ECEG 5352 - Lecture - 5 PDF

Lecture-5
Discrete Time Modeling

(Part-ii)
Nebiyu Tenaye
Addis Ababa University
Addis Ababa Institute of Technology
School of Electrical & Computer Engineering
Digital Control Systems

ECEG-5352
May, 2017
Nebiyu Tenaye (AAiT/SECE) Lecture-5 May, 2017 1 / 39
Overview
Overview
1 Overview
2 Objective
3 Hybrid Control Systems
4 Properties of Sampled Data Systems
5 Discretization Methods
Forward Difference
Backward Difference
Trapezoidal (Tustin) Rule
6 Notes on Discretization Methods
7 Zero Order Hold
8 First Order Hold
9 System Modeling with MATLAB

Objective
Lecture Objectives
In this lecture you will learn the following:

The properties of sampled data systems,
The different discretization techniques,
The First-Order Hold (ZOH), and
Modeling tools in Matlab.

Hybrid Control Systems
Hybrid Control Systems
In practice, several analog controllers and a number of micro-

controllers / control computers (which are synchronized by the
same clock) may be utilized as a part of a complex control scheme.
From the stand-point of control engineering, it is highly desirable
to develop equivalent (discrete-time) transfer functions
representing the essential features of such control systems.

Properties of Sampled Data Systems
Property 1: (sampled signal)
Y (z) = Z{Y ∗ (s)}
Property 2: (re-sampled signal)
Y ∗ (s) = [Y ∗ (s)]∗
Y (z) = Z{Y ∗ (s)}

Property 3: (summation of sampled signals)
Y ∗ (s) = Y1∗ (s) ± Y2∗ (s) ± Y3∗ (s) ± · · · Yn∗ (s)

Y (z) = Z{Y1∗ (s)} ± Z{Y2∗ (s)} ± Z{Y3∗ (s)} ± · · · Z{Yn∗ (s)}
Y (z) = Y1 (z) ± Y2 (z) ± Y3 (z) ± · · · Yn (z)

Property 4: (sampled response)
Y (s) = G(s)X(s)
Y ∗ (s) = [G(s)X(s)]∗
Y (z) = Z{G(s)X(s)}
Notice that: Y ∗ (s) 6= G∗ (s)X ∗ (s)
Property 5: (sampled impulse response)
Y (s) = G(s)X ∗ (s)

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

Property 6: (Cascaded impulse responses))
Y ∗ (s) = G∗n (s) · · · G∗2 (s)G∗1 (s)X ∗ (s)

Y (z) ≡ Gn (z) · · · G2 (z)G1 (z)X(z)
Property 7: (sampled impulse response of a cascaded system)
Y ∗ (s) = [Gn (s) · · · G2 (s)G1 (s)X ∗ (s)]∗

Y ∗ (s) = [Gn (s) · · · G2 (s)G1 (s)]∗ X ∗ (s)
Y ∗ (s) = Z{Gn (s) · · · G2 (s)G1 (s)}X(z)

Example (1)
Consider the block diagram of the illustrated system. Obtain the

following transfer functions:
(a) B(z)/R(z)
(b) Y (z)/R(z)

Example (1)
Solution:
(a) Using Property 3:
E ∗ (s) = R∗ (s) − B ∗ (s) ⇒ E(z) = R(Z) − B(z)
From Property 5:
M ∗ (s) = [Gc (s)E ∗ (s)]∗ = G∗c (s)E ∗ (s) ⇒ M (z) = Gc (Z)E(z)
where Gc (z) ≡ Z{G∗c (s)} = Z{Gc (s)}
Similarly, property 7 yields:
B ∗ (s) = [H(s)Gp (s)Gzoh (s)M ∗ (s)]∗
B ∗ (s) = [H(s)Gp (s)Gzoh (s)]∗ M ∗ (s) ⇒ B(z) = G(z)M (z)
where G(z) ≡ Z{H(s)Gp (s)Gzoh (s)}

Example (1)
Combining the previous (boxed) three equations gives:
[1 + G(z)Gc (z)]B(z) = G(z)Gc (z)R(z)
Thus
B(z) G(z)Gc (z)
=
R(z) 1 + G(z)Gc (z)
where
Gc (z) ≡ Z{Gc (s)}

G(z) ≡ Z{H(s)Gp (s)Gzoh (s)}

Example (1)
(b)
To obtain the desired transfer function Y (z)/R(z), a f ictitious

sampler must be added to the system.

Example (1)
Using Property 3:
E ∗ (s) = R∗ (s) − B ∗ (s) ⇒ E(z) = R(z) − B(z)
From Property 5:
M ∗ (s) = [Gc (s)E ∗ (s)]∗ = G∗c (s)E ∗ (s) ⇒ M (z) = Gc (z)E(z)
where Gc (z) ≡ Z{G∗c (s)}
Similarly, property 7 yields:

Y ∗ (s) = [Gp (s)Gzoh (s)M ∗ (s)]∗
Y ∗ (s) = [Gp (s)Gzoh (s)]∗ M ∗ (s) ⇒ Y (z) = G1 (z)M (z)
where G1 (z) ≡ Z{Gp (s)Gzoh (s)}

Example (1)
Finally, property 7 gives:
B ∗ (s) = [H(s)Gp (s)Gzoh (s)M ∗ (s)]∗
B ∗ (s) = [H(s)Gp (s)Gzoh (s)]∗ M ∗ (s) ⇒ B(z) = G(z)M (z)
where G(z) ≡ Z{H(s)Gp (s)Gzoh (s)}
Dividing the previous (boxed) two equation results:
B(z) G(z)
= ≡ G2 (z)
Y (z) G1 (z)

Example (1)
Combining the previous equation with boxed equations of E(z),
M (z) and Y (z) gives:
Y (z) = G1 (z)Gc (z)[R(z) − G2 (z)Y (z)]
Thus
Y (z)[1 + G1 G2 (z) Gc (z)] = G1 (z)Gc (z)R(z)
| {z }
G(z)
Y (z) G1 (z)Gc (z)

=
R(z) 1 + G(z)Gc (z)
where, Gc (z) ≡ Z{Gc (s)}

G1 (z) ≡ Z{Gp (s)Gzoh (s)}
G(z) ≡ Z{H(s)Gp (s)Gzoh (s)}
Discretization Methods
The design of control system consists of two main steps:

The mathematical model of the plant is obtained to analyze its
behavior.
Using the model, an appropriate controller is designed to get the
desired response from the controlled system.
In continuous-time domain, the dynamics of the plant are

represented by a set of differential equations.

In control literature, there exist various approximation techniques

to convert such equations conveniently into discrete-time forms
without utilizing z-transforms.
The methods discussed here approximate the time derivative
(d/dt) in ODE using a corresponding difference equation:
1. Forward difference (Euler’s rule)

2. Backward difference
3. Trapezoidal (integration) rule
Also known as Tustin or bilinear transformation

Discretization Methods Forward Difference
1. Forward Difference
Utilizing the definition of derivative, we have
dx ∼ x((k + 1)T ) − x(kT ) x(k + 1) − x(k)
= = (1)
dt (k + 1)T − kT T
In terms of forward time-shift operator q:
dx ∼ q − 1 d ∼ q−1
= x(k) ⇒ = (2)
dt T dt T
Since d/dt corresponds to s variable while q operator is equivalent
to z variable, equation-2 can be rewritten as:
z−1
s∼
= (3)
T
The s-variable in the transfer function is conveniently replaced by
this expression leading to a new transfer function of z:

0 ∼ z−1
G (z) = G s = (4)
T
Discretization Methods Backward Difference
2. Backward Difference
Utilizing the definition of derivative, we have
dx ∼ x(kT ) − x((k − 1)T ) x(k) − x((k − 1))
= = (5)
dt kT − (k − 1)T T
In terms of Backward time-shift operator q −1 :
dx ∼ 1 − q −1 d ∼ 1 − q −1
= x(k) ⇒ = (6)
dt T dt T
As d/dt corresponds to s variable while q −1 operator is equivalent
to z −1 variable, equation-6 can be rewritten as:
z−1
s∼
= (7)
Tz
The s-variable in the transfer function is to be replaced by this
expression leading to a new transfer function of z:

0 ∼ z−1
G (z) = G s = (8)
Tz
Discretization Methods Trapezoidal (Tustin) Rule
3. Trapezoidal (Tustin) Rule

Let u(t) = dx/dt. The trapezoidal integration rule leads to
x((k + 1)T ) = x(kT ) + T2 [u((k + 1)T ) + u(kT )]

x((k + 1)T ) − x(kT ) = T2 [u((k + 1)T ) + u(kT )]
in terms of q:
(q − 1)x(k) = T2 (q + 1)u(k)
dx ∼ 2 q − 1 d 2 q−1
u(k) = = · · x(k) ⇒ = ·
dt T q+1 dt T q+1
As d/dt corresponds to s variable 2 q−1
while q operator is equivalent to z s∼
= · (9)
T q+1
variable, we have:
0 ∼ 2 q−1
Therefore, G (z) = G s = · (10)
T q+1
Notes on Discretization Methods
The discretization techniques discussed here approximate the

dynamics of a continuous plant better when the sampling intervals
are kept short.
Forward- and backward difference methods are quite vulnerable to
long sampling periods.
Despite its complexity, the Tustin transformation yields the best
results in most cases.
Since short sampling periods are essential for the successful
application of these techniques, the dynamics of output interfaces
(like ZOH) could be neglected under such circumstances.

Example (2)
Consider the ODE representing the dynamics of a plant:
d2 y dy
+ 3 + 2y = x(t)
dt2 dt
If T = 0.01 s, obtain the discrete-time models for this system using
the following techniques:
(a) Forward difference
(b) Backward difference
(c) Tustin transformation
(d) Z-transform with ZOH

Example (2)
The transfer function of this plant in s-domain becomes:
Y (s) 1
(s2 + 3s + 2)Y (s) = X(s) ⇒ = 2
X(s) s + 3s + 2
(a) Forward Difference:

1
1
G0 (z) ∼
= = 10000
z−1 2 z−1 z 2 − 1.9700z + 0.9702

T +3 T +2
(b) Backward Difference:

1 2
1 10302 z
G0 (z) ∼
= =
z−1 2 z−1 z 2 − 1.9705z + 0.9707

zT +3 Tz +2

Example (2)
(c) Tustin Transformation:
1
1 40601 (z + 1)2
G0 (z) ∼
= 2 =
z 2 − 1.9700z + 0.9704

4 z−1 6 z−1
T 2 z+1
+ T z+1 +2
Be careful when rounding numbers in these transfer functions. Since

we are dealing with relatively small numbers, make sure to perform all
calculations to the highest precision and then display at least 4 digits
after decimal point!

Example (2)
(d) ZOH:
1 − e−sT 1 − e−sT

1 1 1
G(s) = = −
s2 + 3s + 2 s s+1 s+2 s
n o n o
1−e−sT
G0 (s) = Z 1
s+1 − 1
s+2 s = (1 − z −1 )Z 1
s+1 − 1
s+2 s
1
From Z-transform tables:

1 −1 −2T )
z −1 (1 − e−T ) 2 z (1 − e
G0 (z) = −
1 − z −1 e−T 1 − z −1 e−2T

Example (2)
Hence,
B1 z −1 + B2 z −2
G0 (z) =
1 − z −1 (e−T + e−2T ) + z −2 e−3T
where
1
B1 ≡ (1 − e−T ) − 1 − e−2T

2
1
B2 ≡ e−T 1 − e−2T − e−2T (1 − e−T )

2
In terms of the numerical values, the final value will be:
(4.9503 × 10−5 )z + 4.9010 × 10−5

G0 (z) =
Z 2 − 1.9702 + 0.9704

Why Z-transforms?
If compared to the discretization techniques discussed here,

the model development efforts in the previous lecture appear
to be tedious!
However, notice that the discrete-time models (with ZOH)
developed via z-transform techniques are NOT
approximations.
Independent of sampling rate, the outputs of such models
exactly match to those of their continuous-time counterparts
at sampling instances.

Zero Order Hold
Zero Order Hold

Zero-order-hold (ZOH) is the interface
between the digital and analog world.
Consists of a latch and a D/A
converter:
Latch unit holds the binary
signals for one sampling period
(T),
D/A unit converts the binary
representation into a
corresponding output voltage.
Very simple and easy to implement!
If the sampling rate is too slow, the
output of a ZOH looks like a sequence
of boxcars.
Some information is lost during the
process,
Adelay of (T/2) is introduced.
First Order Hold
First Order Hold
First-order-hold (FOH) and other

higher-order methods are developed to
overcome the representation difficulties
associated with ZOH.
FOH (or so called ramp-invariant hold)
does a simple linear interpolation
between m(k) and m(k + 1).
It generates a corresponding output
voltage m(t) when kT < t < (k + 1)T .

First Order Hold
Transfer Function of FOH
The FOH performs interpolation between sampling instances:

t − kT
m(t) = m(k) + [m(k + 1) −m(k)] (11)
T | {z }
future value
where kT < t < (k + 1)T . Without further elaboration, the

transfer function of a FOH can be simply given as:
(1 − e−sT )2 1 (1 − z −1 )2 1
GF OH (s) = · = · 2 (12)
T e−sT s2 T z −1 s
Since it requires the future (one-step ahead) value of its input
m(k), it is very difficult to implement it in practice.

First Order Hold
Practical Implementation of FOH
Industry-standard “casual”
realization of a FOH interface
employs “extrapolation”
techniques.
When kT < t < (k + 1)T , the
slope of the output waveform
is [m(k) − m(k − 1)]/T .
The corresponding transfer
function of this practical
implementation differs from
the one given previously.

System Modeling with MATLAB
Matlab provides excellent toolbox functions to model/design

control systems.
The following Matlab functions are commonly used for this
purpose:
tf: to create the transfer functions of both continuous-time and
discrete-time systems.
c2d: to convert continuous-time models into discrete- time ones.
impulse/step: to simulate the response of a system to
impulse/step inputs.

MATLAB Functions
To illustrate the use of these functions, let us revisit our previous

example:

MATLAB Functions Continued
One can define a transfer function in z-domain as well:

To convert the continuous-time model into a discrete-time one:

To simulate the response of

this system to a step-input:
Other Matlab functions that

might interest you at this
point: poly, roots,
residue.

Example (3)
(Q.) If T = 0.01 s, obtain the transfer function B(z)/M (z) for the
given system using MATLAB.

Example (3)

Questions?

ECEG 5352 - Lecture - 5 PDF

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

ECEG 5352 - Lecture - 5 PDF

Uploaded by

Copyright:

Available Formats

Lecture-5

Discrete Time Modeling

Digital Control Systems

Nebiyu Tenaye (AAiT/SECE) Lecture-5 May, 2017 2 / 39

In this lecture you will learn the following:

Nebiyu Tenaye (AAiT/SECE) Lecture-5 May, 2017 3 / 39

Hybrid Control Systems

In practice, several analog controllers and a number of micro-

Nebiyu Tenaye (AAiT/SECE) Lecture-5 May, 2017 4 / 39

Properties of Sampled Data Systems

Property 1: (sampled signal)

Y (z) = Z{Y ∗ (s)}

Property 2: (re-sampled signal)

Nebiyu Tenaye (AAiT/SECE) Lecture-5 May, 2017 5 / 39

Property 3: (summation of sampled signals)

Y ∗ (s) = Y1∗ (s) ± Y2∗ (s) ± Y3∗ (s) ± · · · Yn∗ (s)

Nebiyu Tenaye (AAiT/SECE) Lecture-5 May, 2017 6 / 39

Property 4: (sampled response)

Notice that: Y ∗ (s) 6= G∗ (s)X ∗ (s)

Property 5: (sampled impulse response)

Y (s) = G(s)X ∗ (s)

Nebiyu Tenaye (AAiT/SECE) Lecture-5 May, 2017 7 / 39

Property 6: (Cascaded impulse responses))

Y ∗ (s) = G∗n (s) · · · G∗2 (s)G∗1 (s)X ∗ (s)

Property 7: (sampled impulse response of a cascaded system)

Y ∗ (s) = [Gn (s) · · · G2 (s)G1 (s)X ∗ (s)]∗

Nebiyu Tenaye (AAiT/SECE) Lecture-5 May, 2017 8 / 39

Consider the block diagram of the illustrated system. Obtain the

Nebiyu Tenaye (AAiT/SECE) Lecture-5 May, 2017 9 / 39

Nebiyu Tenaye (AAiT/SECE) Lecture-5 May, 2017 10 / 39

[1 + G(z)Gc (z)]B(z) = G(z)Gc (z)R(z)

Gc (z) ≡ Z{Gc (s)}

Nebiyu Tenaye (AAiT/SECE) Lecture-5 May, 2017 11 / 39

To obtain the desired transfer function Y (z)/R(z), a f ictitious

Nebiyu Tenaye (AAiT/SECE) Lecture-5 May, 2017 12 / 39

Similarly, property 7 yields:

Nebiyu Tenaye (AAiT/SECE) Lecture-5 May, 2017 13 / 39

Dividing the previous (boxed) two equation results:

Nebiyu Tenaye (AAiT/SECE) Lecture-5 May, 2017 14 / 39

Y (z) = G1 (z)Gc (z)[R(z) − G2 (z)Y (z)]

Y (z) G1 (z)Gc (z)

where, Gc (z) ≡ Z{Gc (s)}

The design of control system consists of two main steps:

In continuous-time domain, the dynamics of the plant are

Nebiyu Tenaye (AAiT/SECE) Lecture-5 May, 2017 16 / 39

In control literature, there exist various approximation techniques

1. Forward difference (Euler’s rule)

Nebiyu Tenaye (AAiT/SECE) Lecture-5 May, 2017 17 / 39

3. Trapezoidal (Tustin) Rule

x((k + 1)T ) = x(kT ) + T2 [u((k + 1)T ) + u(kT )]

Notes on Discretization Methods

The discretization techniques discussed here approximate the

Nebiyu Tenaye (AAiT/SECE) Lecture-5 May, 2017 21 / 39

Consider the ODE representing the dynamics of a plant:

Nebiyu Tenaye (AAiT/SECE) Lecture-5 May, 2017 22 / 39

(a) Forward Difference:

(b) Backward Difference:

Nebiyu Tenaye (AAiT/SECE) Lecture-5 May, 2017 23 / 39

Be careful when rounding numbers in these transfer functions. Since

Nebiyu Tenaye (AAiT/SECE) Lecture-5 May, 2017 24 / 39

From Z-transform tables:

Nebiyu Tenaye (AAiT/SECE) Lecture-5 May, 2017 25 / 39