Classnotes

for
Classical Control Theory

I. E. K ose
Dept. of Mechanical Engineering
Bo gazici University
May 13, 2011
Chapter 1
Modeling of Dynamic Systems
1.1 Mechanical Systems
The governing principles are Newtons laws of motion.
Particles
For a particle,

i
F
i
= ma.
Example 1.1.1. Spring-mass-damper system
F

k
c
m
m

= F k c

.
Example 1.1.2 (Quarter car suspension model). A simplied model is shown below:
m
s
m
u
x
s
x
u
x
g
k
c
k
t
1
The equations of motion are
m
s
x
s
= k(x
u
x
s
) +c( x
s
x
u
)
m
u
x
u
= k(x
u
x
s
) c( x
s
x
u
) +k
t
(x
g
x
u
).
Example 1.1.3 (Lunar landing). Simplied picture:
h
m
g
L
T
Rigid Bodies
For a rigid body,

i
F
i
= ma
G
and

i
(r
i
F
i
) =

H
G
.
In 2D only, these boil down to

i
F
i
= ma
G
and

i
(r
i
F
i
) = J
z
.
Example 1.1.4 (Lateral vehicle dynamics). Consider the free body diagram for the lateral motion
of a vehicle:
x
y
z
F
RRy
F
RRx
F
RLy
F
RLx
F
FLy
F
FLx
F
FRx
F
FRy
X
Y
Z

b
L
R
L
F
Figure 1.1: Free body diagram for lateral vehicle dynamics.
The equations of motion are given by

i
F
i
= m v (translation)

i
r
i
F
i
= J
z

k. (rotation)
2
...
If v is constant, the equations simplify to
mv(

+

) = c
F
_

W

L
F

v
_
+c
R
_
+
L
R

v
_
J
Z

= c
F
L
F
_

W

L
F

v
_
c
R
L
R
_
+
L
R

v
_
.
1.2 Electrical Systems
There are three main passive elements:
(i) Resistor: V = IR
(ii) Capacitor: V =
1
C
_
t
0
I() d
(iii) Inductor: V = L
dI
dt
Two main source elements:
(i) Voltage source,
(ii) Current source.
Basic laws governing electrical circuits:
(i) Kirchhos voltage law (KVL): The algebraic sum of all voltage dierences taken around a
closed loop in a circuit is zero.
(ii) Kirchhos current law (KCL): The algebraic sum of currents leaving a node equals the alge-
braic sum of currents entering that node.
Remark 1 (Equivalence between mechanical force-displacement principles and voltage-current
principles). With v(t) := x(t), can write
(i) For a mass, F = m v,
(ii) For a damper, F = cv,
(iii) For a spring, F = k
_
t
0
v() d.
Example 1.2.1. (Series RLC circuit)
V
s
R
L
C
3
By KVL,
v
s
= v
R
+v
L
+v
C
.
By KCL,
i
R
= i
L
= i
C
=: i.
Hence, we obtain
v
s
= Ri +L
di
dt
+
1
C
_
t
0
I() d.
if we dene q(t) :=
_
t
0
I() d, this gives
v
s
= R q +L q +
1
C
q.
Example 1.2.2. (Parallel RLC circuit) The same elements can be connected in series and driven by
a current source to obtain the parallel RLC circuit shown below:
R L C
I
s
By KVL,
v
R
= v
L
= v
C
=: v.
By KCL,
i
s
= i
R
+i
L
+i
C
.
Example 1.2.3. (DC motor) An electro-mechanical system. The coupling between the electrical
and the mechanical systems is given by
v
emf
= K
m

=: K
m
and T = K
t
I.
V
s
R
L
I

J
The equation for the armature circuit is
v
s
= IR +L

I +K
m
.
If we assume the output shaft has a mass moment of inertia J and is subject to rotational viscous
damping with damping coecient c, we have
J = T c = K
t
I c.
4
1.3 Fluid ow systems
By conservation of mass, for any container,
m = w
in
w
out
Remark 2. Flow through a pipe:
P
1
P
2
w
Assume there is no accumulation in the pipe so that m = 0. Then, w
in
= w
out
=: w, where
w =
1
R
(p
1
p
2
)
1/
= 1: Laminar ow
Example 1.3.1. Consider the following system where = 1 for the outlet pipe.
h
w
in
w
out
A

h = w
in
w
out
= w
in

P
atm
+gh P
atm
R
= w
in

gh
R

h =
g
AR
h +
1
A
w
in
.
1.4 State-space representations
If the governing equations are linear, constant-coecient ODEs of a total order of n, can always
dene state variables x
1
, x
2
, x
n
and input variables u
1
, u
2
, , u
m
such that the equations can
be put in the form
x = Ax +Bu,
where A is n n and B is n m.
5
Example 1.4.1. For the spring-mass-damper system, dene
x
1
:= , x
2
:=

= x
1
and u := F.
Then, can write
_
x
1
x
2
_
=
_
0 1
k/m c/m
__
x
1
x
2
_
+
_
0
1/m
_
u.
Example 1.4.2. For the DC motor, can dene
x
1
:= I, x
2
:= and u := v
s
.
We obtain
_
x
1
x
2
_
=
_
R/L K
m
/L
K
t
/J c/J
__
x
1
x
2
_
+
_
1/L
0
_
u
1.4.1 Linearization about an equilibrium point
Recall that for an analytic function f : IR IR, the Taylor series expansion of f(x) around x
0
is
given by
f(x) =

k=0
f
(k)
(x
0
)
k!
(x x
0
)
k
.
Example 1.4.3. sin =

3
3!
+

5
5!


7
7!

= for small.
Example 1.4.4. cos = 1

2
2!
+

4
4!


6
6!

= 1 for small.
Autonomous Systems
Consider an autonomous (i.e., no-input) nonlinear system. The dynamics are described by
x = f(x) =
_
_
_
_
_
f
1
(x
1
, x
2
, , x
n
)
f
2
(x
1
, x
2
, , x
n
)
.
.
.
f
n
(x
1
, x
2
, , x
n
)
_
_
_
_
_
.
If f is analytic at x
0
, can expand it as
f(x) = f(x
0
) + (Jf)(x)

x=x
0
(x x
0
) +h.o.t.,
where the Jacobian of f is
Jf :=
_
_
_
_
_
_
f
1
x
1
f
1
x
2

f
1
xn
f
2
x
1
f
2
x
2

f
2
xn
.
.
.
.
.
.
.
.
.
.
.
.
fn
x
1
fn
x
2

fn
xn
_
_
_
_
_
_
6
If x
0
is an equilibrium point, i.e., f(x
0
) = 0, we have
f(x)

= (Jf)(x)

x=x
0
(x x
0
).
Then, since x
0
= 0, we have
d
dt
(x x
0
)

= (Jf)(x)

x=x
0
(x x
0
).
Dening x := x x
0
and A := (Jf)(x)

x=x
0
this gives

x = A x.
Example 1.4.5. Linearize the following system about its equilibrium point assuming > 0 and
0 < V/A < 1:
y = V Ae
/y
At equilibrium, we have y = 0, or equivalently,
V Ae
/yeq
= 0 e
/yeq
=
V
A


y
eq
= ln
_
V
A
_
y
eq
=

ln
_
V
A
_.
Non-autonomous systems
Now suppose
x = f(x, u) =
_
_
_
_
_
f
1
(x
1
, , x
n
, u
1
, , u
m
)
f
2
(x
1
, , x
n
, u
1
, , u
m
)
.
.
.
f
n
(x
1
, , x
n
, u
1
, , u
m
)
_
_
_
_
_
By the same procedure as above, suppose
f(x
eq
, u
eq
) = 0
and expand f as
f(x, u) = f(x
eq
, u
eq
) + (Jf)(x, u)

(x,u)=(xeq,ueq)
((x, u) (x
eq
u
eq
)) +h.o.t.
Note that in this case,
(Jf)(x, u) =
_
_
_
_
_
_
f
1
x
1

f
1
xn
f
1
u
1

f
1
um
f
2
x
1

f
2
xn
f
2
u
1

f
2
um
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
fn
x
1

fn
xn
fn
u
1

fn
um
_
_
_
_
_
_
=:
_
(J
x
f)(x, u) (J
u
f)(x, u)
_
.
7
Then, can write
x = f(x, u)
= f(x
eq
, u
eq
) +
_
(J
x
f)(x, u) (J
u
f)(x, u)
_

(x,u)=(xeq,ueq)
((x, u) (x
eq
u
eq
)) +h.o.t.
=
_
(J
x
f)(x, u) (J
u
f)(x, u)
_

(x,u)=(xeq,ueq)
((x, u) (x
eq
u
eq
)) +h.o.t.

=
_
(J
x
f)(x, u) (J
u
f)(x, u)
_

(x,u)=(xeq,ueq)
((x, u) (x
eq
u
eq
))
= (J
x
f)(x, u)

(x,u)=(xeq,ueq)
(x x
eq
) + (J
u
f)(x, u)

(x,u)=(xeq,ueq)
(u u
eq
)
x x
eq

= (J
x
f)(x, u)

(x,u)=(xeq,ueq)
. .
A
(x x
eq
)
. .
x
+(J
u
f)(x, u)

(x,u)=(xeq,ueq)
. .
B
(u u
eq
)
. .
u

= A x +B u.
Example 1.4.6. The governing equation of a nonlinear electrical circuit is given by
L
di
dt
+ 10 ln
i
2
20 = v(t),
where i(t) denotes the current and v(t), the applied voltage.
(a) Put the equation above into the nonlinear state-space form.
(b) Find an equilibrium current for the case when no voltage is supplied.
(c) Obtain a linearized state-space equation around this equilibrium point.
Example 1.4.7. Consider the magnetic levitation system shown below:
R
L
+

v
i
(t)
s(t)
i(t)
m
g
The dynamic equations are given as
v
i
=
L
0
s
di
dt
+Ri
m s =mg K
i
2
s
2
8
where
s : elevation of mass i : current
v
i
: input voltage L
0
: nominal inductance
K : a positive constant g : gravitational acceleration
(a) Put the system equations in state-space form. Make sure you specify your state variables
explicitly.
(b) Find the required equilibrium values for the input and the state variables for the mass to be
in equilibrium at s
eq
.
(c) Linearize the system equations about this equilibrium point.
Example 1.4.8. Consider the liquid container of varying cross-sectional area below:
h
w
in
w
out
A(h)
By the conservation of mass, we obtain
d (V (h))
dt
= w
in
w
out

dV (h)
dt
= w
in

(P
a
+gh P
a
)
1/
R
= w
in

gh
R
if = 1.
Note that
V (h(t)) =
_
h(t)
0
A() d
and recall Leibnitz rule:
d
d
_
_
h()
g()
f(s, ) ds
_
= f (h(), )
dh
d
f(g(), )
dg
d
+
_
h()
g()
f

(s, ) ds.
Therefore,

_
A(h)

h 0 + 0
_
= w
in

g
R
h
9
Chapter 2
The Laplace Transform
2.1 Complex Numbers
Denition 2.1.1. The imaginary unit j is dened as
j :=

1.
Denition 2.1.2. The set of complex numbers is dened as
C := a +bj : a, b IR
Denition 2.1.3. Given z = a +bj, its complex conjugate is dened as
z := a bj.
Denition 2.1.4. Given z = a +bj, its modulus is dened as
[z[ :=
_
a
2
+b
2
=

z z.
Proposition 2.1.5. Complex numbers satisfy the following:
(i) z
1
+z
2
= z
1
+ z
2
(ii) z
1
z
2
= z
1
z
2
Lemma 2.1.6. Given a polynomial with real coecients, its roots are symmetric about the real
axis. (That is, if x

is a root, so is x

.)
2.2 The Laplace Transform
Denition 2.2.1. Given f(t), its Laplace transform is dened as
/[f] :=
_

0
f(t)e
st
dt
=: F(s).
for any s C for which the integral is well-dened.
10
Lemma 2.2.2. The Laplace transform satises the following:
(i) /[af
1
+bf
2
] = aF
1
(s) +bF
2
(s)
(ii) /
_
df
dt
_
= sF(s) f(0)
(ii) /
_
d
n
f
dt
n
_
= s
n
F(s)
n1

k=0
s
n1k
f
(k)
(0)
(iii) /
__
t
0
f() d
_
=
F(s)
s
(iv) /[f(t )] = e
s
F(s)
(v) /
_
e
at
f(t)

= F(s +a)
(vi) /[f
1
f
2
] = F
1
(s)F
2
(s) where
(f
1
f
2
)(t) :=
_

0
f
1
()f
2
(t ) dt
=
_

0
f
2
()f
1
(t ) dt
(vii) /[f
1
f
2
] = (F
1
F
2
)(s) where
(F
1
F
2
)(s) :=
1
2j
_
+j
j
F
1
()F
2
(s ) d
=
1
2j
_
+j
j
F
2
()F
1
(s ) d
(viii) /[tf(t)] =
dF(s)
ds
Proof. (i) Trivial.
(ii) Using integration by parts,
_

0
e
st
df()
d

=t
dt = e
st
f(t)

0

_

0
(s)e
st
f(t) dt
= lim
t
e
st
f(t) e
s0
f(0) +s
_

0
e
st
f(t) dt
= 0 f(0) +sF(s) for Re(s) > 0
11
(iii) By denition,
/
__
t
0
f() d
_
=
_

0
e
st
_
t
0
f() d dt
=
_

0
__

e
st
dt
_
f() d
=
_

0

1
s
e
st

f() d
=
1
s
_

0
(e
s
e
s
)f() d
=
1
s
_

0
e
s
f() d for Re(s) > 0
=
F(s)
s
.
(iv) By denition,
/[f(t )] =
_

0
e
st
f(t ) dt
=
_

e
s(+)
f() d, := t
= e
s
_

e
s
f() d,
= e
s
_

0
e
s
f() d, since f(t) = 0 for t < 0
= e
s
F(s).
(v) By denition,
/
_
e
at
f(t)

=
_

0
e
st
e
at
f(t) dt
=
_

0
e
(s+a)t
f(t) dt
= F(s +a).
(vi) By denition,
f
1
f
2
=
_

f
1
()f
2
(t ) d.
But since we assume f
i
(t) = 0 for t < 0 for i = 1, 2, we have
f
1
f
2
=
_
t
0
f
1
()f
2
(t ) d.
12
Then,
/[f
1
f
2
] =
_

0
e
st
_

f
1
()f
2
(t ) d dt
=
_

0
e
st
_
t
0
f
1
()f
2
(t ) d dt
=
_

0
_
t
0
e
st
f
1
()f
2
(t ) d dt
=
_

0
f
1
()
_

e
st
f
2
(t ) dt d
=
_

0
f
1
()e
s
_

e
s(t)
f
2
(t ) dt d
=
_

0
f
1
()e
s
_

0
e
s
f
2
() dd := t
=
_

0
f
1
()e
s
d
_

0
e
s
f
2
() d
= F
1
(s)F
2
(s).
(vii) Skip.
(viii) By denition,
/[tf(t)] =
_

0
tf(t)e
st
dt
=
_

0
f(t)
_
te
st
_
dt
=
_

0
f(t)
de
st
ds
dt
=
d
ds
_

0
f(t)e
st
dt by Leibnitz rule
=
dF(s)
ds

2.2.1 Examples
Denition 2.2.3. The Dirac delta function (a.k.a. the impulse function) is dened as
(t) := lim
0
P

(t),
where
P

(t) :=
_
1/ for 0 t
0 for t > 0
13
Example 2.2.4. We now show that
/[(t)] = 1.
Consider the following:
/[(t)] =
_

0
(t)e
st
dt
=
_

0
lim
0
P

(t)e
st
dt
= lim
0
_

0
P

(t)e
st
dt
= lim
0
_

0
1

e
st
dt
= lim
0
1

_

0
e
st
dt
= lim
0
1

1
(s)
e
st

0
= lim
0
1
s
(e
s
1)
By LHospitals rule,
lim
0
1 e
s
s
= lim
0
se
s
s
= lim
0
e
s
= 1.
Example 2.2.5. The unit step function is dened as
1(t) :=
_
0 t < 0
1 t 0.
Then, for any Re(s) > 0,
/1(t) =
_

0
e
st
1(t) dt
=
_

0
e
st
dt
=
e
st
s

0
=
_
0
1
s
_
=
1
s
.
14
Example 2.2.6. We now show that
/[sin t] =

s
2
+
2
.
Note that
sin t =
e
jt
e
jt
2j
.
Then, for any Re(s) > 0, we have
/[sin t] =
_

0
_
e
jt
e
jt
2j
_
e
st
dt
=
1
2j
_

0
_
e
(js)t
e
(js)t
_
dt
=
1
2j
_
e
(js)t
j s
+
e
(js)t
j +s
_

0
=
1
2j
_

1
j s

1
j +s
_
=

s
2
+
2
Example 2.2.7. We show that
/
_
e
at
cos t

=
s +a
(s +a)
2
+
2
.
Since
cos t =
e
jt
+e
jt
2
,
we have
_

0
e
st
e
at
e
jt
+e
jt
2
dt =
1
2
_

0
_
e
(s+aj)t
+e
(s+a+j)t
_
dt
=
1
2
_

e
(s+aj)t
s +a j

e
(s+a+j)t
s +a +j
_

0
=
1
2
_
1
s +a j
+
1
s +a +j
_

0
=
s +a
(s +a)
2
+
2
.
2.3 The Inverse Laplace Transform
Denition 2.3.1. A rational function of s is a ratio of two polynomials in s. A rational function
is said to be
Proper if deg(den) deg(num)
15
Strictly proper if deg(den) > deg(num)
Bi-proper if deg(den) = deg(num)
Improper if deg(den) < deg(num)
Now concentrate on strictly proper functions. Suppose a given F(s) is in the form
F(s) =
b
m
s
m
+b
m1
s
m1
+ +b
1
s +b
0
s
n
+a
n1
s
n1
+ +a
1
s +a
0
where n > m. One can rewrite F as
F(s) = K

m
i=1
(s z
i
)

m
j=1
(s p
j
)
Two dierent cases are possible:
Distinct roots: In this case p
j
,= p
k
for any j ,= k and we can nd C
j
s such that
F(s) =
C
1
s p
1
+
C
2
s p
2
+ +
C
n
s p
n
.
Each C
j
can be obtained from
C
j
= (s p
j
)F(s)[
s=p
j
.
After all C
j
s are obtained, the inverse Laplace transform of F(s) becomes
f(t) =
n

j=1
C
j
e
p
j
t
Example 2.3.2. Consider
F(s) =
1
s
2
2s 3
=
1
(s + 1)(s 3)
=
C
1
s + 1
+
C
2
s 3
.
Then,
C
1
= (s + 1)F(s)[
s=1
=
1
s 3

s=1
=
1
4
and
C
2
= (s 3)F(s)[
s=3
=
1
s + 1

s=3
=
1
4
.
Therefore,
F(s) =
1
4
_
1
s 3

1
s + 1
_
so that
f(t) =
1
4
_
e
3t
e
t
_
.
16
Example 2.3.3. Consider
F(s) =
1
s(s
2
+s + 1)
=
C
1
s p
+
C
2
s p
+
C
3
s
,
where p :=
1
2
+

3
2
j. One can show that
C
1
=

C
2
=
1
2
+

3
6
j and C
3
= 1.
Then, the inverse Laplace transform is
f(t) =
_

1
2
+

3
6
j
_
e
(
1
2
+

3
2
j)t
+
_

1
2

3
6
j
_
e
(
1
2

3
2
j)t
+ 1.
After a lengthy calculation, f(t) simplies to
f(t) = 1 e
t/2
cos

3
2
t
1

3
e
t/2
sin

3
2
t.
One could also write
F(s) =
K
1
s +K
2
s
2
+s + 1
+
K
3
s
and solve for K
i
s as K
1
= K
2
= 1 and K
3
= 1. Then, using the facts that
/
_
e
at
cos bt

=
s +a
(s +a)
2
+b
2
and /
_
e
at
sin bt

=
b
(s +a)
2
+b
2
,
we can write
F(s) =
_
s + 1/2
(s + 1/2)
2
+ 3/4
+
1

3
_
3/4
(s + 1/2)
2
+ 3/4
_
+
1
s
and obtain the same result.
Repeated roots: Suppose the rst root is repeated k times, where the others are distinct.
Then, F(s) can be expressed as
F(s) =
C
1
s p
1
+
C
2
(s p
1
)
2
+ +
C
k
(s p
k
)
k
+
C
k+1
s p
k+1
+ +
C
n
s p
n
.
In order to nd C
1
through C
k
, use
C
ki
=
1
i!
_
d
i
ds
i
_
(s p
1
)
k
F(s)
_
_
s=p
1
i = 0 : k 1.
Example 2.3.4. Consider
F(s) =
s + 1
(s + 2)
2
(s 3)
=
C
1
s + 2
+
C
2
(s + 2)
2
+
C
3
s 3
.
17
For the pole at s = 2, we have k = 2. By applying the formula above,
C
1
= C
21
=
1
1!
_
d
ds
_
(s + 2)
2
F(s)
_
_
s=2
=
_
d
ds
_
s + 1
s 3
__
s=2
=
_
4
(s 3)
2
_
s=2
=
4
25
and
C
2
= C
20
=
1
0!
_
d
0
ds
0
_
(s + 2)
2
F(s)
_
_
s=2
=
_
s + 1
s 3
_
s=2
=
1
5
.
Finally,
C
3
= (s 3)F(s)[
s=3
=
s + 1
(s + 2)
2

s=3
=
4
25
.
Then,
F(s) =
4
25
1
s + 2
+
1
5
1
(s + 2)
2
+
4
25
1
s 3
and the inverse Laplace transform is
f(t) =
4
25
e
2t
+
1
5
te
2t
+
4
25
e
3t
.
Theorem 2.3.5 (Final Value Theorem). If all poles of sF(s) lie in the left-half plane, then
lim
t
f(t) = lim
s0
sF(s).
Proof. First note that
/
_
df
dt
_
=
_

0
e
st
df
dt
dt = sF(s) f(0).
Due to the assumption that all poles of sF(s) lie in the left-half plane, we know that the region of
convergence for the Laplace transform of
df
dt
includes s = 0. Hence, we can take the limit of both
sides as s 0 and obtain
lim
s0
_
sF(s) f(0)
_
= lim
s0
_

0
e
st
df
dt
dt =
_
f()
f(0)
df.
Therefore,
lim
s0
sF(s) f(0) = f() f(0).
Hence,
lim
s0
sF(s) = f().

18
2.3.1 Steady-State Behavior with respect to Pole Locations
(i) Real root in C

:
Re
Im
t
=
x(t)
X(s) =
1
sa
, a < 0
x(t) = e
at
, a < 0
(ii) Complex conjugate pair in C

:
Re
Im
t
=
X(s) =

(s+a)
2
+
2
, a > 0
(iii) Single pole at the origin:
t
Re
Im
=
X(s) =
1
s
x(t) = 1(t)
(iv) Repeated pole at the origin:
t
Re
Im
=
X(s) =
1
s
2 x(t) = t
19
(v) Complex conjugate pair on C
0
:
t
Re
Im
=
X(s) =

s
2
+
2 x(t) = sin t
(vi) Real pole in C
+
:
Re
Im
t
=
X(s) =
1
sa
, a > 0
x(t) = e
at
, a > 0
(vii) Complex conjugate pair in C
+
:
2.4 Solution of ODEs via Laplace Transforms
Example 2.4.1. Consider the initial value problem
x +x = 1, with x(0) = 3, x(0) = 2.
Taking Laplace transforms,
s
2
A(s) sx(0) x(0) +A(s) =
1
s
,
so that
s
2
A 3s + 2 +A =
1
s
.
Solving for A,
A =
3s
2
2s + 1
s(s
2
+ 1)
=
1
s
+ 2
s
s
2
+ 1
2
1
s
2
+ 1
.
Taking the inverse Laplace transform,
x(t) = 1 + 2 cos t 2 sin t.
20
Example 2.4.2. Consider the system
F = m x
and solve for x(t) assuming F(t) = F
0
cos t and x(0) = 0, x(0) = 0.
Rewrite as
m x = F
0
cos t
ms
2
A(s) = F
0
s
s
2
+
2
A(s) =
F
0
ms(s
2
+
2
)
=
F
0
m
_
C
1
s
+
C
2
s +C
3
s
2
+
2
_
Example 2.4.3. Consider the system
F = m x
and solve for x(t) assuming F(t) = sin t.
21
Chapter 3
Transfer Functions and Stability
3.1 Obtaining Transfer Functions from ODEs
Example 3.1.1. The governing equation of a spring-mass-damper system is
m x +c x +kx = F.
We assume that x is the output of the system. By taking the Laplace transform of both sides, we
obtain
m(s
2
A(s) sx(0) x(0)) +c(sA(s) x(0)) +kA(s) = T(s).
Assuming x(0) = 0 and x(0) = 0, we obtain
ms
2
A(s) +csA(s) +kA(s) = T(s).
Equivalently,
A(s) =
1
ms
2
+cs +k
T(s).
Hence, the transfer function of the system is
G(s) :=
1
ms
2
+cs +k
.
Example 3.1.2. The equations governing a DC motor are
IR +L
dI
dt
+K
m

= v
i
and J

+b

= K
t
I.
Taking Laplace transforms, we obtain
JR +LsJ +K
m
s = V
i
and Js
2
+bs = K
t
J.
One can solve
J = s
_
J
K
t
s +
b
K
t
_

22
and plug it in the other equation to obtain
=
1
(R +Ls)s
_
J
Kt
s +
b
Kt
_
+K
m
s
V
i
.
Note that with =

, one could also write
=
1
(R +Ls)
_
J
Kt
s +
b
Kt
_
+K
m
V
i
.
Remark 3. (Initial conditions as impulse inputs) Consider the spring-mass-damper system with
no force applied, but a non-zero initial velocity. The Laplace transform gives
m(s
2
A(s) x(0)) +csA(s) +kA(s) = 0.
One can rewrite this as
ms
2
A(s) +csA(s) +kA(s) = m x(0).
Hence, one can think of the eect of the non-zero initial condition x(0) as an impulse input of
intensity m x(0). A similar, but more strained argument can be given for x(0) ,= 0.
3.2 (Bounded Input-Bounded Output) Stability
Denition 3.2.1. A signal x(t) is said to be bounded if there exists an M
x
< such that
[x(t)[ M
x
t.
Denition 3.2.2. A system with input u and output y is said to be bounded input-bounded
output (BIBO) stable if every bounded input u results in a bounded output y.
Theorem 3.2.3. BIBO stable i all poles of G(s) lie in C

.
Proof. (If) Assume that all poles of G(s) lie in C

. Let u be a bounded input, with bound M

u
.
Then,
y(t) =
_
t
0
g()u(t ) d
[y(t)[ =

_
t
0
g()u(t ) d

_
t
0
[g()u(t )[ d
=
_
t
0
[g()[ [u(t )[ d

_
t
0
[g()[ M
u
d
= M
u
_
t
0
[g()[ d
23
If G(s) has poles only in C

, there exists an M
g
< such that
_
t
0
[g()[ d M
g
t.
Now dene M
y
:= M
u
M
g
to conclude that
[y(t)[ M
y
t.
(Only if) If not all poles of G lie in C

, one can always nd a bounded input that will result in

an unbounded output.
Remark 4. Obviously, BIBO stability is equivalent to the condition that the impulse response of
the system dies out as t .
Remark 5. When no input is applied, BIBO stability is equivalent to the fact that the output goes
to zero asymptotically in response to any initial condition.
Remark 6. For a quadratic polynomial of the form
d(s) = s
2
+as +b,
the roots have negative real parts i a > 0 and b > 0. For any polynomial of higher order, we need
the Routh-Hurwitz stability criterion.
3.3 Routh-Hurwitz Stability Criterion
Let
d(s) = s
n
+a
n1
s
n1
+ +a
1
s +a
0
.
We say d is stable if all of its roots lie in C

.
Lemma 3.3.1. If d is stable, then a
i
> 0 for all i = 0 : n 1.
Proof. For a rst-order term with a real, negative root, the result is trivial: d(s) = s a, with the
root a < 0. For a second-order term with a complex conjugate pair of roots a bj with a < 0,
d(s) = (s (a +bj))(s (a bj)) = (s a)
2
(bj)
2
= s
2
2as +a
2
+b
2
.
Since a < 0, all coecients are positive again. Any other polynomial is a product of rst-order and
second order terms. The coecients are only multiplied with and added to each other. Hence, all
coecients remain positive.
Lemma 3.3.2. If n = 2, d is stable i a
1
> 0 and a
2
> 0.
Proof. Necessity is proven already. Suppose a
1
> 0 and a
2
> 0. The roots are
s
1,2
= a
1

_
a
2
1
4a
2
.
If a
2
1
4a
2
0, then 1(s
1,2
) = a
1
< 0.
24
If a
2
1
4a
2
> 0, then, a
1
>
_
a
2
1
4a
2
> 0 so that s
1,2
are both real and negative.

Example 3.3.3. The equations governing the lateral motion of a vehicle are
mv(

+

) = c
F
_

W

L
F

v
_
+c
R
_
+
L
R

v
_
J
Z

= c
F
L
F
_

W

L
F

v
_
c
R
L
R
_
+
L
R

v
_
.
The denominator of the transfer function from to

is
d(s) = s
2
+
(J
z
+mL
2
F
)c
F
+ (J
z
+mL
2
R
)c
R
J
z
mv
. .
a
1
s +
c
F
c
R
L
2
+mv
2
(c
R
L
R
c
F
L
F
)
J
z
mv
2
. .
a
2
.
If v < 0, then a
1
< 0 so that the system is unstable.
If v > 0, then a
1
> 0. Moreover,
If c
R
L
R
c
F
L
F
0, then a
2
> 0 also, so that the system is stable.
If c
R
L
R
c
F
L
F
< 0, then the system is
_
stable if v < v
critical
unstable if v v
critical
,
where v
critical
:=
_
c
F
c
R
L
2
m(c
F
L
F
c
R
L
R
)
.
Remark 7. For n 3, positiveness of the coecients does not imply stability of the polynomial.
Example 3.3.4. Consider
d(s) = s
4
+ 2s
3
+ 3s
2
+ 4s + 5.
Obviously, all coecients are positive, yet the roots are s
1,2
= 1.2880.858j, s
3,4
= 0.2881.416j.
3.3.1 RH Stability Criterion - Version 1
Form the Routh array:
Row n: 1 a
n2
a
n4

Row n 1: a
n1
a
n3
a
n5

Row n 2: b
1
b
2
b
3

Row n 3: c
1
c
2
c
3

.
.
.
.
.
.
.
.
.
.
.
.
Row 1: 0
Row 0:
25
where
b
1
:=
1
a
n1
det
_
1 a
n2
a
n1
a
n3
_
b
2
:=
1
a
n1
det
_
1 a
n4
a
n1
a
n5
_
b
3
:=
1
a
n1
det
_
1 a
n6
a
n1
a
n7
_
c
1
:=
1
b
1
det
_
a
n1
a
n3
b
1
b
2
_
c
2
:=
1
b
1
det
_
a
n1
a
n5
b
1
b
3
_
c
3
:=
1
b
1
det
_
a
n1
a
n7
b
1
b
4
_
and so on.
Lemma 3.3.5. The system is stable if and only if all elements of the rst column of the Routh
array are positive. Moreover, the number of unstable poles is equal to the number of sign changes
in the rst column.
Example 3.3.6. Consider the polynomial d(s) = s
4
+ 2s
3
+ 3s
2
+ 4s + 5 again. The Routh array
becomes
1 3 5
2 4 0
b
1
b
2
0 c
1
0
d
1
We compute
b
1
=
1
2
det
_
1 3
2 4
_
= 1 b
2
=
1
2
det
_
1 5
2 0
_
= 5
c
1
=
1
1
det
_
2 4
1 5
_
= 6
d
1
=
1
6
det
_
1 5
6 0
_
= 5
The resulting Routh array is
1 3 5
2 4 0
1 5 0
-6 0
5
Due to the entry c
1
= 6, the system is unstable. In fact, there are two sign changes in the rst
column (from 1 to -6 and from -6 to 5), indicating that there are two unstable poles. Recall that
the roots are indeed 1.288 0.858j, 0.288 1.416j.
26
3.3.2 RH Stability Criterion - Version 2
Form the Routh-Hurwitz matrix associated with polynomial d as
_
_
_
_
_
_
_
_
_
_
_
_
_
a
n1
a
n3
0 0
1 a
n2
0 0
0 a
n1
0 0
0 1 0 0
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
0 0 a
0
0
0 0 a
1
0
0 0 a
2
a
0
_
_
_
_
_
_
_
_
_
_
_
_
_
.
Let

1
:= a
n1
,
2
:= det
_
a
n1
a
n3
1 a
n2
_
,
3
:= det
_
_
a
n1
a
n3
a
n5
1 a
n2
a
n4
0 a
n1
a
n3
_
_
and so on.
Lemma 3.3.7. The system is stable if and only if all coecients are positive and
_

i
> 0 for i = 2, 4, , n 1 if n is odd

i
> 0 for i = 1, 3, , n 1 if n is even.
Example 3.3.8. Consider d(s) = s
4
+2s
3
+3s
2
+4s +5 again. The Routh-Hurwitz matrix becomes
_
_
_
_
2 4 0 0
1 3 5 0
0 2 4 0
0 1 3 5
_
_
_
_
.
Since n = 4, we need only to check
1
and
3
. Since
1
= a
n1
= 2 > 0, check
3
only.

3
= det
_
_
2 4 0
1 3 5
0 2 4
_
_
= 12 < 0.
Hence, the system is unstable.
Remark 8. The second version is simpler, but provides less information.
3.4 Examples
Example 3.4.1. Find the range of stabilizing K values for the feedback system in Figure 3.1.
27
s+1
s(s1)(s+6)
K
R Y
Figure 3.1:
The transfer function from R to Y is
G
cl
=
K(s + 1)
s(s 1)(s + 6) +K(s + 1)
.
The denominator is
d(s) = s
3
+ 5s
2
+ (K 6)s +K.
For stability, we need
K 6 > 0 and K > 0.
The RH matrix is
_
_
5 K 0
1 K 6 0
0 5 K
_
_
.
Since n = 3, need to check
2
only.

2
= det
_
5 K
1 K 6
_
= 4K 30.
Hence, we additionally require 4K 30 > 0, i.e., K > 7.5. All conditions for stability are satised
if K > 7.5.
Example 3.4.2. Find the range of stabilizing K
p
and K
I
values for the feedback system in Figure 3.2.
1
(s+1)(s+2)
K
p
+ K
I
/s
R Y
Figure 3.2:
The closed-loop system is
G
cl
(s) =
K
p
s +K
i
K
p
s +K
i
+s(s + 1)(s + 2)
.
The closed-loop denominator is
d(s) = s
3
+ 3s
2
+ (2 +K
p
)s +K
i
.
For stability, we rst need
2 +K
p
> 0 and K
i
> 0.
The RH matrix becomes
_
_
3 K
i
0
1 2 +K
p
0
0 3 K
i
_
_
.
28
Since n = 3, we need to check
2
only.

2
= det
_
3 K
i
1 2 +K
p
_
= 6 + 3K
p
K
i
.
Example 3.4.3. Propose a stabilizing control law for the plant G(s) =
1
s
2
+ 1
.
Example 3.4.4 (Cart and pendulum). Consider the system shown in Figure 3.3.

F
m
m
C
L
g
Figure 3.3: Cart-and-pendulum system.
The equations of motion are found to be
mL
2

mgLsin +mLcos

= 0
(m
C
+m)

+mLcos

mLsin

2
= F.
Around the equilibrium point = 0,

= 0, F = 0 and any

, these equations simplify to
mL
2

mgL +mL

= 0
(m
C
+m)

+mL

= F.
The transfer function from F to is found to be
(s) =
1
mLs
2
+ (m
C
+m)(g Ls
2
)
F(s)
=
1
m
C
Ls
2
(m
C
+m)g

F(s),

F := F
=
1
s
2
a

F(s),

F :=
1
m
C
L

F, a :=
(m
C
+m)g
m
C
L
.
Suppose we use a PD control law

F(s) = (K
p
+K
d
s)(s).
The denominator of the closed-loop system transfer function becomes
d
cl
(s) = K
p
+K
d
s +s
2
a = s
2
+K
d
s + (K
p
a).
Therefore, stability is guaranteed as long as K
d
> 0 and K
p
a > 0.
29
Chapter 4
Steady-State Response
4.1 The Reference Tracking Problem
For a unity negative feedback loop,
Y =
GC
1 +GC
R.
With G = n
G
/d
G
and C = n
C
/d
C
,
Y =
n
G
n
C
n
G
n
C
+d
G
d
C
R.
The error can then be shown to be
E =
d
G
d
C
n
G
n
C
+d
G
d
C
R =
d
G
d
C
n
G
n
C
+d
G
d
C
n
R
d
R
.
For zero steady-state error, we want
lim
t
e(t) = 0.
Recall that if sE(s) has poles only in C

, the FVT says

lim
t
e(t) = lim
s0
sE(s) = lim
s0
sd
G
d
C
n
G
n
C
+d
G
d
C
n
R
d
R
.
In short, for asymptotic tracking,
Stabilize the closed-loop system. If d
R
is stable, this is sucient.
If d
R
is not stable, make sure its unstable roots are canceled with the term d
G
d
C
.
Remark 9. Consider dierent possibilities:
(i) Decaying r(t): If d
R
is a stable polynomial,
(ii) Step function: If r(t) = K 1(t)
(iii) Ramp function: If r(t) = Kt
(iv) Sinusoid: If r(t) =
30
4.2 Examples
Example 4.2.1. G(s) =
1
s
. Follow a step.
Example 4.2.2. G(s) =
1
s
2
+4s+5
. Follow a ramp.
Example 4.2.3. G(s) =
1
s(s+1)
. Follow a sinusoid.
31
Chapter 5
Transient Response
5.1 First-Order Systems
A standard rst-order system is any system that has a transfer function in the form
G(s) =
1
s + 1
,
where is called the time constant.
Example 5.1.1. Consider the longitudinal motion of a road vehicle with a resisting aerodynamic
force that is proportional to velocity. The model is
m v = F dv,
where F represents the traction force and dv is the wind resistance. Taking Laplace transforms,
V (s) =
1
ms +d
F(s).
Dening

F :=
F
d
, we obtain
V (s) =
1
s + 1

F, where := m/d.
5.1.1 Impulse Response
Y (s) =
1
s + 1
U(s), u(t) = (t)
=
1
s + 1
1
=
1/
s + 1/
.
Therefore,
y(t) =
1

e
t/
.
The smaller the time constant, the faster the decay.
32
5.1.2 Step Response
Y (s) =
1
s + 1
U(s), u(t) = 1(t)
=
1
s + 1

1
s
=
1
s


s + 1
=
1
s

1
s + 1/
.
Therefore,
y(t) = 1 e
t/
1 as t .
The smaller the time constant, the faster the response.
5.1.3 Ramp Response
Y (s) =
1
s + 1
U(s), u(t) = t
=
1
s + 1

1
s
2
=
1
s
2


s
+

2
s + 1
=
1
s
2


s
+

s + 1/
.
Therefore,
y(t) = t +e
t/
.
Note that
e(t) = t y(t) = e
t/
as t .
5.2 Second-Order Systems
A standard second-order system is any system that has a transfer function in the form
G(s) =

2
n
s
2
+ 2
n
s +
2
n
,
where is the damping ratio and
n
is the (undamped) natural frequency. We assume 0 < 1.
Otherwise, one can express G(s) as the product of two rst-order systems.
Example 5.2.1. Consider the transfer function model of a spring-mass-damper system:
X(s) =
1
ms
2
+cs +k
T(s)
=
k/m
s
2
+sc/m+k/m

T(s),

T := T/k.
33
Dene
n
:=
_
k/m and :=
c
2

km
and assume 0 < 1. We obtain the standard second-order
form
X(s) =

2
n
s
2
+ 2
n
s +
2
n

T(s).
5.2.1 Pole Locations with Varying
Consider the denominator d(s) = s
2
+ 2
n
s +
2
n
. The roots are
s
1,2
=
n
w
n
_

2
1
=
n
j
n
_
1
2
if 0 < 1
=: j
d
.
Re
Im
= 0
= 0
= 1
Figure 5.1: Pole locations with varying .
5.2.2 Impulse Response
(i) Critically damped system ( = 1):
In this case, the denominator is
d(s) = s
2
+ 2
n
s +
2
n
= (s +
n
)
2
.
In this case, with u(t) = (t),
Y (s) =

2
n
(s +
n
)
2
1.
Taking the inverse Laplace transform, we obtain
y(t) =
2
n
te
nt
.
(ii) Overdamped System ( > 1):
(iii) Underdamped System (0 < < 1):
34
Y (s) =

2
n
s
2
+ 2
n
s +
2
n
1
=

2
n
(s +
n
)
2
+ (1
2
)
2
n
=

n
_
1
2
_
1
2

n
(s +
n
)
2
+ (1
2
)
2
n
Therefore,
y(t) =

n
_
1
2
e
nt
sin
_
(1
2
)
n
t
(iv) Undamped System:
5.2.3 Step Response
y(t) = 1 e
nt
_
cos
d
t +

_
1
2
sin
d
t
_
,
where
d
:=
n
_
1
2
.
Rise Time
t
r
:= t
0.9
t
0.1
Can approximate:
t
r

=
1.8

n
for = 0.5

=
2.16 + 0.6

=
1.4
2
+ 1.1 + 1

n
Overshoot
M
p
:= max
t
y(t) 1
= exp
_

_
1
2
_
35
Settling Time
t
s
:= minT : [y(t) 1[ .01 t > T
=
4.6

n
obtained from exp(
n
t
s
) = 0.01.
5.3 PID Controllers
5.3.1 Structure of PID controllers
C(s) = K
p
_
1 +
1
T
i
s
+T
d
s
_
5.3.2 Ziegler-Nichols rules for Tuning PID Controllers
Assume the plant has the structure
G(s) = K
e
sL
s + 1
.
First Method
(i) Sketch step response of the system.
(ii) Dene L as the time-lag, R as the slope between 0 and nal value.
(iii) For a P-controller, use K
p
=
1
RL
(iv) For a PI-controller, use K
p
=
0.9
RL
and T
i
= L/0.3
(v) For a PID-controller, use K
p
= 1.2/RL, T
i
= 2L and T
d
= 0.5L.
Second Method
(i) Use C(s) = K and increase K until K
u
(the ultimate gain), where the response becomes
oscillatory.
(ii) Dene the period of the resulting oscillations as P
u
(the ultimate period).
(iii) For a P-controller, use P = 0.5K
u
(iv) For a PI-controller, use K
p
= 0.45 and T
i
=
Pu
1.2
(v) For a PID-controller, use K
p
= 1.6K
u
, K
i
= 0.5P
u
and K
d
= 0.125P
u
36
Chapter 6
Root Locus
Let
G(s) =
n(s)
d(s)
and d(s) = a(s) +Kb(s),
where K is some variable. Let n := deg(a) and m = deg(b) and m n. Assume
a(s) = s
n
+a
n1
s
n1
+ +a
1
s +a
0
=
n

i=1
(s p
i
)
and
b(s) = s
m
+b
m1
s
m1
+ +b
1
s +b
0
=
m

j=1
(s z
j
).
6.1 Simple Root locus rules
When K = 0, it is obvious that the poles of G are the roots of a(s). Otherwise, we dene L(s) :=
b(s)
a(s)
and investigate the equivalent condition
1 +KL(s) = 0.
6.1.1 K > 0
Any point on the real axis that is to the left of an odd (total) number of poles and zeros of
L is on the root locus.
As K ,
m-many of the n branches approach the m zeros of L
The remaining n m branches go to innity along asymptotes making angles

i
=
2i
n m
for i = 1 : n m
37
with the real axis that meet at the point
:=

i
p
i

j
z
j
n m
The root locus plot is symmetric about the real axis
6.2 Examples
Example 6.2.1 (Standard 2
nd
-order System). We are interested in
d(s) = s
2
+ 2
n
s +
2
n
.
Dene K := 2
n
and rewrite d(s) = 0 as
1 +K
s
s
2
+
2
n
= 0.
Re
Im
= 0
= 0
= 1
Example 6.2.2 (Approximate time delay). Consider the following feedback system:
1
s
e
Ts
1
s+1
We can approximate delay by
(i) e
Ts
=
1
1 +Ts
(ii) e
Ts
=
1 Ts/2
1 +Ts/2
(iii) e
Ts
=
1 Ts/2 + (Ts)
2
/12
1 +Ts/2 + (Ts)
2
/12
38
6.2.1 K < 0
Any point on the real axis that is to the left of an even (total) number of poles and zeros of
L is on the root locus.
As K ,
m-many of the n branches approach the m zeros of L
The remaining n m branches go to innity along asymptotes making angles

i
=
2i 2
n m
for i = 1 : n m
with the real axis that meet at the point
:=

i
p
i

j
z
j
n m
The root locus plot is symmetric about the real axis
In both cases, we have multiple poles when
d
ds
L(s) = 0 a

b ab

= 0.
39
Chapter 7
Frequency Response
7.1 Preliminaries
Let G be stable. Can show that the response at steady-state to the input
u(t) = sin t
will be
y
ss
(t) = [G(j)[ sin(t +G(j)).
Proof.
Y (s) = G(s)U(s)
=
n(s)
d(s)

s
2
+
2
=
n(s)
d(s)
+
C
s +j
+
C
s j
where
C = (s +j)G(s)

s
2
+
2

s=j
= G(j)

2j
= G(j)
1
2j
= G(j)
j
2
= G(j)
j
2
so that
C = G(j)
j
2
.
40
In the time domain, we obtain
y(t) = /
1
_
n(s)
d(s)
_
+Ce
jt
+Ce
jt
.
Since /
1
_
n(s)
d(s)
_
0 as t , we obtain
y
ss
(t) = Ce
jt
+Ce
jt
= C(cos t j sin t) +C(cos t +j sin t)
= (C +C) cos t +j(C +C) sin t
= 2Re(C) cos t + 2Im(C) sin t
= K sin(t + )
where
K :=
_
4Re(C)
2
+ 4Im(C)
2
= 2
_
Re(C)
2
+ Im(C)
2
= 2

Re
_
j
2
G(j)
_
2
+ Im
_
j
2
G(j)
_
2
=
_
Re
_
jG(j)
_
2
+ Im
_
jG(j)
_
2
=
_
Im(G(j))
2
+ Re (G(j))
2
= [G(j)[
and
= tan
1
2Re(C)
2Im(C)
= tan
1
Re
_
j
2
G(j)
_
Im
_
j
2
G(j)
_
= tan
1
Im(G(j))
Re (G(j))
= G(j)

7.1.1 Determination of Signal Spectra via Fourier Transforms

Periodic Signals
With x(t) = x(t +T) for some T (0, ), we have
x(t) =

k=
a
k
e
jk
0
t
,
41
where
0
:=
2
T
and
a
k
=
1
T
_
T
x(t)e
jk
0
t
dt.
Note that if x(t) is real, we can also write
x(t) =

k=
a

k
e
jk
0
t
.
But since
x(t) =

k=
a
k
e
jk
0
t
,
we obtain
a

k
= a
k
.
In that case,
x(t) = a
0
+

k=1
_
a
k
e
jk
0
t
+a
k
e
jk
0
t
_
= a
0
+

k=1
_
a
k
e
jk
0
t
+a

k
e
jk
0
t
_
= a
0
+

k=1
2Re
_
a
k
e
jk
0
t
_
= a
0
+ 2

k=1
Re
_
A
k
e
j
k
e
jk
0
t
_
= a
0
+ 2

k=1
A
k
Re
_
e
j(k
0
t+
k
)
_
= a
0
+ 2

k=1
A
k
cos(k
0
t +
k
)
Non-periodic signals
Dene
X(j) =
_

x(t)e
jt
dt
so that
x(t) =
1
2
_

X(j)e
jt
d.
42
7.2 Bode Plots
Given G(s), its Bode plots are
[G(j)[
dB
:= 20 log [G(j)[ vs. log (Bode magnitude plot)
G(j) vs. log (Bode phase plot)
Note that
[G
1
G
2
[
dB
= [G
1
[
dB
+[G
2
[
dB
and G
1
G
2
= G
1
+G
2
.
7.2.1 Straight Line Approximations
(i) G(s) = K [G(j)[ = [K[.
[G(j)[
dB
= 20 log K.
(ii) G(s) = j [G(j)[ = .
[G(j)[
dB
= 20 log w.
(iii) G(s) = 1 +
j

1
[G(j)[ =

1 +

2

2
1
.
[G(j)[
dB
=
_

_
0 for
1
20 log

= 3 dB for =
1
20 log
1
+ 20 log for
1
(iv) G(s) = 1 +
2

n
j

2

2
n
[G(j)[ =

_
1

2

2
n
_
2
+
4
2

2
n
.
[G(j)[
dB
=
_

_
0 for
1
20 log 2 for =
1
40 log
n
+ 40 log for
1
7.2.2 Bodes Gain-Phase Relationship
Denition 7.2.1. The transfer function G(s) is said to be minimum-phase if it has no open
RHP zeros.
Theorem 7.2.2. Given a stable and minimum-phase G(s),
G(j
0
) =
1

_
d
du
ln [G(j)[
_
ln(coth [u[/2) du,
where u := ln /
0
.
As a consequence, we have
G(j)

= n

2
,
where n is the slope of the Bode magnitude plot.
43
7.2.3 Unstable and/or Non-minimum-phase systems
Example 7.2.3. Consider
G(s) =
1 s
1 +s/10
and G
p
(s) :=
1
(1 +s)(1 +s/10)
.
Note that G
p
is stable and minimum-phase and G
p
= G.
7.2.4 Examples
Example 7.2.4 (Butterworth lter). A low-pass Butterworth lter is a transfer function with the
property
[B(j)[
2
=
1
1 + (/
c
)
2N
,
where N is referred to as the order of the lter. Since [B(j)[
2
= B(j)B(j)

and for a real

lter, B(j) = B(j), then,
R(s) := B(s)B(s) =
1
1 + (s/j
c
)
2N
.
The poles of R(s) are at
p = (1)
1/(2N)
j
c
so that
[p[ =
c
and p =
(2k + 1)
2N
+

2
, k :.
In other words,
p =
c
exp
_
j
_
(2k + 1)
2N
+

2
__
.
Hence, for R(s), there are 2N poles equally spaced in angle on a circle of radius
c
in C. In order to
construct a stable B(s), note that since R(s) = B(s)B(s), if p is a pole, then so is p. Therefore,
we choose the poles of R(s) is the LHP and obtain
For N = 1,
B(s) =
1
1 +s/
c
=

c
s +
c
.
For N = 2,
B(s) =
1
(1 +s/(
c
e
j/4
))(1 +s/(
c
e
j/4
))
=

2
c
s
2
+

2
c
s +
2
c
.
For N = 3,
B(s) =
Example 7.2.5 (Crossover Network). Consider the loudspeaker cross-over designs given below.
44
L
C
v
i
v
o L
C
v
i
v
o
(A) (B)
Each driver is modeled as a resistor of resistance R. Draw Bode magnitude plots for each circuit
and specify which connection yields a low-pass lter and which, a high-pass lter. (I.e., which
driver is a woofer and which is a tweeter?)
(A) By KVL, we have
v
i
= v
L
+v
C
and v
C
= v
o
.
By KCL,
I
L
= I
C
+I
R
.
Hence,
1
L
_
t
0
v
L
() d = C
dv
C
(t)
dt
+
v
o
R

1
L
_
t
0
(v
i
() v
o
()) d = C
dv
o
(t)
dt
+
v
o
R

1
L
(v
i
() v
o
()) = C
d
2
v
o
(t)
dt
2
+
1
R
dv
o
(t)
dt
v
i
(t) = LC
d
2
v
o
(t)
dt
2
+
L
R
dv
o
(t)
dt
+v
o
(t).
Hence, by taking Laplace transforms,
V
o
(s) =
1
LCs
2
+
L
R
s + 1
V
i
(s).
This is a second-order low-pass lter with cut-o frequency
c
=
1

LC
.
(B) By KVL, we have
v
i
= v
C
+v
L
and v
L
= v
o
.
By KCL,
I
C
= I
L
+I
R
.
Hence,
C
dv
C
(t)
dt
=
1
L
_
t
0
v
L
() d +
v
o
R
C
d(v
i
(t) v
o
(t))
dt
=
1
L
_
t
0
v
o
() d +
v
o
R
C
d
2
(v
i
(t) v
o
(t))
dt
2
=
1
L
v
o
(t) +
1
R
dv
o
(t)
dt
45
LC
d
2
v
i
(t)
dt
2
= LC
d
2
v
o
(t)
dt
2
+v
o
(t) +
L
R
dv
o
(t)
dt
By taking Laplace transforms,
V
o
(s) =
LCs
2
LCs
2
+
L
R
s + 1
V
i
(s).
Example 7.2.6 (Time delay). Consider the time-delay transfer function
G(s) = e
sT
.
Then,
G(j) = e
jT
= cos T j sin T.
Hence,
[G(j)[ = 1 and G(j) = tan
1
_

sin T
cos T
_
= T.
However, if we approximate e
Ts
using rational functions (Pade approximations), we obtain
Example 7.2.7 (PID controllers). Consider
C(s) = K
P
+
K
I
s
+K
D
s and C

(s) = K
P
+
K
I
s
+K
D
s
s + 1
.
Compare the frequency responses
Example 7.2.8 (Lead and lag terms).
7.3 Nyquist Plots and the Nyquist Stability Criterion
Theorem 7.3.1 (The Argument Principle). Given G(s) and a simple, closed, CW contour (,
assume ( contains m zeros and n poles of G and that it does not pass through any pole or zero of
G. Then, the number of CW encirclements of the origin by G(s) as s varies along ( is given by
N := m n.
Remark 10. N is referred to as the winding number. If N < 0, it means that the origin is
encircled in the CCW direction.
Example 7.3.2. Consider G(s) =
s 2
(s + 1)(s 1)
. The argument principle is demonstrated for several
dierent contours in the following graph.
46
G(C)
N = 1
N = 1
C
C
G(C)
G(C)
N = 2
N = 0
C
C
G(C)
N = 0
C
G(C)
7.3.1 The Nyquist Stability Criterion
Consider a standard negative unity feedback loop. The closed-loop transfer function is
G
cl
=
GC
1 +GC
.
Now consider the following sequence of statements:
The closed loop is stable i the term 1 +GC has no zeros in the closed RHP.
Note that the poles of 1 +GC are the poles of GC.
Assume the term GC has no poles on the imaginary axis.
Let 1 +GC have m zeros and n poles inside the closed RHP.
Consider the so-called Bromwich contour that covers the entire RHP:
47
R
Re
Im
By the argument principle, 1+GC encircles the origin in the CW direction N = m n times.
The closed-loop system is stable i m = 0.
Equivalently, the closed-loop system is stable i the plot of 1 + GC encircles the origin n
times in the CW direction.
Equivalently, the closed-loop system is stable i the plot of 1 + GC encircles the origin n
times in the CCW direction.
Nyquist Stability Criterion: the closed-loop system is stable i the plot of GC (as s varies
along the Bromwich contour) encircles 1 exactly n times in the CCW direction.
Example 7.3.3. Consider
GC =
K
(s + 2)(s
2
+ 2s + 5)
.
For the rst section of the Bromwich contour, we rst obtain the Bode plot of GC for K = 1:
48
120
110
100
90
80
70
60
50
40
30
20
M
a
g
n
itu
d
e
(d
B
)
10
2
10
1
10
0
10
1
10
2
270
180
90
0
P
h
a
s
e
(d
e
g
)
Bode Diagram
Frequency (rad/sec)
Figure 7.1: Bode plot for K = 1
Based on this Bode plot, we obtain the following Nyquist plot for GC:
1 0.8 0.6 0.4 0.2 0 0.2 0.4 0.6 0.8 1
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1


Nyquist Diagram
Real Axis
Im
a
g
in
a
ry
A
x
is
K=1
K=10
10 8 6 4 2 0 2 4 6 8 10
10
8
6
4
2
0
2
4
6
8
10


Nyquist Diagram
Real Axis
Im
a
g
in
a
ry
A
x
is
K=100
Remark 11. If GC has imaginary-axis poles, modify the Bromwich contour so as to avoid having
poles on the boundary of the contour, but still cover the entire closed RHP.
49
Example 7.3.4. Consider
GC =
K
s(1 +s/4)(1 +s/10)
and modify the Bromwich contour as follows:
R
0
Re
Im
7.4 Control Design via Frequency Response
Note that
[GC[
dB
= [G[
dB
+[C[
dB
and GC = G+C.
Example 7.4.1. Consider an inverted pendulum with the equation of motion
mL
2

= mgLsin +T.
Dening u := T/(mL
2
), for small , we can write

=
g
L
+u,
so that the transfer function from u to becomes
G(s) =
1
s
2
g/L
.
50