Chapter 3

Impulse Sampling
If a continuous-time signal x(t) is sampled in a periodic manner, mathematically the sampled signal may be represented by

x (t) =

x(t) (t kT ) = x(t)

k=

(t kT )

(1)

k=

or
x (t) =

x(kT ) (t kT )

(2)

k=

x*(t) is the sampled version of the continuous-time signal x(t).

Low limits may be changed to k = 0, if x(t) = 0,
t<0

Carrier

Modulator

x(t)
Modulating
Signal

x*(t)
Output

Relationship between z transform and Laplace transform

Taking the Laplace transform of equation (2)
X (s) = L[x (t)] = x(0) L[(t)] + x(T ) L[(t T )] + x(2T ) L[(t 2T )] +
= x(0) + x(T )eT s + x(2T )e2T s +
=

x(kT )ekT s

k=0

Now, dene

eT s = z

so,
s=

1
ln z
T

X (s) |s= 1 lnz =

T

or

x(kT ) z k

k=0

X (s) |s= 1 lnz = X(z)

T

This shows how the z transform is related to the Laplace transform.

Note: the notation X(z) does not signify X(s) with s replaced by z, but
rather X (s = T1 lnz)
1

Data Hold
Data hold is a process of generating a continuous-time signal h(t) from a
discrete-time sequence x(kT). The signal h(t) during the time interval
kT t (k + 1)T may be approximated by a polynomial in as follows:
h(kT + ) = an n + an1 n1 + + a1 + a0
where
note:

0 T
h(kT ) = x(kT )

h(kT + ) = an n + an1 n1 + + a1 + x(kT )

If the data hold circuit is an nth -order polynomial extrapolator, it is called

an nth -order hold . It uses the past n + 1 discrete data x((k n)T ),
x((k n + 1)T ), , x(kT ) to generate h(kT + ).
ZERO-ORDER HOLD
If n = 0 in the above equation, we have a zero order hold so that
h(kT + ) = x(kT )

0 < T,

k = 0, 1, 2,

Transfer Function of ZOH

h(t) = x(0) [u(t) u(t T )] + x(T ) [u(t T )] u(t 2T )] +
x(2T )[u(t 2T ) u(t 3T )] +
=

x(kT ) [u(t kT )] u(t (k + 1)T )]

k=0

Now, L[u(t kT )] =
thus,

ekT s
s

ekT s e(k+1)T s
L[h(t)] = H(s) =
x(kT )
s
k=0
=

1 eT s

x(kT )ekT s
s
   k=0



Gh0 (s)
X (s)

Thus, transfer function of ZOH

=

1 eT s
H(s)
=
X (s)
s

FIRST-ORDER HOLD
h(kT + ) = a1 + x(kT ),

0 < T,

k = 0, 1, 2,

now
h((k 1)T ) = x((k 1)T )
so that
h((k 1)T ) = a1 T + x(kT ) = x((k 1)T )
or

x(kT ) x((k 1)T )

T
x(kT ) x((k 1)T )
,
h(kT + ) = x(kT ) +
T
a1 =

x (kT)

Input to FOH
T 2T

-T

3T

4T kT

h(t)

Output from FOH

0 <T

TRANSFER FUNCTION
Consider a unit step function input

x (t) =

u(kT ) (t kT ) =

k=0

h(t) = (1 +

(t kT )

k=0

(t T )
t
) u(t)
u(t T ) u(t T )
T
T

1
1
1 T s
1 T s
e
H(s) = ( + 2 )
e

2
s Ts
Ts
s
1 eT s
1 eT s
=
+
s
T s2
Ts + 1
= (1 eT s )
T s2

The Laplace transform of unit step

X (s) = L[u (t)] =

1
1 eT s

thus
Gh1 (s) =

Ts + 1
H(s)
= (1 eT s )2

X (s)
T s2

1 eT s
s

2

Ts + 1
T

Obtaining z transform of functions involving the term

1eT s
s

Suppose the transfer function G(s) follows a zero-order hold

G(s)

ZOH

X(s) =

1 eT s
G(s)
s


G(s)
X(z) = Z[X(s)] = (1 z ) Z
s

Suppose a transform function G(s) follows a rst-order hold

G(s)

FOH

X(s) = (

1 eT s 2 T s + 1
)
G(s)
s
T

X(z) = (1 z 1 )2 Z[

Ts + 1
G(s)]
T s2

Reconstructing Original Signals from Sampled Signals

Sampling Theorem:
where T is the sampling period, then if
Dene s = 2
T
s > 21
where 1 is the highest-frequency component present in the continuous-time
signal x(t), then the signal x(t) can be reconstructed completely from the
sampled signal x (t).
Intuitive proof
The frequency spectrum of a sampled signal x (t) is given by
1
X (j) =
T

k=

* See next page

X(j + js k)

Spectrum of Sampled Signal

Fourier series representation of a train of impulses

(t kT ) =

k=

n=

Cn ej(

2n
)t
T

(3)

where
Cn

1
=
T
1
T

T
2

T2

k=

T
2

T2

(t kT )ejn(

(t)ejn(

2t
)
T

2t
)
T

dt

dt

(t) is the only value in the integral limit range. So

1
T
(t) g(t) dt = g(0)

(4)

Cn =

note: by sifting property,

Thus, (3) & (4)



k=

(t kT ) =

1
T

n=

ej(

2n
)t
T

 Fourier series representation of sum of impulses

x (t) =

x(t) (t kT )

k=

where

s =

L {x (t)} =

1

x(t)
ejns t
T n=

esT dt

2
T

X (s) =
=

1

x(t) ejns t est dt
T n=

1

x(t) e(sjns)t dt
T n=

 Laplace transform with a change of variable

X (s) =

1

X(s jns )
T n=

Aliasing

If s < 21 , aliasing occurs

Consider frequency component 2 which falls in the overlap. It has two
components
|X (j2 )|

and

|X (j(s 2 ))|




from spectrum centered at = s

Therefore the frequency spectrum of the sampled signal at = 2 includes
components not only at frequency s but also at frequency s 2 (in general, at ns 2 , where n is an integer).
The frequency s 2 is known as an alias of 2 .




In general ns 2
Aliasing
Two frequencies x(t) and y(t) dier from each other by an integral multiple of s , sampling frequency T = 2
.
s
x(kT ) = sin(2 kT + )
y(kT ) = sin((2 + ns )kT + )
= sin(2 kT + 2kn + )
= sin(w2 kT + )

Two signals of dierent frequencies can have identical samples,

means that we cannot distinguish between them from their samples.

If s < 21 , the original spectrum is contaminated, so x(t) cannot be reconstructed.

If s > 21 , the original signal can be recovered by using an ideal low
pass lter.

GI (j) =

1 12 s 12 s
0
elsewhere

the inverse Fourier transform gives

gI (t) =

1 sin(s 2t )
T
s 2t

unit impulse response

Use a ZOH to approximate ideal low pass lter.

Frequency response of ZOH

Gh 0 =
Since

1 eT s
s
s

eT 2 (eT 2 eT 2 )
1 eT s
=
s
s
s = j

j
T
2

(eT

j
2

eT

j
2

j
sin T
e
2

j
T
2

sin =

ej ej
2j

Gh0 (j) = T

sin(
T2

T
)
2

ej 2

The ZOH does not approximate an ideal low pass lter very well. Higher
order holds do a better job but are more complex and have more time delay,
which reduces stability margin. So ZOHs get used a lot.

10

The Pulse Transfer Function

Convolution Summation

y(kT ) =
=

k=0

g(kT hT ) x(kT )
x(kT hT ) g(kT )

k=0

x(kT ) g(kT )
where g(kT) is the systems weighting sequence. i.e. g(k) = Z 1 {G(z)}
Starred Laplace Transfrom

Y (s) = G(s) X (s)

Note: X (s) is periodic with period
0, 1, 2, . G(s) is not periodic.
Taking the starred transform

2
s

since X (s) = X (s js k), for k =

{ See notes on next page }

Y (s) = [G(s)X (s)] = [G(s)] X (s) = G (s)X (s)

Note:

X (s) =

1
T

k= X(s

+ js k) + 12 x(0+ )

Y (z) = G(z) X(z)

Since the z transform can be seen to be the starred Laplace transform with
eT s replaced by z; i.e. X (s) = X(z)

11

Obtaining The Pulse Transfer Function

The presence or absence of an input sampler is crucial in determining the

pulse transfer function of a system.
For gure a,
Y (s) = G(s) X (s)

Y (s) = G (s) X (s)

Y (z) = G(z) X(z)

For gure b,
Y (s) = G(s) X(s)

Y (s) = [G(s) X(s)] = [GX(s)] = GX(z) = G(z)X(z)

12

Methods for obtaining the z transform

1. X(z) = Z [X(s) expanded into partial f ractions Xi (s)] =
2. X(z) =
3. X(z) =

k=0 x(kT )z

{Z[Xi (s)]}

residue of

X(s)z
zeT s

at pole of X(s)

Pulse transfer function of cascaded elements

For gure (a)

U(s) = G(s) X (s)

U (s) = G (s) X (s)

Y (s) = H(s) U (s)

Y (s) = H (s) U (s)

Y (s) = H (s) G (s) X (s)

= G (s) H (s) X (s)

Y (z) = G(z) H(z) X(z)

Y (z)
= G(z) H(z)
X(z)

For gure (b)

Y (s) = G(s) H(s) X (s) = GH(s) X (s)

Y (s) = [GH(s)] X (s)

Y (z) = GH(z) X(z)

Note that

Y (z)
= GH(z) = Z[GH(s)]
X(z)

G(z) H(z) = GH(z) = Z[GH(s)]

13

Pulse transfer function of closed-loop systems

C(s) = G2 (s) U (s)

C (s) = G2 (s) U (s)

E(s) = R(s) H(s) C(s)

U(s) = G1 (s) E(s) = G1 (s) R(s) G1 (s) H(s) C(s)
= G1 (s) R(s) G1 (s) H(s) G2 (s) U (s)

14

(5)
(6)
(7)

U (s) = [G1 (s) R(s)] [G1 (s) Gs (s) H(s) U (s)]

= [G1 R(s)] [G1 G2 H(s)] U (s)
{1 + [G1 G2 H(s)] } U (s) = [G1 R(s)]

[G1 R(s)]
1 + [G1 G2 H(s)]

U (s) =

(6)&(8)

C (s) =

G2 (s) [G1 R(s)]

1 + [G1 G2 H(s)]

C(z) =

G2 (z) G1 R(z)
1 + G1 G2 H(z)

Example

Find

C(z)
R(z)

C(z)
G(z)
=
,
R(z)
1 + G(z)

for

H(s) = 1

now


G(z) = Z[G(s)] =
=
=
=
closed-loop

K
Z (1 e )
s(s + 1)


K
K
1

(1 z ) Z
s
s+1


K
K
(1 z 1 )

1 z 1
1 eT z 1
K(1 eT )z 1
1 eT z 1
T s

C(z)
K(1 eT )z 1
=
R(z)
1 + [K (K + 1)eT ]z 1

15

(8)

Obtaining response between consecutive sampling instants

The z transform does not give the response between sampling instants.
Can use the following methods to do so.
1. The Laplace transform method
2. The modied z transform method
3. The state space method
Will look only at rst method for now.
1. The Laplace transform method
Example:

Consider the following system

We know that
C(s) = G(s) E (s) = G(s)
Thus

c(t) = L [C(s)] = L

16

R (s)
1 + GH (s)

R (s)
G(s)
1 + GH (s)

Mapping between the s plane and the z plane

z = eT s
let s = + j
z = eT (+j) = eT ejT = eT ej(T +2k)
now

|z| = eT < 1

for < 0, i.e. LHP

j axis in s plane corresponds to |z| = 1 , the unit circle in z plane

17

Constant-attenuation loci

Constant-frequency loci

Constant-damping ratio line

18

Stability analysis of closed-loop systems in the z domain

Consider the following pulse transfer function
G(z)
C(z)
=
R(z)
1 + GH(z)
Stability may be accessed by looking at the roots of the the characteristic
equation.
P (z) = 1 + GH(z) = 0
1. For stability, require that the roots |zi | < 1
2. If simple pole zi = 1 or zi = 1 or (zi = 1 and zi = 1)
or zi where zi complex such that |z| = 1
critical stability
3. Zeros do not aect absolute stability
Stability tests without finding roots
1. Jury test
2. Bilinear transformation and Routh criteria
The Jury test
F (z) = an z n + an1 z n1 + + a2 z 2 + a1 z + a0 = 0
an > 0

19

General Form of the Jury stability table

Row
1
2
3
4
5
6
:
:

z0
a0
an
b0
bn-1
c0
cn-2
:
:

z1
a1
an-1
b1
bn-2
c1
cn-3
:
:

z2
a2
an-2
b2
bn-3
c2
cn-4
:
:

2n-5
2n-4
2n-3

p0
p3
q0

p1
p2
q1

p2
p1
q2

zn-k
an-k
ak
bn-k
bk

cn-2
c0

zn-1
an-1
a1
bn-1
b0

zn
an
a0

:
:
p3
p0

where


 a

 0 ank 


 an
ak 

 b
bn1k

0

 bn1
bk

 c
cn2k

0

 cn2
ck

bk =
ck =
dk =
..
.












q0 =
q2 =












p0 p3 

p3 p0 


p0 p1 

p3 p2 

The necessary and sucient condition for the F(z) to have no roots on
and outside the unit circle are
F (1) > 0

F (1)
|a0 | <
|b0 | >
|c0 | >
|d0 | >
|q0 | >

an
|bn1 |
|cn2 |
|dn3 |
|q2 |

> 0 n even
< 0 n odd

(n 1) constraints

For a second order system, n=2, Jurys table contains only one row
F (1) > 0
F (1) > 0
20

|a0 | < an
Bilinear transformation and Routh stability criterion
Transform the z plane to the w plane by
z=

w+1
w1

z+1
z1
which maps the inside of the unit circle in the z plane into the left half of
the w plane. The unit circle in z plane maps into the imaginary axis in the
w plane and the outside of the unit circle in z plane maps into RHP of plane.

w=

The w plane is similar to s plane (but not quantitatively)

use Routh test.
Let

a0

P (z) = a0 z n + a1 z n1 + + an1 z + an = 0
w+1
w1

n

+ a1

w+1
w1

n1

+ + an1

can

w+1
+ a0 = 0
w1

Q(w) = b0 wn + b1 wn1 + + bn1 w + bn = 0

 Requires a lot of computation but can now use Routh test.

21