Chapter 1

State Space Representation of LTI

Systems

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

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

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

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

1.2 Mathematical Formulation of Control System

Problems
We assume that the dynamics of a system are an evolution of its states overtime
under a given input. The evolution of the state of the system can be viewed as a
function of the state, control input, and time. Hence, in the evolution of a system, we
have system states, state mapping function, control inputs, control input mapping
function. The state mapping function gives us the effect of the status of the states
on the dynamics of a system, while the input mapping function gives us the effect
of control input on the dynamics of a system.
Mathematically
ẋ = F (x(t), u(t), t) (1.1)
Equation 1.1 is called state equation of a dynamic system. The function F (x(t), u(t), t)
can be linear or nonlinear, it can be separable or not(i.e it may or may not writ-

1
Daniel Abebe Modern Control Systems Lecture Note

ten as sum of input mapping function and state mapping function explicitly as
F (x(t), u(t), t) = f (x(t), t) + g(u(t), t)).

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

y(t) = h(x(t), u(t), t) (1.2)

Equation 1.2 is called output equation of a dynamic system. y(t) is an output (ac-
cessible states/controlled states) of a dynamic system and h(x(t), u(t), t) is output
mapping function. The function h(x(t), u(t), t) defines the relation between output
of the dynamic system, state of the system and control input.

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

In this course, we only deal with linear time-invariant (LTI) dynamic systems. LTI
dynamic systems have linear time-invariant state mapping and input mapping func-
tions. This implies that the evolution of LTI dynamic systems is linear with respect
to the state of the system and control input. Which simplifies the state equation
1.1

ẋ(t) = Ax(t) + Bu(t) (1.3)

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

matrix and matrix B is called input matrix. Their size is defined based on the
number of states and inputs a dynamic system has. For a dynamic system with n
number of states and m number of control inputs, matrix A is n × n matrix, while
matrix B is n × m matrix.
The output equation 1.2 is simplified to

y(t) = Cx(t) + Du(t) (1.4)

for LTI dynamic systems. Matrices C, D are proportionality constant matrices.

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

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

ẋ(t) =Ax(t) + Bu(t) (1.5)

y(t) =Cx(t) + Du(t) (1.6)

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

Chapter 1 2
Daniel Abebe Modern Control Systems Lecture Note

space representation of LTI dynamic systems is written as:

ẋ =Ax + Bu (1.7)
y =Cx + Du (1.8)

The bock diagram of state space representation is show in Figure 1.1.

D
u x + y
+
B ʃ C +
+

A
Figure 1.1: State space representation of a dynamic system in block diagram

1.3 Mathematical Modeling and State Space rep-

resentation of Dynamic Systems
1.3.1 State space representation of series RLC circuit
Consider the series RLC circuit in Figure 1.2, if the concern of the designer is the
voltage across the capacitor, From Kirchhoff current law

R L

vs C vo

Figure 1.2: Series RLC circuit

d
vs = iR + L i + vc
dt
d R 1 1
i = − i − vc + vs (1.9)
dt L L L
d
C vc = i
dt
d 1
vc = i (1.10)
dt C

Chapter 1 3
Daniel Abebe Modern Control Systems Lecture Note

The sate space representation will be

   
d R 1   1
 dt i  − L − L   i   L 
 d = 1  + vs (1.11)
vc 0 vc 0
dt C  
h i i
y= 0 1   (1.12)
vc

1.3.2 State space representation of two degree of freedom

quarter car suspension model
A vehicle suspension system is a complex vibration system having multiple degrees
of freedom. The purpose of the suspension system is to isolate the vehicle body
from the road imperfection. Suspension systems serve a dual purpose
• Contributing to the vehicle’s road holding/handling and braking for good ac-
tive safety and driving pleasure, and

• Keeping vehicle occupants comfortable and reasonably well isolated from road
noise, bumps, and vibrations.
The system maybe
• Passive suspension system, uses springs and dampers only or

• Semi-active suspension system include devices such as air springs and switch-
able shock absorbers, various self-leveling solutions, as well as systems like
hydro pneumatic, hydro elastic, and hydro-gas suspensions or

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

coupled with the means to impact vehicle suspension and behavior in real time
to directly control the motion of the car.
Consider a fully active suspension system shown in Figure 1.3, the free body diagram
of the system is shown in Figure 1.4.

y
mc
y1
k2 d
mw
y2
k1
z
x

Figure 1.3: 2-DOF quarter car suspension model

Chapter 1 4
Daniel Abebe Modern Control Systems Lecture Note

fk2 u fd

mc mw
y1 y2

fk2 u fd fk1

Figure 1.4: FBD of 2-DOF quarter car suspension model

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

ΣFy = mc ÿ1
− fk2 − fd + u = mc ÿ1
− k2 (y1 − y2 ) − d(ẏ1 − ẏ2 ) + u = mc ÿ1
k2 k2 d d 1
ÿ1 = − y1 + y2 − ẏ1 + ẏ2 + u (1.13)
mc mc mc mc mc
In the same way the force on the wheel is

ΣFy = mw ÿ2
fk2 + fd − u − fk1 = mw ÿ2
k2 (y1 − y2 ) + d(ẏ1 − ẏ2 ) − u − k1 (y2 − z) = mw ÿ2
k2 − k1 k2 d d k1 1
ÿ2 = y1 − y2 + ẏ1 − ẏ2 + z− u (1.14)
mw mw mw mw mw mw
Let y1 = x1 , ẏ1 = x2 , y2 = x3 and ẏ2 = x4 , the equations 1.13 and 1.14 simplify

ẋ1 =x2
k2 k2 d d 1
ẋ2 = − x1 + x3 − x2 + x4 + u
mc mc mc mc mc
ẋ3 =x4
k2 − k1 k2 d d k1 1
ẋ4 = x1 − x3 + x2 − x4 + z− u (1.15)
mw mw mw mw mw mw
The state space equation of the system is then
        
0 1 0 0 0
ẋ
 1   x1    0 
k d k d 1

   − 2 2
−
     
ẋ2   m  x
    0 
mc mc 2
mc    +  mc  u + 
 
 = c    
z

(1.16)
  
ẋ3   0 0 0 1 

x
  0   0 
  3 
    
  
k2 − k1 d k2 d   1   k1 
ẋ4 − − x4
mw mw mw mw mw mw

Chapter 1 5
Daniel Abebe Modern Control Systems Lecture Note

If the output h is the position of the body of the car with respect to the road x1 ,
the output equation is
 
x
 1
h i 
x2 
h= 1 0 0 0   
 (1.17)
x3 
 
x4
The goal of the designer is to design a controller that keeps the car body in predefined
range irrespective of the roughness of the road.

1.3.3 State space representation of DC motor

Consider armature voltage-controlled dc motor with armature resistance Ra and
armature inductance La used to drive a mechanical load with load torque τL having
rotational inertia J through a coupling gear with DC motor shaft gear to load side
rm
gear ratio at a speed ωL . Let the mechanical friction coefficient between the
rL
rotor and the load be b, as shown by the equivalent circuit in Figure 1.5. The
control input to the DC motor is armature voltage Va and the output is the speed
of the load.

Ra ia La
+ τm , ωm rm
Va M

- τL , ωL
Load

rL

Figure 1.5: Armature voltage controlled DC motor with mechanical load

The desired goal is to find Va that can derive the load at the required speed. To
determine Va we have to know the relation between the speed of the DC motor ωm
and armature voltage Va , which of course needs the mathematical model of the DC
motor.

From electrical characteristics of the DC motor the torque produced by the motor
τm is proportional to armature current ia , while the back emf Eb of the motor is
proportional to the speed of the motor ωm , so we have
τm =kt ia (1.18)
Eb =kb ωm (1.19)
where kt and kb are torque and back emf proportionality constants respectively.
Using Kirchhoff current law
d
Va = La ia + Ra ia + Eb (1.20)
dt

Chapter 1 6
Daniel Abebe Modern Control Systems Lecture Note

and from mechanical analysis of the load side

d
τm = J ωL + bωL + τL (1.21)
dt
Replacing for Eb and τm in equation 1.20 and equation 1.21 respectively and rear-
ranging;
d Ra kb 1
ia = − ia − ωm + Va (1.22)
dt La La La
d kt b 1
ωL = ia − ωL − τL (1.23)
dt J J J
From the relation between gear radius ratio and angular speed of the gears, we know
that
rL
ωm = ωL (1.24)
rm
replacing for ωm in equation 1.22, we have
d Ra kb rL 1
ia = − ia − ωL + Va (1.25)
dt La La rm La
d kt b 1
ωL = ia − ωL − τL (1.26)
dt J J J
If we consider load torque τL as uncontrolled (disturbance) input , the state space
representation of DC motor dynamics is
   Ra kb rL    1 
  
i̇a − − i a 0
 =  kLa La rm 
b    +  La  Va +  1  τL (1.27)
t −
ω̇ − ωL 0
J J J

 
h i ia
y(t) = 0 1   (1.28)
ωL

As we can see from equation 1.27 and equation 1.28, dc motor dynamics has two
states, namely ia and ωL .
If the position of the dc motor is required for control purpose like in servo motors
we add one state θ̇ = ωm . In this case the state space representation becomes
   Ra kb rL    
1  
i̇a − − 0 i 0
   La La rm   a   La 
k b  1
 
   t
   
ω̇L  = 
 − 0 ωL  +  0  Va + −  τL
 (1.29)
   J J      J
rL 
θ̇ 0 0 θ 0 0
rm
 
ia
h i 
y(t) = 0 0 1 ωL  (1.30)
 
 
θ

Chapter 1 7
Daniel Abebe Modern Control Systems Lecture Note

1.3.4 State space representation of inverted pendulum on a

cart (IPC)
An inverted pendulum on a cart is a mechanical system having a freely rotating
inverted pendulum with mass mp and length l mounted on a cart with mass mc .
The main goal of the control engineer is to determine control signal u that drives
the cart to the desired location and at the same time keeps the inverted pendulum
from falling. The system is shown in Figure 1.6.

lsinθ

y
θ
lcosθ

Figure 1.6: Inverted Pendulum on a cart system

Let us denote the distance of the cart from origin by (x, y) and its velocity by (ẋ, ẏ).
For simplicity assume the cart moves along x axis only, then velocity and position
along y axis is zero. If the pendulum deviation from the vertical axis is θ, then the
position of the pendulum (xp , yp ) with respect to the origin is given as
   
x x + l sin θ
 p =   (1.31)
yp l cos θ

With velocity
   
ẋ ẋ + θ̇ cos θ
 p =   (1.32)
ẏp −lθ̇ sin θ

The free-body diagram of the IPC is shown in Figure 1.7. If we denote the coefficient
of friction between the cart and the rail by b, the normal force exerted by a cart on
the pendulum by P and the horizontal force exerted by the cart on the pendulum
by N .

Chapter 1 8
Daniel Abebe Modern Control Systems Lecture Note

lsinθ

Jα

lcosθ
θ

P N
mpg
N
P
u
ff

Figure 1.7: Free body diagram of the Inverted Pendulum system

Using Newtons second law of motion on a pendulum free body diagram we have,
ΣFpx = mp ẍp (1.33)
2
N = mp ẍp = mp (ẍ + θ̈l cos θ − θ̇ l sin θ) (1.34)
ΣFpy = mp ÿp (1.35)
P = mp g − mp θ̈l sin θ − mp lθ̇2 cosθ (1.36)
Using Newton’s second law of motion forces exerted on a cart are
ΣFcx = mc ẍ (1.37)
u = mc ẍ + bẋ + N (1.38)
ΣFcy = mc ÿ = 0 (1.39)
Substituting for N in equation 1.38 we have,
u = (mc + mp )ẍ + bẋ + mp lθ̈ cos θ − mp lθ̇2 sin θ (1.40)
Assuming the friction between the pendulum and the cart is negligible; the momen-
tum of the pendulum rod about the pivot is described by,
Jα = J θ̈ = P l sin θ − N l cos θ (1.41)
Combining equation 1.34 , equation 1.36 and equation 1.41 we have
(J + mp l2 )θ̈ − mp lg sin θ = −mp lẍ cos θ (1.42)
Solving for ẍ and θ̈ in-terms of ẋ, x, θ̇, θ, we get the two governing dynamic equations
for IPC ,
(mp l2 + I)mp lθ̇2 sin θ − (mp l)2 g sin θ cos θ − b(mp l2 + J)ẋ + (mp l2 + I)u
ẍ =
(mp + mc )(mp l2 + I) − (mp l cos θ)2
(1.43)
−(mp l)2 θ̇2 g sin θ cos θ + bẋmp l cos θ + (mp + mc )mp lg sin θ − mp l cos θu
θ̈ =
(mp + mc )(mp l2 + I) − (mp l cos θ)2
(1.44)

Chapter 1 9
Daniel Abebe Modern Control Systems Lecture Note

Using small angle approximation about θ = 0 deg, we know that for small angle
cos θ ≈ 1, sin θ ≈ θ and θ̇ ≈ 0. Replacing cos θ = 1, sin θ = θ and θ̇ = 0 in equation
1.43 and equation 1.44 we have
mp l 2 + J b(mp l2 + J) (mp l)2 g
     
ẍ = u− ẋ − θ (1.45)
q q q
     
bmp l mp lg(mp + mc ) mp l
θ̈ = ẋ + θ̇ − u (1.46)
q q q
where q = (mc + mp )(mp l2 + J) − (mp l)2 .
To represent equation 1.45 and equation 1.46 in state space let x1 = x, x2 = ẋ,
x3 = θ and x4 = θ̇. The state space representation of IPC is;
   
0 1 0 0 0
   
ẋ  x1
 1  2 2 2
      
  0 − b(m p l + J) (m p l) g     mp l + J 
ẋ2   − 0  
 x2  

 = q q   +  q 
u
      
ẋ3  0 0 0 1 x3   0 
            
ẋ4 bmp l mp lg(mp + mc ) mp l
0 x4
   
0 −
q q q
(1.47)
 
x
     1
 
y1 1 0 0 0 x2 
y= =  
  (1.48)
y2 0 0 1 0 x3 
 
x4

1.3.5 State space representation of quadruple tank system

Consider a quadruple tank process shown in Figure 1.8. The objective is to control
the level of the liquid in the tanks by using pump 1 and pump 2. The pumps pump
the liquid from the reservoir. The volumetric flow rate1 q produced by each pump is
proportional to the voltage applied to each pump. Let this proportionality constants
be k1 and k2 for pump 1 and pump 2 respectively. The position of valve 1 controls
the volumetric flow rate to tank 1 and tank 4, while the position of valve 2 controls
the volumetric flow rate to tank 2 and tank 3. Let the position of valve 1 be γ1 and
the position of valve 2 be γ2 .
So we have
q1in = γ1 k1 V1 q2in = γ2 k2 V2
q3in = (1 − γ2 )k2 V2 q4in = (1 − γ1 )k1 V1

From fluid dynamics, the rate of mass accumulation in each tank is equal to the
difference between mass inflow and mass outflow.
ṁ = ρ(qin − qout ) (1.49)
1
Volumetric flow rate is equal to the product of the flow rate of the fluid time the area perpen-
dicular to the flow q = V elocity of the f low ∗ cross sectional area of the pipe.

Chapter 1 10
Daniel Abebe Modern Control Systems Lecture Note

Figure 1.8: Quadruple tank process

where qin is volumetric inflow rate and qout is volmetric outflow rate. The volumetric
outflow rate is determined by considering each tank as an isolated system. Using
Bernoulli equation
1 1
P1 + gh1 + v12 = P2 + gh2 + v22 (1.50)
2 2
If we take the top as point 2 and bottom of the tank as point 1 (h1 = 0), both of
them are exposed to atmospheric pressure so P1 = P2 , similarly since there is no
inflow from the top v2 = 0. The fluid flows out of each tank due to gravity only.
with these assumption’s equation 1.50 simplifies to
p
v1 = 2gh2 (1.51)

Let the outlet has a cross-sectional area a, the outflow rate will be
p
qout = av = 2gh (1.52)
m m
from basic physics ρ = = , rate of accumulation of mass ṁ is
V Ah
dh
ṁ = ρA (1.53)
dt
Combining equation 1.49 and equation 1.53, the dynamic equation for each tank
will be

Aḣ = qin − qout (1.54)

Chapter 1 11
Daniel Abebe Modern Control Systems Lecture Note

The dynamic equation of the quadruple tank system is then

A1 ḣ1 = q1in + q3out − q1out

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

√ √
a 2gh a 2gh1
 
3 3 1
 
 √A1 −  
ḣ1 A
√ 1
 γ k
1 1 0
   a4 2gh4 a2 2gh2  
   
  
ḣ2   A
 −  
0

γ2 k2  V1
 = 2 √ A2  
+
    (1.55)
  
ḣ3   a 3 2gh 3 0 (1 − γ2 )k2  V2

−
 

√A3
     
ḣ4  a4 2gh4  (1 − γ )k
1 1 0
−
A3 4
If the level of the liquid in tank 1 and tank 2 are required to be controlled, the
output equation will be
 
h
     1
 
y1 1 0 0 0 h2 
y= =  
  (1.56)
y2 0 1 0 0 h3 
 
h4

Chapter 1 12
Daniel Abebe Modern Control Systems Lecture Note

1.4 State Space Representation From Transfer Func-

tion
When a transfer function is converted to state space representation, the state space
has commonly four normal forms (canonical forms). The forms depends on the char-
acteristics polynomial of the transfer function or some other conditions (conditions
like controllability and observability which are explained later).

1. Controllable canonical form

Consider a transfer function
Y (s) N (s) bn sn + bn−1 sn−1 + bn−2 sn−2 . . . + b1 s + b0
= = n (1.57)
U (s) D(s) s + an−1 sn−1 + an−2 sn−2 . . . + a1 s + a0

If we take an intermediate function X(s) (state of the system) then we have,

Y (s) X(s) bn sn + bn−1 sn−1 + bn−2 sn−2 . . . + b1 s + b0

· = n (1.58)
X(s) U (s) s + an−1 sn−1 + an−2 sn−2 . . . + a1 s + a0

which can be written as

X(s) 1
= n n−1
(1.59)
U (s) s + an−1 s + an−2 sn−2 . . . + a1 s + a0
Y (s)
= bn sn + bn−1 sn−1 + bn−2 sn−2 . . . + b1 s + b0 (1.60)
X(s)

From equation 1.59

U (s) = (sn + an−1 sn−1 + an−2 sn−2 . . . + a1 s + a0 )X(s) (1.61)

Letting X1 = X(s), X2 = sX(s), X3 = s2 X(s), . . . , Xn = sn−1 X(s), we have

ẋ1 = x2
ẋ2 = x3
..
.
ẋn−1 = xn
ẋn = −an−1 xn − an−2 xn−1 − . . . − a1 x2 − a0 x1 + u (1.62)

which lead to a state equation

     
ẋ 0 1 ... 0 x 0
 1    1   
 ẋ2   0 0 ... 0   x2  0
      
 ..   .. .. .. ..   ..   .. 
      
 . = . . . .   .  + . u (1.63)
      
      
ẋn−1   0 0 ... 1  xn−1  0
      
ẋn −a0 −a1 ... −an−1 xn 1

Chapter 1 13
Daniel Abebe Modern Control Systems Lecture Note

From equation 1.60

Y (s) = (bn sn + bn−1 sn−1 + bn−2 sn−2 + . . . + b1 s + b0 )X(s)
y = bn ẋn + bn−1 xn + bn−2 xn−1 + . . . + b1 x2 + b0 x1
y = (b0 − bn a0 )x1 + (b1 − bn a1 )x2 + . . . + (bn−1 − bn an−1 )xn−1 + bn u
(1.64)
Which gives us an output equation
 
x
 1 
 x2 
 
 . 
h i 
y = (b0 − bn a0 ) (b1 − bn a1 ) . . . (bn−1 − bn an−1 )  ..  + bn u (1.65)
 
 
xn−1 
 
xn

2. Observable canonical form

Factoring sn and rearranging equation 1.57
Y (s) = bn U (s) + s−1 (bn−1 U (s) − an−1 Y (s) + s−1 (bn−2 U (s) − an−2 Y (s) + . . .
+s−1 (b1 U (s) − a1 Y (s) + s−1 (b0 U (s) − a0 Y (s)))))
(1.66)
Let
X1 (s) = s−1 (b0 U (s) − a0 Y (s))
X2 (s) = s−1 (b1 U (s) − a1 Y (s) + X1 (s))
..
.
Xn (s) = s−1 (bn−1 U (s) − an−1 U (s) + Xn−1 (s)) (1.67)
With this equation 1.66 becomes,
Y (s) = bn U (s) + Xn (s) (1.68)
Solving for Y (s) from equation 1.68 and replace for Y (s) in equation 1.67 and
rearranging we have
sX1 (s) = ((b0 − bn a0 )U (s) − a0 Xn (s))
sX2 (s) = ((b1 − bn a1 )U (s) − a1 Xn (s) + X1 (s))
..
.
sXn (s) = ((bn−1 − bn an−1 )U (s) − an−1 Xn (s) + Xn−1 (s)) (1.69)
Which leads to a state equation
      
ẋ 0 0 ... 0 −a0 x b − b n a0
 1    1   0 
 ẋ2  1 0 ... 0 −a1   x2   b1 − bn a1 
      
 ..   .. .. .. .. ..   ..   ..
      
 .  = . + u (1.70)

. . . .  .   .
      
−an−2  xn−1  bn−2 − bn an−2 
      
ẋn−1  0 0 0... 0
      
ẋn 0 0 ... 1 −an−1 xn bn−1 − bn an−1

Chapter 1 14
Daniel Abebe Modern Control Systems Lecture Note

and output equation

 
x
 1 
 x2 
 
 . 
h i 
y = 0 0 . . . 0 1  ..  + bn u (1.71)
 
 
xn−1 
 
xn

As we can see from equation 1.63 , equation 1.65, equation 1.70 and equation
1.71, the system matrix of observable canonical form is a transpose of the
system matrix of controllable canonical form (Ao = ATc ). Input matrix of
observable canonical form is equal to the transpose of the output matrix of
controllable canonical form (Bo = CcT ). The output matrix of observable
canonical form is equal to the transpose of controllable canonical form input
matrix (Co = BcT ). State transmission matrix of observable canonical form is
a transpose of state transmission matrix of controllable canonical form (Do =
DcT ).

3. Diagonal canonical Form

Suppose the transfer function has distinct poles p1 , p2 , . . . , pn and can be fac-
torized to
c1 c2 cn
Y (s) = U (s) + U (s) + . . . + U (s) (1.72)
s + p1 s + p2 s + pn
1 1 1
Letting X1 (s) = U (s), X2 (s) = U (s), . . . , Xn (s) = U (s)
s + p1 s + p2 s + pn
we have

ẋ1 = −p1 x1 + u
ẋ2 = −p2 x2 + u
..
.
ẋn = −pn xn + u (1.73)

Which leads to a state equation

 
ẋ1     
  −p 1 0 . . . 0 x 1
 ẋ2  
    1  
  .
  0 −p . . . 0 
 ..  

2   x2   .. 
 . = . 
.. .. ..   ..  +   u
     (1.74)
   .. . . .   .  1
 
ẋn−1      
  0 0 . . . −pn xn 1
ẋn

Chapter 1 15
Daniel Abebe Modern Control Systems Lecture Note

and output equation

 
x
 1
h i  x2 


y = c1 c2 . . . cn  .. 
 (1.75)
.
 
xn

4. Jordan Canonical Form

If the transfer function can be factorized and have r multiplicities of pole p1 ,
and the rest poles are distinct, the transfer function can be written as
c1 c2 cr
Y (s) = U (s) + 2
U (s) + . . . + U (s)+
s + p1 (s + p1 ) (s + p1 )r
cr+1 cr+2 cn
U (s) + U (s) + . . . + U (s) (1.76)
s + p2 (s + p3 ) (s + p(n−r) )

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

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

ẋ1 = −p1 x1 + u
ẋ2 = −p1 x2 + x1
..
.
ẋr = −p1 xr + xr−1
ẋr+1 = −p2 xr+1 + u
...
ẋn = −pn−r xn−r + u (1.77)

Chapter 1 16
Daniel Abebe Modern Control Systems Lecture Note

Which lead to a state equation

     
ẋ1 −p 0 0 ... 0 0... 0 x1 1
   1    
 ẋ2   1 −p1 0 ... 0 0... 0   x2  0
      
      
 ẋ3   0 1 −p1 ... 0 0... 0   x3  0
      
 ..   .. .. .. .. .. .. ..   ..   .. 
      
 .   . . . . . . .   .  .
 =   +  u
−p1 . . .
      
 ẋr   0 0 0 ... 1 0  xr−1  0
      
−p2 . . .
      
ẋr+1   0 0 0 ... 0 0   xr  1
 ..   .. .. .. .. .. .. ..   ..   .. 
      
 .   . . . . . . .   .  .
      
ẋn 0 0 0 ... 0 0 . . . −pn−r xn 1
(1.78)
The output matrix is similar to diagonal normal form
 
x
 1
h i 
 x2 
y = c1 c2 . . . cn   .. 
 (1.79)
.
 
xn

1.5 Transfer Function From State Space Repre-

sentation
Consider a LTI system represented by state space representation on equation 1.7
and equation 1.8, with initial condition x(0) = x0 , taking Laplace transform of the
equations, we have

sX(s) − x0 = AX(s) + BU (s)

(sI − A)X(s) = x0 + BU (s)
X(s) = (sI − A)−1 (x0 + BU (s)) (1.80)

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

= C(sI − A)−1 (x0 + BU (s)) + DU (s)
= C(sI − A)−1 x0 + (C(sI − A)−1 B + D)U (s) (1.81)

If the initial condition is zero(x0 = 0, which is the case for all transfer functions),
the transfer function from input to output is
Y (s)
= C(sI − A)−1 B + D
U (s)
Cadj(sI − A)B + det(sI − A)D
= (1.82)
det(sI − A)
The determinant of (sI − A) is a characteristic polynomial of the transfer function.

Chapter 1 17
Daniel Abebe Modern Control Systems Lecture Note

1.6 Review on Eigenvalues and Eigenvectors

The roots of the characteristic’s equation of n × n matrix A, which is defined as
det(λI − A) = 0 are called eigenvalues of the matrix A.

Any nonzero vectors νi , qi which satisfies the matrix equation (λi I−A)νi = 0, qi (λi I−
A) = 0, where λi are eigenvalues of the matrix are called left eigenvector and right
eigenvector of matrix A corresponding to eigenvalue λi respectively. The matrix
formed from left eigenvectors (νi ) of the matrix A is called modality matrix of A
and denoted by M .
h i
M = ν1 | ν2 | . . . |νn (1.83)

1.7 Solution of State Space Equations

Consider the time derivative of e−At x, using chain rule we have
d −At d dx
(e x) = (e−At )x + e−At
dt dt dt
= −Ae−At x + e−At (ẋ)
= −Ae−At x + e−At (Ax + Bu)
= e−At (−Ax + Ax) + e−At Bu
= e−At Bu (1.84)

d
From equation 1.84 we can see that (e−At x) = e−At Bu = e−At (ẋ − Ax), using this
dt
the solution to state equation is

e−At (ẋ − Ax) = e−At Bu

Z t Z t
d −At
(e x)dt = e−Aτ Bu(τ )dτ
t=0 dt
Zt=0
t
e−At x |t=t
t=0 = e−Aτ Bu(τ )dτ
Zt=0
t
e−At x(t) − x(0) = e−Aτ Bu(τ )dτ
t=0
Z t
At
x(t) = e x(0) + e At
e−Aτ Bu(τ )dτ
Z t t=0
x(t) = eAt x(0) + eA(t−τ ) Bu(τ )dτ (1.85)
t=0

and the output solution will be

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

Z t
At
y(t) = Ce x(0) + eA(t−τ ) Bu(τ )dτ + Du(t)
Zt=0
t
At
y(t) = Ce x(0) + (eA(t−τ ) B + D)u(τ )dτ (1.86)
t=0

Chapter 1 18
Daniel Abebe Modern Control Systems Lecture Note

The CeAt x(0) inZequation 1.86 is called zero input response of the dynamic system
t
while the term (eA(t−τ ) B + D)u(τ )dτ is called zero initial condition response.
t=0
The frequency domain representation of equation 1.86 is equation 1.81.

The term eAt denoted as Φ(t) is called state transition matrix as it transfer the state
of a dynamic system from any initial condition x(t0 ) to a final state x(t) whether
the dynamic system is driven by an input or not. A few properties of the state
transition matrix are stated below (It is the property due to the exponential nature
of the function).
• Φ(0) = I since eA∗0 = I
• Φ(t2 − t1 )Φ(t1 − t0 ) = Φ(t2 − t0 )
• Φ−1 (t) = Φ(−1) since (eAt )−1 = eA(−t) = Φ(−t)
• Φ(t2 − t1 )=Φ(t1 )Φ(−t2 )
• Φ(t)n =Φ(nt)
The state transition matrix is determined one of the following methods,
1. Using Inverse Laplace transform method

Φ(t) = L−1 ((sI − A)−1 ) (1.87)

2. Using Series expansion method

i=n
At
X (At)i
Φ(t) = e = (1.88)
i=0
i!
where n is the dimension of matrix A.
3. Using Cayley Hamilton Theorem
Cayley Hamilton theorem states that any function of a n × n square matrix
f (A) can be represented by a finite sum of powers of the matrix not exceed-
ing order n − 1. Let the eigenvalues of A be λi ’s where i = 1, 2, . . . n and the
characteristics polynomial corresponding to each λi be pi λ. So we have n simul-
taneous equations to solve for the coefficient of the characteristics polynomial.
As Cayley Hamilton theorem states matrix A satisfies its own characteristics
equation, that is why we say f (A) can be represented by a finite sum of powers
of the matrix not exceeding order n − 1.
f (A) = α0 I + α1 A + α2 A2 + . . . + αn−1 An−1 (1.89)
The constants are determined from simultaneous equation resulting from
p(λi ) = α0 + α1 λi + α2 λ2i + . . . + αn−1 λi = f (λi ) (1.90)
Where i = 1, 2, . . . n.
If the matrix A r multiplicity of eigenvalue, the polynomial with the multi-
plicity eigenvalues is derivated r times and used as an additional equation, to
have an equal number of equations as the number of variables.

Chapter 1 19
Daniel Abebe Modern Control Systems Lecture Note

4. Using Modality Matrix

Let the diagonal matrix formed from eλi t ordered from smallest value of λi to
largest value of λi (ı.e. λ1 < λ2 < · · · < λn ), where λi are eigenvalues of A be
denoted as eΛt , then the state transition matrix can be solved using modality
matrix M and eΛt as
 
λ1 t
e 0 ··· 0
 
λ2 t
···
 
−1 Λt −1 
 0 e 0 
Φ(t) = M e M = M  . .. ..  M
 (1.91)
 .. . ··· . 
 
λn t
0 0 ··· e

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

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

1.8 Similarity Transformation

If there exist a transformation matrix that transform a state x into new state ξ, in
such a way that ξ = T x which means x = T −1 ξ and ẋ = T −1 ξ˙ then the state space
representation of LTI system on equation 1.7 and equation 1.8 will be transformed
into a dynamics equation with different state coordinates of the form
ξ˙ = T AT −1 ξ + T Bu
ξ˙ = Āξ + B̄u (1.92)

y = CT −1 ξ + Du
y = C̄ξ + D̄u (1.93)
where Ā = T AT −1 ,B̄ = T B, C̄ = CT −1 and D̄ = D.
The main purpose of similarity transformation is to simplify dynamic equation of
the system for analysis (converting any state representation of any dynamic system
to controllable or observable canonical form will simplify the analysis of the system).
Some dynamic system are complex in some co-ordinate systems, but simpler in the
other co-ordinate system.

Similarity transformation only changes the coordinates in which the dynamic system
is expressed. It does not change
• Eigenvalues of the system matrix
Proof:- The characteristics equation of the transformed system matrix is
det(λI − Ā) = 0
det(λI − T AT −1 ) = 0
det(T λT −1 − T AT −1 ) = 0
det(T )det(λI − A)det(T −1 ) = 0
det(λI − A) = 0

Chapter 1 20
Daniel Abebe Modern Control Systems Lecture Note

• Transfer function of the dynamic equation

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

Y (s) C̄adj(sI − Ā)B̄

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

• Solution of output equation

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

Y (s) = C̄(sI − Ā)−1 T x0 + (C̄(sI − Ā)−1 B̄ + D̄)U (s)

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

Note Since the state of the system is changed solution to the state equation is differ-
ent. But we can derive a solution for the original state equation from the solution
of the new state equation by x(t) = T −1 ξ(t).

Chapter 1 21
Daniel Abebe Modern Control Systems Lecture Note

1.9 Phase Plane and Stability Analysis of LTI Sys-

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

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

• Laypunov stable if and only if given ϵ > 0 there exist σ > 0 such that ||x0 || < σ
implies that ||x0 || < ϵ for all time t > 0. This means for every initial condition
in a sphere of radius σ, the final state of the system should be in the a sphere of
radius ϵ. LTI system is Laypunov stable if and only if trace2 (A) < 0, det(A) ≥
0 and rank(A) = rank(A2 ) (i.e. All eigenvalues of system matrix A should
have zero or negative real parts).

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

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

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

To determine whether the LTI system is Lyapunov stable or not, Let us take a
quadratic function V = xT P x, where P are symmetric3 positive definite4 matrix,
2
trace(A) is the sum of all diagonal elements of A
3
A matrix P is called symmetric if and only if P = P T
4
A matrix P is positive definite if and only if all eigenvalues of P are positive or leading principal
minors of A are positive

Chapter 1 22
Daniel Abebe Modern Control Systems Lecture Note

then

V̇ = xT P ẋ + ẋT P x
= xT P (Ax) + (Ax)T P x
= xT (P A + AT P )x (1.94)

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

take positive definite symmetric matrix R such that,

P A + AT P + R = 0 (1.95)

Equation 1.95 is called Lyapunov equation.

BIBO/External Stability
It is a stability analysis of the response of the dynamic system to bounded input.
Definition:- A dynamic system is bounded input bounded output stable if and only
if for zero initial condition if bounded input always invoke bounded output.
Consider LTI dynamic system with zero initial condition response
Z t
y(t) = C eA(t−τ ) Bu(τ )dτ + Du(t) (1.96)
0

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

Z t Z t
A(t−τ )
y(t) = Ce Bu(τ )dτ + Du(t) < CeA(t−τ ) Bαdτ + Dα
0 0
Z t
<α CeA(t−τ ) Bdτ + D (1.97)
0

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

Chapter 1 23
Daniel Abebe Modern Control Systems Lecture Note

1.9.2 Phase plane analysis of LTI systems

Phase plane analysis a graphical analysis of second order system with respect to
initial conditions. Consider the following second order system
ẋ1 = f1 (x1 , x2 )
ẋ2 = f1 (x1 , x2 ) (1.98)
where x1 and x2 are states of the system and f1 and f2 are linear (for LTI systems)
and non linear for nonlinear autonomous systems.
Since there is no control input, the phase plane analysis is used to understand the
property of the system under various initial conditions. For linear time invariant
systems equation 1.98 simplifies to
ẋ1 = a11 x1 + a12 x2
ẋ2 = a21 x1 + a22 x2 (1.99)
Geometrically, the state space of this system is a plane having x1 , x2 as coordinates.
This plane is called phase plane. The solution of equation 1.98 or equation 1.99 with
time varies from zero to infinity can be represented as a curve in the phase plane.
Such a curve is called a phase plane trajectory. A family of phase plane trajectories
is called a phase portrait of a system.
Equation 1.99 can be written as
ẍ1 =a11 ẋ1 + a12 ẋ2
ẍ1 =a11 ẋ1 + a12 (a21 x1 + a22 x2 )
ẋ1 − a11 x1
ẍ1 =a11 ẋ1 + a12 (a21 x1 + a22 )
a12
(a11 a12 + a12 a22 ) (a12 a21 − a12 a22 a11 )
ẍ1 = ẋ1 + x1
a12 a12
ẍ1 +ξ1 ẋ1 + ξ2 x1 = 0 (1.100)
(a11 a12 + a12 a22 ) (a12 a21 − a12 a22 a11 )
where ξ1 = and ξ2 = .
a12 a12
The solution of equation 1.100 has the form
x(t) = k1 eλ1 t + k2 eλ2 t f or λ1 ̸= λ2 (1.101)
x(t) = k1 eλ1 t + tk2 eλ2 t f or λ1 = λ2 (1.102)
A point in phase plane where ẋ = 0 is called a singular point or equilibrium point of
the system. For Linear system we have only one singular/equilibrium point, namely
the origin. The trajectories in the vicinity of this singularity point can display quite
different characteristics, depending on the values of λ1 and λ2 . The following cases
can occur
• λ1 ,λ2 are both real and have the same sign (+ or -)
• λ1 ,λ2 are both real and have opposite sign
• λ1 ,λ2 are complex conjugates with non-zero real parts
• λ1 ,λ2 are complex conjugates with real parts equal to 0

Chapter 1 24
Daniel Abebe Modern Control Systems Lecture Note

The characteristics of the system under this cases is described below

1. The first condition corresponds to a node

(a) λ1 < 0,λ2 < 0 singularity point is called stable node. In this every phase
portraits goes to the singularity point as shown in Figure 1.9, which
means the system is asymptotically stable.

Figure 1.9: Stable node

(b) λ1 > 0,λ2 > 0 singularity point is called unstable node. In this every
phase portraits goes away from the singularity point as shown in Figure
1.10, which means the system is unstable.

Figure 1.10: Unstable node

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

rium point (converge to the singular point) at one end and move away from
the equilibrium point (diverge to infinity) in the other end as shown in Figure
1.11.

Figure 1.11: Saddle Point

3. The third case corresponds to a focus.

(a) Re(λ1 < 0,λ2 < 0) : stable focus, where trajectories circulate around
the point and finally tend to the point as shown in Figure 1.12. The
circulation shows that there is an oscillation in the system response and
the convergence show that this oscillations will die-out and the system
becomes stable over time.

Chapter 1 25
Daniel Abebe Modern Control Systems Lecture Note

Figure 1.12: Stable focus

(b) Re(λ1 > 0,λ2 > 0) : unstable focus, where trajectories circulate around
the point and diverge to infinity as shown in Figure 1.13. The circulation
shows that there is an oscillation in the system response and the diver-
gence show that this oscillations grows large and the system becomes
unstable over time.

Figure 1.13: Unstable focus

4. The last condition corresponds to a center point. All trajectories are ellipses
and the singularity point is the center of these ellipses as shown in Figure 1.14.

Figure 1.14: Center point

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

Chapter 1 26
Daniel Abebe Modern Control Systems Lecture Note

1.10 Examples
1. Find the controllable canonical form, observable canonical form and diagonal
normal form or Jordan canonical form of the following transfer function.

Y (s) 2s3 + 5s2 + 3s + 1

= 4
U (s) 4s + 2s2 + 3s + 6
0.5s3 + 1.25s2 + 0.75s + 0.25
=
s4 + 0.5s2 + 0.75s + 1.5
Solution:

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

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

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

(b) Observable canonical form can be determined from controllable canonical

form by Ao = ATc , Bo = CcT , Co = BcT and Do = DcT .
      
ẋ 0 0 0 −1.5 x 0.25
 1    1  
ẋ2  1 0 0 −0.75 x2  0.75
      
 =   +  u
ẋ3  0 1 0 −0.5  x3  1.25
      
      
ẋ4 0 0 1 0 x4 0.5
 
x
 1
h ix2 

y= 0 0 0 1   

x3 
 
x4

Chapter 1 27
Daniel Abebe Modern Control Systems Lecture Note

(c) Since the poles of the system are distinct, p1 = 0.71 + i,p2 = 0.71 − i,
p3 = −0.71 + 0.70i and p4 = −0.71 − 0.70i. We have diagonal normal
form. The transfer function is factorized into
Y (s) 0.30 − 0.33i 0.30 + 0.33i −0.05 − 0.06i −0.05 + 0.06i
= + + +
U (s) s + 0.71 + i s + 0.71 − i s − 0.71 + 0.70i s − 0.71 − 0.70i

which has Jordan canonical form of

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

2. Find state transition matrix for the following dynamic system

 
0 1
ẋ =  x
−2 −3

Solution

(a) Method 1:-Using Inverse Laplace transform

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

2 s+3

Chapter 1 28
Daniel Abebe Modern Control Systems Lecture Note

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

(b) Method 2:-Using Series expansion

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

Chapter 1 29
Daniel Abebe Modern Control Systems Lecture Note

Solving for α0 and α1 , gives a result

α0 = 2e−t − e−2t , α1 = e−t − e−2t

So by Cayley Hamilton Theorem

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

(d) Method 4:-Using Modality Matrix

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

The eigenvector corresponding to λ2 = −1, ν2 is

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

Hence the modality M matrix, it’s inverse M −1 and eΛt are

     
−2t
1 1 −1 1 e 0
M =  , M −1 =  , eΛt =  ,
2 1 2 −1 0 e−t

Chapter 1 30
Daniel Abebe Modern Control Systems Lecture Note

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

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

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

Y (s) Cadj(sI − A)B + det(sI − A)D

=
U (s) det(sI − A)

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

s2 + s − 2
= +2
s3 − 2s2 + 4s − 5

Y (s) 2s3 − 3s2 + 9s − 12

= 3
U (s) s − 2s2 + 4s − 5
4. Find the eigenvalues and their corresponding eigenvectros for a matrix
 
1 −1
A= 
0 −1
Solution
The characteristics equation of the matrix A determined from det(λI − A) = 0
   
λ 0 1 −1
det  −  = 0
0 λ 0 −1
(λ1 − 1)(λ2 + 1) = 0
λ1 = 1, λ2 = 2

Chapter 1 31
Daniel Abebe Modern Control Systems Lecture Note

The eigenvector corresponding to λ1 = −1, ν1 is

(λ1 I − A)ν1 = 0
     
−1 0 1 −1 ν
 −   11  = 0
0 −1 0 −1 ν12
  
−2 1 ν
   11  = 0
0 0 ν12

Which result in

−2ν11 + ν12 = 0 ⇒ ν12 = 2ν11

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

where m is any real number.

The eigenvector corresponding to λ2 , ν2 is

(λ2 I − A)ν1 = 0
     
1 0 1 −1 ν
 −   21  = 0
0 1 0 −1 ν22
  
0 1 ν
   21  = 0
0 2 ν22

Which result in

ν22 = 0, 2ν22 = 0

ν21 can be any value and ν22 = 0. So ν2 is

   
ν21 k
ν2 =   = m  
ν22 0

Chapter 1 32
Daniel Abebe Modern Control Systems Lecture Note

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

Solution

  
  
1 2
λ 0 λ − 1 −2
det( −  = det   = 0
0 λ 3 0 −3 λ
λ(λ − 1) + 5 = 0
λ2 − λ + 6 = 0 ⇒ (λ − 3)(λ + 2) = 0
λ1 = 3 λ2 = −2
the eigenvectors associated with λ1 = 3, ν1 is
(λI − A) ν1 = 0

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

which result in
2ν11 − 2ν12 = 0, −3ν11 + 3ν12 = 0
   
ν11 1
ν11 = ν12 = 1 ν1 ⇒   = ν11  
ν11 1

the eigenvectors associated with λ2 = −2, ν2 is

(λI − A) ν2 = 0

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

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

the modality matrix M is

 
h i 1 −2
M = ν1 ν2 = 
1 3

Chapter 1 33
Daniel Abebe Modern Control Systems Lecture Note

6. For the following matrix determine the state transition matrix Φ(t), eigenvec-
tors
 
−6 2
A= 
−6 1
Solution
     
λ 0 −6 2 λ + 6 −2
det( −  = det   = 0
0 λ −6 1 6 λ−1
(λ + 6)(λ − 1) + 12 = 0
λ2 5λ + 6 = 0 ⇒ (λ + 3)(λ + 2) = 0
λ1 = −3 λ2 = −2
the eigenvectors associated with λ1 = −3, ν1 is
(λI − A) ν1 = 0
    
ν11 −6 2 ν
−3   =    11 
ν12 −6 1 ν12
which result in
− 6ν11 + 2ν12 = −3ν11 , −3ν11 + 2ν12 = 0
− 6ν11 + ν12 = −3ν12 − 6ν11 + 4ν22 = 0
3ν11 = 2ν11 ⇒ ν11 = 2, ν12 = 3
 
2
⇒ ν1 =  
3
the eigenvectors associated with λ2 = −2, ν2 is
(λI − A) ν2 = 0

[λI − A] ν1 = 0
    
ν11 −6 2 ν
−2   =    21 
ν12 −6 1 ν22
which result in
− 6ν21 + 2ν22 = −2ν11 , −3ν11 + 2ν12 = 0
− 6ν21 + ν22 = −2ν12 − 6ν11 + 4ν22 = 0
3ν22 = 2ν21 ⇒ ν21 = 1, ν22 = 2
 
1
⇒ ν2 =  
2

Chapter 1 34
Daniel Abebe Modern Control Systems Lecture Note

the modality matrix M and it inverse is

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

The diagonal of matrix Ā = M −1 AM

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

the state transition matrix Φ(t) is given as

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

7. Given matrix A find Φ(t)

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

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

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

Chapter 1 35
Daniel Abebe Modern Control Systems Lecture Note

Laplace inverse transform.

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

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

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

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

zero input response yzinput is

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

Chapter 1 36
Daniel Abebe Modern Control Systems Lecture Note

zero state response yzs will be

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

9. Check the stability of the following system matrix

 
3 −2
A= 
1 −1

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

using the following command on MATLAB:

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

Chapter 1 37
Daniel Abebe Modern Control Systems Lecture Note

−13
P is symmetric but not positive definite because det(P ) = < 0. Hence
16
the A is not stable. Checking the eigenvalues
√ √
||λI − A|| = 0 ⇒ (λ − 3)(λ + 1) + 2 = 0 ⇒ λ1 = 1 − 2, λ2 = 1 + 2

both eigenvalues has strictly positive real parts, so A is unstable.

10. For dc motor model derive on 1.3.3, check stability with Ra = 0.25, J =
0.0025, La = 0.003, kb = 0.0067, kt = kb .

Solution:
using MATLAB command [V, D] = eig(A), give us the eigenvaluesλ1 = −83.25, λ2 =
−8.08 and corresponding eigenvectors of the dc-motor dynamics
   
−1 0.03
ν1 =   ν2 =  
0.034 −1

Since all eigenvalues has strictly negative real parts the dynamic system is
stable.

11. Check stability of IPC dynamic system derived on 1.3.4, with parameters
mp = 0.5, mc = 0.2, b = 0.1, l = 0.3, j = 0.5mp l2 , g = 9.81.
Solution:
using MATLAB command [V, D] = eig(A), give us the eigenvalues λ1 =
0, λ2 = −0.14 λ3 = −5.21, λ4 = 5.17 and corresponding eigenvectors
of the IPC dynamic system
       
1 −0.99 −0.01661 0.0158
       
       
0  0.1414   0.0862  0.0816
ν1 =   ν2 = 
  
 ν3 = 
 
 ν2 = 
  
 −0.1879 

0  0.0021  0.1891
       
0 −0.0003 0.9782 0.9784

Since all eigenvalues doesn’t have strictly negative real parts the dynamic
system is unstable.

12. Consider the following simple mass measuring system, with system parameters
k = 20N/m, ms = 1kg and d = 9N s/m shown in Figure 1.15. (Mid-Exam
2021)
The system shows the mass of the object to be measured on the scale. The
reading on the scale is proportional to the mass of the object(y = αmo ). Where
α is a proportionality constant (for this problem use α = 1).

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

(b) Find the eigenvalues and eigenvectors of the system.
(c) Determine whether the system is stable or not.
(d) Determine the state transition matrix of the system.

Chapter 1 38