUTOPILOT SYSTEM

ψ −ψ ref
ψ ref

main wave
ψ

R d
um λ ψ

ψ ref
Vp
transmission width
VOR

Figure 2.48: VOR guidance path geometry

Substituting Eq.2.96 to Eq.2.94, the offset angle can be expressed as:

∙ ¸
57.3 ψ (s) − ψ ref
λ (s) = Vp (2.97)
R s
In the process of guidance, in order that the aircraft position can be fixed
to the main bearing line, the offset angle has to be made to zero. Therefore the
task of the automatic guidance system is the make the λ(s) zero. The following
figures illustrate the functional and mathematical diagrams for the VOR hold
system.
Refer to Fig.2.50. The system inside the dashed box represents the heading
hold control system which has been discussed in Sec.2.4.2. Beyond the box is
the flight guidance and navigation.
For the above guidance and navigation system, the transfer functions are
given as follows:

1. Comparator of aircraft heading angle and VOR bearing

εψ (s) = ψ ref − ψ (s) (2.98)

The reference angle ψ ref is the bearing of the wave transmitted from the
VOR ground station which is nominally received the aircraft.
2.4. LATERAL-DIRECTIONAL AUTOPILOT 87

AP computer AP servo and control mechanism aircraft

β (t )
+ + δ a ,δ r p (t )
Amp r (t )
− −
ϕ (t )
ψˆ m

Roll
Control gyroscope
loop Flight
heading
Heading gyroscope

Guidance loop ψˆ m

εˆψ ψˆ ref Main radial

Guidance VOR
ψˆ comm Geometry

Navigation and control computer

Figure 2.49: The functional diagram of VOR-hold guidance-control system

2. Guidance geometry, Gng (s)

From Fig.2.48, the aircraft flight path geometry with respect to the VOR
bearing has been derived. Eq.2.97 gives

λ (s) = Gng (s) εψ (s) (2.99a)

Vp 1
Gng (s) = 57.3 (2.99b)
Rs

3. Offset comparator λ(s)

ελ (s) = λref − λ (s) (2.100)

It is clear that λref = 0, since the objective of the system is to make the
offset angle λ = 0. Therefore,

ελ (s) = −λ (s) (2.101)

4. Coupler Gcoup (s)

The coupler represents the conversion system from the output of the flight
guidance system to the command signal ψ comm (s). This conversion system
can be expressed as combination of integrator and first order proportional
88 CHAPTER 2. AUTOPILOT SYSTEM

δr
ψˆ ref β (s)
+
εˆψ ϕc εˆϕ [ N ld ( s )] p(s)
S amp Gact ( s )
−
δa Δ ld ( s ) r (s)
ϕ ( s)
servo Aircraft
Amplifier ϕˆ m ϕ
coupler Grg ( s )

Gcop ( s )
Roll gyro
ψm ψ
λ ref εˆλ g
+ Ghg ( s )
ψm Vp s
−
λ (s)
Model ψ
57.3 1 εˆψ − +
Vp ψˆ ref From the VOR
R s

Guidance geometry

Figure 2.50: The mathematical diagram of the VOR-hold guidance-control sys-

tem

terms.

ψ com (s) = Gcop (s) ελ (s) (2.102)

∙ ¸
s + 0.1
Gcop (s) = Kc (2.103)
s

5. Aircraft and heading hold system

As explained in Sec.2.4.2, the transfer function of the aircraft and heading

hold system is given by Eq.2.92 as follows:

¡ ¢
ψ (s) = Gclo s, kψi ψ com (s) (2.104a)
kψ Nδϕa (s)
Gclo (s, kψ ) = (2.104b)
Shg ∆clo (s, kψ )

In here the characteristic polynomial ∆cloi (s, kψi ) is given by Eq.2.89.

From the above transfer functions, the mathematical diagram of the VOR
hold can be simplified as one shown in Fig.2.51.
Because the controlled variable is the offset angle λ(s), thus the transfer
function between λ(s) and ψ ref needs to be expressed as follows:
2.4. LATERAL-DIRECTIONAL AUTOPILOT 89

λref ≅ 0 ψ comm ψ ( s)
+
εˆλ
−
Gcop Gclo ( s, k *ψ i )

Aircraft and heading hold

ψ

λ (s) ψˆ ref
+
Gng −

Figure 2.51: Navigation and guidance system: VOR hold

Nψλr (s)
λ (s) = ψ ref (2.105)
∆ng (s)
where:

Vp ³ ´
Nψλr (s) = 57.3 s∆cl s, kψ∗ i (2.106a)
R

³ ´
∆ng (s) = s2 ∆cl s, kψ∗ i − kλ (s + 0.1) Nδϕa (s) (2.106b)
∗
Vp kψi
kλ = 57.3 Kc (2.106c)
R Shg
The characteristic equation for λ(s)/ψ ref (s) is given by

(s + 0.1) Nδϕa (s)

1 − kλ ³ ´ =0 (2.107)
s2 ∆clo s, kψ∗ i

The root locus of the above characteristic polynomial is illustrated in Fig.2.52

for the gain value of kλ < 0. From the root locus, if the following working point
associated with kλ∗ = −1.1152 is taken, the following characteristic polynomial
is obtained:
∆(s, kλ∗ ) = s8 + 12.2329s7 + 24.797s6 + 65.3037s5 + 73.7215s4
+73.6583s3 + 33.434s2 + 7.7049s + 0.7495
90 CHAPTER 2. AUTOPILOT SYSTEM

Root Locus
2

1.5

0.5
Imaginary Axis

-0.5

-1

-1.5

-2
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
Real Axis

Figure 2.52: Root locus diagram for the VOR offset λ(s) with respect to the
bearing Ψref of the N250-100 aircraft

with the roots:

−0.2436; −10.3918
−0.169 ± j1.79; −0.43 ± j1.01; −0.1998 ± j0.1897

The time response of the offset angle λ(t) can be obtained using Eq.2.105

⎡ ³ ´ ⎤
λ (t) s∆clo s, kψ∗ i
= £−1 ⎣ ψ ref ⎦ (2.108)
V
57.3 Rp ∆ng (s, kλ∗ )
2.4. LATERAL-DIRECTIONAL AUTOPILOT 91

1
offset angle lambda(rad)

0.5

-0.5
0 5 10 15 20 25 30 35 40

Figure 2.53: Time response of λ(t) of the VOR Hold system of the N250-100
aircraft with gains of: kλ∗ = −1.1152 (guidance loop), kψi
∗
= 4.5943 (outer loop
control) and ki = 8.9344 (inner loop control)
92 CHAPTER 2. AUTOPILOT SYSTEM
Chapter 3

Stability Augmentation
System

3.1 Introduction
A stability augmentation system for an aircraft is designed to improve dynamic
characteristics of the aircraft in general. The improvement is often also necessary
for a certain critical flight conditions such as take-off, landing or flying in a bad
weather. A number of stability augmentation system (SAS) typically used for
conventional transport aircraft are:

1. Longitudinal mode

(a) Pitch damper

(b) Phugoid damper

2. Lateral Directional mode

(a) Yaw damper

(b) Roll damper
(c) Spiral damper

3.2 Working Principle of Stability Augmenta-

tion System
A stability augmentation system is a built-in system operated only by an on/off
button. Fig.?? shows the yaw damper type stability augmentation system of
a twin turboprop engine transport aircraft, Indonesian Aerospace N250-100
”Krincing Wesi”. To activate the SAS system, the pilot can simply press the
ON button. Once the SAS is engaged, the aircraft stability in a certain mode

93
94 CHAPTER 3. STABILITY AUGMENTATION SYSTEM

will be improved, for instance its damping or natural frequency. While SAS
is working, the pilot can freely move the controller wheel/stick/pedal without
disturbing the SAS system. Thus pilot input can co-exist with SAS operation.

AP + SAS
SAS servo aircraft
computer
δT
θ ref δe
u (t )
+ + α (t )
Amp θ (t )
− −
q(t )
qm
Pitch damper
loop

θm rate gyroscope

Vertical gyroscope

Figure 3.1: The functional diagram of the pitch damper system

For statically stable aircrafts, SAS is typically used only for a certain flight
region or configuration such as approach and landing or high speed cruise. For
other than the above flight zones, the SAS is typically turned-off. Therefore,
for statically stable conventional aircrafts, the SAS is referred to as a limited
authority system. For non-conventional aircrafts where the static stability is
relaxed or completely removed, this stability augmentation system is commonly
called Super Stability Augmented System. This kind of system is definitely
working continuously, full-time, since without the system the aircraft will not
be able to fly. The system works to provide basic stability to the aircraft and
thus can not be over-ridden by the pilot. Therefore, this system is often referred
to as having a full authority. This kind of Super SAS is used for instance in
the fighter aircrafts namely Lockheed Martin F-16 ”Fighting Falcon”, Sukhoi
SU-27/35, Rafale, SAAB-Grappen among others.

3.3 Longitudinal Stability Augmentation System

The longitudinal stability augmentation system has the goal to increase the
damping ratio of the pitch oscillation mode while maintaining the value of the
3.3. LONGITUDINAL STABILITY AUGMENTATION SYSTEM 95

AP servo SAS servo Aircraft

δT
u (s)
θ ref εˆθ qref + εˆ [ N long ( s )]
+
G AP ( s ) GSAS ( s ) α (s)
−
−
δe Δ long ( s ) θ (s)
qm q(s)
Pitch damper
Altitude loop
hold

S rg
Rate gyro
θm
S vg

Vertical gyro

Figure 3.2: The mathematical diagram of the pitch damper system

damping ratio of the phugoid mode, and vice versa. The following Table list the
possible applications of feedback to increase the longitudinal stability, based on
experience:
In the previous chapter, the root locus configuration for longitudinal feed-
back with positive/negative gain has been shown for an ideal condition. The
ideal condition assumed that the dynamics of the control mechanism and sensor
instruments can be represented a constant gain. If the dynamics of the control
mechanism and sensor is taken into consideration, the root locus configuration
can be different.

3.3.1 Pitch Damper System

The pitch damper system is designed to increase the damping ratio of the pitch
oscillation mode. The pitch rate is sensed by the vertical rate gyro and the
measured signal is sent to the SAS computer in order to improve the pitch
oscillation mode. Refer to the Fig.3.1 in the following.
The figure shows the functional diagram describing the use of the pitch
damper system as an inner control loop of the pitch attitude hold autopilot.
The pitch damper measures the pitch rate by using the rate gyro. The mea-
sured signal is then sent to the SAS computer which in turn outputs the control
command signal to the elevator servo. The elevator is moved in a way that
stabilizes the aircraft in the pitch direction. In the above example, the elevator
96 CHAPTER 3. STABILITY AUGMENTATION SYSTEM

Table 3.1: Feedback for improving longitudinal stability

No Objective Feedback Side effect Stability derivative

improvement
1 Increasing q −→ δe natural fre- mq
the pitch quency of the
oscilla- pitch oscilla-
tion damp- tion slightly
ing (pitch increases
damper)
2 Increasing θ −→ δe the damping mu
the phugoid u −→ δe of the pitch
damping oscillation
(phugoid decreases
damper)
3 Increasing α −→ δe mα
the pitch nz −→ δe
oscillation
natural
frequency

servo has a task to receive command from the SAS computer only. In a num-
ber of large size aircrafts, the AP and SAS servos are usually separated. The
mathematical diagram of the pitch damper system is illustrated in Fig.3.2. If
the SAS servo can be assumed to have a fast time response, it can be modeled
as a constant gain:

GSAS (s) = Kq , constant (3.1)

The AP servo has typically a slower time response, thus it can be assumed to
have a first order lag model:
Kθq
GAP (s) = (3.2)
s + 1/τ s

The inner loop of the pitch damper system can be expressed as:

[Nix (s)]
x̄(s) = q̄ref (3.3)
∆i (s, Ks )

where,
£ ¤
[Nix (s)] = Kq Nδxe (s) (3.4a)
∆i (s, Ks ) = ∆long (s, Ks ) + Kq sNδθe (s) (3.4b)
3.3. LONGITUDINAL STABILITY AUGMENTATION SYSTEM 97

Root Locus
2.5
0.85 0.76 0.62 0.44 0.22
0.92
2

1.5
0.965
pitch oscillation
1

0.992 phugoid
0.5
Imaginary Axis

4 3 2 1
0

-0.5
0.992

-1
0.965
-1.5

-2
0.92
0.85 0.76 0.62 0.44 0.22
-2.5
-5 -4 -3 -2 -1 0 1 2
Real Axis

Figure 3.3: Root locus of the inner control loop: pitch damper system of N250-
100 PA-2 at cruise condition with V = 250 KIAS and h = 150000

From the above equation, the pitch damper loop characteristic polynomial
can be given as the following equation.

sNδθe (s)
1 + Kq =0 (3.5)
∆long (s)
As a study case, the pitch damper system design for the N250-100 aircraft
during cruise at V = 250 KIAS and altitude h = 150000 will be investigated.
The modeling data was presented in Eqs.2.9b—2.12 in Sec.2.3.1. The root locus
diagram for the pitch damper system of the N250-100 is presented in Fig.3.3 for
the negative value of the gain. To analyze the effectivity of the pitch damper,
the following gain value is selected and the the associated poles are presented.

Kq∗ = −0.2432 (3.6)

p1,2 = −3.5657 ± j0.5627 ; ς po = 0.988 , ω npo = 3.61
p3,4 = −0.0092 ± j0.0646 ; ς ph = 0.141 , ω nph = 0.0652

The inner loop characteristic equation in conjunction with the selected gain
98 CHAPTER 3. STABILITY AUGMENTATION SYSTEM

Root Locus
0.15

0.1

0.05 phugoid
differentiator
Imaginary Axis

-0.05

-0.1

-0.15

-0.2 -0.15 -0.1 -0.05 0 0.05 0.1 0.15 0.2

Real Axis

Figure 3.4: Root locus of the inner control loop: pitch damper system of N250-
100 PA-2 at cruise condition with V = 250 KIAS and h = 150000 − enlarged
around the phugoid mode

is
∆i (s, Kq∗ ) = s4 + 7.1497s3 + 13.1661s2 + 0.2702s + 0.0554 (3.7)
With the above data, the time response for θ(t) and q(t) with and without
pitch damper can be calculated and compared.
The response of θ(t) and q(t) with the pitch damper is given by the following
equation:
" #
−1
Kq∗ Nδθe (s)
θ (t) = £ q (3.8)
∆i (s, Kq∗ ) ref

∙ ¸
−1
Kq∗ Nδqe (s)
q (t) = £ q (3.9)
∆i (s, Kq∗ ) ref
Fig.3.5 demonstrates the effectiveness of the pitch damper system. The
longitudinal damping for the system with and without pitch damper can be
obtained as the following:
ζ ph|without = 0.0513 −→ ζ ph|with = 0.1410
3.3. LONGITUDINAL STABILITY AUGMENTATION SYSTEM 99

2.5

2 without pitch damper

theta(rad)

1.5

1
with pitch damper
0.5

0
0 1 2 3 4 5 6 7 8 9 10

6
without pitch damper
pitch rate

2 with pitch damper

-2
0 1 2 3 4 5 6 7 8 9 10

Figure 3.5: Time response of θ(t) and q(t) with and without pitch damper of
N250-100 aircraft

ζ po|without = 0.2897 −→ ζ po|with = 0.988

It is evident from the time response that in the short period of time the
effect of ζ po is significant both in dampening the pitch and reducing the peak
of the response.
If the values are acceptable and used for the analysis of the outer loop, pitch
attitude hold, the diagram in Fig.3.2 can be simplified as that shown in Fig.3.6
in the following:
However, as alluded previously, the δ e −→ θ feedback represents the phugoid
damping augmentation system. Thus the diagram in Fig.3.6 can be regarded as
the diagram of phugoid which will be discussed next.

3.3.2 Phugoid Damper

The outer loop of the functional diagram given in Fig.3.1 shows the diagram
for the phugoid damper. The mathematical diagram of the phugoid damper is
presented in Fig.3.6. With the model of the autopilot GAP (s) given in Eq.3.2,
100 CHAPTER 3. STABILITY AUGMENTATION SYSTEM

θ ref εˆθ u (s)

+ * α ( s)
−
G AP ( s ) Gcli ( s , k i ) θ (s)
q(s)
AP servo Aircraft
+ Pitch damper

S vg

Vertical gyro

Figure 3.6: The mathematical diagram of the pitch attitude hold with the pitch
damper as the inner loop

the outer characteristic equation of the phugoid damper is given by:

sNδθe (s)
1 + Kθ =0 (3.10)
∆i (s, Kq∗ ) (s + 1/τ s )
where

Kθ = Kq∗ Kθq (3.11)

The root locus of the phugoid damper system is presented in Fig.3.7. As an

example the following design point is taken:

Kθ∗ = 3.7979 (3.12)

The poles are obtained as:

p1 = −10.2807 (3.13)
p2,3 = −3.3549 ± j1.3043 ; ς po = 0.932
p4,5 = −0.0796 ± j0.0307 ; ς ph = 0.933

With the above design point, the phugoid damper characteristic polynomial
can be written as:
3.3. LONGITUDINAL STABILITY AUGMENTATION SYSTEM 101

Root Locus
10

8
autopilot servo
6

2
Imaginary Axis

-2 phugoid
pitch oscillation
-4

-6

-8

-10
-12 -10 -8 -6 -4 -2 0 2
Real Axis

Figure 3.7: Root locus of outer control loop: phugoid damper of N250-100 PA2
at cruise condition

∆o (s, Kθ∗ ) = s5 + 17.1497s4 + 84.6632s3 + 146.5079s2 + 21.8273s + 0.9710

(3.14)

And the time response θ(t) can be calculated using the following equation:

" #
Kθ∗ Nδθe (s)
θ (t) = £−1 (3.15)
∆o (s, Kq∗ )θref

The following figure describes the time response of θ(t) of the phugoid
damper system. Compare the result to Fig.2.10, the pitch attitude hold without
the phugoid damper. For the pitch attitude hold system without pitch damper
(See Sec.2.3.1), there exits oscillation at the early stage of time response. This
oscillation does not occur in the system with the pitch attitude damper as the
inner loop. However the settling time gets longer. The settling time for a system
without pitch damper is 12 time units whereas for a system with pitch damper
it is 65 time units.
102 CHAPTER 3. STABILITY AUGMENTATION SYSTEM

Root Locus

0.1

phugoid
0.05
Imaginary Axis

-0.05

-0.1

-0.16 -0.14 -0.12 -0.1 -0.08 -0.06 -0.04 -0.02 0 0.02

Real Axis

Figure 3.8: Root locus of outer control loop: phugoid damper of N250-100 PA2
at cruise condition—enlarged around phugoid mode

From the flight test experience, it can be deduced that N250-100 PA2 has a
very good longitudinal dynamic characteristic, thus the pitch damper system is
not needed.

3.4 Lateral-Directional Stability Augmentation

System
The stability augmentation system in the lateral directional mode is designed
primarily to increase the damping ratio of the dutch roll and to reduce the time
response of the roll subsidence and spiral modes without degrading the stabil-
ity of the other modes. Examples of lateral directional stability augmentation
systems are presented in the following table.

3.4.1 The Dutch-roll stability augmentation: Yaw Damper

The damper system is designed to increase the damping ratio of the dutch roll
mode within which the dominant factors are the side slip angle β(t) and the
3.4. LATERAL-DIRECTIONAL STABILITY AUGMENTATION SYSTEM103

Table 3.2: Feedback for improving lateral directional stability

No Objective Feedback Side effect Stability derivative

improvement
1 Increasing r −→ the spiral mode is nr
the dutch roll δr β −→ also stabilized
damping (yaw δr
damper)
2 Lowering the p −→ δa the spiral is stabi- lp
roll subsidence lized and the dutch
time response roll is sustained by
(roll damper) the aileron
3 Stabilizing the ϕ −→ δa the dutch roll time
spiral mode response increases
r −→ δa the dutch rolll time
response decreases,
dutch roll got un-
stable
β −→ δa the dutch roll
got unstable, the
roll time response
slightly decreases
4 Providing co- β −→ the spiral destabi- nβ
ordination to δr ny −→ lized and the dutch
rudder δr roll frequency in-
creases
5 Reducing the δa −→ δr
adverse yaw (ARI)
104 CHAPTER 3. STABILITY AUGMENTATION SYSTEM

0.4

0.35

0.3

0.25
theta(rad)

0.2

0.15

0.1

0.05

0
0 50 100 150

Figure 3.9: Time response of θ(t) of the phugoid damper with Kθ∗ = 3.7979 and
Kq∗ = −0.2432

yawing angle ψ(t). In operation, this system is commonly known as the Yaw
damper. There are two methods to increase the Dutch-roll damping namely by
the feed-back of yaw rate r and by the feedback of side slip rate β̇ to rudder
deflection δr.

The yaw damper through the feedback of r −→ δr

In this type of yaw damper, the yaw rate is measured by the directional rate
gyro and the measurement output signal is then processed in the SAS computer
and sent out as the control command signal to the rudder actuator. Physically
the rudder deflection motion will continuously make corrections or counter the
dutch roll motion (side-slip/yawing) thus providing the damping. Figs.3.10 and
3.11 show the functional and mathematical diagram of the yaw damper with the
yaw rate feedback respectively. Note that an electric circuit component called
the Wash-out circuit is installed in between the rate gyro and SAS computer.
The wash-out circuit is a first order lead-lag designed to make the measured yaw
rate from the rate-gyro transient which means that it only works shortly from
the start of the maneuver that produces r(t). After the steady state is reached,
the signal r(t) is removed by the wash-out circuit. The objective of the rm (t)
3.4. LATERAL-DIRECTIONAL STABILITY AUGMENTATION SYSTEM105

elimination in the steady state is to make the yaw damper not in opposition
to the pilot input during the the stationary turning maneuver. If the rm (t)
still exists, it will work against the pilot and will make the coordinated turn
difficult to achieve. Therefore, this yaw damper is designed only to enhance the
dutch-roll damping in the transient condition such as the period of entering the
gust or turbulence or the landing phase with ever decreasing speed.

SAS computer Rudder servo

aircraft
δa
rref β (t )
+ εr δr p (t )
− r (t )
ϕ (t )

erm er r (t )

Wash-out circuit rate gyroscope

Figure 3.10: The functional diagram of the yaw damper system with r −→ δr
feedback

The components of the yaw damper system transfer function are:

1. Aircraft, GA/C (s), for the r −→ δr then
x
Nld (s) x̄(s)
GA/C (s) = = (3.16)
∆ld (s) δ r (s)
x̄ = { β p r ϕ }T

2. Rate gyro, Grg (s)

Grg (s) = Srg ≡ constant (3.17)
rm (s)
=
r(s)
106 CHAPTER 3. STABILITY AUGMENTATION SYSTEM

Pilot
input
δa β (s)
r ref +
εˆr [ N ld ( s )] p(s)
Gact ( s )
−
δr Δ ld ( s ) r (s)
ϕ (s)
Rudder servo aircraft
+ control mechanism

r m rm
Gwo ( s ) Grg ( s )

Wash-out Rate-gyro

Figure 3.11: The mathematical diagram of the yaw damper system with r −→ δr
feedback

The rate gyro is assumed to have a very fast time response, especially with
the availability of the advanced technology such as ring laser gyro or fiber
optic gyro. Thus it can be modeled by a constant gain

3. Wash-out circuit, GW O (s), to eliminate the stationary cnodition. The

transfer function can be given as follows

nwo (s) s/τ wo

Gwo (s) = = kwo
dwo (s) (s + 1/τ wo )
r̂m (s)
= (3.18)
r(s)

by using the lead-lag term above, in the steady state condition r̂m (s) −→ 0

4. Control mechanism and yaw damper servo

Similar to previous cases, the yaw damper control mechanism is modeled
3.4. LATERAL-DIRECTIONAL STABILITY AUGMENTATION SYSTEM107

as a first order system as:

1/τ s Kct 1
Gact (s) = Kct = (3.19)
s + 1/τ s τ s dact (s)
δ r (s)
=
εr (s)

The polynomial characteristic of the yaw damper system can be written as:

∆cl (s) = dact (s)dwo (s)∆ld (s) + Kψ Nδrr (s)nwo (s) (3.20)
Kct Kwo Srg
Kψ ≡
τ s τ wo

Therefore, the closed loop characteristic equation of the yaw damper system
is given as:

Nδrr (s)nwo (s)

1 + Kψ =0 (3.21)
dact (s)dwo (s)∆ld (s)

The locus plot of the roots of the above equation represents the root locus
of the yaw damper system.
As a case study, the application of the yaw damper to the N250-100 PA-2
Krincing Wesi aircraft will be presented. The aircraft is equipped with the
yaw damper system to increase the ride quality particularly during the flight in
turbulence and low-speed landing phase.
The transfer functions of the N250-100 due to the rudder input δ r is given
as follows:

³ s ´³ s ´³ s ´
Nδβr (s) = Sβδr −1 +1 +1
0.0193 1.8834 38.6989

³ s ´³ s ´
Nδϕr (s) = Sϕδr −1 +1
4.3266 2.6781

³ s ´µ s
¶µ
s
¶
Nδrr (s) = Srδr +1 +1 +1
1.8552 0.0781 + j0.3926 0.0781 − j0.3926

Nδpr (s) = sNδϕr (s) (3.22)

where, the static sensitivity coefficient is given by:

108 CHAPTER 3. STABILITY AUGMENTATION SYSTEM

Root Locus
10
0.66 0.52 0.4 0.26 0.12 No wash-out
8 0.8

6 dutch roll
0.9
4
0.97
2
Imaginary Axis

10 8 6 4 2
0 spiral

-2
0.97
-4
actuator
0.9
-6

-8 0.8 roll
subsidence
0.66 0.52 0.4 0.26 0.12
-10
-10 -8 -6 -4 -2 0 2 4
Real Axis

Figure 3.12: Root locus of the yaw damper system with r −→ δ r feedback of
N250-100 aircraft

Sβδr = −2.44906 (3.23)

Sϕδr = −235.6588 (3.24)
Srδr = −19.560105 (3.25)

and the characteristic polynomial is:

s4 s3 s2 s
∆ld (s) = + + + +1 (3.26)
0.0381 0.01706 0.01544 0.01004

with the following characteristic roots

3.4. LATERAL-DIRECTIONAL STABILITY AUGMENTATION SYSTEM109

Root Locus
4

2 wash-out zero
and spiral pole
dutch roll
1
Imaginary Axis

-1 roll subsidence

-2
wash-out pole
-3

-3.5 -3 -2.5 -2 -1.5 -1 -0.5 0 0.5 1

Real Axis

Figure 3.13: Root locus of the yaw damper system with r −→ δ r feedback of
N250-100 aircraft—enlarged around the dutch-roll mode

p1 = −1.9577, associated with the roll subsidence mode

p2 = −0.0101, associated with the spiral mode
p3,4 = −0.1325 ± j1.3817, associated with the dutch roll mode, where:
ω nDR = 1.39 rad/s
ς DR = 0.0955 (3.27)
Similar to the previous case study, the model for rate gyro is taken as Srg = 1.
This assumed model is considered valid as the N250-100 PA2 uses the ring laser
gyro. the actuator model has the time constant of τ s = 0.1. The transfer
function of the wash-out circuit has the time constant of τ wo = 4. The following
figure presents the root locus of the yaw damper system. Note that the dutch-roll
mode becomes more stable and the roll subsidence has the faster time response.
Fig.?? shows the root locus when the wash-out circuit is omitted.
110 CHAPTER 3. STABILITY AUGMENTATION SYSTEM

Root Locus
10 0.74
0.6 0.48 0.34 0.22 0.12
8

6 0.88

4
0.96
2
Imaginary Axis

10 8 6 4 2
0

-2
0.96
-4

-6 0.88

-8
0.6 0.48 0.34 0.22 0.12
-10 0.74
-10 -8 -6 -4 -2 0 2
Real Axis

Figure 3.14:

It is evident from the root locus that the damping of dutch roll mode is
improved while the time responses for the roll subsidence and spiral become
faster. For sufficiently large gain of kψ , the dutch roll mode can be overdamped
to break into two dutch roll subsidence modes.

The performance of the yaw damper with and without the wash-out circuit
can be compared by taking the following design points in the corresponding root
locus diagrams:

Yaw damper system with wash-out The selected gain and the associated
poles are:
3.4. LATERAL-DIRECTIONAL STABILITY AUGMENTATION SYSTEM111

kψ∗ = −7.6364 (3.28)

p1 = −7.2651
p2 = −0.0063
p3 = −0.01083
p4 = −2.7346
p5,6 = −0.8319 ± j0.5053

The closed-loop characteristic polynomial can then be obtained as:

¡ ¢
∆cl s, kψ∗ = s6 + 12.4829s5 + 47.0047s4 + 73.1907s3 + 53.7354s2 + 15.5727s + 0.0953
(3.29)

The dutch roll damper can be obtained as ζ ∗DR = 0.4273 or an increase of

347.7%

Yaw damper system without wash-out The selected gain and associated
closed-loop poles without the wash-out are:

k̃ψ∗ = −8.6359 (3.30)

p1 = −6.8279
p2 = −2.3636
p3 = −0.1917
p4,5 = −1.1749 ± j0.6620

with the closed-loop characteristic polynomial is obtained as:

³ ´
˜ cl s, k̃ψ∗ = s5 + 12.2329s4 + 46.4516s3 + 72.03s2 + 47.7434s + 6.8169
∆
(3.31)

The dutch roll damper can be obtained as ζ ∗DR = 0.4356 or an increase of

350%.
From the above data, the time response for β(t) and ψ(t) can be calculated
112 CHAPTER 3. STABILITY AUGMENTATION SYSTEM

by the following relations:

⎡ ⎤
β
d wo (s)N (s)
β(t) = £−1 ⎣kψ∗ ³ δr ´ r̄p ⎦ (3.32a)
∆cl s, kψ∗
⎡ ⎤
p
d wo (s)N (s)
p(t) = £−1 ⎣kψ∗ ³ r ´ r̄p ⎦
δ
(3.32b)
∆cl s, kψ∗
⎡ ⎤
ϕ
d wo (s)N (s)
ϕ(t) = £−1 ⎣kψ∗ ³ δr ´ r̄p ⎦ (3.32c)
∆cl s, kψ∗
⎡ ⎤
∙ ¸ ψ
r(s) dwo (s)N (s)
ψ(t) = £−1 = £−1 ⎣kψ∗ ³ δr ´ r̄p ⎦ (3.32d)
s s∆ s, k∗
cl ψ

The results of the calculated is presented in the following figure where three
cases are compared. Note from the side slip angle response that, the application
of the yaw damper enable a very effective suppresion of the response. In the
yaw response, it is evident that for the damper system without the wash-out,
the steady state ψ(t) is sustained constant at ψ ss (t) ∼ = 1 rad. Note, however
that, in the no yaw damper case, the yaw angle ψ(t) can freely increase. The
inclusion of the wash-out circuit enable the yaw damper to work only during
the transient period. After the r̂m signal is phased-out, the yaw angle ψ(t)
can drift and increase without being inhibited by the yaw damper system. It
can be concluded that by installing the wash-out circuit in the yaw damper,
the maneuvers executed by the pilot during the steady state turn will not be
opposed by the yaw damper system. Fig.3.16 shows the phase-portrait of the
roll rate with respect to the roll angle. The effectivity of the yaw damper system
in dampening down the dutch roll mode and the wash-out circuit in transmitting
the yaw rate transiently is evident.
The N250-100 PA2 and N250-100 PA1 both implements the yaw damper
with the wash-out to reduce the pilot load during the turn maneuver and flight
into turbulence.
3.4. LATERAL-DIRECTIONAL STABILITY AUGMENTATION SYSTEM113

2
yaw rate

0 without yaw damper

yaw damper
-2 yaw damper + washout
-4
0 5 10 15 20 25 30
slideslip angle beta

100

-100
0 5 10 15 20 25 30
4
yaw angle psi

0 5 10 15 20 25 30

Figure 3.15: Time response of ψ̇(t), β(t) and ψ(t) of the N250-100 PA2 equipped
with the yaw damper
114 CHAPTER 3. STABILITY AUGMENTATION SYSTEM

6
without yaw damper
yaw damper
5
yaw damper + washout

3
yaw angle

-1
-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5
yaw rate

Figure 3.16: Phase portrait of ϕ(t) vs p(t) for three cases: no yaw damper, with
yaw damper and yaw damper+wash-out
 Part II

Modern Approach

115
Chapter 4

Introduction to optimal
control

This part of the book focuses on the discussion of optimal control, mainly from
a theoretical point, a little from the stand point of computer based algorithms
for actually solving such problems. A typical problem of our interest would be
for example "how do we bring the space shuttle back into the atmosphere in such
a way that we arrive lined up at the runway at Edwards and do so in such a way
that we optimise something (i.e. minimize the total heat on the body,...) subject
to some constraints (i.e. 4g max for normal acceleration or total acceleration)".
In a case like that we formulate the problem with the dynamics of the aircraft,
we might choose angle of attack as control variables, in this way we can deal
with just a point of mass dynamics of the vehicle. We do not have to worry
about the attitude. The solution that we will get would be just a time history
of the optimal angle of attack and bank angle. This will not constitute a control
system in the feedback sense. Most of the results that we derive will be of that
sort, open loop control histories that optimize some criteria.

4.1 Some Preliminaries

There are only a few cases in where the form of the solution comes out to
naturally in terms of feedback law, which then gives rise to a control system.
The cases in which that happens are those where the system dynamics are linear
and the criteria is quadratic. We are going to formulate these problems in two
basically different ways:

1. The classical variational calculus

2. Dynamic programming

117
118 CHAPTER 4. INTRODUCTION TO OPTIMAL CONTROL

We will deal with both if these. When an analytical solution exists, both
will give rise to the same answer as they must. fi rst, let us consider some
preliminary materials. For a problem to be completely defi ned we need to have
the following:
1. Cost function
The criteria that you wish to minimize or maximize. We will be dealing
with the classical and the most common form of optimization, where we
have a single scalar criteria that we wish to minimize or maximize. When
we are presented with a multi-criteria problem, we will treat it by combin-
ing weighted concerns and adding them them up into a single criteria that
may reflect all of them in some way. In an atmospheric re-entry problem,
for instance, we might be concerned with heat load, acceleration history
and control activity.
2. Model of the system to be controlled
3. Description of constraints.
Constraints come in many different forms. We might have limits on our
control varibles. There may also be constraints of a more diffi cult form
such as bounds on system state variables. In re-entry problem for example,
we may want to restrict the normal acceleration to a maximum 4g, this is
a more diffi cult kind of constraint to deal with. Very typically, there will
be terminal constraints which are not so hard to deal with.
Those are all the elements of the description of the problem. We will be
dealing with classical variational problem meaning there is not anything to be
controlled. The typical variational problem would be formulated as follows:
·
J = J(x, x, t) (4.33)
we simply want to fi nd the histories of these functions which minimize or
maximize some criteria. A classical problem of this sort would be: fi nd the
curve on the surface of a sphere which is a minimum length that joints two
points on the surface of the sphere.
If however, we pose a control problem, then typically we will have a criteria
which will depend upon state variables, and may be time (time is usually the
independent variable). In this case:
J = J(x, u, t) (4.34)
But in addition to that, we will have a constraint which is in the form of the
dynamics of the system to be controlled:
·
x = f (x, u, t) (4.35)
In other words the system dynamics constitute a set of contraints that relate
the control variables and state variables. There might be other constraints too.
Also in the classical variational problems, a set of constraints can be present.
4.2. LINEAR SYSTEMS. 119

So historically, the variational calculus was devised to treat problems of type

[4.79], and we will look at those. But we will also consider control problems of
type [4.34] where we do have a set of system dynamics to deal with [4.35]. And as
well we will look at control problems via the method of dynamic programming.
If there are limits on the controls, in a control problem, then we say that
any history of control which satisfi es the constraints will be called admissible
controls. On the other hand, any trajectory that satisfi es all the constraints
will be called a feasible trajectory.
In terms of constraints that are very often given, we need to think about
initial and terminal conditions. When we formulate an optimal control problem,
the initial conditions are usually given. I.e. we are starting from some initial
state, how do we go from there to some terminal state?
The fi nal state is very often either constrained or penalized in the cost
function. A typical form of cost function is the following:
Z tf
J = h[x(tf ), tf ] + g[x(t), u(t), t]dt (4.36)
t0

It is a function of the terminal states and an integral over the interval of the
solution of some function of the states, the controls and time itself. Alterna-
tively, we could have the following form:
Z tf
J= g[x(t), u(t), t]dt (4.37)
t0

m[x(tf ), tf ] = 0 (4.38)
In this case we do not have a terminal term, but we have constraints [4.38].
These two formulations might give rise to almost the same solution. We can
either constrain those terminal conditions if we really want to hold them exactly
[4.37 and 4.38], or in many cases we do not really have to hold the terminal
condition exactly, it is just that we want them to be about something, and in
that case we could give up the constraints and instead just penalize departures
from the desired terminal conditions in the cost function [4.103]. So this is the
nature of a soft constraint, where the penalty in the cost function mean you have
to come close to those conditions but you do not have to meet them exactly. In
many cases, this is a better formulation.
We will be specifi cally dealing with system dynamics model which is uni-
versally treated as a set of ordinary differential equations. We will take lienar
systems as a special case of general systems and the results that we get will sim-
plify considerably in that case. In what follows, we will review some properties
of linear systems.

4.2 Linear Systems.

·
x = A(t)x(t) + B(t)u(t) (4.39)
120 CHAPTER 4. INTRODUCTION TO OPTIMAL CONTROL

One of the main advantages of dealing with linear systems rather than a
general system is that we can write down the form of the solution immediately.
The solution form is in this case:
Z t
x(t) = Φ(t, t0 )x(t0 ) + Φ(t, τ )B(τ )u(τ )dτ (4.40)
t0

The fi rst term represents the transition from the initial time to the fi nal time
of the initial conditions. The second term represents the effect of the control
over that time. Φ is called the transition matrix, which satisfies the following
differential equation:
dΦ(t, τ )
= A(t)Φ(t, τ ) (4.41)
dt
And the boundary conditions at the beginning of the interval:
Φ(τ , τ ) = I (4.42)
We will be able to take advantage of that solution form and we will do that.
When we are linearizing along the dynamic trajectory, it automatically brings
to a time varying linear description. So that will be the most common form of
linearization that we will consider and it will be time varying. If we are dealing
with a plant that truly is invariant and linear, then:
·
x = Ax(t) + Bu(t) (4.43)
where A and B are constant now. In this case the differential equation is a
set of fi rst order differential equations with constant coeffi cients; the solution
of [4.41] is an exponential. So it has the form of the matrix exponential:
1
Φ(t, τ ) = eA(t−τ ) = I + A(t − τ ) + A2 (t − τ )2 + ... (4.44)
2

4.2.1 Controllability
Definition 5 A plant is considered controllable if it is possible to drive any
initial state to the origin in a f inite time with a f inite control.
This defi nition applies to linear systems, nonlinear systems and time varying
systems. In linear system, driving from a general initial point in the state space
to the origin, is equivalent to the ability to drive from anyone initial point in
the state space to any other fi nal point in the state space. This equivalent is
not true for nonlinear systems. Especially for the invariant case, an equivalent
test is in terms of the controllability matrix:
M = [B|AB|A2 B|...|An−1 B] (4.45)
So this is a matrix which has n rows, where n is the dimension of the state
and n or more colums, depending on whether there is more than one control
or not. The system is controllable if the rank of the matrix is n,which is the
highest possible rank.
4.2. LINEAR SYSTEMS. 121

4.2.2 Conventions for Derivatives

Let us defi ne H as a scalar function of the vector x. Then the derivative of the
scalar H with respect to the vector x is :
dH
= Hx (4.46)
dx
It can be rewritten as:
∂H ∂H ∂H
Hx = [ , , ..., ] (4.47)
∂x1 ∂x2 ∂xn
The fi rst variation in this scalar function H can be written as:
∂H ∂H ∂H
δH = δx1 + δx2 + ... + δxn (4.48)
∂x1 ∂x2 ∂xn

or:
∂H
δH = δx = Hx δx (4.49)
∂x
where
⎡ ⎤
δx1
⎢ δx2 ⎥
⎢ ⎥
δx = ⎢ .. ⎥ (4.50)
⎣ . ⎦
δxn

For a function of more than one variable, we can write:

H = H(x, u) (4.51)

where H is a function of both the states and the controls. In this case, we
can defi ne both partial derivatives as:
∂H ∂H ∂H ∂H
= Hx = [ , , ..., ] (4.52)
∂x ∂x1 ∂x2 ∂xn
∂H ∂H ∂H ∂H
= Hu = [ , , ..., ] (4.53)
∂u ∂u1 ∂u2 ∂un
The second mixed derivative can be stated as:
∂2H ∂2H
= Hxu = [ ] (4.54)
∂x∂u ∂xi ∂uj
They can also be taken in either order, so:
∂2H ∂2H T
= Hux = [ ] = Hxu (4.55)
∂u∂x ∂ui ∂xj
122 CHAPTER 4. INTRODUCTION TO OPTIMAL CONTROL

Note that the above expression is true only if the function H is continuously
differentiable to the second order. If we are taking the derivative of a vector
value quantity with respect to a vector value quantity, we will have have a two
dimensional array:

⎡ ⎤
← ∂z∂x →
1

dz ⎢ . ⎥
=⎢
⎣ .. ⎥
⎦ (4.56)
dx
← ∂z m
∂x →

Now, let us take an example of a particular function that we will use fre-
quently:

H(x, u) = xT Au (4.57)

The fi rst derivative with respect to the vector u is a row with the following
elements:

Hu = xT A (4.58)

The derivative with respect to the vector x can be expressed by writing the
transpose of H without changing anything, since it is a scalar:
∂ T T
Hx = (u A x) = uT AT (4.59)
∂x
The second order derivatives are:
∂ ∂ T ∂ T
Hxu = [ (x Au)] = (x A) = A (4.60)
∂x ∂u ∂x

and
∂ ∂ T ∂ T T
Hux = [ (x Au)] = (u A ) = AT (4.61)
∂u ∂x ∂u
When we have a quadratic form that we are going to deal with frequently:

H(x) = xT Ax (4.62)

Taking the derivative with respect to x we have:

d T
Hx = xT A + (x Ax) = xT A + xT AT (4.63)
dx

If A is symmetric then:

Hx = 2 xT A (4.64)
4.2. LINEAR SYSTEMS. 123

Figure 4.1: Minimum of a cost function J(x) at a stationary point

4.2.3 Function Minimization

We are talking about fi nding the minimum value of a function J :

J = J( x) (4.65)

We do not have a dynamic situation and thus J is a static function and x

is a vector of parameters. We want to fi nd the values of those parameters that

gives us the minimum value of J. Let us take the case of a scalar parameter. In
this case we have:

There is a bounded range of x. In general a minimum of J within a bounded

range can occur under anyone of three conditions.

1. The minimum can occur within the admissible range (fi g. 4.1). It occurs
at a stationary point

2. It could happen at the boundary (fi g. 4.2). In this case there are no
stationary point, there might be but they are not the minimum.

3. It could happen at a corner (fi g. 4.3).

In general minima can be local or global. A global minimum is a point

where:

J( x∗ ) ≤ J( x) ∀x admissible (4.66)
124 CHAPTER 4. INTRODUCTION TO OPTIMAL CONTROL

(a)

Figure 4.2: Minimum of J(x) at the boundary

(a)

Figure 4.3: Minimum of J(x) at a corner

4.2. LINEAR SYSTEMS. 125

This is the solution we like to have. However the necessary conditions that
we derive and the algorithm that we define for searching the minima almost
always can only as sure as a local minima. A local minima is a point where:

J( x∗ ) ≤ J( x) ∀x in the vicinity of x∗ (4.67)

That only means that there is some region around x∗ , we do not know if it
is large or small, where the function is larger than at x∗ .
Now we want to concentrate only on stationary conditions which is the most
common case for aerospace problems. We will eventually look at problems where
we have bounded controls and the solution may lay on the boundaries. Now let
us concentrate on stationary points.

Definition 6 The stationary point is a point where the function is stationary

(constant) to f irst order.

That does not mean that the function is constant everywhere, but if it is
constant to fi rst order that means that there is some range around x∗ where an
expansion of the function would be dominated by the fi rst order term and the
fi rst order terms are zero in this case. If x is a scalar :
dJ
|x=x∗ = 0 (4.68)
dx
Now we can write the difference:
1 d2 J
J( x∗ + δx) − J( x∗ ) = |x=x∗ δx2 + ... (4.69)
2 dx2

it is dominated by the second derivative. Thus there will be some region

around x∗ , where the local behavior of the function is described in terms of
the second derivative. Eq. 4.68 is called fi rst order necessary condition. For a
minimum, it is necessary that the second order derivative is positive:

d2 J
|x=x∗ ≥ 0 (4.70)
dx2
This is called second order necessary condition. If the second order derivative
is not zero and it is strictly positive then we are assured that it is a local
minimum. Tha is a second order sufficient condition:
d2 J
|x=x∗ > 0 (4.71)
dx2
For n dimensions, we can extend those ideas by writing the variation of J
in terms of:
dJ
δJ = δx to fi rst order (4.72)
dx
126 CHAPTER 4. INTRODUCTION TO OPTIMAL CONTROL

The fi rst order necessary conditions is:

dJ
|x=x∗ = 0 (n conditions) (4.73)
dx

The second order term can be written as:

1 T
J( x∗ + δx) − J( x∗ ) = δx Jxx δx (4.74)
2

If Jxx is positive semi-defi nite then the quadratic form on the right side of
Eq.4.74 is positive or zero:

Jxx ≥ 0 (4.75)

This is a second order necessary condition for a local minimum. A test for
semidefi niteness like Eq.4.75 would be that all the eigenvalues of Jxx must be
greater or equal to zero. And again if Jxx is strictly positive defi nite.

Jxx > 0 (4.76)

then we have a suffi cient condition for a local minimum. Again that is
really the defi nition of positive defi niteness, it is a matrix such that if you form
quadratic form the result will be positive for all δx.

4.3 Constrained Problems

We left the discussion in the previous session concentrating on stationary points
which are governed by the necessary condition:

dJ
|x=x∗ = 0 (4.77)
dx

That is for the case where there are no constraints in the problem. Now
we want to look at the issue of constrained problems. That is, problems of the
following form:
½
min J(x) (x has dimension n)
(4.78)
subject to: f (x) = 0 (f has dimension m) m < n

Thus the constraints have the form f (x) = 0. In this case, m must be
less than n, because each scalar constraint removes one degree of freedom to
the problem, and if we remove n degrees of freedom we completely specify the
problem if the constraints are independent. Now how do we go about recognizing
constraints? Fi rst, we will look into the possibility of elimination.
4.3. CONSTRAINED PROBLEMS 127

4.3.1 Elimination
What we mean by the elimination is the possibility that you can actually solve
the constraints for m of the x0 s, in terms of the remaining n−m x0 s. Therefore
reducing the dimension of the problem and automatically satisfy the constraints.
Suppose we partition x :
∙ ¸
x1 }m
x= (4.79)
x2 }n − m

Then if it is possible to take the m equations:

f (x) = 0 (4.80)

and solve it for:

x1 = g(x2 ) (4.81)

then we can write the cost function:

J(x) = J(x1 , x2 ) = J[g(x2 ), x2 ] = J 0 (x2 ) (4.82)

so this is just another function of J 0 (x2 ). This is now an unconstrained mini-

mization problem in a lower dimension. For every value of x2 we can fi nd the
value of x1 from Eq.4.81. And therefore the complete solution that does satisfy
the constraints. We can always do this approach if the constraints are linear
and they are independent. However, we might not be able to eliminate the
constraints explicitly for some classes of problems. We are going to need a more
general approach than this and that is the method of Lagrange.

4.3.2 Method of Lagrange

Let us look at a single function J of two variables J(x1 , x2 ). Suppose that the
unconstrained minimum occurs at the point A :
The level curves represent contours of constant value of J. The fi rst order
change in the function J is given by:
dJ
δJ = δx to fi rst order (4.83)
dx

Since these are level curves, the column JxT must be orthogonal to the δx in that
direction. Thus the gradient of the function J is everywhere orthogonal to the
level curves. The constraint is shown by the function f (x) = 0 which is generally
nonlinear. In this case point B, will be the point of constrained minimum.
Right at the solution at this point the two curves are tangent, which means
that the two gradients are colinear. That property that is obvious here in two
dimensions need to be generalized in higher dimensions. The cost gradient and
the constraint gradient, span the same space at point B. In higher dimension
128 CHAPTER 4. INTRODUCTION TO OPTIMAL CONTROL

Figure 4.4: Constrained minimum vs unconstrained minimum

the cost gradient at a constrained minimum must lie in the subspace that is
spanned by all of the constrained gradients. So the fundamental property of a
constrained minimum is that the cost gradient lies in the space spanned by the
constraint gradients. In this case, we can express it as a linear combination of
those constraint gradients:

dJ df1 dfm df
= −λ1 ... − λm = −λT (4.84)
dx dx dx dx

the combination coefficient are λ0 s. So this is a necessary condition at a con-

strained minimum. Together with the constraint itself, it constitutes a complete
set of necessary conditions. The way we actually use this is to defi ne a new
scalar function L :

L = J + λT f (4.85)

Then, condition Eq. 4.84 can be written just as:

dL
=0 (4.86)
dx

And here we see the standard interpretation, that because of the introduction of
some new parameters, as many parameters as there are constraints, we defi ne
a new function, an augmented cost function, which we can treat as if it were
unconstrained. Eq. 4.86 together with the constraints are the total of m + n
4.4. INEQUALITY CONSTRAINTS 129

conditions to defi ne the m + n parameters x and λ.

dL
dx =0
(4.87)
f (x) = 0

The function L is called the Lagrangian function and the λ0 s are called the
Lagrange multipliers. We can also write:
dL
f= (4.88)
dλ
which means that the necessary conditions can be written as:
dL
dx =0
dL (4.89)
dλ =0

If we defi ne a new augmented parameter y:

∙ ¸
x
y= (4.90)
λ

Then both of the statements Eq. 4.89 are included in a single statement:
dL
=0 (4.91)
dy

And this is a complete set of necessary conditions for constrained minimum.

4.4 Inequality Constraints

Now the problem can be stated as:
½
min J(x) (x has dimension n)
(4.92)
subject to: f (x) ≥ 0 (f has dimension m)

In this case we can have m ≤ n or m ≥ n and we can still have a perfectly

well defi ned minimization problem. And that because of the fact that with
inequality constraints, each constraint does not necessarily remove a degree of
freedom for the problem depending on whether the constraint is active or not.
We have a similar picture in two dimensions.
From the picture, we can see that the constraint related to f1 is not active,
it does not take away a degree of freedom out of the situation. The constraint
f2 does constrain the problem, because with respect to J we can not go further
to the left. The cost gradient lies in the space that is spanned by the gradients
of the active constraints:
½ dJ P dfi P
dx = − i λi dx − i λi dfi
dx (4.93)
(active constraints)λi ≤ 0 (inactive constraints)λi = 0
130 CHAPTER 4. INTRODUCTION TO OPTIMAL CONTROL

Figure 4.5: Inequality constraints

There is another extra condition related to the directionality of the gradient.

This puts a sign condition on λ0 s :

λi ≤ 0 (4.94)

In this way, we defi ne the same Lagrangian as before:

(
L = J + λT f ; λi ≤ 0
dL (4.95)
dx = 0

The rest of the conditions can be summarized in different ways. The usual way
of writing this is:

dL
λi =0 (4.96)
dλi

finally we need to also state that the constraints must be satisfi ed, thus:

dL
≥0 (4.97)
dλi

Eqs.[4.95,4.96,4.97] summarizes the fi rst order necessary conditions for an in-

equality constrained problem. In order for the λ0 s to be well defined, the con-
straints have to have a property of regularity which means that the constraint
gradients that are active must be linearly independent.
4.5. SENSITIVITY OF COST TO CONSTRAINT VARIATIONS 131

4.5 Sensitivity of Cost to Constraint Variations

It is common that the constraints are not really precisely defi ned. It is good
to know whether if you make a little adjustment to the constraint values would
that have a big effect or a small effect on the cost.

Figure 4.6:

We know that to fi rst order:

(
dJ
∆J = dx ∆x
df (4.98)
∆f = dx ∆x

At the solution point we have the necessary condition, and thus we can write:
dJ df
= −λT (4.99)
dx dx
Substituting to Eq.4.98, we have:
df
∆J = −λT ∆x = −λT f (4.100)
dx
Thus we have our sensitivity:
dJ
= −λT (4.101)
df
The sensitivity is given by the Lagrange multipliers, it is negative just because
of the convention on the signs from which we started. So the λ0 s with large
magnitude are the one that correspond to constraints that are highly sensitive.
132 CHAPTER 4. INTRODUCTION TO OPTIMAL CONTROL

Figure 4.7: Principle of Optimality

4.6 Dynamic Programming

Dynamic programming derives from a basic principle known as the principle of
optimality that was fi rst stated by Bellman. The principle of optimality can be
described by the following example. Suppose we have a control problem starting
from a given state and the problem was to drive our system there to a set of
terminal conditions.

Suppose that the optimal solution passes through some intermediate point
(x1 , t1 ). The principle of optimality states that if the optimal solution from the
first point passes through (x1 , t1 ) then it must be true that the optimal solution
to the problem starting from (x1 , t1 ) is the continuation of the same path. This
principle leads to two things: a numerical solution procedure and theoretical
results. The numerical procedure is called dynamic programming. Dynamic
programming is a method for solving multistage decision problems.
To illustrate the procedure, we will use the graph shown in Fig. 4.8.

We start at a certain point and at that stage there several decision we can
make. Suppose that we always start at the end of the problem. Each terminal
condition has its associated cost value. Now we are going our way back and
suppose we have gone back to the third stage. We have made decisions to get
to the end point and we found out what is the optimal return that we get
from there. We have stored these values. Now principle of optimality help us
back up one stage and choose the best branch of the path. However we do it
4.6. DYNAMIC PROGRAMMING 133

Figure 4.8: Multistage decision example

134 CHAPTER 4. INTRODUCTION TO OPTIMAL CONTROL

systematically by considering the following sum:

∗
∆J1 + J021
∗ (4.102)
∆J2 + J022
∗ ∗
All we have to do is compare the sum of ∆J1 + J021 and ∆J2 + J022 , and
∗
whichever one of those is better is the better choice of J02 . Thus we only consider
the optimal return from each of those nodes.
Example.

Figure 4.9: Flight Planning application

Suppose we want to plan flight paths across the Pacifi c. The purpose of the
flight plans was to take advantage of the winds. We can solve this problem using
the dynamic programming approach. The variable along the path is a range
variable. And at each one of these stages what we have is a two dimensional
grid of values in altitude and cross-range. The idea was that we are fi xing power
setting, fi nd the path that minimizes time. And with fi xed power, minimizing
time also minimizes fuel. The time taken would depend upon the wind direction
and velocity. So, we start from the end, take each one of the previous nodes
4.6. DYNAMIC PROGRAMMING 135

and fi gure out the time that it is going to take to go from that node to Jakarta.
Then we back up one stage and use the process that we described. We will
illustrate now the case for control problem.
For a control problem, we will choose a one-dimensional, scalar state problem
so that we can draw the complete picture given by Fig. 4.10.

Figure 4.10: One dimensional scalar state problem

We begin by defi ning the grid consisting of time points and state points on
which we are going to solve the problem. In this case, we wish to minimize:
Z tf
min h[x(tf ), tf ] + g[x(t), u(t), t]dt (4.103)
t0

subject to:

⎧ .
⎨ x = a(x, u, t)
x0 = f ixed (4.104)
⎩
tf = f ixed
136 CHAPTER 4. INTRODUCTION TO OPTIMAL CONTROL

and there may be other constraints too. These other constraints may affect
the control (bounded control for example) or could affect the state (there would
be some allowed region).
We have to put down the optimal value of the cost, at each one of the grid
points at tf . The integral part is vanished at this time, thus we have:

Ji∗ = h(xi ) (4.105)

Now we are going to back up one stage and consider all the possible ways of
completing the problem from there. We have to consider all of them as far as
there are no constraints. Normally, we will approximate the integral terms in
the cost function and the differential equation as just rectangular integration:

∆J[(xi , ti ), (xj , tj )] ∼
= g[xi , ui , ti ]∆t (4.106)

where ui has to be the control that actually takes you from xi at ti to xj

at tj . In this case we can solve for it directly, because we have only a scalar
control and a scalar state:

xj (tj ) = xi (ti ) + a(xi , ui , ti )∆t (4.107)

And for any xi and xj :

xj − xi
a(xi , ui , ti ) = (4.108)
∆t
we can solve the above expression and find ui , giving us for any xi and xj
the value of ui . Then we can put this in the expression of ∆J, Eq. 4.106, that
is the cost that we incur in making that transition. Now in order to find the
optimum path to take from this point, we simply fi nd:

min[∆J(xi , xj )] + J ∗ (xj )]
J ∗ (xi , ti ) = (4.109)
xj

Then we label this node with this value. And so we work our way to the left.
If we have constraints, it simplify the problem in general. The great advantage
of this, if we can solve the whole problem, is that once we found the optimal
path from the first stage it will be defi ned throughout the whole space. If at
some later stage, we fi nd ourself to be at another node not the optimal one, we
know how to go in an optimal path also from this other node to the end. We
have all those data stored. This gives us a kind of feedback solution, in that we
can store in table the optimal control action to take from every node. This is
for one dimension, in the next session we will expand this to higher dimension.
Chapter 5

Discrete time Optimal

Control

5.1 Higher Dimension Control Problems

We had just begun to discuss dynamic programming as a numerical method for

solving scalar control problems. The problem was to minimize a cost function
of the form:
Z tf
min J = h[x(tf )] + g[x(t), u(t), t]dt (5.1)
t0

We defined a grid in x and t, and saw how we could use the principle of op-
timality to reduce the total number of computations required to search for an
optimal solution. Note that this was approximation in two senses: first because
we define a discrete grid values of x and t which are normally continuous, and
secondly because the dynamics and cost integral are approximated by a rectan-
gular integration rule. Now we want to generalize the problem that we did in
couple of steps.
The first generalization will be for the case of a free terminal time. The first
thing that happens when the terminal time is free is that the problem statement
may have a terminal function that depends on that time:
Z tf
J = h[x(tf ), tf ] + g[x(t), u(t), t]dt (5.2)
t0

For example, for a minimum time problem, let h = tf , (g = 0), which can also
be done by h = 0, g = 1. Once we have allowed the terminal time to be free,
then we almost certainly have to specify some terminal constraints or else the

137
138 CHAPTER 5. DISCRETE TIME OPTIMAL CONTROL

Figure 5.1: Discrete grid of x and t

problem will not be well posed. For example, if tf is free, and we do not specify
terminal constraints on the states, then the obvious trivial solution is to make:
tf = t0
(5.3)
x(tf ) = x(t0 ) with J = 0
which is not a well-posed solution. So, in addition to this, in a scalar problem,
we have a function m[x(tf ), tf ] = 0, which will specify some terminal conditions
that we have to meet. So we have also,
tf is free
(5.4)
x(0) is free
Now in this approximate procedure we must approximate the acceptable termi-
nal points. At those points, the optimal cost is just the terminal cost since the
integral does not contribute at tf :
J ∗ (x, t) = h(x, t) on m(x, t) = 0 (5.5)
From here on the procedure is the same as before:
min{g(xi , uj , tk )∆t + J ∗ (xj , tk+1 )}
J ∗ (xi , tk ) = (5.6)
xj (tk+1 )
where uj , the control to go from xi to xj , is found from satisfying the dynamics:
xj (tk+1 ) = xi (tk ) + a(xi , uj , tk )∆t (5.7)
5.1. HIGHER DIMENSION CONTROL PROBLEMS 139

Now, so far, so we could draw the pictures, we have been looking at a scalar
state and control. Now let us generalize to higher dimensional problems, both
in x and u. The problem is restated with vector x and u:
Z tf
min J = h[x(tf ), tf ] + g[x(t), u(t), t]dt (5.8)
t0

subject to :
.
x = a(x, u, t)
m[x(tf ), tf ] = 0
(5.9)
x(0) = x0
tf is free

We initialize the solution with same condition we formed previously in Eq.5.3.

Now we have to use a different procedure to work our way backward from those
terminal values of optimal cost. The main difference come from the fact that
before, we could perform our comparison from a node in terms of the x values at
the next stage in time because for each xi and xj we could solve for the control
that would make that transition, so the control would be defined.
Now, we can not do that. Normally it will be the case that the dimension of
u will be less than that of x meaning if we specify an xi and xj , we normally can
not solve u to make that transition, in fact, not all such transitions are possible.
We look at the case shown by the following figure. Then on this grid at every
time, we actually have a 2-dimensional array of x0 s, which can be expanded to
a 2-D grid of x1 and x2 all at a particular time. In this case given a value of x
at a previous stage, there is really only a 1-dimensional surface of x0 s that can
be achieved at the next stage, because there is only 1 DOF in u
So, it is not good to pick x0 s and try to solve for u0 s. Instead, we really have
to pick u0 s and solve for the corresponding x0 s.
The way this is done, is to also define a grid of points on u in addition to x
and t. In other words we will test discrete points along this curve of achievable
values of x(tk+1 ) corresponding to discrete values of uj . We now can write down
the recursion:
min{g(xi , uj , tk )∆t + J ∗ (xj tk+1 )}
J ∗ (xi , tk ) = (5.10)
uj (tk )

where
.
xj = a(xi , uj , tk )∆t + xi (tk ) (5.11)

Referring to Fig.5.3, starting from a node xi at tk , we try all possible trajectories

to go to the next step by testing over not all values of x but of u. We will end up
somewhere in the space of the grid of 2-D x at tk+1 which generally will require
interpolation from among the known costs at the grid points. Any constraints of
x and u reduce the amount of calculations and simplify the problem in general.
If we have solved this problem, then in effect we have solved all the problems
140 CHAPTER 5. DISCRETE TIME OPTIMAL CONTROL

Figure 5.2: Two-dimensional array of states

5.1. HIGHER DIMENSION CONTROL PROBLEMS 141

Figure 5.3: Double interpolation of Cost function

that appear on the grid, meaning that we can save the optimal controls to use
from any one of these grid points. That represents a feedback control law and
it is true for non-linear problems which can be achieved by other approaches to
the optimal control problem. The feedback law is just a look-up table.
If we have several u0 s and several x0 s then, the number of points in the
solution grid is:
Ntime .Nxn (5.12)
which for a hundred time steps, with the states consisting 3 positions and 3
velocities:
Ntime .Nxn = 100 × 1006 = 1014 (5.13)
namely, we need 100 million megawords just to store the values. Bellman refers
to this as curse of dimensionality. He stresses that there is a reduction of the
number of computations because we should choose only the optimal paths.
In interpreting the final solution, once we have solved the problem on the
grid, we will have stored at every grid point, the optimal value of the cost and
control to use at each grid point. But as soon as we use u∗ to start from x0 , we
will go to a place most likely not on a grid point. So it really is more accurate
to interpolate in the grid of u0 s to get the u to use at this point. Thus we can
interpolate to get both the J ∗ and control u∗ to use at that point. Using that
control, we can then solve the approximated differential equation to the value
for the next point.
142 CHAPTER 5. DISCRETE TIME OPTIMAL CONTROL

It is true that the optimal cost and control were found by starting only at
the gridded values of x at each stage, and now we are at some other value of x.
But this is acceptable to the accuracy with which the whole problem is being
solved as long as we do the interpolation at each stage. So we have now covered
the numerical procedure that the principle of optimality was to lead to. We are
going to cover the theoretical results that we get in the next session.

5.2 Discrete time optimal control problem

The theoretical part gives us the direct way of stating a recursion to solve
discrete optimal control problems with no approximations. The problem will
be stated as a regulator problems meaning that we have to get from some given
initial condition to some stated-set of terminal condition in an optimal fashion.
So we have:

x(k + 1) = aDk [x(k), u(k)] (5.14)

N
X −1
J = h[x(N ), N ] + gDk [x(k), u(k)]
k=0
m[x(N ), N ] = 0, N is free

We will not be concerned with how we will solve it numerically for now. We
only want to fi nd a recursion formula for the problem solution. Just as before,
for higher dimension of x:

J ∗ (x, N ) = h(x, N ) on m(x, N ) = 0 (5.15)

And now, the principle of optimality says:

min{gDk (x, u) + J ∗ [aDk (x, u), k + 1]}

J ∗ (x, k) = (5.16)
allowable u(k)

But notice that for any x, the function on the RHS is only a function of u. In
principle, we can fi nd the u that minimizes the function in the brackets; but
with things being non-linear, a numerical search might be required. But there
would be no approximation involved. We would only have to minimize RHS
for every x we were interested in. So unless we can functionalize this J ∗ as a
function of x(k + 1), then we are still stuck with having to solve the problem
for a bunch of choices of x so that we can store them for later interpolation to
solve for J ∗ . Thus, we have stated an exact recursion for a discrete time optimal
control problem, but attempting to apply it leads to dynamic programming
again.
The only way it appears that we can avoid the problem of a very high-
dimension grid to solve the problem happens when we have the discrete-time,
5.2. DISCRETE TIME OPTIMAL CONTROL PROBLEM 143

Figure 5.4: Interpolation in the grid of u0 s

linear, quadratic regulator, i.e. the dynamics are linear:

x(k + 1) = A(k)x(k) + B(k)u(k) (5.17)

N
X −1
1 1
J = x(N )T Hx(N ) + [x(k)T Q(k)x(k) + u(k)T R(k)u(k)]
2 2
k=0
No other constraints, N, terminal state, fi xed

For a well-posed problem, we need:

H = HT ≥ 0 (5.18)
Q(k) = Q(k)T ≥ 0
R(k) = R(k)T > 0

Our boundary condition is the same as before:

∗ 1
JN [x(N )] = x(N )T Hx(N ) (5.19)
2
Now, using the recursion from the previous page, we can back up one stage and
try to fi nd the optimal control to use now for a given x:
⎧ 1
⎨ 2 x(N − 1)T Q(N − 1)x(N − 1)+
∗ min 1
JN −1 [x(N − 1)] = u(N − 1)T R(N − 1)u(N − 1)+ (5.20)
u(N − 1) ⎩ 21 T
2 x(N ) Hx(N )
144 CHAPTER 5. DISCRETE TIME OPTIMAL CONTROL

Using the dynamics we get:

⎧ 1
⎪
⎪ x(N − 1)T Q(N − 1)x(N − 1)+
⎨ 21 T
∗ min 2 u(N − 1) R(N − 1)u(N − 1)+
JN −1 = u 1 (5.21)
N−1 ⎪
⎪ [A(N − 1)x(N − 1) + B(N − 1)u(N − 1)]T ×
⎩ 2
H[A(N − 1)x(N − 1) + B(N − 1)u(N − 1)]

We would like to minimize this and since it is quadratic in u, we take its fi rst
derivative and make it zero to get:
∗
∂JN −1 [x(N − 1)]
= u(N − 1)T R(N − 1) + [A(N − 1)x(N − 1) (5.22)
∂u(N − 1)
+B(N − 1)u(N − 1)]T HB(N − 1)
= 0

Collecting the two terms in u together:

∗
∂JN−1 [x(N − 1)] T
[ ] = [R(N − 1) + B(N − 1)T HB(N − 1)]u(N −(5.23)
1)
∂u(N − 1)
+B(N − 1)T HA(N − 1)x(N − 1)
= 0

The second derivative can readily be written as:

∗
∂ 2 JN−1 [x(N − 1)]
= R(N − 1) + B(N − 1)T HB(N − 1)] > 0 (5.24)
∂u(N − 1)2
which defi nes an upward rising bowl, i.e. a global minimum stationary point.
Now, solving for u(N − 1) = u∗ (N − 1) from Eq.5.23, we have:

u∗ (N − 1) = −[R(N − 1) + B(N − 1)T HB(N − 1)]−1 ×

B(N − 1)T HA(N − 1)x(N − 1) (5.25)
= F (N − 1)x(N − 1) (5.26)

where

F (N − 1) = −[R(N − 1) + B(N − 1)T HB(N − 1)]−1 B(N − 1)T HA(N − 1)

(5.27)

We see here that we have an optimal control which is a direct feedback of our
current state or simply a state feedback optimal control law. So now we can
calculate the optimal cost:
1 T
∗
JN 2 x(N − 1) Q(N − 1)x(N − 1)
−1 [x(N − 1)] = 1
+ 2 x(N − 1)T F (N − 1)T R(N − 1)F (N − 1)x(N − 1)+
1 T
2 x(N − 1) [A(N − 1) + B(N − 1)F (N − 1)]T H[A(N − 1)
+B(N − 1)F (N − 1)]x(N − 1)
(5.28)
5.2. DISCRETE TIME OPTIMAL CONTROL PROBLEM 145

Combining the terms using the common factor, we can rewrite the above equa-
tion as:
∗ 1
JN −1 [x(N − 1)] = − 1)T [Q(N − 1) + F (N − 1)T R(N − 1)F (N − 1)+
2 x(N
[A(N − 1) + B(N − 1)F (N − 1)]T H[A(N − 1)
+B(N − 1)F (N − 1)]]x(N − 1)
(5.29)

Or, simply:
∗ 1
JN −1 [x(N − 1)] = 2 x(N − 1)T P (N − 1)x(N − 1) (5.30)
∗
which is quadratic. Note that at stage N , the cost JN was a quadratic cost of
x(N ),
∗ 1 T
JN [x(N )] = 2 x(N ) Hx(N ) (5.31)

Eq.5.30 and Eq.5.31 suggest that the cost is always quadratic. We can show
this inductively.