4 State Space Analysis

4.1 Introduction
Conventional methods are conceptually simple and require only a
reasonable number of compu- tations, but they are applicable only to
linear time-invariant systems having a single input and single output.
They don’t apply to nonlinear systems except in simple cases. The
conventional methods do not apply to the design of optimal and adaptive
control systems, which are mostly time varying and/or nonlinear.
A modern control system may have many inputs and many outputs, and
these may be interrelated in a complicated manner. The state-space
methods for the analysis and synthesis of control systems are best suited
for dealing with multiple input multiple output systems that are required
to be optimal in some sense.
The state-space method is based on the description of system equations
in terms of n first-order difference equations or differential equations,
which may be combined into a first-order vector- matrix difference
equations or differential equations.The use of vector-matrix notation
greatly simplifies the mathematical representation of the systems of
equations.

First lets define the common terms used in state-space representation.

State: The state of a dynamic system is the smallest set of variables (called
state variables) such that the knowledge of these variables at t = t0, together
with the knowledge of the input for t ≥ t0, completely determines the
behavior of the system for any time t ≥ t0. Note that the concept of state
is by no means limited to physical systems.

State Variables: are the variables making up the smallest set of

variables that determine the state of the dynamic system. If at least n
variables x1, x2, ..., xn are needed to completely describes the behavior of
a dynamic system, then such n variables are a set of state variables. Note
that state variables need not be physically measurable or observable
quantities. Variables that do not represent physical quantities and those
that are neither measurable nor observable can be chosen as state
variables.

State Vector: If n state variables are needed to completely describe the

behavior of a given system, then these n state variables can be
considered the n components of a vector x. Such a vector is called a state
1
vector. A state vector is thus a vector that determines uniquely the
system state x(t) for any time t ≥ t0, once the state at t = t0 is given and
the input u(t) for t ≥ t0 is specified.

State Space: The n-dimensional space whose coordinate axes consist of the x1
axis, x2 axis, . .
. ,xn axis is called a state space. Any state can be represented by a point in the
state space.

State-Space Equations: In state-space analysis we are concerned with three

types of variables

2
that are involved in the modeling of dynamic systems: input variables, output
variables, and state variables.

For time-varying (linear or nonlinear) discrete-time systems, the state

equation may be written as

x(k + 1) = f[x(k), u(k), k]

and the output

equation as
y(k) = g[x(k), u(k), k]

For linear time-varying discrete-time systems, the state equation and output
equation may be simplified to
x(k + 1) = G(k)x(k) + H(k)u(k)
y(k) = C(k)x(k) + D(k)u(k)

where
x(k) = n-vector (state vector) y(k) = m-vector
(output vector)
u(k) = r-vector (input vector) G(k) = n ∗ n matrix (state
matrix) H(k) = n ∗ r matrix (input matrix) C(k) = m ∗
n matrix (output matrix) D(k) = m ∗ r matrix (direct
transmission matrix)

The appearance of the variable k in the arguments of matrices G(k), H(k),

c(k), and D(k) implies that these matrices are time varying. If the variable
k does not appear explicitly in the matrices, they are assumed to be time
invariant, or constant. That is, if the system is time invariant, then the
last two equations can be simplified to

x(k + 1) = Gx(k) + Hu(k) (4.1)

y(k) = Cx(k) + Du(k) (4.2)

4.2 State Space Representation of Discrete-time systems

Canonical Forms for Discrete-Time State-Space Equations: Many
techniques are available for obtaining state-space representation of discrete-
time systems.
Consider the discrete-time system described by

y(k) + a1y(k − 1) + a2y(k − 2) + · · · + any(k − n) = b0u(k) + b1u(k − 1) + · · · + bnu(k −

n) (4.3)

3
where u(k) is the input and y(k) is the output of the system at the kth
sampling instant. Note that some of the coefficients ai (i = 1, 2, ..., n) and bj
(j = 0, 1, 2, ..., n) may be zero.Equation (5.3) can be written in the form of
the discrete transfer function as
Y (z) b0 + b1z−1 + · · · + bnz−n
=
U (z) 1 + a1z−1 + · · · (4.4)
+ an z−n

4
or
Y (z) b0zn + b1zn−1 + · · · +
U (z)= (4.5)
bn
zn + a zn−1 + · · · +
a
1 n

There are many ways to realize state-space representations for the

discrete-time system described by Equation (4.3),(4.4), or (4.5). Here we
will see four types of representation:
1. Controllable canonical form: The state-space representations for the
discrete-time sys- tem given by Equation (4.3),(4.4), or (4.5) may be
put in the form given by the following equations:
 x1(k + 1)   0 1 0 ··· 0 
x1(k)  
x2(k + 1)
   0 0 1 ··· 0   x (k)  0 0 
 . = . . . .  2.
+. u(k) (4.6)
. . . . . .   . 

 xn−1 (k + 
  0
 0 0 · · · 1     
  xn−1(k 0 
1) −an −an−1 −an−2 · · · −a1 ) 1
xn(k + 1)
xn(k)

5
  
x1(k

y(k) =
[b —
b
a —a ··· — a ) + b u(k) (4.7)
b b b ]  x2(k
. 
b  
) .
n n 0 n− n−1 1 1 0
1 0 0
xn(k)
Equations (4.6) and (4.7) are the state equation and output equation,
respectively. The state- space representation given by Equations (4.6)
and (4.7) is commonly called a controllable canonical form.
From Equation (4.4) we have
Y (z)
b0 + b1z−1 + · · · + bnz−n
=
U (z) 1 + a1z−1 + · · · + anz−n
(b1 − a1b0)z−1 + (b2 − a2b0)z−2 + · · · + (bn − anb0)z−n
= b0 + 1 + a z−1 + a z−2 + · · · + a z−n
1 2 n
This equation can be written as follows:
(b1 − a1b0)z−1 + (b2 − a2b0)z−2 + · · · + (bn − anb0)z−n
Y (z) = b0U (z) 1 + a1z−1 + a2z−2 + · · · −n U (z) (4.8)
+ z
+ an

Let us define
ˆ (b1 − a1b0)z−1 + (b2 − a2b0)z−2 + · · · + (bn − anb0)z−n
Y (z) = 1 + a z−1 + a z−2 + · · · −n U (z) (4.9)
+a z
1 2 n

Then Equation (4.8) becomes

Y (z) = b0U (z) + Yˆ (z) (4.10)
Let us rewrite Equation (4.9) in the following form:

(b − a b )z−1 + Yˆ (z) = U (z)

(b — a b )z−2 + · · · — a b )z−n 1 + a z−1 + a z−2 + · · Q(z)=
+ (b ·+a z−n
1 1 2 2 n n 0 1 2 n
0 0

6
 From this last equation the following two equations may be obtained:

Q(z) = −a1z−1Q(z) − a2z−2Q(z) − · · · − anz−nQ(z) + U (z) (4.11)

an
d
Yˆ (z) = (b1 − a1b0)z−1Q(z) + (b2 − a2b0)z−2Q(z) + · · · + (bn − anb0)z−nQ(z)
(4.12)

Now we define the state variables as

follows:
X1(z) = z−nQ(z)
X2(z) =
(4.13)
z−n+1Q(z)
.
.
Xn−1(z) =
z−2Q(z)
Xn(z) = z−1Q(z)

Then clearly we
have zX1(z) = X2(z)
zX2(z) = X3(z)
.
.
zXn−1(z) = Xn(z)
in terms of difference equations, the preceding n − 1 equations become

x1(k + 1) =
x2(k)
x2(k + 1) = (4.14)
x3(k)
.
.
xn−1(k + 1) =
xn(k)

By substituting Equations (4.13) into Equation (4.11). we obtain

zXn(z) = −a1Xn(z) − a2Xn−1(z) − · · · − anX1(z) + U (z)

which may be transformed into a difference equation:

xn(k + 1) = −anx1(k) − an−1x2(k) − · · · − a1xn(k) + u(k)

(4.15) Also, Equation (4.12) can be rewritten as follows:

Yˆ (z) = (b1 − a1b0)Xn(z) + (b2 − a2b0)Xn−1(k) + · · · + (bn − anb0)X1(z)

7
By use of this last equation, Equation (4.10) can be written in the form

y(k) = (bn − anb0)x1(k) + (bn−1 − an−1b0)x2(k) + · · · + (b1 − a1b0)xn(k) + b0u(k)

(4.16)

Combining Equations (4.14) and (4.15) results in the state equation

given by Equation (4.6). The output equation, Equation (4.16), can be
rewritten in the form given by Equation (4.7).

8
2. Observable canonical form: The state-space representations for the
discrete-time sys- tem given by Equation (4.3),(4.4), or (4.5) may be
put in the form given by the following equations:
     
x1(k + 1) 0 0 ··· 0 −an   x1(k) bn − anb0
   0   
 x2(k + 1)  1
· · 0 · 0 0 −an−1   x2(k)   bn−1 − an−1b0 
  
 . = . . . .  + .  u(k)
. .
. . . . . . . .
      
 x n− (k + 1) 0 0 ··· 1 −a2   xn− (k)   b2 − a 2 b0 
1  1
xn(k + 0 0 0 −a xn(k) b1 − a 1 b0
1) ·· 0 1 1 (4.1
· 7)

9
  
x1(k
)
 
y(k) = [0 0 · · · 0 x2(k
1]   + b0u(k) (4.18)
 ) 
 . 
.
xn−1(k)
xn(k)
Equations (4.17) and (4.18) are the state equation and output
equation, respectively. The state-space representation given by
Equations (4.17) and (4.18) is commonly called a observ- able canonical
form. From Equation (4.4) we have
Y (z)
b0 + b1z−1 + · · · + bnz−n
=
U (z) 1 + a1z−1 + · · · + anz−n
Y (z)−b0U (z)+z−1[a1Y (z)−b1U (z)]+z−2[a2Y (z)−b2U (z)]+· · ·++z−n[anY (z)−bnU
(z)] = 0 or

Y (z) = b0U (z)+z−1 b1U (z) − a1Y (z) + z−1 b2U (z) − a2Y (z) + z−1 [b3U (z) − a3Y (z)
}
+···]
(4.19)
Now define the state variables as follows:

Xn(z) = z−1[b1U (z) − a1Y (z) +

Xn−1(z)]
Xn−1(z) = z−1[b2U (z) − a2Y (z) + (4.20)
Xn−2(z)]
.
.
X2(z) = z−1[bn−1U (z) − an−1Y (z) +
X1(z)]
X1(z) = z−1[bnU (z) − anY (z)]

Then Equation (4.19) can be written in the form

Y (z) = b0U (z) + Xn(Z) (4.21)

By substituting Equation (4.21) into Equation (4.20) and multiplying

both sides of the

10
equation by z, we obtain

zXn(z) = Xn−1(z) − a1Xn(z) + (b1 − a1b0)U (z)

zXn−1(z) = Xn−2(z) − a2Xn(z) + (b2 − a2b0)U (z)
.
.
zX2(z) = X1(z) − an−1Xn(z) + (bn−1 − an−1b0)U (z)
zX1(z) = −anXn(z) + (bn − anb0)U (z)

Taking the inverse z transforms of the preceding n equations and

writing the resulting equa- tions in the reverse order, we obtain

x1(k + 1) = −anxn(k) + (bn − anb0)u(k)

x2(k + 1) = x1(k) − an−1xn(k) + (bn−1 − an−1b0)u(k)
.
.
xn−1(k + 1) = xn−2(k) − a2xn(k) + (b2 − a2b0)u(k)
xn(k + 1) = xn−1(k) − a1xn(k) + (b1 − a1b0)u(k)

Also, the inverse z transform of Equation (4.21) yields

y(k) = xn(k) + b0u(k)

Rewriting the state equation and the output equation in the

standard vector-matrix form gives Equation (4.17)and (4.18),
respectively

3. Diagonal canonical form: If the poles of the discrete transfer function

given by Equation (4.3),(4.4), or (4.5) are all distinct, then the state-
space representation may be put in the diagonal canonical form as
follows:
  
x1(k + 1) 0 ··· x21(k)   
  
  0 1
p 0
 x2(k + 1)  0 0  x (k) 1
1
 ···  
 p2
0  u(k)
 .
.
 =  . . .  
.
+ . (4.22)
. .
.
.
.
.   . 
    
 xn−1 (k +   0 0 ·· pn− 0  xn−1(k 
1) 0 0 · 1 pn ) 1
xn(k + 1) ·· 0 xn(k)

11
  
1 


 x 1(k
) 
y(k) = [c c ··· + b u(k) (4.23)

 x2(k)
c ] .
1 2 n 0
 . 
xn(k)
The system discrete transfer function can be modified as follows:
Y (z)
U (z)= b0nz + b1n−1
zn−1 + · · · + bn
n

z +a z +···+a

12
 1 n

(b1 − a1b0)z + (b2 − a2b0)z + · · · + (bn − anb0)

n−1 n−2

= b0 + (z − p1)(z − p2) · · · (z − ) (4.24)

pn

13
 Since all poles of the discrete transfer function Y (z)/U (z) are distinct, Y
(z)/U (z) can be expanded into the following form:
Y (z)
= b0 c1 c2 + · · · + c n
U (z) +z − p +z − p (4.25)
1 2 z − pn
Equation (4.25) can be written in the form
c c c
Y (z) = b U (z) + 1 U (z) + 2 U (z) + · · · + n U (z) (4.26)
0
z − p1 z − p2 z − pn
Let us define the state variables as
follows:
1 U
X1(z) =
p z− 1 (z)
1
X2(z) = U (4.27)
p z− 2 (z)
.
.
1
Xn(z) = U (z)
z − pn
Then Equation (4.27) can be rewritten as

zX1(z) = p1X1(z) + U
(z)
zX2(z) = p2X2(z) + U (4.28)
(z)
.
.
zXn(z) = pnXn(z) + U
(z)

Also, Equation(4.26) can be written as

Y (z) = b0U (z) + c1X1(z) + c2X2(z) + · · · + cnXn(z)

(4.29) The inverse z transforms of Equations (4.28) and (4.29) become

x1(k + 1) = p1x1(k) +
u(k)
x2(k + 1) = p2x2(k) + (4.30)
u(k)
.
.
xn(k + 1) = pnxn(k) +
u(k)

an d
14
 y(k) = c1x1(k) + c2x2(k) + · · · + cnxn(k) + b0u(k) (4.31)
Rewriting the state equation and the output equation in the form of
vector-matrix equations, we obtain Equations (4.22) and (4.23).

15
 4. Jordan canonical form: If the pulse transfer function given by
Equation (4.3),(4.4), or (4.5) involves a multiple pole of order m at z = p1
and all other poles are distinct, then the state equation and output
equation may be given as follows:

      
x1(k + 1) p1 0 ··· 0 |
0 ·· 0 x1(k) 0
·
1
 x2(k +   p1 ··· 0 |
0 ·· 0 x2(k 0
1) . . | . ) .
. . · .  .
 .  0 1 . .
  1 
.   0 . . · · · p1 |
. 0  .
u(k)
 xm(k + = . . xm(k)
0 ·· +
1) 0 
·   1 
 .
0 .
−−−−− − − − − − |
− − −−− −
  − 
  x 
 xm+1(k + ··· 0 ··· . (k)
|
  0 m+1
 1) .  .
 0 0 Pm+1 0  
 
   . . . | . .   
. . . . . . .   .
xn(k + 1) 0 0 ··· 0 |
0 ··· xn(k) 1
(4.32)
pn
0
 
x1(k)
y(k) = [c c 
···  x2(k) + b u(k) (4.33)
c ]
1 2 n  .  0
 . 
xn(k)

4.3 Solution of Discrete Time State Space Equation

In general, discrete-time equations are easier to solve than differential
equations because they can be solved easily by means of a recursion
procedure. The recursion procedure is quite simple and convenient for
digital computations.
Consider the following state equation and output equation:

x(k + 1) = Gx(k) + Hu(k) (4.34)

y(k) = Cx(k) + Du(k) (4.35)

The solution of Equation (4.34) for any positive integer k may be obtained
16
directly by recursion, as follows:
x(1) = Gx(0) + Hu(0)
x(2) = Gx(1) + Hu(1) = G2x(0) + GHu(0) + Hu(1)
x(3) = Gx(3) + Hu(3) = G3x(0) + G2Hu(0) + GHu(1) + Hu(2)
.
.

By repeating this procedure, we obtain

k−1
Σ
x(k) = G x(0) + Gk−j−1Hu(j),
k
k = 1, 2, 3, ... (4.36)
j=0

17
Clearly, x(k) consists of two parts, one representing the contribution of
the initial state x(0) and the other the contribution of the input u(j),
where j = 0, 1, 2, ..., k − 1. The output y(k) is given by
Σ
k−1
y(k) = CGkx(0) + C Gk−j−1Hu(j) + Du(k) (4.37)
j=0

State Transition Matrix. Notice that it is possible to write the solution of

the homogeneous state equation
x(k + 1) = Gx(k) (4.38)

as
x(k) = Ψ(k)x(0) (4.39)
where Ψ(k) is a unique n ∗ n matrix satisfying the condition

Ψ(k + 1) = GΨ(k), Ψ(0) = I (4.40)

Clearly, Ψ(k) can be

given by
Ψ(k) = Gk (4.41)

From Equation (4.39), we see that the solution of Equation (4.38) is

simply a transformation of the initial state. Therefore, the unique matrix
Ψ(k) is called the state transition matrix. It is also called the fundamental
matrix. The state transition matrix contains all the information about the
free motions of the system defined by Equation (4.38). In terms of the
state transition matrix Ψ(k), Equation (4.36) can be written in the form
k−1
Σ
x(k) = Ψ(k)x(0) + Ψ(k − j − 1)Hu(j) (4.42)
j=0

k−1
Σ
= Ψ(k)x(0) + Ψ(j)Hu(k − j − 1) (4.43)
j=0

Substituting Equation (4.42) or Equation (4.43) into Equation (4.37), the

following output equation can be obtained:
k−1
Σ
y(k) = CΨ(k)x(0) + C Ψ(k − j − 1)Hu(j) + Du(k) (4.44)
j=0

k−1
Σ
y(k) = CΨ(k)x(0) + C Ψ(j)Hu(k − j − 1) + Du(k) (4.45)
j=0

z Transform Approach to the Solution of Discrete-Time State Equations:

We next present the solution of a discrete-time state equation by the z
transform method. Consider the discrete system described by
x(k + 1) = Gx(k) + Hu(k) (4.46)

18
Taking the z transform of both sides of Equations (4.46), we get

zX(z) − zx(0) = GX(z) + HU (z)

where X(z) = Z[x(k)] and U (z) = Z[U (k)]. Then

(zI − G)X(z) = zx(0) + HU (z)

By multiplying both sides of this equation by (zI − G)−1, we obtain
X(z) = (zI − G)−1zx(0) + (zI − G)−1HU (z) (4.47)

Taking the inverse z transform of both sides of Equation (4.47) gives

x(k) = Z−1 (zI − G)−1z x(0) + Z−1 (zI − G)−1HU (z) (4.48)

Comparing Equation (4.36) with Equation (4.48), we obtain

Gk = Z−1 (zI − G)−1z (4.49)

and
Σ
k−1

Gk−j−1Hu(j) = Z−1 (zI − G)−1HU (z) (4.50)

j=0

where k = 1, 2,
3, ...
Notice that the solution by the z transform method involves the process
of inverting the matrix (zI − G), Which may be accomplished by analytically
means or by use of a computer routine. The solution also requires the
inverse z transforms of (zI − G)−1z and (zI − G)−1HU (z).
Example: Obtain the state transition matrix of the following discrete-time
system:

x(k + 1) = Gx(k) + Hu(k)

y(k) = Cx(k)

whe " # " #

re 1 1
G= , H= , C = [1 0]
− 1
1

0
−0.1
6
Then obtain the state x(k) and the output y(k) when the input u(k) = 1 for k
= 0, 1, 2, ... Assume that the initial state is given by
" # " #
x1(0) 1
x(0) = =
x2(0
19 −
) 1
From Equation(4.41) and (4.49) the state transition matrix Ψ(k) is

Ψ(k) = Gk = Z−1 (zI − G)−1z

20
Therefore, we first obtain (zI − G)−1:
" #−1
z −1
(zI − G) −1
=
0.1 z +
6 1
 
z+1 1
(z + 0.2)(z + (z + 0.2)(z +
= 0.8) 
0.8) 
 

−0.16  z
(z + 0.2)(z + (z + 0.2)(z + 
 0.8) 0.8)
4/3 −1 / 3 5/3 −5 / 3
+ +
z + 0.2 z + z+ z + 0.8
=  0.8 0.2

 
 −0.8/3 0.8/3 −1/3 4/3 
+ +
z+ z+ z+ z+
0.2 0.8 0.2
The state transition matrix Ψ(k) is now obtained 0.8as follows:

Ψ(k) = Gk = Z−1 (zI − G)−1z

 
4 z 1 z 5 z 5 z
 − − 3 z + 0.8
= Z−1 3z+ 3z+ 3z+ 
0.2 0.8 0.2
 
 0.8 
0.8 z 1
3z+ z 34 z +
z
− 3 z+ − 0.2 + 0.8
z
0.2 +
 3 z+ 
4 1 0.8 5 5
(−0.2) − k
(−0.8)k (−0.2)k − (−0.8)k
3 3 3 3
  (4.51)
=  
0.8 0.8 1 4 
− (−0.2) +
k
− 2) + (−0.8
3 3 3 k 3 )k
(−0.
Equation (4.51) gives the state transition matrix.
(−0.8)k
Next, compute x(k). The z transform of x(k) is given by
Z[x(k)] = X(z) = (zI − G)−1zx(0) + (zI − G)−1HU (z)
= (zI − G)−1[zx(0) + HU (z)]
Since
1 z
U (z) = =
we 1 − z−1 z−1
obtain  

z 
  z2
z−1 z−1
z    
 He nce zx(0) + HU (z) =

21
   =
2
+ 2z
−z
 +  z  
−

z−1 

z
z−1
X(z) = (zI − G)−1[zx(0) + HU (z)]

22
 
(z2 + 2)z

(z + 0.2)(z + 0.8)(z −
=  1)


(−z2 + 1.84z)z 
(z + 0.2)(z + 0.8)(z −
1)
 −17 z 22 z 25 z 
+ +
 6 z + 0.2 9 z+ 18 z − 1 
= 0.8

 3.4 z 
−17.6
z+ 7 z z− 
18
6 0.8 +1
z
+
z+ 9
Thus, the state vector x(k)0.2 is given by
 
17 22 25
− (−0.2)k + (−0.8)k +

x(k) = Z [X(z)] =
−1
6 9 18  
 3.4 17.6 7 
(−0.2)k − (−0.8)k +
6 9 18
Finally, the output y(k) is obtained
as follows:
 
17 22 25
− (−0.2) +
k
(−0.8) +
k

y(k) = Cx(k) = [1 0] 6 9 18 
 
 3.4 17.6 7 
(−0.2)k − (−0.8)k +
6 9 1
17 22 25 8
=− (−0.2)k + (−0.8)k +
6 9 18

23