AE259 Control

Lecture Notes
For Class Room teaching only
AE259: Navigation, Guidance and Control
Part 3: Control Basics
Prof. M. Seetharama Bhat
Department of Aerospace Engineering

Indian Institute of Science
Bangalore 560012, INDIA
18 November 2011
Contents
1 Dynamics of Aerospace Vehicles 1

1.1 Modelling of Dynamical System (LPS) : Inverted Pendulum on a moving cart: 6
1.2 Satellite Attitude Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2.1 Attitude Representation . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2.2 Basics of Satellite Attitude Dynamics . . . . . . . . . . . . . . . . . . . 15
1.2.3 Linear model of Three Axis Stabilized Satellites . . . . . . . . . . . . . 18
1.3 Aircraft Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.3.1 Longitudinal Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.3.2 Lateral Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.4 Dynamics of Rockets and Missiles . . . . . . . . . . . . . . . . . . . . . . . . 36
1.4.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2 System Analysis 41
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.1.1 Navigation, Guidance & Control . . . . . . . . . . . . . . . . . . . . . . 41
2.2 Methods of Linear Systems Analysis . . . . . . . . . . . . . . . . . . . . . . . 43
2.3 Response of a Linear Time Invariant System . . . . . . . . . . . . . . . . . . . 58
2.4 Frequency Response Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
2.5 Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
2.5.1 Bode stability Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 90
2.5.2 Nyquist stability criterion . . . . . . . . . . . . . . . . . . . . . . . . . 94
2.5.3 Root Locus Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
2.5.4 Asymptotic Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
i
ii CONTENTS
2.6 State Space Representation of a LTIV system . . . . . . . . . . . . . . . . . . 115

2.7 Controllability and Observability (LTIV System) . . . . . . . . . . . . . . . . . 118
2.7.1 Passivity/ Dissipativity . . . . . . . . . . . . . . . . . . . . . . . . . . 121
2.7.2 State Space Realizations . . . . . . . . . . . . . . . . . . . . . . . . . . 124
2.7.3 Canonical Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
2.7.4 MIMO Generalized Control Canonical Forms (GCCF) . . . . . . . . . 128
2.8 Methods of Improving System Performance . . . . . . . . . . . . . . . . . . . . 130
2.9 Compensator Synthesis via Root Locus Technique . . . . . . . . . . . . . . . . 136
2.10 PID Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
2.11 Control System Design Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
2.11.1 Singular Values and Singular Value Decomposition (SVD) . . . . . . . 167
2.12 Design of Controllers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
3 Control of Aerospace Vehicles - Classical Control Perspective 173

3.1 Satellite Attitude Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
3.2 Aircraft Dynamics and Control . . . . . . . . . . . . . . . . . . . . . . . . . . 184
3.2.1 Longitudinal Flight Control System . . . . . . . . . . . . . . . . . . . 185
3.2.2 Lateral Flight Control System [35, 7] . . . . . . . . . . . . . . . . . . 219
3.3 Dynamics of Rockets and Missiles : . . . . . . . . . . . . . . . . . . . . . . . . 242
3.4 Analysis of Traditional Missile Autopilot . . . . . . . . . . . . . . . . . . . . . 244
3.5 Attitude Flight Control System for Launch Vehicles . . . . . . . . . . . . . . . 253
3.6 Control - Structure Interactions in Rockets/ Launch Vehicles . . . . . . . . . 255
3.6.1 Effect of Propellant Sloshing : . . . . . . . . . . . . . . . . . . . . . . . 255
3.6.2 Engine Inertia Effects: . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
3.6.3 Vehicle Flexibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
4 Pole placement Techniques 265

4.1 Controller design by Pole Assignment Techniques . . . . . . . . . . . . . . . . 266
4.2 Observer Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
4.2.1 Compensator Design Using Full-Order Observers . . . . . . . . . . . . 277
4.2.2 Combined Regulator - Observer System. . . . . . . . . . . . . . . . . . 278
4.2.3 Multi-input state variable Feedback controller (dyadic form) . . . . . . 281
CONTENTS iii
5 References 283
iv CONTENTS
List of Figures
1.1 Inverted pendulum on a mobile cart. . . . . . . . . . . . . . . . . . . . . . . . 6

1.2 Orbit Fixed reference Coordinate System. . . . . . . . . . . . . . . . . . . . . 15
1.3 Missile Coordinate System [4]. . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.1 Design Steps [45] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

2.2 Design Plant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.3 First-order system response to a unit step [45] . . . . . . . . . . . . . . . . . . 58
2.4 Unit step response of a first order system: a. System showing input and output;
b. pole-zero plot of the system; c. evolution of a system response [45]. . . . . . 59
2.5 Step response components generated by complex poles of a typical under damped
Second-order system [45]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.6 Unit step response of second order system [45]. . . . . . . . . . . . . . . . . . . 61
2.7 Second-order systems, pole plots, and step responses for different damping ratios 62
2.8 Step responses of second-order underdamped systems as poles move: a. with
constant real part; b. with constant imaginary part; c. with constant damping
ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.9 Second-order systems, plots of Percent overshoot vs. damping ratio and Nor-
malized rise time vs. damping ratio for a second-order under damped response 63
2.10 Effect of adding a zero to a two-pole system . . . . . . . . . . . . . . . . . . . 64
2.11 Component responses of a three-pole system: a. pole plot; b. component
responses: non-dominant pole is near dominant second-order pair (Case I), far
from the pair (Case II), and at infinity (Case III). . . . . . . . . . . . . . . . . 65
2
2.12 Magnitude & phase plots of frequency response for G(s) = (s+2)
. . . . . . . . . 67
2
2.13 Polar plots of frequency response for G(s) = (s+2)
. . . . . . . . . . . . . . . . . 68
v
vi LIST OF FIGURES
100∗(s+1)
2.14 Magnitude & phase plots of frequency response for G(s) = (s+10) (s+100) . 69
2.15 Rules for Bode plots of frequency response. . . . . . . . . . . . . . . . . . . . . 71
1
2.16 Bode magnitude plots for G(s) = (1+s/ω ) . . . . . . . . . . . . . . . . . 72
0
1
2.17 Bode phase plots for G(s) = (1+s/ω ) . . . . . . . . . . . . . . . . . . . 73
0
1
2.18 Bode plots for G(s) = (1+s/10) . . . . . . . . . . . . . . . . . . . . . . . . 74
1
2.19 Bode plots for systems with repeated poles/ roots G(s) = . . . 75
(1+s/30)2
2.20 Bode plots for G(s) = (1 + s/ω0 ); ω0 = 30 rad/sec. . . . . . . . . . . . . . . 76
2.21 Normalized and scaled Bode plots for a. G(s) = s; b. G(s) = 1/s; c. G(s) =
(s + a); d. G(s) = 1/(s + a) [45, p.603]. . . . . . . . . . . . . . . . . . . . . . . 78
ωn2
2.22 Bode plots for G(s) = 2 . . . . . . . . . . . . . . . . . . . . 79
(s +2ξωn s+ωn2 ) .
ωn2
2.23 Bode phase plots for G(s) = 2
(s +2ξωn s+ωn2 )
. ........ . . . . . . . . . 81
2.24 Bode plots for G(s) = ( ωsn )2 + 2 ξ ( ωsn ) + 1; ωn = 10, ζ = 0.1. . . . . . . . . . 82
2.25 Normalized and scaled log magnitude response for G(s) = ( ωsn )2 + 2 ξ ( ωsn ) + 1;
for variable ζ and a fixed ωn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
2.26 Scaled Bode phase response for G(s) = ( ωsn )2 + 2 ξ ( ωsn ) + 1; for variable ζ and
a fixed ωn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
2.27 Typical Feedback Control System . . . . . . . . . . . . . . . . . . . . . . . . . 87
2.28 Typical Cascade Control System . . . . . . . . . . . . . . . . . . . . . . . . . . 87
2.29 Gain and phase margins on the Bode diagrams, Source: Fig.10.37, [45] . . . . 91
K
2.30 Bode plots for OLTF G(s) H(s) = [(s+2) (s+4) (s+5)] ; K = 40 [45] . . . . . 92
2.31 Block diagram for Satellite Pitch attitude control example (Bode 1) . . . . . . 93
2.32 Bode plot for Satellite Pitch attitude control example . . . . . . . . . . . . . . 93
2.33 Cauchys principle of argument . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
2.34 Nyquist contour in the s-plane; ∆(s) = 1 + GH(s) . . . . . . . . . . . . . . . . 96
2.35 Phase and gain stability margins depicted in Nyquist plot in (G H) plane . . . 97
2.36 Nyquist contour plot in the s-plane for Example 2 . . . . . . . . . . . . . . . . 98
2.37 MATLAB Nyquist plot for Example 2 . . . . . . . . . . . . . . . . . . . . . . 100
2.38 MATLAB Root locus plot for RL Sketch Example 1 . . . . . . . . . . . . . . . 104
2.40 MATLAB Root locus plot for angle of departure in RL Sketch . . . . . . . . . 107
LIST OF FIGURES vii
2.41 MATLAB Root locus plot for angle of arrival in RL Sketch . . . . . . . . . . . 108
2.42 MATLAB Root locus plot for four simple examples . . . . . . . . . . . . . . . 109
2.44 MATLAB Root locus plot for RL Example 4 . . . . . . . . . . . . . . . . . . . 110
2.45 MATLAB Root locus plot for RL Example 5 . . . . . . . . . . . . . . . . . . . 111
2.46 Uncontrollable and Unobservable System due to Pole - zero cancellation. . . . 120
2.47 Typical Feedback control system with series/ cascade compensator K(s). . . 120
2.48 Traditional Controller Canonical Form. . . . . . . . . . . . . . . . . . . . . . . 126
2.49 Traditional Observer Canonical Form. . . . . . . . . . . . . . . . . . . . . . . . 126
2.50 Second Order Transfer function and Geq - representation. . . . . . . . . . . . . 135
2.51 Lag and lead compensator poles in s-plane. . . . . . . . . . . . . . . . . . . . . 137
2.52 Root locus plot for uncompensated (left) and lag compensated (right) systems 138
2.53 Root locus plot for uncompensated Launch vehicle flight control system . . . . 139
2.54 Root locus plot for Lag-compensated Launch vehicle control; Left-closeup. . . 140
2.55 Root locus for lateral acceleration feedback control for a missile [1] . . . . . . . 142
2.56 Root locus for lateral acceleration feedback control for a missile with lead filter 143
2.57 Block diagram of PID Feedback Control System. . . . . . . . . . . . . . . . . . 145
2.58 Root locus plot of Proportional Feedback Control System. . . . . . . . . . . . 148
2.59 Step response of P, PI and PID Feedback Control System. . . . . . . . . . . . 150
2.60 Response Curve for Ziegler-Nichols First Method. . . . . . . . . . . . . . . . . 152
2.61 Unit step response Curve for Ziegler-Nichols second Method. . . . . . . . . . . 154
2.62 Root locus for the uncompensated system for PID design example. . . . . . . . 158
2.63 Root locus for the PD - compensated system for PID design [45, Fig. 9.34]. . . 160
2.64 Unit step responses for the uncompensated, PD - compensated and PID - com-
pensated system for PID design [45, Fig. 9.35]. . . . . . . . . . . . . . . . . . . 160
2.65 Root locus for the PID - compensated system for PID design [45, Fig. 9.36] . . 161
3.1 Pitch control loop. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175

3.2 Root locus diagram for the pitch loop. . . . . . . . . . . . . . . . . . . . . . . 176
3.3 Pitch response to impulsive disturbance torque. . . . . . . . . . . . . . . . . . 176
3.4 Pitch response to step disturbance torque. . . . . . . . . . . . . . . . . . . . . 177
viii LIST OF FIGURES
3.5 Pitch response to solar pressure torque. . . . . . . . . . . . . . . . . . . . . . . 177

3.6 Root locus (near origin) with u feedback and α - feedback . . . . . . . . . . . 187
3.7 Root locus (near origin) with q = θ̇ and θ - feedback . . . . . . . . . . . . . . . 187
3.8 Bode (frequency) response plots for sensor outputs u and θ . . . . . . . . . . . 189
3.9 Bode (frequency) response plots for sensor outputs q and α . . . . . . . . . . . 189
3.10 Aircraft Altitude - Mach Envelope . . . . . . . . . . . . . . . . . . . . . . . . . 191
3.11 Aircraft Attitude control system block diagram; :
Top: Generic; Middle: Conventional transport; Bottom: Jet transport. . . . . 193
3.12 Root locus for Conventional transport aircraft & Autopilot. . . . . . . . . . . . 194
3.13 Root locus for Conventional transport aircraft & Autopilot. . . . . . . . . . . . 194
3.14 Root locus for jet transport aircraft & pitch rate stabilization system. . . . . . 197
3.15 Block diagram of Pitch stabilization system. . . . . . . . . . . . . . . . . . . . 199
3.16 Inner - loop of Pitch stabilization system (α - feedback). . . . . . . . . . . . . 201
3.17 Root locus for Inner - loop of Pitch stabilization system. . . . . . . . . . . . . 201
3.18 Root locus for Outer - loop of Pitch stabilization system. . . . . . . . . . . . . 202
3.19 Root locus for Outer - loop of Pitch rate Control Augmentation system for F-16.205
3.20 Root locus for Outer - loop of Pitch stabilization system. . . . . . . . . . . . . 206
3.21 Outer - loop of Pitch attitude hold Autopilot system. . . . . . . . . . . . . . . 206
3.22 Root locus for Outer - loop of Pitch attitude autopilot, Kq = 1.19. . . . . . . . 207
3.23 Root locus for Outer - loop of Pitch attitude autopilot, Kq = 1.98. . . . . . . . 208
3.24 Block diagram for Pitch acceleration control system. . . . . . . . . . . . . . . . 208
3.25 Block diagram for Outer loop for the acceleration control system. . . . . . . . 210
3.26 Root locus for Outer loop for the acceleration control system [7, p.81]. . . . . . 210
3.27 Block diagram for landing configuration of jet aircraft [7, p.82]. . . . . . . . . . 211
3.28 Root locus for inner pitch loop of landing configuration of jet aircraft [7, p.83]. 212
3.29 Root locus for outer pitch loop of landing configuration of jet aircraft. . . . . . 213
3.30 Block diagram for velocity control of landing configuration of jet aircraft. . . . 214
3.31 Block diagram for Mach hold mode autopilot. . . . . . . . . . . . . . . . . . . 215
3.32 Expanded view of region near origin of the root locus for pitch rate stabilization
for the jet transport aircraft. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
3.33 Block diagram for Mach Hold Autopilot. . . . . . . . . . . . . . . . . . . . . . 217
LIST OF FIGURES ix
3.34 complementary Root locus for Mach hold mode. . . . . . . . . . . . . . . . . . 217

3.35 Block diagram for Altitude hold mode. . . . . . . . . . . . . . . . . . . . . . . 218
3.36 Block diagram for the outer loop of Altitude hold mode. . . . . . . . . . . . . 219
3.37 Root locus for altitude hold mode. . . . . . . . . . . . . . . . . . . . . . . . . 219
3.38 Bode diagram for ϕ(s)/δr (s) for jet transport aircraft. . . . . . . . . . . . . . . 224
3.39 Bode diagram for ψ(s)/δr (s) for jet transport aircraft. . . . . . . . . . . . . . . 224
3.40 Bode diagram for β(s)/δr (s) for jet transport aircraft. . . . . . . . . . . . . . . 225
3.41 Block diagram of Dutch roll damper. . . . . . . . . . . . . . . . . . . . . . . . 228
3.42 Block diagram of Dutch roll damper for Root locus study. . . . . . . . . . . . 229
3.43 Root locus for Dutch roll damper with τ = 3sec for washout filter. . . . . . . . 230
3.44 Root locus for Dutch roll damper with τ = 0.5sec for washout filter. . . . . . . 230
3.45 Bode plot for Dutch roll damper with τ = 3.0sec for washout filter. . . . . . . 231
3.46 Root locus for Dutch roll damper with τ = 1.0sec for washout filter; top:
normal plot; bottom: closeup view. . . . . . . . . . . . . . . . . . . . . . . . . 233
3.47 Block diagram for turn coordination using sideslip angle β feedback in the outer
loop. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
3.48 Simplified Block diagram for turn coordination using sideslip angle β feedback. 234
3.49 Block diagram of outer turn coordination loop for Root locus study. . . . . . . 235
3.50 Root locus of outer turn coordination loop; left full; right expanded near origin. 235
3.51 Root locus of outer turn coordination loop - expanded near origin. . . . . . . . 236
3.52 Block diagram for β̇ - β SAS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
3.53 Root locus for β̇ - β SAS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
3.54 Block diagram for for turn coordination using lateral acceleration. . . . . . . . 237
3.55 Block diagram for a conventional yaw orientation control system. . . . . . . . 239
3.56 Simplified Block diagram for yaw orientation control system. . . . . . . . . . . 239
3.57 Block diagram for inner loop of Fig. (3.56) for root locus study. . . . . . . . . 240
3.58 Root locus for yaw rate feedback system shown in Fig. (3.57). . . . . . . . . . 240
3.59 Block diagram outer loop of pitch rate feedback for root locus study. . . . . . 241
3.60 Root locus for Block diagram shown in Fig. (3.59). . . . . . . . . . . . . . . . 241
3.61 Root locus for lateral acceleration feedback control for a missile . . . . . . . . 243
3.62 Roll Autopilot for a missile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
x LIST OF FIGURES
3.63 Roll Autopilot for a missile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246

3.64 Lateral Autopilot for a missile . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
3.65 Root locus for Innerloop of simplified Autopilot for a missile . . . . . . . . . . 248
3.66 Root Locus for Inner stabilization loop with pitch rate feedback. . . . . . . . 249
3.67 Block diagram of acceleration autopilot outer loop . . . . . . . . . . . . . . . . 250
3.70 Flight control system block diagram for a rocket; H(s) = (1 + kR s) . . . . . . 254
3.71 Root locus for simplified Autopilot for a rocket, 1/kR = 1/3 . . . . . . . . . . 254
3.72 Root locus for simplified Autopilot for a rocket, 1/kR = 5 . . . . . . . . . . . 255
3.73 Propellant Sloshing schematic representaion. . . . . . . . . . . . . . . . . . . . 256
3.74 Examples of unstable (top) and stable (bottom) pole pairs. . . . . . . . . . . . 257
3.75 Slosh mode stabilization using active control. . . . . . . . . . . . . . . . . . . . 258
3.76 Engine Inertia Effect. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
4.1 Controller + Observer System (dynamic compensator inside dotted block). . . 279
List of Tables
2.1 Example of a stable and an unstable system . . . . . . . . . . . . . . . . . . . 85

2.2 Routh table for Example RL sketch . . . . . . . . . . . . . . . . . . . . . . . . 106
2.3 Effects of changing parameters for manual tuning of PID control . . . . . . . . 152
2.4 Zeigler - Nichols rule for tuning of PID control first method . . . . . . . . . . . 152
2.5 Zeigler - Nichols rule for tuning of PID control . . . . . . . . . . . . . . . . . . 153
2.6 Characteristics of uncompensated, PD-compensated and PID compensated sys-
tems of PID design example . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
3.1 Summary of Roll/Yaw Control System Parameters . . . . . . . . . . . . . . . . 183

3.2 The eigen-structure of the longitudinal dynamics F-16-1 . . . . . . . . . . . . . 186
3.3 FCS catagories [35, p.261] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
xi
xii LIST OF TABLES
Preface
AE 259: Navigation guidance and Control is a three credit core course for Master’s program
of Aerospace Engineering. The control part normally constitutes two third of the entire
course; i.e., two credits. It is taught over approximately 30 hours of class room teaching. This
course is an introductory course on Control of Aerospace vehicles, with students having little
basic understanding in control theory and with or without background knowledge in Flight
mechanics/ dynamics. Chapter 1 forms background/ reading material for the non-aerospace
specialists. Chapter 4 gives registrants a flavor of control system synthesis in state space
domain, but not complete knowhow of design process.
This lecture notes is prepared from the available literature (from open sources,
mainly text books) and research contributions of the author and his research scholars and
graduate students. It is not meant for commercial use and should not be sold. All
research scientists, students and faculty from Academics are free to use it for study purposes.
This lecture notes is being upgraded. It also contains many typographical errors.
Author requests the readers to send the list of errors/ typographical mistakes present in this
lecture notes or suggestions for improvements through e-mail: msbdcl@aero.iisc.ernet.in
xiii
xiv LIST OF TABLES
Acknowledgement
Author wishes to thank all his research students, graduate students and research / project
scientists/ engineers and colleagues for valuable inputs to this effort. Thanks are also due to
Department of Aerospace Engineering, Indian Institute of Science, Bangalore India. Author
wishes to acknowledge the knowledge gained through interactions with Scientists from ISRO
centers, DRDO laboratories, faculty and students of IISc and other academic institutions.
Author wishes to express his gratitude to all the authors of textbooks and papers that appeared
in journals, conference proceedings and technical reports and also the research materials that
are made available in open sources.

27 November 2011
xv
Chapter 1
Dynamics of Aerospace Vehicles
All the flight vehicles, such as rockets, missiles, launch vehicles, satellites and aircraft, need six
degree of freedom (6 DOF) so as to describe transnational and rotational (or attitude) motion.
Transnational motion is the motion of rigid body concentrated at the center of gravity (CG),
while attitude motion is the rotational motion of the rigid body about its CG. In the case of
satellite, where sphericity of earth and large speed of satellite are involved, the reference point
is centre of mass (CM) instead of center of gravity (CG). Also, translational and attitude
motion of the satellite are also known as orbital motion and librational motion. Many of
the present day flight vehicles are designed to be light weight, hence the structural vibration,
propellant/ liquid sloshing also play important role. For the sake of simplicity, discretized
model of continuum mechanics of the structural vibration is used in the analysis. However,
for most part of the discussion that follows, it is assumed that the aerospace vehicle is rigid.
In general, motion of Aerospace Vehicles:
- translational motion (Newton law F = m dv

dt
= d
dt
(mv))
d
- attitude motion (Euler equations T = dt
(Iω))
- Flexural vibration, fuel sloshing, actuator & sensor dynamics.
Missiles, rockets and : translational motion and attitude motion are coupled.
aircraft
Satellites/ ex- : orbital and attitude motion are deemed to be decoupled for
atmospheric vehicles control design purpose
Control system → mostly deals with attitude control of aerospace vehicles.
1
2 CHAPTER 1. DYNAMICS OF AEROSPACE VEHICLES
For majority of satellites, it is possible to assume decoupling between orbital mo-

tion and attitude motion owing large separation time constants involved and also due weak
inertial interaction. It is possible to design controllers independently for these without jeop-
ardizing the total performance. For atmospheric vehicles, however, it is necessary to include
coupling between translational and rotational motion. The rockets, missiles and launch vehi-
cles, generally, have two planes of symmetry, and hence it is possible to design linear controller
separately for roll, pitch and yaw plane dynamics. For aircraft having one plane of symmetry,
pitch plane dynamics or longitudinal dynamics can be assumed to be decoupled from roll-yaw
dynamics or lateral dynamics (at low angles of attack). For all these aerospace vehicles, the
dynamics is governed by the Newton’s laws and resulting equations of motion are coupled,
nonlinear and often time-varying. They are subjected to diverse class of disturbances. In
order to use the powerful linear control theory, these models are linearized about a specific
operating condition/ trim condition/ equilibrium condition. These governing equations are
, therefore, approximation to the actual motion. The flight control system, therefore, is re-
quired to
– stabilize the system
– act against the effect of disturbances(disturbance rejection/alleviation or initialization)
– take care of parameter variation (robustness)
– provide good performance with respect to the command input from pilot/guidance system.
Navigation, guidance and control systems are most critical components of aerospace
and other vehicles. The types of vehicles include airplanes, rockets, missiles, ships, torpedoes,
drones, and material transport vehicles within factories and so forth.
Navigation system is an Apparatus for generating and detecting the path along which a vehicle
or craft is to be guided, often remotely or autonomously. Systems that are intended to have
a high degree of human like intelligence and sensing abilities are usually referred to as a
navigation system. Apparatus for generating and detecting the path along which a vehicle
or craft is guided, often remotely and automatically; determines position, velocity and other
relevant state vectors. For example, an inertial navigation system (INS) provides the position,
velocities and attitude of a vehicle by measuring the accelerations and rotations applied to the
system’s inertial frame. These systems are also referred to as an inertial platform. It is widely
used because it refers to no real-world item beyond itself. It is therefore immune to jamming
3
and deception. The accuracy of the sensor and the properties of the inertial navigation system
are the principal factors. Major remote navigation systems are:
Long Range Navigation (LORAN): This was the predecessor of GPS and was (and to an extent
still is) used primarily in commercial sea transportation. The system works by triangulating
the ship’s position based on directional reference to known transmitters.
Global Positioning System (GPS): This system of satellites provides extremely accurate
position information. The receiver’s position is triangulated using satellites in known orbits.
Commercial receivers are limited in how accurately they may provide position data, as well
as the maximum velocity at which they may operate. This is to prevent their use in weapons.
Radar homing This form of guidance is used exclusively for military munitions. Includes
active (employs own radar to enlighten the target), passive (detects foe radar emission), and
semi-active radar homing.
Laser designation: This form of guidance is also used exclusively for military munitions. A
laser designator device highlights a spot on the target with an encoded laser beam. This
spot provides reference information to an incoming munition that allows it to make in-flight
corrections to its trajectory. The use of an encoded signal reduces the threat of jamming as
well as reducing interference in high-noise combat environments. The primary limitation on
this device is that it requires a line of sight to the target from both the munition and the
designator. More advanced systems use the laser to designate a target, which is acquired by
an orbiting satellite that then feeds GPS target data to a launch facility. This allows potential
targets to be designated long before operations commence as well as eliminating the line-of-
sight requirement for the munition. Optical guidance Another form of guidance used almost
exclusively for military purposes, optically guided missiles use stored images of the terrain they
are to fly over and an external sensor to track their current position. This guidance system was
extremely expensive and not suitable for use in small payload operations. These were used on
cruise missiles before the advent of GPS, which is both cheaper and more accurate. Devices
that implement optical guidance incur high costs because of the high on-board processing
requirements needed to check the current location against the course data. At the time this
type of guidance system was widely used by the military, processors capable of this were very
expensive, although similar processing power is available in embedded architectures today.
Although called optically guided, most designs used infrared, ultraviolet, or radar imaging
to scan the terrain, since the visible spectrum suffers from relatively poor clarity and high
interference (other electromagnetic frequencies can see through dust and clouds, for instance).
Guidance system: Guidance systems are the control devices or group of devices and algorithms
used in guidance of (/ used to navigate) an aircraft or spacecraft or rockets and missiles or ship
or other crafts. Typically this refers to a system that navigates without direct or continuous
human control. One of the earliest examples of a true guidance system is that used in the
German V-1 during World War II. This system consisted of a simple gyroscope to maintain
heading, an airspeed sensor to estimate flight time, an altimeter to maintain altitude, and
other redundant systems.
A guidance system has three major sub-sections: Inputs, Processing, and Outputs.
The input section includes sensors, course data, radio and satellite links, and other information
sources. The processing section, composed of one or more CPUs, integrates this data and
determines what actions, if any, are necessary to maintain or achieve a proper heading. This is
then fed to the outputs which can directly affect the system’s course. The outputs may control
speed by interacting with devices such as turbines, and fuel pumps, or they may more directly
alter course by actuating elevators, ailerons, rudders, or other control surfaces on an airplane,
the rudder on a ship, the control surfaces on a missile or on a torpedo, the gimbal angle of the
motor on a rocket, and others. An inertial guidance system consists of an Inertial Measurement
Unit (IMU) combined with control mechanisms, allowing the path of a vehicle to be controlled
according to the position determined by the inertial navigation system. Inertial guidance
system with INS augmented by external systems like GPS, enable autonomous operation of
a flight vehicle or mobile vehicles/ platforms. For a remotely controlled vehicles, like homing
missiles and command guided missiles, separation between navigation and guidance systems
gets blurred. The Guidance algorithms and computers are utilized to steer a vehicle along a
path. In every case the guidance system utilizes knowledge of the difference between where
the vehicle should be and where it is. The difference between these two vectors is processed
by the guidance algorithm. The output is a steering command intended to reduce the error
between the desired and the actual paths.
Several important performance attributes contribute to the effectiveness of the sys-
tem. These attributes are governed by the guidance system and by the other system compo-
nents, including the vehicle itself and its dynamic behavior. A primary concern is accuracy.
5
Whether the goal is to insert a satellite into synchronous orbit or to try to intercept an enemy
aircraft with an air-to-air missile, the accuracy of the sensor and the properties of the guidance
system are the principal factors. Another concern is speed of response. Here the dynamics
of the vehicle itself can be a limiting factor. The guidance system must compensate to the
extent possible in providing a fast, responsive system. The system should be able to recover
from errors as quickly as possible and return to the desired path. In the case of homing on a
target, this is crucial if the target can maneuver. Coupled with the need for a quick response
is the simultaneous need for a stable response. Another important feature of the system is
its robustness. The guidance system design is based on a mathematical model of the vehicle,
the autopilot, and the sensor. The Guidance system must provide good overall performance
despite this. Reliability is also important. In many cases, backup components are provided
for redundancy.This is frequently the case for the digital computer of the guidance system,
especially for crewed space flight.
Flight Control system: Primary objective of a flight control system is stabilize (or enhance
stability margin) a flight vehicle and to enable the vehicle to perform to a desirable level in
the presence of system uncertainty, external disturbances and sensor noise. A control system
is robust if it remains stable and achieves desired performance criteria in the presence of
possible uncertainties. The robust design, which is the theme of this lecture note, is to find
a controller for a given system such that the closed loop system is robust. Fundamentally,
control system must trade-off performance with robustness against system uncertainty. The
means of steering/control depend on the vehicle and can be the rudder, elevators, and other
control surfaces on an airplane, the rudder on a ship, the control surfaces on a missile or on a
torpedo, the gimbal angle of the motor on a rocket, and others. Structure of controller varies
from one vehicle to another and purpose or utility of the vehicle. Controller can be organized
as feedback or cascade or feedback - feedforward configuration or as one, . . . , four block
configurations. Before dwelling upon the analysis and design control system, it is necessary
to understand the dynamics of a few important class of aerospace vehicles.
A brief account of the linearized equation of motion of an inverted pendulum and
Flight vehicles as well as control system preliminaries are given in the sequel.
1.1 Modelling of Dynamical System (LPS) : Inverted

Pendulum on a moving cart:
Equation of motion:
(M + m) ÿ + ml cos θ θ̈ − mlθ̇2 sin θ = f
ml cos θ ÿ − ml sin θẏ θ̇ + m2 l2 θ̈ − mgl sin θ = 0
Linearized equations:
(M + m)ÿ + mlθ̈ = f, mÿ + mlθ̈ − mgθ = 0
→ Unstable system, hence controller has to stabilize & give good performance, for which
u(t) = f (t) =?
Control input u(t) depends on
;θ
f M
Figure 1.1: Inverted pendulum on a mobile cart.
1. Type and no. of sensors
2. Control law
3. Design methodology
1.2. SATELLITE ATTITUDE DYNAMICS 7
state variable x(t) : θ(t), y(t), θ̇(t), ẏ(t).

dx
ẋ = = A x + bu; u=f
dt
     
 0 0 1 0   0   y 
     

 0 0 0 1 


 0 

 
 θ 
A = 

, b = 
 
, x = 
 


 0 − mg 0 0   1/m   ẏ 
 m     
     
0 (m+m)g
ml
0 0 −1/ml θ̇
     
 0 0 1 0   0    0
     

 0 0 0 1 


 0 


 0 

= 

, b = 
 
 , λi = 
 


 0 −.981 0 0   1   2.3228 
     
     
0 5.3955 0 0 −0.5 −2.3228
On taking Laplace transform
θ(s) −0.5s2 −0.5
= 2 2 = 2 (af ter pole zero cancellation)
u(s) s (s − 5.3955) (s − 5.3955)
=⇒ this system has similarity with the dynamics of statically unstable aircraft/ missile.
1.2 Satellite Attitude Dynamics

The spacecraft dynamics involves - Orbital Motion The space craft dynamics involves - Atti-
tude Motion −→ Rigid body dynamics
The space craft dynamics involves - Flexural vibration - Flexible body dynamics.
For modelling the satellite dynamics, the following idealization is needed : i.e, spacecraft is
modelled as a
– Rigid body
– Elastic body
– Combination of rigid bodies, or elastic bodies or rigid & elastic bodies.
In most of the satellites, coupling between orbital motion and attitude motion
is weak. Owing to the presence of external disturbances, control of orbit & attitude mo-
tion(AOCS - Attitude and Orbit Control System ) station keeping/orbit keeping, attitude
control, shape control/vibration control)are needed. In most of the past and present day
operational satellites, active vibration control/ shape control is not employed. The increas-
ing trend towards building light weight structures vibration frequency and also need to build
more and more accurate control system which results in larger control bandwidth, results in
control-structure interaction problem. At present, the attitude control system are designed in
which alleviation of control-structure interaction is also the design objective.
The attitude angles prescribe the orientation of a satellite with respect to inertial
reference frame or orbit reference frame. Owing to the environmental forces/torques, the
satellite’s attitude gets disturbed. The rotational of motion of satellite with respect to its
centre of mass(CM) is known as attitude motion. The disturbance induced undesirable atti-
tude motion. The disturbance induced undesirable attitude motion is commonly referred as
Liberations. The effect of environmental forces depends on the mission/satellite orbit. For
majority of satellites, orbital and attitude dynamics can be assumed to be independent. In
this chapter, dynamics and control of rigid satellites are discussed.
1.2.1 Attitude Representation
There are several ways of representing the attitude of a satellite Eulerean angle and quater-
nion representations are the most popular among these. Representation of satellite atti-
tude/orientation involve transformation/ mapping a vector from a reference coordinate sys-
tem to a body fixed coordinate system, through a series of attitude parameters. Therefore, a
minimum of three independent parameters are needed to specify the orientation of a satellite
in space. If the number of parameters are more than 3, then there exist constraining relation
amongst them, so that net degree of freedom is equal to 3.
Euler angles are three-parameter representation of satellite attitude : it involves
successive rotation about three axes in which any two successive rotation should not be about
the same axes. The Euler angles are not unique, and in fact there are 12 sets of Euler angles
for such successive rotations about the axes fixed in the body. Similarly, there exist 12 possible
sequence of axis rotation in inertial reference frame; called space-axis rotation.
x y x, x y z, xzx, xzy,
y x y, y x z, yzx, yzy,
z x z, z y z, zyx, zyz,
The discussion on the euler transformation matrix, Eulerean angles, and their rates have
already been given. Note that the Eulerean transformation matrix is a 3 × 3 orthogonal
matrix.
Euler’s eigen-axis rotation theorem states that by rotating a rigid body about an
axis that is fixed to the body and stationery in an inertial reference frame, the rigid body
attitude can be changed from any given orientation to any other orientation. Such axis of
rotation whose orientation remains unaltered with respect to both body and inertial frame is
called the Euler axis or eigen axis is, corresponding representation is Euler axis representation.
1
Recall that the direction cosine matrix for rotation the space object about one its reference
axis, say x-axis is given by
 
 1 0 0
 
 
A1 = 
 0 cosϕ sinϕ 

tr(A1 (ϕ)) = (1 + 2cosϕ)
 
0 −sinϕ cosϕ
Trace of the direction cosine matrix representing the rotation by an angle ϕ about any of X,
Y, or Z axis is same. In general for an eigen axis rotation, axis of rotation will not be one
of the reference axes. The most general direction cosine matrix is written in terms of units
vector along the rotation axis, ê and angle of rotation ϕ as
 
 cosϕ + − cosϕ)
e21 (1 e1 e2 (1 − cosϕ) + e3 sinϕ e1 e3 (1 − cosϕ) − e2 sinϕ 
 
 
 e1 e2 (1 − cosϕ) − e3 sinϕ cosϕ + e22 (1 − cosϕ) e2 e3 (1 − cosϕ) + e1 sinϕ 
A(ϕ) =  
 
e1 e2 (1 − cosϕ) + e2 sinϕ e2 e3 (1 − cosϕ) − e1 sinϕ cosϕ + e23 (1 − cosϕ)
= cosϕ I + (1 − cosϕ) êêT − sinϕ E
where, êêT is the outer product and E is the skew symmetric matrix.
 
 0 −e3 e2 
 
 
E=
 e3 0 −e1 

 
−e2 e1 0
This representation of the spacecraft orientation is called Euler axis or angle representation.
Here,tr(A) = 1+ 2 cosϕ. If sin ϕ ̸= 0, the components of ê are given by
e1 = (A23 − A32 )/2sinϕ
e2 = (A31 − A13 )/2sinϕ
e3 = (A12 − A21 )/2sinϕ
In 1general, a square matrix A is orthogonal if AAT =AT A = I = an identity matrix, so that A−1 = AT .
The quationion also known as the Euler symmetric parameters are defined by
q = [q1 q2 q3 q4 ]T
q1 = e1 sinϕ/2
q2 = e2 sinϕ/2
q3 = e3 sinϕ/2
q4 = cosϕ/2
 

 

 q 1 



 

{ } 
 
q  q2 
& =

 
q4 


q3 




 


 q 
4
The four Euler symmetric parameters ( or quaternions) are not independent (since, only 3
DOF are present), and they satisfy the constraint relation
q12 + q22 + q32 + q42 = 1
The quaternions can also be written in terms of general direction cosine matrix as
1
q4 = (1 + A11 + A22 + A33 )1/2
2
q1 = (A23 − A32 )/4q4
q2 = (A31 − A13 )/4q4
q3 = (A12 − A21 )/4q4

1
q1 = ± (1 + A11 − A22 − A33 )1/2
2
q1 = (A12 + A21 )/4q1
& so on... For z-x-y ( or 312) sequence of rotation (x −→ ϕ, −→ θ, z −→ yaw −→ ψ), the
direction cosine matrix can written as
A312 (ψϕθ) = A2 (θ)A1 (ϕ)A3 (ψ), i.e.,

 
 cosθcosψ − sinϕsinθsinψ cosθsinψ + sinϕsinθcosψ −cosϕsinθ 
 
 
A312 (ψϕθ) = 
 −cosϕsinψ cosϕcosψ sinϕ 

 
sinθcosψ + sinϕcosθsinψ sinθsinψ − sinϕcosθcosψ cosϕcosθ
The rotational angles can be expressed in terms of the elements of direction cosine matrix as
sinϕ = A23
tanψ = − (A21 /A22 )
tanθ = − (A13 /A33 )
if ϕ, θ & ϕ are all small angles (in radians), on using the small angle approximation, the
direction cosine matrix reduces to
 
 1 ψ −θ 
 
A312 (ψϕθ) = 
 −ψ 1

ϕ  ,
 
 
θ −ϕ 1
and quaternions q1 ≈ ϕ/2, q2 ≈ θ/2, q3 ≈ ψ/2, q4 ≈ 1.

Home work : Obtain the relation between quaternions and the Euler rotational angles (ϕ, θ, ψ).
The relation between the body axis fixed velocity vector {ω1 , ω2 , ω3 }T and the
quaternions given by
   

   



ω1 







2(q˙1 q4 + q˙3 q3 − q˙3 q2 − q˙4 q1 ) 



   
 ω2  =  2(q˙2 q4 + q˙3 q1 − q˙1 q3 − q˙4 q2 ) 

 
 
 


   
 ω3 
 
 2(q˙3 q4 + q˙1 q2 − q˙2 q1 − q˙4 q3 ) 

Combining the above with 0 = 2 (q˙1 q1 + q˙2 q2 + q˙3 q3 + q˙4 q4 ),

    

 ω 
 −q2 −q1  
 q˙1 




1 

  q4 q3 






 
   


 ω2  
  −q3 q4 
q1 −q2   q˙2 

 
= 2


 

 ω3 
  q −q1 q4 −q3  
 q˙3 


 
  2  


 
   


 0  
 
q1 q2 q3 −q 4 q˙  4
Since 4 × 4 matrix given above is orthogonal. (Prove !)

    

 q˙1 
 −q3 
 ω1 






  q4 q2 q1  






 
   


 q˙2  
 1 q q4 −q1 
q2  ω2 
 
=  3 
   

 q˙3 
 2  −q2 q3   ω 

 
  q1 q4  3 

 
  





 q˙  
 
4 −q 1 −q2 −q3 q4 0 
q T = [q1 , q2 , q3 ] and ω
or, In terms of − −,
1 1 T
q̇ = q4 ω − ωe ×q and q˙4 = − ω e q,
2   2 
 
 0 −ω3 ω2   
 q1 
 
  

 
e × q =  ω3
where ω 0 −ω1  q
  2 
 






 q3 

−ω2 ω1 0
which is derived from

    

 q˙1 
 −ω2 ω1 
 q˙1 






  0 ω3  


 
   


 q˙2    
 1 −ω3 0 ω1 ω2  q˙2 

= 



  



q˙3 



2  ω2
 −ω1 0 ω3 
 q˙3 



 
  
 



 q˙  
 
4 −ω 1 −ω2 −ω3 0 q˙4 
Unit Quaternion from Rotation Matrix [212] :

It is well known that a 3 X 3 proper orthogonal rotation matrix, or attitude matrix, can be
expressed in terms of a quaternion q
q = [q1 q2 q3 q4 ]T ; as
 
 q12 − q22 − q32 + q42 2(q1 q2 + q3 q4 ) 2(q1 q3 − q2 q4 ) 
 
A = 
 2(q1 q2 − q3 q4 ) −q12 + q22 − q32 + q42 2(q2 q3 + q1 q4 )

 (1.1)
 
 
2(q1 q3 + q2 q4 ) 2(q2 q3 − q1 q4 ) −q12 − q22 + q32 + q42
where the quaternion is assumed to have unit norm. Often one must find the quaternion
corresponding to a given rotation matrix. Of the many methods that have been proposed
for performing this computation, Shepperd’s algorithm (as per [212]), which is singularity
free and requires only one square root, has been the most widely applied. Markley presents a
variant that always produce a normalized quaternion even if numerical errors cause the matrix
A to be only approximately orthogonal in [212]. This modification of Shepperd’s algorithm
also provides a very efficient method for computing an exactly orthogonal matrix that is close
to an approximately orthogonal matrix [212].
Shepperd’s algorithm : The following relations for all the products of two quaternion com-
ponents are derived from the above definition of quaternion and attitude matrix:
4 qi2 = 1 + Aii − Ajj − Akk = 1 − tr(A) + 2Aii (Qa)

4 q42 = 1 + A11 + A22 + A33 = 1 − tr(A) + 2tr(A) (Qb)
4 qi qj = Aij + Aji (Qc)
4qi q4 = Ajk − Akj (Qd)
where {i. j. k}is a cyclic permutation of {1, 2, 3} and tr(A) denotes the trace of A. Shepperd’s
algorithm first compares the right sides of Eqs.(Qa)and (Qb)to see which of the quantities
qi2 for i = 1, 2, 3, 4 is largest.The unusual form of expressing the rightmost member of
Eq.(Qb)shows that this is equivalent to finding the largest of tr(A) and Aii , saving some
computation. Note that all the quantities 4qi2 are positive and that ∥q∥ or Eqs.(Qa) and (Qb)
must be greater than or equal to unity.
In the case that q42 is larger than any of the other qi2 , Shepperd’s algorithm com-
puters q4 from Eq.(Qb) and the other components from Eq.(Qd), giving
1 1
q4 = ± (1 + A11 + A22 + A33 ) 2
2
qi = (Ajk − Akj )/4 q4 ; f or i = 1, 2, 3
Here, the well-known twofold sing ambiguity in the quaternion can be observed. If qi for i ̸= 4
is the largest quaternion component in magnitude, it is computed from Eq.(Qa) and the other
quaternion components are computed from Eqs.(Qc) and (Qd), giving
1
qi = ± (1 + Aii − Ajj − Akk )
2
qj = (Aij + Aji )/4qi
qk = (Aik + Aki )/4qi
q4 = (Ajk + Akj )/4qi
where {i, j, k} is a cyclic permutation of {1, 2, 3} as before. Shuster and Natanson have
commented on the relation between the four possible branches in Shepperd’s algorithm and
the method of sequential rotations [213]. Shepperd’s algorithm is guaranteed to produce a
precisely normalized quaternion only if A ia precisely. Markley’s algorithm in [212] overcomes
this deficiency in the following manner.
Modified Shepperd’s algorithm [212]: For the modification of Shepperd’s algorithm, consider
the four 4-component vectors
∆
x(i) = 4 qi q; f or i = 1, 2, 3, 4. (1.2)
Each of the four components of each x(i) is given by the right side of one of the equations of
Eqs. (Q), so these vectors are easily computable from the components of the rotation matrix.
Explicitly,
 
 1 + A11 − A22 − A33 
 
 
 A12 + A21 
x(1) = 



 A13 + A31 
 
 
A23 − A32
 
 A21 + A12 
 
 
 1 + A22 − A33 − A11 
x(2) = 



 A23 + A32 
 
 
A31 − A13
 
 A31 + A13 
 
 
 A32 + A23 
x(3) = 



 1+A −A −A 
 33 11 22 
 
A12 − A21
 
 A23 − A32 
 
 
 A31 − A13 
x(4) = 


 (1.3)
 A12 − A21 
 
 
1 + A11 + A22 + A33
Equation (1.2) shows that each of the x(i) is a scalar multiple of q, so we can obtain the unit
quaternion by computing and normalizing any one of the x(i)
q = ±x(i) /∥x(i) ∥ (1.4)
As in Shepperd’s method, choosing the x(i) corresponding to the maximum value of qi2 mini-
mizes numerical errors. This selection is made by Shepperd’s procedure of finding the largest
of trA and Aii. This method of extracting a quaternion from a rotation matrix requires
the same number of square roots as Shepperd’s method, namely one, but one more divi-
sion and a few more additions and multiplications. A slightly modified form of this method
was previously used to extract a quaternion from a scalar multiple of the attitude matrix [214].
Orthogonalization : Strapdown inertial systems often employ the numerical integration of a

rotation matrix that can lose orthogonality due to accumulation of roundoff error. Various
methods have been proposed to restore orthogonality, some approximate or iterative and
often requiring matrix inversion. Equations (1.3) and (1.4) provide an exact (within machine
precision) noniterative method for restoring orthogonality by simply substituting the result
of Eq. (1.4) into Eq. (1.1) to produce a new matrix Aorth . The normalization condition
guarantees that Aorth is orthogonal. Because Eq. (1.1) is a homogenous quadratic function of
q, the square root in Eq. (1.4) can be avoided if we only need to compute Aorth and not q; the
matrix elements of Aorth are rational functions of the matrix elements of A. This computation
is much less expensive than any previously proposed for orthogonalization.
1.2.2 Basics of Satellite Attitude Dynamics
Motion of space object [2, 3, 300] can be described in terms of a) Inertial coordinate system;
b) Moving coordinate system, ex: moving along with centre of mass/gravity; c) body fixed
coordinates, ex : coordinates aligned along principal moment of attitude motion are decom-
piled, an orbit fixed moving coordinate is used as a reference frame. The coordinate along
the orbit normal is called pitch axis (Y, θ), axis along local vertical directed away from the
earth’s centre is known as yaw axis (Z, ψ) and the roll axis is perpendicular to both(X, ϕ).
For the circular orbit, the roll axis is tangent to the orbit and is directed along the
velocity vector. For a rigid satellite with constant moment of inertia, the equation of attitude
Roll,X
θ
Z1 Yaw
φ ψ
θ
pitch,Y
Figure 1.2: Orbit Fixed reference Coordinate System.

motion in an inertial frame of reference can be written using Euler’s equations as

dω
I dt
= M + T , M = control torque, T = disturbance torque, I = moment of inertia.
The equation of motion in the moving reference frame is, therefore, I ω̇ + ω × I ω = T.
If the principal moment of inertia axis is referenced, then the product of inertia terms are
zeros. Then, the equation of motion along the principal axes are given by
Roll : Ix ω̇ + (Iz − Iy )ωy ωz = Tx + Mx
P itch : Iy ω˙y + (Ix − Iz )ωz ωx = Ty + My
Y aw : Iz ω˙z + (Iy − Ix )ωx ωy = Tz + Mz
For the near earth satellites, Aerodynamic drag, gravity gradient torque are important sources
of perturbations, while, solar radiation pressure is dominant at high altitude.
The net gravitational forces acting through the centre of gravity produces gravity produces
gravity gradient torque about the centre of mass. If centre of mass centre of gravity and
earth’s centre are collinear then gravity gradient torque is zero. The gravity gradient torques
for a triaxial satellite is given by the expression :
3µe
Tx = (Iz − Iy )Ry Rz
R5
3µe
Ty = (Iz − Ix )Rz Rx
R5
3µe
Tz = (Iy − Ix )Rx Ry
R5
where Rx , Ry , Rz are the components of the satellite position vector R along the principal
body axes. The transformation from the body axes to the orbit fixed axes can be given in
terms of Euler angles (ϕ, θ, ψ) {both the set of symbols are commonly used in the text books}.
As already mentioned, the Euler angles are not unique and depend on sequence of rotation.
The transformation matrix A123 (θ1 , θ2 , θ3 ) is obtained by first rotation about z-axis by an
angle θ3 , next rotation about the resultant y=axis by angle θ2 and finally rotation about the
resultant x-axis (after two rotation) by an angle θ1 . For small angles and angular velocities
and for earth oriented satellites,
ωx ≈ θ̇1 − Ω0 θ3 ; ωy = θ˙2 − Ω0 , ωz = θ̇3 − Ω0 θ1
where, Ω0 = orbital angular velocity (Note : symbol ‘n’ is also used in many text book).
Then, the governing equation is
Roll : Ix ω˙x − (Iy − Iz )ωy ωz = Mx − 3Ω22 (Iy − Iz )θ1
P itch : Iy ω̇y − (Iz − Ix )ωz ωy = My + 3Ω20 (Iz − Ix )θ2
Y aw : Iz ω˙z − (Ix − Iy )ωx ωy = Mz
In terms of Euler angles, the above equations reduce to
Ix θ̈1 − Ω0 (Ix − Iy + Iz )θ˙3 + 4ω02 (Iy − Iz )θ1 = 0
Iy θ̈2 + 3Ω20 (Ix − Iz )θ2 = 0
Iz θ̈3 + 3Ω0 (Ix − Iy + Iz )θ˙1 + ω02 (Iy − Ix )θ3 = 0.
For small angles, θ1 , θ2 , θ3 are after called as roll, pitch and yaw attitude angles of the satellite
respectively.
The presence of environment disturbance forces/ torques, and internal sources, make the
satellite to tumble or more away from the desired orientation (or equilibrium position). Hence,
all satellites need means of stabilization and control. Stabilization implies, the means to bring
the satellite to equilibrium each time it gets disturbed. Many fascinating schemes have been
devised in the past to provide the mechanism of attitude stabilization/control. These can
be broadly classified as - passive, Active, semi possive/semiactive. The active systems use
energy available onboard the satellite. The passive systems exploit the environmental forces
for stabilization and control. The useful environmental forces are : Gravity-gradient, solar
radiation pressure, aerodynamic forces, earth’s magnetic field interaction.
Active Controls : Passive controls :
– mass expulsion system – spin stabilization
– Reaction wheel system – gravity gradient stabilization
– Momentum wheel system
– Gyro-torquers
– semi possive/semiacting controls :
– Dual spin
– aerodynamics
– solar radiation pressure
–gravity gradient with active damping.
1.2.3 Linear model of Three Axis Stabilized Satellites
Most of the current day satellites are three-axis stabilized. They use either reaction wheels
(satellites in inclined orbits like IRS i.e., sun synchronous orbit) or momentum biased wheel
(satellites in equatorial orbit, like, Geostationary communication satellites like INSAT). The
flywheels act like a momentum exchange devices. The reaction wheel has zero nominal speed
(i.e., zero stored angular momentum) while momentum biased wheel has non-zero nominal
speed (i.e., non-zero stored angular momentum) and provide gyroscopic stiffness. In all these
satellites, flywheels act like a primary source for attitude control. For satellites subjected to
large magnitude of disturbances (say manned mission), gyro torquers are used. Here, only
the first two types are considered.
The linearized governing equations of motion are:
Roll : Ix ϕ̈ = −hw Ω0 ϕ + hw ψ̇ + Mx + Tx + Tgx (1.5)
P itch : Iy θ̈ = My + Ty + Tgy (1.6)
yaw : Iz ψ̈ = −hw Ω0 ψ − hw ϕ̇ + Mz + Tz + Tgz (1.7)
where, Ix , Iy , Iz are MI of the satellites about roll, pitch and yaw directions( assumed to be
the principal axes), hw wheel angular moment (nominal), Ω0 = orbital angular velocity Tg
= gravity gradient torque, T = other disturbance torques, M = control torque. For a large
number of satellites, Ix ≈ Iz and hence Tgx ≈ 0. For biased momentum wheels, the torque
about the pitch axis is generated by accelerating or decelerating the wheel (clock wise or anti
clock wise accelerations) and for the other two axes either reaction jets are magnetic torquer
or both are used. For the satellites in the inclined orbits, all the three reaction wheels have
zero stored angular momentum (nominal zero angular speeds). Hence, the term hw = 0, which
results in decoupling of roll - pitch - yaw dynamics and all the three degrees of freedom have
similar equations of motion. Thus, identical controller structure can be used for all axes in
these satellites. For the communication satellite, separate controller for pitch and roll - yaw
dynamics are needed.
Pitch attitude stabilization system
The control torque for pitch motion is obtained by either accelerating or decelerating the fly
wheel so as to over come the effect of disturbances. The disturbance are of two kinds, viz.,
periodic( or harmonic) and secular (monotonic). The fly wheels are extremely effective in an
nulling the periodic disturbance with the proper sizing of the wheel, complete cancellation
of effects of periodic disturbance is possible. On the other hand, the secular disturbance
causes the continued acceleration (or deceleration) of the wheel which will eventually lead to
saturation of wheel speed (close wise or anti-clock wise) after certain time. The saturation of
wheel calls for momentum dumping, i.e., decelerate the wheel rapidly and at the same time
provide counter torque using, say, reaction jets or magnetic torquer. The sizing of the wheel
is done such that the wheel size and nominal speed of the wheel are not too large and the
time internal between two momentum dumping is acceptable.
Analysis of pitch dynamics is straightforward because pitch motion is decoupled
from the others. The linearized pitch equation can be written as
Ty = Iy θ̈ + 3Ω20 (Ix − Iz )θ + My
This is further simplified since the satellite of interest is symmetric i.e., Ix = Iz . Thus,
Ty = Iy θ̈ + My
where My represents rate of change of wheel speed to impose direct torque about the pitch
axis. Then, the transfer function for the idealized pitch dynamics is given by
θ(s) 1
=
My (s) Iy s2
which indicates that the dynamics is similar to the double integrator plant, which is inherently
unstable. This equation (1.8) indicates that no natural damping is available and can be
provided only through the control function hw . A simple PD control in pitch feedback loop
is sufficient to meet the stabilization requirement. Discussions of pitch stabilization control
system is discussed in the Chapter 3.
Coupled Roll-Yaw attitude Dynamics
Coupling between roll and yaw axes is a result of the biased momentum wheel (nonzero
nominal speed) orientation, which is nominally along the orbit normal. An attractive feature
of bias momentum is that it does provide roll / yaw coupling while permitting accurate yaw
control without a direct yaw sensor. Control torques are produced through gimbal deflections.
Since direct roll sensing is available through the horizon sensor, this axis can be controlled
directly in response to roll errors. Yaw deviations are sensed indirectly through coupling with
the roll axis. This technique is sometime referred to as gyrocompassing. Bias momentum
magnitude hw and system gain are selected to make the roll controller sensitive to yaw errors
(i.e., artificially increasing the coupling effect). However, this must be done so that roll errors
are not transformed into yaw errors. The general linearzed equations for roll and yaw were
given as expressions (1.5, 1.7), respectively. Significant coupling effects are achieved through
large values of bias momentum. This implies that a condition may be imposed on the nominal
momentum
hw >> max[Ix Ω0 , Iy Ω0 , Iz Ω0 ]
where, Ω0 = orbital angular velocity. The linearized roll / yaw equations become
Tx = Ix ϕ̈ + Ω0 hw ϕ + hw ψ̇ + Mx (1.8)
Tz = Iz ψ̈ + Ω0 hw ψ − hw ϕ̇ + Mz (1.9)
where hw = nominal wheel angular momentum. It is interesting to briefly consider uncon-

trolled (gimbals fixed) yaw response. Set of equations (1.8,1.9) are expressed in Laplace
variable form as
    
2
 Tx (s)   Ix s + Ω0 hw hw s   ϕ(s) 
 =  
Tz (s) −hw s Iz s2 + Ω0 hw ψ(s)
For controlling the coupled roll - yaw dynamics, one can use one or two sets of -
thrusters or magnetic torquers. Owing to the quarter orbit coupling, roll rate sensor along
with the controller will also null the yaw error. If faster correction is needed, two sets of
actuators are needed. The transfer function for the roll response ϕ(s) for the roll and yaw
axis disturbance torques [Tx (s), Tz (s)] are given by
Tx (s)(Iz s2 + hw Ω0 ) − Tz (s)hw s
ϕ(s) =
Ix Iz s4 + [(Ix + Iz )hw Ω0 + h2w ]s2 + h2w Ω20
Similarly, yaw response to a yaw disturbance torque Tz can be solved from:
ψ(s) Iz s2 + Ω0 hw
=
Tz (s) (Ix s2 + Ω0 hw )(Iz s2 + Ω0 hw ) + h2w s2 )
1.3. AIRCRAFT DYNAMICS 21
A step yaw disturbance torque yield the response
Tz Ix Tz hw
ψ(t) = (1 − cos(Ω0 t)) + 2 [1 − cos( )t]
Ω0 hw hw I x
noting that Ix = Iz . Two phenomena are represented here. The first periodic term is associ-
ated with roll coupling into yaw at orbital rate ω0 , while the second is related to precessional
effects of the momentum wheel. The latter motion is a short period oscillation whose fre-
quency and magnitude depends on the value of nominal momentum.
The characteristic equation for the open loop system is
Ix Iz s4 + [(Ix + Iz )hw Ω0 + h2w ]s2 + h2w Ω20 = 0
or ∆ = (s2 + wN
2
)(s2 + Ω20 ) = 0
hw
√
where ωN = I
= nutational frequency and I = Ix Iz . It may be noted that the characteristic
equation does not contain terms with s3 and s. Hence, roll-yaw motion is an undamped
oscillatory motion. In order to enhance the speed of disturbance of rejection (related to closed
loop frequency) and damping, PD - controller is needed. Chapter 3 deals with the analysis of
SISO PD controller for roll angle feedback stabilization system.
1.3 Aircraft Dynamics

The aircraft dynamics is nonlinear and coupled [7]. To derive the equation of motion body
fixed (at C.G.) cartesian coordinate systems are used. The axis of the coordinate system is
taken with OX towards forward direction, OY out along right wing and OZ downward as seen
by the pilot, to form the right handed axis system. The axes OX and OZ lie in the plane
of symmetry of the aircraft, hence corresponding product of Inertia terms Ixy and Iyz are
zeros. An alternate stability axis system fixed at CG [35, § 2.3] of an aircraft will be discussed
later. Linearized equations of motion [7, 35] during wing level flight about a given operating
conditions (trim condition/ equilibrium condition) are used for the control system analysis and
design. Typically number of operating points vary from tens of thousand to a couple of million
points. The control system design entails the construction of automatic flight control system
(AFCS) at each of these operating points and gain/ controller scheduling these controller
between subsequent operating conditions. With the advent of modern control theory, number
of operating points for the controller design has shrunk to a large extent. The aircraft has one
plane of symmetry, hence pitch dynamics is nearly decoupled from roll-yaw dynamics. These
dynamics are known as longitudinal dynamics and lateral dynamics. Various variables used
in the equations of motion are given in the accompanying table.
Longitudinal Lateral
Rates α: angle of attach β : Side slip angle
q : Pitch rate p : roll rate
u : change in forward speed r : yaw rate
Positions θ : Pitch ϕ : roll angle
h or z : altitude ψ : yaw angle
y : cross -track displacement
acceleration az :longitudinal acceleration ay : lateral acceleration
Controls δe : elevator deflection δa : aileron deflection
δr : rudder deflection
Longitudinal Dynamics :
u̇ = Xu u + Xα α − gθ + Xe δe (1.10)
Zu Zα Ze
α̇ = u+ α + q + δe (1.11)
V V V
q̇ = Mu u + Mα α + Mq q + Me δe (1.12)
Kinematics : θ̇ = q; ḣ = V γ̇ = V (α̇ − q). (1.13)
Lateral Dynamics :
Yβ Yp Yr g Ya Yr
β̇ = β + p + ( − 1)r + ϕ + δa + δr (1.14)
V V V V V V
ṗ = Lβ β + Lp p + Lr r + La δa + Lr δr (1.15)
ṙ = Nβ β + Np p + Nr r + Na δa + Nr δr (1.16)
Kinematics : ϕ̇ = p (1.17)
Kinematics : ψ̇ = r (1.18)
The symbols X, Y, Z, L, M and N with a subscript have become fairly standardized in the
field of aircraft and missile control, even though sign conventions change from one another.
These variables with a subscript are also called aerodynamic stability derivative (S.D).
1.3.1 Longitudinal Dynamics
For stable longitudinal dynamics, aircraft exhibits two distinct modes, viz., short period mode
(faster mode with a short-period oscillations) and phugoid mode (slower mode with a long-
period oscillations). Wikipedia Definition: “the phugoid mode is the one in which there is a
large-amplitude variation of air-speed, pitch angle, and altitude, but almost no angle-of-attack
variation. The phugoid oscillation is really a slow interchange of kinetic energy (velocity) and
potential energy (height) about some equilibrium energy level as the aircraft attempts to
re-establish the equilibrium level-flight condition from which it had been disturbed. The
phugoid motion consists of large oscillatory change in speed u and height h. The motion
is so slow that the effects of inertia forces and damping forces are very low. Although the
damping is very weak, the period is so long that the pilot usually corrects for this motion
without being aware that the oscillation even exists. Typically the period is 20-60 seconds.”
On the other hand, the short period motion is “a rapid pitching of the aircraft about the
center of gravity. The period is so short that the speed does not have time to change, so the
oscillation is essentially an angle-of-attack variation”. Recall that heave velocity w ≃ V α for
small angles. Pitching moment Coefficient Mu (change in pitching moment due to change in
speed) is small or nearly zero for trimmed level flight. Its magnitude can vary significantly
with change in forward speed and influence aero-elastic effects. Change in pitching moment
coefficient with α is Mα , where, Mα > 0 for an unstable aircraft, since Centre of pressure
(CP) is ahead of Centre of gravity (CG), so that aerodynamic force has destabilizing effect
(similar to the gravitational force on an inverted pendulum). Similarly, Mα < 0 for a stable
aircraft, since then, aerodynamic force acts like a restoring force as in a regular pendulum. The
derivative Cmα i.e., Mα is the slope of the curve of static pitching moment coefficient, around
CG, versus α, which controls the neutral. It plays an important role in dynamic behavior of
pitching motion. For an aircraft with horizontal tail, the tail efficiency factor decreases with
the increase in α depending on the degree of coupling between wing and horizontal tail (wash
out effect). Pitch damping derivative Cmq i.e., Mq is normally negative and has influence
on handling qualities. It is estimated from oscillatory motion of aircraft/ aircraft model, or
calculated. For a tail controlled aircraft, substantial damping coefficient can be realized. Pitch
rate determines translational rate of the horizontal tail perpendicular to the relative wind,
which changes the tail angle of attack, hence tail moment about CG. The change in pitching
moment is linearly related to the pitch rate. Mq contributes substantially to the damping of
short period motion. Zα is related to lift curve slope and is change in lift coefficient with a
change in angle of attack α. It is contributed by the wing, tail and fuselage. In conventional
aircraft, wing contributes nearly 85 to 90%. The lift-curve slope CLα is important, since it
also reflects that the wind induced disturbance αw (Note: αtotal = α + αw ) changes lift, hence
affects the level of comfort for the pilot. By the same argument, it affects the maneuverability
of the aircraft. To a smaller extent, it influences damping of pitching motion. Therefore,
lift-curve slope CLα influences the handling quality of the aircraft. For cruise and normal
operating conditions, it is independent of α. Control related coefficient Ze is normally small
for a conventional wing - tail configurations and can be neglected in the first order analysis.
For tailless aircraft, moment arm from CP of the flap to CG can be small, then Ze can also
be comparable to Me and hence can not be neglected. Me is elevator control effectiveness.
When elevator is located aft of CG, Me < 0. The sizing of elevator, therefore, value of Me is
determined by the maximum wing lift and also the range of CG travel that can occur during
a flight. Impact of Important longitudinal stability derivatives (S.D.)[7] are given below.
S. D. Quantity most affected How affected

CLα ⇔ Zα Lift - curve slope Detrmines ride quality
Cmq ⇔ Mq Damping of the short period Increase in Mq increases the damping
Cmα ⇔ Mα Mα > 0: unstable Larger Mα means larger instability
Mα < 0: stable :
Natural frequency of short period Increase in Mα increases frequency ωns
Cxu ⇔ Xu Damping of the phugoid Increase in Xu increases phugoid damping
Czu ⇔ Zu /V Natural Frequency of the phugoid Increase Zu /V increases phugoid frequency
For example, for F-16 aircraft (during landing phase), the stability derivatives at V = 139Kt:
Xu = −0.0507, Xα = −3.861, Xe = 0
Zu Zα Ze
= −0.00117, = −0.5164, = −0.0717
V V V
Mu = −0.000129, Mα = 1.4168, Mq = −0.4932, Me = −1.645
Note : |Me | > |Mα |. The state space representation of the longitudinal dynamics:
     

 u 
 −g 





  Xu Xα 0  Xe 

 
    
 α 
 
  Zu /V Zα /V 1 0 


 Ze /V


x = , u = δe ; A = 

, B = 
 

 (1.19)

 



q 




 Mu Mα Mq 0 


 Me 


 
    

 θ 
 0 0 1 0 0
   
 −0.0507 −3.8610 0 −32.185   0 
   
   
 −0.0012 −0.5164 1.0000 0   −0.0717 
A = 

 B=
 

 (1.20)
 −0.0001 1.4168 −0.4932 0   −1.6450 
   
   
0 0 1.0 0 0
and the corresponding characteristic equation:
∆(s) = s4 + a1 s3 + a2 s3 + a2 s2 + a3 s + a4 = 0;
Zα
where a1 = − − Mq − Xu = 1.0603,
V
Zα Zα Zu
a2 = M q − M α + Xu ( + Mq ) − Xα = −1.1154,
V V V
Zα Zu
a3 = −Xu ( Mq − Mα ) + Xα ( Mq − Mu ) = −0.0565,
V V
Zu Mα − Zα Mu
a4 = g = −0.0512
V
The roots of the characteristic equations [1, p.128] are at -1.705, +0.724, -0.0394 ± 0.20j.
Observe that the short period mode is unstable (since Mα > 0) and the phugoid mode has a
frequency of 0.204 and damping of 0.19. Vehicle is statically unstable. The phugoid motion
is due to change in forward speed which manifest as slight oscillation in altitude. Though the
damping of phugoid is poor, the pilot can handle the statically stable aircraft without much
difficulty, since the time corresponding constants are normally large, in the range of say, 50 to
150 seconds. For tiny UAVs, time constants are short, hence are difficult to fly. Robust flight
flight stabilization is required for an unstable aircraft. Unstable aircraft require less control
effort for maneuvers. The eigenvector (P ) - eigen value (λ) pairs are:
   
 0.9943 0.9995 1.0000 1.0000   −1.7036 
   

 0.0633 −0.0142 0.0005 + 0.0003i 0.0005 − 0.0003i 


 0.7310 

P =

λ=
 


 −0.0740 −0.0165 0.0013 + 0.0002i 0.0013 − 0.0002i   −0.0438 + 0.2065i 
   
   
0.0435 −0.0226 −0.0003 − 0.0064i −0.0003 + 0.0064i −0.0438 − 0.2065i
On transforming state space model to modal canonical form (Use matlab command - [Ad, Bd,
Cd, Dd] = canon(A, B, C, D, ’modal’), where C = I; one gets the canonical realization:
 
 −1.7036 0 0 0 16.4769 
 
 
 0 0.7310 0 0 18.6529 
 
 
 0 0 −0.0438 0.2065 5.5937 
 
   
 
 Ad Bd  
 0 0 −0.2065 −0.0438 −1.2997 

 = 
 
Cd Dd 

0.6223 2.5414 −6.0712 18.2331 0.0 

 
 0.0396 −0.0360 −0.0078 0.0072 0.0 
 
 
 
 −0.0463 −0.0420 −0.0123 0.0231 0.0 
 
 
0.0272 −0.0574 0.1192 0.0342 0.0
Since, Output matrix C = I, components of output vector are state variables. Each row
of Cd matrix (lower left partition of the realization) indicate the relative contribution of
different modes to state variable’s impulse response. The state variables are ordered such that
x = [u, α, q, θ]T . The change in forward speed u and pitch attitude θ are dominated by
phugoid mode; pitch rate p and angle of attack α have larger contributions from short period
modes. Transfer functions for the state variables are
u(s) (s + 81.3145)(s + 0.4145)
= 0.2768 ; (N ote : unstable)
δe (s) ∆(s)
α(s) (s + 23.4365)(s2 + 0.0502s + 0.1057)
= − 0.0717 ;
δe (s) ∆(s)
q(s) s (s + 0.5865)(s + 0.0423)
= − 1.645 ;
δe (s) ∆(s)
θ(s) (s + 0.5865)(s + 0.0423)
= − 1.645
δe (s) ∆(s)
Example 2: Transport aircraft Longitudinal dynamics [7, p.41]:
Equation of motion in Laplace variable for a statically stable transport aircraft (See Blakelock’s
book [7], Chapter 1) are given by
(13.78 s + 0.088) u(s) − 0.392 α(s) + 0.74 θ(s) = 0 (1.21)
1.48 u(s) + (13.78 s + 4.46) α(s) − 13.78 θ(s) = −0.246 δe (s) (1.22)
(0.0552 s + 0.619) α(s) + (0.514 s2 + 0.192 s) θ(s) = −0.71 δe (s) (1.23)
Then, the corresponding transfer functions are:
∆1 (s) = {(s2 + 0.00466s + 0.0053) (s2 + 0.806s + 1.311)}

u(s) (s − 68.8)(s + 0.6)

= − 0.000506 ; (N ote : stable)
δe (s) ∆1 (s)
α(s) (s + 77.8)(s2 + 0.0063s + 0.0057)
= − 0.01785 ;
δe (s) ∆1 (s)
θ(s) (s + 0.3)(s + 0.016)
= − 1.31
δe (s) ∆1 (s)
The short period poles are located at −0.4030 ± 1.0717i and phugoid poles at −0.0023 ±
0.0728i.
Depending on the purpose of analysis or preliminary design study, the longitudinal
dynamics of an aircraft is represented using the short period approximation and phugoid
approximation. It is observed that the state variables α and q influence short period modes
significantly, while, states u and θ influence phugoid mode.
Short period approximation
The short period approximation of pitch plane dynamics is primarily the motion involving
pitch rate q and angle of attack / incidence α with the change in forward speed ∆u = 0 of
forward speed remaining constant. For steady level flight θ is assumed to be constant.
      

 α̇ 
 Zα
1  α 
   Ze 

= 

V
 + V
e δ

 q̇ 
   
Mα Mq  q   Me
    

 α 
   −0.0717 

 −0.5164 1.0000 
=   + δ
    e
1.4168 −0.4932  q   −1.6450 
The characteristic equation, then reduces to
s2 − [Mq + (Zα )/V )]s + [Mq Zα /V − Mα ] = s2 + 2ζs ωs s + ωs2 = 0
where 2ζs ωs = 1.01 and ωs2 = -1.1621. The egenvector (P ) - eigen value (λ) pairs are:
   
 −0.6469 −0.6396   −1.6952 
P =  λ= 
0.7626 −0.7687 0.6856
Since the F-16 aircraft is aerodynamically unstable (CP ahead of CG), ωs2 < 0. The poles are
at : -1.695 and + 0.685. Transfer function:
α(s) Ze [s + V (Me /Ze )] q(s) Me (s − Zα /V )

= ; = ; where, ∆sp (s) = s2 + 2ζs ωs s + ωs2
δe ∆sp (s) δe ∆sp (s)
For stable motion, aerodynamic derivatives provide stiffness and viscous damping. Normally,
both lift curve slope Mα and Mq , which is determined by viscous paddle damping effect of
tail plane, are negative numbers. Mα is a measure of controls fixed stability margin of the
aeroplane. ωs2 is the measure of control fixed maneuver margin of the aircraft.
It may be noted that for pitch autopilot design, it is possible to use short period
approximation during the straight and level flight (say, cruise of a transport aircraft) and get
a simple controller that can give satisfactory performance for the full order model.
Phugoid Approximation
On rearranging the states xT = (α, q, u, θ), and rewriting the equation of motion as
   
 A11 A12   B1 
ẋ =  x +   δe ,
A21 A22 B2
       

Zα
V
1   Zu /V 0   Xα 0   Xu −g 
where, A11 =   ; A12 =   ; A21 =   ; A22 =  
Mα Mq Mu 0 0 1 0 0
   
−0.5164 1  −0.0012 0 
A11 = 


 ; A12 =  ;
1.4168 −0.4932 −0.0001 0
   
 −3.8610 0   −0.0507 −9.8100 
A21 =   ; A22 =  
0 1 0 0
the phugoid approximation (since short period motion is not important) is written as
   

 α 

 u 
0 = A11 + A12  

 q 
 θ
     
 u̇   α   u 
  = A21   + A22  
θ̇ q θ
   
 u̇  −1  u 
theref ore,   = (A22 − A21 A11 A12 )  
θ̇ θ
  
 −0.0530 −9.8100   u 
=   
0.0015 0 θ
The corresponding eigenvalues of phugoid modes of F-16 aircraft are −0.0265±0.1187j (which
are at −0.0394 ± 0.20j for the full order model). In this model, effect of incremental change
in thrust is not included. For conventional aircraft, such an influence is small. There are
several phugoid approximations available in open literature. They can give better frequency
and damping approximation for phugoid than the above simple model.
The characteristic equation and the frequency and damping of phugoid motion are given by
Mu (V Xw − g) g Mu Zα
− V Mw {s2 − [Xu + ] − [Zu − ]} = 0
V Mα V Mα
−g M u Zα Mu (V Xα − g)
2
ωph = [Zu − ]; 2 ζph ωph = − [Xu + ]
V Mα V Mα
where, w ≃ V α, stability derivatives with respect to w nd α are correspondingly scaled.

For the two examples listed above, the phugoid mode is sable with the complex conjugate
eigenvalues. For supersonic aircraft or the aircraft which fly in the transonic regime, the
2
stability derivative Mu takes sufficiently larger negative value (i.e., ωph < 0) to result in
a positive and negative real phugoid eigenvalues. The unstable phugoid mode is known as
divergent tuck mode.
For a level flight, nominal flight path angle γ0 is zero. If Lanchester’s approxima-
− g Zu
2
tion: i.e., Mu = 0 is used, then, ωph = V
; 2 ζph ωph = − Xu . Therefore, the characteristic
equation for phugoid is [s2 − Xu s − g Zu
V
] = 0. Note that Zu ≃ ρS V
m
CL , where, S = reference
area, which is generally normalized with respect to wing area.
weight 2mg
For steady, straight and level flight, the lift coefficient, CL = e
qS
= ρV 2 S
; where, qe =
−2 g
dynamic pressure, m = mass. Thus, Zu = V
. Hence, the simplified expression for phugoid
√
frequency is ωph ≃ 2 Vg .
Assumption of Mu = 0 leads to the assumption the change in coefficient of drag CD due to
change in u is zero. Then, ζph = − 2Xωuph = − (2X√u2)V g .
Since, Xu ≃ 2 eqmSVCD and CL = m e
qS
g
; ⇒ Xu = 2Vg CCDL . On substituting Xu for ζph , one gets,
ζph = −( 2Vg CCDL ) (2 √V2) g = √2 (L/D)
1
, where, L/D = lift/ drag ratio of the aircraft.
For example, for an aircraft flying at V = 210m/sec, and L/D = 12,

√
ζph = √2 1× 12 = 0.06 and ωph
2
= 2 9.81
210
= 0.0661 rad/sec.
1.3.2 Lateral Dynamics
“Lateral-directional” modes involve rolling motions and yawing motions. Motions in one of
these axes almost always couples into the other so the modes are generally discussed as the
“Lateral-Directional modes”. Three types of possible lateral-directional dynamic motion are
roll subsidence mode, Dutch roll mode, and spiral mode. (Refer: Wikipedia).
Roll subsidence mode: Roll subsidence mode is simply the damping of rolling motion. There
is no direct aerodynamic moment created tending to directly restore wings-level, i.e. there is
no returning ”spring force/moment” proportional to roll angle. However, there is a damping
moment (proportional to roll rate) created by the slewing-about of long wings. This prevents
large roll rates from building up when roll-control inputs are made or it damps the roll rate
(not the angle) to zero when there are no roll-control inputs. Roll mode can be improved by
adding dihedral to the aircraft design, such as high wings, dihedral angles or sweep angles.
Spiral mode: If a spirally unstable aircraft, through the action of a gust or other disturbance,
gets a small initial roll angle to the right, for example, a gentle sideslip to the right is produced.
The sideslip causes a yawing moment to the right. If the dihedral stability is low, and yaw
damping is small, the directional stability keeps turning the aircraft while the continuing
bank angle maintains the sideslip and the yaw angle. This spiral gets continuously steeper
and tighter until finally, if the motion is not checked, a steep, high-speed spiral dive results.
The motion develops so gradually, however that it is usually corrected unconsciously by the
pilot, who may not be aware that spiral instability exists. If the pilot cannot see the horizon,
for instance because of clouds, he might not notice that he is slowly going into the spiral dive,
which can lead into the graveyard spiral. To be spirally stable, an aircraft must have some
combination of a sufficiently large dihedral, which increases roll stability, and a sufficiently
long vertical tail arm, which increases yaw damping. Increasing the vertical tail area then
magnifies the degree of stability or instability.
Dutch roll : The second lateral motion is an oscillatory combined roll and yaw motion called
Dutch roll, perhaps because of its similarity to an ice-skating motion of the same name made
by Dutch skaters; the origin of the name is unclear. The Dutch roll may be described as a
yaw and roll to the right, followed by a recovery towards the equilibrium condition, then an
overshooting of this condition and a yaw and roll to the left, then back past the equilibrium
attitude, and so on. The period is usually on the order of 315 seconds, but it can vary from a
few seconds for light aircraft to a minute or more for airliners. Damping is increased by large
directional stability and small dihedral and decreased by small directional stability and large
dihedral. Although usually stable in a normal aircraft, the motion may be so slightly damped
that the effect is very unpleasant and undesirable. In swept-back wing aircraft, the Dutch
roll is solved by installing a yaw damper, in effect a special-purpose automatic pilot that
damps out any yawing oscillation by applying rudder corrections. Some swept-wing aircraft
have an unstable Dutch roll. If the Dutch roll is very lightly damped or unstable, the yaw
damper becomes a safety requirement, rather than a pilot and passenger convenience. Dual
yaw dampers are required and a failed yaw damper is cause for limiting flight to low altitudes,
and possibly lower mach numbers, where the Dutch roll stability is improved.
The state space representation for the lateral dynamics is given by ẋ = Ax + Bu,
where, xT = (β, p, r, ϕ), u = (δa , δr ) and
 
Yβ Yp
( YVr − 1) g/V   
 V V Ya Yr
   

0 
V V
 Lβ Lp Lr   
A= B = 
 La Lr

 (1.24)
   
 N Np Nr 0   
 β 
  Na Nr
0 1 0 0
The effect of few of the most important derivatives in the aircraft dynamics is explained below:
Lp : Change in the rolling moment with the roll rate. This roll damping derivative is required
to be negative. The derivative has the influence on roll mode time constant.
Lr : Change in the rolling moment with the yaw rate. This derivative has a considerable
effect on the spiral stability. For the spiral stability, Lr should be positive and it should be
preferably as small as possible.
Lβ : Change in the rolling moment with the sideslip angle, the derivative is called ‘effective
dihedral’. The derivative has the influence on damping of Dutch roll and spiral modes. A
negative Lβ is desired.
Np : Change in the yawing moment with the roll rate. For rolling about the stability axis,
Np represented in the stability axis should be minimal.
Nr : Change in the yawing moment with the yaw rate. This yaw damping derivative, which
is the main contributor to the Dutch roll damping, is required to be negative. The derivative
also contributes to the stability of the spiral mode.
Nβ : Change in the yawing moment with the sideslip angle. This static directional stability
derivative primarily establishes the natural frequency of the Dutch roll mode and also the
spiral stability. A positive Nβ is desirable.
Yβ : Change in the side force with respect to the sideslip angle. The vertical fin of the
aircraft helps in opposing the side force (sideslip) developed due to the sideslip angle. Hence,
a negative Yβ is desired to resist a sideslip.
Lδa : Change in the rolling moment with the elevon command. The derivative gives
an indication of the elevon effectiveness. Generally, the elevon effectiveness degrades with
reduction in the aircraft speed. At low speed flight conditions, it is important to have adequate
effectiveness of the elevon to counter the asymmetric gusts which tend to roll the aircraft. A
large value of this derivative is preferred for the better handling qualities.
Nδa : Change in the yawing moment with the elevon command. A positive Nδa results in a
proverse yaw whereas a negative value results in an adverse yaw.
Nδr : Change in the yawing moment with the rudder command. The derivative gives an
indication of rudder effectiveness. Rudder effectiveness plays an important role in a cross
wind landing, coordinated turning and rolling around the stability axis. A large value of this
derivative is preferred for better handling qualities.
For a typical aircraft (L1011) ([1], p.185):
   
 −0.746 0.006 −0.999 0.0369   0.0012 0.0092 
   
  
 −12.9 −0.746 0.387 0   6.05 0.952 

A=

B = 
 

 (1.25)
 4.31 0.024 −0.174   
 0   −0.416 −1.76 
   
0 1 0 0 0 0
Eigenvalues for the lateral motion of the aircraft typically consists of
– Lowly damped, complex pole pair – Dutch roll;
– real pole , relatively far from origin – roll subsidence;
– real pole, close to the origin – spiral mode.
For the above example, Dutch roll mode = -0.4473 ± 2.0724j (ζ = 0.211, ω = 2.12rad/sec.)
Roll subsidence mode = - 0.7653
spiral mode = - 0.0062
On examination of A and B matrices, it can be seen that the aileron input (6.05
Vs 0.952) dominantly influences roll rate p, and rudder input (-1.76 Vs -0.416) dominantly
influences yaw rate r. Hence, rudder excites primarily the Dutch roll mode. From the observa-
tion or sensor output point of view, Dutch roll mode is dominantly observed in the yaw rate.
Therefore, for improving the Dutch roll damping, detect/ measure yaw rate and feed it to
the rudder. Similarly, side slip response (most of the time, causes uncomfortable ride quality)
from aileron deflection is caused by the yawing moment, which has to be counteracted.
Diagonal or modal representation of Lateral dynamics for the above given A, B, &
C = I4 , D = 0 is given by
   
 −0.4473 2.0724 0 0   1.544 6.004 
   
   
 −2.0724 −0.4473 0 0   −0.254 −1.3754 
Ad = 

;
 Bd = 



 0 0 −0.7653 0   −7.84 6.97 
   
   
0 1 0 −0.0062 6.259 −5.570
 
 −0.0280 −0.1354 −0.0001 0.0014 
 

 0.8584 0 −0.6075 −0.0062 

Cd = 

;
 Dd = zeros(4, 2).
 −0.2705 0.0839 0.0257 0.0358 
 
 
−0.0854 −0.3957 0.7939 0.9993
Modal representation (Ad , Bd , Cd , Dd ) indicate that rudder has greater influence of Dutch
roll mode than aileron. Aileron and rudder have good influence on roll subsidence and spiral
modes. Side slip angle β is influenced by dutch roll mode, while yaw rate r can influence
Dutch roll damping significantly. Roll subsidence and spiral modes have strong influence on
bank angle output ϕ. Roll subsidence also has strong component in roll rate p output.
The lateral dynamics can also represented by Dutch roll, Spiral and Roll subsidence
approximation. The approximate expression for Dutch roll mode frequency and damping are
Nr Yβ (Nr + yβ /V )
2
ωnd = Nβ + , ζd = −
V 2ωnd
and time constants for spiral and roll subsidence are
Lβ (Nβ − g/V ) − Lp Nβ Lβ Np Lβ g
τs = τr = −Lp (1 − )−( ).
(Lβ Nr − Nβ Lr )g/V Nβ Lp Nβ V
The measurement or output vector is y = [p, r, β, fycg , ϕ]T , where fycg is the lateral acceleration
at the CG of the aircraft. On an aircraft, a subset of above mentioned sensors are normally
available, say, only p, r β are used for feedback. Then, fycg and ϕ are incorporated in the C
matrix to enable monitoring of the handling quality assessment.
For the design and analysis of the controller, a linearized model of say, a generic
fighter aircraft at 0.51 Mach number, 1.38 km altitude corresponding to a 6 g flight condition
is considered. This flight condition is characterized by a high spiral instability, a poor Dutch
roll damping and high roll mode time constant. The trim angle of attack is about 14.4 deg.,
hence the model is suitable to demonstrate the advantages of ‘Rolling around stability axis’.
The corresponding state space matrices are as follows:
   
 −2.85 2.52 −76.11 0.00   −103.6 9.44 
   

 −0.31 −0.46 7.36 0.00 
 
 −6.21 −5.08 
A = 

B = 
 

 (1.26)
 0.25 −0.96 −0.43 0.06   0.09 0.06 
   
   
1.00 0.26 0.00 0.00 0.00 0.00
   
 1.00 0.00 0.00 0.00   0.00 0.00 
   
   
 0.00 1.00 0.00 0.00   0.00 0.00 
   
   
C 
=  0.00 0.00 1.00 0.00  
D =  0.00 0.00  (1.27)
 
   
   
 0.15 −0.03 −37.43 0.00   5.08 −0.12 
   
   
0.00 0.00 0.00 1.00 0.00 0.00
From the state space model, it is seen that the number of state variables, n= 4; the number
of independent input variables, m = 2. Since, only the first three parameters of the matrix C
are available for the feedback, number of independent output variable, r=3. The system has
a rank of 4, the matrices B and C have the ranks of 2 and 3 respectively. Hence, the aircraft
is controllable and observable.
The lateral directional motion of a fighter aircraft is also characterized by the
following three basic modes.
• A highly convergent high frequency mode, called roll mode. The roll mode characterizes
the rate at which the roll rate builds up for elevon command.
• A slowly convergent or divergent low frequency mode called spiral mode . The spiral
mode is characterized by the bank angle ϕ, heading angle ψ and a small sideslip angle β.
The stable spiral mode has the tendency of maintain the wings level on a disturbance.
The unstable spiral mode has the tendency of ‘roll off’, which keeps increasing the bank
angle on a disturbance.
• A pair of lightly damped, low frequency oscillatory mode called Dutch roll mode. The
Dutch roll is a complex oscillatory motion involving rolling motion, yawing motion and
sideslipping. The yawing moment and the sideslip are the major components in the
Dutch roll. The Dutch roll mode characterizes the directional stability of an aircraft.
The aircraft response to a disturbance or a command input consists of all the three
modes. The degree to which a mode responds depends on, whether the elevon or rudder
surface are deflected, the cross coupling between the modes, etc. The eigenvalues and the
corresponding eigenvectors of the open-loop aircraft, considered in the design, are listed in
the following tables.
Open -loop eigenvalues:
Dutch roll mode = −0.8215 ± 4.7995i (ζ = 0.1687, ω = 4.8693)
Roll mode = −2.0095 (τ0 = 0.4976sec.)
Spiral mode = +0.0104 (T imetodouble = 66.63sec.)
Open-loop eigenvectors:
From the above analysis, it can be observed that the Dutch roll mode is lightly damped
Modes/ States Dutch roll mode Roll mode Spiral mode

p -0.9622± 0.0899i -0.8630 -0.0247
r 0.0787 ± 0.1270i -0.1902 0.0552
β 0.0385 ± 0.0613i 0.0037 0.0031
ϕ 0.0447 ± 0.1885i 0.4507 0.9982
and appears dominantly in the roll response as evidenced by the dominant components in
the corresponding eigenvectors. The roll mode mainly appears in the roll rate and roll angle
response. The unstable spiral mode appears almost entirely in the roll angle response. As
most of the aircraft spirally unstable, the aircraft do not have tendency to return to initial
heading and roll angle after it has been initially disturbed from equilibrium. Pilot can make
continually corrections, if the time to double is large.
Algebraic derivation of Lateral - directional transfer functions is given in [35, pp. 270
- 276]. This derivation is simple and can be readily understood by self study and hence not
reproduced here.
1.4 Dynamics of Rockets and Missiles

Majority of rockets and missiles have two planes of symmetry. Their dynamics are character-
ized by large disturbance inputs, rapid variations in mass, inertia and aerodynamic parameters,
aerodynamic instability of air frame, influence of sloshing, and structural vibration resulting
in control structure interaction. For control system study, linear time invariant models are
used. For roll stabilized rockets, roll, pitch and yaw dynamics are decoupled. Note: a very
few missiles are roll rate stabilized. These rockets are controlled by either thrust vectoring or
aerodynamic control surfaces or both. Two important class of thrust vectoring are obtained
from SITVC (side injection thrust vector control) and gimbal system. Flexible nozzles are
used in some cases to produce a side force. The aerodynamic control surface (cruciform con-
figuration) can be configured as a tail, wing or canard, or combination of these. The notation
traditionally used in the rockets and missiles are as follows:
Roll axis pitch axis Yaw axis

xb yb zb
Attitude angles ϕ θ ψ
Angular rates p q r
Velocity U+u v w
Moment components L M N
Force components X Y Z
MI Ix Iy Iz
Control surface deflection ζ η or δ ξ
The accelerations in pitch and yaw plane (i.e., perpendicular to the longitudinal/
body axis) are known as lateral accelerations (in the missile parlance). The rockets and missile
use either lateral acceleration or attitude command following autopilot. Accordingly, state
variable ϕ, θ, ψ or lateral acceleration fy and fz are used in the model. Pitch Dynamics:
mfz = m(ẇ − qU ) = zw w + zq q + zη η
or fz = ẇ − qU = zw w + zq q + zη η (1.28)
Iy q̇ = mw w + mq q + mη η
1.4. DYNAMICS OF ROCKETS AND MISSILES 37
Figure 1.3: Missile Coordinate System [4].
or q̇ = mw w + mq q + mη η (1.29)
Yaw dynamics :
mfy = m(v̇ + rU ) = Yv v + Yr r + Yξ ξ
or fy = v̇ + rU = yv v + yr r + yξ ξ (1.30)
Iz ṙ = Nv v + Nr r + Nξ ξ
or ṙ = nv v + nr r + nξ ξ (1.31)
Roll dynamics :
Ix ṗ = Lp p + Lξ ξ
or ṗ = lp p + rξ ξ (1.32)
On using the notations similar to the aircraft dynamics, the pitch plane dynamic equations
(δ =control input, use total velocity V instead of U) are rewritten as:
Zα Zδ
α̇ = α+q+ δ (1.33)
V V
q̇ = Mα α + Mq q + Mδ δ; q = θ̇ (1.34)
As seen earlier, if Mα > 0, ( i.e., CP is ahead of CG), then the missile is aerodynamically
unstable. The normal acceleration (latax) -in yaw plane fy ≈ −U γ̇, γ = θ − α, implies that
fy ≈ Zα α + Zδ δ (1.35)
The latax feedback system minimizes the structural load acting on the system. For simplified
analysis, the output matrix C = [Zα 0] and D = Zδ and the transfer function from δ to fy is
given by G(s) = C(sI − A)−1 B + D)
  
 s−
Zα
V
1   Zδ /V 
= [Zα 0]    + Zδ
−Mα s − Mq Mδ
Zδ (s2 − Mq s − Mα ) + Zα Mδ
= (1.36)
s2 − (Mq + ZVα )s + ( ZVα Mq − Mα )
For a typical stable missile, Zα , Mα and Mδ are all negative. In the above Transfer function,
characteristic polynomial for stable missile can be written as s2 +2 ζ ω s+ω 2 , where, ω is called
as weathercock frequency and ζ is called weathercock damping . For an unstable missile,
the term ( ZVα Mq − Mα ) is negative, hence implying one stable pole and one unstable pole (as
in the case of inverted pendulum/ unstable short period mode of an aircrft). Weathercock
frequency is similar to the short period mode in longitudinal dynamics of an aircraft. In
missiles, phugoid mode is not significant due to fast changing operating point. However, for
strategic missiles and launch vehicles, aerodynamic delay, which is called drift pole can be
significant. This means that second order effect of phugoid like behavior is replaced by delay
state (integral of pitch angular velocity, by neglecting only the change in forward speed ∆u).
In the Eq. (1.36), the coefficient of s2 in the numerator of G(s) is negative. The constant term
(Zα Mα − Mα Zδ ) is typically positive for tail controlled missile (i.e., control surface behind
CG) and negative for canard controlled missile (control surface ahead of CG). Hence, for the
tail controlled missile (also TVC missiles), the G(s) has one zero on the right half of s - plane
(i.e., non-minimum phase zero). Therefore, the D.C. gain and high frequency gain reverse
their signs. (i.e., presence of overshoot and undershoot). The initial value of step response is
lim
s→∞
[ G(s)
s
] = Zδ < 0 (typical) and final value of the step response is
lim G(s) Z α M δ − M α Zδ
[ ]= > 0 (typically).
s→0 s −Mα
Therefore, latex control system brings in the additional complexity of non-minimum phase
zero to the control system design (for tail controlled missile).
1.4. DYNAMICS OF ROCKETS AND MISSILES 39
For a typical missile, V = 1253 ft/sec (Mach ∼ 1.2),
Zα = 4170f t/sec2 , Zδ = −1115f t/sec2 , (1.37)
Mα = −248rad/sec2 Mδ = −662rad/sec, Mq ≈ 0. (1.38)
Typical actuator dynamics is δ̇ = 1

τa
(u − δ) where, u is the input to the actuator. The time
constant for actuator τa = 0.01 sec. The combined transfer function for missiles with actuator:
−1115K(s2 − 2228)
G(s) = − (1.39)
(0.01s + 1)(s2 + 3.333 + 248)
poles are at −1.67±15.65j, −100 and zeros are at ±47.2. Typically, for tail controlled missiles,
one gets a non-minimum phase zero in the lateral/ normal acceleration frequency response.
Note that dc gain of G(s) is positive: A positive input produces positive response. But high
frequency gain − Zδ /Mα is (typically) negative and hence for high frequencies (initial time
(transient part of) response can be found from initial value theorem) G(s) → −1115/(0.01 s+
1) which produces a negative response (under shoot) for a positive input. The change of sign
in the TF as the frequency is increased is a consequence of the right half zero. Discussion on
flight control system (including analysis of Autopilot) will be dealt with in Chapter 3.
Launch Vehicles
For launch vehicles and long range strategic missiles, normally, attitude control system is
used. The corresponding state variables in pitch dynamics are (α, q, θ). In modeling the
launch vehicle dynamics, slightly different symbols are used [5]:
Lα Xb
lα
lc CP θ
Tc CG . XF
z
δ v
Tr
V
Z
b
Force and moments equations:

−gcosθ0 Tc δ Lα α
α̇ = θ+ − +q (1.40)
V mV mV
Lα lα α Tc lc
q̇ = + δ = µα α + µc δ (1.41)
Iy Iy
On comparing with the earlier equations (1.12 & 1.29 with 1.41) for missile & aircraft equa-
tions (longitudinal), the equivalence between different symbols can be readily established. To
include drift pole effect, kinematic relation θ̇ = q is added. On taking the Laplace transform,
Lα
θ(s) µc [s + mU (1 + llαc )] µc
= 3 Lα 2 µα gcosθ ≈
δ(s) s + mV s − µα s + V (s − µα )
2
As stated earlier, if CP is a head of CG, lα > 0 and hence µα (similar to Mα , Mα > 0) and
the vehicle is unstable. For a typical vehicle data given in Greensite’s book [5],
θ(s) µc (s + 0.0678) 4.54
= ≈ 2
δ(s) (s − 0.0042) (s + 0.0354s − 3.75)
2 (s − 3.75)
Chapter 3 deals with analysis of control system for rigid and flexible vehicles. Modeling of
the flexible rockets along with effects of propellant sloshing and engine inertia effects are also
presented along with control issues and control - structure interaction.
1.4.1 Summary
To summarize, the governing equation for an aerospace vehicle can be written in a standard
state space representation as:
ẋ = A x + B u, where, x ϵ ℜn , u ϵ ℜm (1.42)
and the corresponding measurement equations are
y = C x, where, y ϵ ℜ r , (m, r) ≤ n. (1.43)
The states x include both the state variables for airframe dynamics and actuator. Before
dwelling on the analysis of control systems for aerospace Applications, it is necessary to
investigate the intrinsic properties such as stability, controllability and observability. At the
same time stability and performance robustness also play important role in the flight control
system design. These aspects are discussed in the next chapter for a general linear time
invariant system represented as in the equations (1.42) and (1.43).
Chapter 2
System Analysis
2.1 Introduction
2.1.1 Navigation, Guidance & Control
Navigation : Determines position and velocity and also other

relevant state vectors
Guidance : Control trajectory/path of the vehicle ;
- Open loop Vs closed loop systems
Control/ Autopilot : Closed loop control of attitude motion.
→ Analysis and Design
• Modelling
• Control Structure
• Analysis & Design by

- classical - Frequency domain -T.F.; or
- Modern - Time domain - state space description.
Advantages of Feedback Control (negative feedback):
- improvement in stability
- reduction in sensitivity (robustness) against parameter changes
41
42 CHAPTER 2. SYSTEM ANALYSIS
- reduction in sensitivity against noise/disturbance,
- improvement in system performance (speed of response, % overshoot, steady state error).
Control Problem : Find a policy for automatically choosing control input u(t) so
that y(t), x(t) & u(t) meet certain requirements
Feedback control : Determine u(t) as a function of (closed loop ) x(t) or y(t) and
the reference input r(t)
Requirements : stability and performance
Intrinsic properties : Check stability, controllability, observability, Disturbability etc.,
Methods of Analysis/Design in Frequency and Time Domain
Time domain — Frequency domain

state space model “t” — T.F with “s”
ODE/PDE — Algebraic expressions
Linear, Nonlinear, Time varying/ invari- — LTIV systems
ant systems
Analytical/Numerical methods — Largely graphical + empirical methods
Figure 2.1: Design Steps [45]

2.2. METHODS OF LINEAR SYSTEMS ANALYSIS 43
disturbance
(determinstic/stochastic)
x(t) output
Input
state
Actuator System Sensors
(Control u(t)+ variable y(t)
disturbances) LPS,DPS,Hybrid
noise
Figure 2.2: Design Plant
During the past five decades, lots of analysis are carried out on linear time invari-
ant (LTIV) systems. The LTIV systems have the advantage that they satisfy ubiquitous
superposition principle (with respect to input and initial conditions). First, the methods of
analysing linear systems and study of intrinsic properties of a LTIV system are presented.
2.2 Methods of Linear Systems Analysis

Linear systems can be be analyzed in time domain (Deal with differential equations directly) or
frequency domain (Deal with algebraic equations in terms of complex Laplace variable s).
Laplace transform of a time function f(t)
∫ ∞ ∫ ∞
−st
F (s) = e f (t)dt = e−st f (t)dt
0+ 0
Example
1) f (t) = e−at
∫ ∞ ∫ ∞
e−at e−st dt = e−(s+a)t dt
0 0
1
=
s+a
2) f (t) = sin ωt
∫ ∫
∞
−st ejωt − e−jωt
∞
e sin ωt dt = e−et ()dt
0 0 2j
1 1 1 jω − (−jω) ω
= ( − )= = 2
2j s − jω s + jω 2 2
2j(s + ω ) s + ω2
Show that
ω
3) f (t) = e−at sin ωt ⇒ F (s) =
(s + a)2 + ω 2
dx
4) f (t) = , if L(x(t)) = X(s); then
dt
∫ ∞ ∞ ∫
dx −st
e dt = e−st x|∞
0 − x(t)(−s)e−st dt
0 dt ∫ 0
∞
= −x(0) + s x(t)e−st dt = sX(s) − x(0)
0
5) Take laplace transform of a differential equation - to get input/output relation

Scalar differential equation
(i) ẋ = ax + bu, y = x
On taking laplace transforms
sX(s) − x(0) = aX(s) + bU (s)
(s − a)X(s) = bU (s) + x(0)
Y (s) X(s) b
Transfer function U (s)
= U (s)
= (s−a)
with x (0) = 0 relates input to output through a
multiplicative operation.
(ii) mẍ + cẋ + kx = f (t) = u(t) y = x(t)
ms2 X(s) + csX(s) + kX(s) = U (s)

y(s) X(s) 1
T ransf er f unction G(s) = = =
U (s) U (s) ms2 + cs + k
6) Let f(t) = arbitrary function of time = a0 + a1 t + a2 t2 + a3 t3 + .......
For most of the functions, it is sufficient to take finite number of right hand side terms to
represent f (t) within reasonable limit of accuracy
f (t) = a0 u(t) + a1 r(t) + a2 p(t) + ....
= step ↑ ramp ↑ parabolic ↑
F (s) = a0 U (s) + a1 R(s) + a2 P (s) + . . .
7) Unit step function

∫ ∞
U (s) = u(t)e−st dt
0
∫ ∞ 1 1
= e−st dt = − e−st |∞
0 = .
0 s s
8)Ramp function
∫ ∞
R(s) = te−st dt
0
∫
1 −st 1
= e dt = 2
s s
Transfer function for a first order system
ẋ = −ax + bu, u = unit step (say)
sX(s) = −aX(s) + bU (s)

X(s) 1 1 1
= ⇒ X(s) =
U (s) s+a ss+a
1 1
x(t) = L−1 ( )
ss+a
+1/a −1/a
= L−1 ( + )
s s+a
1 1 1
= + − e−at = (1 − e−at )
a a a
Final value Theorem :
1 1
lim x(t) = x(∞) = lim s X(s) = lim =
t→∞ s→0 s→0 (s + a) a
Final value theorem can be applied only to stable systems.

1
Response of a LTIV System :
ẋ = Ax + Bu : state equation; x(0) = x0 (2.1)
y = Cx + Du : M easurement equation. (2.2)
Solution:
∫ t
x(t) = e−At x0 + e−A(t−τ ) Bu(τ )dτ
0
= T ransient : + F orced : response
T ransientResponse : ẋ(t) = −ax(t) + bu(t); x(0) = x0 , u = 0 ⇒ x(t) = x0 e−at
F orcedResponse : x(0) = 0, u(t) = 1, x(t) = (b/a)(1 − e−at )
x(t) is finite if a > 0 (stable), OR, x(t) → ∞ if a < 0 (unstable).

1
W.L.Brogan: Modern Control Theory, Chs 1, 2, 3, 4, 7, 9, 10, 14; J.J. D’Azzo & C.H. Houpis: Linear
control system Analysis and Design, 4th edition, Chs 6, 7, 8, 10 & 11; N. N. Nise: Control Systems Engineering
In general ẋ = Ax + Bu : state equation; x(t0 ) = x0 and corresponding solution

is given by
∫ t
x(t) = e−A (t−t0 ) x0 + e−A (t−τ ) B u(τ ) dτ (2.3)
t0
Define State transition matrix as
Φ(t, t0 ) = e−A (t−t0 ) ; Similarly Φ(t, τ ) = e−A (t−τ ) (2.4)
so that x(t) = e−A (t−t0 ) x0 = Φ(t, t0 ) x0 ⇒ Transition of state x(τ ) from time t0 to time t.
Properties [46]:
1. Φ(τ, τ ) = In = e−A (τ −τ )
d Φ(t, τ )
2. dt
= A Φ(t, τ )
3. Φ(t2 , t0 ) = Φ(t2 , t1 ) ∗ Φ(t1 , t0 ) = e−A (t2 −t1 ) ∗ e−A (t1 −t0 )
4. Φ(t, 0) = L−1 [(s I − A)−1 ] ⇒ Replace (t − τ ) by t in Φ(t, τ )
The matrix (s I − A) is sometimes called as the characteristic matrix of A. det(s I − A)
is called characteristic polynomial of A and its roots are called the eigenvalues or
characteristic values of A. The eigenvalues may be real or complex conjugate pair -
distinct or repeated. The eigenvalues describe the natural or free response of the system.
Since A is a real valued matrix, complex eigenvalues appear as complex conjugate pairs
(otherwise coefficients of characteristic polynomials will not remain real). The eigenvalues
describe the natural or free or un-forced response. Hence, they are called characteristic or
natural frequencies or modes of the realization. Similar realizations have same natural
frequencies, since,
∆(s) = det(s I − A) = det( T (s I − A) T −1 ) = det(s I − T A T −1 ).
An important property of the characteristic polynomial ∆(s) is that
∆(A) = An + a1 An−1 + . . . + an I
This property is due to Cayley - Hamilton theorem. Another important resolvent formula is
given by (to verify, multiply both sides by (s I − A) ):
Adj(s I − A) = An−1 + (s + a1 ) An−2 + . . . + (sn−1 + a1 sn−2 + . . . + an−1 ) I
N (s)
For a SISO system, system transfer function G(s) = D(s)
, where,
N (s) = c Adj(s I − A) b = sn−1 (cb) + sn−2 (cAb + a1 cb) + . . . + (cAn−1 b + . . . + an−1 cb).
If common factors are canceled between the characteristic polynomial D(s) (or ∆(s)) and the
every element of the numerator polynomial matrix N (s), then, (s I − A)−1 = Γ(s)
µ(s)
, where, µ(s)
is called the minimal polynomial of A.
The homogeneous set of n state equations
ẋ = A x, x(to ) given, A = constant (2.5)
has a solution which is completely analogous to the scalar result:
x(t) = eA(t−to ) x(to ) (2.6)
There are several methods of verifying that this is a solution to the state equations. First,
note that the initial conditions are satisfied. That is,
x(to ) = eA(to −to ) x(to) = e[0] x(to ) = x(to )
On differentiating both sides of equation (2.6), it is easily verified that ẋ = Ax(t). Since
equation (2.6) satisfies the initial conditions and the differential equation, it represents a
unique solution. On taking Laplace transform of equation (2.5) with a non-zero initial con-
dition x(0) = x0 , one gets, s x(s) − x(0) = A x(s), or [s I − A] x(s) = x(0) so that x(s) =
[s I − A]−1 x(0). On taking the inverse Laplace transform gives x(t) = L−1 [s I − A]−1 x(0).
The non-homogeneous set of state equations is considered next. The system matrix
A is still a constant, but B(t) may be time-varying. Components of Bu(t) are assumed to be
piecewise continuous to guarantee a unique solution :
ẋ = Ax + B(t)u(t), x(to ) given (2.7)
The technique used in solving the scalar equation is repeated with only minor dimensional
modifications. Let Φ(t) be an n × n matrix. Pre-multiplying equation(2.7) by Φ(t) and
rearranging gives
Φ(t) ẋ(t) − Φ(t) A x(t) = Φ(t) B(t) u(t)
d[Φ(t)x(t)]
Since dt
= Φ(t)B(t)u(t) dt = Φẋ + Φ̇x, the left-hand side can be written as an exact
(vector) differential provided Φ̇ = − Φ(t) A. One such matrix is Φ(t) = e−A (t−to ) . Agreeing
that this is the Φ matrix to be used, the differential equation can be written as
d[Φ(t)x(t)] = Φ(t)B(t)u(t) dt
Integration gives
∫ t
Φ(t) x(t) − Φ(to ) x(to ) = Φ(τ ) B(τ ) u(τ ) dτ
to
The selected form for Φ always has an inverse, so

∫ t
−1
x(t) = Φ (t) Φ(to ) x(to ) + Φ−1 (t) Φ(τ ) B(τ ) u(τ ) dτ
to
Or
∫
A(t−to )
x(t) = e x(to ) + eA (t−τ ) B(τ ) u(τ ) dτ (2.8)
This represents the solution for any system equation in the form of equation (2.7). Note
that it is composed of a term depending only on the initial state and a convolution integral
involving the input but not the initial state. These two terms are known by various names
such as the homogeneous solution and the particular integral, the force-free response and the
forced response, the zero input response and the zero state response, etc.
Equation (2.7) is considered again. It is emphasized that the matrix A is constant,
but B(t) may be time-varying. Assume that the n eigenvalues λi and n independent vectors,
either eigenvectors or generalized eigenvectors, have been found for the matrix A. These
vectors are denoted by ξi to avoid confusion with the state vector x. Since the set {ξi } is
linearly independent, it is used as a basis for the space Σ. Thus at any given time t, the state
x(t) can be expressed as
x(t) = q1 (t) ξ1 + q2 (t) ξ2 + . . . + qn (t) ξn (2.9)
The time variation of x is contained in the expansion coefficient qi since A, and hence the
ξi , are constant. At any given time t, the vector B(t) u(t) ϵ Σ and therefore it, too, can be
expanded as
B(t)u(t) = β1 (t) ξ + β2 (t) ξ2 + . . . + βn (t) ξn
In fact, βi (t) = < ri , B(t)u(t) >, where ri is the set of reciprocal basis vectors, < ., . > is the
inner product. Using the above expansion, equation (2.7) becomes
q̇1 ξ1 + q̇2 ξ2 + . . . + q̇n ξn = q1 Aξ1 + q2 Aξ2 + + qn Aξn + β1 ξ1 + β2 ξ2 + . . . + βn ξn

Assume for the moment that A is normal (or non-defective, i.e., matrix A has n - independent
set of eigenvectors, irrespective of whether its eigenvalues are distinct or repeated; real or
complex conjugate) so that all ξi are eigenvectors rather than generalized eigenvectors.
Then A ξi = λi ξi , so that
(q̇1 − λ1 q1 − β1 )ξ1 + (q̇2 − λ2 q2 − β2 )ξ2 + . . . + (q̇n − λn qn − βn )ξn = 0 (2.10)
Since the set ξi is linearly independent, this requires that
q̇i = λi qi + βi f or i = 1, 2, . . . , n (2.11)
This demonstrates that when A is constant and has a full set of eigenvectors, the system is
completely described by a set of n uncoupled scalar equations whose solutions are of the form
∫ t
qi (t) = eλi (t−to) qi (to ) + eλi (t−τ ) βi (τ ) dτ (2.12)
to
Of course, qi (to ) = < ri , x(to ) >. The state vector is given by
x(t) = q1 (t)ξ1 + q2 (t)ξ2 + · · · + qn (t)ξn (2.13)
The terms in this sum are called the system modes. The general response of a complicated
system can be broken down into the sum of n simple modal responses. It should be recognized
that equation (2.9) can be written in terms of the concatenated modal matrix M = [ξ1 . . . ξn ]
as x = M q, and as such, represent a change of basis. Using this notation, equation (2.7) is
considered again: ẋ becomes M q̇, since M is constant; Ax becomes AM q; and Bu remains
unchanged (Note that canonical transform changes only the internal descriptions and not the
input and output variables, i.e., input/output descriptions remain invariant). Therefore,
M q̇ = A M q + B u
or q̇ = (M −1 A M ) q + (M −1 B) u = J q + Bn u (2.14)
where Bn = M −1 B and the assumption regarding a full set of eigenvectors is dropped. J is

the Jordan canonical form (or diagonal form in many cases). If the same change of basis is
used in the output equation, then the system is described by the pair of normal form equations.
q̇ = J q + Bn u (2.15)
y = Cn q + D u (2.16)
where Cn = C M . One advantage of the normal form is that the state equations are as nearly
uncoupled as possible. Each component of q is coupled to at most one other component
because of the nature of the Jordan form matrix J. The solution for equation (2.15) can be
written as
∫ t
q(t) = eJ(t−to ) q(to ) + eJ(t−τ ) Bn (τ ) u(τ ) dτ
to
Relating this to the original state vector gives

∫ t
x(t) = M q(t) = M eJ(t−to ) M −1 x0 + M eJ(t−τ ) M −1 B(τ ) u(τ ) dτ (2.17)
to
The above equation uses the fact that M −1 is the matrix of transposed reciprocal basis vectors,
which means that q(t0 ) = M −1 x(to ) Comparing equations (2.16) and (2.8) shows that
eA(t−t0 ) = M eJ(t−t0 ) M −1
Modal decomposition is useful because of the insight it gives regarding the intrinsic
properties of the system. The properties of controllability and observability are more eas-
ily understood and evaluated. The stability properties of the system are also more clearly
revealed. Modal decomposition provides a simple geometrical picture for the motion of the
state vector verses time. By retaining only the dominant modes, a high-order system can be
approximated by a lower-order system.
It should be kept in mind that the modal decomposition techniques is only useful
if A, and thus ξi , λi are constant. It is the invariance of the vector parts of x(t), that is the
ξi terms, that gives value to the method. If the modal matrix were time-varying and had to
be continually reevaluated, the advantages of modal decomposition would be lost.
Example: [46]
A system is described by
      
 ẋ1 
  
 
 
 1 

 0 1  x1
=   + u

 ẋ     
2 8 −2  x2   1 
The initial

conditions

are x(0) = [1 −4]T . Assume that u(t) = 0 and analyze this system.
 With


 1 

0 1 
A=
 , the eigenvalues are λ1 = −4, λ2 = 2. The eigenvectors are ξ1 =

 −4 

8 −2
     

 1 

 1 1 
1 
2 −1 
ξ2 = . The modal matrix and its inverse are M =   ; M −1 = 6  .

 2 
 −4 2 4 1
Any one of several methods gives (Note: fundamental solution/ modes are e−4t and e2t ) :
 
1 2e−4t + 4e2t −e−4t + e2t 
eAt =  
6 −8e−4t + 8e2t 4e−4t + 2e2t
so the homogeneous solution is x(t) = eAt x(0) = [e−4t | − 4e−4t ]T , and the output is
y(t) = C x(t) = [4 1] x(t) = (4e−4t − 4e−4t ) = 0 for all t (∀ t).
Example:
Modal decomposition is now applied in an attempt to gain insight into the unusual
result of above Example. Since the eigenvalues are distinct, M −1 A M = Λ for this system,
and equation (2.15) becomes
      

 q̇1 
 
 
 
 1/6 

 2 0  q1
=  + u

 q̇     
2 0 −4  q2   5/6 
The initial conditions are q(0) = M −1 x(0) = [1 0]T . Equation (2.16) becomes y = [0 6]q. The
state vector x(t) can be written as the sum of two modes,
   

 1 
 
 1 

−4t 2t
x(t) = q1 (0) e + q2 (0)e .

 −4 
 
 2 

The particular initial condition selected here has no component along the direction of mode
2, since, q2 (0) = 0. Thus, the second mode is not excited. The output of this system consists
only of the second mode contribution, as seen by Cn = CM = [0 6]. Mode 1 contributes
nothing to the output and mode 2 is not excited, so the output remains identically zero.
Effect of eigenvalues and eigenvectors on transient response [10, §16.2]
Let us consider a Linear Time Invariant system defined by the equations :
ẋ(t) = A x(t) + B u(t)

(2.18)
y(t) = C x(t)
where, x ∈ Rn , y ∈ Rr , u ∈ Rm , rank B = m, rank C = r.
This system consists of n eigenvalues of A and a corresponding set of n-dimensional eigenvec-
tors. Each set of eigenvalue and eigenvector satisfy the equalities (Note: symbol V ⇔ M ):
A Vi = λi Vi ; i = 1, 2, . . . n ; (2.19)
where λi is the ith eigenvalue, and Vi is the corresponding right eigenvector. The free/ transient
response of the system to non-zero initial condition x0 is given by the equation
x(t) = eAt x0 (2.20)
Assuming eigenvalues of A to be distinct, a nonsingular modal matrix T consisting of eigen-

vectors (Note that Symbols: T, M, V, or P are interchangeably used to represent matrix of
eigenvectors) can be found, where
T = [V1 , V2 , V3 , . . . , Vn ] (2.21)
and note that T represents similarity transformation: x = T q, hence,
A = T Λ T −1 or Λ = T −1 A T (2.22)
where, Λ is the diagonal matrix of eigenvalues, if all the eigenvalues are distinct. For repeated
eigenvalues of the non-defective matrix A, Jordan block replaces the diagonal sub-matrix, size
of which is equal to the multiplicity of a particular eigenvalue. For example, for a matrix with
threeeigenvalues, if λ1 = λ2 ̸= λ3 , then the corresponding Jordan blockis 
 λ1 1 0   λ1 0 0 
   
   
Λ= 0 λ1 0 , compare this with the distinct λ′i s, where, Λ =  0 λ2 0  .
   
   
0 0 λ3 0 0 λ3
Top left (2 × 2) part is called a Jordan block. The homogeneous solution (transient response)
to non-zero initial conditions is sum of fundamental solutions or modal responses, i.e.,
e(λi t) , i = 1, . . . , n. For distinct modes λ1 , λ2 , transient response is linear combination of two
modes, i.e., x(t) = a1 e(λ1 t) + a2 e(λ2 t) . If λ1 = λ2 , then x(t) = a1 e(λ1 t) + a2 t e(λ1 t) . The
super-diagonal term 1, introduces the time variable t into the response. This interpretation
can be generalized the system in Eqn. 2.18. Now, the response (2.20) can be rewritten as
−1 t
x(t) = eT ΛT x0 (2.23)
which is equivalent to
x(t) = T eΛt T −1 x0 (2.24)
Define :
T −1 = [W1T , W2T , ... WnT ]T ; & WiT A = λi WiT because WiT (λi I − A) = 0
∑n (2.25)
θk = j=1 Wkj x0j
Note that rows of T −1 are rows of WiT , which are called reciprocal of left eigenvectors. Vectors
Vi & WiT are orthogonal. Substituting the above expressions in the Eqn 2.24 gives :
xi (t) = V1i θ1 eλ1 t + V2i θ2 eλ2 t + . . . + Vni θn eλn t (2.26)
From the above equation, it can be interpreted that :
1. State variable response consists of a combination of all the existing modes.
2. Each eigenvalue determine the growth/decay rate of the corresponding mode.
3. Amplitude of contribution from a particular mode depends on eigenvectors.
Hence by altering the individual mode amplitudes suitably, it is possible to have various
combination of modes, resulting in different shape of response. The totality of the eigenvalues
and eigenvectors is the eigen-structure of the system.
Example : For the LTIV system [10, pp. 534-535]: ẋ(t) = A x(t) + Bu(t), y(t) = C x(t),
   
 
 −3 0 0   1 0 
   
 1 0 0 
A=
 0 0

1  ; B= 
 0 0 , C =  
   
    0 1 1
0 −8 −6 0 1
The system matrix A has the eigen-spectrum σ(A) = {−2, −3, −4} and the eigenvectors are
such that they produce state responses for which x1 (t) contains only the mode e−3t . Also x2 (t)
and x3 (t) contain only the modes e−2t and e−4t ; thus producing modal decoupling in state
responses. Note that the eigenvector corresponding to eigenvalue λ1 = −2 is [0 1 − 2]T .
Since, the first element of this eigenvector is zero, the mode e−2t does not appear in x1 (t). The
state response of the closed-loop system computed for the initial condition
x(0) = [1 1 2]T is x(t) = £−1 [s I − A]−1 x(0) :

      
−3t −3t
 x1 (t)   e 0 0  1   e 
      
   −2t −4t −2t −4t    
 x2 (t)  =  0 2e −e 0.5e − 0.5e   1  =  3e−2t − 2e−4t 
      
      
x3 (t) 0 −4e−2t + 4e−4t −e−2t + 2e−4t 2 −6e−2t + 8e−4t
Note that the selection of the eigenvectors produces the result that only the mode e−3t appears
in x1 (t). Also, only the modes e−2t and e−4t appear in x2 (t) and x3 (t). The similar separation
of modes appears in the output response for this example obtained from y = C x.
   
−3t
 y1 (t)   e 
 =  (2.27)
−2t −4t
y2 (t) −3e + 6e
This example illustrates how particular modes can be included or excluded from particular
state responses. This is analyzed with reference to Eqn. (2.27) which relates the eigenvector
values to the state responses.
 
 v1 v2 v3 
 

 ↓ ↓ ↓ 
 
 
 a b c  ← x (t)
[ ]   1
 
 
V = v1 v2 v3 =  d e f  ← x2 (t) (2.28)
 
 
 
 g h i  ← x3 (t)
 
 
 ↑ ↑ ↑ 
 
 
eλ1 t eλ2 t eλ3 t
If the modes due to λ1 and λ3 are not to appear in x1 (t), then the eigenvector element values
must satisfy the requirement that a = c = 0 and b ̸= 0. In this example, the mode due to
λ2 must not appear in x2 (t), and x3 (t), which requires that e = h = 0, while d, f , g and i
are not equal to zero. This example shows that particular modes can be eliminated in xi (t),
while maintaining the necessary condition that the matrix V must have a full rank.
Time varying matrix A
The time-varying homogeneous state equation
ẋ = A(t)x, x(t0 ) = x0 . (2.29)
is considered first. In order that this qualify as a valid state equation, it is required that there
be a unique solution for every x(to ) ∈ Σ (x(t) and ξ(t) ∈ the space Σ). This places some
restriction on the kind of time variation allowed on the matrix A. A sufficient condition for the
existence of unique solutions is to require that all elements aij (t) of A(t) be continuous. Weaker
conditions may be found in textbooks on differential equations. Since dim(Σ) = n, n linearly
independent initial vectors xi (to ) can be found, and each one defines a unique homogeneous
solution of equation (2.29), called xi (t), t ≥ to . A particular set of n independent initial
condition vectors is selected as:
x1 (t0 ) = [1 0 0 . . . 0]T , x2 (t0 ) = [0 1 0 . . . 0]T , . . . , xn (to ) = [0 0 0 . . . 1]T

That is, the ith vector has one as its ith component and all other components are zero. The
n solutions corresponding to these initial conditions are used as the columns in forming an
n × n matrix U (t) = [x1 (t)x2 (t) . . . xn (t)]. The matrix U (t) is called the fundamental solution
matrix. Notice that U (to ) = In and U̇ (t) = A(t)U (t)
Assuming that the fundamental solution matrix is available, the solution to equation (2.29)
with an arbitrary initial condition vector x(to ) is
x(t) = U (t)x(to ) (2.30)
The nonhomogeneous time-varying state equation is solved in an analogous manner to the

scalar and constant matrix cases. That is, the equation is reduced to exact differentials so
that it can be integrated. Preliminary to this, it is noted that U −1 (t) can be shown to exist
for all t ≥ to and that
d
U (t) U −1 (t) = In so that
(U (t) U −1 (t) = [0] or
dt
dU −1 dU −1 dU −1 dU −1
U +U = [0] or = −U −1 U
dt dt dt dt
dU −1
Therefore, = −U −1 (t) A(t)
dt
On pre-multiplying the state equation by U −1 (t), post-multiplying above by x(t), and adding
dU −1 d −1
U −1 (t) ẋ + x(t) = U −1 (t) B(t) u(t) or [U (t) x(t)] = U −1 (t) B(t) u(t)
dt dt
The nonhomogeneous solution is obtained by integrating both sides from to to t, that is,
∫ t
U −1 (t) x(t) − U −1 (to ) x(to ) = U −1 (τ ) B(τ ) u(τ ) dτ or
to
∫ t
x(t) = U (t)U −1 (to ) x(to ) + U (t) U −1 (τ ) B(τ ) u(τ )dτ
to
The result again takes the form of a term depending on the initial state and a convolution
integral involving the input function. In fact, since u(to ) + In = U −1 (to ), the first term is the
same as the homogeneous solution.
The State Transition Matrix
The preceding results prompt the definition of an important matrix which can be
associated with any linear system, namely the transition matrix:
△
Φ(t, τ ) U (t) U −1 (τ ) (2.31)
−
This n × n matrix is a linear transformation or mapping of Σ onto itself. That is, in the
absence of any input u(t), given the state x(τ ) at any time τ , the state at any other time t is
given by the mapping
x(t) = Φ(t, τ ) x(τ )
The mapping of x(τ ) into itself requires that ϕ(t, τ ) = In for any τ . This is obviously true
from equation (2.31). Differentiating Φ(t, τ ) with respect to its first argument t gives
dΦ(t, τ ) dU (t) −1
= U (τ ) = A(t) U (t) U −1 (τ )
dt dt
d Φ(t, τ )
so = A(t) Φ(t, τ ) (2.32)
dt
The set of above differential equation, along with the initial condition, equation (2.31), is
often considered as the definition for the state transition matrix Φ(t, τ ).
Two other important properties of the transition matrix are the semigroup prop-
erties while defining ”state”:
Φ(t2 , to ) = Φ(t2 , t1 ) Φ(t1 , tO ) for any to , t1 , t2 (2.33)
and the relationship between Φ−1 and Φ
Φ−1 (t, to ) = Φ(to , t) for any to , t (2.34)
Both of these properties are obvious if the definition of equation (2.31) is considered.
Methods of computing the Transition Matrix:
If the matrix A is constant, then Φ(t, τ ) = eA(t−τ )
Therefore, following methods can be applied for finding Φ [46]:
1. Φ(t, 0) = L−1 [I s − A]−1 . Φ(t, τ ) is then found by replacing t by t − τ , since Φ(t, τ ) =

Φ(t − τ, 0) when A is constant.
2. Φ(t, τ ) = α0 I + α1 A + . . . + αn−1 An−1 , where eλi (t−τ ) = α0 + α1 λi + . . . + αn−1 λn−1

i
and,if some eigenvalues are repeated, derivatives of the above expression with respect
to λ must be used.
3. Φ(t, τ ) = M eJ(t−τ ) M −1 , where J is the Jordan for form (or the diagonal matrix Λ), and
M is the modal matrix.
∑n
4. Φ(t, τ ) = i=1 eλi (t−τ ) Zi (λ), where the matrices Zi are defined in [46, Ch.8].
5. Φ(t, τ ) ∼
= I +A(t−τ )+ 12 A2 (t−τ )2 + 3!1 A3 (t−τ )+. . . This infinite series can be truncated
after a finite number of terms to obtain an approximation for the transition matrix. A
more efficient computational form of this series is given by Φ(t, τ ) ∼
= [I + A(t − τ ) [I +
1
2
A(t − τ ) [I + 3!1 A(t − τ ) [I + . . .]]]]]].
A modification of method 1, using signal flow graphs to avoid the matrix inversion,
can also be used. Since Φij (s) △ Lϕij (t, 0) is the transfer function from the input to the jth
integrator to the output of the ith integrator, that is, the ith state variable xi , Mason’s gain
rule can be used to write the components ϕij (s) directly. Inverse Laplace transformations
then give the elements of Φ(t, 0).
When A(t) is time-varying, the choices for finding Φ(t, τ ) are most restricted:
1. Computer solution of Φ = A(t)Φ with Φ(τ, τ ) = I. This is expensive in terms of com-

puter time if the transition matrix is required for all t and τ . It means solving the matrix
differential equation many times, using a large set of different τ values as initial times.
∫t
2. Let B(t, τ ) = to A(ζ) dζ. Unlike the scalar case, Φ(t, τ ) ̸= eB(t,τ ) unless B(t, τ ) and A(t)
commute. Unfortunately, they generally do not commute, but two cases for which they
do are when A is constant and when A is diagonal. Whenever BA = AB, and methods
may be used for computing Φ(t, τ ) = eB(t, τ )
3. Successive approximations may be used to obtain an approximate transition matrix:

∫ t ∫ t ∫ τo
Φ(t, to ) = In + A(τo ) dτo + [ A(τo )( A(τ1 ) dτ1 ) dτo ]
to to to
∫ t ∫ τo ∫ τ1
+ { A(τo )[ A(τ1 )( A(τ2 ) dτ2 ) dτ1 ] dτo } + . . .
to to to
4. In some special cases, closed form solutions of equations may be possible.
5. When the solution for x(t) is used, the output equation becomes
∫ t
y(t) = C(t)Φ(t, to )x(to ) + C(t)Φ(t, τ )B(τ )u(τ )dτ + D(t)u(t)
to
In some special cases, it may be necessary or desirable to select a set of discrete time points,
tk , such that A(t) can be approximated by a constant matrix over each interval [tk , t(k+1) ].
2.3 Response of a Linear Time Invariant System

Consider a LTIV system [10, 45, 46]:
ẋ(t) = Ax(t) + Bu(t), x(0) = x0
x(t) = Transient response + forced response

Input → sinusoidal function - a cos(ωt)
→ power series function f (t) = a0 + a1 t + a2 t2 + ... .
→ step functions → integral of impulse
→ ramp functions - integral of step
→ parabolic function - integral of ramp
→ impulse function.
Frequency and Time domain response of a first order system
Unit step response (analytical expression) of a typical first order system is given in
Fig. 2.4. One can follow blue arrows to see the evolution of the response component generated
by the pole or zero. For a typical first order system,
C(s) a a 1
G(s) = R(s)
= s+a
=⇒ C(s) = s+a
R(s). For an unit step input, R(s) = s
a A B 1 −1
=⇒ C(s) = s (s + a)
= s
+ (s+a)
= s
+ (s+a)
(By partial fraction expansion)
The unit step response is plotted in the Fig. 2.3. As expected this system does not have
1
overshoot or undershoot. The initial slope of the response = time constant
= a. The response
reaches 63% of final value in one time constant = a1 .
Figure 2.3: First-order system response to a unit step [45]

2.3. RESPONSE OF A LINEAR TIME INVARIANT SYSTEM 59
C(s) (s+2)
Example [45]: Consider a first order system with a zero (Fig. 2.4): G(s) = R(s)
= (s + 5)
(s+2) A B
=⇒ C(s) = s (s + 5)
= s
+ (s+5)
.
−3
By partial fraction expansion, A = (s+2)
|
(s + 5) s=0
= 25 ; B = (s+2)
s
|s=−5 = −5
= 3
5
Hence, C(s) = 2
5s
+ 3
5 (s+5)
=⇒ c(t) = L−1 (C(s)) = 2
5
+ 35 e−5t .
Figure 2.4: Unit step response of a first order system: a. System showing input and output;
b. pole-zero plot of the system; c. evolution of a system response [45].
Response of a second order system
A typical single degree of freedom system can be represented by
ẍ(t) + 2ξωn ẋ(t) + ωn2 x(t) = ωn2 u(t), x(0) = 0, ẋ0 = 0 (2.35)
On taking Laplace transform, Eq. (2.35) results in:
(s2 + 2ξωn s + ωn2 )x(s) = ωn2 u(s) (2.36)

x(s) ωn2
G(s) = = 2 (2.37)
u(s) (s + 2ξωn s + ωn2 )
Prove : ← Geq (s) = 1

s(s2 +2ξωn)
& D.C gain =1
For step function : u(s) = 1/s .
ωn2 (s2 + 2ξωn s + ωn2 ) − (s2 + 2ξωn s)

x(s) = =
s (s2 + 2ξωn s + ωn2 ) s (s2 + 2ξωn s + ωn2 )
1 s + 2ξωn √
= − 2 ; ω d △ 1 − ξ 2 ωn
s s + 2ξωn s + ωn2
√
1 s + ξωn (ξ/ −ξ 2 )ωd
i.e., x(s) = − −
s (s + ξωn )2 + wd2 (s + ξωn )2 + ωd2
Inverse Laplace transform:

√
−ξωn t
x(t) = 1 − e (cos ωd t + (ξ/ 1 − ξ 2 ) sin ωd t)
√
e−ξωn t −1 1 − ξ2
= 1− √ sin[ω d t + tan ( )]
1 − ξ2 ξ
Delay time : td : time to reach 50% of final value for the I st time
Figure 2.5: Step response components generated by complex poles of a typical under damped
Second-order system [45].
Rise time : tr : time required for the response to rise from 10 to 90%, or 5 to 95% or 0 to
100% of final value.
Step Response
1.4
Peak amplitude: 1.09

1.2 Overshoot (%): 9.47
At time: 0.78 sec
2% Settling Time: 1.13 sec DC Gain: 1
Rise Time: 0.554 sec
1
0.8
Amplitude
0.6 Time (sec): 0.271

Amplitude: 0.5
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3
Time (sec)
Figure 2.6: Unit step response of second order system [45].
→ Under damped system (ξ < 0.707) : 0 − 100% : normal

→ over damped system (ξ > 0.707) : 10 − 90% : normal
Peak time : tp : time required to reach first peak value.
Settling time : ts : time required to reach and stay. within 5% or 2% of the deviations about
the final value.
Maximum/peak overshoot Mp : Maximum value above unity: usually represented by percent
(%) overshoot.
Example: For a typical second order system, let ωn = 5rad/sec; ξ = 0.6; ωd = 4rad/sec; then
√
1 −1 1 − ξ2
tr = tan ( ) (f or 0 − 100%) ≈ 0.55sec.
ωd ξ
tp = π/ωd ≈ 0.785sec.
√
−(ξ/ 1−ξ 2 ) π
Mp = e × 100% ≈ 9.5% since ξ = 0.6
2%ts ≈ 4/ξωn ≈ 1.13sec.
5%ts ≈ 3/ξωn ≈ 0.85sec.
Time Constant T :
First order system : x(t) = Ae−at =⇒ −aT = −1 or T = 1/a
For a second order system:
1 1
x(t) = Ae−at sin(ωd t + ϕ); T = =
|a| ξωn
√
Frequency of oscillation of transients = ωd = ωn 1 − ξ 2 rad/sec
Figure 2.7: Second-order systems, pole plots, and step responses for different damping ratios
For an ideal single degree of freedom system (Example: Spring - mass - damper or pendulum
systems) represented by a second order dynamical system (Eqs. 2.35, 2.37; the damping ratio
ξ and natural frequency ωn decide all the response characteristics like rise time, % overshoot,
settling time etc. Likewise, any two of such characteristics decide the system parameters ξ and
ωn uniquely. In control system synthesis, the system specifications on the transient response
will provide the least expected bandwidth of the closed loop system. Hence, to use graphical
procedure to determine the desired pole locations, plots of Percent overshoot vs damping ratio,
Normalized rise time vs damping ratio for a second-order under damped response (given the
Figure 2.8: Step responses of second-order underdamped systems as poles move: a. with
constant real part; b. with constant imaginary part; c. with constant damping ratio
Fig. 2.9), etc are used.
Figure 2.9: Second-order systems, plots of Percent overshoot vs. damping ratio and Normal-
ized rise time vs. damping ratio for a second-order under damped response
Home work: Find Unit step response of a second order system with a zero, i.e.,
x(s) (ω 2 /z) (s + z)
G(s) = = 2 n
u(s) (s + 2ξωn s + ωn2 )
Typical response of a second order system with a zero is given in the Fig. 2.10.
Figure 2.10: Effect of adding a zero to a two-pole system
Unit step response of the third order system:
x(s) ωn2 p
=
u(s) (s2 + 2ξωn s + ωn2 )(s + p)
e−ξωn t βξ[ξ 2 (β − 2) + 1]
x(t) = 1 − 2 {βξ 2 (β − 2) cos ωd t + √ sin ωd t}
βξ (β − 2) + 1 1 − ξ2
e−pt
− ; β = p/ξωn ;
(βξ 2 (β − 2) + 1)
N ote : ξ 2 (β − 2)2 + (1 − ξ 2 ) > 0, since coef f icient of e−pt is always negative.
p > ξωn , real pole to left of complex conjugate pole pair ⇒ reduction in % overshoot, increase
in settling time
p < ξωn , real pole to right of complex conjugate pole pair ⇒ over damped response, with
ripple added to the response. For the system with the real pole to left of complex conjugate
pole pair, responses of the system as the pole moves away from the complex conjugate pole
pair is given in Fig. 2.11), where, note that c(t) represents y(t).
2.4. FREQUENCY RESPONSE ANALYSIS 65
Figure 2.11: Component responses of a three-pole system: a. pole plot; b. component

responses: non-dominant pole is near dominant second-order pair (Case I), far from the pair
(Case II), and at infinity (Case III).
2.4 Frequency Response Analysis

One of the first mathematical analysis of control systems was the frequency-domain ap-
proach. This is based on the developments of Pierre-Simon de Laplace (1749-1827), Joseph
Fourier (1768-1830), Augustin Louis Cauchy (1789-1857), and others. The central concept
of frequency-domain approach is that of transfer function. The transfer function of a linear
time-invariant system is defined as Y (s)/U (s), where Y (s) is the Laplace transform of the
output, and U (s) is the Laplace transform of the input of the system. It turns out that
the transfer function is the Laplace transform of the system impulse response h(t) Therefore,
G(s) = Y (s)/U (s) i.e., G(s) embodies the transfer characteristics of the system. The classical
control theory was expressed in the frequency domain and the s-plane using the methods of
Nyquist, Bode, Nichols and Evans. All that is needed is the magnitude and phase of the fre-
quency response, or the poles and zeroes of the open loop transfer function. This approach is
appropriate for linear time-invariant systems, especially for single-input/single-output systems
where the graphical techniques are very efficient.
Frequency response is the measure of any system’s output spectrum in response to

a sinusoidal input signal. Frequency response curves are often used to indicate the perfor-
mance of systems. Frequency response measurements can be used directly to quantify system
performance and design control systems. Only SISO systems are considered in this discussion.
In the steady state, sinusoidal inputs to a linear system generates a sinusoidal output at the
same frequency as that of the driving input signal 2 . The output differs from the input in
terms of its magnitude and phase angle delay. These differences are function of frequency
and the transfer function of system. The frequency response of LTIV system gives the ratio
of the phaser output to the phaser input of a sinusoidal signal over a band of frequencies.
Laplace variable “s” is replaced by a complex frequency “j ω”. Then the complex variable
representing the signal or the system transfer function can be split into magnitude and phaser
angle for a specified frequency ω. For example the sinusoidal signal M1 cos (ω t + ϕ1 ) can
be represented by M1 ̸ ϕ1 . The system transfer function G(s) can likewise be represented
by a complex number M ̸ ϕ. For a given input sinusoidal Mi ̸ ϕi , the steady - state output
sinusoid is
Mo ̸ ϕo = M ̸ ϕ ∗ Mi ̸ ϕi (2.38)
Mo (ω) √ 2
hence, M (ω) = = {Re[G(j ω)]}2 + {Im[G(j ω)]} (2.39)
Mi (ω)
Im[G(j ω)]
and ̸ ϕ(ω) = ̸ ϕo (ω) − ̸ ϕi (ω) = tan−1 [ ]. (2.40)
Re[G(j ω)]
Equations 2.39 and 2.40 are useful in defining the frequency response; where M (ω) is the
magnitude frequency response and ϕ(ω) is the phase frequency response. The combination
of the magnitude and phase responses is called frequency response and is M ̸ ϕ, which is also
same as the impulse response of the transfer function G(s) in phaser representation. This
means that a steady state output of a system, having transfer G(s) for the sinusoidal input,
has the same frequency as the excitation frequency ω. In polar form, G(j ω) = M ej ϕ . System
output magnitude is scaled by a factor M (ω) and it lags input by ϕ(ω). For example, G(s) =
(k −j ω)
1
s+k
⇒ G(j ω) = 1
(j ω+k)
= 1 (−j ω+k)
(j ω+k) (−j ω+k)
= (ω 2 +k2 )
. For this, M (ω) = √ 1
and phase
(ω 2 +k2 )
lag ϕ(ω) = − tan−1 ωk . The plots of the magnitude M (ω) and the angle ϕ(ω) of G(j ω) vs.
the frequency ω, define the frequency response of a control system.
2
[45] Norman S. Nise: Control Systems Engineering, John Wiley & Sons (Wiley Student Edition), Noida,
2004, Fourth Edition, Chapters 10
Plotting of Frequency Response: Impulse response of G(j ω) = M (ω) ̸ ϕ(ω) can be

plotted as a function of frequency 1) as a separate magnitude and phase plots or 2) as a polar
plot where the phaser length is magnitude and phaser angle is the phase lag or by plotting
the imaginary part of the frequency response against the real part of the frequency response
to obtain a Nyquist plot. When plotting separate magnitude and phase, the magnitude plots
can be plotted in decibels (dB) vs. log ω, where dB is 20 log M (Note: log = log10 ). Phase
plots have linear angles vs. log ω. One way to transform multiplication into addition is by
using the logarithm. Instead of using a simple logarithm, a deciBel (named for Alexander
Graham Bell) is used. The relationship between a quantity, Q and its deciBel representation,
X, is given by X = 20 log10 (Q). So if Q = 100 then X = 40; X = 3 gives Q = 1.41; & so on.
2
Examples : Frequency response plots for the transfer function G(s) = (s+2)
. The magnitude
or Bode plots are given in Figure 2.12 and that of polar plot is given in Figure 2.13. Bode
plots represent the gain and phase of a system as a function of frequency. This is referred to as
√
the frequency domain behavior of a system. For G(s) = 2
(s+2)
, M (ω) ̸ ϕ(ω) = [1/ (ω 2 + 4) −
tan−1 ω
2
for different ω. Note that slope of magnitude plot on the right side of the corner
frequency is 20 dB/decade. The phase lag in a first order system is less than 90◦ . Note that
both type of plots contain same information and can be mapped from one to another.
Bode Diagram
−10
Magnitude (dB)
−20
−30
−40
0
Phase (deg)
−45
−90
−1 0 1 2
10 10 10 10
Frequency (rad/sec)
2
Figure 2.12: Magnitude & phase plots of frequency response for G(s) = (s+2)
.
Given a transfer function, such as
y(s) 100 ∗ (s + 1) 100s + 100)

G(s) = = = 2
u(s) (s + 10) (s + 100) [(s + 110 s + 1000]
Nyquist Diagram
0.5
6 dB
4 dB 2 dB 0 dB−2 dB−4 dB −6 dB
0.4
−10 dB
10 dB
0.3
0.2
Imaginary Axis
20 dB −20 dB
0.1
−0.1
−0.2
−0.3
−0.4
−0.5
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1
Real Axis
2
Figure 2.13: Polar plots of frequency response for G(s) = (s+2)
.
the question naturally arises: ”How can we display this function?” The most useful way to
display this function is with two plots, the first showing the magnitude of the transfer function
and the second showing its phase. One way to do this is by simply entering many values for
the frequency, calculating the magnitude and phase at each frequency and plotting them. For
example, if MATLAB is used by entering the following commands
>> MySys=tf(100*[1 1],[1 110 1000])
100s+100
Transfer function:
s2 +110s+1000
>> bode(MySys); grid
The corresponding Bode plots obtained from MATLAB are given in Figure 2.14
Asymptotic Approximation: Bode Plots
There are many reasons to develop a method for drawing Bode plots/ diagrams manually.
By drawing the plots by hand, as to how the locations of poles and zeros effect the shape of
the plots can be better understood. With this knowledge, a system behavior in the frequency
domain can be predicted by simply examining its transfer function. On the other hand, if
the desired shape of transfer function is known, then one can use this knowledge of Bode
diagrams to generate the transfer function. The first task when drawing a Bode diagram by
hand is to rewrite the transfer function so that all the poles and zeros are written in the form
(1 + s/ω0 ). The reasons for this will become apparent while deriving the rules for a real pole.
Bode Diagram
−10
Magnitude (dB)
−20
−30
−40
90
45
Phase (deg)
−45
−90
−2 −1 0 1 2 3 4
10 10 10 10 10 10 10
Frequency (rad/sec)
100∗(s+1)
Figure 2.14: Magnitude & phase plots of frequency response for G(s) = (s+10) (s+100) .
A derivation can be done using the transfer function from above, but it is also possible to do
a more generic derivation.
The log-magnitude and phase frequency response curves as functions of log ω are
called Bode plots or Bode diagrams. Sketching of Bode plots can be simplified if they can
be approximated as a sequence of straight lines. Straight-line approximations simplifies the
evaluation of frequency responses of individual or combination of transfer functions. The
transfer function is written in a partial fraction terms of poles and zeros. If the magnitude
responses of each of the poles or zeros are known, then total magnitude response is the product
of the magnitude response of each term (for both numerator for zeros and denominator for
poles). On working with the logarithm of magnitude, the process gets simplified since the zeros
term’s magnitude responses get added while pole term’s magnitude responses get subtracted
(instead of getting multiplied and divided respectively) to yield the logarithm of the total
magnitude response written in terms of dB.
100 ∗ (s + 1) 0.1(1 + s/1)

Let G(s) = = (2.41)
(s + 10) (s + 100) (1 + s/10) (1 + s/100)
First note that this expression is made up of four terms, a constant (0.1), a zero (at s = −1),
and two poles (at s = −10 and s = −100). Now the magnitude and phase of this function
(with s = j ω) can be written as four individual phasers:
0.1(1 + j ω/1)
G(s) =
(1 + j ω/10) (1 + j ω/100)
|(1 + j ω/1)| ̸ (1 + j ω/1)
= |0.1| ̸ (0.1)
[|(1 + j ω/10)| ̸ (1 + j ω/10)] [|(1 + j ω/100) ̸ (1 + j ω/100)]
|(1 + j ω/1)|
Hence, |G(j ω)| = |0.1|
[|(1 + j ω/10)|] [|(1 + j ω/100)|]
̸
and (G(s)) = ̸ (0.1) + ̸ (1 + j ω/1) − ̸ (1 + j ω/10) − ̸ (1 + j ω/100)
Drawing the phase is fairly simple. Each phase term can be drawn separately, and then simply
add them. On using logarithm, the magnitude in dB can be written as:
20 log10 (|0.1|) + 20 log10 (|(1 + j ω/1)|) − 20 log10 (|(1 + j ω/10)| − 20 log10 (|(1 + j ω/100)|)
Plotting the constant term is trivial, however the other terms are not so straightforward.
These plots will be discussed below (Also refer figure 2.15). However, once these plots are
drawn for the individual terms, they can simply be added together to get a plot for G(s).
A Constant Term: Consider a constant term G(s) = G(j ω) = K
Magnitude: Clearly the magnitude is constant, |G(j ω)| = K. (Use 20 log10 K in the plot).
Phase: The phase is also constant. If K is positive, the phase is 0◦ (or any even multiple
of 180◦ ). If K is negative the phase is − 180◦ , or any odd multiple of 180◦ . MATLAB uses
− 180◦ . Expressed in radians, if K is positive, then the phase is 0 radians, if K is negative,
then the phase is ± π radians.
Bode Plot of Gain Term
• For a constant term, the magnitude plot is a straight line.
• The phase plot is also a straight line, either at 0◦ (for a positive constant) or ± 180◦ (for a
negative constant).
A Real Pole:
ω0 1 1
Consider a simple real pole G(s) = (s+ω0 )
= (1+s/ω0 )
; G(j ω) = (1+j ωω )
0
The frequency ω0 is called the break frequency or the corner frequency or the 3 dB frequency.
√
Magnitude: The magnitude is given by |G(j ω)| = | 1 + j 1ω / ω0 | = − 20 log10 [ (12 + ω2
ω02
) ] in dB
Consider cases for the three different values of frequency:
Case 1) ω ≪ ω0 . This is a low frequency zone. An approximation for the magnitude of the
transfer function can be written as:
√
|G(j ω)| = − 20 log10 [ (12 + ω2
ω02
) ] f or ω ≪ ω0 ≈ − 20 log10 (1) = 0
Figure 2.15: Rules for Bode plots of frequency response.

The low frequency approximation is shown in blue on the diagram in the Fig. 2.16.
1
Figure 2.16: Bode magnitude plots for G(s) = (1+s/ω ) .
0
Case 2) ω ≫ ω0 . This is the high frequency zone. The magnitude of the transfer function
can be approximated as:
√ √
|G(j ω)| = − 20 log10 [ (12 + ω2
ω02
) ] f or ω ≫ ω0 ≈ − 20 log10 ( ω2
ω02
) = − 20 log10 ( ωω0 )
The high frequency approximation is at shown in green on the diagram in the Fig. 2.16. It is
a straight line with a slope of −20 dB/decade going through the break/ corner frequency at
0 dB. That is, for every factor of 10 increase in frequency, the magnitude drops by 20 dB.
Case 3) ω = ω0 . The break/ corner frequency. At this frequency, i.e., for ω = ω0 ,
√ √
|G(j ω)| = − 20 log10 [ (12 + ωω2 ) ] = − 20 log10 ( 2) = − 3.01 ≈ − 3 dB
2
This point is shown as a red circle on the diagram in the Fig. 2.16.
To draw a piecewise linear approximation, use the low frequency asymptote up to the break
frequency, and the high frequency asymptote thereafter. The resulting asymptotic approxima-
tion is shown highlighted in pink. The maximum error between the asymptotic approximation
and the exact magnitude function occurs at the break frequency and is approximately 3 dB.
The rule for drawing the piecewise linear approximation for a real pole can be stated thus:
For a simple real pole the piecewise linear asymptotic Bode plot for magnitude is at 0 dB until
the break frequency and then drops at 20 dB per decade (i.e., the slope is -20 dB/decade).
Phase: The phase of a single real pole is given by
̸ (G(j ω)) = − ̸ (1 + j ω
ω0
) = − arctan( ωω0 )
The three cases for the different values of the frequency ω are considered once again:
Case 1) ω ≪ ω0 . This is the low frequency case. At these frequencies, an approximation for
the phase of the transfer function can be written as:
̸ (G(j ω)) ≈ − arctan(0) = 0 radians
The low frequency approximation is shown in blue on the diagram in the Fig. 2.17.
1
Figure 2.17: Bode phase plots for G(s) = (1+s/ω ) .
0
Case 2) ω ≫ ω0 . This is the high frequency case. An approximation for the phase of the
transfer function can be written as
̸ (G(j ω)) ≈ − arctan(∞) = − 90◦ = −(π / 2) radians
a straight line with a slope at -90.
Case 3) ω = ω0 . The break/ corner frequency. At this frequency
̸ (G(j ω)) ≈ − arctan(1) = − 45◦ = −(π / 4) radians
This point is shown as a red circle on the diagram in the Fig. 2.17.
A piecewise linear approximation is not as easy in this case because the high and
low frequency asymptotes don’t intersect. Instead we use a rule that follows the exact function
fairly closely, but is also arbitrary. Its main advantage is that it is easy to remember. The rule
can be stated as ‘Follow the low frequency asymptote until one tenth the break frequency (0.1 ω0 )
then decrease linearly to meet the high frequency asymptote at ten times the break frequency
(10 ω0 )’. This line is also shown in the Fig. 2.17. Note that there is no error at the break
frequency and about 5.7◦ of error at one tenth and ten times the break frequency.
Example 1: Real Pole The first example is a simple pole at 10 radians per second.
The low frequency asymptote is the dashed blue line, the exact function is the solid black
line, the cyan line represents 0.
1 1
G(s) = (1+s/10) ; G(j ω) = (1+j ω )
10
1
Figure 2.18: Bode plots for G(s) = (1+s/10) .
Example 2: Repeated Real Pole

The second example shows a double pole at 30 radians per second. Note that the slope of the
asymptote is -40 dB/decade and the phase goes from 0 to −180◦ .
1
G(s) =
(1+s/30)2
Key Concept: Bode Plot for Real Pole
• For a simple real pole the piecewise linear asymptotic Bode plot for magnitude is at 0 dB
until the break frequency and then drops at 20 dB per decade (i.e., the slope is -20 dB/decade).
An nth order pole has a slope of − 20 ∗ n dB/decade.
1
Figure 2.19: Bode plots for systems with repeated poles/ roots G(s) = .
(1+s/30)2
• The phase plot is at 0◦ until one tenth the break frequency and then drops linearly to -90
at ten times the break frequency. An nth order pole drops to − 90 ∗ n.
• The Bode plots for real zeros can be similarly drawn as below, details in Nise’s book [45].
A Real Zero
The piecewise linear approximation for a zero is much like that for a pole.
Consider a simple zero:
ω
G(s) = (1 + s/ω0 ); G(j ω) = (1 + j ω0
)
Magnitude : The development of the magnitude plot for a zero follows that for a pole. Refer
to the previous discussion for details. The magnitude of the zero is given by
√
|G(j ω)| = |1 + j ω / ω0 | = 20 log10 [ (12 + ω2
ω02
)] in dB
Again there are three cases:
1. At low frequencies, ω ≪ ω0 , the magnitude/ gain is approximately zero.
ω2
2. At high frequencies, ω ≫ ω0 , ω02
dominates the magnitude expression, allowing us to
approximate the magnitude as 20 log (ω/ω0 ), the gain increases at 20 dB/decade (increase in
ω by a factor of 10, increases the magnitude by 20 dB.) and goes through the break frequency
at 0 dB.
√
3. At the break frequency, ω ≈ ω0 , the gain is about 3 dB since, 20 log10 [ (12 + ω2
ω02
)] =
√
20 log10 [ 1 + 1] .
The rule for drawing the piecewise linear approximation for a real zero can be stated thus:
For a simple real zero the piecewise linear asymptotic Bode plot for magnitude is at 0 dB until
the break frequency and then increases at 20 dB per decade (i.e., the slope is +20 dB/decade).
Phase : The phase of a simple zero is given by:
ω
̸ (G(j ω)) = ̸ (1 + j ω0
) = arctan( ωω0 )
The phase of a single real zero also has three cases: 1. At low frequencies, ω ≪ ω0 , the phase
is approximately zero. 2. At high frequencies, ω ≫ ω0 , the phase is 90◦ . 3. At the break
frequency, ω ≈ ω0 , the phase is 45◦ . The rule for drawing the phase plot can be stated thus:
Follow the low frequency asymptote until one tenth the break frequency (0.1 ω0 ) then increase
linearly to meet the high frequency asymptote at ten times the break frequency (10 ω0 ).
Example: This example shows a simple zero at 30 radians per second. In the Fig. 2.20the
low frequency asymptote is the dashed blue line, the exact function is the solid black line, the
cyan line represents 0.
Figure 2.20: Bode plots for G(s) = (1 + s/ω0 ); ω0 = 30 rad/sec.
Key Concept: Bode Plot of Real Zero:

• For a simple real zero the piecewise linear asymptotic Bode plot for magnitude is at 0
dB until the break frequency and then rises at +20 dB per decade (i.e., the slope is +20
dB/decade). An nth order zero has a slope of + 20 ∗ n dB/decade.
• The phase plot is at 0◦ until one tenth the break frequency and then rises linearly to 90◦ at
ten times the break frequency. An nth order zero rises to + 90◦ ∗ n.
Fig.2.21 gives summary of Bode plots for the normalized magnitude and scaled frequency
response for four different I st order transfer functions. Scaled frequency plots show the linear
approximation of phase variations in one decade of interval either side of the corner frequency.
Analysis of a Second Order System
Consider a typical second order system (or a single degree of freedom system) given by Equa-
tion (2.35) having a transfer function (On taking Laplace transform of Eq. (2.35)):
x(s) ωn2 1
G(s) = = 2 2
= 2
) (2.42)
u(s) (s + 2ζωn s + ωn ) [(s/ωn ) + 2ζ(s/ωn ) + 1]
with 0 < ζ < 1. This system has a complex conjugate pair of poles. The magnitude and
phase plots of a complex conjugate (under-damped) pair of poles having transfer function
given by Eq.2.42 is more complicated than those for a simple pole.
Magnitude : The magnitude of the transfer function is given by
√
|G(j ω)| = | √ ω
1
ω | = − 20 log10 [ (1 − ( ωω )2 )2 + (2 ζ ωω )2 ] in dB
(1 − ( ωn ) ) + (2 ζ ωn )
2 2 2 n n
Consider the following three cases for the value of the frequency:
Case 1) ω ≪ ωn . This is the low frequency case. An approximation for the magnitude of the
transfer function is |G(j ω)| ≈ − 20 log10 (1) = 0
The low frequency approximation is shown in red on the diagram in the Fig. 2.22.
Case 2) ω ≫ ωn . This is the high frequency case. An approximation for the magnitude of
the transfer function is
|G(j ω)| = − 20 log10 ( ( ωω )2 ) = − 40 log10 ( ωω )
n n
The high frequency approximation is at shown in green on the diagram in the Fig. 2.22. It
is a straight line with a slope of -40 dB/decade going through the break frequency at 0 dB.
That is, for every factor of 10 increase in frequency, the magnitude drops by 40 dB.
Case 3) ω ≈ ωn . It can be shown that a peak occurs in the magnitude plot near the break
frequency. The exact height and location can be determined by differentiating the expression
for the magnitude of the transfer function with respect to frequency and setting it to zero (to
make life easier, square the function before differentiating, since the peak will be in the same
place for a function or its square). The resulting differentiation shows a peak at the frequency
√
given by ωd = ωn (1 − ζ 2 ), where ωd = damped frequency, while ωn = natural frequency.
Figure 2.21: Normalized and scaled Bode plots for a. G(s) = s; b. G(s) = 1/s; c. G(s) =
(s + a); d. G(s) = 1/(s + a) [45, p.603].
ωn2
Figure 2.22: Bode plots for G(s) =
(s2 +2ξωn s+ωn2 ) .
√
The peak has a magnitude of |G(j ωd )| = − 20 log10 (2 ζ (1 − ζ 2 ).
The actual peak frequency is not important when drawing Bode diagrams by hand because if
the peak is large enough to draw, the peak frequency is very near the break frequency. This
point is shown as a blue circle on the diagram in the Fig. 2.22. Note that the peak only
√
exists for 0 < ζ < 0.707 = 1/ 2; and the frequency of the peak is typically very near the
break frequency. For ζ = 0, the peak is exactly at the resonant frequency ωn , but the peak
frequency drops as ζ increases. However even for a fairly large ζ = 0.3 (a small peak of only
√ √
5 dB), the damped resonant frequency is ωd = ωn (1 − ζ 2 ) = ωn (1 − 0.09) = 0.91 ωn ,
which is only a 9% deviation from the break frequency. It is generally accurate enough to put
the peak at the resonant frequency. To draw a piecewise linear approximation, use the low
frequency asymptote up to the break frequency, and the high frequency asymptote thereafter.
Draw a smooth curve between the low and high frequency asymptote that goes through the
peak value. For the curve shown in the Fig. 2.22,
1
G(s) = ωn = 10, ζ = 0.1
[(s/ωn )2 +2 ζ (s/ωn )+1]
The peak will have an amplitude of 5.02 or 14 dB.
The resulting asymptotic approximation is shown as a black dotted line, the exact
response is a black solid line. The rule for drawing the piecewise linear approximation for a
complex conjugate pair of poles can be stated thus:
For the magnitude plot of complex conjugate poles draw a 0 dB at low frequencies, go through
√
a peak of height, |G(j ωd )| = − 20 log10 (2 ζ (1 − ζ 2 ). at the break frequency ωd and then
drop at 40 dB per decade (i.e., the slope is -40 dB/decade). The high frequency asymptote
goes through the break frequency.
Phase :
The phase of a complex conjugate pole is given by
2 ζ ωωn
̸ (G(j ω)) = − arctan
(1 − ωωn 2 )
Consider again the three cases for the value of the frequency:
Case 1) ω ≪ ωn . This is the low frequency case. At these frequencies An approximation for
the phase of the transfer function
̸ (G(j ω)) ≈ − arctan(0) = 0 radians
The low frequency approximation is shown in red on the diagram in the Fig. 2.23.
Case 2) ω ≫ ωn . This is the high frequency case. An approximation for the phase of the
transfer function is
̸ (G(j ω)) = − 180◦
a straight line at − 180◦ .
Case 3) ω ≈ ωn . The break frequency. At this frequency, ̸ (G(j ω)) = − 90◦
This point is shown as a blue circle on the diagram in the Fig. 2.23.
1
For the curve shown in the Fig. 2.23, G(s) = 2 ωn = 10, ζ = 0.1
[(s/ωn ) +2 ζ (s/ωn )+1]
A piecewise linear approximation is not easy in this case, and there are no hard and fast rules
for drawing it. The most common way is to look up a graph in a textbook with a chart that
shows phase plots for many values of ?. Another way is to use connect the low frequency
log10 ( ζ2 )
asymptote to the high frequency asymptote starting at ωℓ = ωn 2 and ending at
2
ωu = ωn . If ζ < 0.02, the approximation can be simply a vertical line at the break
log10 ( ζ2 )
ωn2
Figure 2.23: Bode phase plots for G(s) =
(s +2ξωn s+ωn2 )
2 .
frequency.
The rule for drawing phase of an under damped pair of poles can be stated as:
Follow the low frequency asymptote at 0◦ until ωℓ and then decrease linearly to meet the high
frequency asymptote at − 180◦ at ωu .
Key Concept: Bode Plot for Complex Conjugate Poles:
• For the magnitude plot of complex conjugate poles draw a 0 dB at low frequencies, go through
√
a peak of height, |G(j ωd )| = − 20 log10 (2 ζ (1 − ζ 2 ) in dB at the break frequency and then
drop at 40 dB per decade (i.e., the slope is - 40 dB/decade). The high frequency asymptote
√
goes through the break frequency. Note that the peak only exists for 0 < ζ < 0.707 = 1/ 2.
log10 ( ζ2 )
◦
• To draw the phase plot simply follow low frequency asymptote at 0 until ωℓ = ωn 2
◦ 2
then decrease linearly to meet the high frequency asymptote at − 180 at ωu = ωn .
log10 ( ζ2 )
If ζ < 0.02, the approximation can be simply a vertical line at the break frequency.
A Complex Conjugate Pair of Zeros
Not surprisingly a complex pair of zeros yields results similar to that for a complex pair of
poles. The differences are that the magnitude has a dip instead of a peak, the magnitude
increases above the break frequency and the phase increases rather than decreasing.
Figure 2.24: Bode plots for G(s) = ( ωsn )2 + 2 ξ ( ωsn ) + 1; ωn = 10, ζ = 0.1.
The graph in the Fig. 2.24 corresponds to a complex conjugate zero with
G(s) = ( ωsn )2 + 2 ξ ( ωsn ) + 1; ωn = 10, ζ = 0.1.
The dip in the magnitude plot will have a magnitude of 0.2 or -14 dB.
Key Concept: Bode Plot of Complex Conjugate Zeros:
• For the magnitude plot of complex conjugate zeros draw a 0 dB at low frequencies, go
√
through a dip of magnitude: |G(j ωd )| = 20 log10 (2 ζ (1 − ζ 2 ) in dB at the break frequency
and then rise at +40 dB per decade (i.e., the slope is +40 dB/decade). The high frequency
asymptote goes through the break frequency. Note that the peak only exists for 0 < ζ <
√
0.707 = 1/ 2.
log10 ( ζ2 )
◦
• To draw the phase plot simply follow low frequency asymptote at 0 until ωℓ = ωn 2
◦ 2
then increase linearly to meet the high frequency asymptote at 180 atωu = ωn .
log10 ( ζ2 )
Bode - Magnitude and Phase plots for variable damping ratio ζ for a fixed natural
frequency ωn are given the following figures (Fig. 2.25 & Fig. 2.26; reproduced from Figures
10.16 and 10.17 from Nise’s book [45]).
Figure 2.25: Normalized and scaled log magnitude response for G(s) = ( ωsn )2 + 2 ξ ( ωsn ) + 1;
for variable ζ and a fixed ωn .
Intrinsic properties of a LTIV system are:
- Stability
- Controllability
- Observability
and related intrinsic properties are:
Figure 2.26: Scaled Bode phase response for G(s) = ( ωsn )2 + 2 ξ ( ωsn ) + 1; for variable ζ and
a fixed ωn .
- Stabilzabilty
- Detectability
- Disturbabilty.
2.5 Stability
The response of a LTIV system
ẋ = Ax + Bu, y = Cx (2.43)
2.5. STABILITY 85
is given by
∫ t
A(t−to )
x(t) = e x(to ) + eA(t−τ ) Bu(τ )dτ (2.44)
to
The first part of the equation (2.44) is the homogeneous solution (or complementary function)
due to nonzero initial conditions, while the second part is the forced response (or particular
integral). Actual solution or response x(t) may contain first part (transient response) or sec-
ond part (forced response) or both as in the eqn.(2.44).
Asymptotic stability : Indicator of capability of system to return to equilibrium after being

initially perturbed from equilibrium.
⇒ Asymptotic stability ⇐⇒ Transient (I.C) response, i.e., x(0) ̸= 0, u(t) = 0
stability : The dynamic system, which is initially disturbed from an equilibrium should return
to equilibrium (or to its close vicinity) with the progression of time.
The system described by ẋ(t) = f (x), x(0) = x0 is asymptotically stable with
respect to an equilibrium xe only if for a given ||x0 − xe || < δ there exists a ε(δ) such that
||x(t) − xe || < ε(δ) for t > tf < ∞ and limt→∞ ε (δ) = 0.
Pendulum Inverted Pendulum

Stable unstable
θ̈ + g/l sin θ = 0 θ̈ − g/l sin θ = 0
θ(0) ̸= 0, θ̇(0) = 0 θ(0) ̸= 0, θ̇(0) = 0
Table 2.1: Example of a stable and an unstable system
Bounded input, Bounded Output/State stability :

Forced response =⇒ BIBS/ BIBO stability.
BIBO Stability is often phrased as the bounded response of the system to any bounded input:
Zero initial conditions, & nonzero forcing function.
System is stable if every bounded input produces bounded output state, unstable if some
bounded input produces unbounded output.
⇒ BIBS /BIBO stability, x(0) = 0, ||u|| ≤ M ∀ t.
For LTIV systems, under the conditions of controllability & observability,
Asymptotic stability ⇔ BIBO/BIBS stability
For LTIV systems

Stability ⇒ real part of all the poles are negative
Unstable ⇒ real part of at least one pole is positive
Example: Longitudinal motion of the Aircraft
ẋ = Ax + Bu, x = (u, α, q, θ)
   
 Xu Xα 0 −g   Xe 
   

 Zu /V Zα /V 1 0 

 
 Ze/V 
A = 

, B = 
 


 M Mα Mq 0   M 
 u   e 
   
0 0 1 0 0
The characteristic equation ⇒ det ∥s I − A∥ = 0
s4 + a1 s3 + a2 s2 + a3 s + a4 = 0
Zα
a1 = − − Mq − Xu = 1.0603,
V
Zα Zα Zα
a2 = Mq − Mα + Xu ( + Mq ) − Xα = −1.1154
V V V
Zα Zu
a3 = −Xu ( Mq − Mα ) + Xα ( Mq − Mu ) = −0.0565
V V
(Zu Mα − Zα Mu )
a4 = g( = −0.0512
V
For the data supplied earlier, (AFTI-16) poles are at (roots of the above characteristic equa-
tion): −1.705, +0.724, −0.0394 ± 0.2j. Note one sign change in the coefficient (at s2 - term)
of characteristic equation.
In general, for a LTIV SISO system:
ẋ = Ax + bu, y = cx, x(0) = 0
s x(s) = Ax(s) + bu(s) ⇒ (sI − A)x(s) = bu(s)
Laplace transform :
x(s) = (sI − A)−1 b u(s), y(s) = c x(s) = c (sI − A)−1 b u(s).
SISO System TF
y(s)
= c (sI − A)−1 b = G(s) = {Np (s)}/{Dp (s)}
u(s)
2.5. STABILITY 87
MIMO System : matrix of transfer function (T.F) with ratio of polynomials as elements of
the matrix T.F.
BIBO stability : (Limited to SISO system for this section)
r(s) e u y
k G(s)
+ -
H(s)
Figure 2.27: Typical Feedback Control System
r(s) e(s) y(s)

u G(s)
K H (s)
+ -
Figure 2.28: Typical Cascade Control System
With Feedback control :

Open loop transfer function (OLTF): K G H
Closed loop transfer function (CLTF) : K G/{1 + K G H}. ⇐ Prove!
Cascade/feedforward compensator :
OLTF : K G H ; CLTF : K G H(s)/{1 + K G H(s)} . ⇐ Prove!
Characteristic equation: 1 + KGH = 0 or KGH = −1 = 1 ̸ (180◦ )
Np (s) Nc (s)
Let G = Dp (s)
and H = Dc (s)
, then the characteristic equation is given by
Dp Dc + K N p N c = 0
Stable: all poles have negative real part; Unstable: at least one pole has a positive real part.
Study of stability of a closed loop system by analyzing the properties of its OLTF.
Methods of determining stability :
1. Routh’s criterion determines how many roots have positive real parts (unstable) directly
from the coefficients of the characteristic polynomial of the CLTF.
2. Root locus is a graphical means of factoring characteristic equation: 1+KGH=0 or

KGH=-1. Since GH is a complex number, K|GH| = 1 and ̸ GH = (1+2m) 180◦ for
any integer m. Root locus determines the closed loop stability by working with the open
loop transfer function OLTF : KGH, which is normally available in a factored form.
3. Bode plots provide information on stability in graphical form for minimum phase system
(systems with no open loop poles or zeros in the right half s-plane).
Magnitude and phase angles are considered separately as above equations represent, but
only s is set to s = j ω Decebel units for magnitude and log scale for frequency are
used. The critical point for stability is -1. becomes point of 0 dB and -180◦ phase shift.
4. Nyquist stability criterion : polar plots convey same information as Bode plot’s but
KGH (s=j ω) is plotted as a locus of phasers with frequency, ω as the variable parameter.
The critical point is (-1,0) is KGH (s) -plane. Nyquist stability criterion applies to non
minimum phase system as well.
In principle, Bode plots and Nyquist plots contain same information, selection of a particular
form of presentation depends on the familiarity of the user. In addition to finding stability or
otherwise of a closed loop system by examining the open loop transfer function, these plots
are also useful in determining the relative degree of stability. Two useful measures or figures
of merit for relative stability are gain margin and phase margin; which are discussed next.
Definition of Gain and Phase margins
The phase margin is that amount of additional phase lag at the gain cross over frequency
required to bring the system to the verge of instability. That is, the phase margin is the
change in the open-loop phase shift required at the unity gain to make the closed-loop system
unstable. The gain cross over frequency is the frequency at which |GH(jω)|, the magnitude
of OLTF is unity. The phase margin γ is 180◦ + phase angle ϕ of GH at the gain cross over
frequency i.e., γ = 180 + ϕ.
2.5. STABILITY 89
The gain margin is the change in the open-loop gain (expressed in decibels), required at
− 180◦ of phase shift to make the closed loop system unstable. That is, Gain margin is the
reciprocal of the magnitude |GH(jω)| at the phase cross over frequency, where the phase angle
is −180◦ . Phase cross over frequency is the frequency at which the phase angle ̸ GH = −180◦ .
Upward gain margin and Downward gain Margins: Incremental gain or reduction in gain,
which brings the feedback system to the verge of instability.
Gain margin Kg = 1
|GH| ωp
or Kg |db = 20 log |GH|
1
= −20 log|GH(ωp )|
ωp
For Stable system : P M > 0 & GM > 0
Larger PM & GM ⇒ better stability properties; how ever, for example,
→ larger damping ⇒ sluggish response.
√
(because for ex: ωd = ωn (1 − ζ 2 ) , ts = 4
ζ ωn
,⇒ ζ ↑⇒ ts ↓.
Systems with greater gain and phase margins can tolerate larger changes in system parameters
before being driven to a region of instability.
θ(s) µ
Example of Gain reduction margin : Let G(s) = δ(s) = (s2 −µ c
, where µα > 0. For an
α)
unstable dynamics, proportional + derivative feedback is required for stabilization. Let the
feedback compensator for an ideal (or text book) form of controller H(s) = − (Kp + Kd s).
θ(s) µc ∗(Kp +Kd s)
The open loop transfer function (OLTF) G H(s) = δ(s) = (s2 −µα ) or state space form
of the governing equation is given by
d2 θ d2 θ
d t2
− µα θ = µc δ = µc u = − µc (Kp θ + Kd dθ
dt
). =⇒ d t2
+ µc Kd dθ
dt
+ (µc Kp − µα ) θ = 0 .
Note that µc Kd > 0. For stability (use Routh - Hurwitz Criterion), (µc Kp − µα ) > 0 =⇒
µ µ
Kp > µα . This requirement of Kcritical = µα is called static stability requirement/ Criterion.
c c
If the control gain Kp falls below this critical value, then the system is unstable. Ratio of
designed value of Kp /Kcritical is the measure of gain reduction margin. In summary, gain
reduction margin is critical for unstable systems.
The notions of gain and phase margins, which are commonly used measures of
closeness of a SISO feedback system to stability, have gained acceptance as the specifications
for a control system design. For example, consider military specifications for piloted aircraft:
stability margins required for FCS allow variations in system dynamics: a) Math modelling
and data errors; b) Variations in dynamic characteristics caused by changes in environmental
conditions, manufacturing tolerances, etc; c) maintenance induced errors in calibration, in-
stallation and adjustment. It may be noted that feedback properties have little correlation to
system’s response to command input.
2.5.1 Bode stability Analysis
Bode plots are easier to draw than Nyquist plots, since they can be drawn without the aid of
computational devices or long calculations as in the later case. Using the Bode approximations
as discussed earlier, Bode log-magnitude and phase plots Vs log frequency are drawn. Fig.
2.29 shows the expanded plots of a simplified text book like scenario of Bode plots near cross
over frequencies (drawn only for explanation purpose of evaluating gain and phase margins
from Bode plots). In the Fig. 2.29, M = |G H| and phase = ̸ (G H).
The gain margin is found by using the phase plot to find the phase cross over
frequency ωp = ωGM (in the fig), where the phase angle is 180◦ . At this frequency ωGM , look
at the magnitude plot to determine the gain margin, Kg = GM , which is the gain required
to raise the magnitude curve to 0 dB.
The phase margin is found by using the magnitude curve to find the gain cross
over frequency ωg = ωΦM (in the fig), where the gain is 0 dB. On the phase curve at that
frequency, the Phase margin = ΦM is the difference between the phase value and 180◦ .
K
Example: (Example 10.9 from Nise’s book [45]) G(s) H(s) = [(s+2) (s+4) (s+5)] .
Select K = 40 so that the dc gain is unity or log-magnitude plot starts at 0 dB.
This is a conditionally stable system as seen from the study of characteristic equation
(s + 2) (s + 4) (s + 5) + K = 0
s3 + 11 s2 + 38 s + (40 + K) = 0
At K = 378, s3 + 11 s2 + 38 s + 418 = 0
Corresponding roots of characteristic equation are [−11, 0 ± 6.1644 j], i.e., system is in the
verge of instability. For K > 378, closed loop system is unstable.
In order to draw Bode plots approximately, mark the break frequencies on semi-log graph
sheets, indicating frequency in log-scales along the X − axis. Each division along the Y − axis
for the magnitude plots may indicate 20dB or 10dB as the case may be and like wise for the
phase plots (Refer Fig. 2.30). Draw a vertical line corresponding to the break frequency (Since
the grid lines coincide with the break frequencies of 2 rad/sec, 4 rad/sec and 5 rad/sec in the
Fig. 2.30, no separate vertical lines were drawn). Low frequency gain at low frequency for
2.5. STABILITY 91
Figure 2.29: Gain and phase margins on the Bode diagrams, Source: Fig.10.37, [45]
OLTF G(s) H(s) is obtained by setting s = 0. Since K = 40, log-magnitude plot starts
at 0 dB. At each break frequency corresponding to a pole, an increase of - 20 dB/decade is
effected. The log-magnitude plot is given at the top of the Fig. 2.30.
The phase diagram also begins at 0◦ until a decade below the first break frequency of 2 rad/sec.
Since the OLTF has only three poles, phase angle will finally settle at − 270◦ . At 0.22 rad/sec,
the phase curve decreases at a rate of 45 ◦ /decade, decreasing an additional 45 ◦ /decade at
each subsequent frequency (0.4 rad/sec, 0.5 rad/sec), a decade below each break frequency.
At each decade above each break frequency, the slopes are reduced by 45 ◦ /decade.
Bode log-magnitude should be less than 0 dB (or unity in magnitude) when bode
phase plot is − 180◦ . Hence, the gain cross over frequency is ≈ 7 rad/sec. The corresponding
value at log–magnitude plot is 20 dB (or a gain of 10). Since the gain was scaled at 40, the
required gain K = 400. Hence for stability, 0 < K < 400. Look at the phase plot, reveals that
phase gain cross over frequency is 0 rad/sec and the phase margin is − 180◦ . On performing
the exact phase and gain plots using Matlab (or using the ‘margin’ command of Matlab):
num = 40; den = [1 11 38 40]; [Gm,Pm,Wcg,Wcp] = margin(num, den); =⇒
K
Figure 2.30: Bode plots for OLTF G(s) H(s) = [(s+2) (s+4) (s+5)] ; K = 40 [45]
Gm = 9.4500; Pm = -180; Wcg = 6.1644; Wcp = 0). This validates the Bode approximation.
Note that 378/40 = 9.45, critical gain K = 378 is computed from the characteristic equation.
Example Bode 1: Consider the normalized pitch attitude dynamics and stabilization system
for a satellite (Figs. 3.1, 2.31). Assume τ = 1.8 and Kp = 3.25. The corresponding Bode plot
is given in Figure 2.32. On using the Matlab command num = 3.25*[1.8 1]; den = [1 5 0 0];
[Gm, Pm, Wcg, Wcp] = margin(num, den); =⇒
Gm = 0; Pm = 51.9607; Wcg = 0; Wcp = 1.2436). =⇒ 1
Gm
= ∞ (infinite gain margin)
Exercise: Draw the approximate Bode plot by hand on a semi-log graph sheet and determine
the margins.
2.5. STABILITY 93
Figure 2.31: Block diagram for Satellite Pitch attitude control example (Bode 1)
Bode Diagram
100
50
Magnitude (dB)
−50
−100
−120
Phase (deg)
−150
−180
−2 −1 0 1 2
10 10 10 10 10
Frequency (rad/sec)
Figure 2.32: Bode plot for Satellite Pitch attitude control example
Example Bode 2: Consider the transfer function corresponding to a longitudinal dynamics

and actuator dynamics 15/(s+15), having the short period and phugoid modes, for a statically
stable transport aircraft (See Blakelock [7], Chapter 1) is given by
θ̇(s) 15 s (s + 0.3) (s + 0.016)

= − 1.31
δe (s) {(s + 15) (s + 0.00466s + 0.0053) (s2 + 0.806s + 1.311)}
2
The transfer of the cascade PI (proportional + integral) controller is Kp + KsI . Draw Bode
plot and find margins for
Case 1: Assume KI = 0.6, and Kp as a variable gain. Find range of Kp for which the closed
loop system is stable.
Case 2: Assume Kp = 10.0, and KI as a variable gain. Find range of KI for which the closed
loop system is stable.
Home work Assignment :

Draw the bode plot and obtain the gain and phase margins for the following OLTFs:
−125 ∗ (s−1) −125 ∗ (s+1)
1. GH = (s+1) (s+5) (s+25) 2. GH = (s−1) (s+5) (s+25)
−25 ∗ (s−5) −25 ∗ (s+5)
3. GH = (s+1) (s+5) (s+25) 4. GH = (s+1) (s−5) (s+25)
250 ∗ (s+1) 25 ∗ (s+10)
5. GH = (s+2) (s+5) (s+25) 6. GH = (s+2) (s+5) (s+25)
10 0.01 ∗ (s2 +0.01s+1)
7. GH = 8. GH =
s (s2 +0.5 s+5) s2 [(s2 /4)+0.02 (s/2)+1]
2.5.2 Nyquist stability criterion
A stability test for time invariant linear systems can also be derived in the frequency domain
using Nyquist stability criterion. It is based on the complex analysis result known as Cauchys
principle of argument. Note that the system transfer function is a complex function. By ap-
plying Cauchys principle of argument to the open-loop system transfer function, it is possible
to get information about stability of the closed-loop system transfer function and arrive at the
Nyquist stability criterion (Nyquist, 1932). The importance of Nyquist stability lies in the fact
that it can also be used to determine the relative degree of system stability by producing the
so-called phase and gain stability margins. These stability margins are needed for frequency
domain controller design techniques. Once again recall that Nyquist stability criterion also
determines the stability of the closed loop system (determined by the characteristic equation,
∆(s) = 1 + GH = 0) by working on the open loop transfer function. Roots (or zeros) of the
characteristic equations are poles of the closed loop system. At the same time the poles of
∆(s) are the open-loop control system poles since they are contributed by the poles of (G H),
which can be considered as the open-loop control system transfer functionobtained when the
feedback loop is open at some point. The Nyquist stability test is obtained by applying the
Cauchy principle of argument to the complex function.
2.5. STABILITY 95
Cauchys Principle of Argument

Let F(s) be an analytic function in a closed region of the complex plane given in Figure 2.33
except at a finite number of points (namely, the poles of F(s)). It is also assumed that F(s) is
analytic at every point on the contour. Then, as ‘s’ travels around the contour in the s - plane
in the clockwise direction, the function encircles the origin in the (ℜ{F(s)}, ℑ{F(s)})-plane
[(Re{F(s)}, Im{F(s)})-plane] in the same direction times (see Figure 2.33), with N given
by N = Z − P , where Z and P stand for the number of zeros and poles (including their
multiplicities) of the function F(s) inside the contour. The above result can be also written
as arg{F(s)} = (Z − P ) 2π = 2 π N which justifies “the principle of argument”.
Figure 2.33: Cauchys principle of argument
Nyquist Plot
The Nyquist plot is a polar plot of the characteristic polynomial function ∆(s) = 1+GH when
travels around the contour given in Figure 2.34. The contour in this figure covers the whole
unstable half plane of the complex plane s, R → ∞. Since the function ∆(s), according to
Cauchys principle of argument, must be analytic at every point on the contour, the poles of
∆(s) on the imaginary axis must be encircled by infinitesimally small semicircles.
Nyquist Stability Criterion states that the number of unstable closed-loop poles is equal to
the number of unstable open-loop poles plus the number of encirclements of the origin of
the Nyquist plot of the complex function ∆(s). This can be justified by applying Cauchys
principle of argument to the function with the s-plane contour given in Figure 2.34. Note that
Figure 2.34: Nyquist contour in the s-plane; ∆(s) = 1 + GH(s)
Z and P represent the numbers of zeros and poles, respectively, of ∆(s) in the unstable part
of the complex plane. At the same time, the zeros of ∆(s) are the closed-loop system poles,
and the poles of ∆(s) are the open-loop system poles (closed-loop zeros).
The above criterion can be slightly simplified if instead of plotting the function ∆(s) = 1+GH,
plot only the function (OLTF) (G H) and count encirclement of the Nyquist plot of around
the point (−1, j 0), so that the modified Nyquist criterion has the following form :
ZR = no. of zeros of (1 + G H) in RH s-plane
N = no. of clockwise encirclements
PR = no. of poles of G H in the RH s-plane.
Z = N + PR
If PR ̸= 0, for stable system Z = 0 ⇒ N = −PR , which means no. of counter clockwise
encirclement of (−1, 0) is PR .
Statement of Nyquist stability criterion:
Three possibilities exist in examining stability of linear system
a There is no counter clockwise encirclement of (-1, 0) point. The system is stable if there
are no RH s-plane poles of GH. Otherwise unstable.
b There is a counter clockwise encirclement of (-1,0) point stable if no of encirclement= no.

of RH s-plane poles of GH(s).
c There exists clockwise encirclement of (-1,0) point. The system is unstable.

2.5. STABILITY 97
Institutively speaking, as ω changes from 0+ rad/sec to ∞, system with single pole, introduces
a maximum of 90◦ phase lag,while for two poles, delay is 180◦ . Hence, as ω traverses from
− ∞ to ∞, the Nyquist plot has one encirclement. For addition of every two poles, one will
have one additional encirclement.
Procedure:construct the following Table. Plot the phasers as ω is varied from ∞ to 0
Real GH Im GH |GH| ̸ GH Due to symmetry,

s= j ω1 – – s= j ω1 – – it is sufficient to
OR
s=jω2 – – s=j ω2 – – study Nyquist plots
..
. – – ... – – for ω ∈ [0+ , ∞)
Note: as s → j∞ , both Re(GH), Im(GH) → 0 (origin of (G H) ) for causal systems.

Example 1: G H = k/(1 + sT1 )(1 + sT2 ) ⇒ Unconditionally stable system.
Phase margin (PM) and Gain margin (GM): The phase and gain stability margins
Figure 2.35: Phase and gain stability margins depicted in Nyquist plot in (G H) plane
are presented in Fig. 2.35. They give the degree of relative stability; in other words, they tell
how far the given system is from the instability region. Their formal definitions are given by
Phase margin (Pm) γ = 180◦ + arg{G(j ωcg ) H(j ωcg )}
Gain Margin in dB] = 20 log |G(j ω )1H(j ω )| in dB
cp cp
where ωcg = ωg and ωcp = ωp stand for, respectively, the gain and phase crossover frequencies,
which from Figure 2.35 are obtained as

|G(j ωcg ) H(j ωcg | = 1 =⇒ @ ω = ωcg
and arg{G(j ωcp ) H(j ωcp )} = 180◦ =⇒ @ ω = ωcp
1
Example 2: Consider an OLTF represented by G(s) H(s) = s (s+1) .
Since this system has a pole at the origin, the contour in the s-plane should encircle it with
a semicircle of an infinitesimally small radius. This contour has three parts (a), (b), and (c).
Mappings for each of them are considered below.
(a) On this semicircle, the complex variable is represented in the polar form by s = R e(j ψ)
with R → 0; − π2 < ψ < π
2
. Substituting s = R e(j ψ) into G(s) H(s), it can be seen that
G(s) H(s) → 0. Thus, the huge semicircle from the s-plane maps into the origin in the
G(s) H(s)-plane (see Figure 2.36).
Figure 2.36: Nyquist contour plot in the s-plane for Example 2
(b) On this semicircle the complex variable is represented in the polar form by s = r e(j ϕ)
with R → 0; − π2 < ϕ < π
2
, so that we have G(s) H(s) → 1
r e(j ϕ)
→ − ∞ arg(− ϕ). Since
changes from − π2 at point A to π
2
at point B, arg{G(s) H(s)} will change from π
2
to − π2 .
It can be concluded that the infinitesimally small semicircle at the origin in the s-plane is
mapped into a semicircle of infinite radius in the G(s) H(s)-plane.
(c) On this part of the contour s takes pure imaginary values, i.e. s = j ω with ω changing from
to −∞ to ∞. Due to symmetry, it is sufficient to study only mapping along 0+ ≤ ω ≤ ∞ .
The real and imaginary parts of the function G(j ω) H(j ω) are given by
2.5. STABILITY 99
Re(GH)(j ω) = ω−1 2 +1 , Im(GH)(j ω) = ω (ω−1 2 +1)
From these expressions it can be seen that neither the real nor the imaginary parts can be made
zero for finite value of ω, and hence the Nyquist plot has no points of intersection with the
negative real coordinate axis (negative GH - axis. Point B corresponds to ω = 0+ , and since
the plot at ω = +∞ will end up at the origin, the Nyquist diagram in GH-plane corresponding
to part (c) has the form as shown in Figure 2.36. Note that the vertical asymptote of the
Nyquist plot in Figure 2.36 is given by Re{G(j 0± ) H(j 0± )} = −1, since at those points
Im{G(j 0± ) H(j 0± )} = ± ∞.
From the Nyquist diagram, note that N = 0. Since there are no open-loop poles
in the right half of the complex plane, PR = 0, then, ZR = 0, so that the corresponding
closed-loop system has no unstable poles.
The Nyquist plot is drawn by using the MATLAB function ‘nyquist’
num=1; den=[1 1 0];
nyquist(num,den);
axis([-1.5 0.5 10 10]);
axis([-1.2 0.2 1 1]);
The MATLAB Nyquist plot is presented in Figure 2.37. It can be seen from Figures 2.35 and
2.37 that 1/Kg = 1/Gm = 0, which implies that Gm = ∞. Also, from the same figures it
follows that phase cross-over frequency ωp = ωcp = 0. In order to find the phase margin and
the corresponding gain crossover frequency, use the MATLAB function ‘margin’ as follows
[Gm, P m, wcp, wcg] = margin(num, den)
producing, respectively, gain margin, phase margin, phase crossover frequency, and gain
crossover frequency. The required phase margin and gain crossover frequency are obtained as
P m = 53.41◦ and 0.7862rad/sec.
Example 3:
k
GH =
s(1 + sT1 )(1 + sT3 )
system is unstable for gain k > kcritical (which depends on T1 , T2 )

A feedback controller with good transient properties need not have good feedback properties and
vice versa.
Nyquist Diagram
10
2
Imaginary Axis
−2
−4
−6
−8
−10
−1.5 −1 −0.5 0 0.5
Real Axis
Figure 2.37: MATLAB Nyquist plot for Example 2
Example 4:
K
GH =
[s (s + 1) (s + 5)]
Desirable : PM ε [30 to 60◦ ] and GM ε [ 6 to 10 db] for reasonable good stability & performance.
For K = 10, PM = 21◦ , GM = 8 db (stable); For K = 100, PM = -30◦ , GM = - 12 db (unstable)
Note that, Good PM & GM always do not ensure good performance.
Example 5:
0.486
GH = ; Mp = 1.34 at ωr = 1.2
s2
+ 0.3s + 1
P M = 45◦ , GM ≈ ∞.
Roots of Characteristic equation : |1 + GH| = 0 are − 0.15 ± j1.21

-under damped system, highly oscillatory response; hence, PM & GM will not always ensure
good transient stability.
2.5. STABILITY 101
2.5.3 Root Locus Analysis
Root locus analysis is used study trajectory closed loop poles in complex s-plane by using
the OLTF, which is available in factored form (simple multiplication or cascade of first and
second order systems). Characteristic equation 1 + KG H = 0 can be suitably parameterized
into 1 + Ks Geq = 0, where Ks is any parameter of interest (not necessarily gain in all the
cases) which changes the closed loop poles. Consider a typical feedback control system in
K G(s)
Fig. 2.27, the closed loop transfer function Gc (s) = [ 1+K G(s) H(s) ] and the characteristic
N (s)
equation is [1 + K G(s) H(s)] = 0. Let G(s) = D(s) and the text book form of PD-feedback
control H(s) = (Kp + Kd s); so that the characteristic equation is D(s) + N (s) ∗ (Kp + Kd s) =
0 ⇒ [D(s) + N (s)Kp ] + [N (s)Kd s] = 0 ⇒ 1 + Kd { [D(s)+N (s)K ] = 0}. Now Kd is the
s N (s)
p
parameter that influences the closed loop characteristic equation for which the equivalent Geq
= { [D(s)+N (s)K ] } for the parameterized root locus analysis. In the root locus analysis, the
s N (s)
p
poles and zeros of complex OLTF are marked in the complex s = σ+jω-plane or (σ, jω)-plane.
In general, Let G(s) = K Np (s)/Dp (s), &H(s) = Nc (s)/Dc (s) so that Gc (s) =
K Np (s) Dc (s)
Dp (s) Dc (s)+K Np (s) Nc (s) . Zeros of Gc (s) are roots of [Np (s) & Dc (s)]. Poles of Gc (s) has
to be computed for each value of gain term K, which is cumbersome. Root locus anal-
ysis describes locus of closed loop poles as the term K is varied from 0 to ∞ (from 0
to −∞ for a complementary root locus). The characteristic equation can be recast into
K N (s) N (s)
{Dp (s) Dc (s) + K Np (s) Nc (s)} = 0. The open loop transfer function (OLTF) is D (s)p c
Dc (s) .
p
When the gain (like) term K = 0, then {Dp (s) Dc (s) + K Np (s) Nc (s)} = 0. Hence closed loop
poles are same as that of OLTF (or an equivalent TF). On rewriting characteristic equation,
{[Dp (s) Dc (s)]/ K + Np (s) Nc (s)} = 0. As K → ∞ ⇒ Np (s) Nc (s)} = 0, that is, closed
loop poles terminate at open loop zeros (including those at infinity or infinite horizon; n = m
zeros at finite domain and (n - m) zeros at infinite horizon) ‡.
Any complex variable α + jβ can be represented on a complex Cartesian s-plane or
(X = σ, Y = jω)-plane as a point. For example, H(s) = s + a ⇒ H = (σ + a) + jω, again
a sum of real and imaginary parts. A vector can be defined between any two points in this
complex plane. For example, a vector can be defined between a root of a function and a test
point in s-plane. This concept can be extended to a complex function like a transfer function.
∏m ∏
(s+z ) numerator′ s complex f actors
Let G(s) = ∏ni=1 (s+pi ) = denintor′ s complex f actors .
i=1 i
Sketching of root locus

The following properties help Sketching of root locus with minimal calculations.
Property 1: Number of branches. Each closed-loop pole moves as the gain is varied. There-
fore, number of separate loci (or branches of the root locus) is equal to the number of poles.
Property 2: Symmetry about the real axis. Since the complex poles always appear in complex
conjugate pairs, the root locus must be symmetrical about the real axis.
Property 3: Location of the real-axis segment of the root locus. The root locus on the real
axis always lies in a section of the real axis to the left of an odd number of (open-loop) poles
and zeros. To prove this property, consider figure 1, where poles and zeros location for a
general open-loop system is shown. One can note the following facts.
• The angular contribution of a pair of complex conjugate poles (zeros) is zero. Since complex
poles always appear in complex conjugate pairs, we see that for a respective point on
the real axis the angular contribution of complex poles (zeros) is zero.
• The angular contribution of real poles (zeros) that are to the left of the respective point is
zero.
• Contribution of each real pole (zero) to the right of the respective point is 180◦ .
Combining these three facts, one gets property 3.

Property 4: Starting points and ending points. To determine starting and ending points of
root locus, recall discussion given above: root locus begins at the poles of G(s)H(s) and ends
at the zeros of G(s)H(s) as K increases from 0 to +∞ (inclusive of zeros at ∞ - horizon) ‡.
Property 5: Location of infinite zeros on asymptotes. A function F (s) has a zero (pole) at
infinity if F (s) → 0 (F (s) → ∞) as s → ∞. Every function has an equal number of zeros
and poles if the infinite poles and zeros as well as finite poles and zeros are included. Consider,
for example, a function
K
KG(s)H(s) = . (2.45)
s(s + 1)(s + 2)
This function has three finite poles at p1 = 0, p2 = −1, p3 = −2, and no (finite)
K
zeros. However, if p approaches infinity, the function becomes KG(s)H(s) p.p.p . Each p in
the denominator causes the function to become zero as p approaches infinity. Therefore, the
function has three infinite zeros. Thus, the root locus for equation 2.45 will begin at finite
2.5. STABILITY 103
poles and end in infinite zeros. Where are the infinite zeros located? In general, if a function
has np finite poles and nz finite zeros (np ≥ nz ), then N = (np − nz ) sections (branches) of the
root locus will end at infinite zeros. The following rule helps us to locate the infinite zeros.
The branches that end at infinite zeros approach the zeros along linear asymptotes. These
asymptotes are centered at the point on the real axis given by
∑np ∑
j=0 f inite poles − ni=0z
f inite zeros
σ0 = , (2.46)
(np − nz )
and their angles are
(2k − 1) 180◦
θk = ; k = 1, 2, . . . , (np − nz ). (2.47)
(np − nz )
Property 6: Location of the real-axis breakaway and break-in points. A point on the real axis
is called breakaway point if the root locus departs from the real axis at this point. On the
other hand, a point on the real axis is called break-in point if the root locus arrive to the real
axis at this point. From the symmetry property it follows that the root loci at the breakaway
(break-in) point are symmetrical with respect to the real axis. Also, the following rule is valid.
• The tangents to the loci at a breakaway (break-in) point are equally spaced over 360◦ .
Therefore, the two loci at the breakaway point are spaced 180◦ apart, the four loci are spaced
90◦ apart, etc.
How to find the breakaway (break-in) points? Suppose we have two real axis poles which
move towards each other as gain increases. One can conclude that the gain must be maximal
at the point where breakaway occurs. Thus, the breakaway point is the point of maximum
gain between two open-loop real-axis poles. Analogously, when the complex pair returns to
the real axis, the gain will continue to increase as the closed-loop poles move toward the
open-loop zeros. Therefore, one can conclude that the break-in point is the point of minimum
gain between two real-axis zeros. These considerations allow us to use the following method
to find breakaway and break-in points. As is already know, for any points λ of the root locus,
the following equation
1
K=− (2.48)
G(s = λ) H(s = λ)
is valid. On the real axis λ is real, therefore, H(s) and G(s) are real-valued function. To find
points of maximum and minimum of K one can simply differentiate the equation (2.48) with
respect to λ and set the derivative equal to zero. Consider the following example:
RL Sketch Example 1. Consider a unity negative feedback system with the following open-
K(s−3)(s−5)
loop transfer function KG(s)H(s) = (s+1)(s+2)
, sketch the root locus.
Solution : The system has two poles at p1 = −1, p2 = −2, and two finite zeros at z1 = 3
and z2 = 5.
Root Locus
1
Imaginary Axis
−1
−2
−3
−3 −2 −1 0 1 2 3 4 5 6
Real Axis
Figure 2.38: MATLAB Root locus plot for RL Sketch Example 1
The root locus has two branches that start from the poles, end in the zeros, and
symmetrical with respect to real line. Using property 3, the first real-axis segment of the root
locus is located between -2 and -1, and the second one is located between +3 and +5. For
the points on the root locus, K(s−3)(s−5)
(s+1)(s+2)
= −1. Solving for K yields
−(s +3s+2)
2
K = (s2 −8s+15) .
In order to calculate breakaway and break-in points, differentiate K with respect to s and
setting the derivative equal to zero, to get
dK 11s2 −26s−61
ds = (s2 −8s+15)2 = 0.
Solving for s, roots are s1 = −1.45, s2 = 3.82. Clearly, s1 is the breakaway point, while s2 is
the break-in point. The root locus computed using Matlab is shown in figure 2.38. Another
method to find the breakaway and break-in points is by using the following rule:
• Breakaway and break-in points satisfy the following relationships
2.5. STABILITY 105
∑m 1 ∑n 1
i=1 s−zi = j=1 s−pi ,
1 1 1 1
where zi , and pj are zeros and poles, respectively, of H(s) G(s). Hence, s−3
+ s−5 = s+1
+ s+2 .
On simplifying, 11s2 − 26s − 61 = 0, and solving for s, Breakaway and break-in points are
s1 = −1.45, s2 = 3.82.
Property 7: Points of Imaginary/ jω-axis crossings. The points where the root locus cross
the jω-axis are important because they separate stable parts of root locus from unstable ones.
To find the jω-axis crossing , we can use the Routh-Hurwitz criterion as follows: forcing a
row of zeros in the Routh table will yield the gain: going back one row to the even polynomial
equation and solving for the roots (if possible) yields the points of the imaginary axis crossing.
RL Sketch Example 2. Consider a unity negative feedback system with the following open-
loop transfer function
K(s+3)
K G(s) H(s) = s(s+1)(s+2)(s+4)
.
Sketch the root locus and find the range of K such that the closed-loop system is stable.
Solution : The open-loop system has four poles p1 = 0, p2 = −1, p3 = −2, p4 = −4, and
one finite zero z1 = −3, therefore, there are three infinite zeros. Using property 3, there are
three real axis segments of the root locus: first between 0 and -1, second between -2 and -3,
and the third one is to the left of -4. To find the location of infinite zeros, the asymptotes are
calculated using formulas (2.46), (2.47). The asymptotes are centered at the point
∑ ∑
np f inite poles− nz f inite zeros
σ0 = j=0 i=0
(np −nz )
= (−1−2−4)−(−3)
4−1
= − 43
The angles of asymptotes that intersect at − 43 are
(2k−1) 180◦
θk = (np −nz )
; k = 1, 2, . . . , (np − nz ).
i.e., θ1 = 60◦ , θ2 = 180◦ , θ3 = 300◦ . A breakaway point must exist between the poles -1 and
0, and the corresponding branches tends to infinite zeros along the asymptotes with angles
θ1 and θ3 . To, calculate the points where the root locus crosses the imaginary axis; first, one
can find the transfer function of the closed loop system as follows:
K(s + 3)
Gc (s) = . (2.49)
s4 + 7s3 + 14s2 + (8 + K)s + 3K
The Routh table for the system (2.49) is given in the table 2.2.
It is known that a complete row of zeros yields the possibility for imaginary axis
roots. For this system, only the s1 row can yield a row of zeros. This happens if −K 2 − 65K +
720 = 0. Solving for K, one gets K = 9.65. Going one row back, one can form the following
s4 1 14 3K
s3 7 8+K
s2 90 − K 21K
−K 2 −65K+720
s1 90−K
s0 21K
Table 2.2: Routh table for Example RL sketch
even polynomial (90 − K)s2 + 21K = 0, and for K = 9.65, one gets 80.35s2 + 202.65 = 0.
Solving for s, one gets s1,2 = ±j 1.59. Thus, the root locus crosses the imaginary axis at
± j 1.59 at a gain K = 9.65. The system is stable for 0 < K < 9.65. The root locus is
sketched using the Matlab in figure 2.39.
Root Locus
1.5
0.5
Imaginary Axis
−0.5
−1
−1.5
−2
−4.5 −4 −3.5 −3 −2.5 −2 −1.5 −1 −0.5 0 0.5 1
Real Axis
Property 8: Angles of Departure and Arrival. One can find angles of departure from the
complex poles as well as angles of arrival to the complex zeros as follows. Consider figure
2.40, which shows the open-loop poles and zeros, some of them are complex. Take a point ϵ
of the root locus close to a complex pole, then the sum of angles drawn from all finite poles
and zeros to this pole is equal to an odd multiple of 180◦ . Assume that all angles from all
other poles and zeros are drawn directly to the pole that is near the point. The only unknown
angle is the angle close drawn from the pole that is close to ϵ. We can solve for this unknown
angle, which is actually the angle of departure from this complex pole. For example, for
the pole-zero plot in figure 2.40, −θ1 + θ2 + θ3 − θ4 − θ5 + θ6 = (2k + 1)180◦ , which implies
2.5. STABILITY 107
θ1 = θ2 + θ3 − θ4 − θ5 + θ6 − (2k + 1)180◦ . Angle of arrival to a complex zero can be calculated

analogously. For example (see figure 2.41) θ2 = θ1 − θ3 + θ4 + θ5 − θ6 − (2k + 1)180◦ .
Figure 2.40: MATLAB Root locus plot for angle of departure in RL Sketch
RL Sketch Simple Examples: Sketching the root locus for the following open loop transfer
functions:
K K (s + 3)
a) G H(s) = ; b) G H(s) = ;
s (s + 2) s (s + 2)
K K (s + 3) (s + 4)
c) G H(s) = d) G H(s) =
s (s + 2) (s + 3) s (s + 2)
Example a) contains two poles at origin and -2 and no finite zeros. Example b) has additional
zero at -3 on the other hand, Example c) contains additional pole at -3. Example d) has two
additional zeros at -3 and -4. Hence these three systems illustrate the effect of additional pole
or zero on the root locus, example d) shows the effect of two zeros. The matlab generated
root locus are plotted in the Figure 2.42.
For the Example a), number of poles n = 2; number of zeros m = 0. Two asymptotes intersects
real axis at -1 with subtended angle of ± 90◦ . The system is unconditionally stable. As
amplifier gain increases monotonically to large values, position feedback term dominates over
Figure 2.41: MATLAB Root locus plot for angle of arrival in RL Sketch
the damping term, hence it results increase in natural frequency and reduction in damping.
Example b) exemplifies the effect of a zero to the left of farthest pole. Additional zero
increases the damping, hence results in a highly damped system. For this case, n = 2 and
m = 1. There is only one asymptote that aligned along negative real axis of s-plane. The
numerator polynomial N (s) = (s + 3) and the denominator polynomial is D(s) = s (s + 2).
The break in and break away points are obtained by solving the Equation (Note: D′ (s) =
d/ds[D(s)]; N ′ (s) = d/ds[N (s)])
D′ (s) N (s) − N ′ (s) D(s) = 0
(2 s + 2) (s + 3) − 1 (s2 + 2s) = 0 ⇒ s2 + 6 s + 6 = 0, ⇒ (s + 4.73) (s + 1.27) = 0.
Hence break away and break in points are -1.27 and -4.73 respectively with gain K = 0.536
and 7.46 respectively. For the example c), n = 3 and m = 0, number of asymptotes = 3.
Angle of asymptotes are 60◦ , 180◦ , 300◦ and asymptotes intersect the real axis at -5/3 =
-1.667. Additional pole at -3 reduces the gain margin and makes the system conditionally
stable. The break away point is -0.785 with a gain K = 2.11. The characteristic equation is
given by s3 + 5 s2 + 6 s + K = 0, Therefore, critical gain K = 30 for the system on verge of
instability. At the imaginary axis cross over point, s3 + 5 s2 + 6 s + 30 = 0. On substituting
2.5. STABILITY 109
√
s = j ω and solving for ω gives the imaginary axis cross over frequency at ± j 6 = ± j 2.4495
. Similarly, root locus for example d) is drawn using matlab at the bottom right corner.
Root Locus Root Locus
1.5 System: sys System: sys

Gain: 7.46 Gain: 0.536
1 Pole: −4.73 Pole: −1.27 + 0.0288i
Damping: 1 Damping: 1
Imaginary Axis
0.5 Overshoot (%): 0 Overshoot (%): 0
Frequency (rad/sec): 1.27
0
−0.5
−1
−1.5
−2
−2 −1.5 −1 −0.5 0 0.5 1 −10 −5 0 5
Real Axis Real Axis
1.5
Imaginary Axis
0.5
0
System: sys System: sys System: sys
Gain: 2.11 −0.5 Gain: 19.7 Gain: 0.202
Pole: −0.785 Pole: −3.38 + 0.0321i Pole: −1.43 + 1.01e−008i
Damping: 1 −1 Damping: 1 Damping: 1
Overshoot (%): 0 Overshoot (%): 0 Overshoot (%): 0
Frequency (rad/sec): 0.785 −1.5 Frequency (rad/sec): 3.38
−2
−5 0 5 −6 −5 −4 −3 −2 −1 0 1 2
Real Axis Real Axis
Figure 2.42: MATLAB Root locus plot for four simple examples
RL Sketch Example 3: Sketching the root locus. Problem. Find the root locus for a unity
negative feedback system with the following open-loop transfer function
K (s+4)
G(s) = (s+0.5)2 (s+2)
.
Solution. The root locus is plotted in figure 2.43 using Matlab.
Chapter 1 contains many examples of RL analysis from aerospace applications.
RL Sketch Example 4: Hydraulically actuate gun turret. The transfer function between the
input u and the azimuth angle θ(s) along with controller gain K is given by ([1], p.148):
100980 K
G H(s) =
s (s3 + 140.2s2 + 10449s + 1000980)
Solution: The four open loop poles (P = 4)are p1 = 0; p2 = −11.2, p3, 4 = −64.5 ± 69.6.
Number of zeros = nil ⇒ N = 0. Root locus on the real axis: [-11.2, 0]
Number of asymptotes = (P - N) = 4. Angle of asymptotes: 45◦ , 135◦ , 225◦ and 315◦ .
and asymptotes are centered around (-35, 0) on the real σ-axis. The breakaway point is
Root Locus
1
Imaginary Axis
−1
−2
−3
−4.5 −4 −3.5 −3 −2.5 −2 −1.5 −1 −0.5 0 0.5

Real Axis
Root Locus
60
System: sys
Gain: 69.3
Pole: 0.0119 + 26.8i
40 Damping: −0.000444
Overshoot (%): 100
20 System: sys
Imaginary Axis
Gain: 2.59
Pole: −5.41 − 1.03e−007i
Damping: 1
Overshoot (%): 0
0
−20
−40
−60
−70 −60 −50 −40 −30 −20 −10 0 10

Real Axis
Figure 2.44: MATLAB Root locus plot for RL Example 4
at (-5.41, 0) and the corresponding gain K = 2.59. The imaginary axis cross-over point is
ω = 26.8 j and corresponding critical gain Kcritical = 69.3. Hence the closed system is stable
for 0 < K < 69.3. Root locus is plotted using matlab in the Figure 2.44.
2.5. STABILITY 111
Exercise: Hand calculate all the above values and also angle of departure at −64.5 ± 69.6j.
RL Sketch Example 5: Missile Pitch dynamics and control. The transfer function between the
input δ and the lateral/ normal acceleration az (s) along with actuator dynamics plus controller
gain K is given by ([1], p.154):
1 Zδ s2 + (Zα Mδ − Zδ Mα ) −1115 K (s2 − 2228)

G H(s) = =
(τ s + 1) (s2 + (Zα /v) s − Mα ) (0.01 s + 1) (s2 + 3.33 s + 248)
Solution: The three open loop poles (P = 3)are p1 = −100; p2, 3 = −1.665 ± 15.66.
Root Locus
25
20
15
10 System: sys
Gain: 9.68e−005
Pole: −2.64 + 0.0615i
5 Damping: 1
Overshoot (%): 0
0
System: sys
Gain: 9.98e−005
−5 Pole: −0.0117
Damping: 1
Overshoot (%): 0
−10
−15
−20
−25
−100 −80 −60 −40 −20 0
Real Axis
Figure 2.45: MATLAB Root locus plot for RL Example 5
Number of zeros = two ⇒ N = 2 and are located at ±47.202.

Root locus exists on the real axis between [−47.202, & 47.202] and (− ∞, −100].
Number of asymptotes = (P - N) = 1. Angle of asymptotes: −180◦ , i.e., asymptote is aligned
with the negative real axis of the s − plane.
The d.c. gain of G H(s) is positive, i.e., positive input produces a positive acceleration
response. But for high frequencies, G H(s) → −1115/(0.01s + 1) which produces a nega-
tive response for a positive input. The change in sign in the transfer function as the fre-
quency is increased is another consequence of the right half plane (non-minimum phase)
zeros [1, p.154]. Complementary Root locus K ∈ (−∞, 0] is plotted using matlab in
the Figure 2.45. From Fig. 2.45, it is observed that as the gain is increased from 0 to
−∞, one branch of root locus crosses the imaginary axis at s = 0 and then continues to-
wards the real root at s = −47.22. Define K = 111500 K. Then the characteristic equa-
tion: s3 + (100.33 + K) s2 + 581 s + (24800 − 222 K) = 0. The coefficient of s0 vanishes at
K= 24899
2228
= 11.13 or K = 11.13/111500 = 9.9821e−005 .
The break-in point lies between two consecutive zeros at [−47.202, 47.202] and it is located
at (-2.64, 0) and the corresponding gain K = 2.59. Hence the closed system is stable for
K ≤ 9.9821e−005 or K ≤ 11.13.
2.5.4 Asymptotic Stability
In the present context, the terminology such as measures/ norm/ distance/ metric are inter-
changeably used. A norm is an abstraction of the concept of length. For example, Euclidean
norm/ distance of a state −
x from the equilibrium state −
x is defined as d, where
e
d2 = (x1 − x1e )2 + (x2 − x2e )2 + . . . + (xn − xne )2
On the other hand, weighted norm of −

x from null equilibrium is defined as
||x||2Q = q1 x21 + q2 x22 + .... + qn x2n , xe = 0,
For Euclidean norm, Q = I = Identity. All acceptable norms satisfy four important properties:
1) triangular inequality,
2) positive definite property
−) −→ ∞ and ρ(x
3) ρ(x x −→ 0,
−) −→ 0 as −
x ) = |k| ρ (x
4) ρ (k − −) .
Formal definition for the asymptotic stability due to Liapunov is defined as follows:
Let −
x be the equilibrium state of the system − ẋ (t) = f (x x (to ) ̸= 0. Then, −
−(t), t), − x is said to
e e
be stable equilibrium if for a given any real number ϵ > 0, there exists a real number δ(ϵ, to ) > 0
x (to ) in the sphere of radius δ, i.e., ||x
such that for all initial state − −(to ) − −
x || < δ, the solution
e
x (t) remains with in a sphere of radius ϵ for all t ≥ to i.e., ||x
− −(t) − − x || < ϵ.
e
The stability in the sense of Liapunov states that if an equilibrium state is stable, then every
trajectory starting in the neighborhood of xe must remain close to equilibrium state xe . For
nonlinear systems, multiple equilibria exists. For example, for a pendulum, θ̈+ {g/l} sinθ = 0,
√
the two equilibria are θ = 0 (regular pendulum, ω = g/l ) and θ = π (inverted). For a linear
system, xe = 0 (null equilibrium) is the only equilibrium point.
2.5. STABILITY 113
Asymptotic stability : The system is stable in the sense of Liapunov and x(t) converges to xe
as t −→ ∞. (Note: under bar . is omitted hence forth)
Stability in large: Other stability concept.
Any positive definite functional V(x(t)) can be called Liapunov functional if

V(0) = 0, V(x) −→ ∞ as x −→ ∞ and satisfies triangular inequality. Note
that the Liapunov functional V(x) serves as a suitable norm.
Stability in the sense of Liapunov:
Consider a system ẋ(t) = f (x(t), t), x0 ̸= 0. Suppose there exists a Liapunov functional V(x)
such that V̇ (x) is negative definite, then the corresponding equilibrium state xe is asymptot-
ically stable. If V̇ (x) is negative semi-definite, then the corresponding equilibrium state xe is
stable.
Example 4:
    

 x˙1 
 0 1   x1 
= 
   =⇒ stable system

 x˙ 
2 −1 −1 x2
1)Let V (x) = x21 + x22 = xT x −→ V̇ (x) = −2x22 =⇒ stable
2)V (x) = 2x21 + x22 =⇒ V̇ (x) = 2x1 x2 − 2x22 =⇒ sign indef inite,
→ therefore, V(x) is not a suitable Liapunov functional

 
1 3 1 
3)V (x) = [(x1 + x2 )2 + 2x21 + x22 ] = (3x21 + 2x1 x2 + 2x22 ) = xT 
 x
2 1 2
=⇒ V̇ (x) = −(x21 + x22 ) =⇒ stable.
Example 5:
ẋ = Ax, xo ̸= 0 : V (x) = xT P x,
V̇ (x) = ẋT P x + xT P ẋ = xT (AT P + P A)x = −xT Qx
therefore, Stability =⇒ AT P + P A = −Q, where P and Q are symmetric positive definite.

Start with Q = I, determine P from Liapunov equation and check the sign definiteness of P.
If P is sign indefinite, alter Q.
=⇒ Matrix P is the positive definite if xT P x > 0 for all x ϵ [−∞, ∞]
=⇒ All eigenvalues (real part) of P are greater than 0
=⇒ Use Sylvester’s criteria to check sign definiteness.
Note that Sylvesters criterion is a necessary and sufficient criterion to determine

whether a real symmetric matrix is positive-definite. It is named after James Joseph Sylvester.
The Sylvester criteria can be extended to a complex valued Hermitian matrix.
Sylvester’s criterion states that a Symmetric matrix P is positive-definite if and
only if all the following matrices have a positive determinant:
the upper left 1-by-1 corner of P,
...
P itself.
In other words, all of the leading principal minors must be positive. If a given
(X+X T )
matrix X is not symmetric, then symmetric part of X, i.e., P = 2
should be used.
For a real symmetric matrix, all the eigenvalues are real. The eigenvectors of a symmetric
matrix are orthogonal. If the given matrix is positive definite, then all its eigen values are
positive. Hence, determinant of det|P | is positive.
In Example 4: Select Q = I, then, xT P x is same as third choice of V(x) since:
 
3 1
 2 2 
P = 
1
2
1
Hence, P11 = 1.5; det|P | = 1.25 > 0, satisfies Sylvester’s criterion. Eigenvalue of P =
(0.6910, 1.8090), both of these are also positive. Therefore, P is positive definite matrix.
Home work Assignment: Check stability in the sense of Liapunov for the LTIV System:
ẋ = Ax,where, 
 −0.0507 −3.8610 0 −32.185 
 
 
 
 −0.0012 −0.5164 1.0000 0  −0.5164 1.0000 
1) A = 


 2) A = 
 
 −0.0001 1.4168 −0.4932 0  1.4168 −0.4932
 
 
0 0 1.0 0
 
 −0.746 0.006 −0.999 0.0369   
 
 
 −12.9 −0.746 0.387 0   0.5 1.00 
3) A = 

,
 4) A =  
 4.31 0.024 −0.174 0  −0.4356 −1.4932
 
 
0 1 0 0
2.6. STATE SPACE REPRESENTATION OF A LTIV SYSTEM 115
2.6 State Space Representation of a LTIV system

Let state variable representation of a LTIV system be given by set of n-tuple (vector) differ-
ential equation:
ẋ = Ax + Bu : state equation; x(0) = x0 (2.50)
y = Cx + Du : M easurement equation. (2.51)
Illustrations:
Example 1: Consider a second order system:
ẍ(t) + 2ξωn ẋ(t) + ωn2 x(t) = ωn2 u(t), x(0) = 0 ẋ0 = 0
On defining x1 = x; x2 = ẋ = ẋ1 ; the state space model as a physical variable representation

is given by
      
 ẋ1   0 1   x1   0 
 =  +  u = Ax + Bu
ẋ2 − ωn2 − ξωn x2 ωn2
Example 2: Longitudinal motion of the F-16 Aircraft (Stability axis reference frame)
e = 300 psf, X =
[35, Example 4.4.1 on p.293]: Trim conditions are h = 0 f t, Dynamic pressure Q cg
0.35 c, αtrim = 0.03691 rad = 2.115 deg, VT (trim) = VT = 502f t/sec; θtrim = αtrim , γ = 0.
State variables are x = [vT , α, θ, q]T (Note: vT = ∆ VT ) and control input u = δe .
v˙T = Xv vT + Xα α − gθ + Xe δe
Zv Zα Ze
α̇ = vT + α+q+ δe
VT VT VT
q̇ = Mv vT + Mα α + Mq q + Me δe
Kinematics : θ̇ = q
which results in the following state space model (physical variable representation):
ẋ = Ax + Bu, x = [vT , α, θ, q]T

   
 Xv Xα −g 0   Xe 
   

 Zv /VT Zα /VT 0 1 


 Ze/VT


where, A = 

, B = 
 


 0 0 0 1   0 
   
   
Mv Mα 0 Mq Me
   
 −0.01931 −8.8157 −32.17 −0.575   0.1737

   
   
 −0.00025 −1.0189 0 0.90506   −0.00215 
A = 




B= 

 0 0 0 1.0   0 
   
   
0.000 0.82225 0 −1.0774 −1.7555
   
 q   0 0 0 1 
y =  = x
θ 0 0 1 0
Example 3: The state space representation for the lateral dynamics [1, p.45] is given by
xT = (β, p, r, ϕ), u = (δa , δr ) and
 
Yβ Yp
( YVr − 1) g/V   
 V V Ya Yr
   

0 
V V
 Lβ Lp Lr   
A= B = 
 La Lr


   
 N Np Nr 0   
 β 
  Na Nr
0 1 0 0
For a typical L1011 transport aircraft [1, p. 185]:
   
 −0.746 0.006 −0.999 0.0369   0.0012 0.0092 
   
  
 −12.9 −0.746 0.387 0   6.05 0.952 

A=

B = 
 


 4.31 0.024 −0.174 0   −0.416 −1.76 
   
   
0 1 0 0 0 0
For transport aircraft, the sensors are roll and yaw rate gyroscopes and side slip angle β.
Corresponding C can be easily written and D = null vector.
Example 4: Nonlinear dynamics of satellite in a near circular orbit can be linearized about
the reference circular orbital parameters to get the Clohessy - Wiltshire equation (Note: CW
equations are derived for modeling rendezvous dynamics between two satellites in neighboring
orbits) in terms of Multi-input multi-output perturbed state variable representation with the
following system matrices (r0 = orbital radius of the reference orbit, Ω0 = orbital angular
2.6. STATE SPACE REPRESENTATION OF A LTIV SYSTEM 117
velocity, state: x = [r, ṙ, θ, θ̇, ϕ, ϕ̇]T and control = input thrust, output y = [r, θ, ϕ]T ):
   

0 1 0 0 0 0  
0 0 0 
   
 3Ω2 0 0 2Ω20 0 0   1/m 0 0 
 0   
   
   
 0 0 0 1 0 
0   0 0 0 
A = 


 B=


   


0 −2Ω0 /r0 0 0 0 0 

 0

1/mr0 0 

   
 0 0 0 0 0 
1   0 0 0 
  
   
0 0 0 0 −Ω20 0 0 0 1/mr0
 
 1 0 0 0 0 0 
 
 
C =  0 0 1 0 0 0 
 
 
0 0 0 0 1 0
The two separating lines show that this sixth order equation splits into two decoupled subsets,
describing motion in equatorial plane and azimuthal plane.
Canonical Realizations
A state space representation is a mathematical model of a physical system as a set of input,

output and state variables. The state variable realization for a physical system is not unique.
For example, one can use either angle of attack α or vertical velocity component w as one of
the state variable in the longitudinal dynamics. Many a time, all the state variables may not
have physical meaning and they represent mathematical abstraction of the state. In principle,
one can think of states of realization rather than the states of the system. Different canonical
representations can be interchangeably used and it is possible shift from one representation
to other through similarity transformations. These one to one mappings do not alter internal
(intrinsic) properties as well as they do not alter input - output descriptions of a system.
Selection of a specific realization for a given system depends on its use (purpose), familiarity
of the user and ease of utilization (convenience - numerical / analytical) for a given application.
State variable x of realization 1 can be related to state variable z of realization 2 through a
nonsingular similarity mapping T such that x = T z. For the given realization1 with triplets
(A, B, C), state space equation for realization 2 is given by
ż = T −1 A T z + T −1 B u, y =CT z
Since, there are infinitely many nonsingular matrices, there are multitudes of realizations.
Such transformations can be judiciously chosen to give most convenient state space description
for a given system. Some of the important / standard realizations are (details of which will
be dealt with at a later stage):
1. Controller / Controllability/ phase variable canonical form or realization
2. Observer / Observability canonical form
3. Modal/ Jordan canonical form
4. Parallel and cascade realizations
5. Balanced realizations.
2.7 Controllability and Observability (LTIV System)

Controllability is the ability to move a system from any given state to any desired state in
finite time. Observability is the ability to observe all of the parameters or state variables in
the system from finite duration (or finite number) of measurements. Stability is often phrased
as the bounded response of the system to any bounded input. Any successful control system
will have and maintain all three of these properties. Uncertainty presents a challenge to the
control system engineer who tries to maintain these properties using limited information.
Controllability: The system described by ẋ(t) = Ax(t) + Bu(t) is said to be state controllable
at t = to if it is possible to construct an unconstrained control signal which will transfer
an initial state xo to any final state (normally, null equilibrium) in a finite time interval
to ≤ t ≤ tf . If every state is controllable, then the system is said to be completely controllable.
The solution to ẋ(t) = Ax + Bu, x(o) = xo is
∫ t
x(t) = eAt xo + eA(t−τ ) Bu(τ )dτ (2.52)
o
∫ t1
Let x(t1 ) = 0, then 0 = eAt1 xo + eA(t1 −τ ) Bu(τ )dτ
o
∫ −t
or xo = − e−Aτ Bu(τ )dτ
0
The above relation can be satisfied by unconstrained u(t), 0 ≥ t ≥ tf provided (using Cayley
- Hamilton theorem ) if rank QC = rank [B|AB| . . . An−m B]n×m(n−m+1) = n
for a single input system: rank QC = rank [b|Ab| . . . An−1 b]n×n = n
QC = controllability matrix.
2.7. CONTROLLABILITY AND OBSERVABILITY (LTIV SYSTEM) 119
Example 6 :
     
 1 0   0   0 0 
A= b =   =⇒ [b|Ab] =   =⇒ singular
1 1 1 1 1
Example 7:
     
0 1   0   0 1 
A=
 b =   =⇒ [b|Ab] =   =⇒ nonsingular
−2 −1 1 1 −1
System 1 is uncontrollable (not completely controllable), while system 2 is completely state

controllable.
Observability : The system ẋ(t) = Ax + Bu, y = C x is said to be completely observable if
every initial state xo can be determined from the observation of y(t) over a finite time interval.
∫ t
Solution/ output : y(t) = CeAt xo + C eA(t−τ ) Bu(τ )dτ. (2.53)
o
For the system to be completely observable, prove that
RankQo = rank[C T |AT C T | . . . |(AT )n−p C T ](n×p(n−p+1) = n.
Example 8:
     
[ ]
 1 1   0   0 1 
A =  , b =  , c = 1 0 [b, Ab] =   =⇒ controllable
−2 −1 1 1 −1
 
 1 1 
Qo = [cT |AT cT ] =   =⇒ observable
0 −2
Example 9:
 
 0 1 0 
 
 
A =  0 0 1  , c = [4, 5, 1]
 
 
−6 −11 −6
 
 4 −6 6 
 
 
Qo =  5 −7 5  =⇒ det Qo = 0, rank < 3.
 
 
1 −1 −1
system is not completely observable

Check for observability for c = [0, 0, 1] , c = [0, 1, 1], c = [1, 1, 1].
u s+2 x2 s-1 y=x1

A: s - 1 s+6
u s-1 x2 s+2 y = x1
B : s+6 s-1
Figure 2.46: Uncontrollable and Unobservable System due to Pole - zero cancellation.
Determine stability, controllability & observability of systems (Fig. 2.46). Comment.

For the bounded input bounded output (BIBO) stability, it is well known that
the plant transfer function must have all poles in the left half of s-plane (LHS-plane). The
asymptotic stability requirement also boils down to the requirement of all poles must lie in
the LHS - plane.
Therefore, if the system is asymptotically stable (internally stable), then it implies that it is
BIBO stable. If the system is completely controllable and observable, then BIBO stability
implies asymptotic stability. Cancelation of pole-zero pairs may sometime hide internal insta-
bility since then the system will loose either complete controllability or observability (as the
following two examples indicate).
The feedback system is called internally stable if each of the four transfer functions mapping
disturbance d and measurement noise η to control input u and system output y are stable.
The following example will further substantiate this statement:
d
r t y
K G
- controller Plant
Figure 2.47: Typical Feedback control system with series/ cascade compensator K(s).
Example 10: Consider

s−1 1
K(s) = ( ) and G(s) = 2
s+1 s −1
The closed loop transfer function (TF) from r to y is stable, but that from d to y is not (check
your self). This is due to cancellation of the controller zero and plant pole at s = 1. Therefore,
the system is not internally stable.
Example 11: Plant TF G(s) = (s/(s + 1)); and controller TF K(s) = (s + 3)/s produces
[ ]
stable sensitivity TF (1 + GK)−1 = s+1
s (s+2)
while mapping from d to y. However the closed
[ ]
loop TF from d to u is [K (1 + GK)−1 ] = (s+1) (s+3)
2 s (s+2)
, which is unstable due to pole at the
origin. Hence, the closed loop system is not internally stable, though it is input-output stable.
2.7.1 Passivity/ Dissipativity
In physics, dissipation characterizes the loss of energy over time in a dynamical system where
important mechanical modes, such as waves or oscillations, typically due to the action of
friction or turbulence.
Concept of passivity or dissipativity is extension of Liapunov theory to study BIBO
stability of linear or nonlinear systems. This energy/ norm based method also gives sufficient
conditions for stability.
In systems and control theory, dissipative systems are dynamical systems with a
state x(t), inputs u(t) and outputs y(t), such that there exist so-called storage functions
V (x, t) and supply rates w(u, y) such that: V (., .) is a nonnegative function, and for any time
dV (x(t),t)
t, one has dt
= v̇(x, t) less than u(t).y(t), where . is the scalar product. The physical
interpretation is that V(x) is the energy in the system, whereas u.y is the energy that is
supplied to the system. This notion has a strong connection with Lyapunov stability, where
the storage functions may play, under certain conditions of controllability and observability of
the dynamical system, the role of Lyapunov functions. Roughly speaking, dissipativity theory
is useful for the design of feedback control laws for linear and nonlinear systems.
In systems and control theory, dissipative systems are dynamical systems with
a statex(t), inputs u(t) and outputs y(t), such that there exist so-called storage functions
V (x, t) and supply rates w(u, y) such that: V (., .) is a nonnegative function, and for any time
t, one has dV (x(t), t)/dt less than [u(t).y(t)], where . is the scalar product. The physical
interpretation is that V (x) is the energy in the system, whereas u.y is the energy that is
supplied to the system. This notion has a strong connection with Lyapunov stability, where
the storage functions may play, under certain conditions of controllability and observability of
the dynamical system, the role of Lyapunov functions. Roughly speaking, dissipativity theory
is useful for the design of feedback control laws for linear and nonlinear systems. Dissipative
systems theory has been discussed by V.M. Popov, J.C. Willems [51], D.J. Hill and P. Moylan.
In the case of linear invariant systems, this is known as positive real transfer functions, and
a fundamental tool is the so-called Kalman-Yakubovic-Popov lemma which relates the state
space and the frequency domain properties of positive real systems.
To motivate dissipativity definition, let Z be the impedance of a linear, passive
circuit, which maps the vector of port currents i to the vector of port voltages v. Then
∫ T
< Z i, i >[0,T ] = iT (t) v(t)dt (2.54)
0
is the energy consumed by the circuit over the time interval [0, T ]. Since a passive circuit
never produces energy, the above integral in Eq. (2.54) must be nonnegative. A strictly
passive circuit consumes energy for any port current i ̸= 0 and all terminal times T , so the
integral in Eq. (2.54) must be positive for any i ̸= 0. By evaluating the inner product in
the frequency domain, one can show that the (time-invariant) circuit is incrementally strictly
passive if and only if the transfer function matrix Z satisfies Z(jω) + Z ∗ (jω) ≥ 2 ϵ I. Since
this inequality is equivalent to the requirement that the real part of Z(jω) is larger than ϵ for
all ω, such transfer function matrices are called strictly positive real.
Now, the key results connecting the notion of passivity to strict positive realness
and strict bounded realness, respectively, of a linear dynamical system is presented.
Theorem : Consider the dynamical system
 
min  A B 
G(s)   (2.55)
∼ C D
with input u(.) and output y(.). Then the following statements are equivalent:
(i) G(s) is strictly positive real;
∫T
(ii) G(s) is passive; that is, 0 uT (t)y(t)dt ≥ 0;
(iii) there exist matrices P ∈ ℜn×n , L ∈ ℜp×n , and W ∈ ℜn×m , with P positive definite, such
that
0 = AT P + P A + LT L, (2.56)
0 = P B − C T + LT W, (2.57)
0 = D + DT − W T W. (2.58)
Furthermore, G(s) is strongly positive real if and only if there exists n × n positive-definite
matrices P and R such that
0 = AT P + P A + (B T P − C)T (D + DT )−1 (B T P − C) + R. (2.59)
Remark: The proof of the above Theorem assumes that the storage function belongs to C 3 .
Remark: The dual version of the above Theorem can be obtained by replacing A by AT
and B by C T . In particular, G(s) is strictly positive real if and only if there exist matrices
Q, L and W such that:
0 = A Q + Q AT A + L L T ,
0 = Q CT − B + L W T ,
0 = D + DT − W W T .
Next, an analogous result for strictly bounded real systems is presented.

Theorem : Consider the dynamical system
 
min  A B 
G(s)  
∼ C D
with input u(.) and output y(.). Then the following statements are equivalent:
(i) G(s) is strictly positive real; (ii) G(s) is finite gain,
∫T ∫T
i.e., 0 y T (t) y(t)dt ≤ γ 2 0 uT (t) u(t)dt, T ≥ 0
(iii) there exist matrices P ∈ ℜn×n , L ∈ ℜp×n , and W ∈ ℜn×m , with P positive definite, such
that
0 = AT P + P A + C T C + LT L,
0 = P B − C T D + LT W,
0 = γ 2 Im − DT D − W T W.
The salient feature of positive real transfer functions is that they are dissipative
and phase bounded. Hence the feedback interconnection of positive real transfer functions is
guaranteed to be stable without requiring that a small gain condition be satisfied. Positive
real design is thus potentially less conservative in the presence of phase information.
To be added
2.7.2 State Space Realizations
Given the transferfunction/


matrix G(s), there exists (∃) a one to one mapping to (A, B, C, D),
 A B 
i.e., G(s) ⇐⇒  . A state space realization (A, B, C, D) is minimal if and only
C D
if (A, B) is a controllable pair and (C, D) is an observable pair. If (A1 , B1 , C1 , D) and
(A2 , B2 , C2 , D) are two minimal realizations of G(S), then, ∃ a unique nonsingular mapping
T such that A2 = T A1 T −1 , B2 = T B1 , C2 = C1 T −1 .
For a linear dynamic system: ẋ = A x + B u, y = C x + D u; transfer function/
 A B 
matrix is given by G(s) = C (s I − A)−1 B + D or   := C (s I − A)−1 B + D.
C D
Given G(s), realization (A, B, C, D) can be obtained (use Matlab command: tf2ss).
SISO Example : Consider a SISO system with the following transfer function:
 
β1 s n−1
+ β2 s + · · · + βn−1 s + βn
n−2 A b 
G(s) := Then, G(s) ⇐⇒ 
  , where,
sn + a1 sn−1 + · · · + an−1 s + an c 0
   
 −a1 −a2 · · · −an−1 −an   1 
   

 1 0 ··· 0 0 

 0 


A :=  .  
, b :=  . 
 .. .. .. .. 


 .. 
 . ··· . .   
   
0 0 ··· 0 0 0
[ ]
c := β1 β2 · · · βn−1 βn
MIMO Example : Consider a 3 × 3 transfer matrix:
 
1 2 s+1 s
 (s+1) (s+2) (s+1) (s+2) (s+1) (s+2) 
 
 s2 +5 s+3 
G(s) = 

1
(s+1)2 (s+1)2
s
(s+1)2


 
1 2 +1 s
(s+1)2 (s+2) (s+1)2 (s+2) (s+1)2 (s+2)
On mapping into state space model, a 4th order minimal state space realization can be derived
as (Refer to Book by Zhou and Doyle, [50]):
 
 −1 1 0 0 0 3 1 
 

 0 −1 0 0 1 −1 −1 

 
 
 0 0 −1 0 −1 0 1   
 
 
 A B 
G(s) ⇐⇒ 
 0 0 0 −2 1 −3 −2

 =  .
 
  C D
 
 0 1 0 −1 0 0 0 
 
 
 1 0 0 0 0 1 0 
 
 
1 0 1 1 0 0 0
2.7.3 Canonical Forms
Canonical realization (or canonical simulation, Refer: Kailath [48], pp. 37 - 49) is the repre-
sentation which uses minimum number of integrators or differentiator in the analog simulation
of dynamic system represented by its transfer function.
For a single input - single output systems, the generic transfer function is given by
b1 sn + b2 sn−1 + . . . + bn+1 s0
G(s) =
sn + a1 sn−1 + a2 sn−2 + . . . + an
For the sake of demonstration, consider the following second order system [48, p. 38]:
y(s) b1 s2 + b2 s1 + b3 s0
= G(s) = 2
u(s) s + a1 s1 + a2 s0
{s2 + a1 s1 + a2 s0 } y(s) = {b1 s2 + b2 s1 + b3 s0 } u(s)
which is written as a differential equation on taking inverse Laplace transform as:
ÿ + a1 ẏ + a2 y = b1 ü + b2 u̇ + b3 u (2.60)
The related system is given by
ẍ + a1 ẋ + a2 x = u; y = b1 ẍ + b2 ẋ + b3 x (2.61)
where all zero initial conditions are assumed. On the right hand side, Eq. (2.60) has number
of differentiators. These can be eliminated by moving the lines with differentiators backwards
over the requisite number of integrators corresponding to the same order of denominator
polynomial. Note that if x is the output of the last integrator, then input to this integrator
is ẋ and input to the preceding integrator is ẍ and so on. The desired linear combination
of (x, ẋ, . . .) produces the output y in Eq.2.61. The simulation block diagram is shown in
Fig. 2.48. This realization is also known as controller canonical form or realization. Similarly,
observer canonical form can be worked out (Home work problem). Details refer [48, pp.35-48].
Figure 2.48: Traditional Controller Canonical Form.
Figure 2.49: Traditional Observer Canonical Form.
Transforming a (SISO) system to a control canonical form:
F or ẋ = Ax + bu, letQc = [b|Ab| . . . |An−1 b]n×n
Let x = T xc =⇒ ẋc = T −1 AT xc + T −1 bu or ẋc = Ac xc + bc u.
Qcc = [bc |Ac bc | . . . |An−1

c bc ] ; then, prove that T = Qc Q−1
cc
procedure to find T −1 : 1. Form augmented matrix [Qc , I]

2. Perform elementary row operation (Hermite normal form) until this matrix has the form
[Qcc T −1 ]: Ref.[48].
Alternately, ẋ = Ax + bu =⇒ T.F. (Characteristic equation) =⇒ control canonical form

Algorithm to find (SI − A)−1 = E1 sn−1 + . . . + En /(sn + a1 sn−1 + . . . + a); where,
E1 = I, a1 = −tr(AE1 )
−1
E2 = a1 I + AE1 , a2 = tr(AE2 )
2
−1
Ei = ai−1 I + AEi−1 , ai = tr(AEi )
i
−1
En = an−1 I + AEn−1 , an = tr(AEi )
n
En+1 = 0 =⇒ measure of convergence =⇒ check.
c[E1 sn−1 + ... + En ]b
G(s) = c (SI − A)−1 b = n
s + a1 sn−1 + ....an
Jordan canonical (Modal) form: Given a matrix A, let P = matrix of eigenvectors, &
matrix of eigenvalues
 
 λ1 0 
 
F or distinct λ1 , λ2 , λ3 , Λ = [P −1
AP ] = 
 λ2


 
 
0 λ3
 
 λ1 1 0 
 
 
f or λ1 = λ2 , and λ3 distinct, Λ =  0 λ1 0 
 
 
0 0 λ
 
λ1 t
At −1 Λt Λt  e 0  ′
P rove that e =P e P, then, e =  λi s distinct
0 eλ2 t
   
 a + jb 0   a b 
Λ=  ←→   λ1 , λ2 complex conjugate
0 a − jb −b a
   
at
 a + jb 0   e cosbt eat sinbt 
 ⇐= P =? f ind.
Λt
Λ=  =⇒ e = 
0 a − jb −eat sinbt eat cosbt
   
 λ1 0 0   b̃1 
   
   
ẋ = Ax + bu =⇒ z = P x =⇒ ż = Λ z + P −1 bu =  0 λ2 
0  z +  u
  b̃2 
   
0 0 λ3 b̃3
Completely controllable system =⇒ b̃1 ̸= 0, b̃2 ̸= 0, b̃3 ̸= 0;

smaller b̃1 =⇒ larger control effort and vice versa
=⇒ particular mode has poor controllability.
2.7.4 MIMO Generalized Control Canonical Forms (GCCF)
A physical variable representation can be converted into GCCF using the following procedure:
ẋ = Ax + Bu =⇒ ż = Ag z + Bg v
Ag = block diagonal (Aγ1 , Aγ2 , . . . Aγm )
Bg = block(bγ1 bγ2 . . . bγm

   
 0 1 0 ... 0   0 
   
   
 0 0 1 ... 0   0 
   
   
A γi = 
 −− −− −− −− −− 
 ; bγi =  −− 


   
   
 0 0 0 ... 1   0 
   
   
0 0 0 ... 0 1
Procedure [10]:
1. A change of basis in the state variable x, i.e.,
x = T z, where det(T ) ̸= 0 (2.62)
ż = T −1 AT z + T −1 Bu = Āg z + B̄g u
2. A change of basis in the control variable, u, i.e.,
u = F w, where, detFm×m ̸= 0 (2.63)
ż = Āg z + B̄g Fw = Āg z + Bg w.
3. Introduction of state feedback
w = v − Hz = v − HT −1 x, where H( m × n) and vm×1 redef ined input vector(2.64)
ż = [Āg − Bg H]z + Bg v = Ag z + Bg v = [T −1 AT − T −1 BF H]z + [T −1 BF ]v (2.65)
Control Invariants:
Controllability matrix Qc = [B|AB| . . . |An−1 B](n×nm) . For a multi-input sys-
tem, controllability can also be evaluated from the modified controllability matrix Q̄c =
[B|AB| . . . An−m B] . The columns of Q̄c are expanded as
Q̄c = [b1 b2 . . . bm |Ab1 . . . Abm |A2 b1 . . . An−m bm ] (2.66)

For controllable system, rank Qc = rank Q̄c = n =⇒ i.e., there exists at least n - linearly
independent columns in Qc and Q̄c . These independent n-vectors are rearranged in the form:
Q̃c = [b1 Ab1 . . . Aγ1 −1 b1 |b2 . . . Aγ2 −1 b2 |bm . . . Aγm −1 bm ] (2.67)
This selection of Q̃c ensures that all m-input are used to control the system. The γi′ s are the
control invariants of the system. If needed, the columns of B are permutated so that control
invariants are in decreasing order.
Example 12 [10]:
   
 3 5 1   0 1  λ1 = 1
   
   
A= 0 0 1  , B= 0 0  λ2 = 2
   
   
0 −2 3 1 3 λ3 = 3
System with n = 3, m = 2 is controllable .

   
 0 1 1 6   0 1 1 
   
   
Q̄c =  0 0 1 3  =⇒ Q̃c = [b1 Ab1 |b2 ] =  0 1 0  =⇒ γ1 = 2, γ2 = 1
   
   
1 3 3 9 1 3 3
     
 −3 0 1     |  eT11
     

 0 1 0 

 T 
 e12 
 
 |eTij  j = 1, . . . γi
Q̃c 
=      
= = 
 −− −− −−   −−   −−  i = 1, . . . m.
     
     
1 −1 0 eT21 |
Use bottom row of each partitioned sub-matrix Qc to form transformation matrix T −1 as:
   
T  
 e12   0 1 0 
     1 0 1 
 T  
 e12 A   0 0 1 
  
T −1 =  = , T = 
 1 0 0 

T hen,
     
 −−   −− −− −−   
   
    0 1 0
eT21 1 −1 0
   
 0 1 0   0 0 
   
Āg = T −1 AT =  −2 3 0  , B̄g = 
  
 1 3 
   
   
8 0 3 0 1
 
 f11 f12 
u = F w, F =   ; because m = 2
f21 f22
2×2
   
 0 0   0 0 
     
  
 f11 + 3f12 f12 + 3f22   1 0 
 1 −3 
Bg = B̄g F =  =  , theref ore F = 
 ,
   
 −−−−− −−−−−   −− −−  0 1
   
   
f21 f22 0 1
 
 h11 h12 h13 
let H =  
h21 h22 h23
m×n
 
 
 0 1 0 
 0 1 0   




  00 0 

Ag = Āg − Bg H =  −2 − h11 3 − h12 −h13 =

 , hence,

   
   −− −− −− 
8 − h21 −h22 3 − h23  
0 0 0
 
−2 3 0 
=⇒ H = 
 
8 0 3
In general, T −1 = [eT1γ1 eT1γ1 A . . . eT1γ1 Aγ1 −1 | . . . |eTmrm . . . eTmrm Aγm −1 ]T (2.68)
For a SISO system, How controllability and observability canonical forms can be found?
Determine the transformation matrix using the above procedure.
2.8 Methods of Improving System Performance

• Simple adjustment of gain K : From root locus analysis, it is obvious that the gain
adjustments only shift the closed loop poles along the well defined loci. No major changes
in pole - zero locations are possible.
• Minor changes in system structure : such as using additional feedback components,

like say, velocity feedback.
• Major changes in system structure or components.
• Addition of compensating networks to alter characteristic at critical frequency range.

Series/ cascade compensation or feedback/ parallel compensation or both. Typical com-
pensators : lead or PD (proportional + derivative) ; lag or PI (proportional + integral);
lead-lad and lag lead or PID (proportional + integral + derivative) compensation.
2.8. METHODS OF IMPROVING SYSTEM PERFORMANCE 131
Compensating networks can be placed either in feedback path or cascade/ feed-forward path or
both. Structure of such a controller depends on designer’s choice and familiarity (& also liking)
as well as hardware configuration that implements the controller. The control algorithm has
to be suitably programmed for efficient implementation based on control structure. Therefore,
selection of cascade Vs feedback configuration (or control structure) depends on:
1. Selection based on implementability.
2. Size and weight consideration, requirement of amplification in cascade (to get say, desired
dc gain from reference input to the desired output).
3. Aerospace systems, stabilization systems with feedback compensator (or state variable
feedback) has greater robustness against parameter variations.
4. Selection depends on influence of noise and disturbances which in turn depends on type of
compensator used.
5. For speed up of response (say, with lead compensator), feedback compensator preferred.
6. Good tracking requirement, cascade compensator with lag/PI compensator is preferred.
7. Feedback compensators are more effective for open and closed loop systems having complex
pole pairs.
Performance of a control system based on step input :
1. A given system is stable, its transient response is satisfactory; but its steady state error
is large.
2. A given system is stable, but, its transient response is unsatisfactory.
3. A given system is stable; but, both transient response and steady state response, are
unsatisfactory.
4. given system is unstable for all values of forward/ cascade gain K.
Common types of compensators (passive) :
1. Lag network/compensator (acts like a Low pass filter) having the transfer function Gc (s):
A (1 + T s) A s + 1/T
Gc (s) = = ; 1 < α ≤ 10; typical α ∈ [4, 8]. (2.69)
(1 + α T s) α s + 1/α T
=⇒ can be used for systems of category 1 (listed above).

2. Lead network/ compensator (acts like a high pass filter) :
A α (1 + T s) s + 1/T
Gc (s) = =A ; 1 > α ≥ 0.1 typical α ∈ [0.2, 0.4]. (2.70)
(1 + α T s) s + 1/α T
=⇒ can be used for systems of category 2 or 4 .
3. Lag - Lead network/ compensator :
A (1 + T1 s) (1 + T2 s) A (s + 1/T1 ) (s + 1/T2 )
Gc (s) = = ; α > 1, T1 > T2 (2.71)
(1 + α T1 s) (1 + (T2 /α) s) (s + 1/α T1 ) (s + α/ T2 )
=⇒ can be used for systems of category 3.
4. PID controller (text book type) : Gc (s) = Kp + Ki /s + Kd s.

=⇒ used largely by industrial/ process control.
Features of these compensators will be discussed later.

Response of a stable control system based on step input :
⋆ Inputs: step, ramp (t), parabolic (t2 ), sinusoidal etc
⋆ output response: transient + steady state
⋆ Tracking/ servo systems (Type 1 system: to be discussed next):
→ Error to step input = 0;
→ Error to ramp input = finite/ zero.
The transient response characteristics are important performance features control systems.
Another important feature is concerned with system error. System error could arise due to -
change in reference input; steady state error; imperfections in system components which can
cause error in steady state. For a stable system, steady state error is obtained using the final
value theorem, which is repeated here as:
lim f (t) = lim s F (s); F (s) has all poles in LH s-plane

t→∞ s→0
A good stable, controllable and observable control system should have a most fa-
vorable transient as well as steady state responses. Apart from this attribute, control system
should also have ability to reject effects of disturbance and noise. Since the physical systems
are constrained due to the limited number of inputs and outputs/ measurements, there is
always a compromise on achievable robust stability and performance (transient and steady
state) and disturbance rejection (or accommodation) and noise suppression.
Root locus analysis gives a general idea of about attainable transient response and
relative stability. As the gain is varied, the closed loop poles move along the well defined
locus or paths in the s-plane. Flexibility in the control system design process can be be
increased if roots can be placed at a desired pole location that is outside the locus. For
example, increasing the speed of response without effecting the percent overshoot can not be
accomplished by mere gain adjustment. Rather changing the gain, if system can be augmented
or compensated with additional poles and zeros so that the closed loop poles pass through the
desired pole locations. Possible disadvantage of augmenting a compensating system is that
it increases the order of the system and may adversely effect the frequency response outside
the design frequency band. However, compensators can be used for independently improving
transient properties as well as steady state characteristics.
Steady - state Error Analysis
Steady state error is a measure of the tracking accuracy of a stable system. Steady state
performance (long after transient has died down) of a stable system is generally judged by
the steady state error due to step, ramp, or acceleration and higher order polynomial in time
inputs. Static error coefficients are concerned with such errors. Control systems are classified
according to their ability to follow step, ramp and higher order polynomial in time inputs.
The actual input can be considered as a linear combination of such inputs. The magnitude of
the steady state errors are indicative of goodness of the system.
For the typical cascade control configuration in the Figure (2.28), the input - output
transfer function is given by
y(s) K G(s) H(s)

=
r(s) 1 + K G(s) H(s)
From the Figure (2.28), tracking error e(s) = r(s) − y(s) and the corresponding transfer
function for the reference input r(s) is given by
e(s) 1
=
r(s) [1 + K G(s) H(s)]
r(s) r(s)
e(s) = =
1 + K G(s) H(s) [1 + Geq (s)]
Geq (s)
Gc (s) = = Closed loop transfer function
1 + Geq (s)
Steady state error

s r(s)
ess = lim e(t) = lim s e(s) = lim [ ].
t→∞ s→0 s→0 1 + K G(s) H(s)
The static (position, velocity, acceleration) error coefficients (Kp , Kv , Ka , . . .) are figures of
merit of a control system. The higher the coefficients, the smaller the steady state error.
For a step input, (r(s) = 1s )

s 1 1
ess = lim =
s → 0 [1 + K G(s) H(s)] s [1 + K G(0) H(0)]
define Kp = lim [K G(s) H(s)] = K G(0) H(0) = Geq (0)
s→0
1
⇒ ess =
1 + Kp
Therefore, ess → 0 requires static position error coefficient/ constant Kp → ∞.
Since, Geq (s) = K G(s) H(s), Geq (s) should have at least one integrator, i.e., term ( 1s ). For
Type 1 systems to follow the step input with zero steady state error, hence, its
open loop transfer function must have at least one integrator (Kp is ∞). Let the OLTF
K (1 + Ta s) (1 + Tb s) · · · (1 + Tm s)
G(s) H(s) = , (2.72)
sN (1 + T1 s) (1 + T2 s) · · · (1 + Tn s)
where N = number of integrators = pole multiplicity at the origin; and
N = 0, 1, 2, . . . for Type 0, 1, 2, . . . systems, respectively.
K (1+Ta s) (1+Tb s) ··· (1+Tm s)
For a Type 0 system, Kp = lims → 0 (1+T1 s) (1+T2 s) ··· (1+Tn s) = K, which is finite.
K (1+Ta s) (1+Tb s) ··· (1+Tm s)
For a Type 1 system, Kp = lims → 0 s1 (1+T1 s) (1+T2 s) ··· (1+Tn s) = ∞.
Example : Single Degree of freedom system with second order transfer function (Refer Fig.
ω2
2.50 ) Gc (s) = (s2 +2 ζ ω+ω 2 )
. On converting this transfer function into Geq - form (see bottom
block diagram of Fig. 2.50), there exists a pole at origin. This property is true for all transfer
functions having D. C. gain = unity, and such closed loop systems are Type 1 systems.
1
Similarly, for a ramp input: (r(s) = s2
)
s 1 1
ess = lim = ,
s → 0 1 + K G(s) H(s) s2 s [1 + K G(s) H(s)]
define Kv = lim [s K G(s) H(s)]
s→0
⇒ static velocity error coefficients Kv → ∞ ⇒ ess → 0
Hence, Geq (s) should have at least two integrators. Type 2 systems follow the ramp input
with zero steady state error, hence, its open loop transfer function must have at least two
Figure 2.50: Second Order Transfer function and Geq - representation.
integrators. If static velocity error constant Kv is ∞, then Kp is also ∞. However, for Type
1 systems, Kv is finite.
s K (1+Ta s) (1+Tb s) ··· (1+Tm s)
For a Type 0 system, Kv = lims → 0 (1+T1 s) (1+T2 s) ··· (1+Tn s) = 0.
s K (1+Ta s) (1+Tb s) ··· (1+Tm s)
For a Type 1 system, Kv = lims → 0 s1 (1+T1 s) (1+T2 s) ··· (1+Tn s) = K, which is finite.
s K (1+Ta s) (1+Tb s) ··· (1+Tm s)
For a Type 2 system, Kv = lims → 0 s2 (1+T1 s) (1+T2 s) ··· (1+Tn s) = ∞.
1
For acceleration / parabolic input (r(s) = s3
), Ka = lims → 0 [s2 K G(s) H(s)].
For Type 3 systems, acceleration error constant Ka → ∞. If Kv → ∞, then Kp → ∞, but
Ka is finite for Type 2 systems; and so on.
K (c1 + c2 s + · · · + cm sm−1 + cm+1 sm )
Let, Geq (s) =
(b1 + b2 s + · · · + bn sn−1 + sn )
e(s) (sn + bn sn−1 + · · · + b1 )
then, = = α0 + α1 s + α2 s 2 + · · · ,
r(s) n
(s + an s n−1 + · · · + a1 )
where, a1 = (b1 + K c1 ); a2 = (b2 + K c2 ), · · ·
e(s) 1
since, =
r(s) (1 + Geq (s))
ess = lim [α0 + α1 s + α2 s2 + · · ·] = α0
s→0
For Type 1 systems, Kp → ∞ ⇒ α0 = 0; b1 = 0.
i.e., integrator or one s-term can be taken out from denominator of Geq (s). Note that
y(s) y(s) y(0)
CLTF: =1− = 1 − (α0 + α1 s + α2 s2 + · · · , ) ⇒ = 1 − α0 .
r(s) r(s) r(0)
y(0)
If α0 = 0, then, r(0) = 1 = steady state or d.c. gain = unity.
y(s) Kp K (s + δ1 ) (s + δ2 ) · · · (s + δm )
=
r(s) (s + λ1 ) (s + λ2 ) · · · (s + λn )
∏
y(0) Kp K m i=1 δi
= ∏n
r(0) j=1 λj
∏m
i=1 δi
therefore, Kp = ∏n ∏
{ j=1 λj − m i=1 δi }
Similarly, the static velocity error constant/ coefficient Kv and static acceleration
error constant/ coefficient Ka can be computed as (Home work !):
1 ∑n
1 ∑m
1
= −
Kv j=1 λj i=1 δi
2 1 ∑
n
1 ∑
m
1
− = + −
Ka Kv j=1 λj i=1 δi2
2 2
Implications of steady state tracking and steady state error Kp for tracking are as follows: For
tracking step input with zero steady state error, Geq should have at least one integrator. Since
poles of the plant are not at the hands of a designer, lag filter has to be used for systems
of category 1 (stable system with satisfactory transient response but unsatisfactory steady
state response, and since ideal integrator 1s is impossible to realize in implementation). Lag
compensator, then should have a pole close to origin, which helps in keeping the associated
‘zero’ close to origin, so that phase contribution due to ‘pole - zero’ pair of the lag filter is
small (say, around 2◦ to 5◦ at the closed loop poles of the system without a lag compensator).
Thus, the closed loop poles do not move far away due to addition of lag filter/ compensator,
and hence, transient response does not deteriorate for the compensated system.
2.9 Compensator Synthesis via Root Locus Technique

Features of (lag, lead) cascade compensators/ filters:
Lag compensator : (Characteristic similar to Low pass filter - LPF)
H(s) = Gc (s) = A (1+T s)
(1+α T s)
= A s+1/T
α s+1/α T
; 1 < α ≤ 10; typical α ∈ [4, 8].
Lag filter pole at pc = α−1T and zero at zc = −1 T (Eqn. 2.69) are placed close to the origin
(Refer left part of Figure 2.51). In many practical designs, α ∈ [4, 8]. The closeness of pole
and zero makes the angle contributed by the filter ϕc (∈ [2◦ , 5◦ ] ) small. The new closed loop
pole practically remains unchanged from the uncompensated value. The transient response,
therefore, does not change. The value of static error coefficient increases by a factor α,
hence helps in reducing steady state error with respect to step input. Therefore, α should be
2.9. COMPENSATOR SYNTHESIS VIA ROOT LOCUS TECHNIQUE 137
made as large as practically possible. However, larger the value of α, larger will be reduction
in phase margin and hence larger is the amplitude of transient.
Summary: Addition of Lag filter results in: 1. large increase in Kp (by a factor ≈ α), which
means a smaller steady state error; 2. decrease in natural frequency ωn , and therefore has
disadvantage of increase in settling time (sluggish).
Figure 2.51: Lag and lead compensator poles in s-plane.
Example Lag1: Consider an OLTF of a unity gain feedback system [45, §9.2]:
K
G(s) H(s) =
(s + 1) (s + 2) (s + 10)
whose root locus plot is shown in left part of Figure 2.52. Objective is to improve the steady
state error by a factor of 10, while the system is operating with a damping ratio of 0.174.
The uncompensated system error ess = 0.108 with Kp = 8.23. A ten fold decrease in steady
1
state error implies that ess = e(∞) = 0.108/10 = 0.0108. Since, ess = 1+K = 0.0108; solving
p
for Kp = 1−ee
ss
= 1−0.0108
0.0108 = 91.59. The improvement in Kp from uncompensated system to
ss
the compensated system is the required ratio of the compensator zero to a compensator pole
zc 91.59
= = 11.13.
pc 8.23
On selecting an arbitrary value of pc = 0.01, results in zc = 11.13 ∗ 0.01 = 0.1113. Hence, the
OLTF for the compensated system is
K (s + 0.1113)
G(s) H(s) =
(s + 0.01) (s + 1) (s + 2) (s + 10)
The root locus for the compensated system is shown on the right side of Figure 2.52.
Root Locus Lag−compensated system
6
System: sys System: sys
Gain: 161 Gain: 156
Pole: −0.685 + 3.88i Pole: −0.674 + 3.81i
Damping: 0.174 Damping: 0.174
Overshoot (%): 57.4 Overshoot (%): 57.4
Frequency (rad/sec): 3.94 Frequency (rad/sec): 3.87
4
Imaginary Axis
0
−2
−4
−6
−8 −6 −4 −2 0 2 −12 −10 −8 −6 −4 −2 0 2
Real Axis Real Axis
Figure 2.52: Root locus plot for uncompensated (left) and lag compensated (right) systems
On comparing the root locus plot of uncompensated system with that of lag-compensated
system, it can be seen that the dominant poles at damping ratio of ζ = 0.174 are located at
−0.694±j 3.926 with a gain of 164.6 which is very close to the dominant pole at −0.678±j 3.836
with a gain of 158.1 (gains differ slightly in Fig.2.52, due to error in marking on MATLAB
plot). For the compensated system, the third and fourth poles are at -11.55 (against -11.61
for uncompensated system) and -0.101 respectively. The pole at -0.101 is close to the zero
at -0.1113 (near pole - zero cancelation) and hence it is a non-dominant pole and does not
contribute significantly to the step response. Hence, the transient response for both systems
are nearly the same, but the steady state error for the compensated system approximately one
tenth of uncompensated case. Note that further bringing the pole - zero pair of compensator
does make the controller sluggish since the fourth pole will have relatively larger residue and
has slower transient. This makes the system to reach steady state value rather late.
Example Lag2: Consider a unstable rocket dynamics (Refer: Chapter 1), whose transfer
function (including actuator transfer function and PD feedback (θ, θ̇ - feedback):
KA kc kR µc (s + 1/kR ) K A (s + 3)
GH = = ; since (2.73)
(s − KA ) (s + kc )
2 (s − 3.75) (s + 20)
2
1
µα = 3.75, µc = 4.54, = 3, kc = 20, K A = KA ∗ kc ∗ kR ∗ µc = 30.26667 ∗ KA .
kR
The minimum value of gain KA that stabilizes the system without lead compensator =
Kcritical = µα /µc = 3.75/4.54 = 0.826. For the gain KA = 4.13, the closed loop poles are
located at −4.02 ± 2.987 j (ζ = 0.803, ω = 5.0rad/sec) and −11.96 and the percent overshoot
for the third order system is 1.42%.
Based on the second order model, desired overshoot ≤ 2 % ⇒ ζ ≈ 0.5. Rise time ≈ 1 sec
√
⇒ ωn ≈ 5 rad/sec ⇒ ζ ωn = 2.5; ωd = ωn (1 − ζ 2 ) = 4.35.
In the above example, the real pole is close to the damping factor of the complex pole pole
Root Locus
System: sys
Gain: 4.13
Pole: −4.02 + 2.98i
Damping: 0.803
Overshoot (%): 1.45
Frequency (rad/sec): 5
System: sys
Gain: 0.828
Pole: −0.00762
Damping: 1
Overshoot (%): 0
−15 −10 −5 0
Real Axis
Figure 2.53: Root locus plot for uncompensated Launch vehicle flight control system
pair, hence it has significant effect on the transient response. Though the damping ratio of
the complex pole pair is ζ = 0.803, the overshoot is 1.42%, which is acceptable. Since, this is
Type 0 system, lag filter is needed. The closed loop transfer function for the uncompensated
system is
125.0 (s + 3)
Gc (s) = ; ⇒ d.c. gain = 1.25
(s3 + 20 s2 + 121.2513 s + 300)
Select α = 10, so that the steady state error reduces by a factor of 10. On selecting the lag
(s+0.1)
compensator transfer function H(s) = A (s+0.01) instead of KA , the corresponding root locus
is drawn in Fig. 2.54. (Home work: Draw the root locus using step by step rules.) For getting
the frequency of 5rad/sec, the gain is A = 4.15 (instead of 4.13 for uncompensated system).
Even the value of critical gain has not changed significantly (Kcritical = 0.865 instead of 0.826).
System: sys
Gain: 4.15
Root Locus Pole: −3.94 + 3.07i Root Locus
8 4 Damping: 0.789
Overshoot (%): 1.77
6 3
System: sys
4 Gain: 0.865 2
Pole: 0.00313 + 0.561i
Damping: −0.00558
2 Overshoot (%): 102 1
Imaginary Axis
Imaginary Axis
0 0
−2 −1
−4 −2
−6 −3
−8 −4
−20 −15 −10 −5 0 −4 −3 −2 −1 0 1 2
Real Axis Real Axis
Figure 2.54: Root locus plot for Lag-compensated Launch vehicle control; Left-closeup.
The open loop and closed loop transfer functions for the compensated system are:
30.2667 A (s + 3) (s + 0.1)
G H(s) = ;
(s2 − 3.75) (s + 20) (s + 0.01)
125.6067 (s + 3) (s + 0.1)
Gc (s) =
(s + 3.9432 ± 3.0687 j) (s + 12.0) (s + 0.1233) (s + 0.01)
(125.6067 s2 + 389.3807 s + 37.682)
=
(s4 + 20.01 s3 + 122.0567 s2 + 314.3432 s + 36.932)
The right side root locus plot indicates that the dominant closed loop complex pole pair is at
(−3.9432 ± 3.0687 j), i.e., ζ = 0.789, ω = 5.0rad/sec, percent overshoot is 1.72%. The steady
state for a unit step input is around 2%, which is also acceptable as compared to around
≈ 25% without lag compensator.
Lead compensator (Characteristic similar to High pass filter - HPF)

H(s) = Gc (s) = A α (1+T s)
(1+α T s)
=A s+1/T
s+1/α T
; 1 > α ≥ 0.1 typical α ∈ [0.2, 0.4]. Make α
sufficiently small, i.e., the non-dominant pole at pc = −1/α T is far to the left and has little
effect on the important part of root locus. The net angle contribution is predominantly due
to the zero (Refer right part of Figure 2.51). Determine zc by trial and error for best location.
For Type 1 or higher order stable systems, if zero zc = −1/T is super imposed/ and
possibly cancels the dominant real pole (excluding the pole at origin) of the original transfer
function, a good improvement in transient response can be obtained.
For Type 0 systems, better time response can be obtained by placing compensator zero zc
close or over the second largest real stable pole of the original transfer function.
In other cases, zc is moved to the left or right until desired transient response and desired
loop gain are obtained.
Desired root (closed loop pole) pd is selected based on performance specification for the system.
The lead compensator should make pd a closed loop pole. Compute total angle contributions
from all poles and zeros of the original system = 180◦ + ϕ. The lead compensator should
then contribute a lead of ϕc = −ϕ, so that total angle at the desired pole location pd of the
compensated system is 180◦ and the desired root is a point on the locus.
Many combinations of (pc , zc ) can contribute ϕc = −ϕ. Cancelation of a real pole of the
original system by a filter zero may simplify locus and reduce the complexity of the system.
This will also result in slight increase in Kp (hence, steady state accuracy is improved).
Lead compensator increases ωn ⇒ reduction in settling time.
For a Passive lead network/ filter, addition to gain A to compensate for α is more than increase
in error constant Kp .
Example Lead1: Consider a missile dynamics (Refer: Chapter 1), flying at Mach ∼ 1.2,
Zα = 4170f t/sec2 , Zδ = −1115f t/sec2 ,
Mα = −248rad/sec2 Mδ = −662rad/sec, Mq ≈ 0.
Typical actuator dynamics is δ̇ = 1

τa
(u − δ) where, u is the input to the actuator. The time
constant for actuator τa = 0.01 sec. The combined transfer function for missiles with actuator
(Eq. 1.39 is reproduced here):
−1115K(s2 − 2228)
G(s) = −
(0.01s + 1)(s2 + 3.33 s + 248)
jω
20 -3
K = 11.13 or K~
~ 10
−100 −47 47 σ
−20
Figure 2.55: Root locus for lateral acceleration feedback control for a missile [1]
Detailed analysis of acceleration feedback control system is discussed in the §3.3.

Complementary root locus plot of the acceleration feedback loop is shown in Fig. 2.55. The
system becomes unstable at low gain values. Hence, lead compensator may be employed to
enhance stability margins and speed of response. Let the desired dominant pole be located
at pd = −20 ± 20 j. The angle contributions from the poles at −1.665 ± 15.6596 j, −100 at
the desired pole are ϕ1 = 166.6817, ϕ2 = 117.2107, ϕ3 = 14.0362. The angle contributed by
the zeros are at ± 47.2017 be equal to θ1 = 163.4260, θ2 = 36.3268. For the complementary
root locus, net angle contribution is on the locus is 0◦ . Therefore, the net angle contribution
by three poles and two zeros is (166.6817 + 117.2107 + 14.0362) − (163.4260 + 36.3268) =
98.1758◦ . The pole - zero pair of the lead compensator should contribute 98.1758◦ . Let
the angles contributed by the zero zc be 110◦ and hence by the non-dominant pole be
11.8242◦ . Hence the pole zero combination of the lead compensator are at pc = −115.533
and zc = −12.7206, α = zc /pc = 0.1101. At gain = 0.001082345, the closed loop poles are
located at −29.0919 ± 83.7307 j, −20.0004 ± 19.9989 j. Refer Fig. (2.56).
For the uncompensated system, forward gain = 0.001; hence A α = 1.082345, i.e.,
1.082345 (s+12.7206)
A = 1.082345/0.1101 = 9.83. Therefore, compensator transfer function is H(s) = (s+115.533)
.
Root Locus
100
80
System: sys
Gain: 0.00108
60 Pole: −19.8 + 20i
Damping: 0.704
40 Overshoot (%): 4.45
Imaginary Axis
20
−20
System: sys
−40 Gain: 0.00108
Pole: −29.5 − 83.5i
Damping: 0.333
−60
Overshoot (%): 33
−80
−100
−100 −50 0 50 100 150

Real Axis
Figure 2.56: Root locus for lateral acceleration feedback control for a missile with lead filter
4
Example Lead2: Consider a stable dynamics G(s) = s (s+2) with unity gain feedback control.
y(s)
Closed loop transfer function r(s) = (s2 +24 s+4) . For this system, d.c. gain = 1, hence,
√
Kp = ∞. Eigenvalues of the closed loop system are at −1 ± j 3. Let the desired closed
√
loop poles be located at −2 ± j 2 3 (i.e., ζ = 0.5, ω = 4.0rad/sec) so as to have a faster
√
settling time. At s = −2 + j 2 3, total angle contribution from the open loop poles at 0
and −2 are +210◦ . Lead network must contribute ϕ = 30◦ at the desired closed loop location
√
s = −2 + j 2 3. If the zero is set at zc = 1/T = −2.9 erad/sec. Angle contributed by this
√
zero at the closed loop pole at s = −2 + j 2 3 is 75.5◦ . Select the pole pc such that it subtends
√
an angle of 45.5◦ at the desired closed loop pole. Hence, pc = 5.4. Then at s = −2 + j 2 3,
the magnitude condition | s (s+2) (s+5.4) | = 1; results in A = 18.7. Therefore gain constant
A (s+2.9)
for the compensator is 18.7/4 = 4.68.

Homework: For a system having transfer function G(s) = 1/[s2 (s + 5)], determine a lead
compensator such that ζ = 0.5; ωn = 0.5 rad/sec.
2.10 PID Control

Consider a Linear Time Invariant system defined by the equations:
ẋ(t) = A x(t) + B u(t)

(2.74)
y(t) = C x(t)
where, x ∈ Rn , y ∈ Rr , u ∈ Rm , rank B = m, rank C = r.

For example, satellite attitude dynamics θ̈ = u, with the state variables x1 = θ, x2 = θ̇ and
the output y = θ, can be written in the state space form as:
        

 ẋ1    

 0 1 
 x1   0 
 
 x1 

=  + u; y = [1 0]

 ẋ  
 x    1 
 
 x 
2 0 0 2 2
Solution to Eqn. (2.74) is given by

∫ t
A(t−to )
x(t) = e x(to ) + eA(t−τ ) Bu(τ )dτ (2.75)
to
Response of a linear time invariant system: ẋ(t) = Ax(t) + Bu(t) y(t) = Cx(t) can be
1. stable, but transient response unsatisfactory; or
2. stable but steady state accuracy is poor; or
3. stable, but both transient response and steady state accuracy unsatisfactory; or
4. unstable.
By mere gain adjustment of controller, it may not be possible to improve performance (also
stabilize better). One may need to use PID (proportional - integral - derivative) controller
or its derivatives in order to meet the design objectives. In case 1, PD controller or lead
compensator is preferred, for case 2, PI controller or lag compensator is better, for case
3, PID controller or lag - lead or lead - lag (depending on placing such a compensator in
cascade of feedback path) compensator is needed. For the last case one requires either PD
or PID controller. PID controllers are most popular with the process industries. but are not
considered as favorably in aerospace vehicles. Note that the use of the PID algorithm for
control does not guarantee optimal control of the system or system stability. With the wide
spread use of Numerical optimization technique and combining concepts of optimization (say,
2.10. PID CONTROL 145
H2 or H∞ or LMI sense), PID controller will become more acceptable in the future. Most of
the PID tuning are done by heuristic approaches and are applicable to largely SISO systems.
Extension to MIMO systems by combining modern optimal technique is being investigated.
A typical parallel form of PID control topology is shown in Fig. (2.57).
Figure 2.57: Block diagram of PID Feedback Control System.
A proportional-integral-derivative controller (PID controller) is a generic control

loop feedback mechanism (controller) widely used in industrial control systems. It is initially
developed for SISO systems. A PID controller attempts to correct the error between a mea-
sured process variable and a desired set point by calculating and then outputting a corrective
action that can adjust the process accordingly. The PID controller calculation (algorithm)
involves three separate parameters; the Proportional, the Integral and Derivative values. The
Proportional value determines the reaction to the current error, the Integral value determines
the reaction based on the sum of recent errors, and the Derivative value determines the re-
action based on the rate at which the error has been changing. The weighted sum of these
three actions is used to adjust the process via a control element. By “tuning” the three con-
stants in the PID controller algorithm, the controller can provide control action designed for
specific process requirements. The response of the controller can be described in terms of the
responsiveness of the controller to an error, the degree to which the controller overshoots the
set point and the degree of system oscillation. Note that the use of the PID algorithm for
control does not guarantee optimal control of the system or system stability. Note that the
terms - controller, compensator, filter are interchangeably used to break the monotonicity.
A brief account of analog PID control There are many version PID controllers available
in literature. Textbook form is easy to understand while depending on the PD, or PI or
PID form of controller and also on type of process or plant (including types of sensors and
actuators), different implementable forms have been developed in the past. Here, only text
book type of PID control is discussed for the sake of simplicity. Here too, simplest topology is
called non-interacting PID controller. The output of the three contributing terms are added
to produce a manipulated variable or input to the plant. Input to the PID is an error signal
e(t) which desired/ reference/ set point value minus actual measured value.
Proportional term: The proportional term (sometimes called gain) makes a change to the
output that is proportional to the current error value. The proportional response can be
adjusted by multiplying the error by a constant Kp , called the proportional gain (which is a
tuning/ adjustable parameter).
A high proportional gain results in a large change in the output for a given change
in the error. If the proportional gain is too high, the system can become unstable (See the
section on loop tuning). In contrast, a small gain results in a small output response to a
large input error, and a less responsive (or sensitive) controller. If the proportional gain is
too low, the control action may be too small when responding to system disturbances. In the
absence of disturbances, pure proportional control will not settle at its target value, but will
retain a steady state error that is a function of the proportional gain and the process gain.
Despite the steady-state offset, both tuning theory and industrial practice indicate that it is
the proportional term that should contribute the bulk of the output change.
Integral term: The contribution from the integral term (sometimes called reset) is proportional
to both the magnitude of the error and the duration of the error. Summing the instantaneous
error over time (integrating the error) gives the accumulated offset that should have been
corrected previously. The accumulated error is then multiplied by the integral gain and added
to the controller output. The magnitude of the contribution of the integral term to the overall
control action is determined by the integral gain, Ki (which is also a tuning/ adjustable
∫t
parameter). The integral term is given by: Iout = Ki e(τ ) dτ .
0
The integral term (when added to the proportional term) accelerates the movement
of the process towards set point and eliminates the residual steady-state error that occurs with
a proportional only controller. However, since the integral term is responding to accumulated
errors from the past, it can cause the present value to overshoot the set point value (cross over
the set point and then create a deviation in the other direction). For further notes regarding
integral gain tuning and controller stability, see the section on loop tuning.
Derivative term: The rate of change of the process error is calculated by determining the slope
of the error over time (i.e. its first derivative with respect to time) and multiplying this rate
of change by the derivative gain Kd . The magnitude of the contribution of the derivative term
(sometimes called rate) to the overall control action is termed the derivative gain, Kd . The
d e(t)
derivative term is given by: Dout = Kd dt
. The derivative term slows the rate of change of
the controller output and this effect is most noticeable close to the controller set point. Hence,
derivative control is used to reduce the magnitude of the overshoot produced by the integral
component and improve the combined controller-process stability. However, differentiation of
a signal amplifies noise and thus this term in the controller is highly sensitive to noise in the
error term, and can cause a process to become unstable if the noise and the derivative gain
are sufficiently large.
To summarize, the proportional, integral, and derivative terms are summed to
calculate the output of the PID controller. Defining u(t) as the controller output, the three
term formula of the PID control is:
∫t
d e(t)
u(t) = Kp e(t) + Ki e(τ ) dτ + Kd ; (2.76)
dt
0
1 ∫
t
d e(t)
= Kp [e(t) + e(τ ) dτ + Td ].
Ti dt
0
• Kp : Proportional gain - larger Kp typically means faster response since the larger the
error, the larger the Proportional term compensation. An excessively large proportional
gain will lead to process instability and oscillation.
• Ki : Integral gain - larger Ki implies steady state errors are eliminated quicker. The
trade-off is larger overshoot: any negative error integrated during transient response
must be integrated away by positive error before we reach steady state.
• Kd : Derivative gain - larger Kd decreases overshoot, but slows down transient response
& may lead to instability due to signal noise amplification on differentiation of the error.
1
Preliminary Example on PID: Consider a third-order plant model given by G(s) = (s+1)3
for studying he behavior of the proportional, integral, and derivative actions individually. If a
proportional control strategy is selected, i.e., Ki → 0 or Ti → ∞ and Kd = Td → 0 in the PID
control strategy, for different values of Kp , then the closed-loop responses of the system can
Kp
be obtained from the closed loop transfer function Gc (s) = (s+1)3 +Kp
. In order to evaluate the
Kp , the following root locus plot (Fig. 2.58) can be used. Note: n − m = 3, σ0 = −3/3 = −1.
Closed loop poles for three different gain values of Kp are given below:
0.1 −1.4642 −0.7679 ± 0.4020 j
. 1.0 −2.0000 −0.5000 ± 0.8660 j
8.0 −3.0000 −0.0000 ± 1.7321 j
Unit step response of the proportional feedback system is shown in top plot of Fig. 2.59.
3
Root locus for 1/(s + 1) with proportional feedback
2
1.5 System: sys

System: sys Gain: 7.99
Gain: 0.1 Pole: −0.000362 + 1.73i
Pole: −0.768 + 0.402i Damping: 0.000209
Damping: 0.886
Overshoot (%): 0.249 Frequency (rad/sec): 1.73
0.5
Imaginary Axis
System: sys
0 Gain: 1
Pole: −0.499 − 0.867i
Damping: 0.499
−0.5 Overshoot (%): 16.4
−1
−1.5
−2
−2 −1.5 −1 −0.5 0 0.5
Real Axis
Figure 2.58: Root locus plot of Proportional Feedback Control System.
However, it can be seen that when Kp increases, the response speed of the system increases, the
overshoot of the closed-loop system increases and the steady-state error decreases. However
when Kp is large enough, the closed-loop system becomes unstable, which can be directly
concluded from the root locus analysis. When Kp is outside the range of (0, 8), the closed-
loop system becomes unstable.
In order to reduce steady state error, integral component is added. Let Kp = 1 and
on applying a PI (proportional plus integral) control strategy for different values of Ki or Ti ,
closed-loop step responses of the example system can be generated as shown middle plot of
in Fig. 2.59. The most important feature of a PI controller is that there is no steady-state
error in the step response if the closed-loop system is stable. Further examination shows that
if Ti is smaller than 0.6, the closed-loop system will not be stable. It can be seen that when
Ti increases, the overshoot tends to be smaller, but the speed of response tends to be slower.
After fixing both Kp and Ti at 1, i.e., T i = Kp = 1, when the PID control strategy
is used with different T d, the unit step response of the closed-loop system is shown at the
bottom plot of Fig. 2.59. Clearly, when Td increases, the response has a smaller overshoot
with a slightly slower rise time but similar settling time.
Eq. 2.76 represents text book form of PID control. The derivative term in Eq. 2.76
is non causal, hence modified derivative terms are sometimes, used as in the transfer function:
Td s
U (s) = Kp [1 + T1i s + Td 1+(Td /N ) s
] E(s). Different versions of practical implementation of PID
have also been developed. For example: series and parallel PI and PD blocks in implementable
version. Extensive study of PID control is done by Aström, et al (for ex: [11]).
Methods of synthesizing/ Tuning PID Controller:
• Ziegler Nichols Method
• Kappa tau method
• Pole Placement
• Design on Gain & Phase margin Spec.
• D partitioning
• OLDP method
• Interval Polynomial technique
• Root Locus technique
• Bode/ Nyquist based design
• Genetic Algorithm
• Fuzzy PID
• Multi-objective control
Figure 2.59: Step response of P, PI and PID Feedback Control System.
PID Loop tuning
If the PID controller parameters (the gains of the proportional, integral and derivative terms)
are chosen incorrectly, the controlled process input can be unstable, i.e. its output diverges,
with or without oscillation, and is limited only by saturation or mechanical breakage. Tuning
a control loop is the adjustment of its control parameters (gain/proportional band, integral
gain/reset, derivative gain/rate) to the optimum values for the desired control response.
The optimum behavior on a process change or setpoint change varies depending on the ap-
plication. Some processes must not allow an overshoot of the process variable beyond the
setpoint if, for example, this would be unsafe. Other processes must minimize the energy ex-
pended in reaching a new setpoint. Generally, stability of response (the reverse of instability)
is required and the process must not oscillate for any combination of process conditions and
setpoints. Some processes have a degree of non-linearity and so parameters that work well
at full-load conditions don’t work when the process is starting up from no-load. This section
describes some traditional manual methods for loop tuning.
There are several methods for tuning a PID loop. The choice of method will depend largely
on whether or not the loop can be taken ”offline” for tuning, and the response time of the
system. If the system can be taken offline, the best tuning method often involves subjecting
the system to a step change in input, measuring the output as a function of time, and using
this response to determine the control parameters.
Manual tuning If the system must remain online, one tuning method is to first set the I and
D values to zero. Increase the P until the output of the loop oscillates, then the P should be
left set to be approximately half of that value for a “quarter amplitude decay” type response.
Then increase I until any offset is correct in sufficient time for the process. However, too much
I will cause instability. Finally, increase D, if required, until the loop is acceptably quick to
reach its reference after a load disturbance. However, too much D will cause excessive response
and overshoot. A fast PID loop tuning usually overshoots slightly to reach the set-point more
quickly; however, some systems cannot accept overshoot, in which case an “over-damped”
closed-loop system is required, which will require a P setting significantly less than half that
of the P setting causing oscillation.
Ziegler-Nichols method Another tuning method is formally known as the Ziegler-Nichols
method, introduced by John G. Ziegler and Nathaniel B. Nichols.
Ziegler-Nichols first method:
The first method is applied to systems with over damped step responses of the form displayed
in Fig. 2.60. Typically, first order systems delay in process control applications, having
K e−s L
transfer function with transportation delay, G(S) = (T s+1)
have such responses. The response
Effects of increasing parameters

Parameter Rise Time Overshoot Settling Time S.S. Error
Kp Decrease Increase Small Change Decrease
Ki Decrease Increase Increase Eliminate
Kd Small Decrease Decrease Decrease None
Table 2.3: Effects of changing parameters for manual tuning of PID control
is characterized by two parameters, L = the delay time and T = the time constant. These
are found by drawing a tangent to the step response at its point of inflection and noting
its intersections with the time axis and the steady state value. It should be noted that the
response curve of Fig. 2.60 is also typical of over damped second order systems.
Figure 2.60: Response Curve for Ziegler-Nichols First Method.
Ziegler-Nichols first method

Control Type Kp Ki Kd
T
P L
- -
PI 0.9 TL L
0.3
-
PID 1.2 TL 2L 0.5 L
Table 2.4: Zeigler - Nichols rule for tuning of PID control first method
Ziegler-Nichols second method:

This is a more popular technique. The second method targets plants that can be rendered
unstable under proportional control. The technique is designed to result in a closed loop
system with 25% overshoot. This is rarely achieved as the PID parameters are adjusted based
on a specific plant model.
The steps for tuning a PID controller via the 2nd method is as follows:
1. Reduce the integrator and derivative gains to 0 (only proportional feedback control).
2. Increase Kp from 0 to a critical value Kp = Kc at which sustained oscillations occur. If it
does not occur then an alternate method has to be applied.
3. Note the value Kc and the corresponding period of sustained oscillation, Pc .
The controller gains are now specified based on two parameters Kc and the oscillation period
Pc as shown in Table (2.5):
Ziegler-Nichols second method

Control Type Kp Ki Kd
P 0.5Kc - -
PI 0.45 Kc 1.2 Kp / Pc -
PID 0.6 Kc 2 Kp / P c Kp Pc / 8
Table 2.5: Zeigler - Nichols rule for tuning of PID control
Example on PID tuning by Z-N second method:

1
Consider a process with transfer function G(s) = (s+1) (s+3) (s+5)
. The limiting/ critical gain
for stability (before oscillations) can be determined by use of the Routh-Hurwitz condition.
The characteristic equation with Proportional control is: 1 + K G(s) = 0 ⇔
(s + 1) (s + 3) (s + 5) + K = 0 ⇔ [s3 + 9 s2 + 23 s + (15 + k)] = 0.
The corresponding Routh array is
s3 1 23 0
s2 9 15 + K 0
.
s1 192 − K 0
s0 15 + K
The range of K for stability is (15 + K) > 0 ⇒ K > −15 and (192 − K) > 0 ⇒ K < 192.
So the critical gain that drives the system to the verge of instability is Kc = 192. When
Kc = 192, characteristic equation has imaginary roots since the s1 row is identically 0. The
corresponding auxiliary equation is 9 s2 + (15 + 192) = 0 with roots at s = ± j4.8. Since this is
a quadratic factor of the characteristic polynomial ⇒ the sustained oscillation at the limiting
value of K, i.e., Kc, is at 4.8 rad/s. Thus, Pc = 1.31 sec and Kc = 192. This gives for full
PID control from the Table 2.5 as
Kp = (0.6 Kc ) = 115.2; Ki = 2 Kp / Pc = 177.2; Kd = Kp Td = Kp /(8Pc ) = 18.3.
Analysis: The closed loop step response (Fig. 2.61) shows an overshoot performance of 50%,
which is 100% over the targeted value. Given the dependence of the technique on a generic
model, it is not surprising that the design objectives will almost always not be met. The
technique, however, does provide an effective starting point for controller tuning.
Figure 2.61: Unit step response Curve for Ziegler-Nichols second Method.
Modifications to the PID algorithm

The basic PID algorithm presents some challenges in control applications that have been
addressed by minor modifications to the PID form. One common problem resulting from the
ideal PID implementations is integral windup. This problem can be addressed by:
• Initializing the controller integral to a desired value
• Increasing the setpoint in a suitable ramp
• Disabling the integral function until the PV has entered the controllable region
• Limiting the time period over which the integral error is calculated
• Preventing the integral term from accumulating above or below pre-determined bounds
Many PID loops control a mechanical device (for example, a valve). Mechanical maintenance
can be a major cost and wear leads to control degradation in the form of either stiction or
a deadband in the mechanical response to an input signal. The rate of mechanical wear is
mainly a function of how often a device is activated to make a change. Where wear is a
significant concern, the PID loop may have an output dead band to reduce the frequency of
activation of the output (valve). This is accomplished by modifying the controller to hold
its output steady if the change would be small (within the defined dead band range). The
calculated output must leave the dead band before the actual output will change.
The proportional and derivative terms can produce excessive movement in the output when a
system is subjected to an instantaneous ”step” increase in the error, such as a large setpoint
change. In the case of the derivative term, this is due to taking the derivative of the error,
which is very large in the case of an instantaneous step change. As a result, some PID
algorithms incorporate the following modifications:
• Derivative of output - In this case the PID controller measures the derivative of the
output quantity, rather than the derivative of the error. The output is always continuous
(i.e., never has a step change). For this to be effective, the derivative of the output must
have the same sign as the derivative of the error.
• Setpoint ramping - In this modification, the setpoint is gradually moved from its old
value to a newly specified value using a linear or first order differential ramp function.
This avoids the discontinuity present in a simple step change.
• Setpoint weighting - Setpoint weighting uses different multipliers for the error depending
on which element of the controller it is used in. The error in the integral term must be
the true control error to avoid steady-state control errors. This affects the controller’s
setpoint response. These parameters do not affect the response to load disturbances and
measurement noise.
PID control synthesis using Bode/ Nyquist plots - to be added

Recall that a good control system should have a favorable transient as well as steady state
responses. Compensators are traditionally designed in a set of two sequential design processes
(design is actually an iterative process where in two sequential processes are repeated). The
first step is design a compensator that stabilizes (for unstable plant) and improves transient
response. The second step is to improve the steady state response. Alternately, for stable
plants, these two steps can be reversed. Design of lag and lead compensators are discussed
in the §2.9. PID controller can also be designed by similar two step procedures, design of PI
and PD controllers. It is also possible to design PID controllers in a single step; this design
step is iterated until all the design specifications (primary and secondary) are satisfactorily
met. Importance to primary specifications is given in the design process. In general, PID
controller can be implemented in a series form or cascade form. The practically imple-
mentable structure of PI or PD controller is not discussed at present, as the text book form
of compensators are easy to understand/ interpret.
PID controller design using Root Locus Technique :

A text book form of PID is depicted in Figure 2.57. Its input-output equation and corre-
sponding transfer function is repeated (Eqn. 2.76) here as:
∫t
d e(t)
u(t) = Kp e(t) + Ki e(τ ) dτ + Kd . (2.77)
dt
0
Ki Kp s + Ki + Kd s 2 τi
H(s) = Kp + + Kd s = = Kp {1 + + τd s} (2.78)
s s s
which has two zeros plus one pole at the origin. In this text book form, controller transfer
function is non-causal. There are many different forms of implementable causal transfer
functions for PID compensators, which are not discussed here, since basic interpretation of
stability and performance of compensated system do not change significantly for the practical
PID case. The design technique comprises the following eight steps [45, p. 532]:
1. Evaluate the performance of the uncompensated system to determine quantum of im-

provements in transient response required.
2. Design the PD controller to meet the transient response specifications. The design
involves the zero location and the loop gain.
3. Simulate the system to be sure that all requirements are satisfactorily met.
4. Redesign if simulation shows non-compliance of specifications.

5. Design the PI controller to yield the required steady state error.
6. Determine the gains, Kp , Ki , Kd as shown in Figure 2.57.
7. Simulate the system to be sure that all requirements have been met.
8. Redesign if simulation shows that requirements have not been met.
PID controller design via an illustrative example [45, Example 9.5]: Given the system and unity
gain feedback system having an OLTF:
K (s + 8)
G(s) H(s) = , (2.79)
(s + 3) (s + 6) (s + 10)
design a PID controller so that the system can operate with a peak time that is two - thirds
that of uncompensated system at 20% overshoot and with zero steady state error for the step
input. Solution is obtained using the eight design steps enunciated above.
Step 1 To evaluate the uncompensated system operating at 20% overshoot, vary the cascade
gain K and plot the root locus. For a damping ratio of 0.456 (0.455 in the Figure 2.62)
corresponding to 20% overshoot, the dominant poles are located at −5.41 ± j 10.6 with a gain
of 121.5 (122 in the figure). The third closed loop pole is located at −8.169 (not marked in the
figure) which lies in the locus segment [−8, −10] with a gain of 121.5. The estimated peak
time for the uncompensated system is 0.297 sec with an overshoot of 20%. The performance
characteristics of uncompensated system is given in the first column of Table 2.6.
Step 2 To compensate the system to reduce the peak time to two - thirds that of uncompen-
sated system at 20% overshoot, it is necessary to find compensated system’s dominant pole
location. The imaginary part of the compensated dominant pole ωd is related to the peak
time by a simple formula:
π π
ωd = = = 15.87 rad/sec
Tp (2/3) ∗ (0.297)
The real part of the complex pole pair is determined by the per cent overshoot of 20%, (i.e.,
tan−1 (ζ) = tan−1 (0.456) = 117.13◦ , angle being measured from positive real axis in s-plane):
σ = ωd /tan(117.13) = − 8.13
Therefore, the desired dominant pole is located at − 8.13 ± j 15.87 . Sum of all angle con-
tribution at the dominant pole of the compensated system should be −180◦ . Sum of angles
Root Locus
15
10
System: sys
Gain: 122
Pole: −5.41 + 10.6i
Damping: 0.455
Imaginary Axis
−5
−10
−15
−12 −10 −8 −6 −4 −2 0 2
Real Axis
Figure 2.62: Root locus for the uncompensated system for PID design example.
contributed by the uncompensated poles at −3, −6, −10 and the zero at −8 at the desired
compensated pole is equal to −198.37◦ . Thus angle contribution needed from the compen-
sator zero at −zc is 198.37 − 180 = 18.37◦ . Angle subtended by the zero −zc at the pole
− 8.13 ± j 15.87 is given by
15.87
= tan 18.37◦ , ⇒ zc = 55.92
(zc − 8.13)
Thus, the PD compensator (text book type) is GP D (s) = (s + 55.92). The root locus plot of
PD-compensated system is plotted in the Figure 2.63 [45, Fig. 9.34]. The gain at the design
point − 8.13 ± j 15.87 is 5.34. The performance characteristics of PD-compensated system
are given in the second column of Table 2.6.
Step 3 \& 4 Unit step response of the PD-compensated systems are given in the Figure
2.64. This figure shows marked improvements in peak time (by 2/3 factor) and steady state
error over the uncompensated system.
Step 5 Following the design of PD-controller, an ideal integral compensator is appended to
reduce the steady sate error to zero for a step input. Any ideal integrator (PI control) will
work, as long as pole - zero pair is placed close to the origin. Choosing the ideal integrator at
s + 0.5
GP I (s) =
s
⇒ OLTF for PID compensated system
Uncompensated PD-compensated PID-compensated

K (s+8) K (s+8) (s+55.92) K (s+8) (s+55.92) (s+0.5)
Plant and compensator (s+3) (s+6) (s+10) (s+3) (s+6) (s+10) (s+3) (s+6) (s+10) s
Dominant poles −5.41 ± j 10.6 − 8.13 ± j 15.87 −7.516 ± j 14.67

K 121.5 5.34 4.6
ζ 0.456 0.456 0.456
ωn 11.88 17.83 16.49
% OS 20 20 20
Ts 0.739 0.492 0.532
Tp 0.297 0.198 0.214
Kp 5.4 13.27 ∞
ess = e(∞) 0.156 0.070 0
Other poles -8.169 -8.079 -8.009, -0.468
Zeros -8.0 -8, -55.92 -8, -55.92 -0.5
Comments Second - order Second - order Zeros at -55.92 and
approx - OK. approx - OK. -0.5 not canceled
Table 2.6: Characteristics of uncompensated, PD-compensated and PID compensated systems

of PID design example
K (s + 8) (s + 55.92) (s + 0.5)
G(s) H(s) =
(s + 3) (s + 6) (s + 10) s
The figure 2.65 ([45, Fig. 9.36]) shows root locus plot for PID compensated system by varying
K. On tracing the closed loop poles along the root locus corresponding to the dominant poles
till the damping ratio line for 0.456, the dominant pole pair is found at −7.516 ± j 14.67, with
an associated gain K = 4.6. The performance characteristics of PID-compensated system are
given in the third column of Table 2.6. As expected, addition of lag due PI controller, makes
the system sluggish, thereby, increasing the peak time from 0.198sec to 0.214sec, which is an
increase of 0.016sec. This slight increase may be acceptable. If not, PD controller has to be
redesigned such that peak time is reduced by slightly more than 0.016sec to a requirement of
Tp ≤ 0.182sec.
Step 6 From the above discussion transfer function for the PID controller is
K (s + 55.92) (s + 0.5) 4.6 (s + 55.92) (s + 0.5) 4.6 (s2 + 56.42 s + 27.96)
GP ID (s) = = =
s s s
Figure 2.63: Root locus for the PD - compensated system for PID design [45, Fig. 9.34].
Figure 2.64: Unit step responses for the uncompensated, PD - compensated and PID - com-
pensated system for PID design [45, Fig. 9.35].
Figure 2.65: Root locus for the PID - compensated system for PID design [45, Fig. 9.36]
On comparing/ matching with the Equation (2.78) of like powered s-terms, the PID filter
coefficients are obtained as Kp = 259.5, Ki = 128.6 & Kd = 4.6.
Step 7 \& 8 On examining the unit step response in the Figure 2.64, it is observed that
The steady state error is zero and it will remain so independent of input magnitude. Owing
to finite error coefficient, PD compensated system will exhibit larger magnitude of error for
larger input signal. More over, the PD compensated system will have larger error in tracking
velocity or ramp input. However, the peak time for PID compensated system is slightly
higher than the PD compensated system. On iterating the steps 2 and 5, one can reach the
desired specifications on peak time too. From the Figure 2.64, it is seen that the final value
is reached after a long period of 3 seconds. If this is undesirable, the speed of the system
must be increased by redesigning the ideal derivative controller or moving PI controller zero
little further from the origin. For better understanding of the higher order system of PID
compensated system, simulation is important since the compensated system does not behave
like a dominant second order system.
2.11 Control System Design Issues
The flight control systems/ autopilots are designed to improve the stability and performance.
Some of the current day vehicles have multiple actuation or control inputs and multiple outputs
or sensors. The typical inputs could be from thrust vector control, aerodynamic control from
either or combination of tail, wing or canard. The sensors such as position and rate gyros,
accelerometers, angle of attack and mach number sensors are used in various combinations in
aerospace vehicles. In order to effectively make use of the multiple degrees of freedom available
for the flight control system/ autopilot design, it is desirable to use modern control synthesis
procedures. Even for the single input systems, the modern control synthesis can provide
better solutions over classical frequency domain (one loop at a time / SISO) techniques.
The modern techniques for deterministic/ stochastic or quasi- stochastic processes encompass
modal control, use of positivity concepts, pole placement/ eigen structure assignment strategy,
optimal control, H2 /H∞ optimization and numerical optimization. A perspective of past
developments and challenges to control is well summarized in the Editorial [11]. There exists
a number of references that derive benefits from cross-fertilization of ideas in their design
philosophies.
In designing the autopilot, an important property requirement is that the perfor-
mance of the closed loop system should not deteriorate substantially when small changes
in the operating condition occurs. The variations in the operating conditions reflect on the
parameters of the linearized plant. However, most practical controllers in future require
maintenance of high performance even in the presence of large changes in the operating con-
ditions. Therefore, in designing the autopilot, robustness for stability and performance is
the most important singular criterion that must be addressed to [12, 13]. The present day
vehicles are either statically unstable or having poor static stability margins. Uncertainties
in mathematical model describing the physical behavior of a system arise mainly from two
sources: inadequate information of the system being modelled, like unpredictable dynamics
and unknown or unpredictable inputs, including external disturbances and noise. The modern
synthesis techniques offer significant promise in attainable vehicle performance.
Almost all of the earlier methods of controller design are carried out in continuous/
Laplace (s) domain. One of the very early work on frequency response analysis is compiled
2.11. CONTROL SYSTEM DESIGN ISSUES 163
by Oldenburger [14]. It is the development of classical analysis techniques which enabled the
designer to identify the set of primary and secondary system specifications and combine with
the system requirements along with tertiary system requirements. These are also some times
grouped under time domain and frequency domain specifications [15, Ch.8,9; 16, Ch. 4; 10,
Ch. 4,6]. Typical frequency domain specifications are [16, 10] :
• Control bandwidth,
• M - peak,
• Stability requirements
• Gain and phase margins at all cross over frequencies.
The modern trend is to combine the various methods in order to take advantage of all points
of view. In fact it is realized that none of these approaches is wholly satisfactory because
the criteria used are artificial. Rarely of ever is a user of a control system interested in
phase margin or M peak as such. The analytic-design or state space in principle appears
to be improvement; however, manipulative difficulties are not entirely absent with presently
used criteria. A list of the more common control-system specifications and certain of their
attributes is given below.
State space or matrix methods were are used in time-domain analysis and synthesis.
All the classical design methods were designed primarily for control systems with a single input
and a single output. However, many practical systems have multiple inputs and outputs. Or,
saying this another way, there are cases in which several adjacent systems will have cross-
feeding of signals. If this cross-feeding or interacting is strong enough to be a first-order effect,
it can result in deterioration of performance if not controlled. The time domain specifications
are [16, p.161]:
1. Transient response specifications
• rise time, (or delay time, or peak time),
• percent over shoot,
• settling time
2. Steady state and tracking specifications
• steady state tracking accuracy,
• steady state error coefficients;
3. Dynamic error specifications,
4. Disturbance rejection,
• steady state,
• transient,
5. Control effort.
• maximum magnitude of u,
∫
• energy: K {u2 dt} .
6. Sensitivity to parameter changes.
The spectrum of frequency domain methods can be put under two categories, viz., classical
or modern. The Bode diagram, the Nyquist diagram, the Nichols chart, and the root-locus
techniques are graphics oriented. The advantages of the classical methods are that:
- guaranteed robustness against the plant parameter variations in the plant model can be
ensured,
- limited robustness against disturbances can be achieved,
- methodology of design are well established, i.e., confidence level on the controller design
procedure is high,
- simple to use graphical procedures,
- stability margins can be pre-specified,
- compensator can be included to improve the system performance.
The major disadvantages of the classical techniques are that:
- it is better suited for Single Input Single Output (SISO) systems,
- it produces conservative design,
- only linear time invariant (LTIV) system can be handled,
- low order dynamical system can only be tackled comfortably as complexities in terms of the
compensator/ controller design increase with the dimensions of the plant model,
- translation of system specification to a large order model is non-unique,
- lots of trial and error is required.
To arrive at a good design for controller K, a priori information about the expected
disturbances and reference inputs, and of the plant model (G) and disturbance model (Gd )
are needed. A major source of difficulty is that the models (G, Gd ) may be inaccurate or may
change with time. In particular, inaccuracy in G may cause problems because the plant will
be part of a feedback loop. To deal with such a problem, the concept of model uncertainty
is made use of. Therefore, dealing with and understanding the effects of uncertainty are
important tasks for a control engineer. Reducing the effects of some forms of uncertainty
(initial conditions, low-frequency disturbances) without catastrophically increasing the effects
of other dominant forms (sensor noise, model uncertainty) is the primary job of the feedback
control system.
Closed-loop stability is the way to deal with the (always present) uncertainty in
initial conditions or arbitrarily small disturbances. High-gain feedback in low-frequency ranges
is a way to deal with the effects of unknown biases and disturbances acting on the process
output. In this case, one is forced to use roll-off filters in high-frequency ranges to deal with
high-frequency sensor noise in a feedback system.
Finally, notions such as gain and phase margins (and their generalizations) help
quantify the sensitivity of stability and performance in the face of model uncertainty, which is
the imprecise knowledge of how the control input directly affects the feedback variables. The
stability margins are defined with reference to Fig. (2.47). The positive phase margin is the
smallest value of ϕ > 0 such that the compensator K(s) with additional phase lag = ejϕ in
verge of instability. The upward gain margin (GM) is the smallest value of additional value
of ∥K(s)∥ = constant > 0 for which the become unstable (GM expressed in decibels). The
downward GM or gain reduction margin is similarly defined.
The graphical techniques are difficult to apply for multi-input/multi-output, or
multi-loop systems. Due to the interactions among the control loops in a multivariable sys-
tem, even that each single-input/single-output transfer function has acceptable properties
concerning the step response and robustness, the whole system can fail to be acceptable.
Limitations of GM and PM:

For MIMO systems, closing one loop at a time and designing the controllers for each of these
loops individually, can limit the achievable performance of the closed loop system as compared
to a multivariable design. For example, consider a two input two output system [34]:
 
1 0.5
 s+1 s+1 
G(s) =  
1
0 s+2
which has a high degree of coupling. For a separate two loop controller from y1 to u1 and
from y2 to u2 , the feedabck gain K is obtained successively by:
 
k1 0 
K=
 .
0 k2
For controller K in cascade path, the closed loop transfer function matrix
Gc = (I + GK)−1 GK and
 
k1 0.5k2 (s+2)
 (s+1+k1 ) {(s+1+k1 ) (s+2+k2 )} 
Gc (s) =  
k2
0 (s+2+k2 )
The outputs y1 and y2 are coupled and coupling becomes stronger if mode 2 at -(2 + k2 ) is
made faster (so that second mode has little effect on the first output). On the other hand, if
the controller K is selected such that:
 
k1 k3 
K=
 ,
0 k2
then the closed loop system will have the following transfer function matrix:
 
k1 (k3 +0.5k2 ) (s+2)
 (s+1+k1 ) {(s+1+k1 ) (s+2+k2 )} 
Gc (s) =  .
k2
0 (s+2+k2 )
By making k3 = - 0.5 * k2 , the two modes get decoupled for the nominal plant. For small
plant parameter perturbations, coupling will be small.
For SISO system, gain margins and phase margins are used as a good figure of merit
to indicate robustness. For weakly multivariable systems, these can be used with caution.
Consider a perturbed plant:
 
1 0.5
G∆ (s) = 

s+1 s+1 
.
ϵ 1
s+2 s+2
For a multivariable system break one loop at a time and stability margins are computed for
each loop. That is, for multi-loop systems, variations are made with all gain and phase values
in the feedback paths The OLTF for the two loops are held at nominal values except for one
path under investigation. These can give misleading results because they fail to to check the
effect of simultaneous variations in several paths. For example, once again consider above
system [34] with a decoupled controller. The OLTF for the two loops are k1 /(s + 1) and
k1 /(s + 2). These Transfer functions indicate infinite gain margin and a minimum of 90 deg
phase margins. Hence, it is expected that the closed loop system is highly robust.
The corresponding characteristic equation for the closed loop is given by:
s2 + (3 + k1 + k2 + ϵ k3 ) s + (2 + k2 + + ϵ k3 + 2 k1 + k1 k2 − k1 k2 ϵ) = 0
If k3 ≫ k1 , k2 , then a small negative ϵ can make the closed system unstable since (3 + k1 +
k2 + ϵ k3 ) can become negative. Hence, the closed loop system has very little robustness. This
conclusion can not be derived from the individual stability margins of two loops.
Since, Gain margins and phase margins can not be used reliably for all MIMO
systems as a measure of robustness (figure of merit); Concept of singular values are used by
extending many frequency domain concepts and various measures of robustness are defined
based on these singular values. Singular vectors are good indicators of strong/ weak input or
output directions.
2.11.1 Singular Values and Singular Value Decomposition (SVD)
In linear algebra, the singular value decomposition (SVD) is an important factorization of a

rectangular (i.e., not necessarily square) real or complex matrix. Let An×n be any matrix of
rank ’r’, Un×n and Vn×n be the unitary matrices such that (U ∗ U = I, V ∗ V = I). For any
matrix, there exists unitary matrices U and V such that
∑
A=U V∗ (2.80)
∑ ∑∗ ∑∗ ∑
where U satisfies AA∗ U = U and V satisfies A∗ AV = V
∑
where has the canonical structure :
 
∑
∑ 0  ∑
=
o
 , = diag [σ1 , . . . , σr ],
0 0 o
σi > 0, σi+1 > σi , i = 1, 2, . . . r.

The values σi , i = 1, . . . , r are called singular values of A. σ(A) = max. singular value = σ1 .
Similarly, σ = σmin is defined. The singular values σ1 , i = 1, . . . r can also be calculated from
σi2 = λi {AAT } = ith nonzero eigenvalue of (AAT )
= ith nonzero eigenvalue of (AT A).
For non-square matrix A, AT A and AAT will have different number of nonzero eigenvalues
(hence singular values) which is equal to rank of A. [Note A∗ = AT for real A]. Non-degenerate
singular values always have unique left & right singular vectors, except for a scalar multiplying
factor. SVD is numerically stable and well-conditioned. The columns of U & V are, respec-
tively, left- and right-singular vectors for the corresponding singular values. In Eq.(2.80),
- The columns of V form a set of orthonormal “input” or “analysing” basis vector directions
for A. (These are the eigenvectors of A∗ A.)
- The columns of U form a set of orthonormal “output” basis vector directions for A. (These
are the eigenvectors of A A∗ .)
∑
- The diagonal values in matrix are the singular values, which can be thought of as scalar
“gain controls” by which each corresponding input is multiplied to give a corresponding out-
put. (These are the square roots of the eigenvalues of A A∗ that correspond with the same
columns in U and V .)
- An m × n matrix A has at least one and at most p = min(m, n) distinct singular values.
Example 13 :
A = [11]
∗
∑∑
T
AA = 2 hence, U = 1 and σ12 =2=
     
 1 1   1 1   2 0 
AT A =   =⇒   [v
−1 , −
v ] = [v
− v ]
− 
2 1 2
1 1 1 1 0 0
 
 1 1 
=⇒ [v1 , v2 ] =  
1 −1
   
∗ ∑
∑
1 1 1  2 0 
Af ter normalizing V =    & =
 
2 1 −1 0 0
∑ √ ∑ √
T heref ore = [ 2, 0] and = 2
0
 
√  1 1  1
hence A = (1) [ 2, 0]   √
1 −1 2
Example 14 [27] :
   
 1 1  T  2 2 
A =   =⇒ AA =  
2 0 2 4
√ √
λ{AAT } = 3 + 5, 3 − 5
 √ 
 3 + 5 0  √ √
√  V, σ1 = 3 + 5, σ2 = 3 − 5
2 2
A = U
0 3− 5
U and V are left and right eigen vectors of a square real matrix AAT . Another example of
SVD of a matrix A is given by
     
 5 4   0.872 0.490   7.343 0   0.794 −0.608 
A =   =    
3 2 0.490 −0.872 0 0.272 0.608 0.794
The singular value decomposition of a matrix A can calculated using the MATLAB command:
[U, S, V ] = svd(A)
If ||x|| denotes any vector norm, then an induced matrix norm is defined by
||Ax||
||A|| = sup
||x||
if Euclidean norm is used, then
√
||x|| = x∗ x (if x is complex)
√
= xT x (if x isreal),
and the induced matrix norm is the Hilbert or spectral norm
||G||s = σ̄
where σ̄ 2 is maximum eigen value of G∗ G or GT G (or GGT or GG∗ ). Instead of constant

matrix G, if the matrix transfer function (MIMO system) G(s) is used, then set s = j ω
(o ≤ ω < ∞) , then singular values of G(j ω) are functions of ω. They are called principal
gains of G(s), and are denoted by σi (ω), i = 1, 2, . . . , m (m < r)
A useful characteristics of a matrix A is its condition number which is defined as
σ̄(A)
cond(A) = , σ
−(A) = minimum singular value of the matrix A.
σ
−(A)
Note that in numerical analysis, the condition number measures the difficulty of inverting a
matrix. Properties: For square matrices A, B: |σ(A + B) − σ(A)| ≤ σ(B); σ(A−1 ) = 1
σ(A)
if
A−1 exists; and σ(A B) ≥ σ(A)σ(B).
Physical interpretation of SVD to the frequency response of MIMO system G(s)
with m - inputs and r - outputs is as follows: Consider a fixed frequency where G = G(jω) is
a constant p × m complex matrix. Then, G = U Σ V H where, Σ is a p × m with ℓ = min(r,
m) non-negative singular values, σi arranged in descending order along its main diagonal; the
other entries being zero. U and V are unitary matrices. The singular values are positive
√
square roots of the eigenvalues of GH G, i.e., σi (G) = λi (GH G).
The singular values are also known as principal values or principal gains and associated di-
rections are called principal directions. The column vectors of U , i.e., ui represent the output
directions of the plant. They are orthogonal and of unit length (ortho-normal), i.e., ∥ui ∥2 = 1
i ui = 1; ui uj = 0; i ̸= j. Similarly, the column vectors of V , i.e., vi are ortho-

and uH H
normal vectors representing the input directions of the plant. SVD of G can be written as
G V = U Σ, which is written in i − th component form as G vi = σi ui , where σi is a scalar
and ui , vi are vectors. The ith singular value gives directly the gain of the matrix G(jω) (unlike
∥G vi ∥2
ith eigenvalue); i.e., σi (G) = ∥G vi ∥2 = ∥vi ∥2
.
However, it is useful to combine the properties eigenvalues and singular values through the
bounding relations: σ[G] ≤ |λ[G]| ≤ σ[G], which holds for any λi of a square matrix.
Maximum and minimum singular values: correspond to maximum and minimum gains. The
vector v corresponds to the input direction with largest amplification and u corresponds to the
output direction in which the inputs are most effective. Directions involving v, u corresponds
are called ‘strongest’ or ‘high-gain’ or ‘most important’ directions. Similarly, the minimum
singular values σ(G) of the plant, evaluated as a function of frequency, is a useful measure of
evaluating the feasibility of achieving acceptable control. Generally, σ(G) should be as large
as possible. Larger σ(G) also minimizes occurrence of actuator saturation.
Example 15: Consider the following model of short period dynamics of F-16 aircraft [1]:
 
 −0.5164 1.0000 
A =  
1.4168 −0.4932
     
−0.5600 0.8285   1.7546 0   0.8338 0.5521 
=⇒ svd(A) = 
     
0.8285 0.5600 0 0.6623 −0.5521 0.8338
2.12. DESIGN OF CONTROLLERS 171
From the first singular vector, v = [0.8338 − 0.5521]T and corresponding gain is 1.7546. On
the other hand, from the second singular vector, v = [0.5521 0.8338]T and corresponding
gain is only 0.6623. Reason for this is that the inputs to the two modes counteract with each
other. That is, if gain to q-feedback is increased, it enhances damping but relatively reduces
effect of stiffness and α - feedback increases speed of response with reduced damping.
It was believed in the late 60’s that the traditional optimal control/ LQG based
controller provided enough stability robustness for the closed loop system. The positive defi-
nite Q and R weighting matrices are expected to provide a minimum of 6db gain margin and
60◦ phase margin. In these designs, the variation in system parameters are finite but unknown
in its value. Analysis by Stein et al in 1981 [12, 26] showed the weakness of LQG design. Then,
the search for robust controller began and robustness requirement are introduced as a part of
the specifications. Hence, it is necessary to understand various issues related to robustness
(i.e., robust stability and robust performance) against uncertainty.
To summarize, the robust multivariable feedback control system design is, therefore, can be
deemed to be extension of loop shaping (Bode design) of SISO system, with additional freedom
for optimization in the design process (due to larger design freedom as compared to SISO).
2.12 Design of Controllers

• State Variable Feedback
• Output Feedback
• Dynamic Output Feedback
Methodology of Control System Design
− Pole Assignment and eigenstructure Assignment
− Positivity/Liapunov Method
− Singular value Decomposition
− Stable Factorization
− Quantitative Feedback Theory

− Optimal Control :
Linear Quadratic Regulator (LQR)
Linear Quadratic Gaussian (LQG)
Linear Quadratic Gaussian/ Loop Transfer Recovery (LQG/LTR)
− Variable Structure Control
− H2 /H∞ Control
− Adaptive Control -
Self Tuning Model Following
Implicit/Explicit
− Linear Matrix Inequality (LMI) Control
− Fuzzy control/ Neural network + AI control
− Genetic Algorithm Based optimal Control
Selection of particular technique for controller synthesis depends on vehicle specific appli-
cations, user requirement and most importantly familiarity of the technique to the system
designer.
Chapter 3
Control of Aerospace Vehicles -

Classical Control Perspective
Primary objective of a flight control system is stabilize (or enhance stability margin) a flight
vehicle and to enable the vehicle to perform to a desirable level in the presence of system
uncertainty, external disturbances and sensor noise.
A brief account of the linearized equation of motion of Flight vehicles are given in
the Chapter 1. System analysis in frequency and time domain, Asymptotic and BIBO stability
analysis are briefly discussed in the Chapter 2. Chapter 2 also gives a brief overview of PID
controller, which are popular with process industries. Analysis of flight control systems for
aerospace vehicles using tools discussed in the Chapter 2 will be presented in the sequel.
3.1 Satellite Attitude Control

Linear model of Three Axis Stabilized Satellites
The linearized equations of motion for a three-axis stabilized satellites are rewritten as:
Roll : Ix ϕ̈ = −hw Ω0 ϕ + hw ψ̇ + Mx + Tx + Tgx (3.1)
P itch : Iy θ̈ = My + Ty + Tgy (3.2)
yaw : Iz ψ̈ = −hw Ω0 ψ − hw ϕ̇ + Mz + Tz + Tgz (3.3)
where, Ix , Iy , Iz are MI of the satellites about roll, pitch and yaw directions( assumed to be
the principal axes), hw wheel angular moment (nominal), Ω0 = orbital angular velocity Tg
173
174CHAPTER 3. CONTROL OF AEROSPACE VEHICLES - CLASSICAL CONTROL PERSPECTIVE
= gravity gradient torque, T = other disturbance torques, M = control torque. For a large
number of satellites, Ix ≈ Iz and hence Tgy ≈ 0. For biased momentum wheels, the torque
about the pitch axis is generated by accelerating or decelerating the wheel (clock wise or anti
clock wise accelerations) and for the other two axes either reaction jets are magnetic torquer
or both are used. For the satellites in the inclined orbits, (Example: IRS satellites) all the
three reaction wheels have zero stored angular momentum (nominal zero angular speeds).
Hence, the term hw = 0, which results in decoupling of roll - pitch - yaw dynamics. Hence,
all the three degrees of freedom have similar equations of motion. Thus, identical controller
structure can be used for all axes in these satellites. For the communication satellite, like
INSAT satellites, separate controller for pitch and roll - yaw dynamics are needed. (Control
analysis is taken from Kaplan’s book on Spacecraft Dynamics and Control [8, Ch. 6, §6.2].)
Pitch attitude stabilization system
Design of an automatic pitch control system is straightforward because pitch motion is de-
coupled from the others. The linearized pitch equation can be written as
Ty = Iy θ̈ + 3Ω20 (Ix − Iz )θ + My
This is further simplified since the satellite of interest is symmetric i.e., Ix = Iz . Thus,
Ty = Iy θ̈ + My
where My represents rate of change of wheel speed so as to impose direct torque about the
pitch axis. Then, the transfer function for the idealized pitch dynamics is given by
θ(s) 1
= (3.4)
My (s) Iy s2
which indicates that the dynamics is similar to the double integrator plant. This equation
(3.4) indicates that no natural damping and stiffness terms are available and they can be
provided only through the control function hw . A typical pitch control loop is depicted in the
following Figure. A satisfactory form of control law is
My = Kp (τp θ̇ + θ)
where the attitude rate term θ̇ introduces damping. Kp and τp are the pitch autopilot gain
and time constant, respectively. Pitch motion now becomes that of the classic, damped second
3.1. SATELLITE ATTITUDE CONTROL 175
Pitch − Control +
+ − Satellite
Sensor logic dynamics
Figure 3.1: Pitch control loop.
order system
Ty = Iy θ̈ + Kp τp θ̇ + Kp θ (3.5)
The associated transfer function is obtained directly as
θ(s) 1
= 2
(3.6)
Ty (s) Iy s + Kp τp s + Kp
In the nominal operating mode, θref = 0, and sensor errors are ignored in preliminary consid-
erations. Thus, pitch motion has a natural frequency and damping factor given by
√ √
Kp τp Kp
ωp = , ζp =
Iy 2 Iy
Since overshoot is not desirable for such applications, parameter value selections are begun
by setting ζp = 1 (critical damping). A greater value would slow system response, while a
smaller value permits overshoot. Thus, equation (3.8) becomes
θ(s) 1
=
Ty (s) Iy (s + ωp )2
Selection of pitch autopilot gain Kp is based on two considerations, steady state error and
response time. Since the limit allowed in pitch is specified, steady state errors must be well
below this. The maximum magnitude of solar pressure torque is expected to be 10−4 N.m
about the pitch axis (See Table 6.1 in Kaplan’s book on Spacecraft Dynamics and Control).
Assuming τp is much less than the orbital period, the maximum steady state error is estimated
through the final value theorem as
10−4
Θss = rad
Iy ωp2
Selecting a gain value of Kp = 0.275N.m/rad gives ωp = 0.025/rad/s and τp = 80s. This

results in a maximum expected pitch error of 0.02◦ and validates the assumption. Note that
τp << orbital period.
A root locus diagram for the pitch loop is shown in Figure 3.1 to check stability
and evaluate gain value selections. This form is typical of second systems with two open loop
poles at s = 0 and a single zero on the negative real axis. It is apparent that the system is
stable for all values of Kp . The design value of 0.275N.m/rad is depicted on the real axis,
which is consistent with critical damping. ζp = 1.
0.015
0.010
Design point 0.005
Kp = 0.275 Nm rad
a
−0.030 −0.025−0.020−0.015 −0.010 −0.005
−0.0
05
−0.010
−0.015
Figure 3.2: Root locus diagram for the pitch loop.

(seconds)
0.2
T τ/I
N
θ
0.1
0
40 80 120 160 200 240
Time ( Seconds)
Figure 3.3: Pitch response to impulsive disturbance torque.

0.05
0.04
0.03
0.02
0.01
0
40 80 120 160 200 240
Figure 3.4: Pitch response to step disturbance torque.
0.05
0.04
0.03
0.02
0.01
0
6 12 18 24 30 36 42 48
−0.01
−0.02
−0.03
−0.04
−0.05
Figure 3.5: Pitch response to solar pressure torque.
Responses produced by the pitch autopilot, with the gain and damping values
selected above, are predictable analytically or numerically. Results for impulsive, step, and
cyclic disturbance torques for the satellites of mass 716kg, Ix = Iz = 200N.msec2 ; Iy =
400N.msec2 are plotted in Figures (3.3, 3.4, 3.5). Since it is difficult to estimate the magnitude
of impulsive torques which may act on the satellite, the response shown in Figure (3.3) is
parameterized by the factor Ty τp /Iy . A step input of magnitude 8.5 ×10−5 N.m.representing
as estimate of thruster misalignment torque about the pitch axis, produced the response of
Figure(3.4). Both impulsive and step disturbances result in stable responses with the latter
yielding a steady state error of 0.0177◦ well below the allowable limits of 0.05◦ .
A cyclic disturbance torque due to solar pressure causes the periodic response of
Figure (3.5) with amplitude of about 0.02◦ . Noting that the control term. My can be expressed
in terms of wheel speed, equation (3.4) becomes
Iw Ω̇(s) = Kp (τp Θ̇ + Θ) (3.7)
where Ω is the wheel speed and Iw is the wheel moment of inertia about its symmetry axis.
Effects of disturbance torques on wheel speed can be determined by applying the final value
theorem. Assuming the initial wheel speed is nominal Ωw the Laplace transform of expression
(3.7) gives
Kp Ωw
Ω(s) = (τp s + 1)Θ(s) +
Iw s s
Apply equation (??) to eliminate Θ(s).
Kp (τp s + 1) Ωw
Ω(s) = 2
Ty (s) +
Iw Iy [s (s + ωp ) ] s
An impulsive disturbance torque of magnitude Ty leads to
Ty
Ωss = + Ωw
Iw
indicating that a net change in wheel speed occurs even though θss = 0. A step disturbance
torque leads to infinite wheel speeds if constant counter torquing is not provided through
reaction jets or other devices. Thus, step disturbance lead to continual momentum dumping
or wheel saturation. Cyclic disturbance torques result in cyclic wheel speeds centered about
Ωw . For solar torques considered here the amplitude of wheel momentum oscillation with
respect to hw is 1.37 N.m.s. This knowledge will aid in sizing hw and the wheel torque motor
as well as influencing saturation levels.
Roll-Yaw attitude stabilization system
Now that the pitch loop has been designed, control of the coupled roll/yaw system can be
attempted. Coupling between roll and yaw axes is a result of the biased momentum wheel
(nonzero nominal speed). An attractive feature of bias momentum is that it does provide roll
/ yaw coupling while permitting accurate yaw control without a direct yaw sensor. Control
torques are produced through gimbal deflections. Since direct roll sensing is available through
the horizon sensor, this axis can be controlled directly in response to roll errors. Yaw deviations
are sensed indirectly through coupling with the roll axis. This technique is sometime referred
to as gyrocompassing. Bias momentum magnitude hw and system gain are selected to make the
roll controller sensitive to yaw errors (i.e., artificially increasing the coupling effect). However,
this must be done so that roll errors are not transformed into yaw errors. The general linearized
equations for roll and yaw were given as expressions (1.8 or 3.1, and 1.9 or 3.3), respectively.
The linearized roll / yaw equations (3.1 & 3.3) are repeated here for convenience as:
Tx = Ix ϕ̈ + Ω0 hw ϕ + hw ψ̇ + Mx (3.8)
Tz = Iz ψ̈ + Ω0 hw ψ − hw ϕ̇ + Mz (3.9)
where hw = nominal wheel angular momentum. It is interesting to briefly consider uncon-

trolled (gimbals fixed) yaw response. Set of equations (3.8, 3.9) are expressed in Laplace
variable form as
    
2
 Tx (s)   Ix s + Ω0 hw hw s   ϕ(s) 
 =  
Tz (s) −hw s Iz s2 + Ω0 hw ψ(s)
For controlling the coupled roll - yaw dynamics, one can use one or two sets of - thrusters
or magnetic torquers. Owing to the quarter orbit coupling, roll rate sensor along with the
controller will also null the yaw error. If faster yaw correction is needed, two sets of actuators
are needed. Since, reaction thrusters use consumable / propellants stored onboard, on-off
control such as PWPFM (pulse width pulse frequency modulation) controller is being used.
These controllers are fuel efficient version of an equivalent PID / PD- controller. The transfer
function for the roll response ϕ(s) for the roll and yaw axis disturbance torques [Tx (s), Tz (s)]
are given by
Tx (s)(Iz s2 + hw Ω0 ) − Tz (s)hw s
ϕ(s) = (3.10)
Ix Iz s4 + [(Ix + Iz )hw Ω0 + h2w ]s2 + h2w Ω20
Similarly, yaw response to a yaw disturbance torque Tz can be solved from:
ψ(s) Iz s2 + Ω0 hw
=
Tz (s) Ix Iz s4 + [(Ix + Iz )hw Ω0 + h2w ]s2 + h2w Ω20
A step yaw disturbance torque yield the response

Tz Ix Tz hw
ψ(t) = (1 − cos(Ω0 t)) + 2 [1 − cos( )t]
Ω0 hw hw Ix
noting that Ix = Iz . Two phenomena are represented here. The first periodic term is associ-
ated with roll coupling into yaw at orbital rate ω0 , while the second is related to precessional
effects of the momentum wheel. The latter motion is a short period oscillation whose fre-
quency and magnitude depends on the value of nominal momentum.
The characteristic equation for the open loop system is
∆OL = Ix Iz s4 + [(Ix + Iz )hw Ω0 + h2w ]s2 + h2w Ω20 = 0
or ∆ = (s2 + wN
2
)(s2 + Ω20 ) = 0
hw
√
where ωN = I
= nutational frequency and I = Ix Iz . It may be noted that the characteristic
equation does not contain terms with s3 and s. Hence, roll-yaw motion is an undamped
oscillatory motion. In order to enhance the speed of disturbance of rejection (related to closed
loop frequency) and damping, PD - controller is needed.
The roll control law is considered first because direct measurement of ϕ is available
from the horizon sensor. Control terms in the equation (3.8) must be related to ϕ such that roll
responses to disturbance torques are fast and well damped. This will also minimize coupling
of roll errors into yaw. For the sake of illustration, assume reaction torquer about the roll
axis which use the following single axis feedback law (implemented normally by a PWPFM
control)
Mx = (k1 + k2 s)ϕ(s); Define k1 = K; k2 = Kτ
or Mx = Kτ ϕ̇ + Kϕ − Ω0 hw ϕ
where Mx is the roll control torque. K and τ are the roll autopilot gain and time constant,
respectively. Notice that the first term on the right side introduces damping while the second
and third terms transform the coefficient of ϕ in equation (3.8) from ω0 hw to roll autopilot
gain resulting in
Tx = Ix ϕ̈ + K τ ϕ̇ + K ϕ + hw ψ̇ (3.11)
This represents a classic second-order system with driving function Tx and −hw ψ̇ the gyro-
scopic coupling term which permits yaw errors to appear in roll sensor outputs.
The ideal yaw response would be well damped and decoupled form roll. However,
the yaw control law cannot provide direct damping because yaw angles are not measurable by
this system. Therefore, the control law uses roll angle and a pseudo-rate modulator (PWPFM
modulator) to form the complement of equation (3.11), derive yaw torque using roll angle ϕ
and pseudo-angular rate ϕ̇ as: Mz = hw ϕ̇ − k (K τ ϕ̇ + K ϕ).
To investigate system stability and responses, roll/ yaw equations with the above
control law for Mx and Mz , governing equations are rewritten in Laplace form as [8, Eqn.6.74]:
    
2
 Tx (s)   Ix s + K(τ s + 1) hw s   Φ(s) 
 =  
Tz (s) −kK(τ s + 1) Iz s2 + Ω0 hw Ψ(s)
which yield a fourth-order characteristic equation,
C.E. = Ix Iz s4 + KIz τ s3 + (KIz + Ix Ω0 hw + kKτ hw )s2
+ (Ω0 hw Kτ + kKhw )s + Ω0 hw K = 0 (3.12)
The closed loop system, now contains terms in s3 and s and it is damped. Owing to the
quarter orbit coupling, the yaw error also gets minimized. However, a MIMO controller can
be designed to improve the robustness and performance of the roll - yaw control system. The
transfer functions for roll and yaw angle outputs associated with roll and yaw axis disturbance
torques are now easily derived. Roll response is obtained from
Tx (s)(Iz s2 + Ω0 hw ) − Tz (s)hw s
Φ(s) = (3.13)
C.E.
When Tz = 0
Φ(s) Iz s2 + Ω0 hw
=
Tx (s) C.E.
When Tx = 0
Φ(s) hw s
=
Tz (s) C.E.
Yaw response is obtained from
Tz (s)(Ix s2 + Kτ s + K) + K(τ s + 1)Tx (s))
Ψ(s) =
C.E.
When Tx = 0.
Ψ(s) Ix s2 + Kτ s + K
=
Tz (s) C.E.
When Tz = 0.
Ψ(s) K(τ s + 1)
=
Tx (s) C.E.
Selection of roll / yaw control parameter values begins by examining the characteristic equa-
tion. which can be factored into the product of two quadratics,
C.E. ∼
= (Ix s2 + Kτ s + K)(Iz s2 + khw s + Ω0 hw ) = 0
provided that
Kτ Iz >> khw Ix
The roots of this equation yield natural frequencies and damping rates of each mode.
√ √
K τ K
ω1 = ; ζ1 =
Ix 2 Ix
√ √
Ω0 hw k hw
ω2 = ; ζ2 =
Iz 2 Ω0 Iz
The high frequency roots(ω1 , ζ1 ) are characteristic of roll dynamics while the low frequency
roots(ω2 , ζ2 ) dominate yaw motion during yaw error correction. A large value of hw will result
in fast yaw corrections with respect to orbit period.
Roll autopilot gain K is selected to limit steady- state roll error caused by a constant
roll torque and to provide fast response. from equation (3.13) the steady-state roll error
produced by a constant roll torque is
Tx
ϕss =
K
Based on the maximum roll torque resulting from solar pressure and the allowable roll error
of 0.05◦ . this yields a minimum required value of K = 0.07 N.m/rad with a corresponding
natural frequency of ω1 = 0.006rad/s. Large values of K will result in a small steady-state
error and faster response time. Thus, a natural frequency of ω1 = 0.025 rad/s is selected
with a corresponding roll autopilot gain of K = 1.25 N.m/rad/s. It should be noted that this
choice is preliminary and must satisfy stability condition. Final selection of K will depend on
a root-locus analysis of the system. values of τ roll time constant and k, yaw- to - roll gain
ratio are selected to provide critical damping of roll and yaw dynamics. Thus, for ζ1 = ζ2 = 1
and hw = 200N .m. s. τ = 80sandK = 0.054. To justify the selection of hw , the steady-state
yaw offset expression is obtained for constant torques.
Tz − kTx
Ψss =
Ω0 hw
Since the contribution of roll torque is small ( because k = 0.054), only yaw offset resulting
from a yaw disturbance torque need be considered. The resulting condition on hw is
Tz|
hw >>
Ω0 Ψss |max
Taking the maximum amplitude of Tz = 5 × 10−5 N.m, and the specified yaw error limit of
0.40◦ leas to a minimum value for nominal momentum
hw ≥ 99N.m.s
Even though, the satellite experiences a low level of disturbances, the long period of operation
Table 3.1: Summary of Roll/Yaw Control System Parameters

Parameter Value
hw nominal wheel momentum magnitude 200 N. m. s
K, roll autopilot gain 1.56 N. m/rad
τ, roll time constant 80s
k, yaw-to-roll gain ratio 0.054
ω1, roll dynamics natural frequency 0.0286 rad/s
ζ1, roll dynamics damping ratio 0.998
ω2, yaw dynamics natural frequency 0.00264 rad/s
ζ2, dynamics damping ratio 0.985
coupled with the requirement of accurate attitude pointing (better than 0.1◦ ) and very low
jitter rates (lower than 10−3 deg/sec), make the control system design more challenging. Owing
to low level of attitude rates, rate gyros are not used for attitude control (due to low levels of
signal, low stability of signals and larger noise level inside the satellite). Position or Rate
integrating gyro sensors are used for attitude control. In order to obtain high precision
pointing, robust attitude control system is required.
Home work: Derive characteristic equation for Mx = Kτ ϕ̇ + Kϕ; Mz = 0 and Roll and Yaw
responses and two natural frequencies and damping. Compare the results with the above.
3.2 Aircraft Dynamics and Control

The aircraft dynamics is nonlinear and coupled [7]. Linearized equations of motion about a
given operating conditions (trim conditions)/ equilibrium conditions) are used for the control
system analysis and design. Typically number of such operating point vary from tens of
thousand to a couple of million points for a combat aircraft. The control system design entails
the construction of automatic flight control system (AFCS) at each of these operating points
and gain/ controller scheduling these controller between subsequent operating conditions.
With the advent of modern control theory, number of operating points for the controller
design has shrunk to a large extent. The aircraft has one plane of symmetry, and hence pitch
dynamics is nearly decoupled from roll-yaw dynamics. These dynamics are also known as
longitudinal dynamics and lateral dynamics (Details are given in Chapter 1). Equations of
Longitudinal and Lateral dynamics are repeated in this chapter for the sake of convenience.
Longitudinal Dynamics :
u̇ = Xu u + Xα α − gθ + Xe δe (3.14)
Zu Zα Ze
α̇ = u+ α + q + δe (3.15)
V V V
q̇ = Mu u + Mα α + Mq q + Me δe (3.16)
Kinematics : θ̇ = q; ḣ = V γ̇ = V (α̇ − q). (3.17)
Lateral Dynamics :
Yβ Yp Yr g Ya Yr
β̇ = β + p + ( − 1)r + ϕ + δa + δr (3.18)
V V V V V V
ṗ = Lβ β + Lp p + Lr r + La δa + Lr δr (3.19)
ṙ = Nβ β + Np p + Nr r + Na δa + Nr δr (3.20)
Kinematics : ϕ̇ = p (3.21)
Kinematics : ψ̇ = r (3.22)
The different control modes for an aircraft are: stability augmentation system
(SAS), Attitude control system (ACS), flight path control system (PPS), Active control system
(ACT), Automatic take-off and Landing control systems. Some of these flight control systems
will be discussed for longitudinal and lateral dynamics in the sequel.
3.2. AIRCRAFT DYNAMICS AND CONTROL 185
3.2.1 Longitudinal Flight Control System
This section gives a number of examples pertaining to different class of aircraft to illustrate
the diverse control issues that vary from one class of aircraft to another.
Example 1: For the landing phase of F-16 aircraft, the stability derivatives at V = 139Kt are
Xu = −0.0507, Xα = −3.861, Xe = 0
Zu Zα Ze
= −0.00117, = −0.5164, = −0.0717
V V V
Mu = −0.000129, Mα = 1.4168, Mq = −0.4932, Me = −1.645
(Note : Me > Mα ). The state space representation of the longitudinal dynamics:

   
 −0.0507 −3.8610 0 −32.185   0 
   
   
 −0.00117 −0.5164 1.0000 0   −0.0717 

A=   
 B=  (3.23)
 −0.000129 1.4168 −0.4932   
 0   −1.6450 
   
0 0 1.0 0 0
and the corresponding characteristic equation:
∆(s) = s4 + a1 s3 + a2 s3 + a2 s2 + a3 s + a4 = 0;
Zα
where a1 = − − Mq − Xu = 1.0603,
V
Zα Zα Zu
a2 = Mq − Mα + Xu ( + Mq ) − Xα = −1.1154,
V V V
Zα Zu
a3 = −Xu ( Mq − Mα ) + Xα ( Mq − Mu ) = −0.0565,
V V
Zu Mα − Zα Mu
a4 = g = −0.0512
V
i.e., ∆(s) = (s2 + 0.981 s − 1.2344) (s2 + 0.0788 s + 0.04155) = 0
The roots of the characteristic equations are at -1.705, + 0.724, - 0.0394 ± 0.20j. Observe
that the short period mode is unstable (since Mα > 0) and the phugoid mode has a frequency
of 0.204 and damping of 0.19. Vehicle is statically unstable.
Transfer function for the state variables(Note: unstable system) are
u(s) (s + 81.3145)(s + 0.4145) (21.2513 s + 26.1282) −(21.2513 s + 6.6784)

= 0.2768 = 2 + 2 ;
δe (s) ∆(s) (s + 0.981 s − 1.2344) (s + 0.0788 s + 0.04155)
α(s) (s + 23.4365)(s2 + 0.0502s + 0.1057)
= − 0.0717
δe (s) ∆(s)
−(0.0957 s + 1.7839) (0.0240 s + 0.0838)

= + 2 ;
(s + 0.981 s − 1.2344) (s + 0.0788 s + 0.04155)
2
q(s) s (s + 0.5865)(s + 0.0423) −(1.5537 s + 0.7956) −(0.0913 s + 0.0268)

= − 1.645 = 2 + 2 ;
δe (s) ∆(s) (s + 0.981 s − 1.2344) (s + 0.0788 s + 0.04155)
θ(s) (s + 0.5865)(s + 0.0423) −(0.6445 s + 2.1860) (0.6445 s − 0.0405)
= − 1.645 = 2 + 2
δe (s) ∆(s) (s + 0.981 s − 1.2344) (s + 0.0788 s + 0.04155)
Root locus plots (region near origin of s - plane) for the above four cases are given in the Figs.
3.6 & 3.7. The pole - zero maps for the four transfer functions show relative contributions of
short period and phugoid modes. This aspect gets amplified in the partial fraction expansion
of transfer functions into short period and phugoid modes (right most terms). These plots
indicate strength and weakness of using only the constant gain output feedback (one loop
at a time) control. In majority of cases (specially for an unstable aircraft), dynamic output
feedback, and most importantly, MIMO form of dynamic compensators can produce better
stability and performance. Discussions on different SISO flight control structures in frequency
domain will be discussed later in this section. The eigen-structure (eigenvalue - eigenvector
pairs) for the system matrix of the longitudinal dynamics is given below:
Table 3.2: The eigen-structure of the longitudinal dynamics F-16-1

Eigenvalues: -1.7036 0.7310 -0.0438 + 0.2065i -0.0438 - 0.2065i
Eigenvectors: 0.9943 0.9995 1.0000 1.0000
0.0633 -0.0142 0.0005 + 0.0003i 0.0005 - 0.0003i
-0.0740 -0.0165 0.0013 + 0.0002i 0.0013 - 0.0002i
0.0435 -0.0226 -0.0003 - 0.0064i -0.0003 + 0.0064i
Example 2: Transport aircraft Transfer functions for a statically stable transport air-
craft (See Blakelock’s book [7], Chapter 1, pp. 41-46.) are given by
∆1 (s) = {(s2 + 0.00466s + 0.0053) (s2 + 0.806s + 1.311)} (3.24)

u(s) (s − 68.8)(s + 0.6)
= − 0.000506 ; (N ote : stable) (3.25)
δe (s) ∆1 (s)
α(s) (s + 77.8)(s2 + 0.0063s + 0.0057)
= − 0.01785 ; (3.26)
δe (s) ∆1 (s)
θ(s) (s + 0.3)(s + 0.016)
= − 1.31 (3.27)
δe (s) ∆1 (s)
Root Locus
Root Locus
0.1
4
0.05
2
Imaginary Axis
1
Imaginary Axis
0
0
−1
−0.05
−2
−3
−0.1
−4
−5
−2 −1.5 −1 −0.5 0 0.5 1 1.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2

Real Axis Real Axis
Figure 3.6: Root locus (near origin) with u feedback and α - feedback
Root Locus
Root Locus 2
0.15
1.5
0.1
1
Imaginary Axis
0.05
Imaginary Axis
0.5
0
−0.05 0
−0.1
−0.5
−0.15
−1
−0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6
Real Axis Real Axis
Figure 3.7: Root locus (near origin) with q = θ̇ and θ - feedback
The short period poles are located at (−0.4030 ± 1.0717i) and phugoid poles at
(−0.0023 ± 0.0728i). The transfer function in Eqn. (3.25) indicates the presence of right half
plane zero. Unit step response of this transfer function (3.25) will have both overshoot and
undershoot. The dc gain / steady-state sign of the transfer function (use final value theorem)
is positive. For the positive step elevator input δe , steady state value of change in forward
speed u will be positive. This is expected since, positive elevator will make the aircraft to
pitch down (can be inferred while analyzing the pitch attitude transfer function), which in
turn make the aircraft to loose height and gain speed.
The transfer function for pitch attitude in Eqn. (3.27) has two zeros at −0.016 and −0.3. The
θ(s)
steady state gain values for the transfer function δe (s)
is negative, which implies that positive
elevator input produces negative or pitch down motion. The pitch attitude rate q(s) = s θ(s)
has three zeros at [0.0; −0.016, −0.3]. The two zeros of q(s)
δe (s)
at [0.0; −0.016] effectively cancel
the phugoid poles at (−0.0023 ± 0.0728i). The residues corresponding to phugoid mode will be
small. Hence feedback the pitch rate sensor will not have significant influence on the phugoid
mode. The third zero of q(s)
δe (s)
at −0.3 will have an effect introducing lead by feedback action,
hence, the pitch rate feedback enhances short period mode damping. It is used in the stability
augmentation system (SAS). The effect of feedback of state variables x = [u, α, q, θ]T on the
short period and phugoid modes can be seen from the frequency response (Bode) plots. Bode
plots can be generated using Matlab commands as under.
Matlab command to generate Bode plots are (Refer Eqns: 3.24, 3.25, 3.27 respectively):
den = conv([1 0.00466 0.0053], [1 0.806 1.311]) =⇒ Characteristic polynomial;
num1 = -0.000506 *conv([1 -68.8], [1 0.6]) =⇒ Numerator polynomial for the state u;
num2 = -1.31*conv([1 0.016], [1 0.3]) =⇒ Numerator polynomial for the state θ;
subplot(1,2,1); bode(num1, den); grid
num3 = -1.31*conv([1 0.016], [1 0.3 0]) =⇒ Numerator polynomial for the state q = s θ;
num4 = -0.01785 * conv([1 77.79], [1 0.0063 0.0057]) =⇒ Numerator polynomial for α;
Use commands like “title”, “xlabel” and “ylabel” to add title and labels to each figure. Spec-
tral content of change in forward speed u and pitch attitude angle θ for harmonic elevator
input δe is shown in the Fig. (3.8). Similar Bode plots for the state variables q & θ are dis-
played in the Fig. (3.9). Note that
q(s) = s θ(s) = s − 1.31 (s+0.3)
∆1 (s)
(s+0.016)
= − 1.31 (s2 +0.3 s+0) (s+0.016)
(s2 +0.00466s+0.0053) (s2 +0.806s+1.311)
.
Figs. 3.8 & 3.9 indicate that the frequency response of state variables u, θ, α, q
for the harmonic excitations at varied frequencies. Responses for u, θ have dominant phugoid
modes ωnp . Similarly, on looking at the magnitude as well as phase contributions from feedback
of α, q, they contribute more significantly towards short period modes ωns . Figure 3.8 shows
u(s)
two bode plots for the transfer functions , θ(s) .
δe (s) δe (s)
Left plot shows that the the amplitude
u(s)
of δe (s)
response is very small at the short period mode at ωns , which implies small change
ωnp ¯ Bode Diagram

50 ωns 50 ωnp ¯ ω
θ / δe amplitude ratio (dB)

u/δe amplitude ratio (dB)
ns
0
¯ ¯
−50 0
−100
−150 −50
−200
−250 −100
450 270
Phase lag (deg)

phase lag (deg)
360
180
270
180
90
90
0 0
−2 0 2 4 −3 −2 −1 0 1 2
10 10 10 10 10 10 10 10 10 10
Frequency (rad/sec) Frequency (rad/sec)
Figure 3.8: Bode (frequency) response plots for sensor outputs u and θ
Bode Diagram Bode Diagram

20 20
0
α / δe amplitude ratio (dB)
q / δe amplitude ratio (dB)
0
−20
−20 −40
−40 −60
−80
−60
−100
−80 −120
360 180
315
135
Phase (deg)
Phase (deg)
270
225 90
180
45
135
90 0
−2 0 2 −2 0 2 4
10 10 10 10 10 10 10
Frequency (rad/sec) Frequency (rad/sec)
Figure 3.9: Bode (frequency) response plots for sensor outputs q and α
in forward speed during short period oscillation. Similar observation holds good for the right
side plot of Fig. 3.8. Fig. 3.9 shows that considerable variations in pitch rate q and α at
short period and phugoid frequencies. The above observations on the transfer functions and
frequency responses of four state variables enable the following approximations/ simplifications
for the longitudinal dynamics.
Short period approximation : (Repeated here for convenience from the Chapter 1)
The short period approximation of pitch plane dynamics is primarily the motion
involving pitch rate q and angle of attack / incidence α with the change in forward speed
∆u = 0 of forward speed remaining constant. For a steady level flight θ is assumed to be
constant. Here, it is assumed that the dynamic stability derivatives Zα̇ = 0 and Zq = 0.
      

 α̇ 
 Zα 
 
 
 Ze 

1  α
=

V
 + V
δ

 q̇ 
     e
Mα Mq  q   Me 
The characteristic equation, then reduces to
s2 − [Mq + (Zα )/V )]s + [Mq Zα /V − Mα ] = s2 + 2ζs ωs s + ωs2 = 0
Transfer functions:
α(s) Ze [s + V (Me /Ze )] q(s) Me (s − Zα /V )

= ; = ; where, ∆sp (s) = s2 + 2ζs ωs s + ωs2
δe (s) ∆sp (s) δe (s) ∆sp (s)
In pitch autopilot design, it is possible to use short period approximation during the straight
and level flight (say, cruise of a transport aircraft) and get a simple controller that gives
satisfactory performance when analyzed with the full order model.
Now, consider Example 2 (Transport aircraft) given above in Eqn. 3.24 (Refer:
Blakelock [7], pp. 46 - 50 and Stevens and Lewis [35], pp. 263 - 265). The state variables are
(α, q)T and the input is elevator deflection δe . The transfer function ([7], Eqn. 1-140)
θ(s) 1 q(s) Me (s − Zα /V ) −1.39 (s + 0.306)

= = = (3.28)
δe (s) s δe (s) s(s − [Mq + (Zα )/V )]s + [Mq Zα /V − Mα ]
2 s(s2 + 0.805 s + 1.325)
θ(s) q(s)
The transfer functions δe (s)
and δe (s)
will be used for flight control system analysis in the next
subsection.
Longitudinal Flight Control System (FCS)
Modern aircraft, both military and civil, require flight control system to enlarge flight envelope
and to provide better stability and performance characteristics. For military aircraft, specially
for the fighter aircraft, widening of the flight envelope (Fig. (3.10)) is crucial. The aircraft
envelope encompass a wide range of dynamic pressure regime. FCS is designed at different
operating point within this envelope, but the actual flight conditions can vary significantly
from the design conditions resulting in large variations in stability derivatives/ coefficients
of dynamic equations. This results in large uncertainty in system parameters. Therefore,
robust control is of paramount importance for modern aircraft. In this section, different FCs
architectures are discussed and some of them are analyzed.
Figure 3.10: Aircraft Altitude - Mach Envelope
Different types of flight control systems (FCS) can be categorized as - stability

augmentation system (SAS); Control augmentation system (CAS); and Autopilots, etc (Table
3.3 [35, p.261]). The eigenvalue analysis of longitudinal dynamics carried out earlier has shown
that damping of short period and phugoid modes are poor. There is a need to enhance short
period damping and also improve stability margins. Short period frequency is also inadequate.
Such control system is known as stability augmentation system (SAS) Phugoid frequency
is normally very small (with a time period of 100 seconds or so. Therefore, pilot can maneuver
the stick while performing combat operations. If the augmentation system is required to
provide pilot with a specified type of response based on handling/ flying quality requirements,
the such a FCS is known as control augmentation system (CAS). An example of this is the
normal acceleration CAS, wherein normal acceleration along the negative z − axis is limited/
restricted. Slow modes such as phugoid and spiral modes are normally controlled by pilots.
However, during critical operations, like say, weapon release, landing etc., controlling the slow
modes as well as flight, pilot relief can be provided by autopilots. Similarly, autopilots
are used during long cruising phase of a transport aircraft.
SAS CAS Autopilots

Pitch damper Pitch rate Pitch attitude hold
Roll damper Roll rate Altitude hold
Yaw damper Normal acceleration Speed/ Mach hold
Lateral/ directional Automatic landing
Roll angle hold
Turn Coordination
Heading hold/ VOR hold
Table 3.3: FCS catagories [35, p.261]
Example 2 (transport aircraft) is considered first for demonstrating the need for
stability or damping augmentation. Block diagram for a simple displacement autopilot is given
in the Fig. 3.11 [7, pp.63-68]. Middle and bottom block diagrams represent the conventional
and jet transport aircraft. Autopilot is designed to hold the transport aircraft in straight and
level flight. Aircraft use the conventional vertical gyro, which outputs the error signal eg .
This is a type 0 system since there is no integrator in the forward loop. Disadvantage of this
autopilot (with only θ - feedback) is characterized through the root locus analysis.
The transfer function for short period model of conventional aircraft flying at 240km/hour
at sea level and the corresponding actuator are given by
θ(s) − (s + 3.1) δe (s) −Kc

= 2
; = (3.29)
δe (s) s(s + 2.8 s + 3.24) eg (s) (s + 12.5)
Since the plant transfer has negative sign, the negative sign is attached to the amplifier, so
that the forward loop gain is positive. Hence, positive reference attitude θ command produces
positive attitude change/ pitch up of aircraft. Then, the open loop transfer function is
Kc (s + 3.1)
OLTF: G H(s) =
s (s + 12.5) (s2 + 2.8 s + 3.24)
It may be recalled that the root locus is locus of closed loop poles in s − plane as the gain
(in this case Kc ) or equivalent gain like parameter) varies from zero to ∞. Number of poles
n = 4, which are located at 0, −12.5, & − 1.4 ± 1.1314 j. Number of locii = n = 4. Number
Figure 3.11: Aircraft Attitude control system block diagram; :

Top: Generic; Middle: Conventional transport; Bottom: Jet transport.
of zeros m = 1, which is located at −3.1. The pole - zero excess is n − m = 3, which is equal
to number of zeros at the infinite horizon of s - plane. Number of asymptotes = 3. The three
zeros are located along the three asymptotes in the infinite horizon. Root locii start at finite
poles and end at the zeros as the gain Kc is varied from zero to ∞. The real axis segment of
root locus lies to the left of an odd number of poles and zeros. Therefore, root locii on the
real axis lie between [0, −3.1] and [−12.5, ∞). Since, there are no two consecutive poles or
zeros on the real axis, there may not be real axis break-away or break-in points.
All the asymptotes intercept the real axis at a point σ0 with an angle θk , k =
1, . . . , (n − m); where,
∑ ∑
n f inite poles− m f inite zeros (2k−1) 180◦
σ0 = j=0
(n−m)
i=0
and θk = (n−m)
; k = 1, 2, . . . , (n − m).
For the current example σ0 = 0−12.5−1.4−1.4+3.1
3
= −4.067. Angle of asymptotes θk =
(2k−1) 180◦
3
; (k = 1, 2, 3) = ±60◦ , 180◦ . Four root locii start at the four finite poles. Locus
Root Locus
6
ζ = 0.6
4
Kc = 77.5
2
Kc = 8.8
jω
−2
−4
−6
−14 −12 −10 −8 −6 −4 −2 0
σ
Figure 3.12: Root locus for Conventional transport aircraft & Autopilot.
Root Locus
Kc = 38.4
2
Imaginary Axis
jω
−2
−4
−6
−12 −10 −8 −6 −4 −2 0
σ
Real Axis
Figure 3.13: Root locus for Conventional transport aircraft & Autopilot.
starting at origin terminates at the finite zero at −3.1. Locus starting at −12.5 moves along
the asymptote making 180◦ , i.e., negative real axis of s-plane and terminates at infinite horizon
zero. Two locii starting at complex conjugate poles at −1.4 ± 1.1314 j eventually move along
the asymptote as gain Kc tends towards ∞. See the root locus diagram in the Fig. (3.12).
The asymptotes inclined at ±60◦ intercept the imaginary axis at ± 7.03 j. The two
root locii will cross the imaginary axis within the cone bounded by the above asymptotes. The
point of imaginary axis crossing can be obtained using the Routh - Hurwitz criterion, since
imaginary crossing point corresponds neutrally stable system. The characteristic equation for
the closed loop system is
[s (s + 12.5) (s2 + 2.8 s + 3.24)] + Kc (s + 3.1) = 0
s4 + 15.3 s3 + 38.24 s2 + (Kc + 40.5) s + 3.1 Kc = 0
Using the Routh Table, critical value of gain Kcritical = 77.5. The root locus crosses the
imaginary axis at ± 2.78 j. The other two real poles of the closed loop system are located
at −2.36, −12.94. The angle of departure of the root locii at the complex conjugate poles,
say, at −1.4 + 1.1314 j is given by ϕ1 = θ1 − ϕ2 − ϕ3 − ϕ4 − (2k + 1)180◦ , where, θ1 =
angle subtended by the zero at −3.1, ϕ2 , ϕ3 , & ϕ4 are angles subtended by the poles at
0, −1.4 + 1.1314 j, & − 12.5 respectively at the pole in question −1.4 + 1.1314 j (See Fig.
2.40). Hence, ϕ1 = tan−1 ( 3.1−1.4
1.1314
) − (180 − tan−1 ( 1.1314
1.4
)) − 90 − tan−1 ( 12.5−1.4
1.1314
) − (2k + 1)180◦ =
33.64 − 180 + 38.94 − 90 − 5.82 − (2k + 1)180◦ |k=0 = −383.24 = 336.76◦ = −23.24◦ (note:
discount integer multiples of ±360◦ ). Matlab generated root locus is plotted in Fig. (3.12).
It may be noted that the damping of short period modes (open loop pole at −1.4 +
1.1314 j, having the damping ratio of 0.78) decreases as the amplifier gain Kc increases. When
Kc is increased to 8.8 deg/volt, the damping ratio decreases to a satisfactory value of 0.6.
On the other hand, the transfer function for short period model of jet aircraft
flying at 600ft/sec (658km/hour) at 40,000ft (≈ 12km) and the corresponding actuator are
given by
θ(s) − 1.39 (s + 0.306) δe (s) −Kc

= ; = (3.30)
δe (s) s(s2 + 0.805 s + 1.325) eg (s) (s + 10)
The corresponding open loop transfer function is
Kc 1.39 (s + 0.306)
OLTF: G H(s) =
s (s + 10) (s2 + 0.805 s + 1.325)
The root locus diagram can be plotted as detailed above. Matlab generated root locus is
plotted in the Fig. (3.13), which is grossly similar to the Fig. (3.12). Note that damping
is already low for the plant/ open loop transfer function (poles are at −0.4025 ± 1.0784 j;
i.e., damping ratio ζ = 0.35 and frequency ω = 1.15rad/sec). The root locus plot shows
that for jet transport aircraft, damping decreases rapidly with the increase in gain Kc . The
critical gain on the verge of instability in this case is only Kcritical = 38.4deg/volt as against
77.5deg/volt for the conventional transport. Therefore, displacement autopilot alone is not
good enough for jet transport. Additional feedback control of pitch rate in the inner loop is
needed for damping enhancement (hence, stability augmentation).
Stability Augmentation System
Majority of civil and military aircraft are statically stable, but they exhibit poor damping
characteristics during large part of the flying within the flight envelope. However, the modern
combat aircraft have been designed deliberately to statically unstable for better maneuver-
ability. Now a days, all the stable or unstable aircraft invariably use pitch rate feedback for
damping enhancement. Unstable aircraft use α - feedback in the inner loop for enhancing
the loop stiffness and hence performance. For attitude hold mode, θ - feedback is used in the
outer loop instead of α - feedback in the inner loop. Next, FCS with pitch rate feedback is
analyzed using the root locus analysis for the jet transport aircraft of Example 2.
Pitch rate feedback FCS:
Consider the plant transfer for pitch including the actuator transfer function and
rate feedback gain Krg once again for Example 2 (jet aircraft):
q(s) − 1.39 (s + 0.306) δe (s) −10

= 2 ; = ; Krg = rate gyro feedback gain(3.31)
δe (s) (s + 0.805 s + 1.325) eg (s) (s + 10)
The corresponding open loop transfer function is
Krg 13.9 (s + 0.306)

OLTF: G H(s) =
(s + 10) (s2 + 0.805 s + 1.325)
Number of poles = Number of locii = n = 3 and Number of zeros = Number of finite horizon
zeros m = 1. The number of asymptotes = n − m = 2. Hence, angle of asymptotes = ±90◦ .
−10−0.403−0.403+0.306
Real axis intercept of the asymptote is at σ0 = 2
= −5.25. Root locus can
be plotted using the MATALB command:
≫ num = 13.9 * [1 0.306]; den = conv([1 10], [1 0.805 1.325]); rlocus(num,den)
Root locus diagram is shown in the Fig. (3.14). On the real axis of s − plane,
root locus exists between [-10, -0.306]. It is not easy to guess that there exists break-in and
break-away points at the real axis. To find out such possibility, one has to use root locus rules
for determining these points (§2.5.3, Property 6), which requires solution of a third order
Inner loop pitch rate feedback for jet aircraft

3
System: sys
Gain: 1.21
1 Pole: −1.54
Damping: 1
Imaginary Axis
Overshoot (%): 0
0
−1
−2
−3
−12 −10 −8 −6 −4 −2 0 2
Real Axis
Figure 3.14: Root locus for jet transport aircraft & pitch rate stabilization system.
polynomial equation. If the solution for the polynomial equation falls within the admissible
range [-10, -0.306], then there exists break-in and/or break-away point(s). If at all, break-away
point exists, it must be close to the intersection of asymptote at −5.25. Find it iteratively;
say the point is s = −b. The remaining two solutions of quadratic part of the equation
(s + b) (s2 + a1 s + a2 ) = 0 make up three roots. These are break-in and/ or break-away points
of the root locus (roots in the admissible region) and complementary root locus. As is already
stated in the Property 6, for any points λ of the root locus, the following equation
1 (s + 10) (s2 + 0.805 s + 1.325)

Krg =− =
G(s = λ) H(s = λ) 13.9 (s + 0.306)
is valid. On the real axis s = λ is real, therefore, H(s) and G(s) are real-valued function.
To find points of maximum and minimum of K = Krg , differentiate K with respect to s and
setting the derivative equal to zero, to get
dK (s+10) (s2 +0.805 s+1.325)∗d/ds[13.9 (s+0.306)]−d/ds[(s+10) (s2 +0.805 s+1.325)]∗[13.9 (s+0.306)]
ds = [13.9 (s+0.306)]2 =0
Solving for s for the polynomial equation (27.80 s3 + 162.9497 s2 + 91.916 s − 144.2994) = 0,
roots are −4.9906, −1.5444 + 0.6735. Therefore, s = −1.5444 is the break-in point and
s = −4.9906 is the break-away point (closer to the asymptote). The third point s = +0.6735
corresponds to the complementary root locus. At the break-in point, the gain Krg = 1.21,
and repeated eigenvalues/ poles are λ = −1.5444. The third pole origination from the pole
at s = −10.0 moves to s = −7.8 at Krg = 1.21. Normally, the outer loop, say attitude
hold mode autopilot mode, adds lag into the loop transfer function, hence the damping of
short period mode with the outer loop will be lower than that of inner loop alone. Hence, a
larger damping ratio of short period is set. However, increasing the rate gain further, makes
one closed loop pole move towards the zero closer to the imaginary axis, hence it is counter
productive. There is no strict guideline for the selection of rate gyro gain Krg . Therefore,
gain selection is subjective and depends on the designer’s expertise.
It may be noted that the closed loop system with pitch rate feedback (also note
that plant is it self stable) is unconditionally stable, i.e., stable for all values of Krg ∈ [0, ∞).
The system has infinite gain margin (both upward and downward). This conclusion is based
on the reduced order short period model of the aircraft. However, the addition of stable
phugoid mode and un-modeled dynamics will not change the stability properties at low gain
operation. Before implementation of simplistic controller for SAS, detailed analysis with full
fledged model and inclusion of nonlinearity like say, saturation of actuators is needed. These
aspects are beyond the purview of this short course.
Homework: Obtain the Bode plot and Nyquist plot for the above OLTF with Krg = 1.21 by
manual step by step procedure outlined in Chapter 2, §2.5.2. Check the result using Matlab.
Example 3: F-16 Fighter aircraft [35, pp. 292 - 299]
The above example on stability augmentation system (SAS) for transport aircraft
primarily focussed on damping augmentation. However, for combat aircraft, enhancement of
both natural frequency and damping of dominant short period mode is critical. With the
increased tendency to build aircraft with larger instability, stabilizing the aircraft dynamics
plays an important role. Therefore, majority of fighter aircraft of recent origin (including
Light combat aircraft, Tejas, though not the earlier version of Mirage 2000), have stability
augmentation system having feedback of angle of attack α and pitch rate q. In real design
exercise, feedback gains Kα & Krg (or Kq = Krg ) are determined by modern control technique.
Since this chapter deals with the SISO system, analysis of one loop at a time is undertaken.
Figure 3.15: Block diagram of Pitch stabilization system.
The system matrices for F-16 aircraft model in the nominal flight condition are:
   
 u α θ q   δe 
   
   


−0.019311 8.8157 −32.17 0.57499 



0.1737 

   
A = 
 −0.00025 −1.0189 0 0.905  ; B =  
 −0.0021499  ; (3.32)
   
   
 0 0 0 1   0 
   
   
2.994 ∗ 10−12 0.82225 0 −1.0774 −0.17555
 
 0 57.29578 0 0  α
C =   (3.33)
0 0 0 57.29578 q
The output variables are in written in the units of degree (◦ ), hence the factor 57.29578◦ /rad.
The input is elevator angle δe is in degrees.
α(s)
The following transfer function δe (s)
is used in the study of inner loop (Fig. 3.15)
α(s) −0.232 (s + 75) (s + 0.00982 + 0.09379 j) (s + 0.00982 − 0.09379 j)

=
δe (s) (s − 0.09755) (s + 1.912) (s + 0.1507 + 0.1153 j) (s + 0.1507 − 0.1153 j)
This transfer function differs from conventional stable transfer function for longitudinal dy-
namics of having stable short period and phugoid modes. The pole at s = + 0.09755 indicates
an unstable exponential mode with a time constant of ≈ 10 seconds. The complex pole pair
has a damping ratio of 0.79 and oscillatory period of 33 seconds, which is like a phugoid with
large damping ratio like that of short period. This is a marginally unstable aircraft with
a small static instability margin (CP is slightly ahead of CG). The elevator transfer func-
δe (s) 20.2
tion is ue (s)
= (s+20.2)
. Most of the angle of attack sensors are noisy, hence a low pass filter
αf (s) 10.0
α(s)
= (s+10.0)
is used. By augmenting the actuator and sensor filter states ue , αf with the
airframe dynamics, the state space model with state variables x = [u, α, θ, q, δe , αf ]T and
actuator input ue is
    

−0.019311 8.8157 −32.17 0.57499 0.1737 0  
u  
0
     
 −0.00025 −1.0189 0 0.905 −0.00215 0   α   0 
     
     
     
 0 0 0 1 0 0   θ   0 
     
ẋ =     +   ue
     
 2.99 ∗ 10−12 0.82225 0 −1.0774 −0.17555 0   q   0 
     
     
 0 0 0 0 −20.2 0   δ   20.2 
   e   
     
0 10.0 0 0 0 −10.0 αf 0
which can be written as ẋ = Aa x + ba ue . The goal of the α - feedback is to shift the
unstable pole at s = 0.09755 to the left-half of s-plane. Here, since the filtered angle of
attach sensor output is in degrees, then output αf is multiplied by a factor 57.29578, i.e.,
Ca = [0, 0, 0, 0, 0, 57.29578], da = 0. In modern aircraft, there will be a Kalman filter which
has much better noise rejection capability and less lag at the dominant modes, thus the lag
sensor filter need not be used in the analysis. In such a case, ideal sensor can be assumed, i.e.,
sensor delay can be neglected. This is a large order system, hence, it is not possible to draw
root locus plot within a short time. The following Matlab program can be used [35, p.295]:
≫ k = logspace(-2, 1, 20000);
≫ r = rlocus(Aa, -ba, ca, 0, k);
≫ plot(r)
≫ grid on
≫ axis([-20, 1, -10, 10])
Fig. (3.16) shows the complementary root locus plot (note the negative sign in the
plant transfer function) for the angle of attack/ inner loop on two scales (left and right plots).
The expanded plot of root locus near origin shows that the α feedback makes locii from the
third modes poles converge on to the real axis at −0.19515 at gain = Kα = 0.0187. The
branch going to right, as Kα increases, terminates on the complex conjugate zeros near the
origin. These poles for finite Kα corresponds to phugoid poles. The branch going right breaks
away at −0.9907 at gain Kα = 0.0697. These pole pairs after the branch-off correspond to the
Inner−loop root locus plot for pitch SAS − Example 3 Expanded inner−loop root locus plot
10 5
8 4
6 3
Imaginary axis (j ω) −−−>
Imaginary axis (j ω) −−−>

4 2
2 1
0 0
−2 −1
−4 −2
−6 −3
−8 −4
−10
−20 −15 −10 −5 0 −9 −8 −7 −6 −5 −4 −3 −2 −1 0 1
Real Axis (σ) −−−> Real Axis (σ) −−−>
Figure 3.16: Inner - loop of Pitch stabilization system (α - feedback).
Inner−loop root locus plot (full) for pitch SAS − Example 3 Expanded/ closeup inner−loop root locus plot
30 0.5
0.4
20
0.3
10
Imaginary Axis
Imaginary Axis
0.2
0
0.1
−10 0
−0.1
−20
−0.2
−30
−100 −90 −80 −70 −60 −50 −40 −30 −20 −10 0 10 −2.5 −2 −1.5 −1 −0.5 0 0.5
Real Axis Real Axis
Figure 3.17: Root locus for Inner - loop of Pitch stabilization system.
short period modes. Note that the no of poles of the augmented system is six (4 - plant + 1 -
actuator + 1 - sensor) and no. of zeros are three. The no of asymptotes are three and they are
along ±60◦ , 180◦ . Fig. 3.17 gives larger plot of the root locus (left) and further close up view
(right). Matlab commands are (≫ [num, den] = ss2tf(Aa, -ba, ca, 0); num(2)=0; num(3)=0;
rlocus(num, den); axis([-100 10 -30 30]) (for left plot); axis([-2.5 0.5 -0.25 0.5]) (for right plot).
The root locus breaks away at −16.7785 with gain of Kα = 3.94 will break in at −97.2 with
gain Kα = 1.12e5 after the far left zero at −75 with one branch going towards finite zero
and another towards infinite horizon zero. However, this part of locus is only of academic
interest since the gain will be very large to be of practical use. The root locus breaking
away at −0.195 crosses the imaginary axis at 3.87 j at a gain of Kα = 1.98. Therefore, the
α feedback has stabilizing effect. For the Kα = 0.5, the short period modes are located at
−0.7 ± 2.0 j; ⇒ ζ ≈ 0.33, ωn ≈ 2.2rad/sec. At this gain value, short period frequency is
acceptable, but damping is poor. Hence, pitch rate outer feedback loop is recommended.
The closed loop transfer for the inner loop with Kα = 0.5 and Kq = 0 [35, p.297] is
q(s) 203.2 s (s + 10) (s + 1.027) (s + 0.02174)

= (3.34)
ue (s) (s + 20.01) (s + 10.89) (s + 0.600 + ±2.03 j) (s + 0.0085 ± 0.08269 j)
No of poles = 6 and no of zero = 3. Once again, the no of asymptotes are three and they
are oriented along ±60◦ , 180◦ . The outer loop feedback gain is Kq . The output vector
Cq = [0, 0, 0, 57.29578, 0, 0, ] Root locus plot by varying Kq ∈ [0. ∞) is also drawn using
the Matlab program [35, p.297] given below:
≫ Aclq = Aa + ba * 0.5 * ca; % Kα = 0.5
≫ [num, den] = ss2tf(Aclq, ba, cq,0); num(2)=0 % num(2) ̸= 0 or ϵ due to round off errors
≫ rlocus(-num, den);
≫ axis([-22, 1, -10, 10])
Note that the factorizing the numerator (num) and the denominator (den) gives the same
expression as given in Eqn. (3.34)
Ouer−loop root locus plot for pitch SAS Ouer−loop root locus plot for pitch SAS − expanded view near origin
20 0.2
15 0.15
10 System: sys System: sys 0.1 System: sys

Gain: 0.322 Gain: 0.447 Gain: 27.8
Pole: −13.7 Pole: −3.33 + 1.75e−007i Pole: −0.0108 + 1.61e−010i
5 Damping: 1 Damping: 1 0.05 Damping: 1
Imaginary Axis
Imaginary Axis
Overshoot (%): 0 Overshoot (%): 0 Overshoot (%): 0

Frequency (rad/sec): 13.7 Frequency (rad/sec): 3.33 Frequency (rad/sec): 0.0108
0 0
−5 −0.05
−10 −0.1
−15 −0.15
−20 −0.2
−20 −15 −10 −5 0 −0.05 −0.04 −0.03 −0.02 −0.01 0 0.01
Real Axis Real Axis
Figure 3.18: Root locus for Outer - loop of Pitch stabilization system.
Fig. (3.18) gives regular and eaxpanded view of rootlocus for the outer loop. The
expanded view of root locus shows that phugoid poles move slowly with increase in gain Kq .
Only at the large (/impractical) gain of Kq = 27.8, phugoid poles reach the real axis. This is
expected since influence of q - feedback on phugoid is small. It may be noted that for many
stabilization systems, decent value of pitch rate feedback gain lies in the range of [0.2, 1.0].
At Kq = 0.25, short period poles are at −2.02 ± 1.94 j. This corresponds to a damping ratio
of ζ = 0.72 and short period natural frequency of ωn = 2.8 rad/sec. The closed loop transfer
function can be obtained as
q(s) 203.2 s (s + 10) (s + 1.027) (s + 0.02174)

=
r(s) (s + 16.39) (s + 11.88) (s + 2.018 + ±1.945 j) (s + 0.00878 ± 0.06681 j)
The open loop actuator pole has moved over from s = −20.2 to s = −16.39 and filter poles
have moved from s = −10.0 to s = −11.88. The closed loop transfer is similar to the stable
transfer function in Eqn. (3.34), except improvement in damping of the short period mode.
Control Augmentation System (CAS)
Under normal circumstances, stability augmentation system will suffice. This is more so for
manual flying of aircraft. Whenever, the combat aircraft is required to perform extreme ma-
neuvers or pilot need to operate risky operations, like target tracking and weapon release
and so on; Control Augmentation System (CAS) becomes handy. Kind of FCS used in the
feedback loop or the forward gains have constant magnitude. Augmentation of control func-
tions can be done by using dynamic compensators (lag, lead, lag-lead, or PI, PD or PID
compensators). Pitch-rate augmentation system [35, pp.309 - 313] is demonstrated next.
Stability augmentation system discussed in Example 2 and Example 3 above are
Type 0 systems. They may not be satisfactory since large loop (amplifier/ control) gains may
lead to large control activity and large control errors. In order to improve steady state error, PI
controllers are recommended. Analysis of pitch rate CAS is explained through an illustrative
example pertaining to F-16 longitudinal dynamics (Eqn. 3.33 so that its functionality can be
compared with that of SAS of Example 3 above. For the sake of simplicity in the analysis,
short period approximation of longitudinal dynamics is used, while full model can be used for
simulation purpose. The state variables for the short period models are x1 = α, x2 = q. As
a first step inner loop with the α feedback is closed, followed by compensator for q feedback.
State, input and output matrices (Ap, Bp, Cp) are

   
α q δe  
   
    57.29578 0 α
     
Ap =  −1.0189 0.905  ; Bp =  −0.0021499  ; Cp =  
   
    0 57.29578 q
0.82225 −1.0774 −0.17555
On augmenting the actuator and sensor dynamics at the input and α sensor output:
   
 α q δe αf   ue 
    α
     
 −1.019 0.905 −0.0022 0   0 
   
    57.3 0 0 0  q
Aa =   ; Ba =  0  ; Ca = 
 0.8223 −1.077 −0.1756 0     
    0 0 0 57.3 δe
   
 0 0 20.2 0   20.2 
   
    αf
10.0 0 0 −10.0 0
On closing the α - feedback loop as earlier, the closed loop matrix for the inner loop is
Acl1 = Aa − Ba ∗ [Kα 0] ∗ Ca. Select Kα = 0.2, then, the outer loop plant transfer function is
q(s) 203.2 (s + 10) (s + 1.029)
=
ue (s) (s + 10.38) (s + 20.13) (s + 0.8957 + ±1.152 j)
(s+3)
Let the PI compensator is given by H(s) = Kp s
, then the OLTF is
q(s) 203.2 Kp (s + 3) (s + 10) (s + 1.029)
= ; e = r − q;
e(s) s (s + 10.38) (s + 20.13) (s + 0.8957 + ±1.152 j)
Kp (203.2 s3 + 2850.7 s2 + 8814.2 s + 6272.8)
= 5
s + 32.3014 s4 + 265.7344 s3 + 439.2794 s2 + 444.9332 s + 0
The root locus diagram in Fig. (3.19) shows that dominant short period modes at gain value
of Kq = 0.5 are located at s = (−3.24 ± 3.41 j). At the gain Kq = 0.497, combined actuator -
sensor poles break away at −12.5. As the gain increases, at the gain value of Kq = 0.823, the
short period modes merge at s = −6.83. At this large gain, the response to the step input is
sluggish, hence, large gain values are not recommended. With Kq = 0.5, the slow integrator
pole reaches s = −1.02. At Kq = 0.5, actuator - filter pole pairs are on verge of becoming
complex conjugate. For Kq 0.5, actuator - filter pole pairs become complex conjugate, but
such oscillatory response for this mode is not recommended. The unit step response (not
given, take it as a Homework) shows overshoot of about 20%, which is little high. If the zeros
(s + 3) removed and α - gain, i.e., Kα is reduced to 0.08, then the step response is satisfactory
(though it takes slightly larger rise time, the settling time is nearly same as earlier).
Homework: Determine the closed loop transfer function for H(s) = ( Ksp ), Kα = 0.08, Kq = 0.5.
Root locus for pitch rate CAS

8
4 System: sys System: sys

Gain: 0.497 Gain: 0.823
Pole: −12.5 − 4.73e−007i
Pole: −6.83
2 Damping: 1 Damping: 1
Imaginary Axis
Overshoot (%): 0 Overshoot (%): 0

System: sys
0 Gain: 0.501
Pole: −3.24 − 3.41i
Damping: 0.689
−2 Overshoot (%): 5.04
−4
−6
−8
−20 −18 −16 −14 −12 −10 −8 −6 −4 −2 0 2
Real Axis
Figure 3.19: Root locus for Outer - loop of Pitch rate Control Augmentation system for F-16.
A few Longitudinal Autopilots given in the Table (3.3) are discussed below (due to
limitation of time and space).
Pitch Autopilots
Pitch Attitude hold autopilot:

Block diagram of conventional attitude hold autopilot is given in Fig. (3.20). The discussion
on the inner loop containing pitch rate feedback for Example 3 of jet transport aircraft has
already been discussed under stability augmentation system. It may be recalled that the rate
gyro gain is selected such that resulting inner closed loop is an over damped system, so that
the lag due to integrated pitch rate θ(s) = q(s)/s), does not make the closed loop system too
oscillatory. A gain of Kq = 1.2, results in shifting poles to (−1.5, −1.5, −7.8). The feedback
action does not alter the location of plant zero. In the conventional system, the pitch angle
error output is directly amplified. For analysis of outer loop, OLTF (Fig. 3.21) is given by
θ(s) 13.9 Kθ (s + 0.306)

=
eg (s) s (s + 7.8) (s + 1.5)2
The pole - zero excess is (4 - 1) = 3 = number of asymptotes with the angle of asymptotes =
−1.5−1.5−7.8+0.306
±60◦ , 180◦ , and which intercept the real axis at σ0 = 3
= −3.5. As expected,
the damping ratio of the closed loop system decreases with the increase in cascade amplifier
Figure 3.20: Root locus for Outer - loop of Pitch stabilization system.
Figure 3.21: Outer - loop of Pitch attitude hold Autopilot system.
gain Kθ . At Kcritical = Kθ = 15.2, the root locus crosses over to the right side of s − plane.
For Kθ = 1.41, damping ratio is 0.6 (See Fig. 3.22). If the nominal system gain increased by
a factor of 10.78, then the closed system is driven to the verge of instability. Hence, the gain
margin is 20log10 10.78 = 20.65 dB. It may be noted that this is a Type 0 system, since there
is no integral action in the controller. As is done for the control augmented system (CAS), if
Figure 3.22: Root locus for Outer - loop of Pitch attitude autopilot, Kq = 1.19.
integral part of the controller is added, then, Type 1 system with good performance can be
achieved. Analysis of PI compensator has already been discussed and hence, is not repeated.
For the sake of better understanding, let the pitch rate feedback gain be increased
from 1.2 to 1.98. The repeated real root splits, and faster root merges with the actuator pole
to produce a damped oscillatory mode. Then, the OLTF for the outer loop is given by
θ(s) 13.9 Kθ (s + 0.306)

=
eg (s) s (s + 0.8) (s2 + 10 s + 29)
Matlab drawn root locus diagram is given in Fig. (3.23)(similar figure in [7, p.67]). The
corresponding gain margin ≈ 22 dB. Higher the gain of the inner loop implies higher the
stability margin. This is achieved at the cost of higher control activity.
Longitudinal Acceleration Control System:
During the high maneuver of the combat aircraft, it is necessary to control acceleration along
the Z-axis (longitudinal acceleration). Thre is special requirement to limit the negative ac-
celeration. The block diagram of the acceleration control system is given in Fig. (3.24).
The feedback of angle of attack or acceleration helps in stiffening the loop trans-
fer matrix, i.e., they enhance loop gain and hence control bandwidth. Therefore, many
fighter aircraft use acceleration feedback to reduce maneuver load on an aircraft. The
Figure 3.23: Root locus for Outer - loop of Pitch attitude autopilot, Kq = 1.98.
Figure 3.24: Block diagram for Pitch acceleration control system.
acceleration along Z-direction is known as longitudinal acceleration in aircraft parlance, since

it is related to Longitudinal dynamics. (In missile/ rocket terminology, acceleration perpen-
dicular to the body axis along Z - direction or Y - direction is known as lateral acceleration,
latax for short). Acceleration az or fz is not a state variable, hence it has to be written in
terms of state variables x = [u, α, q, θ]T . For the nominal forward speed V , az = V (α̇ − θ̇),
az (s) α̇ θ̇
=V ( − )
δe (s) δe δe
The question that arises [7, p.77] is whether full state equation or short period approximation
has to be used. It was shown by Blakelock [7, §1-10] that influence phugoid oscillation on α̇
and θ̇ are almost zero. Hence short period approximation for the plant model can be used in
the preliminary analysis (apart from ease of understanding). For a typical combat aircraft,
α̇ − s (0.1 s + 15) θ̇ −(15 s + 6)
= ; = 2 ⇒
δe (s2 + 0.9 s + 8)) δe (s + 0.9 s + 8)
az − 77.7 (s2 − 60)
=
δe (s2 + 0.9 s + 8))
On factorizing the numerator,
az − 77.7 (s + 7.75) (s − 7.75)
=
δe (s2 + 0.9 s + 8)
This is a lowly damped mode with the damping ratio ζ = 0.133 and frequency of ωn =
3 rad/sec. There is a need to improve both damping and frequency. Therefore, a two loop
autopilot, with inner loop feedback of pitch rate for damping augmentation and an outer loop of
acceleration feedback is needed. Analysis of pitch stabilization/ damping augmentation system
has already been presented earlier.
−15 (s+0.4)
The OLTF for the inner loop is given by G H|i (s) = Kq 10
(s+10) (s2 +0.9 s+8)
. {Homework:
Conduct root locus analysis for Kq ∈ [0, ∞). Find the closed inner loop transfer function for
Kq = 0.23.} The closed loop transfer function for the inner loop with Kq = 0.23 is given by
θ̇(s) −150 (s + 0.4)
= (3.35)
ea (s) (s + 3) (s2 + 7 s + 24)
Block diagram of outer acceleration control system is shown in Fig. (3.25), where the
rate gain Kq = 0.23 is already absorbed. For the autopilot/ servo/ tracking system of Type 1
system, an integrator with an amplifier, representing ideal / textbook type of cascade com-
pensator is used. On using the inner loop feedback, the zeros of the closed system does not
change, but the poles do get altered. To analyze the unity gain acceleration feedback, transfer
function ae z = az ÷ eθ̇ is required. Transfer function e (s) is given in eqn.(3.35) &
θ̇(s)
a θ̇ a a
az −0.042 (s + 7.75) (s − 7.75)

=
θ̇ −15 (s + 0.4)
Figure 3.25: Block diagram for Outer loop for the acceleration control system.
Figure 3.26: Root locus for Outer loop for the acceleration control system [7, p.81].
The factor −77.7 is changed to −0.042 due to change unit for az from f t/sec to g. Product
of two transfer functions az
θ̇
& θ̇
ea
augmented with Integral compensator Ksc results in
az Kc −0.042 (s + 7.75) (s − 7.75)
=
εa s (s + 3) (s2 + 7 s + 24)
where, ε = az(comm) − az , az(comm) = commanded acceleration. Note that the change of output
variable changes only the zeros of the transfer function (Note: TF: C (s I − A)−1 B, poles
depend on det|(s I − A)| and not on C or B matrices). Therefore, poles as indicated in eqn.
(3.35) will remain the same. Root locus plot for this OLTF (for unity gain feedback system) is
given in Fig. (3.26). The closed loop poles for the forward amplifier gain Kc = 2.2 is marked
by a box on the locus. The corresponding closed loop transfer function is [7, p.80, (eqn.2-12)]:
az −0.93 (s + 7.75) (s − 7.75)

=
az(comm) (s2 + 2.2 s + 2.4) (s2 + 7.8 s + 24)
The acceleration autopilot is not generally used in stand alone mode due to operational dif-
ficulties (Refer Blakelock [7, p.80]). However, acceleration feedback is used along with pitch
rate in many modern combat aircraft for maneuver load alleviation. These designs are based
on MIMO state space design techniques.
Automatic landing : Glide slope coupler and Flare control: Automatic land-
Figure 3.27: Block diagram for landing configuration of jet aircraft [7, p.82].
ing system guides an aircraft to a predetermined glide slope at a specified altitude and make
it slide along glide till the aircraft nears the ground. In order to reduce impact load on the
landing gear, landing system will flare out the touch down trajectory. These two trajectories
require control of flight path angle γ. This requires control of airspeed u, either manual or
automatic. Therefore, the landing control system has two distinct control loops, one
Figure 3.28: Root locus for inner pitch loop of landing configuration of jet aircraft [7, p.83].
loop is similar to the pitch attitude hold autopilot and the second one is Mach hold (feedback
of airspeed to engine thrust control input. The specification for pitch attitude autopilot dif-
fers substantially from that discussed earlier, though block diagram is similar. Here, attitude
displacement/ attitude sensor is needed, while in the former case one could do with a rate
integrating gyro (Refer [7, §2.2]. Basic block diagram of automatic control aircraft for ≈ sea
level at 85m/sec is shown in Fig. (3.27). For analysis, three degree of freedom math model
(Eqn. 3.36) with state variables x = [u, α, θ]T is used.
(7.4 s + 0.15) u(s) − 0.25 α(s) + 0.85 θ(s) = CxT δ(rpm)(s)
1.7 u(s) + (7.4 s + 4.5) α(s) − 7.4 θ(s) = −0.246 δe (s)
(0.118 s + 0.619) α(s) + (0.59 s2 + 0.41 s) θ(s) = −0.71 δe (s) (3.36)
The eigenvalues of the plant are located at [0, −0.7500 ± 0.9526 j]. The OLTF for the inner
12 (s+0.585)
loop is G H(s) = (s+10) (s2 +1.5 s+1.47)
. The root locus for pitch inner loop is plotted in Fig.
(3.28). The pitch rate feedback gain Kq is selected such that the complex poles lie on the
outward going branch of locus approaching asymptotes at ±90◦ , with a damping ratio of
around or higher than 0.7 (as against a double pole pair at the break-in point in the earlier
analysis). The operating point marked (△) on locii has a damping of 0.866 for the dominant
mode at the gain value Kq = 2.5 volt/(rad/sec). For analyzing the outer loop, pitch rate
feedback loop is merged to get closed inner loop transfer function. The outer loop OLTF for
the root locus analysis is given by

Kc 12 (s + 0.585) 1 12 s + 7.02
G H(s) = =
(s + 0.9) (s2 + 10.4 s + 36) s s4 + 11.3 s3 + 45.36 s2 + 32.4 s
Number of asymptotes = 3, angle of asymptotes = ±60◦ , 180◦ which intercept the real axis
Root Locus
8
6 System: sys
Gain: 3.88
Pole: −2.68 + 1.98i
4 Damping: 0.804
Overshoot (%): 1.42
2
Imaginary Axis
−2
−4
−6
−8
−8 −7 −6 −5 −4 −3 −2 −1 0 1
Real Axis
Figure 3.29: Root locus for outer pitch loop of landing configuration of jet aircraft.
at −3.5717. The location of the dominant pole for the cascade gain Kc = 3.88 is marked on the
Matlab drawn root locus in the Fig. (3.29). The characteristic equation for Kc = 3.88 volt/volt
is s4 + 11.3 s3 + 45.36 s2 + 78.96 s + 27.2376 = 0. The closed loop system eigenvalues/ poles
are located at [−5.4980, −2.6771 ± 1.9738 j, −0.4478]. For the dominant complex conjugate
poles the damping ratio ζ = 0.805 and natural frequency ωn = 3.3 rad/sec. Reason for this
selection of gain Kc = 3.88 will be clear after the discussion on outer glide slope coupler is
analyzed through root locus analysis.
Homework: The state equations (Eqn. 3.36) in frequency domain for aircraft flying at ≈ sea
level with a speed of 85m/sec in landing configuration is given in Blakelock [7, p. 82, eqn.(2-
13)]. Obtain the unit step response for all the three state variables and flight path angle γ,
with no control, with inner loop stabilization Kq = 2.5 volt/(rad/sec) and with inner and
outer pitch attitude control (Kc = 3.88 volt/volt).
Figure 3.30: Block diagram for velocity control of landing configuration of jet aircraft.
Note that oscillation response for γ reduces with the addition of inner loop. The
flight path angle γ shows no oscillations (except initial undershoot) with the two loop autopi-
lot. The steady state value of flight path angle γ is 0.25◦ , which ensures a gentle descent.
However, the steady state pitch attitude θ is positive (instead of having same sign as that of
γ). Hence some form of velocity control needs to be exercised. Fig. (3.30) shows the block
diagram of a typical speed control system for jet aircraft described above. The engine thrust
is controlled by varying the engine rpm as the input. A first order lag model with a time
constant of 10sec is used. The reader is advised to read the glide slope and flare control from
the Reference (Blakelock, [7, pp. 84 - 98]).
Flight path stabilization [7, pp. 98 - 106]:
Two modes of flight path control related operations are “Mach Hold” and “Altitude Hold”,
which are normally operated during cruise operation. In Mach Hold mode, the aircraft is
made to fly at a constant speed by controlling the flight path angle through elevators and the
engine power is adjusted to yield desired speed. Altitude Hold mode is useful for long range
operation, in say, transport aircraft. For analysis of these two modes, transfer function for
the jet transport cruising at 600ft/sec (≈ 183m/sec) at an altitude of ≈ 12.2km (Example 2).
Mach Hold Mode Autopilot :

Fig. (3.31) is a block diagram of the Mach hold mode autopilot/ flight control system. The
q(s) s θ(s)
transfer function δ (s) = δ (s) with 4 state longitudinal model is (Eqn. 3.27):
e e
q(s) − 1.31 s (s + 0.3)(s + 0.016)

= 2
δe (s) (s + 0.00466s + 0.0053) (s2 + 0.806s + 1.311)
Figure 3.31: Block diagram for Mach hold mode autopilot.
The root locus for the inner loop for full state model (Fig. 3.32 for expanded root
locus near origin)differs slightly from that of reduced order short period model given in the
Fig. 3.14. (Note small deviation in poles for full order Vs short period model of Example
1 for F-16). For the short period model, at Kq = 1.98, the dominant poles were located at
−5 ± 2 j. The closed loop transfer function for Kq = 2.1 for the full model is given by
q(s) 13.1 s (s + 0.3) (s + 0.016)
=
ea (s) (s + 0.8) (s2 + 10 s + 29) (s2 + 0.01 s + 0.003)
The phugoid modes are at −0.0050 ± 0.0545 j. For the outer loop, forward speed is fed-back,
u(s) u(s) θ̇(s)
hence, the transfer function
θ̇(s)
= q(s) is required so that it can be multiplied to δe (s) to
u(s)
get δ (s) in the top branch of aircraft dynamics. The transfer function is
e
u(s) 0.23 (s − 68.8) (s + 0.6)

=
θ̇(s) 57.3 s (s + 0.016) (s + 0.3)
Figure 3.32: Expanded view of region near origin of the root locus for pitch rate stabilization
for the jet transport aircraft.
Dominant poles at [(s2 + 10 s + 29)] are same for both the models (Note: gain has changed
from 1.98 to 2.1). The factor 57.3 is required to convert from radians to degrees. Block
diagram of outer loop (after merging pitch loop with gain of 2.1) is given in Fig. (3.33) and
the corresponding complementary root locus in Fig. (3.34). Owing to the negative gain of
−0.525, the cascade amplifier gain Kc is varied from (∞, 0] or −Kc ∈ [0, ∞). As the gain
is increased, root locus crosses the imaginary axis, hence the closed system is conditionally
stable. The gain Kc = 1.19 is selected (closed loop poles is marked in the figure). Poles
corresponding to phugoid mode (Obtain expanded view of root locus) near the origin move
aound, break into the real axis and become dominant roots for the time response analysis.
Altitude Hold Mode Autopilot :

The block diagram of the altitude hold mode is drawn in Fig. (3.35). Altitude hold mode
implies forcing the flight path angle γ to zero. For analysis of such a system, short period
model is sufficient, so that analysis can be hand calculated too. The inner loop design is same
as discussed in pitch rate feedback for jet transport; refer Fig. 3.14. For, Kq = 1.98, inner
Figure 3.33: Block diagram for Mach Hold Autopilot.
Figure 3.34: complementary Root locus for Mach hold mode.
loop (pitch rate augmented) closed loop system transfer function is

q(s) 13.9 (s + 0.306))
=
ea (s) (s + 0.8) (s2 + 10 s + 29)
Aircraft dynamics block in Fig. (3.35) has two outputs, viz., q = θ̇ and altitude h. Even
Figure 3.35: Block diagram for Altitude hold mode.
after augmenting the inner loop, sensor outputs remain the same. Since this analysis deals
h(s) h(s)
with the SISO system, an additional function θ̇(s)
= q(s)
is to be cascaded at the output of
h(s)
q = θ̇, so that the total transfer function becomes δ (s) as in Fig. (3.36).
e
The rate of change of altitude is related to flight path angle by the relation ḣ = V γ.
Longitudinal acceleration az = V γ̇. Since az is taken positive downward and h is taken
positive upwards; ḧ = −az . In frequency domain; h = − az / s2 . Therefore, additional
transfer function is
h(s) 1 az (s) −0.135 (s − 4.89) (s + 4.89)

=− 3 =
θ̇(s) s θ(s) s2 (s + 0.306)
The complementary root locus is plotted in Fig. (3.37). The closed loop poles for the dominant
short period mode is marked for the gain Kc = 0.493.
So far, a limited number of longitudinal flight control systems (FCS) have been
discussed. These analysis demonstrate the use root locus plots. Reader can refer to the
Bibliography for further details and also additional FCS. Next, Lateral a few Lateral FCS are
discussed from the view point of application of root locus analysis.
Figure 3.36: Block diagram for the outer loop of Altitude hold mode.
Figure 3.37: Root locus for altitude hold mode.
3.2.2 Lateral Flight Control System [35, 7]
The state space representation for the lateral dynamics is given by xT = (β, p, r, ϕ), u = (δa , δr )
 
Yβ Yp
( YVr − 1) g/V   
 V V Ya Yr
   

0 
V V
 Lβ Lp Lr   
A= B = 
 La Lr

 (3.37)
   
 N Np Nr 0   
 β 
  Na Nr
0 1 0 0
Three independent modes (viz., roll, spiral & Dutch roll) of lateral dynamics correspond
to damped response in roll rate p, to a long term tendency either to maintain wing level or
roll-off to divergent spiral and to weather cocking directional stability.
Example 4: Transport aircraft For a typical aircraft (L1011) [ Ref??]:
   
 −0.746 0.006 −0.999 0.0369   0.0012 0.0092 
   
  
 −12.9 −0.746 0.387 0   6.05 0.952 

A=

B = 
 

 (3.38)
 4.31 0.024 −0.174 0   −0.416 −1.76 
   
   
0 1 0 0 0 0
Eigenvalues for the lateral motion of the aircraft consists of

Dutch roll mode = -0.4473 ± 2.0724j (ζ = 0.211, ω = 2.12rad/sec.)
Roll subsidence mode = - 0.7653
spiral mode = - 0.0062
On examination of A & B matrices, it can be seen that aileron input (6.05 Vs 0.952)
dominantly influences roll rate p, and rudder input (-1.76 Vs -0.416) dominantly influences
yaw rate r. Hence, rudder excites primarily the Dutch roll mode. From the observation
or sensor output point of view, Dutch roll mode is dominantly observed in the yaw rate.
Therefore, for improving the Dutch roll damping, detect/ measure yaw rate and feed it to
the rudder. Similarly, side slip response (most of the time, causes uncomfortable ride quality)
from aileron deflection is caused by the yawing moment, which has to be counteracted.
Diagonal or modal representation of Lateral dynamics for the above given A, B, &
C = I4 , D = 0 is given by
   
 −0.4473 2.0724 0 0   1.544 6.004 
   
   
 −2.0724 −0.4473 0 0   −0.254 −1.3754 
Ad = 

 ; Bd = 
 


 0 0 −0.7653 0   −7.84 6.97 
   
   
0 1 0 −0.0062 6.259 −5.570
 
 −0.0280 −0.1354 −0.0001 0.0014 
 

 0.8584 0 −0.6075 −0.0062 

Cd = 

 ; Dd = zeros(4, 2).

 −0.2705 0.0839 0.0257 0.0358 
 
 
−0.0854 −0.3957 0.7939 0.9993
Modal representation (Ad , Bd , Cd , Dd ) indicate that rudder has greater influence of Dutch
roll mode than aileron. Aileron and rudder have good influence on roll subsidence and spiral
modes. Side slip angle β is influenced by dutch roll mode, while yaw rate r can influence
Dutch roll damping significantly. Roll subsidence and spiral modes have strong influence on
bank angle output ϕ. Roll subsidence also has strong component in roll rate p output.
The measurement or output vector is y = [p, r, β, fycg , ϕ]T , where fycg is the lateral
acceleration at the CG of the aircraft. On an aircraft, a subset of above mentioned sensors are
normally available, say, only p, r β are used for feedback. Then, fycg and ϕ are incorporated
in the C matrix to enable monitoring of the handling quality assessment.
Example 5: A generic fighter aircraft [Ref ??]: For the design and analysis of
the controller, a linearized model of say, a generic fighter aircraft at 0.51 Mach number, 1.38
km altitude corresponding to a 6 g flight condition is considered. This flight condition is
characterized by a high spiral instability, a poor Dutch roll damping and high roll mode time
constant. The trim angle of attack is about 14.4◦ , hence the model is suitable to demonstrate
the advantages of ‘Rolling around stability axis’. The corresponding state space matrices are:
   
 −2.85 2.52 −76.11 0.00   −103.6 9.44 
   
   
 −0.31 −0.46 7.36 0.00   −6.21 −5.08 
A=

B = 
 

 (3.39)
 0.25 −0.96 −0.43 0.06   0.09 0.06 
   
   
1.00 0.26 0.00 0.00 0.00 0.00
   
 1.00 0.00 0.00 0.00  0.00 0.00 
   
   
 0.00 1.00 0.00 0.00   0.00 0.00 
   
   
C=
 0.00 0.00 1.00 0.00  D = 
 0.00 0.00 
 (3.40)
   
   
 
0.15 −0.03 −37.43 0.00   
  5.08 −0.12 
   
0.00 0.00 0.00 1.00 0.00 0.00
From the state space model, it is seen that the number of state variables, n= 4; the number
of independent input variables, m = 2. Since, only the first three parameters of the matrix C
are available for the feedback, number of independent output variable, r=3. The system has
a rank of 4, the matrices B and C have the ranks of 2 and 3 respectively. Hence, the aircraft
is controllable and observable.
The aircraft response to a disturbance or a command input consists of all the three
modes. The degree to which a mode responds depends on, whether the elevon or rudder
surface are deflected, the cross coupling between the modes, etc. The eigenvalues and the
corresponding eigenvectors of the open-loop aircraft, considered in the design, are listed in
the following tables.
Open -loop eigenvalues:
Dutch roll mode = −0.8215 ± 4.7995i (ζ = 0.1687, ω = 4.8693)
Roll mode = −2.0095 (τ0 = 0.4976sec.)
Spiral mode = +0.0104 (T imetodouble = 66.63sec.)
Open-loop eigenvectors: From the above analysis, it can be observed that the Dutch
Modes/ States Dutch roll mode Roll mode Spiral mode

p -0.9622± 0.0899i -0.8630 -0.0247
r 0.0787 ± 0.1270i -0.1902 0.0552
β 0.0385 ± 0.0613i 0.0037 0.0031
ϕ 0.0447 ± 0.1885i 0.4507 0.9982
roll mode is lightly damped and appears dominantly in the roll response as evidenced by the
dominant components in the corresponding eigenvectors. The roll mode mainly appears in
the roll rate and roll angle response. The unstable spiral mode appears almost entirely in the
roll angle response.
Example 6: A generic jet transport aircraft [7, Ch.3]:

Linearized lateral dynamics for jet transport flying straight and level at 300mph at sea level
is given Laplace domain as:
(0.02725 s2 + 0.0553 s) ϕ(s) − (0.0128 s) ψ(s) + 0.057 β(s) = 0.0131 δr (s) + 0.68 δa (s) (3.41)
(0.00338 s) ϕ(s) + (0.0585 s2 + 0.0158 s) ψ(s) − 0.096 β(s) = −0.08 δr (s) − 0.01 δa (s) (3.42)
−0.344 ϕ(s) + (4.71 s) ψ(s) + (4.71 s + 0.6) β(s) = 0.171 δr (s) (3.43)
s ϕ(s) = p(s) ⇒ kinematics (3.44)
s ψ(s) = r(s) ⇒ kinematics (3.45)
The state variable ψ(s) does not appear in Eqns. (3.41, 3.42 & 3.43), hence is not considered in
the state space model used for analyzing the system response for short period of time (similar
to variable h in longitudinal dynamics). Since, p(s) = s ϕ(s) ⇒ (0.02725 s + 0.0553) p(s) =

(0.02725 s2 +0.0553 s) ϕ(s), eqn. (3.44) is already embedded in the first three equations, hence
only Eqns. (3.41, 3.42 , & 3.43) are used to find the following transfer functions.
ϕ(s) 0.485 (s + 1.53) (s − 2.73)

= (3.46)
δr (s) (s + 0.380 s + 1.813) (s + 2.09) (s − 0.004)
2
s
134 ( 1.53 s
+ 1) ( 2.73 − 1)
= (3.47)
s 2 2∗0.14∗s s
[( 1.345 ) + ( 1.345 ) + 1] ( 2.09 + 1) ( 0.004 s
− 1)
ψ(s) −1.38 (s + 2.07) (s2 + 0.05 s + 0.066)
= (3.48)
δr (s) s (s2 + 0.380 s + 1.813) (s + 2.09) (s − 0.004)
−12.4 ( 2.07 s s
+ 1) [( 0.257 )2 + ( 2∗0.097∗s
0.257
) + 1]
= (3.49)
s 2 2∗0.14∗s s
s [( 1.345 ) + ( 1.345 ) + 1] ( 2.09 + 1) ( 0.004 s
− 1)
β(s) 0.0364 (s − 0.01) (s + 2.06) (s + 37.75)
= (3.50)
δr (s) (s2 + 0.380 s + 1.813) (s + 2.09) (s − 0.004)
s s s
1.87 ( 0.01 + 1) ( 2.06 + 1) ( 37.75 + 1)
= (3.51)
s 2 2∗0.14∗s s
[( 1.345 ) + ( 1.345 ) + 1] ( 2.09 + 1) ( 0.004 s
− 1)
2
ϕ(s) 22.1 (s + 0.40 s + 1.67)
= (3.52)
δa (s) (s + 0.380 s + 1.813) (s + 2.09) (s − 0.004)
2
s 2
2440 [( 1.3 ) + ( 2∗0.154∗s
1.3
) + 1]
= (3.53)
s 2∗0.14∗s s
[( 1.345 )2 + ( 1.345 ) + 1] ( 2.09 + 1) ( 0.004 s
− 1)
ψ(s) −0.171 (s − 1.14) (s + 9.29) (s + 1.45)
= (3.54)
δa (s) s (s2 + 0.380 s + 1.813) (s + 2.09) (s − 0.004)
−173 ( 1.14s s
+ 1) ( 9.29 s
+ 1) ( 1.45 + 1)
= (3.55)
s 2 2∗0.14∗s s
s [( 1.345 ) + ( 1.345 ) + 1] ( 2.09 + 1) ( 0.004 s
− 1)
β(s) 0.171 (s + 18.75) (s + 0.15)
= (3.56)
δa (s) (s2 + 0.380 s + 1.813) (s + 2.09) (s − 0.004)
s s
31.7 ( 18.75 + 1) ( 0.15 + 1)
= (3.57)
s 2∗0.14∗s s
[( 1.345 )2 + ( 1.345 ) + 1] ( 2.09 + 1) ( 0.004 s
− 1)
The modes are identified as: (s2 + 0.380 s + 1.813), ⇒ ζ = 0.141; ω = 1.345 is the Dutch roll;
(s + 2.09) is the roll subsidence; (s − 0.004) is the unstable spiral divergence. The time to
double for the unstable spiral is very large and hence pilot can handle it manually. However,
the Dutch roll of 0.141 is small to be comfortable from piloting point of view, which calls for
augmentation of damping by feedback action. The roll subsidence is the rolling response of
an aircraft to aileron input. Figs. (3.38, 3.39, 3.40) are respectively Bode plots for transfer
function for ϕ(s), ψ(s), β(s) for the rudder input. Similar Bode plots for aileron inputs can
be drawn (not shown for the sake of brevity).
Homework: Draw Bode plot by hand using the Bode approximations described in §2.5.1.
Bode diagram for φ(s)/ δr(s)
100
80
60
Magnitude (dB)
40
20
−20
−40
90
45
Phase (deg)
−45
−90
−135
−180
−2 −1 0 1 2
10 10 10 10 10
Frequency (rad/sec)
Figure 3.38: Bode diagram for ϕ(s)/δr (s) for jet transport aircraft.
Bode diagram for ψ(s)/ δr(s)
60
40
Magnitude (dB)
20
−20
−40
180
135
Phase (deg)
90
45
−45
−2 −1 0 1
10 10 10 10
Frequency (rad/sec)
Figure 3.39: Bode diagram for ψ(s)/δr (s) for jet transport aircraft.
Bode diagram for β(s)/ δr(s)
20
0
Magnitude (dB)
−20
−40
−60
−80
45
0
Phase (deg)
−45
−90
−135
−180
−2 −1 0 1 2
10 10 10 10 10
Frequency (rad/sec)
Figure 3.40: Bode diagram for β(s)/δr (s) for jet transport aircraft.
Example 7: F-16 fighter aircraft - Landing phase[Ref 35, p. 302]:

The state space model for Lateral dynamics of F-16 fighter aircraft flying sea level with a
speed of 205 ft/sec (α = 18.8◦ ) is given (where, xT = (β, ϕ, ψ, p, r), and u = (δa , δr )) by
   
 −0.1315 0.1486 0.0 0.3243 −0.9396   0.0001205 0.000329 
   
   
 0.0 0.0 0.0 1.0 0.3398   0.0 0.0 
   
   
A=
 0.0 0.0 0.0 0.0 
1.0561  B = 
 0.0 0.0  (3.58)

   
   
 −10.614 0.0 0.0 −1.1793 
1.0023   −0.1032 0.02099 
  
   
0.9966 0.0 0.0 −0.001817 −0.2586 −0.002133 −0.01072
   
 0.0 0.0 0.0 57.2959 0.00   0.0 0.0 
C= D =   (3.59)
0.0 0.0 0.0 0.0 57.2959 0.0 0.0
The third null column in the A matrix shows that the state ψ is coupled back to any other
states, and it can be omitted from the state equations when analyzing the stabilization and
control augmentation systems. The outputs p, r are in degrees/sec and inputs are in degrees.
Hence, input and output matrices are scaled by 1/57.2959 and 57.2959, respectively. The
transfer functions of primary interest are:
p(s) −5.911 (s − 0.05092) (s + 0.2370 ± 1.072 j)

= (3.60)
δa (s) (s + 0.06789) (s + 0.6960) (s + 0.4027 ± 2.012 j)
r(s) −0.1222 (s + 0.4642) (s + 0.3512 ± 4.325 j)
= (3.61)
δa (s) (s + 0.06789) (s + 0.6960) (s + 0.4027 ± 2.012 j)
p(s) 1.202 (s − 0.0528) (s − 2.177) (s + 1.942)
= (3.62)
δr (s) (s + 0.06789) (s + 0.6960) (s + 0.4027 ± 2.012 j)
r(s) −0.6139 (s + 0.5078) (s + 0.3880 ± 1.5439 j)
= (3.63)
δr (s) (s + 0.06789) (s + 0.6960) (s + 0.4027 ± 2.012 j)
(3.64)
The aircraft is stable since all the poles are on LHS of s-plane. The spiral mode is at −0.06789,
roll subsidence at −0.6960 (which is low), and Dutch roll poles ate at (−0.4027 ± 2.012 j) ⇒
ζ = 0.196, ω = 2.05rad/sec. Dutch roll poles ar not canceled by zeros in the transfer
function for roll rate p/δa ; hence, coupling exists between rolling and yawing motions and
Dutch roll will involve secondary effects of rolling motion. At lower angle of attack α, Dutch
roll poles get nearly canceled by the zeros of transfer function, leaving only roll subsidence
and spiral modes as dominant. The two roll rate transfer functions given by Eqns. (3.60,
3.63) contain non-minimum phase (NPZ) zeros close to the origin. This is because gravity
will cause the aircraft to sideslip as it rolls. Then, if the dihedral derivative Cl β is negative
(positive roll stiffness), the aircraft will have a tendency to roll in the opposite direction. The
rudder to roll rate transfer has another NPZ zero away from the origin which has has faster
NMP effects. Positive deflection of rudder produces a positive rolling moment and a negative
yawing moment. A negative yawing moment rapidly leads to positive sideslip, which will in
turn produce negative rolling moment if the aircraft has positive stiffness. This effect tends
to cancel the initial positive roll. NMP zero is the manifestation of these competing effects.
The Handling Qualities and Stability requirements :
The stability and control aspects are of utmost importance in the flight control system (FCS)
design. An aircraft, possessing all the stability characteristics, may not be flight worthy until
it meets certain minimum requirements of Handling Qualities. A special issue on the Aircraft
Handling Qualities was published in the Journal of Guidance, Control and Dynamics during
the year 1986. The issue contains many papers on the aircraft’s flying qualities and handling
qualities. In the editorial of the journal, the handling qualities of an aircraft is defined as:
‘Those qualities of an aircraft which govern the ease and precision with which a pilot is able to
perform his mission with the aircraft which has a set of flying qualities’. The flying qualities
are usually characterized by a number of frequency domain parameters such as damping ratio,
natural frequency, etc. of various modes of the aircraft. Knowledge of these parameters help
the designer to visualize the aircraft response to a command or a disturbance input. In a
flight, human pilot forms the dynamic element for closing the outer loop of the FCS. The
handling qualities, therefore, depend not only on the flying qualities but also on the visual
motion cues available to the pilot and the cockpit ergonomics. Depending on the ease and
precision with which a pilot can carry out a mission, the handling qualities are rated on
a scale of 1 to 10 called Cooper-Harper rating. Rating 1 corresponds to the excellent and
highly desirable handling qualities, whereas rating 10 corresponds to a major deficiencies in
the aircraft. Handling quality specifications become important particularly when the aircraft
exhibits unwanted characteristics such as roll ratcheting, pilot induced oscillations, etc. The
aircraft handling qualities are characterized by a number of parameters in complex frequency
domain such as damping factor, natural frequency of different modes, etc. Extensive research
has been carried out to translate the handling quality specifications to the design specifications.
Lateral Flight Control System (FCS)
Majority of current day aircraft have poor Dutch roll damping. For the transport aircraft,
wing leveler and coordinated turn are critical. For combat aircraft, lateral acceleration and
hence side force has to be severely controlled during large maneuvers, hence coordinated turn
in which the aircraft is rolled about the velocity axis. Normally, Dutch roll damping in an
aircraft is inadequate. As a result, whenever, rudder is activated, the mode gives oscillatory
yawing motion with coupling into the rolling motion, where coupling depends on the relative
size of stability derivative Lr . Stability augmentation system artificially increases damping by
augmenting the term Nr . Body axis roll rate is fed back to the ailerons to modify the roll -
subsidence mode. Yaw rate is fed back to the rudder to modify the Dutch - roll mode (known
as Yaw damper).
Damping of Dutch roll mode (Yaw damper):
While discussing the state space model, it is noted that rudder excites primarily the Dutch
roll mode. At the sensor output, Dutch roll mode is observed in yaw rate and side slip angle.
The purpose of yaw damper feedback is make the rudder to generate a yawing moment that
opposes any yaw rate build up from the Dutch roll mode. Velocity or rate feedback enhances
the damping. If a (near) constant yaw rate (rate of heading angle, for say, circular path/ in
a coordinated steady state turn) has to be executed, then with yaw rate feedback operating
with a constant non-zero value opposes the circular motion, in the sense that it drives yaw
rate towards null value. For example, a positive yaw rate produces positive deflection, which
would result in uncoordinated maneuver. Therefore, the pilot must apply a much larger than
the normal rudder pedal input (increases pilot work load). Therefore, a washout (high pass)
filter is added in the feedback loop, so that the output of the filter is zero for constant yaw rate
‘r’. Feedback produces output only during the transient period. Block diagram of Dutch roll
damper is given in Figs. (3.41, 3.42). Please note the positive sign at the summing junction
for the feedback signal. The transfer function for the washout filter is τ τs+1
s
. To illustrate the
impact of time constant τ of washout filter on the damping augmentation, two case studies
are with two distinct time constant are evaluated. For SISO analysis, only the rudder input
is included in Fig. (3.42).
Figure 3.41: Block diagram of Dutch roll damper.
r(s)
The transfer function δ (s) for a jet transport (after pole-zero cancelation in eqn.3.49) is:
r
r(s) −1.38 (s2 + 0.05 s + 0.066)

=
δr (s) (s − 0.004) (s2 + 0.380 s + 1.813)
Figure 3.42: Block diagram of Dutch roll damper for Root locus study.
The OLTF with a first order actuator lag τ = 3sec and τ = 0.5sec are respectively:
3s −13.8 (s2 + 0.05 s + 0.066)

OLT F G H(S) = with τ = 3sec.
1 + 3 s (s + 10.0) (s − 0.004) (s2 + 0.380 s + 1.813)
0.5 s −13.8 (s2 + 0.05 s + 0.066)
OLT F G H(S) = with τ = 0.5sec.
1 + 0.5 s (s + 10.0) (s − 0.004) (s2 + 0.380 s + 1.813)
The root locus for these two cases are given in Figs. (3.43, 3.44) as rate gyro gain Kc is varied.
As seen from the Fig. (3.44), if the time constant is not large enough, little increase in
damping ratio can be achieved. For larger time constants τ (Fig. 3.43), then considerable
improvement can be obtained. Decent damping of Dutch roll mode is obtained for Kr = 1.04.
Corresponding closed loop poles of the inner loop is used in the outer loop analysis at a later
−(41.4 s3 +2.07 s2 +2.7324 s+0)
discussion. For τ = 3sec, the OLTF G H(S) = (3.0 s5 +32.128 s4 +27.0904 s3 +59.8941 s2 +17.89 s−0.0725)
.
Bode plot is given in Fig. (3.45). Noting the positive sign (against conventional negative sign)
at the summing junction of the feedback loop, characteristic equation for the closed loop is:
Dencl = Den − Kr ∗ N um = 3.0 (s5 + 32.128 s4 + 27.0904 s3 + 59.8941 s2 + 17.89 s − 0.0725) −
1.04 ∗ (−(41.4 s3 + 2.07 s2 + 2.7324 s + 0))
= 3 ∗ (s5 + 10.7093 s4 + 23.3821 s3 + 20.6823 s2 + 6.9106 − 0.0242) = 0.
Roots of the characteristic equations are : [−8.1347; −0.9622; −0.8080 ± 0.4891 j; +0.0035].
The dominant pole has a damping of ζ = 0.855 and frequency of ω = 0.945 rad/sec. However,
Figure 3.43: Root locus for Dutch roll damper with τ = 3sec for washout filter.
Figure 3.44: Root locus for Dutch roll damper with τ = 0.5sec for washout filter.
the spiral mode for the closed loop system is still unstable (pole at +0.0035), since yaw rate
feedback to rudder can not stabilize or enhance stability margin for the spiral mode. Since,
the corresponding time to double is large and the closed loop spiral mode less unstable as
compared to the open loop mode, this low level of instability of spiral mode is acceptable.
Bode Diagram
Gm = Inf dB (at Inf rad/sec) , Pm = −97.2 deg (at 0.898 rad/sec)
20
0
Magnitude (dB)
−20
−40
−60
−80
−100
180
90
Phase (deg)
−90
−180
−4 −3 −2 −1 0 1 2 3
10 10 10 10 10 10 10 10
Frequency (rad/sec)
Figure 3.45: Bode plot for Dutch roll damper with τ = 3.0sec for washout filter.
Roll damper/ Yaw damper or stability augmentation for F-16 aircraft in Example7
Fig. (3.41) gives the block diagram of Dutch roll damping augmentation system (also called
yaw damper). The washout filter time constant is a compromise selection; too large a value
is undesirable since yaw damper will then interfere with the entry into turns. Small value
of the washout filter time constant (τ ) will reduce the achievable Dutch roll damping. Root
locus study can be conducted with a couple of value of τ in the range of say, [0.5, 4.0] to choose
the appropriate value of τ . For the current example, τ = 1.0 sec is chosen. Experience [35,
p.303] has shown that damping augmentation system is less critical loop, hence selection of τ
need not be done to first/ second decimals. The actuator model is a first order lag transfer
function with a pole at −20.2. The open loop transfer function (OLTF) for root locus study:
s −12.40 (s + 0.5078) (s + 0.3880 ± 1.5439 j)

G H(S) = with τ = 1.0 sec.
1 + s (s + 20.2) (s + 0.06789) (s + 0.6960) (s + 0.4027 ± 2.012 j)
−12.4 (s4 + 15.9191 s3 + 36.31 s2 + 15.957 s)
=
(s6 + 22.7693 s5 + 58.3417 s4 + 138.2573 s3 + 167.62 s2 + 69.9539 s + 4.0187)
Six Poles of OLTF are at [−20.20, −0.4027 ± 2.0120 j, −1.0000, −0.6960, −0.0679] and four
zeros at [0, −0.3880 ± 1.5439 j, −0.5078]. Number of poles = 6, number of zeros = 4, hence,
n − m = 2 = number of asymptotes at ± 90◦ . The root locus diagram obtained using the
Matlab is shown in top part of Fig. (3.46). The break away point is at −10.6 with a gain
of Kr = 7.34. The bottom part of plot shows the closeup view of the root locus. Maximum
damping of 0.303 at ω = 1.85 rad/sec is achieved for τ = 1.0 sec. Improvement in Dutch
roll damping can be achieved by increasing the value of τ . (Homework: Conduct root locus
analysis for τ = 2.0 sec and τ = 3.0 sec and comment on the maximum achievable Dutch roll
damping. Obtain the closed loop poles at this maximum damping condition.)
Turn Coordination Control :
The lateral acceleration experienced by an aircraft depends on the sideslip angle. Crew comfort
demands small/ near zero sideslip angle, since human body is accustomed to vertical gravity
force and not so much on the side force. The yaw rate feedback in the inner loop reduced
the transient due to Dutch role mode, but not control the sideslip angle β. There are many
ways of achieving coordination, say by feedback of side slip angle β, or its rate β̇, or lateral
acceleration ay . The closed loop poles of the inner loop is the open loop poles for the outer
loop.
Block diagram for coordinated turn using β feedback is shown in Fig. (3.47). By
redrawing this figure, a simplified block diagram is obtained (Fig. 3.48) such that output of
the inner loop is yaw rate r. An independent output β(s) = β(s)
r(s)
× r(s), where,
β(s) −0.0364 (s − 0.01) (s + 2.06) (s + 37.75)

=
r(s) 1.38 (s + 2.07) (s2 + 0.05 s + 0.066)
To simplify the analysis, pole at −2.07 can be canceled with the zero at −2.06, and pole at
+0.01 with zero at +0.04 of the inner loop transfer function. Further simplified block dia-
gram is redrawn as in Fig. (3.49). Poles of the OLTF are located at [−8.1000; −0.8500 ±
0.5268 j; −0.9300] and zeros at [−37.55; −0.33]. The corresponding root locus is shown in
Fig. (3.50). Various break away point (−5.87 @ Kc = 4.37) and break in points (−5.87 @ Kc =
4.37 and −5.87 @ Kc = 4.37)are also shown in this figure. Fig. (3.51) show that the closed
loop system is unstable for Kc ≥ 1.56. For the damping ration of ζ = 0.7, the gain Kc is only
0.373. Such a low gain is not nice.
The short comings of the r - β feedback in the two loop autopilot (Fig. 3.47) can be
minimized by using two loop autopilot with β̇ - β feedback. Block diagram of two loop stability
augmentation system is shown in Fig. (3.52). The inner loop feedback requires the following
Root Locus
3
System: sys
Gain: 7.34
1 Pole: −10.6 − 4.2e−007i
Damping: 1
Imaginary Axis
Overshoot (%): 0
0
−1
−2
−3
−20 −18 −16 −14 −12 −10 −8 −6 −4 −2 0
Real Axis
Root Locus
2.5
1.5 System: sys

Gain: 3.54
Pole: −0.56 + 1.76i
1 Damping: 0.303
Overshoot (%): 36.8
Imaginary Axis
0.5 Frequency (rad/sec): 1.85
−0.5
−1
−1.5
−2
−2.5
−1 −0.8 −0.6 −0.4 −0.2 0 0.2
Real Axis
Figure 3.46: Root locus for Dutch roll damper with τ = 1.0sec for washout filter; top: normal
plot; bottom: closeup view.
transfer function for F-15 aircraft flying at Mach 0.8 and altitude of 20,000ft ( 6.1km):
˙
β(s) 0.0573 (s − 0.0061) (s + 2.783) (s + 79.77)
=
δr (s) (s + 0.024) (s + 2.95) (s2 + 0.492 s + 7.46)
Since β̇(s) = s β(s), the two loops with individual gains Kβ and Kβ̇ can be merged to get
Figure 3.47: Block diagram for turn coordination using sideslip angle β feedback in the outer
loop.
Figure 3.48: Simplified Block diagram for turn coordination using sideslip angle β feedback.
Kβ̇ [s+(Kβ /Kβ̇ )]

the common feedback control transfer function H(s) = s
. For simplicity, assume
that Kβ = Kβ̇ , so that root locus can be drawn by varying a single feedback gain parameter
Kβ̇ . The root locus for the two loop stability augmentation system is shown in Fig. (3.53).
The location of closed loop poles for Kβ = Kβ̇ = 3 is marked on the locus by a black square.
Figure 3.49: Block diagram of outer turn coordination loop for Root locus study.

1
20
0.8
15
0.6
10 System: sys System: sys System: sys

Gain: 2.17e+004 Gain: 4.37 0.4 Gain: 1.66
Pole: −54.8 − 6.47e−007i Pole: −5.87 Pole: 0.0118 + 1.17e−008i
Damping: 1 Damping: 1 Damping: −1
Imaginary Axis
Imaginary Axis
5 0.2
Overshoot (%): 0 Overshoot (%): 0 Overshoot (%): Inf
Frequency (rad/sec): 54.8 Frequency (rad/sec): 5.87 Frequency (rad/sec): 0.0118
System: sys
0 0
Gain: 0.373
Pole: −0.432 − 0.441i
−0.2 Damping: 0.7
−5
Overshoot (%): 4.59
−0.4
−10
−0.6
−15
−0.8
−20
−1
−60 −50 −40 −30 −20 −10 0 −1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1
Real Axis Real Axis
Figure 3.50: Root locus of outer turn coordination loop; left full; right expanded near origin.
Although, β can be measured with a vane or similar sensor, β̇ cannot, and differentiating the
β signal would amplify the noise significantly from an already noisy signal. Therefore, β̇ must
be computed using algebraic relations:
ay + cos Θ cos Φ
β̇ = P sin α − R cos α + 57.3 g
V
which is a nonlinear relation, where, symbols in capital letters indicate total variable (against
Root Locus
1
0.8
0.6 System: sys

Gain: 1.56
Pole: 4.34e−005 + 0.138i
0.4 Damping: −0.000314
Overshoot (%): 100
Imaginary Axis
0.2
−0.2
−0.4
−0.6
−0.8
−1
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1
Real Axis
Figure 3.51: Root locus of outer turn coordination loop - expanded near origin.
Figure 3.52: Block diagram for β̇ - β SAS.
perturbed state variables of the linearized model). For high performance aircraft, β̇ - β SAS
is superior, but more difficult to implement. Alternate method of achieving coordination by
feeding back the lateral acceleration ay . From implementation point of view, sideslip angle or
acceleration feedback systems are preferred.
By definition, total acceleration lies in the plane of symmetry (longitudinal plane),
Figure 3.53: Root locus for β̇ - β SAS.
which requires that there is no lateral acceleration. By feeding lateral acceleration ay to the
rudder, ay can be driven towards null in a course of time. Typical block diagram for turn
coordination using lateral acceleration is shown in Fig. (3.54), where relation between lateral
ay (s) Sq β(s) f t/sec2 Sq β(s) g
acceleration ay and β, given by δr (s)
= m
Cy β δr (s) rad
= (32.2) (57.3) m
Cy β δr (s) deg
, is used.
Figure 3.54: Block diagram for for turn coordination using lateral acceleration.
Yaw Orientation Control System:

Automatic flight path control require altitude hold mode and heading control. For transport
aircraft, constant heading control relieves/ reduces pilot work load during long cruise between
two cities. Before landing at an airport, aircraft may need to circle many times around the
airport if landing gets delayed. This requires preplanned heading maneuvers specified by
ICAO and air traffic control (ATC) rules. In early days of control implementation, yaw rate
was specified and rate integrating gyro will produce heading error by integration of error
between commanded yaw rate and sensed yaw rate. This integrating process introduces the
integrator ( 1s ) in the OLTF, thus making it a type 1 system. Fig. (3.55) is the block diagram
of a four loop autopilot for yaw orientation control system, which includes Dutch roll damper
and sideslip turn coordination systems. The slowest two outer loops consist of yaw rate (r(s))
integrating gyro and roll rate feedback in the aileron feedback loop. In a steady turn, r(s)
signal is driven to zero, so that signal to the yaw rate integrated gyro is the commanded yaw
rate. The analysis of two loop sideslip SAS is already discussed above, whence attendant closed
loop transfer function can be used as the open loop TF for the outer most loop. However,
such a transfer function is complex, hence a simplified transfer function for the inner loop can
be used for analysis and ease of understanding. Therefore, two inner loop SAS is replaced by
a first order lag model for the outer loop analysis with roll rate as given by
ϕ̇ (s) 23 ϕ(s) 23
= ; =
δa (s) (s + 2.3) δa (s) s (s + 2.3)
For yaw rate feedback
r(s) g ϕ(s)
[T F ] = , V = 440 f t/sec.
δa (s) V δa (s)
Simplified block diagram for outer two loop yaw orientation control system is redrawn in Fig.
(3.56). Simplified block diagram for inner loop of yaw orientation control is shown in Fig.
(3.57). The OLTF consists of double pole at the origin, two poles on the negative real axis
at −2.3 and −10. The OLTF has four poles and no zeros. Number asymptotes are four,
with angles at ±45◦ and ±135◦ . Between two success poles at −2.3 and −10, there will be a
break away point and the root locus branches move towards asymptotes at ±135◦ orientation
(Fig. 3.58)). However, the root locus branch will break away at the origin and move towards
the asymptotes at ±45◦ , thus dual complex conjugate poles are unstable even at low rate
integrating gyro gain value.
Figure 3.55: Block diagram for a conventional yaw orientation control system.
Figure 3.56: Simplified Block diagram for yaw orientation control system.
The roll rate feedback control of the outer loop of the yaw orientation control system
is shown in Fig. (3.59). This loop is required to stabilize the instability introduced by the
yaw rate integrating loop of yaw orientation system. The closed loop poles of inner loop of
yaw orientation control is OLTF for the loop in Fig. (3.59). At rate integrating gyro amplifier
Figure 3.57: Block diagram for inner loop of Fig. (3.56) for root locus study.
Figure 3.58: Root locus for yaw rate feedback system shown in Fig. (3.57).
gain of 0.4 volt/(deg/sec), from root locus (Fig. 3.58), transfer function is
r(s) 6.74 ϕ̇(s) V r(s)

= ; =s ; p(s) = ϕ̇(s)
irg (s) (s + 2.45) (s + 9.99) (s − 0.20 s + 0.28) irg (s)
2 g irg (s)
p(s) 92 s
=
irg (s) (s + 2.45) (s + 9.99) (s2 − 0.20 s + 0.28)
Three asymptotes are ± 60◦ and 180◦ . The transfer function has a zero at the origin, which
is on the left of the unstable poles due to (s2 − 0.20 s + 0.28). The angle of departure from
the unstable pole is towards the left half of s-plane. As the gain Kϕ̇ is increased, the effect of
the pole at −2.45 becomes prominant, hence the locus of this branch turns around and locus
will eventually tend toward the asymptotes at ± 60◦ . For large gain of Kϕ̇ , the closed loop
system once again becomes unstable.
Figure 3.59: Block diagram outer loop of pitch rate feedback for root locus study.
Figure 3.60: Root locus for Block diagram shown in Fig. (3.59).
Root locus study for the yaw orientation control system using one loop at a time
design and analysis technique clearly demonstrate the complexity of the process and limitation
on achievable stability and performance. Majority of such drawbacks can be overcome by
MIMO state space design techniques. They are beyond the scope of this chapter.
3.3 Dynamics of Rockets and Missiles :

Chapter 1, §1.4 presented the basic dynamics a typical tactical missile. For such missile
lateral acceleration tracking control or Lateral acceleration (Latax) autopilots are normally
used. This section discusses, various issues related to control of tactical missiles.
Example TM1: Missile Control: For a typical missile, V = 1253 ft/sec (Mach ∼ 1.2),
Zα = 4170f t/sec2 , Zδ = −1115f t/sec2 , and
Mα = −248rad/sec2 , Mδ = −662rad/sec, Mq ≈ 0. (3.65)
Typical actuator dynamics is

1
δ̇ = (u − δ), where, u is the input to the actuator
τa
The time constant for actuator τa = 0.01 sec. The combined transfer function for missiles
with actuator is
−1115K(s2 − 2228)
G(s) = −
(0.01s + 1)(s2 + 3.333 s + 248)
Note that dc gain of G(s) is positive: A positive input produces positive response. But high
frequency gain − Zδ /Mα is (typically) negative and hence for high frequencies (initial time
(transient part of) response can be found from initial value theorem) G(s) → −1115/(0.01 s+
1) which produces a negative response (under shoot) for a positive input. The change of sign
in the TF as the frequency is increased is a consequence of the right half zero. For a constant
gain latax control system, the root locus is as shown below. The figure indicates that the
closed loop system is conditionally stable. Rate q - feedback is essential. It can also be
observed from root locus that as the gain is increased from zero in the positive direction, one
branch of root locus crosses the imaginary axis at s = 0 and then real root continues to move
towards s = 47.2. The characteristic equation
s3 + (100.33 + K) s2 + 581 s + 24800 − 2228 K = 0 , K = 111500 K
The coefficient of s0 vanishes at K = 24800/2228 = 11.13 and hence there is a pole at the
origin for this value of K.
3.3. DYNAMICS OF ROCKETS AND MISSILES : 243
jω
20 -3
K = 11.13 or K~
~ 10
−100 −47 47 σ
−20
Figure 3.61: Root locus for lateral acceleration feedback control for a missile
For tactical missiles, reference command is lateral acceleration (generated by the

guidance system and transformed to body fixed coordinates). Many text books refer lateral
acceleration as a normal acceleration with a symbol aN = fy . It is preferable to write equation
motion in terms of state variable fy instead of angle of attack α. Differentiation of Eq. (1.35)
yields f˙y = Zα α̇ + Zδ δ̇. The state space model in terms of fy , q, δ can be written as:
Zα Zδ Zδ
f˙z = fz + Zα q − δ+ u (3.66)
V τa τa
Mα
q̇ = (fz − Zδ δ) + Mq q + Mδ δ; (3.67)
Zα
1
δ̇ = (u − δ) (3.68)
τa
Exercise: Using the above three equations, derive the characteristic equation and show it to
be same as in Eq. (1.36).
A majority of the missiles can be classified into two main streams (categories):-
strategic and tactical missiles. Flight control system (autopilot or servo/ tracking control) is
an inner loop that makes the missile to follow the command generated by an outer guidance
loop (apart from from enhancing stability characteristic). For the strategic missiles, command
input is the attitude of the missile while it is lateral acceleration for tactical missile.
3.4 Analysis of Traditional Missile Autopilot

Missile Autopilot Design Objective
An autopilot [4, p.92] is a closed loop tracking system and it is a minor loop inside the main
guidance loop. Most of the current missiles have an autopilot. Missile lateral autopilot for
pitch and yaw are similar, but the roll autopilot or roll stabilization system is far simpler and
less demanding. At first, objectives of a lateral autopilot is presented.
The maintenance of near-constant steady state aerodynamic gain :
Performance of autopilot to guidance input should be made less sensitive to parameter varia-
tion during missile flight, for avoidance of serious degradation of the performance. Since, the
missile speeds and altitude of flight can change over a large range, autopilot gains are sched-
uled as function of Mach number or dynamic pressure. Reduction of autopilot sensitivity
enhances robustness and reduce the control design efforts due to smaller number of operating
points.
To increase the weathercock frequency : since missiles are inertia dominated vehicle, a high
weathercock frequency is essential. From tracking accuracy point of view also calls for higher
bandwidth ⇒ increase in the weathercock frequency.
To increase the weathercock damping : The weathercock damping is under damped especially,
with large static margin and at high altitude (low dynamic pressure). This has several unde-
sirable/ adverse effects. Badly damped oscillatory mode produces large output to broadband
noise and also significant reduction in range due enhanced induced drag. Accuracy or miss-
distance of missile will also be compromised. Large missile oscillations will also increase
dynamic structural load acting on the missile.
To assist gathering : Missiles are launched usually some distance off the line of sight (LOS). If
there are no disturbances, the missile will align with LOS in the presence of thrust misalign-
ment, biases in sensors, cross winds by minimizing the resulting dispersions.
There are many autopilot versions, a few will discussed after dealing Roll stabilization.
Roll Stabilization for Rockets
Roll stabilization for Rockets and missiles depends on type of rockets - Roll angle or roll rate
stabilization. Roll stabilization is most critical in terms of performance of the missile, though
3.4. ANALYSIS OF TRADITIONAL MISSILE AUTOPILOT 245
mathematically and theoretically it is the simplest. Therefore, roll stabilization gets only
cursory treatment, but as a result, many missiles have highly under performed and in one or
two cases led to instability.
Three important modes of roll control for missile are:
Roll on demand (ϕd = ϕ ) or
Roll stabilized (ϕd = 0 ) or
Roll rate stabilized autopilots (p = ϕ̇ = 0)
The state space equations representing roll dynamics in continuous time is given below:
ṗ = lp p + lξ ξ (3.69)
ϕ̇ = p (3.70)
For a small tactical missile, typical values of stability derivatives are lp = −30 sec−1 ; lξ =
70N m/rad; lξ ∈ [70, 7000]. Data for small missiles are also available in text book by Garnell
[4, p.127] and Blakelock [7, §7.2]. The actuator dynamics can be represented either first order
kc
or second order model. For the class room study, first order model [ (s+k ] will suffice.
c)
Roll rate stabilization:

Some of the small missiles, like say, shoulder fired missiles, do not have a roll gyro. Since the
flight duration of such missiles are of the order a few seconds, roll rate stabilization (rate gyro
output feedback) is good enough. Since a majority of the missiles use roll angle stabilization,
further discussion on rate stabilization is not presented.
Roll angle stabilization:
On examination of roll dynamics in Eqs. 3.69 and 3.70, the poles are at origin and at −lp . A
constant gain feedback of roll angle will not provide required speed of response. Recall that
k k l
the open loop transfer function with proportional feedback control is G H(s) = [s (s+kA )c(s+l
ξ
.
c p )]
There are three poles and no zeros, hence the root locus has three asymptotes (± 60◦ , 180◦ ).
As the gain is increased (to enhance loop stiffness), the damping reduces and which finally
leads to instability. Hence, either roll rate feedback has be augmented or at least lead network
needs to be added to the roll angle feedback loop (See Fig. 3.62). Let lp = −30 sec−1 ; lξ =
(s+2)
70N m/rad; kc = 80; and lead circuit transfer function is (s+25) . The root locus plot is
then given in Fig. 3.63. For the gain of KA = 0.722, the closed loop poles are located at
−63.3 ± 64.3 j; −8.37.
Figure 3.62: Roll Autopilot for a missile
Root locus for missile roll autopilot with lead compensator

250
200
System: sys
Gain: 0.723
150 Pole: −63.2 + 64.3i
Damping: 0.701 System: sys
Overshoot (%): 4.56 Gain: 0.722
100
Frequency (rad/sec): 90.1 Pole: −8.37
Damping: 1
Imaginary Axis
50 Overshoot (%): 0
0
−50
−100
−150
−200
−250
−90 −80 −70 −60 −50 −40 −30 −20 −10 0 10
Real Axis
Figure 3.63: Roll Autopilot for a missile
The steady state accuracy of this system is poor. Hence, lag-lead controller is
needed in the place of lead network.
Homework: Study the Roll autopilot for a missile with a redesigned lag-lead network Refer [4,
Fig. 6.10-1 on page 127]. Note that for a roll stabilization system, lag - lead network can be
put in the feedback path or cascade path (preferable).
Missile Lateral Autopilot using One Accelerometer and One rate Gyro
Accelerometer feedback predominantly improves weathercock frequency and rate gyro output
improves damping. If the accelerometer is placed in front of centre of gravity (CG), then a
marginal increment in damping can be obtained. Larger the distance from the CG, larger will
be enhancement in damping. Output of an accelerometer at distance ℓ ahead of CG gives an
output
fys = fysensed = fy |CG +ℓ θ̈ = fy + ℓ q̇, q̇ = Mα α + Mq q + Mδ δ
The term ℓ Mq q is positive if the accelerometer is ahead of CG, otherwise negative. Hence, ac-
celerometer placed behind CG has a slight destabilizing effect due to reduced damping. Since,
the distance ℓ is small, damping enhancement or reduction is small with one accelerometer
feedback. Typical latax (lateral acceleration) autopilot for the missile is given in the Fig.3.64.
Figure 3.64: Lateral Autopilot for a missile
The transfer function q/δ for the missile data in Eq. (3.65) is given by
q −662 (s + 2.9946)
= 2
δ (s + 3.328 s + 248)
It may be noted that positive fin deflection produces negative pitching moment (hence, neg-
ative q). Root locus analysis (Fig. 3.65) is done to get enhanced damping. Alternately,
negative gain may assigned to the servo amplifier and regular root locus analysis can be car-
100
ried out. Augmenting the actuator dynamics ( (s+100) ), the root locus diagram (inner loop)
with q feedback is given in the Fig. (3.65).
Root Locus
150 140
0.62 0.48 0.36 0.26 0.16 0.08
120
100
100
0.78
80
60
50 40
0.94
Imaginary Axis
20
20
0.94 40
−50
60
80
0.78
−100
100
120
0.62 0.48 0.36 0.26 0.16 0.08
−150 140
−120 −100 −80 −60 −40 −20 0 20
Real Axis
Figure 3.65: Root locus for Innerloop of simplified Autopilot for a missile
As the feedback gain kq (Fig. 3.65) is increased from zero to a large value, the
damping of the closed loop eigenvalue increases and at kq = kq−critical , damping ratio becomes
1.0. For determining the outer loop gain ka by root locus analysis, larger damping is preferred
since the outer loop reduces the effective damping while the bandwidth get increased. How-
ever, it may be noted that the missile autopilot in Fig. (3.64) is a type 0 system. If Type 1
system is needed, the PI controller block has to be added in the cascade loop. State space
model with inner loop gain kq is given by:
Zα Zδ
α̇ = α+q+ δ
V V
q̇ = Mα α + Mq q + Mδ δ;
1
δ̇ = (u − δ − kq q)
τa
fz
The open loop transfer function δec
for the outer loop with kq = kq−critical can be obtained as
earlier and root locus analysis carried out to find the gain ka so as to meet design requirements.
Example: Pitch Latax Autopilot:
Consider a cruciform missile in the boost phase of the trajectory. For data, refer [7, §7.3, Case
4] (maximum dynamic pressure regime). It has been concluded from the above discussion
that pitch rate and pitch lateral acceleration feedbacks are needed. The aerodynamic transfer
function for the pitch rate output Gq (s) and the lateral acceleration Ga (s) are given by:
θ̇(s) −469.6 (s + 1.2) az (s) −0.964 (s + 34.41) (s − 34.41)

Gq (s) = = 2 ; Ga (s) = = (3.71)
δt (s) (s + 1.27 s + 72.25) δt (s) (s2 + 1.27 s + 72.25)
az (s) 0.0020528 (s + 34.41) (s − 34.41)
which implies that = (3.72)
θ̇(s) (s + 1.2)
−75
The actuator transfer function is (s+75) . The sign of throttle servo is made negative, so that
positive input produces positive output for the combined actuator - aerodynamic transfer
function. The open loop transfer function for the inner rate feedback loop is given by:
35220 (s+1.2)
GH(s) = (s+75) (s2 +1.27 s+72.25) . Number of poles = 3 and zeros = 1. Two asymptotes of
the root locus have angles of ± 90◦ and intersect the real axis at −37.535. Root locus plot
is shown in Fig. (3.66). Two real repeated roots occur at the gain = 0.042097 and 0.03323
respectively, with corresponding closed loop poles are located at −35.2363; −35.2363; −5.7973
and −53.7322; −11.2689; −11.2689. For the higher gain, dominant pole is at −5.7973, while
dual dominant poles at a lower gains are at −11.2689. Hence, gain of 0.0332 is preferred.
Root Locus
10
6
System: sys System: sys
4 Gain: 0.0421 Gain: 0.0332
Pole: −35.2 Pole: −11.2 + 5.46e−007i
Damping: 1 Damping: 1
Imaginary Axis
2 Overshoot (%): 0 Overshoot (%): 0

Frequency (rad/sec): 35.2 Frequency (rad/sec): 11.2
0
−2
−4
−6
−8
−10
−70 −60 −50 −40 −30 −20 −10 0
Real Axis
Figure 3.66: Root Locus for Inner stabilization loop with pitch rate feedback.
The characteristic equation for the inner stabilization loop for the pitch rate gain of 0.15 [7,
p. 242], is (s3 + 76.27 s2 + 5450.5 s + 11758.35) = 0, hence, the closed loop poles are located
at −37.0227 ± 62.5708 j; −2.2245 (point not marked in Fig. 3.66). Owing to the zero at −1.2,
residue for pole −2.2245 will be low. The closed loop transfer function for the inner loop is
θ̇(s) 35220 (s + 1.2)

= . (3.73)
ea (s) (s + 2.2245) (s2 + 74.045 s + 5285.78)
On using Eqs. 3.72and 3.73, outer loop plant model (35220 * 0.0020528 = 72.3) is
az (s) 72.3 (s + 34.41) (s − 34.41)

= (3.74)
ea (s) (s + 2.2245) (s2 + 74.045 s + 5285.78)
For better robustness and increment in damping, accelerometers are normally placed ahead
of CG. Additional signal due to rigid body rotation (angular acceleration) is ℓ θ̈, where, ℓ =
distance of accelerometer ahead of CG. For this analysis, let ℓ = 0, i.e., sensed acceleration
= acceleration at CG. The compensator has a pole at origin, so that the closed loop system
is Type 1. The zero of the lead compensator is put at −2, and the pole at −20 , so that
it nearly the cancels pole at −2.2245. The negative gain −KA is to get positive output for
positive lateral acceleration command. Therefore, the open loop transfer function used for
the root locus analysis (Fig. 3.67) is given by
az (s) + 72.3 KA (s + 2) (s + 34.41) (s − 34.41)

=
e(s) s (s + 20) (s + 2.2245) (s2 + 74.045 s + 5285.78)
The root locus plot generated using Matlab command is shown in the Fig. 3.68. At the
Figure 3.67: Block diagram of acceleration autopilot outer loop
gain KA = 5.78, the closed loop poles are located at −36.6290 ± 58.1305 j; −10.5675 ±
0.2101 j; −1.8764. The pole at −1.8764 is non-dominant due to the zero at −2. The unit
step response of the closed loop system is shown in Fig. 3.69. The undershoot in the response
is due to non-minimum zero at +34.41. The rise time (10% to 90%) of 0.35sec is a good value
for tactical missile applications.
Root Locus
60
40
System: sys
Gain: 5.78
Pole: −10.6
20 Damping: 1
Imaginary Axis
Overshoot (%): 0
0
−20
System: sys
Gain: 5.9
Pole: −36.6 − 58i
−40 Damping: 0.534
Overshoot (%): 13.8
−60
−60 −40 −20 0 20 40 60 80

Real Axis
Step Response
1.2
System: sys
Time (sec): 0.429
1
Amplitude: 0.9
0.8
Amplitude
0.6
0.4
System: sys
Time (sec): 0.076
0.2
Amplitude: 0.102
−0.2
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
Time (sec)

Missile Lateral Autopilot using Two Accelerometers
For small missiles with short range of operation, two accelerometers can be used since ac-
celerometer is smaller and less expensive than rate gyros. An accelerometer is placed at a
distance ℓa ahead of CG with its output axis along lateral (z-) direction produces a signal equal
to ka (fz +ℓa q̇). Second accelerometer is placed at a distance ℓb behind the CG with its output
axis along lateral (z-) direction produces a signal equal to ka (fz + ℓb q̇). The second compo-
nent of accelerometer output (ℓa q̇) has a stabilizing effect, while later one has destabilizing
effect and hence, it appears that second accelerometer is not useful. However, by ingeniously
mixing signals from two accelerometers, i.e., forward accelerometer gain is increased to 3 ka
and the rear accelerometer gain is increased to -2 ka (i.e., positive feedback) and adding the
two amplified signals, results in negative feedback as in
fz−equivalent = 3 ka (fz + ℓa q̇) − 2 ka (fz − ℓb q̇) = ka [fz + (3 ℓa + 2 ℓb ) q̇]
This is equivalent of having one accelerometer much farther of CG at a distance of (3 ℓa +2 ℓb ).

If one increases the gains of accelerometers to very large value, then the noise in accelerometers
nz1 and nz2 ) will not exactly cancel at all times due to mixing of signals (it could even be
adding up). For example, if the gains have raised to say 11 and 10 times, the equivalent
distance of accelerometer location ahead of CG will be (11 ℓa + 10 ℓb ) but the equivalent noise
will be (11 nz1 + 10 nz2 ), which could be much larger than the composite accelerometer signal.
It is to be noted the stabilizing term (3 ℓa + 2 ℓb ) q̇ = (3 ℓa + 2 ℓb ) (Mα α + Mq q + Mδ δ) largely
affects the coefficient of s2 term in the characteristic polynomial of the closed loop system
rather than s term as in the rate gyro feedback case (due to absence of kq Mδ term). Now,
suppose, the signal from the rear accelerometer is lagged by passing it through a filter having
the simplified transfer function ( 1+T1ra s ). For the moment, angular acceleration term may
be ignored (both accelerometers are assumed to be at CG) for the sake of simplicity and
better appreciation, then total negative feedback of composite/ mixed accelerometer signal
has a gain 3 ka − ( 1+T
2 ka
ra s
), which on rearranging results in ka ( 1+3 Tra s
1+Tra s
). This is equivalent
to a 3 to 1 phase advance network. On including the q̇ term, the feedback term will be
ka ( 1+3 Tra s
1+Tra s
) fz + ( (1+2 ℓ1+T
b )+3 ℓa Tra s
ra s
) q̇. This technique was successfully used in small British
missiles several decades ago.
3.5. ATTITUDE FLIGHT CONTROL SYSTEM FOR LAUNCH VEHICLES 253
3.5 Attitude Flight Control System for Launch Vehicles

For rockets such as launch vehicles and surface to surface ballistic missiles, normally, attitude
control system is used. The corresponding state variables in pitch dynamics are (α, q, θ). In
modeling the rockets/ launch vehicle dynamics, slightly different symbols are used [5]:
Lα Xb
lα
lc CP θ
Tc CG . XF
z
δ v
Tr
V
Z
b
Force and moments equations:

−gcosθ0 Tc δ Lα α
α̇ = θ+ − +q (3.75)
V mV mV
Lα lα α Tc lc
q̇ = + δ = µα α + µc δ (3.76)
Iy Iy
On comparing with the earlier equations (3.16 & 1.29 with 3.76) for missile & aircraft equa-
tions (longitudinal), the equivalence between different symbols can be readily established. To
include drift pole effect, kinematic relation θ̇ = q is added. On taking the Laplace transform,
Lα
θ(s) µc [s + mU (1 + llαc )] µc
= 3 Lα 2 µα gcosθ ≈
δ(s) s + mV s − µα s + V (s − µα )
2
As stated earlier, if CP is a head of CG, lα > 0 and hence µα (similar to Mα , Mα > 0) and
the vehicle is unstable. For a typical vehicle data given in Greensite’s book [5],
θ(s) µc (s + 0.0678) 4.54
= ≈ 2
δ(s) (s − 0.0042) (s + 0.0354s − 3.75)
2 (s − 3.75)
A typical simplified control system for a launch vehicle is shown in the diagram (3.70).
The rate gyro together with the attitude gyro act as a lead compensator having the
transfer function (1 + kR s). The characteristic equation for the closed loop system is given by
s3 + kc s2 + (kA kc kR µc − µα )s + kc (kA µc − µα ) = 0
F or stability, kA µc > µα and kc > 1/kR

θc + θe δc Kc δ µc θ
M
KA
S - µα
S + Kc 2
-
Actuator Airframe
KR S
+ Rate Gyro
M
+ 1
Attitude Gyro
Figure 3.70: Flight control system block diagram for a rocket; H(s) = (1 + kR s)
The static stability criterion: kA µc > µα implies that restoring moment per unit amplitude
change in “δ” must be greater than the over turning moment (due to aerodynamic moment).
This requirement is independent of kR . KAN = µα /µc = Gain for neutral stability = 0.826.
The dynamic stability criterion: kc > 1/kR implies that the lead from the rate gyro must be
greater than the lag from the actuator. Open-loop transfer function for this system is :
kA kc µc (1 + kR s) kA kc kR µc (s + 1/kR )
G H (s) = =
(s + kc ) (s − µα )
2 (s + kc ) (s2 − µα )
Root locus diagrams for this system for two values of 1/kR = 1/3 & 5 are shown in Figs. (3.71
and 3.72). Two diagrams differ only in the relative values of µa and kR . In both the cases,
KAN = µα /µc is the critical gain for neutral stability.
Root Locus
10
4
Imaginary Axis
−2
−4
−6
−8
−10
−16 −14 −12 −10 −8 −6 −4 −2 0 2 4
Real Axis
Figure 3.71: Root locus for simplified Autopilot for a rocket, 1/kR = 1/3
3.6. CONTROL - STRUCTURE INTERACTIONS IN ROCKETS/ LAUNCH VEHICLES 255
Root Locus
40
30
20
10
Imaginary Axis
−10
−20
−30
−40
−16 −14 −12 −10 −8 −6 −4 −2 0 2 4
Real Axis
Figure 3.72: Root locus for simplified Autopilot for a rocket, 1/kR = 5
The condition for stability (which can be collaborated by root locus plot) indicates
that there is no upper limit on kA . However, if higher order dynamics and/ or cascade PI
controller are introduced, then, the system will tend to be unstable for large value of kA .
3.6 Control - Structure Interactions in Rockets/ Launch

Vehicles
The control - structure interactions and effect of sloshing substantially influence the stability
of performance of autopilots designed based on rigid body models. Though the structural
vibrations and propellant (fuel, oxidizer) sloshing are represented by continuum, discretized
model with first few modes are sufficient for the analysis and design of autopilots. Since, these
modes are orthogonal (called normal modes), it is sufficient to include one mode at a time for
the analysis that helps in understanding the effect of control - structure interaction.
3.6.1 Effect of Propellant Sloshing :
The forces and moments produced by sloshing of liquid propellants is modeled by an equivalent
mass and pendulum whose parameters depends on tank shape, liquid level, etc. The
simplified equations of motion are:
∑ ∑
mo (ẇ − Uo θ̇) = Tc δ − Lα α + mpj U̇o Γpj = Fz (3.77)
j=1
l pj
lpj
n
Figure 3.73: Propellant Sloshing schematic representaion.
∑ ∑
Iyy θ̈ = My = Tc lc δ + Lα lα α − mpj lpj U̇o Γpj (3.78)
j=1
∑
1 Fz (lpj − Lpj ) ∑
Γ̈pj = [− + My − U̇o Γpj ] (3.79)
Lpj mo Iyy
Assume α ≈ θ
∑ mpj lpj U̇o

θ̈ = µc δ − µα θ − µpj Γpj , µpj = (3.80)
Iyy
1 Tc Lα ∑ mpj
Γ̈pj = [− δ + α− U̇o Γpj (3.81)
Lpj mo mo mo
(lpj − Lpj ) ∑
− (Tc lc δ − Lα lα θ) − mpi lpi U̇o Γpi ]
Iyy i
or on considering only one mode ( since orthogonality of different slosh modes)
θ̈ = µc δ − µp Γ, (Assume µα = 0) (3.82)
1 Tc 1 mp
Γ̈ = − [ − (lp − Lp )µc ]δ − [(1 + )U̇o + (lp − Lp )µp ]Γ (3.83)
Lp mo Lp mo
On taking Laplace transform and eliminating Γ, one gets:
θ µ(s2 + ωr2 ) 2 1 mp µp r2 Iyy

= 2 2 2
, ω p = [(1 + ) U̇o + (lp − Lp )µp ], ω 2
r = ωp
2
− (lp − L p − ), r = (3.84)
δ s (s + ωp ) Lp mo Lp lc mo
Actuator transfer function (TF) = kc /(s + kc )

Forward gain = KA , Feedback TF = (1 + kR s)
For ωp > ωr =⇒ ϵ = ψR − ψc
For ωp < ωr =⇒ ϵ = ψR − ψc + 180◦
Im S
Effect of sloshing
pendulum
E pole-zero pair
wb ο example wp > wr
90
wr
Angle of departure
Dipole ε is such that root locus
ψc YR 2 poles into LH of s-plane
Res .
-1 .. stable system.
-K c
Kp
Therefore, for ωp > ωr , as long as kc kR > 1, =⇒ stable system.

r2
i.e., ωp > ωr =⇒ µp
(l
Lp p
− Lp − lc
) >0
If the pendulum hinge point is behind C.G., lp < 0 and µp < 0. The above inequality is always
satisfied and system is stable. When hinge point is ahead of CG, lp > 0 and the system is
r2
Stable when : (lp − Lp ) > r2 /lc | U nstable when : (lp − Lp ) < , lc < 0
lc
The point on vehicle at a distance “r2 /lc ” forward of CG = “ Centre of percussion”
If the hinge point is ahead of CG, the pendulum mass must “straddle” the centre of percussion
for the motion to be stable. When multiple propellant (fuel/ oxidizer) tanks are present, no.
Im S
Dipole
Res
Figure 3.74: Examples of unstable (top) and stable (bottom) pole pairs.
of modes = (no. of pendulums) × no of tanks. Sloshing motion should be damped. Simplest

means of stabilization =⇒ passive i.e., use baffles, separators etc. Passive control are more
robust but result in large weight penalty and larger cost.
OR stabilize the unstable slosh motion with compensators. Introduce two real poles at −1/τ1
and −1/τ2 as shown in the figure, so that departure angle at slosh pole is into LHS-plane.
wr
wp
Dipole
-k - 1 -1 -1 Res
c τ1 τ2 kR
Figure 3.75: Slosh mode stabilization using active control.
But this results in deterioration of rigid body response (attitude response) since then the
system can become unstable for large gain. Therefore, use both Passive and Active control.
3.6.2 Engine Inertia Effects:

z}|{
Iyy θ̈ = Tc lc δ + Lα lα θ + mR lR lc δ̈ + IR δ̈; [IR δ̈] ≈ 0 because IR << MR lR lc (3.85)
θ mR lR (s2 + Tc /mR lR )
= µc (3.86)
δ Tc (s2 − µα )
Engine inertia effect manifests in terms of a pair of zero on the imaginary axis. Therefore,
Im S
Tc
m
R lR
R
e
-k c -1
µα µ
kR
- α
Figure 3.76: Engine Inertia Effect.

√
closed loop frequency (short period) is always less than Tc /mR lR , irrespective of the forward
gain.
√
Tc /mR lR = Tail Wags Dog (TWD) zero (frequency).
It is the frequency at which the transverse inertia forces resulting from gimballing of the rocket
engine cancel the thrust forces due to the rocket angular deflection.
This effect is not important for rigid vehicles, though the effect becomes significant
for large engines and having higher actuator bandwidth (larger IR δ̈). Then control frequency
bandwidth (dominant mode) is same order of magnitude as TWD zeros. However, TWD zero
have dominant/ important effect in Bending stability problem.
3.6.3 Vehicle Flexibility
Structural vibration influence the control problem primarily because the sensors pick up rigid
body motion plus local deformation due to elastic distortion. The launch vehicle being a
continuum exhibits theoretically large no. of deformation (bending) modes. Often, three to
five modes are enough in the analysis and design of control system. For ease of understanding,
consider only one mode of vibration.
θc δ=δc
+ θe θ(S)
- K (s+KI )
A S
θf θRG
CS
+ KR
-2 ξ(l t) +
2l
Xb la
ξ(l t)
Uo α lα
α
Uo+ u r θ
θo δ
2ξ (l t)
ξ 1(lT1 t) = T1
2l
lc
Zb wo
Iyy θ̈ = Tc lc δ + Lα lα θ + mR lR lc δ̈ + IR δ̈ (3.87)
−1
qï + 2ξi ωi q˙i + ωi2 qi = (mR lR δ̈ + Tc δ) (3.88)
Mi
M easurements : θP G = θ + σgi qi ; (3.89)
˙ + ω 2 θRG = kR ω 2 (θ̇ + σgi q̇i )
θ̈RG + 2ξR ωR θRG (3.90)
R R
or θRG = (q + σgi dotqi ); kR = rate gyro gain (3.91)

s + kI
F eedback signal θf = θRG + θpg ; where, θc − θf ; δc = kA ( )θe .
s
Simplified open loop transfer function (for small sensor delay)
θf KT (s + kI )(s + 1/kR )(s2 + Tc /(mR lR + IR ))

= (3.92)
θE s(s2 − µα )(s2 + 2ξR ωR s + ωR2 )(s3 + 2ξe ωe s2 + ωe2 s + k̄c ωc2 )
(s2 + 2ξk ωk s + ωk2 )
×
(s2 + 2ξi ωi s + ωi2 )
KT (s + kI )(s + 1/kR )(s2 + Tc /(mR lR + IR )) (s2 + 2ξk ωk s + ωk2 )
≈ × 2
s(s2 − µα )(s + k̄c ) (s + 2ξi ωi s + ωi2 )
mR lR Ai
KT = kA k̄c kR µc mR lR ωc2 ωR2 )Ai /Tc = kA kc kR µc (3.93)
Tc
Ai = 1 − (Tc σgi /µc Mi ) (3.94)
ωk2 = [ωi2 + (Tc σgi /µc Mi )µα )/(1 − (Tc σgi /µc µi )] (3.95)
ξk ωk = ξi ωi /Ai F ind ξk (3.96)
By neglecting the higher order dynamics ( non dominant modes), but including TWD zero,
obtain open loop TF as :
θF kT (s + kI )(s + 1/kR )(s2 + 2ξk ωk s + ωk2 )

= (3.97)
θE s(s2 − µα )(s + k̄c )(s2 + 2ξi ωi s + ωi2 )
Derive the open loop transfer function starting from the equation of motion (Home work)
• accounting for one bending mode
• additional pole-zero pair (TWD) near j ω-axis. . Note that:
• Analysis of one bending more at a time is permissible since they are orthogonal coupling
occurs through aerodynamic and engine inertia force are small
• In assessing stability, (Tc σgi /µc mi ) is a critical parameter. Neglect Higher order dynam-
ics + TWD + Non dominant pole pair.
On neglecting the effect of TWD zero, the open loop TF :
θF kA kR µc (s + 1/kR )(s2 + 2ξk ωk s + ωk2 )

=
θE (s + kc )(s2 − µα )(s2 + 2ξi ωi s + ωi2 )
Case I: Critical parameter (Tc σgi /µc Mi ) < 0 i.e., σg < 0 gyros are placed ahead of nodal point
Ai > 0 and ωk < ωi
As long as ψR > ψc , angle of departure at ωi is into left-half plane. The system is stable as
shown in the figure.
Im S
wi
nt
g poi wK
n
rati
ope
ψR µα
ψc
-k c -1 - µ Res
KR α
aeromargin related to ImGh

Case I ( µα )
µc
(-1,0)
Re GH
Iml
wk
w
kd
operating point
wi
wi
ψR µα
ψc
-k c -1 R
KR N µα ImGH
Case II
(-1,0) Re GH
Case II : 0 < (Tc σgi /µc mi ) < 1 → Ai > 0 and ωk > ωi

The angle of departure at ωk is into right half plane (when damping = 0). The system is
unstable for all loop given values
Note : When bending mode damping is present, ωid and ωkd are inside left half of s-plane for
low gain values, the system can be stable.

Gain stabilizing bending modes: useful for stabilizing higher modes and when aerodynamic
Im S
wi
ψc ψR + wk
-k c -w k -1 k
R - µ µc Res
c
ImGH
(-1,0)
gainstabilized
Re GH
effects are small (µα ≈ 0) gain stabilization =⇒ use notch filters how (HW) ?
Case III : (Tc σgi /µc Mi ) > 1 =⇒ Ai < 0, ωk2 < 0; σgi is negative.
Bending modes zeros lie on the real axis. Open loop gain is negative =⇒ 0-angle
criterion (not 180 criterion) should be used in Root Locus analysis.
Coupling between bending vibration modes and engine inertia effects:
Case I : σg is negative for increasing ωi , departure angle rotate clockwise because of influence
of higher order dynamics, when bending mode crosses TWD zero, 180◦ phase reversal takes
place.
Case II σg > 0 (positive) departure angle vary by 180◦ compared to case I.
Therefore, simple checking of sign of σgi is not enough.

=⇒ phase stabilize
=⇒ gain stabilize
=⇒ use both.
I mS
TWD
Actuator poles
-1/k R µα Re
-k c µα
Gyro
Aerodynamic
I mS
( ξc, ωc)
µα
-1/k R µα Res
-k c
To summarize, the governing equation for an aerospace vehicle can be written in a

standard state space representation as:
ẋ = A x + B u, where, x ϵ ℜn , u ϵ ℜm (3.98)
and the corresponding measurement equations are
y = C x, where, y ϵ ℜ r , (m, r) ≤ n. (3.99)
The states x include both the state variables for airframe dynamics and actuator.
Before dwelling on the control system design techniques [215], it is necessary to investigate
the intrinsic properties such as stability, controllability and observability. At the same time
stability and performance robustness also play important role in the flight control system
design. These aspects are discussed in the next chapter for a general linear time invariant
system represented as in the equations (3.98) and (3.99).
Chapter 4
Pole placement Techniques
The aerospace vehicle autopilots are designed to improve the stability and performance. The
dynamics of the aerospace vehicle is coupled, nonlinear and time varying. For the aerospace
vehicles, the allowable angle of attack (in pitch and yaw) under controlled condition has to be
restricted. Since most of the currently available missiles have cruciform wings, it is possible
to design decoupled controller separately for roll, pitch and yaw dynamics for a short period
of time, while for the aircraft having only one plane of symmetry, separate longitudinal and
lateral controller are needed. The linear time invariant model developed for the design purpose
can deliver satisfactory stability and performance.
Some of the current day aerospace vehicles have multiple actuation or control inputs
and multiple outputs or sensors. The typical inputs could be from thrust vector control,
aerodynamic control from either or combination of tail, wing or canard. The sensors such as
position and rate gyros, accelerometers, angle of attack and mach number sensors are used
in various combinations in aerospace vehicles. In order to effectively make use the multiple
degrees of freedom available for the flight control system/ autopilot design, it is desirable to use
modern control synthesis procedures. This chapter describes procedure to design controller
using pole placement / pole assignment technique.
Among various controller design techniques, the pole assignment method has a
fairly long history. Generally, the class of controllers can be put under either regulator or
servo / tracking systems. The regulator is primarily concerned with the closed loop poles
or eigenvalues alone. The problem of pole assignment basically deals with the determination
of feedback action so as to achieve a set of eigenvalues that will meet the transient response
265
266 CHAPTER 4. POLE PLACEMENT TECHNIQUES
specifications. However, it must be remembered that the locating the eigenvalues is rarely
adequate in achieving the control objectives in aerospace systems, since the eigenvectors, the
zeros, the cost function, and the output correlation are all ignored in this exercise.
The intrinsic properties of the system, viz., controllability, stabilizability, observ-
ability, disturbability play very important role in pole placement techniques. Before one
proceeds with the design, care should be taken to check for controllability and observability.
Once the desired pole locations are specified, the problem of pole assignment by state variable
feedback for single input deterministic system is discussed in many text books.
4.1 Controller design by Pole Assignment Techniques

The major difficulty in the design of controllers by most of existing pole placement techniques
relates to determination of desired pole location for higher order systems. In principle, the
closed loop poles are selected so that the design specifications are met while control effort
is kept to a minimum that is enough to meet the specifications. A proper root selection is
expected to keep the action of actuators to a minimum thus reducing the cost of controller
implementation. Based on the transient and steady state accuracy specifications, method of
obtaining a second order or third order Transfer function (TF ) is well known. Simple to use
graphs can be used to arrive at a suitable TF of order two or three [15].
The control system design strongly depends on control system specifications which
in turn can be used to get ⇒ desired system representation. In order to simplify the design
yd (s)
procedure for SISO system, desired system transfer function (TF) Gd (s) = r(s)
⇒ should be as simple as possible; i.e. TF Should be least complex. For example, consider a
single degree of freedom physical system having second order TF of the form:
ωn2
Gd (s) =
s2 + 2ζωn s + ωn2
System response yd (t) depends on ζ, ωn , t and u(t). Let s̃ = s/ωn , t̃ = ωn t
yd (s̃) 1
= 2
r(s̃) s̃ + 2ζ s̃ + 1
Step response to the input r(s) = (1/s) is given by
√
y(t) = 1 − e−ζωn t sin(ωd t + ϕ); ωd = ωn 1 − ζ 2
4.1. CONTROLLER DESIGN BY POLE ASSIGNMENT TECHNIQUES 267
1 √
−ζ t̃
y(t̃) = 1 − √ e sin( 1 − ζ 2 t̃ + ψ)
1 − ζ2
√
ψ = tan−1 ( 1 − ζ 2 /ζ)
td = delay time = time to 50% of the steady state

ts = settling time, tr = rise time
P O =Percentage Over Shoot
Consider typical Time Domain Specifications which are related to ζ, ω of the desired TF
√
P O = 100e−(ζπ/ 1−ζ )
2
3 7.04ζ 2 + 0.2
5% t̃s ≈ , t̃r ≈
ζ 2ζ
√
t̃d ≈ 1 + 0.7ζ; ψ = tan−1 ( 1 − ζ 2 /ζ)
Step1: Obtain Desired TF

(a) transfer the transient response specifications from a time description to an s-plane pole-
location requirement
1
y(s) = s
[( s )2
ωn ωn
+ 2ζ( ωsn ) + 1]
Response of a second order system with no finite zeros, and unit d.c. gain with varying and
it can be seen that the major influence on the percent overshoot in the damping ratio and a
simplified relationship. % overshoot ≈ (1 − ξ
0.6
)100 (f or ξ < 0.65) for a second order system
with no finite zeros. Thus given a requirement on percent overshoot, ξ ≥ 0.6(1 − %overshoot
100
).
Another feature of interest is the rise time {normalized} of the response towards its final value.
By observing the step response, with ωn t as, the time scale, the rise time is shorter as ωn is
increased. While, there is some dependence of rise time on ξ one may take the curve ξ = 0.5
to about the centre of the distribution and thus approximate the rise time tr ≈ 2.5/ωn . where
tr is the time necessary for the response to rise from 0.1 to 0.9. A requirement on tr thus
becomes a requirement that ωn satisfies ωn ≥ tr .
Example :
To realize ” PO 25%, select (using the plots of (PO, t̃d , t̃r , t̃s ) Vs damping ratio, see [15])
⇒ ζ = 0.4
⇒ t˜s = 7.4
t˜s = ωn ts
t̃s 7.4
ts = 5sec ⇒ ωn = = = 1.48
ts 5
y(s) (1.48)2 2.2
= 2 =
r(s) s + 2 × 0.4 × 1.48 + (1.48)2 s2 + 1.18s + 2.2
⇒ Transfer function based on overshoot and settling time only.
Error Constants
Kp = ∞ for the above TF (note: dc gain of the desired TF Gd (s)= 1)
1 ∑n
1 ∑n
1
= −
Kv j=1 λj j=1 zj
√
s̃ = −ζ ± j 1 − ζ 2
K̃v = Kv /ωn
1 1 1
= √ + √ = 2ζ
K̃v ζ + j 1 − ζ2 ζ − j 1 − ζ2
1
K̃v =
2ζ
K̃v should be large ⇒ ζ has to be small ⇒ large overshoot → not acceptable.
For example, ζ = 0.4 ⇒ K̃v = 1/0.8 =1.25
Normally, K̃v ∼ 50 to 100 is needed in many aerospace vehicles during the maneuvers. Intro-
duce a zero to enhance the velocity error constant
2
yd (s) ( ωzn (s + z)) z
= 2 2
; Def ine z̃ =
r(s) s + 2ζωn s + ωn ωn
yd (s̃) (s̃ + z̃) 1 (s̃ + z̃)
= 2 = 2
r(s̃) s̃ + 2ζ s̃ + 1 z̃ (s̃ + 2ζ s̃ + 1)
√ √
1 + z̃ z̃ − 2ζ −ζ t̃ √
y(t̃) = 1 + { √ }e sin( 1 − ζ 2 t̃ + ψ)
z̃ 1 − z̃ 2
√
1 − ζ2
where, ψ = π − tan−1 ( )
z̃ − ζ
PO is function of ζ and z̃ and
z̃
K̃v = = velocity error constant (normalized)
2ζ z̃ − 1
Example :
P O = 15%
z̃ = 0.6, ζ = 0.9, K̃v → ∞
However, f or z̃ = 2.0, ζ = 0.55, K̃v ≈ 2.0

Pole Placement Technique for SISO sytems: The problem of control law generation is to
find the gain vector K in u = Kx, K = [k1 , k2 , ..., kn ], for LTIV systems ẋ = Ax+Bx, y = Cx;
(Estimate x̂ from y if state variables x are not available for feedback and then, use u = - k x̂)
such that the eigenvalues of (A − BK) are the same as desired.
Method 1: Indirect Design
In General, convert the LTIV system to control/ phase variable canonical form and use the
control law of the form:
 
[ ]
x1 
 
 
u=− k1 . . k n  . 
 
 
xn
   
 0 1 0 . . 0   0 
   
   
 0 0 1 . . 0   0 
A =  ;B =  
   
 . . . . . .   . 
   
   
−a1 −a2 −a3 . . −an 1
which results in a closed loop system
 
 0 1 0 . . 0 
 
 
 0 0 1 . . 0 
A − BK =  
 
 0 0 0 . . 1 
 
 
−(a1 + k1 ) −(a2 + k2 ) −(a3 + k3 ) . . −(an + kn )
 
 0 1 0 . . 0 
 
 
 0 0 1 . . 0 
= [Ad ] =  ;
 
 . . . . . . 
 
 
−d1 −d2 −d3 . . −dn
where the desired characteristic polynomial equation sn + dn sn−1 + · · · + d1 = 0.
On comparing the coef f icients of like powers of characteristic polynomials
(a1 + k1 ) = d1 , (a2 + k2 ) = d2 , . . . (an + kn ) = dn
or k1 = (d1 − a1 ), k2 = (d2 − a2 ) . . . kn = (dn − an )
The desired characteristic equation is obtained based on system specifications, stability re-
quirements, noise & disturbance rejection and control effort. For example for n = 2, d1 =
2ζωn , d2 = ωn2 . For single input systems having n - states, there are n - gain parameters that
assign n - closed loop poles. Hence, all the n - poles can be arbitrarily placed in s - domain
by sate variable feedback. For output feedback, number of design freedom (gain parameters)
are limited, hence not all poles can be arbitrarily located. This method is easy to use. But
for a general form of matrix ’A’, it is difficult or computationally expensive or numerically
inefficient to obtain control canonical forms.
Alternate Method 2 Let u = - K x
Closed loop system matrix Ac = (A − BK)
Determine the characteristic equation |λI − (A − BK)| = 0 , with K = (k1 , . . . , kn ) as control
gain parameters.
Equate the coefficient of the characteristic equation, with the characteristic equation of the
desired TF ; one has n-equations for n unknowns (k1 , . . . , kn ). These are linear equation in K.
(Provided the system is completely controllable). If these exist a pole-zero cancellation, then,
one will have less number of equations for large variables and no solution could be found out
readily.
For uncontrollable systems :
Check (1) whether the uncontrolled modes are stable?
if yes, design controller (/regulator) for the controllable modes. Then, one may get near
satisfactory performance; if no, then redesign the plant, as no regulator can be designed.
Method 3: Direct Design
The famous Ackermann’s formula [10, 16, 49], though in principle relies on the
canonical form indirectly, gives a convenient to use equation:
K = [0 ... 0 1] Q−1
c Gd (A), or K = [0, ....0, 1][BAB...A
n−1
B]−1 Gd (A)
where, Qc is a controllability matrix, and Gd (A) is obtained by substituting A for ’s’ in the
desired characteristic polynomial Gd (s) or αc (s) to form :
Gd (A) = An + adn An−1 + · · · + ad2 A + ad1 I, or αc (A) = An − α1 An−1 − · · · , −αn I
where, the adi or αi , i = 1, . . . , n are the polynomial coefficients of the characteristic equa-
tion. It should be noted that the controllability matrix Qc should not be inverted if it is poorly
conditioned. The solution can be obtained by a stable method such as Gaussian elimination
with pivoting or singular value decomposition with a suitable normalization procedure. Algo-
rithms have been developed for pole assignment in real Schur form for its numerical stability
and structural properties. They use orthogonal similarity transformation U for any matrix
A such that U T AU is quasi upper triangular having only 1 × 1 or 2 × 2 blocks on the diag-
onal corresponding to real or to complex conjugate eigenvalues, respectively. The similarity
transformation preserves all the intrinsic properties of the system.
Ackermann’s formula makes use of the above properties; i.e., by using the control-
lability matrix, it is possible to transform a general A matrix into controllable form.
Existence of K: Test for complete controllability.
Exercise : Write a simple program to evaluate Ackermann’s formula.
Example : Missile Autopilot
1
δ̇ = (u − δ)
τ
az
α̇ = q +
V
e = azc − az ⇒ ė = ȧzc − ȧz
Assume command acceleration = const ⇒ ȧzc = 0
ė = −ȧz = −Zα α̇ − Zδ δ̇
1 Zδ
= −Zα q + (azc − e) + (δ − u)
V τ
1 1
α = (az − Zδ δ) = (azc − e − Zδ δ)
Zα Zα
Mα
q̇ = (azc − e − Zδ δ) + Mq q + Mδ δ
Zα
 
 e 
 
x =  
 q  ; ẋ = Ax + Bu + Br azc
 
 
δ
     
 Zα /V −Zα Zδ /τ   −Zδ /τ   −V
Zα

     
A = 
 −Mα /Zα Mq M̄δ
 
=B 0
 
 Br =  Mα /Zα


     
     
0 .0 − τ1 1
τ
0
where M̄δ = Mδ − Mα
Z
Zα δ
V = 1253ft/sec (≈ Mach 1.1) ; Zα = −4170f t/sec2 ( per radian of the angle of attack)
Zδ = −1115f t/sec2 ; Mα = −248rad/sec2 ; Mq ≈ 0 ; Mδ = −662 rad/sec2 ; τ = 0.01 sec
The characteristic equation of the plant:

1 Zα
(s + )(s2 + s − Mα ) = 0
τ V
(s + 100)(s2 + 3.33s + 248) = 0
or, s3 + 103.33s2 + 581s + 24800 = 0
with roots at s = - 1000 (due to actuator)

s = -1.67± 15.65 j (due to airframe; i.e., ωn ≈15.65 rad/sec ≈ 2.5 Hz)
Design objective is to increase the damping ratio to 0.707 and frequency to 30 read/sec.
Dominant closed loop characteristic polynomial
√
s2 + 2 ∗ 30 2s + (30)2
Therefore, the closed loop poles should satisfy the characteristic equation (Please check ??)
s3 + 144.8528s2 + 5991.1688s + 54000 = 0
Using the Ackerman’s formula:
G = [−0.6366 × 10−4 , −0.3929 × 10−1 ≈ 0]
Gain margin = 22db at ω = 187 rad/sec.

For a typical dual input (wing-tail control) missile,
x = [θ, q, α, δw δT ]T , u = [δwa , δT a ]T where,

   
 0 1 0 0 0 θ   0 0 
   
   


0 −3.9 −793.4 −17.2 115.0 q    0

0 

   
A =  0.055 0.997 −1.82 
0.2239 0.0995 α  B =  
  0 0 
   
   
 0 0 0 −80 0 δw   80 0 
   
   
0 0 0 0 −100 δT 0 100
   
 0 −0.24 −1236 189 84   0 0 
C =  ,D =  
0 10 0 0 0 0
eig(A) = −0.5487
− 2.58571 ± 28.08i(ζ = 0.09168, ω = 28.201)
− 80
− 100
4.2. OBSERVER DESIGN 273
Use ”Place” : MATLAB routine for the ”Ackernmann’s formula”

Closed loop poles with δT feedback alone with poles at {−0.6, −25 ± 20j, −80, −90} results
in Gain K = [ 1.2553, 0.3937, 1.1683, 0.0 , 0.3488 ] where as the closed loop poles with both
δT and δw feedback with
{ }
poles at −0.6, −25 ± 30i, −70 −90
 
 0.2278 0.3593 −0.3131 0.027 0.6443 
Kwt =  
0.06027 0.3492 2.2036 0.122 0.2272
{ }
P oles at −0.6, −35 ± 30i, −70 −90
 
 0.0561 0.2200 −1.606 −0.0313 0.4487 
Kwt =  
0.9038 0.5660 5.2689 0.1611 0.4739
Closed loop eigen values at (-0.55 , −25± 30i, -70, -90) are due to gain
 
 0.2203 0.2112 −1.8033 0.0004 0.4737 
K= 
0.4722 0.3628 2.2134 0.1240 0.2479
Low gain since the closed loop drift pole is kept close to the open loop drift pole.
The implementation of the control law requires the knowledge of state variables.
The reconstruction of state by the estimators will also provide smoothing of measurements
which are often contaminated by random noise. Then, eventually the estimated state replaces
the true state in the control law implementation. The design procedure for the observer is
similar to the control law evaluation and hence only a brief discussion is given.
4.2 Observer Design

Most cases, all the states are not available for feedback. State variables are necessarily con-
structed from the measurements. Observers are used as estimators for deterministic systems.
The observer parameters therefore do not depend on disturbance and measurement noise
statistics. In general, an observer is a dynamical system with input of (u, y) and output of
say, x̂, which asymptotically estimates the state x, i.e., {x̂ − x} → 0 as t → ∞ for all initial
states x0 and for every input u. For the LTIV system,
ẋ = Ax + Bu, y = Cx
the open loop only mathematical model based computation of state x̂(t) by solving
x̂˙ = Ax̂ + Bx, y = C x̂
will result in the error estimate x̃(t) = x(t) − x̂ having the dynamics
x̃˙ = Ax̃; with x̃(0) ̸= 0 since initial conditions x̂(0) are unknown
If system response for the given ’A’ is stable and satisfactory, then x̃(t) will tend towards
zero. If ’A’ is not satisfactory ( other wise controller need not be designed ), the error will be
continued to be present (or may increase beyond prescribed limit, i.e., where A has unstable
poles). Estimator tries to overcome this difficulty by using the regulator principle i.e., error
x̃(t) will be driven to zero. Controller and estimator (observer) are duals for the system. In
the estimator design, check a priori for the complete observability of the system. Then,
˙
x̂(t) = Ax̂(t) + Bu(t) + L[y(t) − C x̂(t)]
= [A − LC]x̂(t) + Bu(t) + Ly(t)
By suitably choosing L, the pole location of (A-LC) can be arbitrarily placed such that (or,
A + L C, L ← -L)) is stable.
˙
The error equation :x̃(t) = (A − LC)x̃(t) i.e., using L, the error can be reduced at any desired
rate. Usually, they are reduced at rate approximately (5-10) times faster compared to the
system’s time constant (ie for example rise time specification for the regulator)
These estimators are known as prediction estimators (Based on the measurement, prediction
error is computed - dynamical delay is not considered. Selection of L:
Method 1: Expand the determinant |sI − A + LC| and match the coefficient in the like
powers of s-characteristic polynomial (of desired TF ) α0 (s) = (s − p1 )...(s − pn ), where pi are
obtained from the desired pole locations
Method 2: Use Observer Canonical form
Method 3: Use Ackermannn’s formula
 −1  
 C   0 
   
   
 CA   0 
   
   
L − αo (A) 
 CA2 

 . 
 
   
   
 .   0 
   
   
CAn−1 1
The conventional pole assignment methods for SISO systems proceed on a paradigm based
on central limit theorem that the control law can be derived independently from estimator/
observer. The corresponding design procedure is undertaken in two independent steps. The
first step assumes that all state variables are available for feedback. Since, it may not be prac-
tical/ desirable to measure all the state variables, the remaining step proceeds on estimator/
observer design which reconstructs the state variables from the measurements. In order that
this two step procedure is successful, the set of eigenvalues of closed loop system has to be
well separated from that of observer, and the zones containing observer poles are further away
from the origin than the closed loop poles in s - domain. This is necessary, since then, the
estimation error decays at a faster rate dependent on the eigenvalues of the observer. Fast
roots in an observer do not burden the actuators, but they create larger sensitivity between
sensor errors and estimation errors. On the other hand, the optimal estimator such as Kalman
filtering is designed with roots varying in accordance with the ratio of plant model errors to
sensor errors. The larger the ratio between the errors, the larger will be the estimator gains
in order to use the measurements more effectively. While, the lower gains make the estimator
to rely more on the plant model than the sensor output. In any case, the estimator roots
should be faster than the control roots, so that the total system response is dominated by the
control roots and estimation error transients die down quickly. The final controller consists
of combination of control law and estimator. The evaluation of controller and observer for a
deterministic system by these methods have become a matter of routine affair.
Pole Placement for single output systems:
The design of observer for a single output system can be considered as a dual problem of
design of feedback controller for single input system. In particular, we have following counter
parts. Bass - Gura Formula for observer Design:
L = [(N W )T ]−1 (â − a)

 
 â1 
â = coefficients of the desired polynomial represented in a vector form =  
ân
sn + aˆ1 sn−1 + . . . + ân = 0  
 â1 
a = coefficients of the plant characteristic polynomial represented in a vector form =  
ân
sn + a1 ss−1 + ...... + an = 0
Full State Feedback Observer

ẋ = Ax + Bu ẋ = Ax + Bu
u=-Kx+r y = Cx
closed loop system closed loop observer
ẋ = (A − BK)x + Br x̂˙ = (A − LC)x̂ + Bu + Ly(t)
A Replace A byAT in controller gain formula
B Replace B by C in controller gain formula
K Replace K by L in controller gain formula
N = [C T AT C T ...(AT )n−1 C T ] = observability matrix

 
 1 a1 . . an−1 
 

 0 1 . . an−2 

W =



 . . . . . 
 
 
0 0 . . 1
Example : Motor Driven Inversted Pendulum with slowly varying Disturbance:
θ̂ = ω
ω̂ = Ω2 θ − αω + u + d
d˙ = 0
α = F riction coef f ient

   
 0 1 | 0   0 
     
 2
 Ω −α | 1 


 b 
 0
A =  B =  ; E = 


D = 0
   
 − − − −   −−−  1
   
   
0 0 0 0
The error dynamics ė = Ae + Bu + Exo

 
[ ]
θ  [ ]
 
 
y = 1 0 0  c = 1 0
 ω  0
 
d
 ⇒s(s2 +αs−Ω2 )  
 s −1 0   1 0 Ω 
2
   
   
SI − A =  −Ω2 s+α −1  ; N =  0 1 −α  ;
   
   
0 0 s 0 0 1
⇒a1 =α,a2 =−Ω2 ,a3 =0
     
 1 α −Ω  2
 1 α 0   1 −α 0 
     
    −1  
W =  0 1 α N W =  0 1 
0  ; (N W ) =  0 1 0 
    
     
0 01 0 0 1 0 0 1
      
 1 0 0   â1 − α   â1 − α   ℓ1 
      
      
 −α
Gain L =  1 0   2  =  â2 + Ω2 − α(aˆ1 − α)  =  ℓ2 
  â2 + Ω     
      
0 0 1 â3 aˆ3 ℓ3
Note: Regulator + Estimator ⇒ controller

Difficulties:
1. order of controller high
2. Many times result in unstable control estimator combination though Closed Loop Systems
are stable.
4.2.1 Compensator Design Using Full-Order Observers
The separation principle allows independent design of complete state variable feedback (gain)
and the observer (gain) by pole placement or any other technique (restricted conditions for
the independent design will be discussed during the study of LQG/TLTR theory).
Consider the dynamic system:
ẋ = Ax + Bu, y = Cx (4.1)
Complete (full state feedback control law
u = −Kx (4.2)
and the observer dynamics
x̂˙ = Ax̂ + Bu + K(y − C x̂) (4.3)
Since x is not available for feedback, use instead x̂ in the place of x to get
u = −K x̂; (4.4)
which closes the loop
r = −y − C x̂ = C(x − x̂) = C x̃ (4.5)
4.2.2 Combined Regulator - Observer System.
The dynamic compensator is the combinations eqns(3.3) and (3.4). The constant gain output
feedback is a special case of Ot h order dynamic compensator (Note for example PI and PD
controller are 1st order compensators or low order equivalent observers).
The dynamic behaviour of 2 nth order system is dependent on the closed loop system
ẋ = Ax − BK x̂ (4.6)
and x̂˙ = Ax̂ − BK x̂ + L(Cx − C x̂) (4.7)

    

 ẋ  

 A −BK 
 x 

or   =  
 
(4.8)
 x̂˙  LC A − LC − BK  x̂ 
Observer based controller:
x̂˙ = Ax̂ − BK x̂ + L(Cx − C x̂) (4.9)
u = K x̂ or u = F (s) y (4.10)
 
A − BK + LC −L 
where controller realization F (s) = 
  (4.11)
K 0
The observer error x̃ = x − x̂
and error dynamics x̃˙ = (A − LC)x̃.
ẋ = Ax − BK x̂ = Ax − BK(x − x̃) = (A − Bk)x + BK x̃ (4.12)

    

 ẋ 

 A − BK −BK   x 

=   (4.13)

 x̃˙ 
  
0 A − LC  x̃ 
The plant or state dynamics ”x” and error dynamics state ”e” are decoupled. If e → 0 much
faster than state x, then the closed loop system is likely to behave like a idealized regulator.
The transfer function of the compensator is obtained by the use of eqns (3.3, 3.4) as:
x̂˙ = (A − BK − LC)x̂ + Ly (4.14)
x̂(s) = (sI − A + BK + LC)−1 Ly(s) (4.15)
u(s) = −K x̂(s) = −K(sI − A + BK + LC)−1 Ly(s) = −F y(s) (4.16)
theref ore F (s) = K(sI − A + BK + LC)−1 L (4.17)

Plant y
x= Ax + Bu
u y=cx
-K
+ r
>
x
c
Model
.
>
>
x = A x + Bu
>
+ L (y-C x )
Figure 4.1: Controller + Observer System (dynamic compensator inside dotted block).
The design of regulator will ensure that (A-Bk) is Hurwitz (all poles on LHS-plane) and that
of observer ensures that (A-LC) is also Hurwitz. However, the roots(poles) of the dynamic
compensator depends on the eigenvalues of (A - BK - LC) or roots of dt |(sI − A + BK + LC)|
which do not occur at poles of plant, or regulator or observer. Therefore, there is a possibility
that one or more poles of the compensator may be unstable (RHP - poles), i.e.,
⇒ compensator could turn out to be unstable.
⇒ It is possible that unstable compensator may stabilize the unstable plant, or stable plant
with unstable zero, and in poorly judged system, unstable controller for a stable proper plant
(plant with minimized phase zero).
Note :
Unstable plant ⇒ can not be avoided unstable compensator preferably avoided any satura-
tion/large signal will produce large transient/instability of the system open loop property:
plant with unstable controller or compensator is relatively more unstable than the plant alone
⇒ which is highly undesirable.
⇒ unstable compensator ⇒ closed loop system is conditionally stable. Ref: example 8A in
Friedland’s book [1], pp.295 - 298 for further details.
Example:
   
1 2   1 
Let A = 
 , B =   , and C = [1 0].
1 0 0
Design u = K x such that the closed loop poles are at {-2, -3}. Use Matlab command: K
= -place(A, B, [-2, -3]). Then K = [-6, -8]. Suppose observer poles are placed at {-10, -10}.
Then, on using Ackermann’s formula, L = [-21, -51]’ (i.e., on using the Matlab command: L
= - acker(A’, C’, [-10, -10])’). The closed loop controller is derived as (Eqn. 4.17):
−534 (s + 0.6966)
F (s) =
(s + 34.6564) (s − 8.6564)
which implies that the stabilizing controller itself is unstable. This may not be desirable in
practice.
Reduced order Observer
The observers are designed to estimate entire state from the measurement. Given the mea-
surement of some state directly, there is no need to estimate them. Hence, it is possible to
use reduced order observer. In cases where the observations are such that
 
 x1 
y = C1 x1 = (C1 | 0)  
x2
where C1 is non singular the x1 = {C1−1 y} , and again, there is no need to estimate states
x1 . The separation principle which is applicable for full order observer is also applicable to
reduced observer (same limitations hold good). The design of reduced order observer portion
the system into
      

 ẋ1 

 A11 A12  
 x1 
 
 B1 

 
=  
 
+ u
 ẋ  A21 A22  x2   B 
2 2
 
[ ]
 x1 

y = I 0 
 x 
2
C1 = I is used in this case with no loss of generality, theref ore

{ }
ẋ2 = A22 x2 + A21 x1 + B2 u
(x̂˙ 1 (t) − A11 x1 (t) − B1 u(t)) = A12 x2 (t)
| {z }
known /equivalent measurement
Therefore regular & replaced reduced order observers have following counter parts
x ← x2 | Bu(t) ← A21 x1 (t) + B2 u(t)
A ← A22 | y(t) ← x1˙(t) − A11 x( t) − B1 u(t)
C ← A12
4.2.3 Multi-input state variable Feedback controller (dyadic form)
For a typical wing-tail controlled missile,

[ ]T
x = θ1 q1 α1 δw δ˙w δT δ˙T ; u = δW a , δ T a
   
 0 1 0 0 0 0 0   0 0 
   
   
 0 −0.39 −793.4 −17.2 0 115.0 0   0 0 
   
   
 0.055 0.997 −1.80 0.2239 0 0.09953 0   0 0 
   
   
   
A = 
 0 0 0 0 1 0 0 B =  0
  0 

   
   
 0 0 0 900 −42 0 0   900 0 
   
   
 0 0 0 0 0 0 1   0 0 
   
   
0 1 0 0 0 −14400 −168 0 14400
   
 0 −0.24 −1236.1 189.0 0 84.0 0   0 0 
C =  ; D =  
0 1 0 0 0 0 0 0 0
A − Bk2 C(2) = [ ] = Ac
eig(Ac ) = −0.0052; −4.29 ± 4.9j; −42.18 ± 68.38j; −21 ± 21.42j
eig(A) = −0.0056; −1.1 ± 2.82j; −21 ± 21.42j; −84 ± 85.7j
Pole Placement : [A, B(:, 1), C(1, :)]

K = Place (A,B,P); P = vector of eigen values of closed loop system where only the wing
actuator is considered :
[ ]
P = −0.006 −15 ± 15j 80 ± 80j
 
 −0.0678 −9.48 610 17.71 0.162 
K =  
(large ↑) (not ↑ acceptable) (large ↑)
Extension of SISO controller design to MIMO systems is not very simple. In a pole placement
procedure, a possible method of using large design freedom is to assign dyadic structure to
the feedback matrix.
ẋ = Ax + Bu; u = Kx; K = f dT ; f & d − vectors
ẋ = Ax + Bf dT x
= Ax + B1 dx,
= Ax + B1 u, B1 = Bf, u1 = dT x
Here, the feedback gain is first expressed in a dyadic form: K = q f, such that the equivalent
single input v is defined by the relation B u = (B q) v. Then, the gain vector f is found as
usual, but there exists many possibilities in picking q. Use pole placement by Ackermann’s
formula to find the gain vector ’d’
dT = [0. . . . 01]Q−1
c αc (A).
while, determine the vector ’f’ with the help of some constraints
→ Discussion on selection of f will be differed.
⇒ For example : select f such that
a condition number of the controllability matrix (A, B) is minimized
b control effort for the step response is minimized
c steady state error is minimized &so on → many possibilities do exist.

Chapter 5
References
1. B. Friedland : Control System Design, McGraw Hill, NY, 1987.
2. M. J. Sidi : Spacecraft Dynamics and Control, Cambridge University Press, 1997.
3. Bong Wie : Space Vehicle Dynamics and Control, AIAA Education Series, 1998
4. P. Garnell : Guided Weapon Control Systems, Pergamon Press, 1980.
5. A. L. Greensite : Analysis and Design of Space Vehicle Flight Control System, Spartan
Books, Vol. II, 1970.
6. A. E. Bryson : Control of Spacecraft and Aircraft, Princeton Univ. Press, NJ, 1994.
7. J. H. Blakelock : Automatic Control of Aircraft and Missiles, John Wiley, NY, 1991.
8. M. H. Kaplan : Modern Spacecraft Dynamics and Control, John Wiley and Sons, NY,
1976.
9. B. Wie and K. W. Byan : New Generalized Structural Filtering Concept for Active
Vibration Control Synthesis, J. of Guidance and Control, Vol. 12, No. 2, 1989, pp. 147
- 154.
10. J. J. D’Azzo, and C. H. Houpis : Linear Control System Analysis and Design, Conven-
tional and Modern, McGraw-Hill, NY, 1995, 4th Edition.
11. K. J. Åström and T. Hägglund: PID Controllers: Theory, Design and Tuning, ISA,
1995, 2nd Edition.
283
284 CHAPTER 5. REFERENCES
12. P. Dorato (Ed.): Robust Control, IEEE Press, NY, 1987.
13. P. Dorato : A Historical Review of Robust Control, IEEE Control System Magazine,
Vol. 7, No. 2, 1987, pp. 44-47.
14. R. Oldenburger (Ed.): Frequency Response, Macmillan Co.,NY,1956
15. J. L. Melsa and D. C. Schultz : Linear Control Systems, McGraw Hill, NY, 1967
16. G. F. Franklin, J. D. Powell and M. L. Workman : Digital Control of Dynamical Systems,

2nd Ed., Addison Wesley, 1990.
17. Special Issue on Aircraft Handling Qualities, J. of Guidance and Control, Vol. 9, 1986.
18. Z. Bien and J. Lee: A Note on a Computer-Aided Root-locus Method, IEEE Trans on
Automatic Control, Vol. AC-30, No.6, 1985, pp.574-577
19. F. N. Bailey, D. Panzer, and G. Gu : Two Algorithms for Frequency Domain Design of
Robust Control Systems, Int. J. Control, Vol. 48, No. 5, 1988, pp. 1787 - 1806.
20. J. C. Doyle, B. A. Francis, and A. R. Tannenbaum : Feedback Control Theory, Macmil-

lan Publishing Co., NY, 1992.
21. R. E. Skelton: Dynamic System Control, John Wiley & Sons, NY, 1988.
22. J. M. Maciejowski: Multivariable Feedback Design, Addison-Wesley, Wokingham, 1989.
23. A. E. Bryson and Y. C. Ho: Applied Optimal Control, Hemisphere, Washington, 1975.
24. B. D. O. Anderson and B. C. Moore: Linear Optimal Control, Prentice Hall, 1990.
25. F. L. Lewis: Optimal Control, Wiley, NY, 1986.
26. C. F. Lin: Advanced Control System Design, Prentice Hall, NJ, 1994
27. B. C. Kuo : Digital Control Systems, Holt, Rinehart and Winston, Inc., Tokyo, 1980.
28. H. Kwakernaak and R. Sivan: Linear Optimal Control Systems, Wiley, NY, 1972.
29. M. V. Cook: Flight Dynamics Principles, Elseier Ltd., Oxford, 2007, Second Edition.
285
30. F. L. Lewis, L. Xie and D. Popa: Optimal and Robust Estimation, CRC Press, Boca
Raton, 2008.
31. R. G. Brown and P. Y. C. Hwang: Introduction to Random Signals and Applied Kalman
Filtering, John Wiley & Sons, NY, 3rd Ed., 1997.
32. K. Åström and T. Hägglund: PID Controllers: Theory , Design and Tuning, Instrument
Society of America, NC, 1995, Second edition.
33. M. R. Ananthasayanam: Fascinating Perspective of State and Parameter Estimation

Techniques, AIAA, 2000.
34. M. Green and D. J. N. Limebeer : Linear Robust Control, Prentice Hall, Ennglewood
Cliffs, 1995.
35. B. L. Stevens and F. L. Lewis: Aircraft Control and Simulation, Wiley India Pvt. Ltd.,
New Delhi, 2003, Second Edition.
36. B. Restic: Beyond the Kalman Filter, Artech House, Boston, 2004 .
37. J. C. Willems: Dissipative Dynamic Systems, Part 1: General Theory; Part II: Linear
Systems with Quadratic Supply Rate, Archive for Rational Mechanics and Analysis,
Vol.45, 1972, pp. 321-351 & pp. 352-393
38. D.J. Hill and P.J. Moylan: Dissipative Dynamical Systems: Basic Input-Output and
State Properties. J. Franklin Institute, Vol. 309, No. 5, 1980, pp. 327 - 357.
39. R. H. Barnard and D. R. Philpott: Aircraft Flight, Pearson Education, Ltd., Delhi, 3rd
Edition, 2004
40. C. F. Lin: Robust Control Design, An Optimal Control Approach, John Wiley & Sons,
Chichester, West Sussex, 2007.
41. F. L. Markley: Unit Quaternion from Rotation Matrix, Journal of Guidance, Control,
and Dynamics, Vol. 31, No. 2, 2008, pp. 440 - 442.
42. M. D. Shuster, and G. A. Natanson: Quaternion Computation from a Geometric Point

of View, Journal of the Astronautical Sciences, Vol. 41, No. 4, 1993, pp. 545 - 556.
286 CHAPTER 5. REFERENCES
43. F. L. Markley: New Quaternion Attitude Estimation Method, Journal of Guidance,

Control and Dynamics, Vol. 17, No. 2, 1994, pp. 407-409
44. W. S. Levine (Ed.): The Control Hand Book, IEEE Press/ CRC Press, Boca Raton,
Florida, 1996.
45. Norman S. Nise: Control Systems Engineering, John Wiley & Sons (Wiley Student
Edition), Noida, 2004, Fourth Edition.
46. W.L. Brogan: Modern Control Theory, Second Edition, Prentice-Hall, Inc, Englewood
Cliffs, NJ., 1985
47. Donald McLean: Automatic Flight Control Systems, Prentice Hall, Englewood Cliffs,
N.J/ Hertfordshire, UK, 1990.
48. T. Kailath: Linear Systems, Prentice-Hall, Englewood Cliffs, N.J., 1980.
49. G. F. Franklin, J. D. Powell, and Emami - Naini : Feedback Control of Dynamic Systems,
Addison Wesley, 1986.
50. K. Zhou and J. C. Doyle: Essentials of Robust Control, Prentice Hall, NJ, 1998.
51. J.C. Willems: Dissipative dynamical systems, part I: General theory; part II: Linear
systems with quadratic supply rates. Archive for Rationale mechanics Analysis, vol.45,
1972, pp.321-393;
D.J. Hill and P.J. Moylan. Dissipative dynamical systems: Basic input-output and state
properties. J. of the Franklin Institute, Vol. 309, 1980, pp.327-357.
52.

AE259 Control

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

AE259 Control

Uploaded by

Copyright:

Available Formats

Lecture Notes

For Class Room teaching only

AE259: Navigation, Guidance and Control

Part 3: Control Basics

Prof. M. Seetharama Bhat

Department of Aerospace Engineering

1 Dynamics of Aerospace Vehicles 1

2.6 State Space Representation of a LTIV system . . . . . . . . . . . . . . . . . . 115

3 Control of Aerospace Vehicles - Classical Control Perspective 173

4 Pole placement Techniques 265

1.1 Inverted pendulum on a mobile cart. . . . . . . . . . . . . . . . . . . . . . . . 6

2.1 Design Steps [45] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

3.1 Pitch control loop. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175

3.5 Pitch response to solar pressure torque. . . . . . . . . . . . . . . . . . . . . . . 177

3.34 complementary Root locus for Mach hold mode. . . . . . . . . . . . . . . . . . 217

3.63 Roll Autopilot for a missile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246

2.1 Example of a stable and an unstable system . . . . . . . . . . . . . . . . . . . 85

3.1 Summary of Roll/Yaw Control System Parameters . . . . . . . . . . . . . . . . 183

Prof. M. Seetharama Bhat

Prof. M. Seetharama Bhat

Dynamics of Aerospace Vehicles

- translational motion (Newton law F = m dv

For majority of satellites, it is possible to assume decoupling between orbital mo-

1.1 Modelling of Dynamical System (LPS) : Inverted

(M + m) ÿ + ml cos θ θ̈ − mlθ̇2 sin θ = f

ml cos θ ÿ − ml sin θẏ θ̇ + m2 l2 θ̈ − mgl sin θ = 0

(M + m)ÿ + mlθ̈ = f, mÿ + mlθ̈ − mgθ = 0

Control input u(t) depends on

Figure 1.1: Inverted pendulum on a mobile cart.

1. Type and no. of sensors

state variable x(t) : θ(t), y(t), θ̇(t), ẏ(t).

1.2 Satellite Attitude Dynamics

1.2.1 Attitude Representation

e1 = (A23 − A32 )/2sinϕ

e2 = (A31 − A13 )/2sinϕ

e3 = (A12 − A21 )/2sinϕ

q12 + q22 + q32 + q42 = 1

q2 = (A31 − A13 )/4q4

q3 = (A12 − A21 )/4q4

A312 (ψϕθ) = A2 (θ)A1 (ϕ)A3 (ψ), i.e.,

tanψ = − (A21 /A22 )

tanθ = − (A13 /A33 )

and quaternions q1 ≈ ϕ/2, q2 ≈ θ/2, q3 ≈ ψ/2, q4 ≈ 1.

Combining the above with 0 = 2 (q˙1 q1 + q˙2 q2 + q˙3 q3 + q˙4 q4 ),

Since 4 × 4 matrix given above is orthogonal. (Prove !)

which is derived from

Unit Quaternion from Rotation Matrix [212] :

4 qi2 = 1 + Aii − Ajj − Akk = 1 − tr(A) + 2Aii (Qa)

4 q42 = 1 + A11 + A22 + A33 = 1 − tr(A) + 2tr(A) (Qb)

4 qi qj = Aij + Aji (Qc)

4qi q4 = Ajk − Akj (Qd)

qk = (Aik + Aki )/4qi

q4 = (Ajk + Akj )/4qi

q = ±x(i) /∥x(i) ∥ (1.4)

Orthogonalization : Strapdown inertial systems often employ the numerical integration of a

1.2.2 Basics of Satellite Attitude Dynamics

Figure 1.2: Orbit Fixed reference Coordinate System.

motion in an inertial frame of reference can be written using Euler’s equations as

Roll : Ix ω̇ + (Iz − Iy )ωy ωz = Tx + Mx

P itch : Iy ω˙y + (Ix − Iz )ωz ωx = Ty + My

Y aw : Iz ω˙z + (Iy − Ix )ωx ωy = Tz + Mz

ωx ≈ θ̇1 − Ω0 θ3 ; ωy = θ˙2 − Ω0 , ωz = θ̇3 − Ω0 θ1

Then, the governing equation is