EE5103/ME5403 Lecture Two

Computer Control
and Discrete-Time Systems
Xiang Cheng
Associate Professor
Department of Electrical & Computer Engineering
The National University of Singapore

Phone: 65166210 Office: Block E4-08-07

Email: [email protected]
1
• We discussed lots of basic concepts last time.
• Systems
• Input and output
• Control
• Feedback
• Superposition
• Laplace Transform
• Transfer function
• Poles and zeros
• Stability
• State

• If I want you to choose ONLY ONE that you think is the most important
concept for control theory, what would you say?

Feedback

2
Control of the robotic manipulator

For small y =θ −π

Angle
r=0 INPUT Robotic OUTPUT
REFERENCE Difference: Torque manipulator
ERROR

1
Controller
+ s2 −1
–

Objective: To bring the system OUTPUT to zero (upright position) for any small
nonzero initial conditions.
3
 Open-loop control
r u 1 y
C(s)
s2 −1

The transfer function from r to y is 1

C (s)
s2 −1
Is it possible to choose the controller C(s) such that the whole system is
stable?
No.
s −1 1
We tried C (s) = s − 1 2
≠
s −1 s +1

Last time, we showed that the zero-input-responses (u=0) are different!

How about the zero-state-responses? Are they the same or different?
s −1 1
2
U (s) ⇔ U (s) The same!
s −1 s +1
Those common factors can always cancel out if they are only signals 4
instead of T.F.
If the system is stable, what is the steady state value of the zero-input-response ?

Zero. So it is possible to cancel out the poles and zeros if the whole system is stable.

s +1 1
≈
( s + 1)( s + 2) s + 2

If the whole system is stable, then the common poles and zeros can cancel out
each other such that the reduced T.F. is good approximation of the original one.

5
 Let’s take a look at another example.
r u s−2 y
C(s) 2
s −1
Let 1
C (s) =
s−2 s−2
The transfer function from r to y is
( s − 2)( s 2 − 1)
Should we cancel out this common factor (s-2)?
Let’s write down the differential equation relating y and u
s−2
Y (s) = 2
U (s) ( s 2 − 1)Y ( s ) =
(s− 2) U(s)
s −1
Let’s write down the equation relating r and u
1
U (s) = R( s) (s− 2) U(s) =
R(s)
s−2
What is the differential equation relating r and y?
1
What is the transfer function from r to y?
( s 2 − 1)
We need to cancel out this common factor (s-2) in order to get the correct answer! 6
 Let’s compare the two examples carefully.

r 1 y r 1 u s−2 y
s-1
s2 −1 s−2 s2 −1
The transfer function from r to y is
s −1 1
( s 2 − 1) ( s 2 − 1)

This common factor (s-1) This common factor (s-2) has

cannot cancel out. to cancel out.

What is the difference between these two common factors?

And when should we cancel them out?

The answer lies in the different ways of these factors to be associated

with the signals. Can you figure it out?

7
r u 1 y r 1 u s−2 y
s-1
s2 −1 s−2 s2 −1
The transfer function from r to y is
s −1 1
( s 2 − 1) ( s 2 − 1)
U(s)= ( s − 1) R( s ) (s− 2) U(s) =
R(s)

The first (s-1) is operating on r. The first (s-2) is operating on u.

( s 2 − 1)Y ( s ) =
( s + 1)( s − 1)Y ( s ) =
U(s) ( s 2 − 1)Y ( s ) =
(s− 2) U(s)

The second (s-1) is operating on y. The second (s-2) is also operating on u.

Overall, we have
Overall, we have
2
( s − 1)Y ( s ) =
( s − 1) R( s ) ( s 2 − 1)Y ( s ) =
R( s )
This common factor (s-1) This common factor (s-2) has
cannot cancel out. to cancel out.

If the common factors are operating on the same variable, they should cancel out.8
If the common factors are operating on different variables, they should remain.
 What is a feedback control system?

aircraft, missiles, Information

Desired INPUT
economic systems, about the
Difference: to the
performance: cars, etc system:
ERROR system
REFERENCE OUTPUT

Controller System to be controlled

+
Compare –
Correct
Measure

•Trio: Measure —Compare –-Correct

•Fundamental Concept of Control: Feedback

9
r y r y
C(s) H(s) C(s) H(s)
+ –

Y (s)
= C (s) H (s) Y ( s) C ( s) H ( s)
R( s) =
R( s) 1 + C ( s) H ( s)
• There are so many wonderful properties about the feedback controller.
• What is the most important one?

• If the system is unstable, is it possible to use open-loop control to make it stable?

• No. Only feedback controller can change the dynamic property of the whole system!
Therefore, an unstable system can be stabilized by feedback!
• But it is a double-edged sword! It can also destabilize the system!
• Whenever a feedback controller is used, the dynamic property of the system has
been changed!
• Therefore, you have to design your feedback controller carefully because it can
make the performance better, and also possibly worse.
• However, there is one obvious advantage for open loop controller from the point of
view of implementation, what is that?
It is much easier to implement!
Drawing the feedback link in one second, which may cost the engineers one year!
10
 What is a Computer Control System?
The controller is implemented by a computer instead of an analog circuit!
How to convert an analog controller into computer control system?
Replace the analog circuit controller with a computer!
Difference: INPUT
Torque Angle
r=0 ERROR
OUTPUT
REFERENCE

KComputer
p + Kd s robotic manipulator
+
–

What is the input to the computer? The error signal.

Is it continuous time or discrete-time? It is continuous-time.
Can you feed a continuous-time signal directly into a computer?
No. Computer can only deal with a sequence of digits.
What is the output from the computer program?
A sequence of numbers---discrete-time signal
What is the input to the robotic manipulator?
The torque is supposed to be continuous-time signal! 11
So there is another mismatch!
 Algorithm
A-D D-A Process
+ (Computer)
–

How to do the A-D?

Sampling!

• The conversion is done at the sampling instants.

• The computer interprets the converted signal, y (t k ) ,as a sequence of
numbers.
• t k is called the sampling instants, and the time between successive
samplings is called the sampling period and is denoted as h. 12
• The sampling period h is normally a constant - uniform sampling
 How to do the D-A? What’s the natural way of converting the discrete-time
signal into continuous-time signal?

Zero-order Hold

Is the resulting analog signal continuous

or dis-continuous?
It is dis-continuous.

Can we make it piece-wise continuous?

First-order Hold

Why don’t we use First-order Hold to convert the control signal

from discrete-time to continuous time?
For feedback control system, the control signal depends upon the measured output.
At present, can we measure the output at next step?
13
No.
Next we are going to derive the equations for sampled system. But
first we need to know the solution for the continuous-time system.
How did we solve the differential equation? What is the tool?
We did it last time using Laplace Transform. Let’s do it again.

Since we need a precise solution, we need to consider the initial conditions.

sX ( s ) − AX ( s ) =x(0) + bU ( s )
sX ( s ) − x(0) = AX ( s ) + bU ( s )
X ( s ) = ( sI − A) −1 x(0) + ( sI − A) −1 bU ( s )
Y ( s ) = cX ( s ) = c( sI − A) −1 x(0) + c( sI − A) −1 bU ( s )

Let’s find out the solution in time-domain!

( sI − A) −1 e At
t
x(t ) = e x(0) + ∫ e A(t −τ ) bu (τ )dτ
At

0 t
y = cx = ce x(0) + c ∫ e A( t −τ ) bu (τ )dτ
At

Zero-input response Zero-state-response 14

Now we are going to use the solution in the continuous-time to derive the state space
model of the sampled system. t
x(t ) = e At x(0) + ∫ e A(t −τ ) bu (τ )dτ
0 t
y = cx = ce At x(0) + c ∫ e A( t −τ ) bu (τ )dτ
0

Assume that the state at time t k is known, how to calculate the future state at
time t, x(t), where t is in the interval [t k , t k +1 ] ?
t
x(t ) = e A( t −t k )
x(t k ) + ∫ e A(t −τ )bu (τ )dτ
tk

Is u(t) a constant in the whole interval [tk , tk +1 ) ?

Yes. Zero-order-hold.
t
x(t ) = e A( t −tk ) x(t k ) + ( ∫ e A(t −τ ) dτ )bu (t k )
tk
Next, let’s try to simplify the expression of the integral.
t 0 t −t k
By change of variable v = t −τ ∫ ∫ dv = ∫ dv
A ( t −τ )
e d τ = − e Av
e Av

tk t −t k 0
t −t k

x(t ) = e A(t −tk ) x(t k ) + ( ∫ e Av dv)bu (t k ) 15

0
Now we are ready to compute the state at t k +1 , putting t = tk +1
t k +1 −t k

x(t k +1 ) = e A(tk +1 −tk ) x(t k ) + ( ∫ dv)bu (tk )

Av
e
0

For periodic sampling with period h, we have t k = kh and t − t = h

k +1 k

h
x(kh + h) = e x(kh) + ( ∫ e Av dv)bu (kh) = Φx(kh) + Γu (kh)
Ah

0
Choose the sampling period as the time unit

State-space Model

Zero-order hold x(k + 1) = Φx(k ) + Γu (k )

y (k ) = cx(k )
Φ = e Ah
h
Γ = ( ∫ e Av dv)b
0

How to calculate e At ? ( sI − A) −1 e At
16
State-space Model
x(k + 1) = Φx(k ) + Γu (k )
y (k ) = cx(k )
Given x(k) and u(k), how to predict the future values of x(k)?

To get x(k+1) is simple:

x(k + 1) = Φx(k ) + Γu (k )

How about x(k+2), …?

x(k + 2) = Φx(k + 1) + Γu (k + 1)
= Φ (Φx(k ) + Γu (k )) + Γu (k + 1)
= Φ 2 x(k ) + ΦΓu (k ) + Γu (k + 1)
How about x(k+n), …?

zero-input response
zero-state response

Given x(k) and input signal, can you predict the whole future of the system?
17
Yes. Therefore, it is verified that this is indeed the state space model.
•Input-Output Models
For continuous-time, we have differential equations. For discrete-time, we have
y (k + 1) = a1 y (k ) + a2 y (k − 1) + b0u (k ) + b1u (k − 1)
For continuous-time, we have differentiation operator. For discrete-time, we have
•Shift-Operator ---to simplify the representation

forward-shift operator q q f (k ) = f (k + 1)
−1
backward-shift operator or the delay operator q q −1 f (k ) = f (k − 1)
•Operator calculus gives compact descriptions of systems

y (k + 1) = a1 y (k ) + a 2 y (k − 1) + b0 u (k ) + b1u (k − 1)
y (k + 1) − a1 y (k ) − a 2 y (k − 1) = b0 u (k ) + b1u (k − 1)
•Use of the shift operator gives

qy (k ) − a1 y (k ) − a 2 q −1 y (k ) = b0 u (k ) + b1 q −1u (k )
(q − a1 − a2 q −1 ) y (k ) = (b0 + b1q −1 )u (k )

18
 (q − a1 − a2 q −1 ) y (k ) = (b0 + b1q −1 )u (k )
−1
How to get rid of q ? q (q − a1 − a2 q −1 ) y (k ) = q (b0 + b1q −1 )u (k )

(q 2 − a1q − a2 ) y (k ) = (b0 q + b1 )u (k )

In terms of q A(q ) y (k ) = B(q )u (k )

A(q ) = q 2 − a1q − a2
B(q ) = b0 q + b1

This is a more compact representation than the original difference equation:

y (k + 1) = a1 y (k ) + a2 y (k − 1) + b0u (k ) + b1u (k − 1)

19
 How to solve difference equation?
y (k + 1) = a1 y (k ) + a2 y (k − 1) + b0u (k ) + b1u (k − 1)
Given input u(k), how to find out output y(k) from this equation?

In continuous-time, what is the tool to solve differential equation?

We rely on Laplace Transform.
∞
F ( s ) = ∫ f (t )e − st dt
0

Can we directly apply Laplace transform to discrete-time signal? No.

What is the fundamental basis for Laplace Transform? Why is Laplace

Transform so powerful for linear system?

Superposition Principle

Is the superposition principle still true for discrete-time linear system?

Yes. 20
 Fundamental property of linear system----Superposition Principle
Why is superposition principle so important?

Suppose you try to figure out the response of the system to a complicated input u.
u (t ) ???
Let input u be resolved into a set of basis functions

φ1 (t ) y1 (t )
φ2 (t ) y2 (t )
if the system is linear

21
 The trick of superposition principle is
to choose proper basis functions whose response can be easily obtained.
What did we choose for the continuous time system?

In the discrete-time, what basis function should we use?

We can still try the exponential function just replacing t with t=kh!

Then we can make a wild guess that

a k H (a)a k
Does it really work?
We need to verify whether they satisfy the equation or not.
y (k + 1) = a1 y (k ) + a2 y (k − 1) + b0u (k ) + b1u (k − 1)
u (k ) = a k , y (k ) = H (a)a k
y (k + 1) − a1 y (k ) − a2 y (k − 1) = b0u (k ) + b1u (k − 1)
b0 + b1a −1
H (a)a k +1 k
− a1 H (a )a − a2 H (a )a k −1 k
= b0 a + b1a k −1
H (a) =
a − a1 − a2 a −1 22
So it works again!
 k
Can we decompose any discrete-time signal f(k) by the exponentials a ?
1
∫
k −1
f (k ) = F ( z ) z dz zk → ak
2πi

What is F(z)?

F(z) is the Z- transform of f(k)

The equivalent of Laplace Transform in continuous-time.

23
 Break

•10 Amazing Robots

24
•The z-Transform

where z is a complex variable, just like in F(s), where s is complex variable.

•Examples. What is z-transform of unit impulse signal δ (k ) ?

Just apply the definition we have,

∞
F ( z ) = ∑ f (k ) z − k = f (0) = 1
k =0

25
•The z-Transform

•What is Laplace transform of e at ? at 1

e →
s−a
What is z-transform of ak ?

What is the little trick to calculate this type of summation (Geometric series)?

1 z
F ( z) = =
a z−a
1−
z 26
 What is the most important property of Laplace transform?
The relationship between y(t) and y’(t) in s-domain!
L( y (t )) = Y ( s ) d
⇔s
L( y ' (t )) = sY ( s ) − y (0) dt

Similarly, the most important property of z-transform is about

the relationship between x(k) and x(k+1) in the z-domain

How do we relate these two?

X(0)=0 z{x(k + 1)} = zX ( z )

Compare with q {x(k)}=x(k+1) q ( forward − time − shift ) ⇔ z
q −1 (backward − time − shift ) ⇔ z −1 27
 Now we are ready to use z-transform to solve difference equations; for instance,

y (k + 1) = a1 y (k ) + a2 y (k − 1) + b0u (k ) + b1u (k − 1)
y (k + 2) − a1 y (k + 1) − a2 y (k ) = b0u (k + 1) + b1u (k )

If the z-transform of both sides is taken, assuming zero initial conditions

z 2Y ( z ) − a1 zY ( z ) − a2Y ( z ) = b0 zU ( z ) + b1U ( z )

It is very easy to get the solution of Y(z) in the z-domain.

b0 z + b1
Y ( z) = 2 U ( z ) = H ( z )U ( z )
z − a1 z − a2

What is the solution in the time-domain?

28
 How to get y(k) in time-domain? How to do inverse z-Transform?
1
∫
k −1
y (k ) = Y ( z ) z dz
2πi This is the hard way!

What is the easy way?

Partial Fraction Expansion:

z +1 −1 2
Y ( z) = =( + )
( z − 0.5)( z − 2) z − 0.5 z−2

Use the z-Transform table : z

→ ak
z−a
There is a mismatch! There is one
little trick to make it work!
Y ( z)
Use instead ! Y ( z) z +1 1 2 1
z = =( − + )
z z ( z − 0.5)( z − 2) z z − 0.5 z − 2
y ( k ) = δ ( k ) − 2(0.5) k + 2k , k ≥ 0
29
There is another way to get the answer using Partial Fraction Expansion directly:
z +1 −1 2 z
Y ( z) = =( + ) → ak
( z − 0.5)( z − 2) z − 0.5 z−2 z−a
In order to use the z-Transform table , let’s rewrite

Using the z-transform table, we get

What is y(k) then?

q −1 (backward − time − shift ) ⇔ z −1

What is y(0)? Let’s try to find it out from the definition of z-transform

30
y(0)=0.
 Let’s apply z-transform to the state-space equation
x(k + 1) = Φx(k ) + Γu (k )
y (k ) = cx(k )
z ( X ( z ) − x(0)) = ΦX ( z ) + ΓU ( z )
zX ( z ) − ΦX ( z ) = zx(0) + ΓU ( z )

X ( z ) = ( zI − Φ ) −1 zx(0) + ( zI − Φ ) −1 ΓU ( z )
What is the output?

Y ( z ) = cX ( z ) = c( zI − Φ ) −1 zx(0) + c( zI − Φ ) −1 ΓU ( z )

•If the initial conditions are zero, x(0)=0

Y ( z ) = c( zI − Φ ) −1 ΓU ( z ) = H ( z )U ( z )
•What is the transfer function? H ( z ) = c( zI − Φ ) −1 Γ
U (z ) H(z) Y(z) 31
Transfer function: H(z) -----Transfer the input into output

U (z ) H(z) Y(z)

Since Y(z)=H(z)U(z), does the same input always produce the same output?

Y(z)=H(z)U(z) only if the initial conditions are all zero.

If intial conditions are not zero, there are extra terms in the solution!.

Y ( z ) = cX ( z ) = c( zI − Φ ) −1 zx(0) + c( zI − Φ ) −1 ΓU ( z )

zero-input response zero-state response

32
 Transfer function:
A representation of the system which is equivalent to difference equation.

q ( forward − shift − operator ) ⇔ z

Given a transfer function, how to write down the difference equation?

Y ( z) z + 2
= 2 z 2Y ( z ) − Y ( z ) = zU ( z ) + 2U ( z )
U ( z) z −1

q⇔z y (k + 2) − y (k ) = u (k + 1) + 2u (k )

Given a difference equation, how to write down the transfer function?

y (k + 1) = y (k ) + y (k − 1) + u (k − 2) y (k + 1) − y (k ) − y (k − 1) = u (k − 2)

qy (k ) − y (k ) − q −1 y (k ) = q −2u (k ) q ⇔ z zY ( z ) − Y ( z ) − z −1Y ( z ) = z −2U ( z )

z −2 1
Y ( z) = U ( z) Y ( z) = 3 2
U ( z)
z −1 − z −1 z −z −z 33
What is the unit impulse response?

u (k ) = δ (k )

U ( z ) = 1, Y ( z ) = H ( z )

Transfer function is the unit impulse response in z-domain!

 •What is the response to exponential a k ?
k
a k H (a)a
•The special case: a=1

1 H (1)

The static gain (the steady-state gain, or DC gain)------ H(1)

C H (1)C
What is steady-state gain in continuous time?

s=0
1 H (0)
35
 U (z ) H(z) Y(z)

−1 Q( z )
H ( z ) = c( zI − Φ ) Γ =
P( z )
•What are the poles? P( z ) = 0

•What are the zeros? Q( z ) = 0

What is P(z)?

* *
* *
( zI − Φ ) −1 =   P ( z ) = det{( zI − Φ )}
det{( zI − Φ )}

How to get the eigenvalues of the matrix Φ?

det{(λI − Φ )} = 0
36
The eigenvalues of Φ The poles of the system
 The relationship between poles and stability:
Partial Fraction expansion of Transfer function (Impulse response)
1 1 1 1 z +1
H (s) = = ( − ) H ( z) =
2
s −1 2 s −1 s +1 ( z − 0.5)( z − 2)

If λ is the pole, then If λ is the pole, then

1 1
must be one component in H(s) z −λ must be one component in H(z)
s−λ

e λt is component of the impulse response λk is component of the impulse response

λ = σ + jω eσt --stability
λ = ρe jθ ρk ---stability

e λt = eσt e jωt e jωt ---oscillation λk = ρ k e jθk ---oscillation

e jθk
σ <0 eσt → 0 Stable ρ <1 ρk → 0 Stable

σ > 0 eσt → ∞ Unstable ρ >1 ρk → ∞ Unstable

σ = 0 eσt = 1 Marginally Stable ρ =1 ρk =1 Marginally Stable

37
 Multiplicity of the poles
Partial Fraction expansion of Transfer function (Impulse Response)
3s + 1 1 1 1
H (s) = = − +
( s − 1)( s + 1) 2 s −1 s + 1 ( s + 1) 2
If λ is the multiple pole, then
1 1
must be one component in H(s) must be one component in H(z)
(s − λ )2 ( z − λ )2

te λt is component of the impulse response kλk is component of the impulse response

λ = σ + jω teσt ---stability
λ = ρe jθ kρ k ---stability
te λt = teσt e jωt e jωt ---oscillation kλk = kρ k e jθk ---oscillation
e jθk
σ < 0 teσt → 0 Stable ρ < 1 kρ k → 0 Stable

σ > 0 teσt → ∞ Unstable ρ >1 kρ k → ∞ Unstable

σ = 0 teσt = t → ∞ Unstable ρ =1 kρ k = k → ∞ Unstable

38
 Stability criterion

Continuous-time Discrete-time

λ = σ + jω λ = ρe jθ

σ <0 Stable ρ <1 Stable

σ >0 Unstable
ρ >1 Unstable

σ =0 ρ =1
Multiple Multiple
Single Single

Marginal Stable Unstable Marginal Stable Unstable

For control system design, “marginal stable” is not considered as “stable”. 39

 Now we know that the poles decide the stability. What about zeros?
•Meaning of zeros

The zeros decide the weights of the components in the impulse response.

There is another important meaning for zeros.

a k H (a)a k
What is the output if a is the zero of the system, ie. H(a)=0?
Zero! Signal blocking property
Example: H(1)=0

1 ???
0
It means that the effect of the input can be completely rejected!
When do we want this to happen?
40
We will use this property for disturbance rejection later!
 Zeros are related to an important concept in control system design:
•Inverse of a system
Q( z )
If the transfer function of a system is H ( z) =
P( z )
The inverse of a system is defined as 1 P( z )
=
H ( z ) Q( z )
The inverse of a system can be imagined as the original system with inverse direction
of the signal flow: the input becomes output while the output becomes input!
Y ( z ) = H ( z )U ( z ) 1
U ( z) = Y ( z)
H ( z)
U (z ) H(z) Y(z) U (z )
1
H ( z) Y(z)

•The inverse of a system may not have any physical meaning.

•It is just a mathematical concept which is useful for control system design.

41
•Inverse of a system
Q( z )
If the transfer function of a system is H ( z) =
P( z )
The inverse of a system is defined as 1 P( z )
=
H ( z ) Q( z )

Systems with stable and unstable inverses

•System with stable inverse (1/H(z) is stable): All the zeros are stable.

System with unstable inverse (1/H(z) is unstable):

The system contains at least one unstable zero.

•If a system has unstable zeros, does it imply that the system is unstable ?

No. It only implies that the inverse of the system is unstable.

This concept is important for controller design, we will discuss it next week.

42
 What happens to the poles of the system after sampling?

Zero-order hold x(k + 1) = Φx(k ) + Γu (k )

y (k ) = cx(k )
Φ = e Ah

What are the eigenvectors and eigenvalues of A?

Av = λv
Is v also the eigenvector of Φ ?

Av = λv, Ak v = λk v

If Av = λv Then Φv = e hλ v
43
λ ⇒ e hλ
 λ ⇒ e hλ
λ = σ + jω ⇒ e hλ = e h (σ + jω ) = e hσ e jhω
•This is the mapping from the continuous-time poles to the discrete-time poles.

•If p is the stable pole, is it

stable in the discrete-time?

Yes.

•If p is the unstable pole,

is it unstable in the
discrete time?

Yes.

44
 The map of the poles: λ ⇒ e hλ
•How about zeros? Can we use the same map to get zeros?
•Example Second-order system

Using Table 2.1 , we can get the discrete transfer function

What is the zero for the sampled system?

Z=-1.

•Are there any zeros in the original continuous-time system?

•Zeros are introduced by sampling!

It is not possible to give a simple formula for the mapping of

zeros under zero-order-hold. 45
Q & A…

46
 b1 z + b2
H ( z) = 2
z + a1 z + a2

return

47
 b1 z + b2
H ( z) = 2
z + a1 z + a2

return

48
return

return

49