www.EasyEngineering.

net

9 wState Space Analysis of

w w.E Systems
Control
asy
9.1 Introduction
En
gin
Mathematical modelling of a system plays an important role in the analysis and design of control
systems. Transfer function is one such model, which we have used for analysis and design of control

eer
systems in the previous chapters. This is a useful representation if the system is linear, time invariant
and has a single input and single output (SISO). It is also defined for systems with zero initial

in
conditions only. The tools developed, viz, root locus technique, Bode plot, Nyquist plot, Nichol's

g.n
chart etc are powerful in the analysis and design of control systems. But transfer function representation
is not useful for,
1.
2.
3.
Systems with initial conditions
Nonlinear systems
Time varying systems and
et
4. Multiple input multiple output (MIMO) systems
further, the output for a given input can only be found. It does not throw any light on the variation
of internal variables. Sometimes this information is necessary because, some internal variables may
go out of bounds, eventhough the output remains within the desired limits. The methods discussed
so far are known as classical methods. In this method, the output only is fedback to obtain the
desired performance of the system. This may not result in the best or optimum performance of the
system. It may be desirable to feedback additional internal variables to achieve better results. The
design procedures in classical theory are mostly trial and error procedures.
A need for a represenation which overcomes all the above draw backs was felt and the state space
representation of the system was evolved. This representation forms the basis for the development of
modern control systems. This representation contains the information about some of the internal
variables along with the output variable and is amenable for analysis and design using digital computer.
It is suitable for representing linear, nonlinear, time invariant, time varying, SISO and MIMO systems.

www.EasyEngineering.net
 www.EasyEngineering.net

State Space Analysis of Control Systems 319

9.2 State Variables

Consider an RLC network excited by an input vet) as shown in Fig. 9.1.

L i(t)
r>r>nC">~

v(t)

Fig. 9.1 An RLC circuit

ww
The dynamic behaviour of this circuit can be understood by considering the loop equation,

w.E di 1
Ri + L dt + C
t
Ii dt = vet) ..... (9.1)

asy -00

Differentiating eqn. (9.1) we get a second order differential equation,

d 2i di i
E
L dt 2 + R dt + C =
dv

ngi
dt ..... (9.2)

nee di
The solution of eqn. (9.2) requires two initial conditions, namely, i(o) and dt (0). i(o) is the

current throught the inductor at t = 0 and from eqn. (9.1) with t = 0, we have,
rin
v(O) - Ri(O) _l fi dt
o
g.n
o
di
dt (0) = L
C -00

d'
et
The quantities ~ Ii dt is the voltage across the capacitor at t = 0 and hence d: (0) is dependent
-00

on the inital voltage across the capacitor in addition to v(O) and i(O). Thus, if we know vet) for t ~ 0,
i(O), the current through the inductor at t = 0 and vc(O), the voltage across the capacitor at t = 0, the
dynamic response of the system can be easily evaluated. Inductor and the capacitor are the two
energy storing elements in the network which are responsible for the behaviour of the network
alongwith the input. Thus we can treat the current through the inductor and voltage across the
capacitor, as the characterising variables of the network. If these variable are known at any time
t = to' the network response can be easily found out. Hence these two variables describe the ~te of
the network at any time t and these are the minimum number of variables that should be known at
t = to to obtain the dynamic response of the network.

www.EasyEngineering.net
 www.EasyEngineering.net

320 Control Systems

Now we can define the state and state variables as :

The minimum number of variables required to be known at time t = to alongwith the input for
t 2: 0, to completely determine the dynamic response of a system for t > to,are known as the state
variables of the system. The state ofthe system at any time 't' is given by the values of these variables
at time 't'.
If the dynamic behaviour of a system can be described by an nth order differential equation, we
require n initial conditions of the system and hence, a minimum of n state variables are required to be
known at t = to to completely determine the behaviour of the system to a given input. It is a standard
practice to denote these n state variables by xl(t), xit) ..... ~(t) and m inputs by ul(t), l1:2(t), .....
um(t) and p outputs by YI(t), Y2(t) ..... yp(t).
The system is described by n first order differential equations in these state variables:

ww d~l = Xl = fl (XI' x2 ... ~; u I' 11:2 ... urn' t) ..... (9.3)

w.E dx
.
...
·•
..
.
dtn =
asy Xn = fn (xl' x2 ... ~; ul' 11:2 ... urn' t)
The functions f l _f2 ... fn may be time varying or time invariant and linear or nonlinear in nature.

E
Using vector notation to represent the states, their derivatives, and inputs as :

ngi
nee ..... (9.4)

rin
where X(t) is known as state vector and U(t) is known as input vector. We can wrie eqns. (9.3) g.n
in a compact form as,
X(t) = f [X(t), U(t),t] et ..... (9.5)

f1 [X(t), u(t), t]1

where f [X(t), U(t),t] = ~2 [X(t), u(t), t]
[
fn [X(t), u(t), t]

If the functions f are independent of t, the system is a time invariant system and eqn. (9.5) is
written as,
x(t) f [X(t), U(t)]
= ..... (9.6)
The outputs YI' Y2 ... yp may be dependent on the state vector X(t) and input vector U(t) and may
be written as,
yet) = g [X(t), U(t)] ..... (9.7)

www.EasyEngineering.net
 www.EasyEngineering.net

State Space Analysis of Control Systems 321

YI (t)
Y2 (t)
where yet) =

is known as output vector.

For a single input single output systems (SISO) U(t) and yet) are scalars. Once the system state is
known at any time t, the output can be easily found out since eqn. (9.7) is only algebraic equation and
not a dynamic relation.

9.3
ww
State Equations for Linear Systems
For linear systems, in eqn. (9.3), the derivatives of state variables can be expressed as linear

w.E
combinations of the state variables and inputs.
XI = all xI + a l2 x 2 + .,. + a ln ~ + b ll u l + b l2 u 2 + ... + b lm urn

asy
x2 = azl XI + az2 x2 + .,. + azn xn + b21 u l + b22 u2 + ... + b2m urn

E ngi
Xn = ~I xI + an2 x 2 + .,. + ann xn + bnl u l + b n2 u 2 + ... + b nm u m.... ·(9.8)

a;/ and bz/ are constants.

where
Eqns. (9.8) can be written in a matrix form as, nee
X (t) = A X (t) + BU (t)
rin ..... (9.9)
where X(t) is a n
A is a n
x 1 state vecotr
n constant system matrix g.n
et
x

B is a n x m constant input matrix

U(t) is a m x 1 input vector.

a l2 ... ll b12 . .. lm
[an ... _ [b ...
A= .
a 21 a 22 al'l
a 2n .
··
·
b21
, B - ..
.
b 22 b 2m
b 1
ani a n2 ann b nl b n2 b nm

Similarly, the outputs can also be expressed as linear combinations of state variables and inputs as,
YI(t) = C ll xI + C l2 X2 .. , + C ln Xn + d ll UI + d l2 U2 + ... + dim urn

www.EasyEngineering.net
 www.EasyEngineering.net

322 Control Systems

Eqns. (9 -10) can be written compactly as,

yet) = C X(t) + D U(t) ..... (9.11)
Where Y (t) is a p x 1 output vector
C is a p x n output matrix
D is a p x m transmission matrix

C= .
c pl[
~ll ... ~ln 1
.
C
[~ll
. D= .
pn ' d pl
. . ~:1
The complete state model of the linear system is given by eqn. (9.9) are (9.11).

ww X(t) = A X(t) + BU (t)

C X(t) + DU (t)
..... (9.12)

w.E yet) = ..... (9.13)

Eqn. (9.12) is known as the state equation,

asy
and eqn. (9.13) is known as the output equation.
For a single input single output system

yet) =
E
X(t) = A X(t) + bu
C X(t) + du
ngi
.... (9.14)
..... (9.15)

nee
Where band dare (n x 1) and (p x 1) vectors respectively. u(t) is the single input and yet) is the
single output. In most of the control systems, the output is not directly coupled to the input and hence
yet) is not dependent on u(t). Hence eqn. (9.15) is written as,
yet) = C X(t) rin ..... (9.16)

g.n
For time invariant systems, the matrices A, B, C and D are constant matrices. For time varying
systems, the elements of A, B, C and D matrices are functions of time. In this book we will be
concerned with linear, time invariant, single input single output systems only.
et
State variable representation of a system is not unique. For a given system we may define different
sets of variables to describe the behaviour of the system. In all such different representation the
number of state variables required are the same, and this number is known as the order of the system.
Let us consider the RLC circuit again, shown in Fig. 9.1. If the current, i, through the inductor and
voltage, vC' across the capacitor are taken as state variables, we have,
dVe
C dt = i(t) ..... (9.17)

di
L dt = vL(t) = v - i R - ve ..... (9.18)

Defining ve(t) ~ xl(t)

i(t) ~ x 2(t)

vet) ~ u(t)

www.EasyEngineering.net
 www.EasyEngineering.net

State Space Analysis of Control Systems 323

1
We have x\ (t) = C x2 (t) ..... (9.19)

1 R 1
x2 (t) = - L x\(t) - L xit ) + L u(t) ..... (9.20)
In matrix form, eqns. (9.19), (9.20) can be written as,

..... (9.21)

or
ww X(t) = A X(t) + b u(t)

where w.E ..... (9.22)

asy
Now, consider the eqn. (9.1) for the RLC circuit,

E
di 1
Ri + L - + -
dt C
t
Sidt = v
-co ngi ..... (9.23)

The charge q is given by,

t
nee
q(t) = Sidt
rin ..... (9.24)

. dq
-co

g.n
and 1= -
dt
In terms of the variable q, eqn. (9.23) can be written as
d 2q di q
et
..... (9.25)

L -2+ R - + - = v ..... (9.26)

dt dt c
Now if we defme the state variables as:
x\(t) = q(t) ..... (9.27)

and x 2(t) = ~; = i(t) = x\ (t) ..... (9.28)

Thus x\ (t) = x2 ..... (9.29)

d 2q q
x2 (t) = dt 2 LC

..... (9.30)

www.EasyEngineering.net
 www.EasyEngineering.net

324 Control Systems

Eqns. (9.29) and (9.30) can be put in matrix form as,

XI(t)]
[ X2(t) [°1 -r::-lR]
= - LC
[XI]
x2 +
[~l
L u(t)
..... (9.31)

Eqns (9.21) and (9.31) give two different representations of the same system in state variable
form. Thus, the state variable representation of any system is not unique.

9.4 Canonical Forms of State Models of Linear Systems

Since the state variable representation of a system is not unique, we will have infinite ways of choosing
the state variables. These different state variables are uniquely related to each other. If a new set of
state variables, are chosen as a linear combination of the given state variables X, we have

ww X=PZ ..... (9.32)

w.E
where P is a nonsingular n x
Z=P- 1 X.
n constant matrix, so that
. .... (9.33)
From eqn. (9.32),
asy
x =P Z =AX+Bu
E
=APZ + Bu
ngi
Z=AZ+13u.
nee
Eqn. (9.34) is the representation of the same system in terms of new state variables Z and
. .... (9.34)

A =p- 1 AP
rin
and 13 = p- 1 B.
g.n
Since P is a non singular matrix, p- 1 exists. Now let us consider some standard or canonical forms
of state models for a given system.

9.4.1 Phase Variable Form

et
When one of the variables in the physical system and its derivates are chosen as state variables,
the state model obtained is known to be in the phase variable form. The state variables are themselves
known as phase variables. Usually the output of the system and its derivates are chosen as state
variables. We will derive the state model when the system is described either in differential equation
form or in transfer function form.
A general nth order differential equation is given by
(n) (n-I) . _ (m) (m-I) b
y + a 1 y + ... + ~ _ 1 Y + ~y - b o u + b l u + ... + m-I Ii + bmu ..... (9.35)
Where aj S and bJ S are constants and m and n are integers with n 2. m.
(n) LlA _
dny
_ an
d (m) L\ -
dmu
-
Y = dt n u = dt m

www.EasyEngineering.net
 www.EasyEngineering.net

State Space Analysis of Control Systems 325

The n initial conditions are y(o), Y(0), ... , (n~l) (0). If, for the present, we assume the initial
conditions to be zero, we can obtain the transfer function of the system from eqn. (9.35) as,
Y(s) bos m + bls m- I -+- ... + bm_Is + b m
T(s) = U() = n n-I ..... (9.36)
s s +als + ... +an_Is+a n
The initial conditions can be taken care of, after obtaining the state space representation.
Case (a) m= 0
It the derivatives of the inputs are not present in eqn. (9.35) and it can be written as,
(n) (n-I)
y + a l y + ... + ~ _ I Y + ~y = bou ..... (9.37)

ww
Let us defme the state variables as,
xI =y

w.E "-2=x l
x3 = x2 = Y
=y

asy .

E
(n-I)
~=xn_l= y ..... (9.38)

and xn =
(n) ngi
From eqn. (9.37) we have,
y
nee
(n-I) n-2 .
X n = bou - a l y - ~ y ... - ~ _ 1 Y - an Y
rin
bou - a l ~ - ~ ~ _ 1 ..• - an _ I x2 - ~ xI
=
g.n
The above equations can be written in a matrix form as,

o
o o
o
et
..... (9.39)
o 0

In vector matrix notation we have,

X =AX+bu ..... (9.40)

In this equation, we can observe that the system matrix A is in a special form. the diagonal above
the main diagonal of the matrix contains alIi S and the last row contains the negatives of the coefficients,
elj , of the differential equation. All other elements are zeros. Such a form of the matrix is known as
companion form or Bush's form. Similarly, the vector b contains all elements to be zero except the
last one. Hence, in view of these observations, the state space equations given by eqn. (9.39) can be
written down directly from the differential eqn. (9.37).

www.EasyEngineering.net
 www.EasyEngineering.net

326 Control Systems

The output y is equal to xI and hence the output equation is given by,
y=CX ..... (9.41)
where c = [1 0 0 ..... 0]
If the transfer function is known instead of the differential equation, we can easily obtain a differential
form as shown below.
For m = 0, we have from eqn. (9.36)

Yes) bo
T(s) = U() = n n-I ..... (9.42)
S S +a1s + .... +a n

ww (Sn + a l Sn-I + ... + an) YeS) = bo U(s)

or
w.E (n)
y (t) + a l
(n-I)
Y + ... + an y = bou ..... (9.43)

asy
Eqn. (9.43) is the same as eqn. (9.37) and hence the state space model is again given by
eqns. (9.39) and (9.41). A block diagram of the state model in eqn. (9.39) is shown in Fig. (9.2).

state variable. E
Each block in the forward path represents an integration and the output of each integrator is takne a

ngi
nee
--,- J

rin
~ g.n
et
I
\ I
... ----~----"

L..------~IIuf---------l

Fig. 9.2 Block diagram representation of eqn. (9.39).

We can also represent eqn. (9.39) in signal low graph representation as shown in Fig. (9.3).

y
? 0

Fig. 9.3 Signal flow graph representation of eqn. (9.39).

www.EasyEngineering.net
 www.EasyEngineering.net

State Space Analysis of Control Systems 327

If we have non zero initial conditions, the initial conditions can be related to the initial conditions
on the state variables. Thus, from eqn. (9.38).
Xl (0) = Y (0)
x2 (0) = y (0)

(n-l)
"n (0) = y (0) ..... (9.44)

Example 9.1
Obtain the phase variable state model for the system

ww
Solution
y+2y+3y+y=u

w.E
Here n = 3 and a l = 2, az = 3, a3 = 1 and b = 1. Hence the state model can be directly written down as,

asy[~ ~ ~] + [~] u
[ ::] =

E
[::]
X3 -1 - 3 - 2 X3 1
X(O) = [yeO), ngi
y (0), Y(O)]T
and y = [1 0 O]X
nee
Example 9.2
rin
Obtain the companion form of state model for the system whose transfer function is given by,
Yes) 2 g.n
Solution:
Case a :
T(s) = -U-(s-) = --'s3=-+-s"""2-+-2-s-+-3
et
The state model in companion form can be directly written down as,

x=[~ 0
-3 -2 -1
~]X+[~]U 2
Y = [1 0 O]X
Case b: m:;t:O
Let us consider a general case where m = n.
The differential equation is given by
(n) (n-I) n-2 n-I (n) (n-I)
y +a l y +az Y +"'+~-l Y +any=bo u +b l u + ... +bnu .....(9.45)

www.EasyEngineering.net
 www.EasyEngineering.net

328 Control Systems

The transfer function of the system represented by eqn. (9.45) is,

n1
Yes) b o sn + b1s - + ... + bn-1s + b n
T(s) = U( ) = n n-l ..... (9.46)
S S + a1s + ... + a n-l S + an

Eqn. (9.46) can be written as

where ..... (9.47)

and
ww Yes) _ n
Y (s) - bOs + b i s
n--I
+ ... + bn ..... (9.48)

w.E 1
Eqn. (9.47) is same as eqn. (9.42) with b =1.

a
Hence its state space representation is given by eqn. 9.39.

syE 0

[~:l=
0 1 0 0 0

ngi
nee [}l
0 0 0 0 0
u ..... (9.49)

0 0 0 0 1 0
-an -an-l -a n-2
rin
-a 2 -a1

and YI= xl
g.
The signal flow graph of this system is the same as in Fig. (9.3) with y = YI and b o = 1.
net
..... (9.50)

From eqn. (9.48), we have

Yes) = (b o sn + blsn - I + ... + bn) YI(s)

(n) (n-l)
or Y = b o Y1 + b i Y 1 + .. , + b n _ I Y1+ bn YI
From eqn. (9.49) and (9.50), we have

Y = bo xn + b i ~ + ... + bn _ I x2 + bn xl

Substistuting for x n from eqn. (9.49), we have

Y = b o (-an xl - an _ I x2 + .... - ~ ~-I - a l ~ + u) +
b i ~ + ... + bn _ I X2 + bn Xl
Y = Xl (b n - bo an) + X2 (b n _ I - bo an_I)
+ ... + ~ (b i - bo a I) + bou ..... (9.51)

www.EasyEngineering.net
 www.EasyEngineering.net

State Space Analysis of Control Systems 329

..... (9.52)

The signal flow graph of the system of eqn. (9.45) is obtained by modifying signal flow graph
shown in Fig. (9.3) as,

ww U(s)

w.E
asy
E ngi
Fig. 9.4 Signal flow graph of system of eqn. (9.45)

y = [bn bn _ 1 ... bd X nee

For the more common case, where m ~ n - 1 in eqn. (9.35), bo = 0 and eqn. (9.52) can be written as
..... (9.53)

rin
Where b! are the coefficients of the numerator polynomical of eqn. (9.46). In this case, the state

g.n
space representation of eqn. (9.35) can be written down by inspection.
Example 9.3

y + 2 Y+ 3 y + 6 y = ii - U + 2u
et
Obtain the state space representation of the system whose differential equation is given by,

Also draw the signal flow graph for the system.

Solution
In the given differential equation,
al= 2, ~ = 3, a3 = 6
and bo = 0, b l = 1, b2 = -1 and b3 = 2
Substituting in eqns. (9.49) and (9.52), we have

X {~6 -~3 -~J Xm + u

y = [2 -1 1] X

www.EasyEngineering.net
 www.EasyEngineering.net

330 Control Systems

The signal flow graph of the system is given in Fig. (9.5).

u y

ww
-6

Fig. 9.5 Signal flow g'~ for Ex. (9.3)

w.E '~

Phase variable representation is a simple method of obtaining the state space representation of a

a
system. This plays a very important role in the design of a control systems in state space. But this

syE
representation is not useful in practice, as these phase variables often do not represent physical
variables and hence are not available for measurement or control. They are given by the output and its

ngi
derivatives. Higher order derivatives of the output are difficult to obtain in practice. Hence let us
consider another more useful representation of state model of a system.

9.4.2 Diagonal Form

nee
This form is also known as canonical variable form or normal form. The system matrix A in this case

rin
is obtained as a diagonal matrix. Let us consider the transfer function of the system given by
n nI
Yes) bos + bls - + ... + bn_Is + b n
T(s) = U() = n n-I g.n ..... (9.54)
S S + als + ... + an-Is + an
et
Case a : All the poles ofT(s) are distinct and given by -PI' -P2' ... - Pn' Expanding T(s) in partial
fractions, we have,
Yes) n k·
T(s) = - - = b + :E _1_ ..... (9.55)
U(s) 0 i=l S + Pi

Eqn. (9.55) can be represented by a block diagram in Fig. (9.6) (a) and signal flow graph in
Fig. (9.6) (b).
Defming the output of each integrator as a state variable, as shown in Figs. (9.6) (a), (b) we have,
XI =u-PI Xl
X2 =u-P2 X2 ..... (9.56)

Xn = U'- Pn Xn
y = kl 'Xl + k2 X2 + ... + ~ ~ + bO U ..... (9.57)

www.EasyEngineering.net
 www.EasyEngineering.net

State Space Analysis of Control Systems 331

r---------~~bor-------------

+
U(s)

- - - - - - - - - ..... --- /'

/'

ww
w.E
asy
Fig. 9.6 (a) Block diagram representation of eqn. (9.55).

E ngi
nee
rin
g.n
Fig. 9.6 (b) Signal flow graph of eqn. (9.55).
et
Expressing eqn. (9.56) and (9.57) in matrix form.

..... (9.58)

..... (9.59)

www.EasyEngineering.net
 www.EasyEngineering.net

332 Control Systems

In eqn. (9.58) we observe that the system matrix A is in diagonal form with poles of T(s) as its
diagonal elements. Also observe that the column vector b has all its elements as IS. The n equations
represented by eqn. (9.58) are independent of each other and can be solved independently. These
equations are said to be decoupled. This feature is an important property of this normal form which
is useful in the analysis and design of control systems in state variable form. It is also pertinent to
mention that the canonical variables are also not physical variables and hence not available for
measurement or control.
Example 9.4
Obtain the normal form of state model for the system whose transfer function is given by
Yes) s+1

ww
T(s)- ----- ..... (9.60)
U(s) s(s + 2)(s + 4)
Solution

w.E
T(s) can be expanded in partial fractions as,
I 1 3

asy
T(s) == 8s + 4(s + 2) - 8(s + 4) ..... (9.61)

E
The state space representation is given by,

ngi
nee
[i ±-i] rin
Y==
Case b: Some poles of T(s) in eqn. (9.54) are repeated.
X
g.n
Let us illustrate this case by an example.
Yes) s
et
..... (9.62)
T(s) == U(s) = (s + 1)2(s + 2)
The partial fraction expansion is given by,
2 1 2
T(s) == s + 1 - (s + 1)2 - s + 2 ..... (9.63)

The simulation of this transfer function by block diagram is shown in Fig. (9.7).
Defining the output of each integrator in Fig. (9.7) as a state variable, we have
Xl == x2 - Xl

X2 == U-X2
X3 == u- 2x3 ..... (9.64)
and Y == -Xl +. 2x2 - 2x3 ..... (9.65)

www.EasyEngineering.net
 www.EasyEngineering.net

State Space Analysis of Control Systems 333

U(s)

Fig. 9.7 Block diagram representation.

ww
Eqns. (9.64) and (9.65) can be put in matrix form as,
1
w.E X
. = [-0 1
0
~ ~ 1X [~]
0 -2
+
I
u ..... (9.66)

asy
Y = [-1 2 -2] X
The above procedure can be generalised to a system with more repeated poles. Let Al be repeated
..... (9.67)

E
twice, A2 be repeated thrice and A3 and A4 be distinct in a system with n = 7. The state model for this

ngi
I
system will be,

XI
x2
Al
0 Al
0
0
0
0
0
0 nee
0
0
0
0 xXl2
0
I

X3
x
0 0 A2 0 0
rin
0 X3 0

+ 0 u
X5
4
0
0
0
0
0
0
A2
0 A2
0
0
0
0
Ix 4
g.n ..... (9.68)

X6
X7
0
.0
0
0
0
0
0
0
0
0
A3
0
0
A4
lX'
X6
x7
Matrix eqn. (9.68) is partitioned as shown by the dotted lines and it may be represented by
et
eqn. (9.69).

[~l[~ J~l [~} [~ l

0 0
J2 0
u ..... (9.6?)
0 J3
0 0
Eqn. (9.69) is in diagonal form. The sub matrices J I and J2 contain the repeated pies Al and A.2
respectively on their diagonals and the super diagonal elements are all ones. J I and J2 are known as
Jordon blocks. The matrix A itself is known to be of Jordon form. J I is a Jordon block of order
2 and J2 is a Jordon block of order 3. J 3 and J4 corresponding to non repeated pies A3 and A4 are said
to be of order 1. The column vectors b l and b2 have all zero elements except the last element which
is a 1. b 3 and b4 are both unity and correspond to non repeated roots A3 and A4.
Eqn. (9.68 )is said to be the Jordon form representation of a system.

www.EasyEngineering.net
 www.EasyEngineering.net

334 Control Systems

9.5 Transfer Function from State Model

From a given transfer function we have obtained different state space models. Now, let us consider
how a transfer function can be obtained from a giveh state space model. If the state model is in phase
variable form, the transfer function can be directly written down by inspection in view of eqns.
(9.39) and'(9.42) or eqns. (9.46), (9.49) and (9.52).
Alternatively, the signal flow graph can be obtained from the state model and Mason's gain formula
can be used to get the transfer function.
Example 9.5
Obtain the transfer function for the system,

ww [=:] [~ ~ ~] [:~] [~l

= + u

w.E ° X3

Y = [1
-1 - 2 - 3

0] X
X3 1

asy
The transfer function is of the form, for n = 3

E
3 2
bos + b 1s + b 2s + b 3
T(s) -
ngi
- S3 + a 1s 2 + a2s + a 3
From the matrix A, b, and C we have
a l = 3, ~= 2, a3 = 1 nee
bo = 0, b l = 0, b2 = 0, b 3 = 1
rin
Hence 2
1
T(s) - -=-----c:-----
-s3+3s +2s+1 g.n
Example 9.6
Obtain the transfer function of the system
et
x=[~ ~ ~]x+[~]u
-2 -4 -6 1
Y = [1 -2 3] X + 2u
Here a l =6
b o= 2
bl - bo at = 3
b3 = 1 + 2(2) = 5
b2 = - 2 + 2(4) = 6
bl = 3 + 2(6) = 15

www.EasyEngineering.net
 www.EasyEngineering.net

State Space Analysis of Control Systems 335

2s 3 +15s 2 +6s+5
T(s) = 2
S3 +6s +4s+2

If the state model is in Jordon form, we can obtain the transfer function as shown in Ex 9.7.
Example 9.7
Obtain the transfer function for the system

-1 1 °"j [0]
X= [°° -°l O-2X + °1 u

ww Y = [-1 3 3] X
Solution
w.E
From the system matrix A, the three poles of the transfer function are -1, -1 and -2. The residues at

-1 asy
these poles are -1, 3 and 3 (from the matrix C). The transfer function is, therefore, given by
3 3
T(s) = - +
E +--
s+l (s+1)2 s+2
ngi
2S2 + 6s + 7
(s + 1)2(s + 2) nee
rin
If the matrix A is not in either companion form or Jordon form, the transfer function can be
derived as follows.

X =AX+bu g.n ..... (9.70)

Y=CX+du
Taking Laplace transform of eqn. (9.70), assuming zero initial conditions, we have
et
..... (9.71)

s Xes) = A Xes) + b U(s)

(sl - A) Xes) = b U(s)
Xes) = (sl - Ar! b U(s) ..... (9.72)
Taking Laplace transform of eqn. (9.71) and substituting for Xes) from eqn. (9.72), we get
yes) = C (sl - Ar! b U(s) + d U(s) ..... (9.73)
= [C (sl -Ar! b + d] U(s)
T(s) = C (sl - Ar! b + d
If d is equal to zero, for a commonly occuring case,
T(s) = C (sl -Ar! b ..... (9.74)

www.EasyEngineering.net
 www.EasyEngineering.net

336 Control Systems

Since a transfer function representation is unique for a given system, eqn. (9.74) is independent of
the form of A. For a given system, different system matrices may be obtained, but the transfer
function will be unique.
-I Adj(5I - A)
Since (51 - A) = lsi - A I
the roots of lsI -Ai = ° ..... (9.75)
are the poles of the transfer function T(s) and eqn. (9.75) is known as the characteristic equation of
the matrix A.
Example 9.8

ww
Obtain the transfer function for the system,

w.E [-1° ° -1] [0]

X= -1 1 X + 1 u

as°y Y = [1
1 -2 -3

1] X
1

Solution En
° gi
(sl - A) =
s+1
° s+1 -1
1 : nee
[
-1 2 s+3
rin
g.n
2
-(s+l)l
l
(sI-Ar = s3+5s 2 +10s+6
S2 + 4s + 4
- 2(s + 1) et
s+l
(s + 1)2

The transfer function is given by,

S2 +4s + 5 2
-(s+1)l [0]
T(s) = _1_ [1 0 1] 1 S2 +4s+4 s +1 1
L1(s) [ 1
s+ - 2(s + 1) (s+1)2 1

Where L1(s) = s3 + 5s2 + lOs + 6

T(s) = _1_ [1
L1(s)
° 1]
[
1-S
52 +5s+5
]

(s + 1)(s -1)

s(s -1)

www.EasyEngineering.net
 www.EasyEngineering.net

State Space Analysis of Control Systems 337

9.6 Diagonalisation
The state model in the diagonal or Jordon form is very useful in understanding the system properties
and evaluating its response for any given input. But a state space model obtained by considering the
real physical variables as state variables is seldom in the canonical form. These physical variables can
be readily measured and can be used for controlling the response of the system. Hence a state model
obtained, based on physical variables, is often converted to canonical form and the properties of the
system are studied. Hence we shall consider the techniques used for converting a general state space
model to a Jordon form model. Consider a system with the state space model as,

X =AX + bu ..... (9.76)

y = CX + du ..... (9.77)

ww
Let us define a new set of state variables 'Z', related to the state variables X by a non singular
matrix P, such that

w.E X=PZ
X =PZ
..... (9.78)
..... (9.79)

asy
Substituting eqns. (9.78) and (9.79) in eqn. (9.76), we have
PZ =APZ + bu

or
also y = CPZ + du
E
Z = p- 1 APZ + p- 1 bu
ngi ..... (9.80)
..... (9.81)
1
nee
Let us select the matrix P such that p- AP = J where J is a Jordon matrix.
Z= bu
Thus JZ +
y = cZ + du rin ..... (9.82)
..... (9.83)

where b= P-1b
g.n
and c =CP e
Now how to choose the matrix P such that p- 1 AP is a Jordon matrix? In order to answer this t
question, we consider the eigenvalues and eigenvectors of the matrix A.

9.6.1 Eigenvalues and Eigenvectors

Consider the equation
AX=Y ..... (9.84)
Here an n x n matrix A transforms an n x 1 vector X to another n x 1 vector Y. The vector X has
a direction in the state space. Let us investigate, whether there exists a vector X, which gets transformed
to another vector Y, in the same direction as X, when operated by the matrix A.
This means Y=AX where A is a scalar.
AX=AX ..... (9.85)
or [A - AI] X = 0 ..... (9.86)

www.EasyEngineering.net
 www.EasyEngineering.net

338 Control Systems

Eqn. (9.86) is a set of homogeneous equations and it will have a nontrivial (X"# 0) solution if and
only if,
IA - All = 0 ..... (9.87)
Eqn. (9.87) results in a polynomial in A given by,
An + a l An - I + ~ An - 2 + ... + an_IA + an = 0 ..... (9.88)
There are n values of A which satisfy eqn. (9.88). These values are called as eigen values of the
matrix A. Eqn. (9.88) is known as the characteristic equation of matrix A. Since eqns. (9.87) and
(9.75) are similar, the eigen values are also the poles of the transfer function.
For any eigenvalue A = Ai' from eqn. (9.86) we have,

ww [A - Ai I] X = 0
For this value of Ai' we know that a nonzero vector Xi exists satisfying the eqn. (9.89).
..... (9.89)

w.E
This vector X\ is known as the eigenvector corresponding to the eigenvalue \. Since there are
n eigenvalues for a nth order matrix A, there will be n eigenvectors for a given matrix A. Since
IA - All = 0, the rank of the matrix (A - AI) < n.

asy
If the rank of the matrix A is (n - 1), there will be one eigenvector corresponding to each A\. Let
these eigen vectors corresponding to AI' A2, ... An be m l , m2, ... mn respectively.
Then, from eqn. (9.85),
Ami =
E Al m l ngi
Am2 = A2 m 2
nee ..... (9.90)

Amn = An mn
rin
Eqn. (9.90) can be written in matrix form as,
A [m\, m2, ... m n] = [AI m l , A2 m 2, ... An m n] g.n ..... (9.91)

et
I
We can write eqn. (9.91) as,

A [m l , m2, ... m n] = [m l , m 2, ... ~]

[ ~:.1:2
o 0
~
An
..... (9.92)

Defming,
where M is called as the modal matrix, which is the matrix formed by the eigen vectors, we have,
AM = MJ ..... (9.93)
Where J is the diagonal matrix formed by the eigen values as its diagonal elements.
From eqn. (9.93) we have
J = ~I AM ..... (9.94)
This is the relation to be used for diagonalising any matrix A. Ifthe matrix P in eqn. 9.80 is chosen
to be M, the model matrix of A, we get the diagonal form.

www.EasyEngineering.net
 www.EasyEngineering.net

State Space Analysis of Control Systems 339

If all the eigenvalues ofA are distinct, it is always possible to find n linearly independent eigenvectors
and the J matrix will be diagonal. On the other hand if some of the eigenvalues are repeated we may
not be able to get n linearly independent eigenvectors. Corresponding to the repeated eigen values, we
will have to obtain, what are known as generalised eigenvectors. These vectors also will be linearly
independent and the resulting J matrix will be in the Jordon fonn as given in eqn. (9.68). Let us
illustrate these different cases by some examples.
Example 9.9
This examples illustrates the case when all the eigen values are distinct. Consider the matrix

ww
Solution
w.E
Since A is in triangular fonn the eigen values are the values on diagonal.

Consider
Al =

IA - All asy
1, A2 = 2 and A3 = 3

IA - All =
E 1- A
0
o
2-A ngi 2
1
o o 3-A
nee
rin
It can be shown that the eigen vector can be obtained as any non zero column of adj (A - AI).
Thus,

(2 - A) (3 - A) o g.n1
-2 (2-A)
adj IA - All =
[
~ (I-A) (3-A)
o et
-(I-A)
(I-A) (2-A)

Let us find the eigen vector corresponding to A = 1.

2 0 - 21
adj (A - I) =
[o0 00 00
The two non zero column of adj (A - I) are linearly dependent and hence anyone of them can be
taken as an eigen vector.

(Any constant multiple of the vector

also qualifies as an eigen vector)

www.EasyEngineering.net
 www.EasyEngineering.net

340 Control Systems

Similarly for A = 2 and A = 3 we have

10-1]

ww ~I =
.
0 1 -1
[0 0 1

w.E ~IAM=
1 0
0 1 =~l[~0 ~ ~][~ ~ ~]
asy [
o 0 -1 0 3 0 0 1

E
Carrying out the multiplication, we have,

n
J~[~ ~~] gin
Example 9.10 eer
ing
Let us consider the case when roots are repeated. Consider

[~ ~ ~] .ne
Solution:
A=
t
The eigen values of A are
Al = 1 and 1..2 = 1..3 = 2
The adj (A - AI) is given by
(2 - 1..)2 -(2-1..) 1-2 (2-1..) 1
Adj(A - AI) = ~ (1-1..) (2-1..) -(1-1..) ..... (9.95)
[
o (1-1..) (2-1..)

For 1..=1

Adj(A-I)=[~ ~l ~ll
www.EasyEngineering.net
 www.EasyEngineering.net

State Space Analysis of Control Systems '. l41

1
",
. '-,' .
Any non zero column of Adj (A - AI) qualifies as an eigen vector.
~ ,

1] }_
; . it . l
,~
.. m
I
=
[00 ~~~.
',.

Since A = 2 is a repeated eigen value, it is possible to get either two li~eatl, iAdtJ'ltft.~ ~'_~ri
vectors or one linearly independent eigen vector and one generalised eigen vector. In ~etWt\~ I~A: is
repeated r times, the number of linearly independent eigen vectors is eqwli to the ril18t1 of,<AJ..'1J,
or [n - rank of (A - Ai I)] where n is the order of the matrix A. • ~ '\ , /.~j.,
In the present example, ' •
..
t;, '
'f"
-1 I 2-

ww [A - 21] =
[
~ o
o
I ;
"

:'
.
'
.
\ :

The rank w.E P [A - 21] = 2

0 , i

, , ~
~'

, I

:, nullity
asy
11 [A - 21] = 3 - 2 = 1
We can find only one linearly independent vector for the repeated eiae6
j I, ",
v;1ll,.../l.:l: ~' • .t~tbe
W, b~; ;
' " '

E
obtained by any non zero column of adj [A - AI]. From eqn. (9.95) with A = t, , ,.

Adj[A-21] = [~ ~ni]gin \,~:,

1]
eer .
, ';
:. m2 =
[o1 ing ;
~
,
;' , t; ~
'. ~ • •
~

\ '\
i

.nJ'"e ~.9~); ,

• 0;'
" •

with respect to A and putting A = 2.

~J J
The third eigen vector can be obtained by considering the differential of.. jol·,kA
~ 1)~ t L~; ~~, ,
~ f(

m3 = -
d [1-2(1(2-A)]
- - ~)
/I,
. ~
'" ".!~ "
J ., ~ 'f '
~ of. ~'
I
I

dA (1- A) (2 - A) A=2
.i)
'I "/f~'i;;
. },~' ..•
.' ' J,,_i
~, ~ .;;. t::' "

~~ .,. '.

': , f ','
'. -, ~ I

www.EasyEngineering.net
 www.EasyEngineering.net

342 Control Systems

The modal matrix is given by

M=
1 1 2]
0 1 1
[ 001

1-1 -1]
~I =
[o
0 1
0
-1
1

ww J~~I ~ [~ -~l ~:] [~ ~ ~] [~AM 0 :]

w.E
as~y[~ ~ !]
En
This is the required result.
Example 9.11
gin
Consider the matrix,

1 1 2]
eer
[ A= 0 2 0
002
in g.n
Obtain the Jordon form of the matrix.
Solution:
et
The eigen values of A are,

The adj (A - Ai) is,

(2 - 1..)2
o
- (2 - A)
(1-1..) (2-1..)
-2(2-1..)
- (I-A)
1
[ o 0 (1-1..) (2-1..)

The eigen vector of Al = 1 is any non zero column of Adj (A - Ai I).

-1 1 2]
Adj(A - I) =
[o 0 0
0 0 0

www.EasyEngineering.net
 www.EasyEngineering.net

State Space Analysis of Control Systems 343

The three columns of (A - I) are linearly dependent.

For the repeated eigen value 1..2 = 1..3 = 2

(A-2I){f f ~1
ww
The rank
and nUllity
p (A - 21) = 1
11 (A - 21) = 3 - 1 = 2

w.E
Hence we can find two linearly independent eigen vectors for the repeated eigen value A = 2. The
two eigen vectors can be found by considering the equation

asy (A - AI) X = 0 with A = 2

E ngi
or
nee
Choose xl' x2 and x3 to satisfy this equations. Since there is only one equation and 3 variables, we
can choose two variables arbitrarily.
rin
We have xl = x2 + 2x3
g.n
One solution is obtained by taking xl = 0, x 2 = 2 and x3 = -1. Another linearly independent solution
is obtained by taking
Xl = 1, x2 = 1 and x3 = 0 et
The modal matrix is,

M=[~ ~l i1
-1
o -21
-1
2

www.EasyEngineering.net
 www.EasyEngineering.net

Control Systems
" ~ ,,~

, '; -N' ,,(~do, Ntf

' ix is given by,
n-II ... f I
~ , ,~ '. J = M" AM
(P , ' ."

:rfC
>"'j'j'"
1. .- ~
'. ", '"

,
4'
~I =[ ~ -~
0
~~1[~ ~ ~1[~
2 0 0 2
0
2 :
0 -1 0
1
" ,' . fL ' \
"'"
[~ ~ ~1
"

~ ,
:'::: . "
,~ ">,

.' ~Yf 'ot~e that the Jordon matrix is purely a diagonal matrix eventhough a root is repeated.
, ~; ~~ ww
." "'.', ..i(>a f~ the m~~ix A, when it .is in the companion form deserves special attention.

w
-"l~~ ~fJ~ ~ 1-04'11 of A ~re dIstm~t and are gIven by AI' 1..2' ... An

.Ea
I ', 'i:J:l t;thll~" ~ modal matnx can be shown to be equal to,
'~f" .. .r;!t·r'f "
. .,. 1 1 ... 1

syE
1. V";', ." ••
", ~.
f • "I'

Al 1..2 An
4 '. ' ( :
..
1..2I 1..22 1..2n
ngi
" '

j " j • • ", M=

. ., I ,

,
, -',

~ . ~ ~ ~ is t special matrix and is known as Vander Monde Matrix.

An- I
I
An- I
2 An- I
n
nee
;, ~
. ~ ', ~¥¥~r<lOts of A are repeated. Let us consider an example in which the order of the
rin
, ' . ;~ . . . ItBd the roots are
, ;'/i ', ," " ,. .. AI' AI' AI' 1..2' t.. 2 , 1..3 g.n
• i
.
i" ~\ ~e'm.atrix Ais in companion form, for an) repeated root of A, we can find only one
. ;~ly ~pendent eigenvector and the other eigenvectors corresponding to this root are
". . .r,a~f'" ,eig,~vectors. The modal matrix in this case can be obtained as,
et
, •
0 0 0
,! , '

Al 0 1..2 1 1..3
}..
1..2I 21..1 A~ 21..2 1..23
M=
, , ,
~(An-I) ~£(t..n-I) An-I ~(An-I) An-I
i ; -,
An-I
I dA I I 22! dA~ dA 2 3
, . I 2

, . 1 ! ,¥the
t~ ,l..j" i~ r~eated r times, the qth eigen vector (q ~ r) corresponding to this eigen value
~_i;,: 'll'~"b)' t
~;' .·.L~ t 1 ' .~ ~
dq- I
!:';~' ~~in
\ \.:"
" \,.' 1
(q -1) dAr l [1 t..j A~ t..rIf
.:,~~ ~f~ 1h~S obtained is known as the modified Vander Monde matrix.

1.. nt'~'
~ t..
! _f
" '1
'
•
www.EasyEngineering.net
 www.EasyEngineering.net

State Space Analysis of Control Systems 345

9.7 Solution of State Equation

The state equation is given by,

x(t) = AX (t) + bu (t) ..... (9.96)

With the initial condition X(o) = Xo
We can write eqn. (9.96) as,

X (t) -
A X (t) = bu (t) ..... (9.97)
Multiplying both sides of Eqn. (9.97) by e-A\ we have
e-At [X (t) - A X (t)] = e-At bu (t) ..... (9.98)

ww
Consider ~
dt
[e-At X (t)] = e-At X(t) - Ae-At X (t)

w.E
From eqn. (9.98) and (9.99), we can write,
= e-At (X (t) - A X (t)] ..... (9.99)

a dt
..!
syE
[e-At X (t)] = e-At bu (t) ..... (9.100)

nf gi
Integrating both sides of eqn. (9.100) between the limits 0 to t, we get
t t

e-At X (t) I
o
=
o nee
e-Ar b u (-t) d,

t
e- At X (t) - Xo = fe-At b u (,) d, rin
o
g.n
et
t

X (t) = eAt Xo + eAt fe-At b u (,) d,

o
t

or X(t)=eAtxo+ f eA(t-t)b u(,) d,. . .... (9.101)

o
The first term on the right hand side of eqn. (9.101) is the homogeneous solution of eqn. (9.96)
and the second term is the forced solution. The matrix exponential eAt is defined by th~nfinite series,
At _ A2 t 2 Ai t i
e - I + At + - - + .. ... +. - - + .....
2! j !
Eqn. (9.10 I) can be generalised to any "-state at t = 10 rather than t = 0, as
1

X(t)=eA(t-to)X(to)+ J eA(t-t)bu(,)d,. . .... (9.102)

10

Let us concentrate on the homogenous solution.

www.EasyEngineering.net
 www.EasyEngineering.net

346 Control Systems

If u(t) = 0, we have
.....(9 .103)
This equation gives the relation relation between the initial state Xo and the state at any time t. The
transition from the state Xo to X(t) is carried out by the matrix exponential eAt. Hence this matrix
function is known as the state transition matrix (STM). If the STM is known for a given system the
response to any input can be obtained by using eqn. (9.101). This is a very important concept in the
state space analysis of any system.

9.7.1 Properties of State Transition Matrix

Let the state transition matrix be denoted by,

ww <I> (t) = eAt

Some useful properties of STM are listed below.
1.
2. w.E <I> (0) = e
AO = I; I is a unit matirx
<1>-1 (t) = [eAtrl = e-At = <I> (- t)
3.
asy <I> (tl +~) = eA(tl + t2) = eAt I. eAt2
(~) (~)

E = <I> (t l )

n
<I> = <I>
The solution of the state equation can be written in terms of the function
<I> (t l )
<I> (t) as,

f gi
t

X(t) = <I> (t) Xo +

o
<I> (t - .) b u (.) d.
nee . .... (9.104)

rin
In the solution of state equation, the STM plays an important role and hence we must find ways of
computing this matrix.

9.7.2 Methods of Computing State Transition Matrix g.n

There are several methods of computing the matrix exponential eAt. Let us consider some of them.
(a) Laplace transform method
et
Let X =AX; X(O) = Xo ..... (9.105)
Taking Laplace transform of eqn. (9.105), we have,
(s X(s) - Xo) = A Xes)
This equation can be written as,
[sI - A] X (s) = Xo
X (s) = (sI - Arl Xo ..... (9.106)
Taking inverve Laplace transform of eqn. (9.106) we get,
X(t) = f.-I (sI - Arl Xo .....(9.107)
Comparing eqn. (9.107) with eqn. (9.103), it is easy to see that,
eAt = f.- I [(sI -Arl] ..... (9.108)

www.EasyEngineering.net
 www.EasyEngineering.net

State Space Analysis of Control Systems 347

Example 9.12
Find the homogenous solution of the system,

x=[_~ _~]x;
Solution
The solution of the given system is given by,
X(t) = eAt Xo
Let us compute the state transition matrix eAt using Laplace transform method.

ww eAt = 1',-1 [(sl - Arl]

[~ ~] ~ 3]
w.E (sl - A) = - [ _02

a syE [s
= 2 s+3
-1]

(sl -Ar l
1ngi [s +3
~]
= S2 +3s + 2 - 2
nee
1) r2)]in
(5+ :(5+
g.n
eAt = 1',-1 [sl -Ar l
(s + 1) (s + 2) (s + 1) (s + 2)
et
2e-I -e -21 e-I -e -21 1
= [ -2e- +2e- 21 -e-I +2e- 21
'
The homogeneous solution of the state equation is given by,
X(t) = eAt Xo
-I
2e -e
-21

= [ -2e- +2e- 21
I
e- _e-
-e- I
21

+2e- 21
1[01]
'

www.EasyEngineering.net
 www.EasyEngineering.net

348 Control Systems

The solution can be obtained in the Laplace transfer domain itself and the time domain solution can
be obtained by finding the Laplace inverse. This obviates the need for finding Laplace inverse of all
the elements of eAt, Thus
X(s)= (sI-Arl Xo

[~]
s+3

X(s) =
(s + l)(s + 2)
-2
(S+I):(S+2)1
[
(s + l)(s + 2) (s + l)(s + 2)

ww =
s+3
(S+I~c;+2)
1
w.E [
(s + l)(s + 2)

a syE 1
2e -t - e -2t
X(t) = [ _2e- t +2e- 2t

(b) Cayley - Hamilton Technique ngi

n eer
Any function of a matrix f(A) which can be expressed as an infinite series in powers of A can be
obtained by considering a polynomial function g(A) of order (n - 1), using Cayley Hamilton theorem,
Here n is the order of the matrix A.
ing
Cayley - Hamilton theorem states that any matrix satisfies its own characteristic equation.
The characteristic equation of a matrix A is given by,
.ne
q(A) = IAI - AI = An + a l An - I + .... + ~ _ I A + an = 0
The Cayley Hamilton theorem say:, that,
q(A) = An + a l An -I + .... + an _ I A + an I = 0
t ..... (9.109)

..... (9.110)
Let n
f(A) = b o I + b l A + b2 A2 + .... + bnAn + b n + I A + 1+..... . .... (9.111)
Consider a scalar function,
f(A) = bo + b l A + b2 A2 + .... + bn An + bn + I An + 1+..... ..... (9.112)
Let f(A) be divided by the characteristic polynomial q(A) given by eqn. (9.109). Let the remainder
polynomial be g(A). Since q(A) is of order n, the remainder polynomial g(A) will be of order (n - 1)
and let Q(A) be the quotient polynomial. Thus,
f(A) = Q(A) q(A) + g(A) ..... (9.113)
Let g(A) = a o + a l A+ a l A2 + ..... + an_I An-I
The function f(A) can be obtained from eqn. (9.113) by replacing A with A
Thus f(A) = Q(A) q(A) + g(A) ..... (9.114)

www.EasyEngineering.net
 www.EasyEngineering.net

State Space Analysis of Control Systems 349

But q(A) = ° by Cayley Hamilton theorem.

f(A) = g(A) ..... (9.115)
-- a + a A + a A2 + ..... + an - I An-I ..... (9.116)
o l 2
Since q(A 1) = ° for i = 1,2, ... n
we have from eqn. (9.113),
f(A) = g(A) for i = 1, 2, ... n ..... (9.117)
Eqn. (9.117) gives n equations for the n unknown coefficients a o' ai' ... an _ I of g(A)
We notice that g(A) is a polynomial of order (n - 1) only and hence f(A) which is of infinite order

ww
can be computed interms of a finite lower order polynomial.
To summarise, the procedure for finding a matrix function is :
(i)
(ii) w.E
Find the eigen vlaues of A.
(a) If all the eigenvalues of A are distinct solve for the coefficients a of g(A) using
l

a f(A)
syE
= g(A) for i = 1, 2, ... n
(b) If some eigenvalues are repeated, the procedure is modified as shown below.
Let A = Ak be repeated r times.
ngi
Then
dJq(A)1
~
A=Ak
= ° .
nee
for J = 0, 1, ... r - 1

Using this equation in eqn. (9.114), we get

rin
djf(A)1 = dJg(A)1 g.n
for j = 0, I, ..... r - 1
dAJ
A=Ak
d~J
~ A=Ak

This gives the required r equations corresponding to the repeated eigen value Ak. Proceeding
et
in a similar marmer for other repeated eigenvalues and using eqn. (9.116) we get the necessary
equations to evaluate,
a l , for j = 0, 1, 2 ... n - 1
(iii) The matrix function f(A) can be computed using the relation,
f(A) = g(A)
The procedure is illustrated using a few examples.
Example 9.13
Find f(A) =A4 + 2A3

where A=[_~ _~]

www.EasyEngineering.net
 www.EasyEngineering.net

350 Control Systems

Solution:
The eigenvalues of A are obtained from,

I=~
1
q(A.) = _5 _A.I = 0
A. (A. + 5) + 6 = 0
2
A. + SA. + 6 = 0
or A. = -3,-2
Since the order of A is 2,
g(A.) = 0.0 + 0. 1 A.
Now
ww f(A.) = A.4 + n 3

w.E g(A.)
g(-3)
g(-2)
=
=
f(A.) for A. = -3, -2
0.0 - 30. 1 = (_3)4 + 2 (_3)3
0.0 - 20. 1 = (_2)4 + 2 (_2)3
= 27
0

asy
= =

Solving for 0.0 and ai' we get

0. 0
g(A.)
fCA)
=
E ~
- 54; 0. 1 =
- 54 - 27 A.
g(A) = ngi
- 27

54 1- 27 A
nee
= -

[-5~ _5~]-27 [_~ _~]

f(A)
rin
g.n
Example 9.14
et
For
A=[ -6~ -110 -6~l
Find f(A) = eAt

Solution:
The characteristic equation is given by
-A. 1 0
q(A.) = 0 - A. 1 = 0
-6 -11 -6-A.
The eigenvalues are
A. = -1, -2, and -3
f(A.) = eAt

www.EasyEngineering.net
 www.EasyEngineering.net

State Space Analysis of Control Systems 351

Since A is of order 3, the remainder polynomial is

g(A.) = <Xo + <XI A. + <X2 A.2
g(A. I ) = f(A. I ) for i = 1, 2, 3
-t
<Xo- <XI + <X2 = e
<X0- 2<x
12
+ 4<x = e-2t
<X - 3<x + 9<x = e-3t
012
Solving these equations for <Xo' <XI and <X2' we have,
<Xo = 3e-t - 3e-2t + e-3t

3
ww <XI = 2.5e-t - 4e-2t + "2 e-3t

w.E <X2 = 0.5e-t - e-2t + 0.5 e-3t

f(A) = eAt = g(A)

asy = <XoI + <XI A+ ~A2

Substituting for A we get
E <Xl ngi
<Xo -Il<X2
-II<Xl + 60<X 2 nee
Substituting the values of <Xo' <XI and <X2' we get, rin
3e-t _ 3e -2t + e -3t 2.5e -t _ 4e -2t + l.5e -3t t g.n
2t 3t
1
eAt = -3e- t +6e- 2t +3e-3t
[ 3e-t -12e- 2t +ge-3t
-2.5e- t +8e- 2t -4.5e- 3t
2.5e- t -I6e-2t + 13.5e-3t
0.5e- _e- +0.5e-
et
_ 0.5e -t + 2e -2t _ 1.5e -3t
0.5e- t _4e- 2t +4.5e-3t

Example 9.15

Find eAt for A = [ 0 I]

-4 -4

Solution:
The characteristic equation is
A.2 + 4A. + 4 = 0
(A. + 2)2 = 0 or A. = -2, -2
We have f(A.) = eA.t and g (A.) = <Xo + <XI A.

www.EasyEngineering.net
 www.EasyEngineering.net

352 Control Systems

Since the eigen value is repeated,

f(A) = g(A)

and
df("-l) dg("-J
--=--
d\ d"-l
f(-2) = g(-2)
a o - 20. I = e-2t

and t eAtl = a
A=-2 I

t e-2t = 0. 1
or
ww a o= e-2t + 2t e-2t

w.E f(A) = eAt = g(A) = 0.01 + alA

asy =
0. 0
[ 0 U
0]
o + ul
[ 0
- 4
1]
- 4

E ngi
nee
rin
Example 9.16
Obtain the solution of the state equation, g.n
. [0-1
X= et
Xo = [0 1] T and u is a unit step input.
Solution:
First the STM is computed. Using Laplace transform method

= S2
1 [s-1+ 2 s1]
+ 2s + 1
Consider the state equation,
X =AX+bu

www.EasyEngineering.net
 www.EasyEngineering.net

State Space Analysis of Control Systems 353

Taking Laplace transfonn of this equation,

sX(s) - Xo = AX (s) + b U(s)
or (sl - A) X(s) = Xo + b U(s)
X(s) = (sl - Arl Xo + (sl - Arl b U(s)
Substituting the relevent values,

X(s) S+2 1] [0]+[S2

S2 +2s + I [ -2 s 1 -1
:~:+l
S2 + 2s + 1

ww
w.E=[:] ['(':1)'] +

asy (s+li

En
+~~Il)l gin
e
Taking inverse Laplace transfonn, we get,
eri
X(t) = [
1
~~
-tj n g.n
Example 9.17
Find the state response of the system,
et
l
0 1
X= 0 0
-6 -11
For Xo = [1 0 O]I and a unit step input.
Solution:
We have already calculated the STM for the A matrix of this example, in Ex. 9.12. Using this result in
the response of the system.

X(t)=eAtXo+ J
t

eA(t-t)bu(t)dt
o

www.EasyEngineering.net
 www.EasyEngineering.net

354 Control Systems

21
3e-1 -3e- +e-
eAIXO= -3e-I+6e-21+3e-31
31
1
From Ex. (9.14),
[ 3e-1 -12e- 21 +ge- 31

O.5e-(t-t) - e- 2 (t-t) + O.5e-3 (t-t) 1

and eA(t--r) bu = _O.5e-(t-t) + 2e-2 (t-t) _1.5e-3 (t-t)
[ O.5e-(t-t) _ 4e- 2 (t-t) + 4.5e-3 (t-t)

Integrating between the limits 0 to t, we get

ww O.5e-(I-t) _ e
-2(I-t) -3(I-f)
e__
+ O.5_

w.Ef I

eA(I- -r) bu(t)d. =

2 3
_ O.Se -(I-t) + e -2(I-t) _ O.5e -3(I-t)

asy
o
O.5e -(I-t) _ 2e -2(t-t) + 1.5e -3(I-f)
o

E ngi lOS
- - . e -lOS
+ . e -21 --e
1 -31
6
nee
O.5e -I _ e -21 + O.Se -31
_ O.5e -I + 2e -21 -l.Se -31
6

rin
I

f eA(t--r)b u (t) d. g.n

X(t) = eAt xo +
o et
1 - 0 .5e -lOS
- + . e -21 --e
1 -31
6 6
O.Se -I _ e -2t + O.Se -31
- O.Se-1 + 2e-2t -l.Se-31

1 2 .Se -I - 2 .Se -21 +-e

-+ S -31
6 6
_ 2.Se -I + Se -21 + 3.Se -31
2.Se -I -1 Oe -21 + 7 .Se -31

www.EasyEngineering.net
 www.EasyEngineering.net

State Space Analysis of Control Systems 355

Example 9.18
Obtain the solution of the following state equation by first obtaining the canmical form.

With the initial condition X(O) = [1 0 - l]T

Solution:
First let us obtain the canmical form of the given state equation. The system matrix A is given by,

ww [~ ~ ~]
w.E A=

Let
asy X=PZ
Where P is the modal matrix .

En
Modal matrix for this matrix A was obtained in Ex 9.9. Using the result, we have

gin
[~ :]
0

eeP= 1
0
rin
[I 0-I]
p- 1 = 0 1 -1 g.n
and
o
Using the transformation, we have
0 1 et
Z = p-l A P Z + p-l bu

(.: p-l AP gives the diagonal matrix with the eigen values 1,2 and 3 on the diagonal)

1 0 0] [-1]
[
Z= ~ ~ ~ Z+ -~ U

www.EasyEngineering.net
 www.EasyEngineering.net

356 Control Systems

and Z(O) = p-l X (0)

ww
Solution of the above equation is given by,

w.E Z(t) = eJt Z(o) + feJ(t-t)

t

a
bU('t)d't

asy
where E ngi
nee
The state transmition matrix eJt can be easily obtained since J is in diagonal form.
It is given by,
rin
Jt [e'~ 0
e 2t
g.n
e =
0 e;' ] et
Thus

e;'][~] f~'
0 0

Z (t) =
[e'
~ e 2t
+
e2(t-t)

0 e"L] [=:J dt
0

Z (t) = nfe(H']
0
o
+ t - e 2(t-t)
a e 3(t-t)
d't

www.EasyEngineering.net
 www.EasyEngineering.net

State Space Analysis of Control Systems 357

e(t-t)

[~]
e2(t-t)

Z (t) = + 2
_ e3(t-t)
3 0

ww Z (t) = m + 2 2

w.E 1 e
--+-
3 3
3t

asy
E .!..(l-e 2t )
2 ngi
..!..C-1+e 3t )
3
nee
rin
g.n
XCt) = PZ Ct),

We have
.!..(l_e 2t )
2
et
..!..Ce 3t -1)
3

e3t 2
-+-
3 3
1 1 2t 1 3t
X(t) = ---e +-e
6 2 3
3t
..!..Ce -1)
3

www.EasyEngineering.net
 www.EasyEngineering.net

358 Control Systems

9.8 Qualitative Analysis of Control Systems

Sofar we have discussed,
(a) how to model a given physical system and
(b) solution of such a system for a given input, i.e., -quantitative analysis of the system.
We will now discuss some fundamental concepts about the control of these systems, i.e., qualitative
aspects of these systems, when a controller is to be designed to obtain a desired response.
There are two basic questions to be answered before we design a suitable controller for a given
system, namely,
(i) can we transfer the system from any initial state to any desired fmal state in fmite time by

ww
applying a suitable unconstrained control.
(ii) by measuring the output for a finite length of time, can we determine the initial state of the
system.
w.E
These aspects were first introduced by Kalman and are defined as controllability and observability

asy
of the system respectively. These aspects playa very important role in the design of a controller for
a given system. In this section we will only give elementary treatment of these aspects of the system.

9.B.l State Controllability

Definition
E ngi
nee
A system is said to be state controllable at t = to if it is po~sible to transfer the initial state X (to) to any

rin
final state X (tf ) in fmite time, u&ing an unconstrained control signal u(t). If every state is controllable,
the system is said to be completely state controllable.
Kalman's test for control/ability
g.n
Consider the dynamical system described by the state equation,

X =AX+Bu et
..... (9.118)
where X is a n-dimensional state vector
u is an scalar control signal
A is an n x n system matrix
B is an n x I input matrix
Without loss of generality, let us consider the initial time 10 = 0 and the fmal desired state to be the
origin of the state space.
The solution of the state eqn. (9.118) is given by,

X(t) = eAtX(O) + f eA(t--c)B uCr) dt

o

www.EasyEngineering.net
 www.EasyEngineering.net

State Space Analysis of Control Systems 359

Since the fmal state at t = t f Le., X(tf) = 0, we have,

t
f
X(tf) = 0 = e Atf X(O) + f eA(tf-t) B u(.) d.
o
tf

or X(O) = _e- Atf f eA(tf-t) B u(.) d.

o
t
f
X(O) = - fe-At B u(.) d. ..... (9.119)
• 0

ww
If we are able to obtain a u(t) for any given X(O) from eqn. (9.119), the system is controllable.
Using eqn. (9.116) with f(A) = e-At, we can write e-At as,

w.E ..... (9.120)

asy
Substituting eqn. (9.120) in eqn. (9.119), we have,

E ngi
nee ..... (9.121)

rin
Let·
g.n
Then

Eqn. (9.122) can be written in the matrix form,

et..... (9.122)

~o
~I
X(O) = [B : AB : A 2B : ..... : An-IB ] ~2 ..... (9.123)

A control u(t) exists, which transfers any given initial state X(O) to the origin of state space, if the
vector [~o' ~l' ... ~n- d can be obtained from eqn. (9.123). It requires that the n x n matrix
T

[B : AB : A2B : ... : An-IBJ

be non singular, i.e., its rank be equal to n.

www.EasyEngineering.net
 www.EasyEngineering.net

360 Control Systems

This result can be extended to a more general case where the input vector is of dimension r. In the'
equation

X =AX+Bu ..... (9.124)

u is an m-vector and
B is an n x m matrix
The controllability condition in this case can be stated as :
The (n x nm) matrix

..... (9.125)

ww has a rank n.
The matrix Qc is known as the controllability matrix. For complete state controllability, therefore,

w.E
all the columns of Qc must be linearly independent. This test for deciding the controllability is known
as Kalman's test for controllability.
Gilbert's test for Controllability

asy
Let us now consider an alternate test for examining the controllability of a given system. Let us
suppose that the eigenvalues of A in eqn. (9.124) are distinct. It is possible to transform the A matrix

X=PZ E
into a diagonal matrix using a transformation matrix P. As already discussed in section 9.6, if

ngi
eqn. (9.124) can be transformed to

Z = p- l APZ + p-l Bu nee ..... (9.126)

rin
where P matrix is the modal matrix and p-l AP is a diagonal matrix with eigen values on the
diagonal. Eqns. (9.126) are first order uncoupled differential equations. The kth equation is given by

Zk = Ak ~ + a kl u l + a k2 ~ + ... + akIn urn g.n ..... (9.127)

where Ak is the kth eigenvalue of A
and a kl , a k2 , ... akIn' are the kth row elements of p-l B et
If all the elements ofkth row ofP-l B, i.e.,
a kl = a k2 = ... = akIn = 0
then no control signal exists in eqn. (9.127) and hence the state variable Zk cannot be controlled.
Hence the condition for complete state controllability is that none of the rows of P-lB in
eqn. (9.126) should contain all zeros.
The above treatment can be extended to the case where the matrix A has repeated eigen values.
In this case the matrix A can be transformed to Jordon's cononical form. If P is the modal matrix
which transforms A into its Jordon form, let
X=PZ
and hence
Z =p- l APZ+p-l Bu
=JZ+p-1 Bu ..... (9.128)

www.EasyEngineering.net
 www.EasyEngineering.net

State Space Analysis of Control Systems 361

For a 7th order system with eigen values AI' AI' A2, A2, A2, A3, and A4 , J is of the form,

Al 1 1 0 0 0 0 0
o All 0 0 0 0
----+------,
0
0 o 1 A2 1 o1 0 0
1 1
J= 0 0 1
0 A2 1 1
0 0 ..... (9.129)
0 0 1
L
0 0
______
A2 +_,
1 0 0
0 0 0 0 o 1LA3. _1 r -0 ,
0 0 0 0 0 o 1• A_4 1

ww
Each square block is called a Jordon block, as discussed earlier. J can thus be written as,

w.E
where J 1 is of order 2
asy
and
J2 is of order 3
J3, J4 are each of order 1
E ngi
nee
The condition for complete state controllability by this method can be stated as follows:
The system is completely state controllable if and only if,

rin
1. No two Jordon blocks in eqn. (9.129), are associated with the same eigen value.

g.n
2. The elements of the row of p-l B corresponding to the last row of any Jordon block of J are
not all zero. If some of the Jordon blocks are of order 1, i.e., the corresponding eigen values

et
are distinct, the rows in p-l B corresponding to these eigen values must not contain all zeros.
It is to be noted that application of Gilbert's method requires that the equations be put in the
Jordon's form. Kalman's test can be applied to any representation of the given system.
Output Controllability
In practical systems, we may have to control the output rather than the states. In such cases, we can
define output controllability. State controllability is neither necessary nor sufficient for controlling the
output.
Consider the system
X =AX+Bu
Y=CX+Du ..... (9.130)
x - is an n vector
u - is an m vector
Y - is a p vector

www.EasyEngineering.net
 www.EasyEngineering.net

362 Control Systems

The above system is completely output controllable if it is possible to find an unconstrained

control vector u(t) which will transfer any given initial output Y (to) to any final output Y (9 in fmite
time.
The test for complete controllability of the system in eqn. (9.130) is that the rank of the
[p x (n + 1) m] matrix,

..... (9.131)
be equal to p.
If in a particular system D = 0 i.e., the system output does not directly depend on the input, the
rank of the matrix,

ww ..... (9.132)
must be p.
Example -9.19 w.E
asy
Test the following systems for controllability using (i) Kalman's test (ii) Gilbert's test.

(a) A=[j -1~ J E

B=m ngi ~l B= Hl
(b)
A= [~
0
2
0

nee
[~ ~ ~l B= [i] A= [~ ~lr B= Hl
(c) A =
002
(d)
ing 2
0

Solution:
.ne
(a) A=[j -1~ J B=m
t
(i) Kalman's test
The controllability matrix,

Qc =[B:AB:A2B]

=
[
0
0
0
1-6
II
1 -6 25
The rank of Qc is 3.
Hence the system is controllable.

www.EasyEngineering.net
 www.EasyEngineering.net

State Space Analysis of Control Systems 363

(ii) Gilberts test

The eigen values are -I, -2,-3
The modal matrix P is given by,

3 2.5 0.5]
ww p-l =
[
-3 -4 -1
1 1.5 0.5

w.E
p-l B = as[-~y~ ~~] [~]
En
1 1.5 0.5 1

-_ [0.5] gin
-1
0.5 eer
ing
Since all the eigenvalues are distinct and none of the elements of p-l B is zero, the system is
completely state controllable.
.ne
(b) t
(i) Kalman's test
The controllability matrix,

The rank of this matrix is 2 and hence the system is not controllable.

www.EasyEngineering.net
 www.EasyEngineering.net

364 Control Systems

(ii) Gilbert's method

From example 9.9, we have the eigen values to be 1,2, and 3 and the modal matrix P to
be,

=[~
0

p
1
0
:]
p-l =
[1 0 -1]
0 1 -1

ww o 0 1

and w.E =[~ =:J [-:]

p-l B
0

asy 0

=H] En
gin
eer
Since the first element in p-1B is zero, the system is not controllable.

ing
.ne
(i) Kalman's test
The controllability matrix, t
Qc = [i ~ ~]
The rank of this matrix is 1 and hence the system is not controllable.
(ii) Gilbert's test
The eigen values are 1, 2 and 2.
From example 9.10, the modal matrix Pis

P=
1 11 2]1
[o 0 1
0

www.EasyEngineering.net
 www.EasyEngineering.net

State Space Analysis of Control Systems 365

and the Jordon form of A is J ~ l~ ~ r]

~l~ l~ -1]
-1
1 -1
0 1

l~ ~:Hl]
-1
p-I B

ww and 1
= 0

w.E ~m
asy
E
Since the first element of p-I B corresponding to the distinct eigen value, Al = 1, is zero

ngi
and the last row element of p-I B corresponding to the repeated eigen value, ~ = 2, is
also zero the system is not controllable.

n
A~l~ ~~] B~H] eer
(d)
ing
(i) Kalman's test
The controllability matrix, .ne
Qc =H-~ ~] t
Since the rank of this is 1, the system is not controllable.
(ii) Gilbert's method
From example 9.11, the eigen values, the model matrix P, and its inverse are,
Al = 1, A2 = A3 = 1 ;

~ ~] ; p-I = r~ ~1 -2]
-1
-1 0 lo 2

www.EasyEngineering.net
 www.EasyEngineering.net

366 Control Systems

The Jordon fonn of A is,

0
J ~ p-l AP~ [~ 2
0 ~]
0
~ [~ J2
0 n
ww Since the Jordon block J2 and J3 are associated with the same eigen value 1..2 = 1..3 = 2,
the system is )1ot controllable. In this case, it is not necessary to fmd the elements of

Example 9.20
w.E
P-lB. For any B, the system is not controllable.

asy
For the following system, detennine controllability.

2 0 0]E [0 1]
+ 1 0 [~l
X= 0 2 0
[o 3 1 X
0 1ngi ]

nee
2

Solution:
The controllability matrix,
rin
Qc = [B : AB : A2B]
g.n
e t
The rank of this matrix is 3 and hence the system is completely state controllable.

9.8.2 Observability
Now let us define the second aspect of the system, i.e., observability.
Definition
A system is said to be completely observable, if every state X(to) can be detennined from the
measurement of the output Y(t) over a finite time interval, to ~ t ~ t f .
This property of the system is very useful, because the state vector can be constructed by observing
the output variables over a finite interval of time. Therefore even if some states are not measurable,
they can be estimated using the measurable output variables over a finite time. Estimation of all the
state variables is required in some cases, when these state variables have to be fedback to obtain a
desired control.

www.EasyEngineering.net
 www.EasyEngineering.net

State Space Analysis of Control Systems 367

Kalman's test for observability

Consider the following system,

X =AX+Bu
Y=CX ..... (9.133)
The condition for complete state observability is that the observability matrix,

C
CA
Qo = CA 2 ..... (9.134)

ww CA n - 1

has a rank n.
w.E
Gilbert's test for observability

asy
The state equations represented by eqn. (9.133) can be transformed to its canonical form using the
transformation,
X=PZ
E
z = p-l APZ + p-l Bu ngi ..... (9.135)
Y=CPZ
nee
rin
Since p-lAP is a diagonal matrix, the states are decoupled. Information about one state is not
available in the other state equations. Since Y is a combination of these decoupled states, if any

g.n
column ofCP in eqn. (9.135) is zero, the corresponding state variable does not affect the output Y,
i.e., if the jth column of CP contains all zeros, then the state variable ZJ is not observable from the
output. This is the condition if all eigen values of A are distinct.
et
If some of the eigen values are repeated, the matrix p-l AP is in Jordon form and the conditions for
complete state observability are stated as follows:
The system is completely state observable if,
I. no two Jordon blocks in J = p-l AP are associated with the same eigen value.
2. no columns of CP that correspond to the first row of each Jordon block consist of zero
elements. If some of the eigen values are distinct, the Jordon blocks corresponding to these
eigen values are of order I and hence the columns of CP corresponding to these block must
not be all zero.
The above conditions can be best illustrated by the following examples:

(a) X= [-2 0]
0 -4 X Y = [2 -2] X

The system is observable since C has no zero elements.

www.EasyEngineering.net
 www.EasyEngineering.net

368 Control Systems

(b) . [-1 0]
X = 0 -1 X Y=[l -l]X

Not observable since two Jordon blocks are associated with the same eigen value-1.

(c) x = [-~ -~ ~] X Y = [a b c] X
o 0-2
(i) If a = 0 and / or c :t:. 0 the system is not observable.
(ii) If a :t:. 0 and c:t:. O. the system is observable.

ww
w.E [
=-1~ ~ __ 0__ ~I
. 0 1 -2 1 0
(d) X r- 0: 0 -2 1 X
1
o 0 0-2

asy
1

Since all the elements of first column of C of Jordon block corresponding to the :!igenvalue

E
A= -1 are non zero and all the elements of first column ofC of the Jordon block corresponding

ngi
to the repeated eigenvalue A = -2 are non zero, the system is completely observable.

0 01
(e) X = -
-3
-
1
~o ~3_0 1-2
~ ~ - - ~1 X
1

nee 0 1 1 2 -1]

1o 0
1
0 -2
[
Y=OO:O
rin 1 X

g.n
1

Since all the elements of column 1 of C corresponding to the Jordon block of the repeated

Problem 9.21
eigenvalue A = -3 are zero, the system is not observable.
et
Comment on the complete state observability of the following systems using,
(i) Kalman's test
(ii) Gilbert's test

(a) x= [j Jx -11
0 Y = [1 -1 1] X

(b) X
= [~ 2
o
I 2]1 X
2
Y = [1 -1 0] X

www.EasyEngineering.net
 www.EasyEngineering.net

State Space Analysis of Control Systems 369

(c) Y = [1 0 0] X

(b) Y=[~ 0 ~]X

Solution:

(a) ww x= [~ ~ ~] X Y = [1 0 0] X

(i) w.E
Kalman's test
-6 -11 -6

a
The observability matrix,

syE
[C~ ] r~ 0n~]
QO =
CA 2
=
0 0 1 gin
The rank of Qo is 3 and hence the system is observable.
eer
(ii) Gilbert's test
The model matrix for this system is, ing
.ne
p+: -~ -i] t
CP = [I 0 0] [-: -~ -i]
= [1 1]
Since none of the columns of CP are zero, the system is completely observable.

(b) X=
1 21 2]1 X
0 Y = [1 -1 0] X
[0 0 2

www.EasyEngineering.net
 www.EasyEngineering.net

370 Control Systems

(i) Kalman's test

The observability matrix,

The rank of Q o is 2 and hence the system is not observable.

(ii) Gilbert's test
The modal matrix Pis,

ww
w.E
The transformed equations are,

asy
Y=CPZ
E ngi
[~o ~ ~l
nee
= [1 -1 0] Z
rin
= [1 0 1] Z
0 1
g.n
Since the first column of CP corresponding to the Jordon block of eigen value A = 2 is
zero, the system is not observable. et
(c) Y = [1 0 0] X

(i) Kalman's test

The observability matrix,

Qo = [~A 1=[~ ~ ~l
CA 2 1 3 6

Since the rank of Qo is 2, the system is not observable.

www.EasyEngineering.net
 www.EasyEngineering.net

State Space Analysis of Control Systems 371

(ii) Gilbert's test

The modal matrix for the system is,

P=
1 02 11]
0
[o -1 0
The Jordon form of the system is

I'0 0]
Z= Oi2~0
[o 0 +,2_ z
L _

ww
w.E Y= [1 0 0]
[~1 _021 ~1]
=
asy
[1 0 0] Z
Since two Jordon blocks correspond to the same eigen value A = 2, the system is not
observable.
E
ngi[~ ~]
x=[~ ~]
0
0
(d)
nee
2
0
X
Y= X

(i) Kalman's test

rin
The observability matrix,
g.n
0
0 1
0
et
1 0 5
Qo =
0 2
1 0 17
0 4 5

The rank of Qo is 3 and hence the system is observable.

(ii) Gilbert's test
The modal matrix P and the Jordon form of the equations are,

1 0 1]
P=
[o 0 1
0 1 1

www.EasyEngineering.net
 www.EasyEngineering.net

372 Control Systems

Z~[~ 2 ~l z
0
and
0

~l [~
0
[~
0
Y=
0
:Jz
[~ o 2]
ww = lIZ
Since none of the columns ofY contain all zero elements, and the eigenvalues of A are

w.E
distinct, the system is observable.

9.8.3 Duality Principle

asy
Kalman has introduced the duality principle to establish the relationship between controllability and
observability.
Consider system 1,

X =AX+Bu
E ngi
Y=CX
nee
and its dual defined by system 2,

Z =A* Z+C* V rin

W=B*Z g.n
where A*, B* and C* are conjugate transpose of A, B and C respectively.
The duality principle states that,
et
(i) if the pair (A, B) is controllable in system I, the pair (A*, B*) is observable in its dual
system 2.
and (ii) if the pair (A, C) is observable in system 1, the pair (A* C*) is controllable in its dual
system 2.

www.EasyEngineering.net
 www.EasyEngineering.net

State Space Analysis of Control Systems 373

Problems
9.1 Obtain the state space representation of the electrical system shown in Fig. P 9.1.

2n

+
+
v IF

ww Fig. P.1

Take
w.E v=u and

asy
9.2 Obtain the state space representation for the mechanical system shown in Fig. P 9.2 taking
the displacement and velocity of the mass as state variables

E ngi
B
nee
rin
x
g.n
Fig. P.2
et
9.3 The block diagram ofa position control system is shown in Fig. P 9.3. Obtain the state space
representation of the system.

Fig. P.3

Choose the state variables as

www.EasyEngineering.net
 www.EasyEngineering.net

374 Control Systems

9.4 Obtain the state space representation of the system shown in Fig. P 9.4.
Take c = xl' C = x 2 and m = x3

u 10 m C(s)
s+4 S2 +s+1

ww
Fig.P.4

9.5 Obtain the state space representation in,

w.E
(i) Companion form or Bush's form
(li) Diagonal/Jordon form
for the following systems. asy
(a)

(b)
y x + 23 x + 15 = u
+9

Y+ 9 x + 23 x + 15 = ti
E- 2 U + 4u ngi
9.6 Obtain the state space representation in,
nee
(i) Phase variable form
(ii) Jordon's form
and
rin
for the systems whose transfer functions are given by,
g.n
(a)
s+1
s(s + 4)(s + 5)
(b) s+5
(s + 1)2(S + 4)
(c)
10
(s + 1)3 et
9.7 Obtain the transfer functions of the following systems.

1. X= [~ ~ ~j [~j
-2 -3 -4
X+
1
u 2. X= [~ ~ ~j [~j
-4 -6 -8
X+
1
u

y = [1 0 0] X Y = [-2 4 0] X + 2u

3. X= n-~ J [~j
y = [-1 2 4] X
X+ u

www.EasyEngineering.net
 www.EasyEngineering.net

State Space Analysis of Control Systems 375

9.8 Find the transfer function of the systems given below.

2. X~[! -; ~]x+[_~]u
y = [1 0] X y = [1 1 -1] X
9.9 Transform the following matrices into Diagonal/Jordon form representations.

[~-1 -3~ -3~]

0 1
(i) A= 0 0 (ii) A=

ww
[
6 5

w.E
(iii) A =
-2
- ~ (iv) A= [-~ -~ ~]
[
asy 1 -1 -2

E ngi (vi)
-1 0]
-2 0

n
9.10 Find the state transition matrix for the following systems. eer
o -1

X = [~ ~] X X =[ 1
i n[~g.~ ~]
(i) (ii)

9.11 Obtain the solution of the state equation

-1 -2
0] X (iii) X=
net
-2 -5 -4
X

X= [~ _~] X + [~] u
Xo
[1 l]T and u(t) is a unit step input.
=
9.12 Obtain the Jordon's cononical form of the following system and obtain its solution for a step
input

X = [~~ ~] [~]
6 5 -2
X+
1
u

X(o) = [1 0 O]T

www.EasyEngineering.net
 www.EasyEngineering.net

376 Control Systems

9.13 Determine whether the following systems are completely state controllable and observable
using (i) Kalman's test and (ii) Gilbert's test.

Y = [1 1 0] X

Y=[~ 0 ~]X

ww[~ ~ _~] {l
w.E
(e) X= X
Y=[O 1 -l]X

(d) X=
a
[~ ~ Jx+Hu syE Y = [0 1 -1] X

n
-1 -3

gin
9.14 Determine output controllability of system in problem P. 9 .13 (b).

eer
ing
-djl- .ne
t

www.EasyEngineering.net