Plan of the Lecture

I Review: state-space models of systems; linearization

I Today’s topic: linear systems and their dynamic response

Goal: develop a methodology for characterizing the output of a

given system for a given input.
State-Space Models
ẋ = Ax + Bu
y = Cx
where:
I x(t) 2 Rn is the state at time t
I u(t) 2 Rm is the input at time t
I y(t) 2 Rp is the output at time t
and -> system Matrix
I A 2 Rn⇥n is the dynamics matrix
I
->
B 2 Rn⇥m is the control matrix input matrix
->
matrix
I C 2 Rp⇥n is the sensor matrix output
How do we determine the output y for a given input u?
Reminder: we will only consider single-input, single-output (SISO)
systems, i.e., u(t), y(t) 2 R for all times t of interest. (m = p = 1)
Impulse Response
(Review from ECE 210)

u ẋ = Ax + Bu y
y = Cx

Unit impulse (or Dirac’s -function):

(t)
1. (t) = 0 for all t 6= 0
Z a
2. (t)dt = 1 for all a > 0
a t
0

It is useful to think of (t) as a limit of impulses of unit area:

area = 1
1/" as " ! 0, the impulse gets taller

" 0 " t
(1/" ! +1), but the area
under its graph remains at 1
Impulse Response <xt)

1
I

I
·

u ẋ = Ax + Bu y
0

y = Cx x(t- 2) = ?
x2( + 2) = ?

zero initial condition: x(0) = 0

Consider the input

I
u(t) = (t ⌧) unit impulse applied at t = ⌧

The system is linear and time-invariant (LTI), with zero I.C.:

x(0)=0; LTI system

u(t) = (t ⌧) ! y(t) = h(t ⌧)

The function h is the impulse response of the system.

Impulse Response

u ẋ = Ax + Bu y
y = Cx

zero initial condition: x(0) = 0

x(0)=0; LTI system

u(t) = (t ⌧) ! y(t) = h(t ⌧)

Questions to consider:
1. If we know h, how can we find the system’s response to
other (arbitrary) inputs?
2. If we don’t know h, how can we determine it?

We will start with Question 1.

Impulse Response

u ẋ = Ax + Bu y
y = Cx

zero initial condition: x(0) = 0

Question: If we know h, how can we find the system’s response

to other (arbitrary) inputs?
By the sifting property, for a general input u(t) we can write
Z 1
u(t) = u(⌧ ) (t ⌧ )d⌧.
1

Now we recall the superposition principle: the response of a

linear system to a sum (or integral) of inputs is the sum (or
integral) of the individual responses to these inputs.
Impulse Response

u ẋ = Ax + Bu y
y = Cx

zero initial condition: x(0) = 0

The superposition principle: the response of a linear system to

a sum (or integral) of inputs is the sum (or integral) of the
individual responses to these inputs.
Z 1 Z 1
u(t) = u(⌧ ) (t ⌧ )d⌧ ! y(t) = u(⌧ ) h(t ⌧ ) d⌧
1 1 | {z }
response to
(t ⌧)

— the integral that defines y(t) is a convolution of u and h.

Impulse Response

u ẋ = Ax + Bu y
y = Cx

zero initial condition: x(0) = 0

Conclusion so far: for zero initial conditions, the output

is the convolution of the input with the system impulse
response:
Z 1
y(t) = u(t) ? h(t) = h(t) ? u(t) = u(⌧ )h(t ⌧ )d⌧
1

Q: Does this formula provide a practical way of computing the

output y for a given input u?
A: Not directly (computing convolutions is not exactly
pleasant), but ...we can use Laplace transforms.
Laplace Transforms and the Transfer Function
Reminder : the two-sided Laplace transform of a function f (t) is
Z 1
F (s) = f (⌧ )e s⌧ d⌧, s2C
1

time domain frequency domain

u(t) U (s)
h(t) H(s)
y(t) Y (s)
convolution in time domain ! multiplication in frequency domain
y(t) = h(t) ? u(t) ! Y (s) = H(s)U (s)
The Laplace transform of the impulse response
Z 1
H(s) = h(⌧ )e s⌧ d⌧,
1

is called the transfer function of the system.

Laplace Transforms and the Transfer Function

Z 1
s⌧
Y (s) = H(s)U (s), where H(s) = h(⌧ )e d⌧
1

Limits of integration:
I We only deal with causal systems — output at time t is
not a↵ected by inputs at future times t0 > t
I If the system is causal, then h(t) = 0 for t < 0 — h(t) is
the response at time t to a unit impulse at time 0
I We will take all other possible inputs (not just impulses) to
be 0 for t < 0, and work with one-sided Laplace transforms:
Z 1
y(t) = u(⌧ )h(t ⌧ )d⌧
0
Z 1
H(s) = h(⌧ )e s⌧ d⌧
0
Laplace Transforms and the Transfer Function
Z 1
s⌧
Y (s) = H(s)U (s), where H(s) = h(⌧ )e d⌧
1

Given u(t), we can find U (s) using tables of Laplace transforms

or MATLAB. But how do we know h(t) [or H(s)]?

I Suppose we have a state-space model:

u ẋ = Ax + Bu y
y = Cx

In this case, we have an exact formula: *

ext=
ent e e
1
H(s) = C(Is A) B (matrix inversion) *

h(t) = CeAt B, t 0 (matrix exponential)

C

— will not encounter this until much later in the semester.

Laplace Transforms and the Transfer Function
Z 1
s⌧
Y (s) = H(s)U (s), where H(s) = h(⌧ )e d⌧
1

I So, how should we compute H(s) in practice?

Try injecting some specific inputs and see what happens at
the output.
Let’s try u(t) = est , t 0 (s is some fixed number)
W *
Z 1 - -
=
m
e meere ~

y(t) = h(⌧ )u(t ⌧ )d⌧ (because u ? h = h ? u)

Z0 1
= h(⌧ )es(t ⌧ ) d⌧
0
Z 1
st
=e h(⌧ )e s⌧ d⌧
0
= est H(s)
– so, u(t) = est is multiplied by H(s) to give the output.
Example
ẏ = ay + u (think y = x, full measurement)
st
u(t) = e (always assume u(t) = 0 for t < 0)
st
y(t) = H(s)e — what is H?

Let’s use the system model:

d
ẏ(t) = H(s)est = sH(s)est
dt
Substitute into ẏ = ay + u:

e⇢
sH(s)⇢st
= e⇢
aH(s)⇢st
e⇢
+⇢st
(8s; t > 0)
sH(s) = aH(s) + 1

1 est
H(s) = =) y(t) =
s+a s+a
Example (continued)

ẏ =ay + u
1
H(s) =
s+a
Now we can fund the impulse response h(t) by taking the
inverse Laplace transform — from tables,
(
e at , t 0
h(t) =
0, t<0
Determining the Impulse Response
u h y

x(0)=0; LTI system

u(t) = est , t 0 ! y(t) = est H(s)

Back to our two questions:

1. If we know h, how can we find y for a given u?
2. If we don’t know h, how can we determine it?

We have answered Question 1. Now let’s turn to Question 2.

One idea: inject the input u(t) = est , determine y(t), compute

y(t)
H(s) = ;
u(t)

repeat for all s of interest. Q: Is this a good idea?

Determining the Impulse Response

u(t) = est h y(t) = est H(s)

y(t)
compute H(s) = , repeat for as many values of s as
u(t)
necessary

Q: Is this likely to work in practice?

A: No — est blows up very quickly if s > 0, and decays to 0
very quickly if s < 0.
So we need sustained, bounded signals as inputs.
This is possible if we allow s to take on complex values.
Review: Complex Numbers

s = |{z}
a +j |{z}
b — rectangular form
real imaginary
part part

Polar form:
Im(s)
s = rej'
b s p
2
b r = |s| = a 2 + b2
+
p a2 r sin ' (magnitude)
✓ ◆
r= ' ' = \s = tan 1 b
Re(s) a
0 r cos ' a (phase)

Euler’s formula: ej' = cos ' + j sin '

Frequency Response
u h y

u(t) = A cos(!t) A – amplitude; ! – (angular) frequency, rad/s

From Euler’s formula:
Reall :

A j!t St
= u(t) = A cos(!t) = e + e j!t UCt) = e

2
By linearity, the response is
A⇣ ⌘
y(t) = H(j!)ej!t + H( j!)e j!t
2 Z 1
j!⌧
where H(j!) = h(⌧ )e d⌧
0
Z 1
H( j!) = h(⌧ )ej!⌧ d⌧ = H( j!)
0 | {z }
complex
conjugate

(recall that h(⌧ ) is real-valued)

Frequency Response
u h y

A⇣ ⌘
u(t) = A cos(!t) ! y(t) = H(j!)ej!t + H( j!)e j!t
2

H(j!) 2 C =) H(j!) = M (!)ej'(!)

j'(!)
H( j!) = M (!)e
Therefore,
A h i
y(t) = M (!) ej(!t+'(!)) + e j(!t+'(!))
2
= AM (!) cos !t + '(!) (only true in steady state)

The (steady-state) response to a cosine signal with amplitude

A and frequency ! is still a cosine signal with amplitude
AM (!), same frequency !, and phase shift '(!)
Frequency Response
u h y

u(t) = A cos(!t) ! y(t) = A M (!) cos !t + '(!)

| {z } | {z }
amplitude phase
magnification shift

Still an incomplete picture:

I What about response to general signals (not necessarily
sinusoids)? — always given by Y (s) = H(s)U (s)
I What about response under nonzero I.C.’s?— we will see
that, if the system is stable, then
transient response steady-state response
total response = +
(depends on I.C.) (independent of I.C.)

— need more on Laplace transforms Buct const rounded

-bre
oscillana
illafory
 -

MIMO
-
-
ISO

- admissible"
system

Jem ↳
U
->
D
⑧ -I -
........
(i) ei
T
o
=

ci f(x u) E

E
=

y h(x u)

( )e
=

h :x RM -> RP=

CL ·

x = As+
- -
Bu

lit-I
Live Y =2x Du
+

-
C

.
i 4
=

in =
5

i
=
u