Chapter 2 Systems & Control Theory

C.S.Tong 41 HKBU
Time-varying case (A, B are now functions of t)
Ref: B[Ch9]
The homogeneous part of the state equation is
( ) ( ) ( ) x t A t x t !
Suppose we know n linearly independent solutions to the above
equation, call them, say,
1 2
, ,
n
" , then we can construct the
Wronskian (or fundamental) matrix
( ) ( ) ( ) ( )
1 2
, , ,
n
t t t t 1
]
"
Example
( )
1 1 1 1
2 2 2
1 0 0
0 2 2
t
x x x e x
A
t x t x x
_



,
!
!
! 2 linearly independent solutions are
( )
2 2 1 2
0 0
,
0
0
t
t
t t
e
e
t
e e

_ _ _



, , ,
"Exercise: check that ( ) ( ) ( ), 1, 2
i i
t A t t i !
Chapter 2 Systems & Control Theory
C.S.Tong 42 HKBU
#The Fundamental matrix is not unique since we can
choose any set of linearly independent combinations of
1

and
2

#The transition matrix ( )

0
, t t can be obtained from the
Wronskian, viz.,
( ) ( ) ( )
1
0 0
, t t t t


Proof
(i) clearly, ( ) , t t I
(ii)
( ) ( ) ( )
1
0 0
,
d d
t t t t
dt dt

( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
1
1 2 0
1
1 2 0
, , ,
, , ,
n
n
t t t t
A t A t A t t

1
]
1
]
! ! ! "
"
(since each
i
satisfies
i i
A ! )
( ) ( )
( )
1
0
0
,
A t t
A t t

Chapter 2 Systems & Control Theory

C.S.Tong 43 HKBU
# ( )
0
, t t describes the transition from time t
o
to time t, i.e.
( ) ( ) ( )
0 0
, x t t t x t
#Usually, t
0
= 0, and it is more convenient to write ( ) t
instead of ( ) , 0 t
Example
( ) ( )
( )
2
0 0
2 2 2 2
0 0
1
0
0 1 0
0 2
0
0 0 0
,
0 0 0
t
t
t t t t
t t t t
e
A t t
t
e
e e e
t t
e e e

_
_



,
,
_ _ _


, , ,
Check:
( ) ( )
( )
0
2 2
0
0 0
0
0 1 0
, ,
0 2
0 2
,
t t
t t
e
d
t t t t
t dt
te
A t t

_
_



,
,

#In general, time-varying state equations are difficult to

solve since the Wronskian is difficult to obtain
Chapter 2 Systems & Control Theory
C.S.Tong 44 HKBU
#In practice, use numerical techniques or exploit special
properties of certain class of state matrices
Example: suppose A(t) commutes with ( )
0
t
t
A d

, i.e.
( ) ( ) ( ) ( ) ( ) ( )
0 0
t t
t t
A t A d A d A t , then
( )
( )
0
0
,
t
t
A d
t t e

Proof
(i) clearly, ( )
0
0 0
, t t e I ,
(ii) ( ) ( ) ( )
0
t
t
t
d
A d A A t
dt

,
(fundamental theorem of calculus)
(iii)
( )
( )
0 0
0
( )
!
t
t
t k
A d t
k
A d
d d
e
dt dt k

' ;


( )
( ) ( )
0
0
2
2 !
t
t
t
t
A d
d
I A d
dt

+ + +
' ;


"
Chapter 2 Systems & Control Theory
C.S.Tong 45 HKBU
( ) ( ) ( )
( ) ( )
( ) ( )
0
0
0
3 1
1
3
0 ( ) ( )
3 !
( )
!
t
t
t
t
k
t
t
A d
A t A t A d A t
k A d
A t
k

+ + + +

+ +
"
"
( )
( ) ( ) ( ) ( )
( )
( )
0 0
0
0
2
0
( )
2! !
( )
( ) ,
t
t
k
t t
t t
t
t
A d
A d A d
A t I A d
k
A t e
A t t t

+ + + +
' ;

" "
Example
1 0
0 2
A
t
_

,
Let ( )
0
0
2 2
0
0
0
t
t
t t
H A d
t t

_

,
Since A, H are both diagonal, they commute:
( )
0
0
2 2 2 2
0
0
0
0
1 0
0 2 0 2
0
t t
t t
HA AH
t t t t
t t
_

_
_

,
,
,
Hence using the theorem above, we have
Chapter 2 Systems & Control Theory
C.S.Tong 46 HKBU
( )
0
2 2
0
0
0
,
0
t t
H
t t
e
t t e
e

_


,
Once we know the transition matrix, then the solution to the
state equation x Ax Bu + ! is
( ) ( ) ( ) ( ) ( )
0
0 0
, ,
t
t
x t t t x t t Bu d +

Proof by direct substitution

( )
( )
( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) { } ( )
( ) ( )
0
0
0
0
0
0 0
0 0
,
,
, , ,
, ,
t
t
t
t
t
t
t
d t t
d d
x t x t t Bu d
dt dt dt
d
A t t x t t Bu d t Bu
dt
A t t x t t Bu d Bu t
Ax t Bu t

_
+

,
_
+ +

,

+ +
+
Chapter 2 Systems & Control Theory
C.S.Tong 47 HKBU
Discrete-time state equation
x(k+1) = A(k) x(k) + B(k) u(k) (*)
! As before, we use the concept of transition matrix ( , ) k j
! Since by definition ( ) ( , ) ( ) x k k j x j , we have
(0) (0) (0, 0)
(1) (0) (0) (1, 0) (0)
(2) (1) (1) (2,1) (1)
(2) (1) (0) (0) (2, 0) (1) (0)
(3) (2) (2)
(3) (2) (2)
(3)
x x I
x A x A
x A x A
x A A x A A
x A x
x A x
x

#
By induction, we have
1
( , ) ( ) ( 1) ( 2) ( )
p k
p j
k j A p A k A k A j

$ ()
! Clearly, ( , ) k j satisfies the homogeneous state equation:
1
( 1, ) ( ) ( ) ( ) ( ) ( , )
p k p k
p j p j
k j A p A k A p A k k j


+

Chapter 2 Systems & Control Theory
C.S.Tong 48 HKBU
! As before, once the transition matrix ( , ) k j is found, we
can immediately write down the solution to the general
discrete-time state equation (*)
1
0
( ) ( , 0) (0) ( , 1) ( ) ( )
k
j
x k k x k j B j u j

+ +

(**)
Proof by direct substitution
As before, we show that the above formula for x(k)
satisfies the state equation (*).
0
1
0
( 1) ( 1, 0) (0) ( 1, 1) ( ) ( )
( ) ( , 0) (0) ( 1, 1) ( ) ( )
( 1, 1) ( ) ( )
( ) ( ) ( ) ( )
j k
j
j k
j
x k k x k j B j u j
A k k x k j B j u j
k k B k u k
A k x k B k u k

+ + + + +
+ + +
+ + +

Chapter 2 Systems & Control Theory

C.S.Tong 49 HKBU
Time-invariant case: A(k) = A, B(k) = B, (A, B constant)
! From (), the transition matrix is
( , ) ( , 0) ( )
k j
k j k j k j A


! From (**), the general solution to the state equation is
1
0
1
1
0
( ) ( , 0) (0) ( , 1) ( )
(0) ( )
(0) ( 1) ( )
k
j
k
k k j
j
k
x k k x k j Bu j
A x A Bu j
A x k Bu k

+ +
+
+

! The remaining problem is how to evaluate

k
A efficiently?
[direct matrix multiplication is very time-consuming for
large n n matrices]
! As in the continuous case (when we evaluate e
At
), we have
3 methods: Similarity Transform, Cayley Hamilton method,
and Z-transform
Chapter 2 Systems & Control Theory
C.S.Tong 50 HKBU
(1) Similarity Transformation
First find matrix P to diagonalize A, i.e.
1
1 2
( , , , )
n
P AP diag

$
where
n
, , ,
2 1
$ are eigenvalues of A. Then from
Chapter 1, we have
1 1
1 2
( , , , )
k k k k k
n
A P P Pdiag P

$
Note that not all matrices are diagonalizable. The next
best thing is that all matrices can be transformed to a
Jordan Normal Form
1
, J. I.e. there exists matrix P s.t.
1 1 1 k k k k
P AP J P A P J A PJ P


k
J is very close to being a diagonal matrix but is more
difficult to evaluate than
k
, nevertheless it is still better
than evaluating A
k
directly.
1
Outside curriculum
Chapter 2 Systems & Control Theory
C.S.Tong 51 HKBU
(2) Cayley-Hamilton Method
As before, every power of an n n matrix A can be
written as a linear combination of
2 1
, , , ,
n
I A A A

" ; i.e.
1
0
n
k i
i
i
A A

, for some (unknown) coefficients
j
.
Now let
j
e be the eigenvector corresponding to the
eigenvalue
j
, i.e.
m m
j j j j j j
Ae e A e e . Then we
have
1 1
0 0
n n
k i k i
j i j j j i j j
i i
A e A e e e




.
If all the eigenvalues are distinct, then we can replace the
matrix A in the above expansion by all its eigenvalues:
1
0
, 1, 2, ,
n
k i
j i j
i
j n

$ .
This is a set of n simultaneous linear equations in
i
and
can be solved for
i
using e.g. Gaussian elimination, just
as in the continuous case.
Chapter 2 Systems & Control Theory
C.S.Tong 52 HKBU
(3) Z-transform(discrete version of Laplace transform)
Sampling (Analog to Digital conversion)
$ With the advent of cheap, powerful micro processors,
digital signal processing is more and more important.
$ Need to convert an analog signal to a discrete one by
sampling at regular intervals
u(t)
0 T 2T . kT t
{u(0), u(T), u(2T), ,u(kT), } are the discrete sampled
values of u(t). Symbolically:
u u* sampled signal at intervals T
T
For convenience, usually assume T = 1 time unit.
Chapter 2 Systems & Control Theory
C.S.Tong 53 HKBU
Definition: the Z-transform of u(t) is defined as
U(z) = Z[u(t)]
0 k

% u(kT) z
k
Examples

0
[1]
1
k
k
z
Z z
z

u(t) = t u(kT) = kT

0 0
[ ] [ ]
k k
k k
Z t Z kT kTz T kz





but
1
0 1 0
1
1
k k k
k k k
d z d
z kz kz
dz z dz z



_
_

,
,

2
[ ]
1 ( 1)
d z Tz
Z t Tz
dz z z
_



,
( ) ( )
t kT
u t e u kT e


( )
0 0
k
kT kT k T
T
k k
z
Z e e z e z
z e

1
]

Chapter 2 Systems & Control Theory
C.S.Tong 54 HKBU
Inverse Z.T. (from now on, let T = 1)
The inverse Z.T. can be obtained simply by expanding U(z)
into a power series of z:
0
( )
k
k
k
U z a z

then by definition of Z.T., we can just read off u(k) from the
Laurent expansion of U(z), i.e.
( )
k
u k a
From definition, it is clear that Z-transform obey very
similar theorems as Laplace transform
#Linearity theorem
(i) Z [ r(k)] = Z [r(k)] , constant
(ii) Z [r
1
(k) r
2
(k)] = Z [r
1
(k)] Z [r
2
(k)]
(proof: trivial)
Chapter 2 Systems & Control Theory
C.S.Tong 55 HKBU
#Time Translation & Time Delay (c.f. differentiation &
integration in the continuous case)
Given a series r (k), k = 0, 1, 2, we are interested in
series of the form r(kn) (time delay) and r(k + n), for
constant n. As in the continuous case, we shall implicitly
assume that all series are zero for negative indices, i.e.
( ) 0, for 0 r k k < .
k r (k) r (k - 2) r (k + 2) r (k - n) r (k + n)
-3 0 0 0 0 r (n - 3)
-2 0 0 r (0)
#
r (n - 2)
-1 0 0 r (1)
#
r (n - 1)
0 r (0) 0 r (2)
#
r (n)
1 r (1) 0 r (3)
# #
2 r (2) r (0)
# # #
3 r (3) r (1)
# # #
# # # #
0
#
n r (n) r (n - 2) r (n + 2) r (0)
#
n + 1 r (n + 1) r (n - 1) r (n + 3) r (1)
#
# # # # # #
Chapter 2 Systems & Control Theory
C.S.Tong 56 HKBU
r(k n) r(k + n)
0 n k -n 0 k
time-delay time-translation
Z [r(k n)] = z
n
Z [r (k)] , fixed n
Proof
0
[ ( )] ( )
, sub
, ( ) 0 for 0
[ ( )]
k
k
n
Z r k n z r k n
k k n
r k k
z Z r k

<

&
Remarks
For n = 1, the effect of time-delay by 1 time unit is to
multiply the Z-transform by z
1
. Hence z
1
is often used to
represent the unit time-delay operator (c.f. Ch 1, p.15).
Chapter 2 Systems & Control Theory
C.S.Tong 57 HKBU

1
0
[ ( )] [ ( )] ( )
n
n k
k
Z r k n z Z r k z r k

_
+

,

Proof
0
1
0
[ ( )] ( )
, sub
[ ( )] ( )
k
k
n
n k
k
Z r k n z r k n
k k n
z Z r k z r k

+ +
+

_


,

Examples: Let r(k) = k , R(z) = Z [r(k)] =

( )
2
1
z
z
Z [k + 1] = z (R(z) r(0))
( )
2
2
1
z
z

, since r(0) = 0
Z [k + 2] =

Chapter 2 Systems & Control Theory

C.S.Tong 58 HKBU
#Initial & Final Value Theorems (Let R(z) = Z [r (k)])
(i)
0
lim
k
r(k)
lim
z
R(z), (if limit exists)
(ii)
lim
k
r(k)
1
lim
z
(1 z
1
) R(z), (if limit exists)
#Complex translation
( ) ( )
kT T
Z e r k R ze
t
1
]
'
#Convolution theorem
Definition: we define convolution for discrete sequence
by
( )
1 2 1 2
0
1 2 2
0
( ) ( ) ( )
( ) ( ), since ( ) 0 for 0
N
k
k
r r N r k r N k
r k r N k r m m

<

%
then
1 2 1 2
[ ] [ ]. [ ] Z r r Z r Z r
Chapter 2 Systems & Control Theory
C.S.Tong 59 HKBU
Proof
( ) ( )
( ) ( )
1 2 1 2
0 0
1 2
0
1 2
[ ]
interchange order of
,
double summation
, by the time delay theorem
[ ]
[ ]. [ ]
j
j k
k
k
Z r r z r k r j k
r k z Z r
Z r Z r

_


,

Corollary
Since
1 2 2 1 1 2 2 1
[ ]. [ ] [ ]. [ ] Z r Z r Z r Z r r r r r
Z.T. can be looked up in tables
Z.T. can be used to solve linear difference equation (cf
solution of differential equation using Laplace
Transform)
Chapter 2 Systems & Control Theory
C.S.Tong 60 HKBU
Example 1
u(k+1) + u(k) = 0 , u(0) = a
Take Z. T. of the above equation, we have
(time delay theorem)
[ ( 1)] [ ( )] 0
,
[ ( )]
1
( ) ( 1)
k
Z u k Z u k
az
Z u k
z
u k a
+ +

+

Example 2: Fibonacci series
r
n+2
= r
n
+ r
n+1
, r
0
= r
1
= 1 (*)
The first few terms are 1, 1, 2, 3, 5, 8, 13, 21, 34,
This series is related to many natural phenomena and has
important significance in arts, music, and architecture. [See
http://www.ee.surrey.ac.uk/Personal/R.Knott/Fibonacci/fibna
t.html]
Chapter 2 Systems & Control Theory
C.S.Tong 61 HKBU
We shall now use Z-transform to solve the Fibonacci difference
equation. Let R(z) = Z[r
n
]. Take Z-transform of equation (*),
and by using the Time-delay theorem, we have
2 1
0 1 0
2 2
2
( ) ( )
( 1)
( )( )
z R r r z z R r R
z z R z
z
R
z z

+



where
1 5
2

+
,
1 5
2

Re-arranging and expanding the terms in power series, we have

1 1
1 1
0
1 1
5
5
n n
n
n
z
R
z z
z



+ +

_
_ _




, ,
,
_


,
Since
0
n
n
n
R r z

, (definition of Z-transform)
1 1
5
n n
n
r

+ +

Chapter 2 Systems & Control Theory
C.S.Tong 62 HKBU
Applying Z-transform to the Transition matrix
The Z-transform of the transition matrix: (k) = A
k
, where A is
n x n matrix, is
0 0
( ) [ ( )]
k
k k
k k
A
z Z k A z
z

_


,
2 3
2 3
1
1
( )
( ) ( )
A A A
I
z z z
A
I z I
z
A
z I z zI A
z

+ + + +

_


,
_


,
"
1
( ) ( ) z z zI A

can be evaluated efficiently using the
resolvent algorithm, as in the continuous case. The transition
matrix can be obtained by inverse Z-transform.
Note that by the time-delay theorem, we have
1
[ ( 1)] ( ) Z k zI A


Chapter 2 Systems & Control Theory
C.S.Tong 63 HKBU
Solution of state equation using Z-transform
Consider the time-invariant discrete time state equation
x(k+1) = A x(k) + B u(k)
Let X(z) = Z[x(k)], U(z) = Z[u(k)]. Taking Z.T., we have
(time delay theorem)
[ ( 1)] [ ( )] [ ( )]
,
( )
[ ( )] [ ( )] (0) [ ( 1)] [ ( )]
Z x k AZ x k BZ u k
X z
Z x k Z k x Z k Z Bu k
+ +

+
Taking inverse Z.T. and using the Convolution Theorem:
1
1
0
( ) ( ) (0) ( 1) ( )
(0) ( )
k
k k j
j
x k k x k Bu k
A x A Bu j

+
+

(Same result as before! See p.49.)

Chapter 2 Systems & Control Theory
C.S.Tong 64 HKBU
Data hold [Ref: Kuo Chapter 3.12]
In many applications, there is a need for digital-to-analogue
conversion to generate analogue output. A data hold is just
such an interface. The simplest example is the zero-order
hold (z.o.h) which holds the magnitude of the impulse at time
t = kT until the next impulse at t = (k+1)T.
T
z.o.h. h(t) analogue output
r(t) r*
r(t) r* h(t)
t t t
T 2T T 2T T 2T
Thus the z.o.h. output h(t) is a staircase approximation of r(t).
Higher order holds use higher order polynomials to fit the
data points. E.g. the first-order hold join the sample points by
straight lines to give a continuous approximation.