Control Systems I
Lecture 5: Transfer Functions
Readings:
Emilio Frazzoli
Institute for Dynamic Systems and Control
D-MAVT
ETH Zürich
October 20, 2017
E. Frazzoli (ETH) Lecture 5: Control Systems I 20/10/2017 1 / 24
Tentative schedule
# Date Topic
1 Sept. 22 Introduction, Signals and Systems
2 Sept. 29 Modeling, Linearization
3 Oct. 6 Analysis 1: Time response, Stability
4 Oct. 13 Analysis 2: Diagonalization, Modal coordi-
nates.
5 Oct. 20 Transfer functions 1: Definition and properties
6 Oct. 27 Transfer functions 2: Poles and Zeros
7 Nov. 3 Analysis of feedback systems: internal stability,
root locus
8 Nov. 10 Frequency response
9 Nov. 17 Analysis of feedback systems 2: the Nyquist
condition
10 Nov. 24 Specifications for feedback systems
11 Dec. 1 Loop Shaping
12 Dec. 8 PID control
13 Dec. 15 Implementation issues
14 Dec. 22 Robustness
E. Frazzoli (ETH) Lecture 5: Control Systems I 20/10/2017 2 / 24
Vibrations of a string
https://www.youtube.com/watch?v=BSIw5SgUirg
https://www.youtube.com/watch?v=TJ533qfpp0o
https://www.youtube.com/watch?v=4BoeATJk7dg
E. Frazzoli (ETH) Lecture 5: Control Systems I 20/10/2017 3 / 24
Today’s learning objectives
Response to exponential inputs
Transfer functions
Conversions from State-Space models to/from Transfer Functions
Laplace transform and inverse Laplace transform
E. Frazzoli (ETH) Lecture 5: Control Systems I 20/10/2017 4 / 24
Motivation
Real System:
Modeling:
Linearization/Normalization:
Easy
Hard
Transfer Function(Today!)
Easy
Control System Design
E. Frazzoli (ETH) Lecture 5: Control Systems I 20/10/2017 5 / 24
State-space formulation of LTI response
Recall: Z t
y (t) = Ce At x(0) + C e A(t ⌧)
Bu(⌧ ) d⌧ + Du(t).
0
This formula gives a complete characterization of the output response of a
LTI system.
We have fully described the homogeneous solution (Ce At x(0)), but the
convolution integral is hard to interpret.
In order to gain a better understanding, we will study the response to
elementary inputs of a form that is
particularly easy to analyze: the output has the same form as the input.
very rich and descriptive: most signals/sequences can be written as linear
combinations of such inputs.
Then, using the superposition principle, we will recover the response to
general inputs, written as linear combinations of the “easy inputs.”
E. Frazzoli (ETH) Lecture 5: Control Systems I 20/10/2017 6 / 24
Exponential inputs
Let us choose as elementary input u(t) = e st , where s 2 C is a complex
number.
If s is real, then u is a simple exponential.
If s = j! is imaginary, then the elementary input must always be
accompanied by the “conjugate,” i.e.,
u(t) + u ⇤ (t) = e j!t + e j!t
= 2 cos(!t);
in other words, if s is imaginary, then u(t) = e st must be understood as a
“half” of a sinusoidal signal.
if s = + j!, then
u(t) + u ⇤ (t) = e t e j!t + e t e j!t
=e t
e j!t + e j!t
= 2e t
cos(!t),
and the input u is a “half” of a sinusoid with exponentially-changing
amplitude.
E. Frazzoli (ETH) Lecture 5: Control Systems I 20/10/2017 7 / 24
Output response to elementary inputs (1/2)
Recall Z t
At
y (t) = Ce x(0) + C e A(t ⌧)
Bu(⌧ ) d⌧ + Du(t).
0
Plug in u(t) = e st :
Z t
At
y (t) = Ce x(0) + C e A(t ⌧)
Be s⌧ d⌧ + De st
0
Z t
At At
= Ce x(0) + Ce e (sI A)⌧
B d⌧ + De st
0
If (sI A) is invertible (i.e., s is not an eigenvalue of A), then
h i
y (t) = Ce At x(0) + Ce At (sI A) 1
e (sI A)t
I B + De st .
E. Frazzoli (ETH) Lecture 5: Control Systems I 20/10/2017 8 / 24
Output response to elementary inputs (2/2)
Rearranging:
⇥ ⇤ ⇥ ⇤
y (t) = Ce At x(0) (sI A) 1
B + C (sI A) 1
B + D e st .
| {z } | {z }
Transient response Steady state response
If the system is asymptotically stable, the transient response will converge to
zero.
The steady state response to an input u(t) = e st can be written as:
yss = G (s)e st , G (s) 2 C,
1
where G (s) = C (sI A) B + D is a complex number.
The function G : s ! G (s) is also known as the transfer function: it
describes how the system transforms an input e st into the output G (s)e st .
E. Frazzoli (ETH) Lecture 5: Control Systems I 20/10/2017 9 / 24
Frequency response
Consider the case in which the input is a sinusoidal signal (s = ±j!):
u(t) = e j!t + e j!t
= 2 cos(!t).
The output is
u(t) = G (j!)e j!t + G ( j!)e j!t
= Me e j!t + Me e j!t
= 2M cos(!t + ).
In other words, the output is another sinusoid of frequency !, the magnitude
of which is M = |G (j!)| times the magnitude of the input, and with phase
leading the input by = \G (j!).
E. Frazzoli (ETH) Lecture 5: Control Systems I 20/10/2017 10 / 24
Linearity Review
Using linearity and the solution for u(t) = e st we can find the solution to
more complex inputs.
Remember, if
y1 (t) is the output for u1 (t),
y2 (t) is the output for u2 (t),
then
↵y1 (t) + y2 (t) is the output for ↵u1 (t) + u2 (t).
E. Frazzoli (ETH) Lecture 5: Control Systems I 20/10/2017 11 / 24
Building more general inputs from e st
Suppose we have two inputs
u1 (t) = U1 e s2 t and u2 (t) = U2 e s2 t .
The corresponding outputs are
y1 (t) = g (s1 )U1 e s1 t and y2 (t) = g (s2 )U2 e s2 t ,
Then if
u(t) = U1 e s1 t + U2 e s2 t ,
we get the output
y (t) = g (s1 )U1 e s1 t + g (s2 )U2 e s2 t .
More generally,
X X
u(t) = U i e si t y (t) = g (si )Ui e si t (1)
i i
E. Frazzoli (ETH) Lecture 5: Control Systems I 20/10/2017 12 / 24
Example (1/4)
A = [ 1]
g (s) = C (sI A) 1 B + D
ẋ = x + u, ! B = [1] !
= 1 · (s ( 1)) 1 · 1 + 0
y = x. C = [1] 1
= s+1
D = [0]
The transfer function of an LTI system will always be a ratio of two
polynomials in s.
What can you say about the eigenvalues of A and the values of s where
g (s) ! 1?
Let’s find the steady state output y (t) for the input
u(t) = cos(!t)
E. Frazzoli (ETH) Lecture 5: Control Systems I 20/10/2017 13 / 24
Example (2/4)
We can write u(t) as
u(t) = cos(!t)
1 j!t 1 j!t
= e + e
|2 {z } |2 {z }
u1 (t) u2 (t)
1
Notice that U1,2 = 2 and s1,2 = ±j!
Then
y1 (t) = g (s1 )U1 e s1 t y2 (t) = g (s2 )U1 e s2 t
= g (j!) 12 e j!t = g ( j!) 12 e j!t
e j!t e j!t
= 2(1+j!) = 2(1 j!)
E. Frazzoli (ETH) Lecture 5: Control Systems I 20/10/2017 14 / 24
Example (3/4)
y (t) = y1 (t) + y2 (t)
e j!t e j!t
= 2(1 j!) + 2(1+j!)
e j!t (1 j!)+e j!t (1+j!)
= 2(1+! 2 )
j!t
e +e j!t +j!(e j!t
e j!t )
= 2(1+! 2 )
1
= 1+! 2 (cos(!t) + ! sin(!t))
Sinusoid in - sinusoid out
The system changes the amplitude and phase
E. Frazzoli (ETH) Lecture 5: Control Systems I 20/10/2017 15 / 24
Example (4/4)
E. Frazzoli (ETH) Lecture 5: Control Systems I 20/10/2017 16 / 24
From State-space to Transfer Function
Recall, the transfer function is defined as
1
g (s) = C (sI A) B +D
If A is diagonal, with eigenvalues 1, 2, ..., n, this simply becomes
c1 b 1 c2 b 2 cn b n
g (s) = + + ... + .
s 1 s 2 s n
In general, the transfer function is a rational function of the form
n 1
bn 1s + bn 2 s n 2 + . . . + b0
g (s) = + d.
sn + an 1 s n 1 + . . . + a0
Note: the denominator of g (s) is the characteristic polynomial of the matrix
A, i.e., det(sI A).
E. Frazzoli (ETH) Lecture 5: Control Systems I 20/10/2017 17 / 24
From Transfer Function to State-space (1/2)
Given a transfer function g (s), there are many state-space models
(A, B, C , D) such that g (s) = C (sI A) 1 B + D.
We are interested in minimal realizations, i.e., (A, B, C , D) matrices that are
controllable, observable, and generate the given transfer function.
If the transfer function is written as a partial fraction expansion of the form
p1 p2 pn
g (s) = + + ... + + d,
s 1 s 2 s n
then a realization is
2 3 2p 3
1 p1
6 .. 7 6 .. 7
A=4 . 5, B=4 . 5
p
n pn
⇥p p ⇤
C= p1 ... pn , D = d.
E. Frazzoli (ETH) Lecture 5: Control Systems I 20/10/2017 18 / 24
From Transfer Function to State-space (2/2)
In the general case
n 1
bn 1s + bn 2 s n 2 + . . . + b0
g (s) = + d,
sn + an 1 s n 1 + . . . + a0
you can verify that the following is a minimal realization of g (s):
2 3 2 3
0 1 0 0 ... 0 0
6 0 0 1 0 ... 0 7 607
6 7 6 7
A=6 . .. 7, B = 6.7
4 .. . 1 5 4 .. 5
a0 a1 ... an 1 1
⇥ ⇤
C = b0 b1 ... bn 1 , D = [d];
This particular realization is called the controllable canonical form.
E. Frazzoli (ETH) Lecture 5: Control Systems I 20/10/2017 19 / 24
The Laplace Transform (1/3)
Can all inputs u(t) be expressed as a (infinite) sum of complex exponentials?
Yes! The tool for this is the Laplace transform and the inverse Laplace
transform.
To denote the Laplace transform of a signal u, we write
L [u] = U, (2)
The Laplace transform of u is defined (don’t worry you will never have to
calculate these integrals),
Z 1
U(s) := u(t)e st dt (3)
0
E. Frazzoli (ETH) Lecture 5: Control Systems I 20/10/2017 20 / 24
The Laplace Transform (2/3)
The inverse Laplace transform is denoted
1
L [U] = u, (4)
and is defined, Z +j!
1
u(t) = lim U(s)e st ds (5)
2⇡j !!1 j!
Notice that the inverse Laplace transform tells us how to write u(t) as a
linear combination of e st weighted by U(s).
E. Frazzoli (ETH) Lecture 5: Control Systems I 20/10/2017 21 / 24
The Laplace Transform (3/3)
Remember, X X
u(t) = U i e si t y (t) = g (si )Ui e si t
i i
Similarly we have (notice that the sum is replaced by an integral),
Z +j! Z +j!
1 st 1
u(t) = lim U(s)e ds y (t) = lim g (s)U(s)e st ds.
2⇡j !!1 j! 2⇡j !!1 j!
We can easily obtain the output for (almost) any input by using the Laplace
transform.
E. Frazzoli (ETH) Lecture 5: Control Systems I 20/10/2017 22 / 24
Systems represented in the “frequency domain”
We can write the output as
Z +j!
1
y (t) = lim g (s)U(s)e st ds.
2⇡j !!1 j!
Notice that this is
1
y (t) = L [gU] (t)
Taking the Laplace transform we find
L[y ](s) = g (s)U(s)
) Y (s) = g (s)U(s)
Once we are comfortable with looking at the signals in the “frequency
domain,” we obtain the (steady-state) output by simply multiplying the input
by the transfer function!
E. Frazzoli (ETH) Lecture 5: Control Systems I 20/10/2017 23 / 24
Today’s learning objectives
Response to exponential inputs
Transfer functions
Conversions from State-Space models to/from Transfer Functions
Laplace transform and inverse Laplace transform
E. Frazzoli (ETH) Lecture 5: Control Systems I 20/10/2017 24 / 24