Chapter 3

Input Output Representation of Linear Systems

A system is a collection of objects that interact with each other as well as with the environment that they are con-
tained in. Examples are mechanical systems, electrical systems, thermodynamical systems, biological systems
etc. Two types of quantities are required to describe the evolution of a system. Quantities that do not change
with time, called system parameters and quantities that vary with time (quantities that are a function of time),
called system variables. The system variables that are directly influenced by the interaction of the system with
its environment (external surrounding) are called the inputs of the system while the output of the system is some
function (or in the case of multiple outputs, a collection of functions) of the system variables.

3.1 Graphical Description of a Input Output System

In this class we will graphically represent a system by a box with inputs coming in and outputs leaving the box.
This allows us to think of a system as a black box that processes an input signal to give an output signal. A
system with a single input and a single output is called a single input single output (SISO) system while a system
with multiple inputs and multiple outputs is called a multi input multi output (MIMO) system. In this class we
will consider only SISO systems.

Fig. 3.1 Block diagram representation of a SISO system.

3.2 Mathematical Description of a System

Let U be the set of all possible signals (That is all possible functions of time). Then we can think of a system,
denoted by S , as a function that takes an input function, u(t), from the set U as the argument and gives the
output function, y(t), as its result. That is,

89
90 3 Input Output Representation of Linear Systems

Fig. 3.2 Block diagram representation of a three input two output MIMO system.

y(t) = S (u(t)). (3.1)

This is consistent with our desire to graphically represent a system as a box that processes the input to give us
the output. Let y1 (t) = S (u1 (t)) and y2 (t) = S (u2 (t)). That is y1 (t) is the output of the system to the input
u1 (t) while y2 (t) is the output of the system to the input u2 (t). Then, If the system satisfies the following two
properties,

S (u1 (t) + u2 (t)) = y1 (t) + y2 (t), (3.2)

S (k u1 (t)) = k y1 (t) (3.3)

it is referred to as a Linear system. A system is said to be time invariant if the system response to any given input
is exactly the same irrespective of the time at which the input was applied. For instance the behavior (response)
of a time invariant system to a sinusoidal input should be exactly the same whether you apply the input at time
t = 0 or t = 10 seconds. In this class we will consider only Linear Time Invariant (LTI) systems. Usually LTI
systems result in constant coefficient linear ordinary differential equation (ODE) models. For instance consider
a simple forced spring mass damper system shown in figure 3.4.
Let the displacement x(t) from its equilibrium be the desired output and f (t) be the input. That is y(t) = x(t),
u(t) = f (t). Applying Newton’s equations yield the following ODE model of the system in terms of the input
and output variables.
M ÿ +Cẏ + Ky = u(t). (3.4)

Fig. 3.3 The Block Diagram Representation of the Spring Mass Damper System.

However we would like to represent this system as in figure 3.3. How do we make this mathematical model
fit into our desired graphical representation? or more precisely, how can we write down the function S (•)
explicitly. The mathematical tool required for this is Laplace transforms. In the next couple of sections we will
briefly and quickly review this mathematical background. We will do so using two typical examples: a first order
system and a second order system. Any linear system can be thought of as higher order coupling of first order
and second order systems.
3.2 Mathematical Description of a System 91

3.2.1 First Order Linear Time Invariant System

Consider the single water tank system with an inflow and an outflow. Let the height of the water level of the tank
be h(t) and let it be the desired output, the cross sectional area of the tank is A, the inflow rate is qi (t). Assuming
that the outflow rate, qo (t), is a linear function of the height, that is q0 (t) = rh(t) where r is a positive constant.
Applying the principle of conservation of mass, the governing equation, ODE model of the system is,

Aḣ(t) + rh(t) = qi (t). (3.5)

r
Dividing by A and setting p1 = A and u(t) = A1 qi (t) in (3.5) we have

ḣ(t) + p1 h(t) = u(t). (3.6)

For a given u(t) that does not grow faster than an exponential function the output will also not grow faster than
an exponential function (this is a consequence of linearity of the system). Thus taking Laplace transform of both
sides of (3.6) we have

L ḣ(t) + p1 h(t) = L {u(t)}, (3.7)
(s + p1 )H(s) − h(0) = U(s). (3.8)

This yields
1 1
H(s) = h(0) + U(s). (3.9)
s + p1 s + p1
The polynomial Δ (s) = s + p1 is the characteristic polynomial of the system and

Δ (s) = s + p1 = 0, (3.10)

is the characteristic equation. The output h(t) can be calculated by taking the inverse Laplace transform of H(s).
That is
 
1 1
h(t) = L −1 {H(s)} = L −1 h(0) + U(s) , (3.11)
s + p1 s + p1
 
1
= h(0)e−p1 t 1(t) + L −1 U(s) , (3.12)
s + p1

The first part of the output

hIC (t) = h(0)e−p1 t 1(t) (3.13)
is called the initial condition response, since it depends only on the initial conditions and not on the input.
Observe that p1 = Ar depends only on the system parameters. The part
 
−1 1
h f (t) = L U(s) (3.14)
s + p1

is called the forced response, since it depends only on the input and has nothing to do with the initial conditions.
In this notation the complete response is

h(t) = hIC (t) + h f (t).

Note that if we set the initial condition to zero h(0) = 0 then the response of the system is simply the forced
response. Thus let us look at the forced response a little more closely.
Most of the inputs of interest to control Engineers result in a Laplace transform that is a proper rational
function of the complex variable s. That is U(s) is of the form

N(s)
U(s) = (3.15)
(s + β1 )(s + β2 ) ∙ ∙ ∙ (s + βk )
92 3 Input Output Representation of Linear Systems

where β1 , β2 , ∙ ∙ ∙ βk are real or complex constants and N(s) is a polynomial of the complex variable s. For example
if u(t) = e−3t 1(t) then U(s) = s+3 1
and if u(t) = e−2t cos (5t) 1(t) then U(s) = (s+2)
s+2 s+2
2 +52 = (s+2+5 j)(s+2−5 j) .
Let us substitute (3.15) in the forced response (3.14) and take the partial fraction expansion of it under the
assumption that all p1 , β1 , ∙ ∙ ∙ , βk are distinct,
   
1 α1 r1 rk
h f (t) = L −1 U(s) = L −1 + +∙∙∙+
s + p1 s + p1 s + β1 s + βk
h f (t) = α1 e−p1 t 1(t) + r1 e−β1 t 1(t) + ∙ ∙ ∙ + rk e−βk t 1(t), (3.16)

where α1 , r1 , ∙ ∙ ∙ rk are constants determined in the process of partial fraction expansion,


α1 = U(s) s=−p1 ,

U(s) 
ri = (s + βi ) .
(s + p1 ) s=−βi
The forced response given by (3.16) can further be broken up into two parts,

h(t) = htr (t) + hss (t).

Where htr (t), given by,

htr (t) = α1 e−p1 t 1(t), (3.17)
is called the the transient response. Observe that the transient response htr (t) is qualitatively similar to the initial
condition response hIC . The hss (t), given by,

hss (t) = r1 e−β1 t 1(t) + ∙ ∙ ∙ + rk e−βk t 1(t), (3.18)

is referred to as the steady state response. To illustrate why we use these names consider the following. If the
real part of p1 is positive then
lim htr (t) = lim α1 e−p1 t 1(t) = 0
t→∞ t→∞

and thus it decays to zero. Hence the name transient response. Also because of this

lim h(t) = lim htr (t) + lim hss (t) = lim hss (t).
t→∞ t→∞ t→∞ t→∞

and thus hss (t) is the response of the system when t becomes very large or in other words is the steady state
response.

3.2.2 Second Order Linear Time Invariant System

Consider a simple forced spring mass damper system shown in figure 3.4. Let the displacement x(t) from its
equilibrium be the desired output and f (t) be the input. That is y(t) = x(t) and u(t) = y(t). Newton’s equations
yield the following ODE model of the system.

M ÿ(t) +Cẏ(t) + Ky(t) = f (t). (3.19)

Dividing by M and setting ωn2 = K

M and 2ζ ωn = C
M in (3.19) we have

1
ÿ(t) + 2ζ ωn ẏ(t) + ωn2 y(t) = u(t). (3.20)
M
For a given u(t) that does not grow faster than an exponential function the output will also not grow faster than
an exponential function (this is a consequence of linearity of the system). Thus taking Laplace transform of both
sides of (3.20) we have
3.2 Mathematical Description of a System 93

Fig. 3.4 Forced Spring Mass Damper System

 1
L ÿ(t) + 2ζ ωn ẏ(t) + ωn2 y(t) = L { u(t)}, (3.21)
M
1
(s2 + 2ζ ωn s + ωn2 )Y (s) − ẏ(0) − (s + 2ζ ωn )y(0) = U(s). (3.22)
M
This yields
1 s + 2 ζ ωn 1/M
Y (s) = ẏ(0) + y(0) + 2 U(s). (3.23)
s2 + 2ζ ω n s + ωn
2 s2 + 2ζ ωn s + ωn
2 s + 2ζ ωn s + ωn2
The polynomial Δ (s) = s2 + 2ζ ωn s + ωn2 is the characteristic polynomial of the system and

Δ (s) = s2 + 2ζ ωn s + ωn2 = (s + p1 )(s + p2 ) = 0, (3.24)

is the characteristic equation where −p1 , −p2 are the roots of the characteristic equation polynomial Δ (s). Ex-
plicitly p
−p1 = (−ζ + ζ 2 − 1)ωn , (3.25)
p
−p2 = (−ζ − ζ 2 − 1)ωn (3.26)
and they will be either real and distinct, real and repeated, complex conjugate or purely imaginary and conjugate
depending on ζ .
Over Damped If ζ > 1 then both −p1 , −p2 will be real and distinct.
Under Damped If 0 < ζ < 1 p they are complex conjugate of each other and −p1 = −ζ ωn + iωd , and −p2 =
−ζ ωn − iωd , where ωd = ωn 1 − ζ 2 is called the damped natural frequency of the system.
Un Damped If ζ = 0 then −p1 = iωn , −p2 = −iωn and they are purely imaginary and conjugate of each
other.
Critically Damped If ζ = 1 then −p1 = −p2 = −ζ ωn and they are real and repeated.
Using (3.24)

1 s + 2ζ ωn 1/M
Y (s) = ẏ(0) + y(0) + U(s). (3.27)
(s + p1 )(s + p2 ) (s + p1 )(s + p2 ) (s + p1 )(s + p2 )

The output y(t) can be calculated by taking the inverse Laplace transform of Y (s). That is
94 3 Input Output Representation of Linear Systems

Fig. 3.5 The location of the system poles when 0 < ζ < 1.

 
−1 −1 1 s + 2ζ ωn 1/M
y(t) = L {Y (s)} = L ẏ(0) + y(0) + U(s) , (3.28)
(s + p1 )(s + p2 ) (s + p1 )(s + p2 ) (s + p1 )(s + p2 )
     
−1 1 −1 s + 2 ζ ωn 1 −1 1
= ẏ(0)L + y(0)L + L U(s) ,
(s + p1 )(s + p2 ) (s + p1 )(s + p2 ) M (s + p1 )(s + p2 )
(3.29)

The part of this output that depends only on the initial conditions is called the initial condition response, yIC ,
   
1 s + 2ζ ω n
yIC (t) = ẏ(0)L −1 + y(0)L −1 (3.30)
(s + p1 )(s + p2 ) (s + p1 )(s + p2 )

and the last part of the output  

1 −1 1
y f (t) = L U(s) (3.31)
M (s + p1 )(s + p2 )
is called the forced response, since its behavior depends only on the input.

3.2.2.1 Initial Condition Response

Let us look a bit closely at the initial condition response. The partial fraction expansion of a rational function
takes distinct forms depending on whether the poles are distinct or not. Thus we will have to consider these two
cases separately.

All roots of the denominator of Y (s) are distinct

Observe that when ζ 6= 1 the roots −p1 , −p2 of the characteristic polynomial are distinct. In cases I,II, and III
we have that ζ 6= 1. In these cases from partial fraction expansions
3.2 Mathematical Description of a System 95
 
1/(p2 − p1 ) 1/(p1 − p2 )
yIC (t) = ẏ(0)L −1 +
(s + p1 ) (s + p2 )
 
(2 ζ ω n − p 1 )/(p 2 − p1 ) (2ζ ωn − p2 )/(p1 − p2 )
+y(0)L −1 + .
(s + p1 ) (s + p2 )

Explicitly;
   
1 −p1 t 1 −p2 t (2ζ ωn − p1 ) −p1 t (2ζ ωn − p2 ) −p2 t
yIC (t) = ẏ(0) e + e 1(t)+y(0) e + e 1(t).
(p2 − p1 ) (p1 − p2 ) (p2 − p1 ) (p1 − p2 )

Now let us investigate the initial condition response for each of the three cases I,II and III separately.

Over damped: ζ > 1

p p
When ζ > 1 then both
p −p1 , −p2 are real and distinct and p2 − p1 = ωn ζ 2 − 1, 2ζ ωn − p1 = (ζ + ζ 2 − 1)ωn ,
2ζ ωn − p1 = (ζ − ζ 2 − 1)ωn . Then from (3.32)
!
1 1
yIC (t) = ẏ(0) p e−p1 t + p e−p2 t 1(t) +
ωn ζ 2 − 1 −ωn ζ 2 − 1
p p !
(ζ + ζ 2 − 1)ωn −p1 t (ζ − ζ 2 − 1)ωn −p2 t
y(0) p e + p e 1(t).
ωn ζ 2 − 1 −ωn ζ 2 − 1

Under damped: ζ > 1

p 0 < ζ < 1 then −p1 and −p2 are complex conjugate of each other and p2 − p1 = iωd , where ωd =
When
ωn 1 − ζ 2 is the damped natural frequency of the system, 2ζ ωn − p1 = (ζ ωn + iωd ), 2ζ ωn − p1 = (ζ ωn − iωd ).
Then from (3.32)
   
1 −p1 t 1 −p2 t (ζ ωn + iωd ) −p1 t (ζ ωn − iωd ) −p2 t
yIC (t) = ẏ(0) e + e 1(t) + y(0) e + e 1(t).
iωd −iωd iωd −iωd

Observe that the partial fraction expansion coefficients are complex conjugate of each other. Now plugging in
−p1 = −ζ ωn + iωd and −p2 = −ζ ωn − iωd in the above equation we have that
 
1 iωd t 1 −iωd t
yIC (t) = ẏ(0)e−ζ ωn t e + e 1(t)
iωd −iωd
 
(ζ ωn + iωd ) iωd t (ζ ωn − iωd ) −iωd t
+y(0)e−ζ ωn t e + e 1(t). (3.32)
i ωd −iωd

1 −iπ /2 (ζ ωn +iωd ) ζ
Now observing that 1
iωd = ωd e and that iωd = reiφ where r = √ 1 and φ = arctan √1− ζ2
equation
1−ζ 2
(3.32) reduces to
 
ẏ(0) −ζ ωn t −ζ ω n t
yIC (t) = e cos (ωd t − π /2) + y(0)re cos (ωd t + φ ) 1(t). (3.33)
ωd

Undamped: ζ = 0

In this case ζ = 0 and the corresponding initial condition response can be derived from (3.33) by setting ζ = 0
in it. Then ωd = ωn , r = 1, and φ = 0. This gives,
 
ẏ(0)
yIC (t) = cos (ωnt − π /2) + y(0) cos (ωnt) 1(t). (3.34)
ωn
96 3 Input Output Representation of Linear Systems

All roots of the denominator of Y (s) are coinciding

Coinciding roots occur when ζ = 1. That is when the system is critically damped, case IV. In this case let
−p1 = −p2 = −p = −ζ ωn . Therefore the partial fraction expansion yields
   
−1 1 −1 1 (2ζ ωn − p)
yIC (t) = ẏ(0)L + y(0)L + .
(s + p)2 (s + p) (s + p)2

Explicitly;


yIC (t) = ẏ(0) te−pt 1(t) + y(0) e−pt + (2ζ ωn − p) te−pt 1(t),
= e−ζ ωn t (ẏ(0) t + y(0)(1 + ζ ωn t)) 1(t). (3.35)

where the last equality follows from plugging in −p = −ζ ωn .

3.2.2.2 The Forced Response

In this section we will look at the forced response a bit more closely. If we choose all the initial conditions of
the system to be zero, that is y(0) = 0 and ẏ(0) = 0 then the initial condition response is identically equal to zero
and the response of the system is only the forced part of the response.
Most of the inputs of interest to control Engineers result in a Laplace transform that is a proper rational
function of the complex variable s. That is U(s) is of the form

N(s)
U(s) = (3.36)
(s + β1 )(s + β2 ) ∙ ∙ ∙ (s + βk )

where β1 , β2 , ∙ ∙ ∙ βk are real or complex constants and N(s) is a polynomial of the complex variable s. For example
if u(t) = e−3t 1(t) then U(s) = s+3 1
and if u(t) = e−2t cos (5t) 1(t) then U(s) = (s+2) s+2 s+2
2 +52 = (s+2+5 j)(s+2−5 j) . Refer
to table A.1 on page 813 in the text book for more examples on standard inputs and their Laplace transforms.
Substituting (3.36) in (3.31) gives
 
−1 1/M N(s)
y f (t) = L . (3.37)
(s + p1 )(s + p2 ) (s + β1 )(s + β2 ) ∙ ∙ ∙ (s + βk )

To evaluate this we need to take the partial fraction expansion of the right hand side. The form of the partial
fraction expansion differs depending on if the denominator contains repeated factors or not.

All roots of the denominator of Y (s) are distinct

Under this assumption the partial fraction expansion gives

 
−1 −1 −1 1/M N(s)
y f (t) = L {Y (s)} = L {G(s)U(s)} = L
(s + p1 )(s + p2 ) (s + β1 )(s + β2 ) ∙ ∙ ∙ (s + βk )
 
−1 α1 α2 r1 rk
=L + + +∙∙∙+
(s + p1 ) (s + p2 ) (s + β1 ) (s + βk )
1 1 1 1
= α1 L −1 { } + α2 L −1 { } + r1 L −1 { } + ∙ ∙ ∙ + rk L −1 { }
(s + p1 ) (s + p2 ) (s + β1 ) (s + βk )
 
= α1 e−p1 t 1(t) + α2 e−p2 t 1(t) + r1 e−β1 t 1(t) + ∙ ∙ ∙ + rk e−βk t 1(t) . (3.38)

where α1 , r1 , ∙ ∙ ∙ rk are constants determined in the process of partial fraction expansion,


α1 = (s + p1 )Y (s) s=−p1 ,
3.2 Mathematical Description of a System 97

α2 = (s + p2 )Y (s) s=−p2 ,

ri = (s + βi )Y (s) s=−βi .
The forced response given by (3.38) can further be broken up into two parts,

y(t) = ytr (t) + yss (t).

The transient response, ytr (t), is given by,

ytr (t) = α1 e−p1 t + α2 e−p2 t 1(t), (3.39)

and the steady state response, yss (t), is given by,

 
yss (t) = r1 e−β1 t + ∙ ∙ ∙ + rk e−βk t 1(t). (3.40)

To illustrate why we use these names consider the following. If the real parts of −p1 , −p2 are negative (observe
that this is always the case when you have nonzero positive damping, ie ζ > 0) then

lim ytr (t) = lim (α1 e−p1 t + α2 e−p2 t ) 1(t) = 0

t→∞ t→∞

and thus it decays to zero. Hence the name transient response. Also because of this

lim y(t) = lim ytr (t) + lim yss (t) = lim hss (t).
t→∞ t→∞ t→∞ t→∞

and thus yss (t) is the response of the system when t becomes very large or in other words is the steady state
response.
Let us now consider in detail the three cases that correspond to distinct roots. That is when ζ 6= 1. Specifically
in the three cases where ζ > 1 (over damped), 0 < ζ < 1 (under damped) and ζ = 0 (undamped) the transient
response will be given by (3.39). Specifically,

Over damped: ζ > 1

When ζ > 1 then both −p1 , −p2 are real and distinct. Then from (3.39)


ytr (t) = α1 e−p1 t + α2 e−p2 t 1(t), (3.41)

Under damped: 0 < ζ < 1

When 0 < ζ < 1 then −p1 and −p2 are complex conjugate of each other and −p1 = −ζ ωn + iωd , −p2 =
−ζ ωn − iωd . Then from (3.39)

ytr (t) = α1 e−p1 t + α2 e−p2 t 1(t). (3.42)
Recall that α1 and α1 are complex conjugate of each other. Then setting α1 = reiφ , where r = |α1 | and φ is the
angle of the complex number α1 , we will have that

ytr (t) = re−ζ ωn t cos (ωd t + φ ) 1(t). (3.43)

Please refer to section 3.2.2.1 for further similar details of this calculation. Also compare this with the initial
condition response (3.33) and observe the similarity in behavior.

Undamped: ζ = 0

This case is similar to case II above the only difference is that now since ζ = 0 ωd = ωn , and therefore 3.43
gives,
98 3 Input Output Representation of Linear Systems

ytr (t) = cos (ωd t + φ ) 1(t). (3.44)

Of the roots of the denominator of Y (s), Roots −p1 and −p2 coincide but all the others are distinct

Let −p1 = −p2 = −p and −p, −β1 , ∙ ∙ ∙ , −βk are all distinct then the partial fraction expansion gives
 
−1 −1 −1 1/M N(s)
y f (t) = L {Y (s)} = L {G(s)U(s)} = L
(s + p)2 (s + β1 )(s + β2 ) ∙ ∙ ∙ (s + βk )
 
−1 α1 α2 r1 rk
=L + + +∙∙∙+
(s + p) (s + p)2 (s + β1 ) (s + βk )
1 1 1 1
= α1 L {−1
} + α2 L {
−1
} + r1 L −1 { } + ∙ ∙ ∙ + rk L −1 { }
(s + p) (s + p)2 (s + β1 ) (s + βk )
 
= α1 e−pt 1(t) + α2te−pt 1(t) + r1 e−β1 t 1(t) + ∙ ∙ ∙ + rk e−βk t 1(t) . (3.45)

where α1 , r1 , ∙ ∙ ∙ rk are constants determined in the process of partial fraction expansion,

d  
α1 = (s + p)2Y (s) s=−p ,
ds

α2 = (s + p)2Y (s) s=−p ,

ri = (s + βi )Y (s) s=−βi .
The forced response given by (3.45) can further be broken up into two parts,

y f (t) = ytr (t) + yss (t).

The transient response, ytr (t), is given by,

ytr (t) = e−pt (α1 + α2t) 1(t), (3.46)

and the steady state response, yss (t), is given by,

 
yss (t) = r1 e−β1 t + ∙ ∙ ∙ + rk e−βk t 1(t). (3.47)

Critically damped: ζ = 1

In this case −p1 = −p2 = −p = −ζ ωn and then

ytr (t) = e−ζ ωn t (α1 + α2t) 1(t). (3.48)

Compare this with the corresponding initial condition response (3.35) and observe the similarity in the behavior.

3.3 Transfer Function Model of Linear Time Invariant Systems

In the previous two examples of the first order system and the second order system we have seen that the initial
condition response and the transient response both have the same qualitative behaviour. Therefore analysing one
suffices. Therefore we loose no information by neglecting the initial condition response. This can be naturally
done if we set all the initial conditions to zero in our analysis. In that case the response of the system is only
the forced response part. In addition to providing us with a simplification of analysis it also gives us a way of
representing linear time invariant (LTI) systems in a way that will fit into the framework discussed in section 3.2.
3.3 Transfer Function Model of Linear Time Invariant Systems 99

In general the modelling process of a linear nth order time invariant SISO system without time delays will
result in a single nth order linear constant coefficient ODE or a set of such coupled ODEs. For convenience of
introducing ideas lets only consider the former type:

dn y d n−1 y dm y d m−1 y
+ an−1 + ∙ ∙ ∙ + a 0 y = b m + b m−1 + ∙ ∙ ∙ + b0 u, (3.49)
dt n dt n−1 dt m dt m−1
where n ≥ m.
Note that a0 , a1 , ∙ ∙ ∙ an−1 , b0 , b1 , ∙ ∙ ∙ bm are constants and depend only on the system parameters. For a given
u(t) that does not grow faster than an exponential function the output will also not grow faster than an exponential
function (this is a consequence of linearity of the system). Thus taking Laplace transform of both sides of (3.49)
and setting the initial conditions to zero we have
 n   
d y d n−1 y dm u d m−1 u
L + a n−1 + ∙ ∙ ∙ + a 0 y = L b m + b m−1 + ∙ ∙ ∙ + b 0 u , (3.50)
dt n dt n−1 dt m dt m−1
(sn + an−1 sn−1 + ∙ ∙ ∙ + a0 )Y (s) = (bm sm + bm−1 sm−1 + ∙ ∙ ∙ + b0 )U(s) (3.51)

This yields
bm sm + bm−1 sm−1 + ∙ ∙ ∙ + b0
Y (s) = U(s). (3.52)
sn + an−1 sn−1 + ∙ ∙ ∙ + a0
Let
bm sm + bm−1 sm−1 + ∙ ∙ ∙ + b0
G(s) = ,
sn + an−1 sn−1 + ∙ ∙ ∙ + a0
and now we have that
Y (s) = G(s)U(s). (3.53)
The polynomial Δ (s) = sn + an−1 sn−1 + ∙ ∙ ∙ + a0 is the characteristic polynomial of the system and

Δ (s) = sn + an−1 sn−1 + ∙ ∙ ∙ + a0 = 0, (3.54)

is the characteristic equation.

Looking closely at equation (3.53) we see that we have arrived at a representation of LTI systems that fit the
block diagram representation described in section 3.2. Where now what the block does is to simply multiply the
Laplace of the input U(s) by the rational function G(s) to yield the Laplace transform of the output Y (s). The
rational function G(s) is referred to as the transfer function model of the system. Remember that this rational
function is a function of the complex variable s.
Revisiting the two previous examples we see that when we set all initial conditions to zero we obtain for the
first order tank system
1
H(s) = U(s), (3.55)
s + p1
and for the second order spring mass damper system

1/M
Y (s) = U(s). (3.56)
(s + p1 )(s + p2 )

Comparing with equation (3.53) we see that the transfer function model of the single tank system is
1
G(s) = , (3.57)
s + p1
and the transfer function model of the second order SMD system is

1/M
G(s) = . (3.58)
(s + p1 )(s + p2 )
100 3 Input Output Representation of Linear Systems

3.3.1 Forced response of the nth order system

In the absence of the initial conditions y(t) = y f (t). Thus the Laplace transform of the forced response of the
system is what is given by (3.52). For completeness lets look at this in detail.
Let −p1 , −p2 , ∙ ∙ ∙ , −pn be the roots of the characteristic polynomial. Thus

Δ (s) = sn + an−1 sn−1 + ∙ ∙ ∙ + a0 = (s + p1 )(s + p2 ) ∙ ∙ ∙ (s + pn ), (3.59)

In this notation (3.52) becomes

bm sm + bm−1 sm−1 + ∙ ∙ ∙ + b0
Y (s) = U(s). (3.60)
(s + p1 )(s + p2 ) ∙ ∙ ∙ (s + pn )

The output y(t) can be calculated by taking the inverse Laplace transform of Y (s). That is
 
bm sm + bm−1 sm−1 + ∙ ∙ ∙ + b0
y(t) = L −1 {Y (s)} = L U(s) , (3.61)
(s + p1 )(s + p2 ) ∙ ∙ ∙ (s + pn )

Most of the inputs of interest to control Engineers result in a Laplace transform that is a proper rational function
of the complex variable s. That is U(s) is of the form

N(s)
U(s) = (3.62)
(s + β1 )(s + β2 ) ∙ ∙ ∙ (s + βk )

where β1 , β2 , ∙ ∙ ∙ βk are real or complex constants and N(s) is a polynomial of the complex variable s1 . Let
us substitute (3.62) in (3.61) and take the partial fraction expansion of it under the assumption that all
p1 , ∙ ∙ ∙ , pn , β1 , ∙ ∙ ∙ , βk are distinct,
 
bm sm + bm−1 sm−1 + ∙ ∙ ∙ + b0 N(s)
y(t) = L −1 {Y (s)} = L −1 ,
(s + p1 )(s + p2 ) ∙ ∙ ∙ (s + pn ) (s + β1 )(s + β2 ) ∙ ∙ ∙ (s + βk )
( )
n
αi k
ri
=L −1
∑ +∑
i=1 s + pi i=1 s + βi
n k
= ∑ αi e−pit 1(t) + ∑ ri e−βit 1(t). (3.63)
i=1 i=1

The constants α1 , α2 , ∙ ∙ ∙ , αn and r1 , rk , ∙ ∙ ∙ , rk are determined in the process of partial fraction expansion as
follows, 
αi = (s + pi )Y (s) s=−pi ,

ri = (s + βi )Y (s) s=−βi ,
Let
y(t) = ytr (t) + yss (t),
where
n
ytr (t) = ∑ αi e−pi t 1(t),
i=1

is the transient response of the system and

n
yss (t) = ∑ ri e−βi t 1(t),
i=1

is the steady state response of the system.

1 For example if u(t) = e−3t 1(t) then U(s) = 1

s+3 and if u(t) = e−2t cos (5t) 1(t) then U(s) = s+2
(s+2)2 +52
= s+2
(s+2+5 j)(s+2−5 j) .
3.5 Bounded Input Bounded Output Stability 101

3.4 Poles and Zeros of an Input Output System

The transfer functions of all real physical systems are strictly proper or proper rational functions. The poles are
the poles of it’s transfer function and the zeros of the system are the zeros of it’s transfer function. Recall that by
definition the number of zeros and number of poles of a transfer function are the same2 .
Thus the system given by the transfer function

bm sm + bm−1 sm−1 + ∙ ∙ ∙ + b0 bm sm + bm−1 sm−1 + ∙ ∙ ∙ + b0

G(s) = = ,
sn + an−1 sn−1 + ∙ ∙ ∙ + a0 (s + p1 )(s + p2 ) ∙ ∙ ∙ (s + pn )

where n ≥ m has n poles and zeros. The poles are −p1 , −p2 , ∙ ∙ ∙ , −pn and the system has m finite zeros, since
the numerator polynomial of G(s) is of order m. The remaining n − m zeros are at infinity.
For example in the case of the single tank system,
1
G(s) = ,
s + p1
the system has one pole and one zero. The pole is finite and has value −p1 while the zero is infinite. For the
spring mass damper system,
1/M
G(s) = ,
(s + p1 )(s + p2 )
the system has two poles and two zeros. The poles are finite and have values −p1 , −p2 while the two zeros are
at infinity.

3.4.1 The Effect of the Poles of the System

Observe that if all the poles of the system −p1 , −p2 , ∙ ∙ ∙ , −pn are in the strict left half complex plane, then

lim ytr (t) = 0,

t→∞

and
lim y(t) = lim yss (t).
t→∞ t→∞

On the other hand if any one of the −pi are in the right half complex plane then

lim ytr (t) = ∞,

t→∞

and
lim y(t) = ∞.
t→∞

(For instance if −p3 = 2 − 3 j, then limt→∞ e−p3 t = limt→∞ e(2−3 j)t = limt→∞ e2t e−3 jt = ∞.)
Thus the long term behavior (the behavior as limt → ∞) of the transient response depends entirely on the
location of the poles the system. If any of the poles of the system are in the right half plane then the transient
response of the system will grow unboundedly as limt → ∞.

3.5 Bounded Input Bounded Output Stability

If u(t) is bounded then all −β1 , −β2 , ∙ ∙ ∙ , −βk are in the strict left half complex plane or on the imaginary axis
(repeated values on the imaginary axis are not allowed). For example u(t) = 1(t) is bounded U(s) = 1s , u(t) =

2 See note by Benjamin C. Kuo and Farid Golnaraghi

102 3 Input Output Representation of Linear Systems

sin (ω t) 1(t) is bounded and U(s) = s2 +ωω 2 , u(t) = e−t 1(t) is bounded and U(s) = s+1
1
. Note that u(t) = t 1(t) is
1
not bounded and U(s) = s2 , thus repeated poles of U(s) on the imaginary axis do not occur if u(t) is bounded.
Now consider what happens when a bounded input is applied to a system. For simplicity assume all
p1 , p2 , ∙ ∙ ∙ , pn , β1 , β2 , ∙ ∙ ∙ , βk are distinct. Then from (3.63)
n k
y(t) = ytr (t) + yss (t) = ∑ αi e−pi t 1(t) + ∑ ri e−βi t 1(t).
i=1 i=1

Since all −β1 , −β2 , ∙ ∙ ∙ , −βk are on either the strict left half complex plane or on the imaginary axis with no
repeated values on the imaginary axis the last k terms
k
∑ ri e−βit 1(t) = yss (t),
i=1

remains bounded as limt → ∞. Thus the whole response will remain bounded only if the transient response,
n
∑ αi e−pit 1(t) = ytr (t)
i=1

remains bounded. This will happen if all −p1 , −p2 , ∙ ∙ ∙ , −pn lie in the left half complex plane.
Therefore if the system response to any given bounded input is bounded then the poles of the system,
−p1 , −p2 , ∙ ∙ ∙ , −pn , will necessarily have to be in the strict left half complex plane. This notion of stability
is called Bounded Input Bounded Output (BIBO) stability.
Definition 3.1. A system is said to be Bounded Input Bounded Output (BIBO) stable if the output of the system
to all bounded inputs remain bounded.
Theorem 3.1. A system is BIBO stable if and only if all the poles of the system lie in the strict left half complex
plane.
s−1 s−1
Examples: The system G(s) = (s+1)(s+2)(s+7) is BIBO stable while the systems G(s) = (s+1)(s−2)(s+7) , G(s) =
s+1 1
s(s+1)(s+2) , G(s) = (s2 +1)
are not BIBO stable.

3.6 Exercises

Exercise 3.1. Identify the, system parameters, system variables, inputs and possible outputs of: a) a forced spring
mass damper system. b) a heat Exchanger. c) a water tank system.
Exercise 3.2. Consider the spring mass damper system shown in figure 2.17 with the external force f (t) = sin ω t
and M = 1. The governing equations of this system in the normal form is given by

ẍ + 2ζ ωn ẋ + ωn2 x = sin ω t.

1. Find analytically x(t) for initial conditions x(0) = 0 and ẋ(0) = 0.

2. Using the above results show that

lim x(t) = M(ω ) sin (ω t + φ (ω ))

t→∞

where  
1 2ζ ωn ω
M(ω ) = p , φ (ω ) = tan −1
(ωn2 − ω 2 )2 + 4ζ 2 ωn2 ω 2 ωn2 − ω 2
3. For ωn = 1 plot M(ω ) Vs ω and φ (ω ) Vs ω for the three cases,
(a) Over damped system (1 > ζ )
(b) Under damped system (0 < ζ < 1)
(c) Critically damped system (ζ = 1)
3.6 Exercises 103

4. Numerically simulating the system for ωn = 1, and ζ = 0.05, plot the response x(t) for each of the cases
ω = 0.01, 0.1, 0.9, 1, 1.1, 10, 100. Compare your results with the graph obtained 3(b) above.

Exercise 3.3. Find the transfer function model of the system system shown in figure 3.6 from input u(t) = f (t)
to output y(t) = x2 (t).

Fig. 3.6 A 2-DOF spring mass damper system. Assume that the damping force fv is proportional to the separation velocity with
proportionality constant C.