AUTOMATIC CONTROL

Exercises

Department of Automatic Control

Lund University, Faculty of Engineering
June 2017

1
1. Model Building and Linearization

1.1 Many everyday situations may be described and analyzed using con-
cepts from automatic control. Analyze the scenarios below and try to
give a description that captures the relevant properties of the system:
• What should be controlled?
• What control signals are used?
• What measurement signals are available?
• Is the system affected by any disturbances?
• Is feedback or feedforward used for control?
• Draw a block diagram that describes the system. The block dia-
gram should show how the measurement signals, control signals,
and the disturbances are connected to the human (which here is
the controller), and to the process.

a. You take a shower and try to get desired temperature and flow of the
water.

b. You drive a car.

c. You boil potatoes on the stove.

1.2 A car drives on a flat road and we assume that friction and air
resistance are negligible. We want to study how the car is affected by
the gas pedal position u. We assume that u varies between 0 and 1,
and that the acceleration of the car is proportional to the gas pedal
position, a = ku.

u y
Bil

a. Write down the differential equation that describes the relation be-
tween the gas pedal position u and the velocity v of the car.

b. Let instead the position p of the car be the output signal, y = p.

Introduce the states x1 = p and x2 = v, and write the system on
state-space form.
c. Assume that the car is affected by air resistance that gives a counter
force that is proportional to the square of the velocity of the car. With
the gas pedal position as control signal and the velocity of the car as
measurement signal, the system may now be written as

ẋ = −mx2 + ku
y=x

The system is no longer linear (why?). Let k = 1 and m = 0.001. Find

the stationary velocity y0 that corresponds to the gas pedal being 10%
down, u0 = 0.1.

3
Chapter 1. Model Building and Linearization

d. Linearize the system around the stationary point in c.

1.3 In the right figure, a mass m is attached y( t)

to a wall with a spring and a damper.
The spring has a spring constant k and k
the damper has a damping constant c. It m
is assumed that k > c2 /4m. An external f ( t)
force f is acting on the mass. We denote c
the translation of the mass from its equilibrium position by y. Further,
we let f ( t) be the input signal and y( t) be the output signal. The force
equation gives
m ÿ = − ky − c ẏ + f
a. Introduce the states x1 = y and x2 = ẏ and write down the state
space representation of the system.

b. Assume that the system is at rest at t = 0 and that f ( t) changes

from 0 to 1 as a step at t = 0. What is the resulting y( t)? Sketch the
solution.

1.4 In the RLC circuit to the right, i R

the input and output voltages are + +
given by vin ( t) and vout ( t), re-
spectively. By means of Kirchhoff’s vin C vout
voltage law we see that
− L −

di
vin − Ri − vout − L =0
dt

For the capacitor, we additionally have

C v̇out = i

Introduce the states x1 = vout and x2 = v̇out and give the state space
representation of the system.

1.5 A cylindrical water tank with cross section q in

A has an inflow q in and an outflow q out. The
outlet area is a. Under the assumption that
the outlet area is small in comparison to
the cross
√ section of the tank, Torricelli’s law
vout = 2 ˆ h is valid and gives the outflow h
rate.
a. What would be a suitable state variable for q out
this system? Determine a differential equa-
tion, which tells how the state variable depends on the inflow q in .
b. Asume that measurement signal y is given by level h. Give a state-
space representation of the system.

c. Let the inflow be constant, q in = q0in . Determine the corresponding

constant tank level h0 and outflow q0out . Linearize the system around
this stationary point.

4
 Chapter 1. Model Building and Linearization

u( t)

r( t)

Figure 1.1 Satellite orbiting the earth.

1.6 Give the state space representation of the system

...
y + 3 ÿ + 2 ẏ + y = u

where u( t) and y( t) are the input and output, respectively. Choose

states x1 = y, x2 = ẏ and x3 = ÿ.

1.7 A process with output y( t) and input u( t) is described by the differ-

ential equation
√
ÿ + y + y ẏ = u2

a. Introduce states x1 = y, x2 = ẏ and give the state space representa-

tion of the system.

b. Find all stationary points ( x01 , x02 , u0 ) of the system.

c. Linearize the system around the stationary point corresponding to

u0 = 1.

1.8 Linearize the system

√
ẋ1 = x21 x2 + 2 sin u ( = f1 ( x1 , x2 , u))
√
ẋ2 = x1 x22 +2 cos u ( = f2 ( x1 , x2 , u))
x2
y = arctan + 2u2 ( = ˆ( x1 , x2 , u))
x1

around the stationary point u0 = π /4.

1.9 A simple model of a satellite, orbiting the earth, is given by the

differential equation

β
r̈( t) = r( t)ω 2 − + u( t)
r2 ( t)

where r is the satellite’s distance to the earth and ω is its angular

acceleration, see Figure 1.1. The satellite has an engine, which can
exert a radial force u.

5
Chapter 1. Model Building and Linearization

a. Introduce the state vector

 
 r( t) 
x( t) = 
 

ṙ( t)

and write down the nonlinear state space equations for the system.

b. Linearize the state space equations around the stationary point

r, ṙ, u = r0 , 0, 0
! " ! "

Consider r as the output and give the state space representation of

the linear system. Express r0 in β and ω.

6
2. Dynamical Systems

2.1 A dynamical system may be described in various ways – with a transfer

function, with a differential equation, and with a system of differential
equations on state-space form. In this problem we transfer between
the different representations for four examples of dynamical systems,
from biology, mechanics, electronics, and economics.

a. A model for bacterial growth in a bioreactor is given by

   
 10 1  0
ẋ = 
 x+
   u
−1 −1 1
 
y = 1 0 x

where u is the inflow of glucose to the reactor, and y is the bio

mass. Determine the transfer function from u to y, and a differential
equation that determines the relation between the input and output
signals of the system.

b. A simple model of a telescope is given by

d2 y dy
J 2
+D =u
dt dt

where yis the angle of the telescope to the earth surface, and u is
the torque from the motor that controls the telescope. Determine the
transfer function from u to y and write the system on state-space
form.

c. An electronic low pass filter is used at recordings to attenuate high

frequency noise. The input u is the original noisy signal, and the
output y is the recorded signal. The filter is given on state-space form
as

dx 1 1
=− x+ u
dt k k
y=x

Determine the transfer function from u to y.

d. The transfer function for a model that describes economical growth is

given by
γ
G(s) = 3
s + αs2 + β s
where the input u is the difference between savings and investments
in the economy, and the output y is GDP. Write the system on state-
space form.

2.2 Determine the transfer functions and give differential equations, de-
scribing the relation between input and output for the following sys-
tems, respectively.

7
Chapter 2. Dynamical Systems

a.    
 −2

ẋ = 
0  5
x+
   u
0 −3 2
 
y =  −1 1  x + 2u

b.    
 −7 2  3
ẋ = 
 x+
   u
−15 4 8
 
y =  −2 1  x

c.    
 −1 0  3
ẋ = 
 x+
   u
0 −4 2
 
y =  1 0  x + 5u

d.    
 1 4   − 1
ẋ = 
 x+
  u

−2 −3 1
 
y =  1 2  x + 3u

2.3 Determine the impulse and step responses of the systems in assign-
ment 2.2.

2.4 Derive the formula G(s) = C (s I − A)−1 B + D for a general system

ẋ = Ax + Bu
y = C x + Du

2.5 Consider the system

1
G(s) =
s2 + 4s + 3
a. Calculate the poles and zeros of the system. Is the system stable?

b. What is the static gain of the system?

c. Calculate the initial value and final value of the step response of the
system.

d. Calculate the initial value and final value of the impulse response of
the system.

e. Calculate the initial derivative of the step response of the system.

2.6 Consider the system

0.25
G(s) =
s2 + 0.6s + 0.25

a. Calculate the poles and zeros of the system.

8
 Chapter 2. Dynamical Systems

b. What is the static gain of the system?

c. Calculate and sketch the step response of the system.

2.7 Determine the transfer function and poles of the oscillating mass in
assignment 1.3. Explain how the poles move if one changes k and c,
respectively. Can the poles end up in the right half plane?

2.8 Determine the transfer function of

a. the RLC circuit in assignment 1.4,

b. the linearized tank in assignment 1.5.

2.9 Consider the linear time invariant system

   
dx  0 −1  1
=
 x+
   u
dt 1 0 0
 
y =  1 −1  x

a. Is the system asymptotically stable?

b. Is the system stable?

2.10 Does the transfer function

s+4
G(s) =
s3 + 2s2 + 3s + 7

have any poles in the right half plane?

2.11 Determine which five of the following transfer functions correspond to

the step responses A–E below.

0.1 4
G1 (s) = G2 (s) =
s + 0.1 s2
+ 2s + 4
0.5 −0.5
G3 (s) = 2 G4 (s) = 2
s − 0.1s + 2 s + 0.1s + 2
1 4
G5 (s) = G6 (s) = 2
s+1 s + 0.8s + 4
2
G7 (s) = 2
s +s+3

9
Chapter 2. Dynamical Systems

y y
A B
1 1

1 t 1 t

y y
C D
1 1

1 t 1 t

y
E
1

1 t

2.12 Pair each of the four pole-zero plots with the corresponding step re-
sponses A–G.

1 2
1 1
+

+
+

−3 −2 −1 1 −3 −2 −1 1
−1 −1
+

3 4
1 1
+

+
+

−3 −2 −1 1 −3 −2 −1 1
−1 −1
+

10
 Chapter 2. Dynamical Systems

y A y B

1 1

1 t 1 t

y C y D

1 1

1 t 1 t

y E y F

1 1

1 t 1 t

y G

1 t

2.13 Glycemic index (GI) is a measure of how fast carbohydrates in food

are processed by the body. To obtain the glycemic index, the system
in Figure 2.1 is studied, where G(s) is different for different types of
carbohydrates.

food intake glucose level

G(s)

Figure 2.1

a. Figure 2.2 shows the impulse response from food intake to glucose
level for two types of food: whole grain pasta with low GI (solid line)
and lemonade with high GI (dashed line). Which of the following
transfer functions may be used to model the uptake of whole grain
pasta and lemonade, respectively?

11
Chapter 2. Dynamical Systems

1 1
G1 (s) = s+1
G2 (s) = s/3+1
1 1
G3 (s) = (s+1)2
G4 (s) = (s/3+1)2
1 1
G5 (s) = s(s+1)
G6 (s) = s(s/3+1)

1.5
pasta
lemonade
glucose level

0.5

0
0 1 2 3 4 5 6
Time (h)

Figure 2.2

b. Why is it more relevant to look at the impulse response rather than

the step response for this application?

2.14 Determine the transfer function from U to Y for the systems below.

a.
U + G1 Y

G2

b.
H1

U G1 + G2 Y

H2

c.
G3 +

U + G1 G2 Y

12
 Chapter 2. Dynamical Systems

d.
U + G1 + G2 Y

− H2

− H1

2.15 The block diagram in Figure 2.3 describes temperature control in a

room. The measurement signal y is the temperature in the room. The
control signal u is the power of the radiators. The reference value r
is the desired temperature. A controller G R (s) controls the power of
the radiators based on the difference of the desired and measured
temperature, e. The temperature of the room is also affected by the
outdoor temperature, which may be seen as a disturbance, d.

D(s)

R(s) E (s) U (s) Y (s)

+ G R (s) + G P (s)

−1

Figure 2.3

a. Determine the transfer function from R(s) to Y (s).

b. Determine the transfer function from D(s) to Y (s).

c. Determine the transfer function from R(s) to E (s).

d. Determine the transfer function from D(s) to U (s).

2.16 Consider the transfer function

s2 + 6s + 7
G(s) =
s2 + 5s + 6

Write the system on

a. diagonal form,

b. controllable canonical form,

c. observable canonical form.

13
3. Frequency Analysis

3.1 Assume that the system

0.01(1 + 10s)
G(s) =
(1 + s)(1 + 0.1s)

is subject to the input u( t) = sin 3t, −∞ < t < ∞

a. Determine the output y( t).

b. The Bode plot of the system is shown in Figure 3.1. Determine the
output y( t) by using the Bode plot instead.

−1
10
Magnitude

−2
10

−3
10
−3 −2 −1 0 1 2 3
10 10 10 10 10 10 10

90

45
Phase

−45

−90
−3 −2 −1 0 1 2 3
10 10 10 10 10 10 10
Frequency [rad/s]

Figure 3.1 The Bode plot in assignment 3.1.

3.2 We analyze the two systems in Figure 3.2; the sea water in Öresund
and the water in a small garden pool. The input signal to the systems
is the air temperature and the output is the water temperature.

a. Figure 3.3 shows two Bode diagrams. Which diagram correpsonds to

which system?

b. We assume that the air temperature has sinusoidal variations with

a period time T = 1 year. The greatest temperature in the summer
is 19○ C and the lowest temperature in the winter is −5○ C. What is
the difference between the greatest and lowest sea water temperature
over the year? Use the Bode diagram.

14
 Chapter 3. Frequency Analysis

Figure 3.2

1
10

0
10
pG(iω )p

−1
10

−2
10

−3
10
0
arg(G(iω ))

−45

−90
−4 −3 −2 −1 0 1
10 10 10 10 10 10
ω (rad/h)

Figure 3.3 Bode diagram in problem 3.2.

c. During a summer day we assume that the air temperature has sinu-
soidal variations with a period time T = 1 day. The greatest temper-
ature of the day (at 13.00) is 27○ C, and the lowest temperature (at
01.00) is 14○ C. At what time during the day is the water in the garden
pool the warmest?

3.3 Assume that the oscillating mass in assignment 1.3 has m = 0.1 kg,

15
Chapter 3. Frequency Analysis

c = 0.05 Ns/cm and k = 0.1 N/cm. The transfer function is then given
by
10
G(s) = 2
s + 0.5s + 1
a. Let the mass be subject to the force f = sin ω t, −∞ < t < ∞.
Calculate the output for ω = 0.2, 1 and 30 rad/s.

b. Instead, use the Bode plot of the system in Figure 3.4 to determine
the output for ω = 0.2, 1 and 30 rad/s.

2
10
1
10
Magnitude

0
10
−1
10
−2
10
−3
10
−2 −1 0 1 2
10 10 10 10 10

−45
Phase

−90

−135

−180
−2 −1 0 1 2
10 10 10 10 10
Frequency [rad/s]

Figure 3.4 The Bode plot of the oscillating mass in assignment 3.3.

3.4 Draw the Bode plots corresponding to the following transfer functions

a. 3
G(s) =
1 + s/10

b. 10
G(s) =
(1 + 10s)(1 + s)

c. e−s
G(s) =
1+s

d. 1+s
G(s) =
s(1 + s/10)

e.
2(1 + 5s)
G(s) =
s(1 + 0.2s + 0.25s2 )

16
 Chapter 3. Frequency Analysis

3.5 Exploit the results from the previous assignment in order to draw the
Nyquist curves of

a.
3
G(s) =
1 + s/10

b. 10
G(s) =
(1 + 10s)(1 + s)

c.
e−s
G(s) =
1+s

3.6 The Bode plot below was obtained by means of frequency response
experiments, in order to analyze the dynamics of a stable system.
What is the transfer function of the system?
0
10

−1
10
Magnitude

−2
10

−3
10

−4
10
−2 −1 0 1 2 3 4
10 10 10 10 10 10 10

−30
Phase

−60

−90
−2 −1 0 1 2 3 4
10 10 10 10 10 10 10
Frequency [rad/s]

3.7 Measurements resulting in the Bode plot below have been conducted
in order to analyze the dynamics of an unknown system. Use the Bode
plot to determine the transfer function of the system.

17
Chapter 3. Frequency Analysis

1
10

0
Magnitude 10

−1
10

−2
10

−3
10
−1 0 1 2 3
10 10 10 10 10

90

0
Phase

−90

−180

−270
−1 0 1 2 3
10 10 10 10 10
Frequency [rad/s]

18
4. Feedback Systems

4.1 Assume that the air temperature y inside an oven is described by the
differential equation

ẏ( t) + 0.01y( t) = 0.01u( t)

where u is the temperature of the heating element.

a. Let u be the input and y the output and determine the transfer
function G P (s) of the oven.

b. The oven is to be controlled by a P controller, G R (s) = K , according

to the block diagram below. Write down the transfer function of the
closed loop system.

r e u y
Σ GR GP

−1

c. Choose K such that the closed loop system obtains the characteristic
polynomial
s + 0.1

4.2 The below figure shows a block diagram of a hydraulic servo system
in an automated lathe.
f
r e u y
Σ GR Σ GP

−1

The measurement signal y( t) represents the position of the tool head.

The reference tool position is r( t), and the shear force is denoted f ( t).
G R is the transfer function of the position sensor and signal amplifier,
while G P represents the dynamics of the tool mount and hydraulic
piston
1
G P (s) =
ms2 + ds
where m is the mass of the piston and tool mount, and d is the
viscous damping of the tool mount. In the assignment it is assumed
that r( t) = 0.

a. How large does the deviation e( t) = r( t) − y( t) between the reference-

and measured tool head position become in stationarity if the shear
force f ( t) is a unit step? The controller is assumed to have a constant
gain G R (s) = K .

19
Chapter 4. Feedback Systems

b. How is this error changed if the amplifier is replaced by a PI controller

with transfer function G R (s) = K 1 + K 2 /s?

4.3 A process is controlled by a P controller according to the figure below.

It is assumed that r = 0.
n
r u y
Σ GR GP Σ

−1

a. Measurements of the process output indicate a disturbance n. Calcu-

late the transfer functions from n to y and n to u.
b. Let G P (s) = s+1 1 and assume that the disturbance consists of a sinu-
soid n( t) = A sin ω t. What will u and y become, after the decay of
transients?
c. Assume that K = 1 and A = 1 in the previous sub-assignment.
Calculate the amplitude of oscillation in u and y for the cases ω = 0.1
and 10 rad/s, respectively.

4.4 The below figure shows a block diagram of a gyro stabilized platform.
It is controlled by an motor which exerts a momentum on the platform.
The angular position of the platform is sensed by a gyroscope, which
outputs a signal proportional to the platform’s deviation from the
reference value. The measurement signal is amplified by an amplifier
with transfer function G R .
M
θ ref θ
1
Σ G R (s) K Σ Js2

−1

It is desired that step changes in the reference θ ref or the disturbance

momentum M on the platform do not result in persisting angular
errors. Give the form of the transfer function G R , which guarantees
that the above criteria hold. Hint: Postulate G R (s) = Q(s)/ P (s)

4.5 When heating a thermal bath, one can assume that the temperature
increases linearly with 1○ C/s. The temperature is measured by means
of a thermocouple with transfer function
1
G(s) =
1 + sT
with time constant T = 10 s.
After some initial oscillations, a stationary state, in the sense that the
temperature measurement increases with constant rate, is reached.
At a time instant, the temperature measurement reads 102.6○ C. Cal-
culate the actual temperature of the bath.

20
 Chapter 4. Feedback Systems

4.6 Consider the system G0 (s) with the following asymptotic gain curve.
Assume that the system lacks delays and right half plane zeros.

log G0

incline: −2

G0 = 1
incline: −1

ω=1 ω=5
log ω

Further assume that the system is subject to negative feedback and

that the closed loop system is stable. Which of the following setpoints
can be tracked by the closed loop system, without a stationary error?
Assume r( t) = 0 for t < 0, and that the constants a, b and c ,= 0.

a. r( t) = a

b. r( t) = bt

c. r( t) = ct2

d. r( t) = a + bt

e. r( t) = sin( t)

4.7 In a simple control circuit, the process and controller are given by
1
G P (s) = and G R (s) = 6.5, respectively.
(s + 1)3
a. Determine the sensitivity function S (s).
The gain plot of the sensitivity function is given below.

b. How much are low-frequency load disturbances damped by the control

circuit (in closed loop, as compared to open loop)?

c. At which angular frequency does the control circuit exhibit the largest
sensitivity towards disturbances and by how much are disturbances
amplified at most?
1
10
Magnitude

0
10

−1
10

−1 0 1
10 10 10

21
Chapter 4. Feedback Systems

4.8 The below figure shows the gain curves of the sensitivity function S
and complementary sensitivity function T for a normal control circuit.

1
10
0
10
Magnitude

−1
10
−2
10
−3
10
−4
10
−2 −1 0 1
10 10 10 10
Frequency [rad/s]
1
10
Magnitude

0
10

−1
10

−2
10
−2 −1 0 1
10 10 10 10
Frequency [rad/s]

a. Determine which curve corresponds to the sensitivity function and

complementary sensitivity function, respectively.

b. Give the frequency range where disturbances are amplified by the

feedback loop, and the frequency range where they are damped by the
feedback loop. What is the maximum gain of disturbance amplifica-
tion?

c. Give the frequency ranges where the output exhibits good tracking of
the reference signal.

d. What is the minimal distance between the Nyquist curve of the open
loop system and the point −1 in the complex plane? What does this
say about the gain margin?

4.9 In a simple control loop, the open loop transfer function is given by

K
Go (s) = G R (s)G P (s) =
s(s + 2)

Draw the root locus of the characteristic equation of the closed loop
system, with respect to the gain parameter K .

4.10 A simple control loop has the open loop transfer function

K (s + 10)(s + 11)
Go (s) = G R (s)G P (s) =
s(s + 1)(s + 2)

a. Which values of K yield a stable closed loop system?

22
 Chapter 4. Feedback Systems

b. Sketch the characteristics of the root locus.

4.11 The figure below shows the block diagram of a printer.

a. Which values of the gain K yield an asymptotically stable system?

b. The goal is to track a reference which increases linearly with rate

0.1 V/s, and guarantee a stationary error of less than 5 mV. Can this
be achieved by adequate tuning of the gain K ?

r y
K 1
Σ
s+2 s(s + 1)

−1

4.12 Consider the Nyquist curves in Figure 4.1. Assume that the corre-
sponding systems are controlled by the P controller

u = K (r − y)

In all cases the open loop systems lack poles in the right half plane.
Which values of K yield a stable closed loop system?

4.13 The transfer function of a process is given by

1
Gp (s) =
(s + 1)3

The loop is closed through proportional feedback

u = K (r − y)

Use the Nyquist criterion to find the critical value of the gain K (i.e.
the value for which the system transits from stability to instability).

4.14 The Nyquist curve of a system is given in Figure 4.2. The system is
stable, i.e. lacks poles in the right half plane.
Assume that the system is subject to proportional feedback

u = K (r − y)

Which values of the gain K result in a stable closed loop system?

23
Chapter 4. Feedback Systems

Figure 4.1 Nyquist curves in assignment 4.12.

Figure 4.2 Nyquist curve of the system in assignment 4.14.

24
 Chapter 4. Feedback Systems

4.15 In order to obtain constant product quality in a cement oven, it is cru-

cial that the burn zone temperature is held constant. This is achieved
by measuring the burn zone temperature and controlling the fuel flow
with a proportional controller. A block diagram of the system is shown
below.

reference fuel burner zone

temperature flow temperature

Σ GR GP

-1

Find the maximal value of the controller gain K , such that the closed
loop system remains stable? The transfer function from fuel flow to
burn zone temperature is given by

e−9s
G P (s) =
(1 + 20s)2

4.16 In a distillation column, the transfer function from supplied energy to

liquid phase concentration of a volatile component is

e−sL
G P (s) =
1 + 10s

where time is measured in minutes. The process is controlled by a PI

controller with transfer function

1
3 4
G R (s) = 10 1 +
2s

What is the maximal permitted transportation delay L, yielding at

least a 10○ phase margin?

4.17 A process with transfer function G P (s) is subject to feedback according

to Figure 4.3. All poles of G P (s) lie in the left half plane and the
Nyquist curve of G P is shown in Figure 4.4. It is assumed that G P (iω )
does not cross the real axis at other points than shown in the figure.
Which of the below alternatives are true? Motivate!

a. The gain margin Am < 2 for K = 1.

b. The phase margin φ m < 45○ for K = 1.

c. The phase margin decreases with decreasing gain K .

d. For K = 2 the closed loop system becomes unstable.

25
Chapter 4. Feedback Systems

r y
Σ K G P (s)

−1

Figure 4.3 The closed loop system in assignment 4.17.

Figure 4.4 Nyquist curve of the process G P (s) in assignment 4.17.

4.18 The Bode plot of the open loop transfer function, Go = G R G P , is shown
in Figure 4.5. Assume that the system is subject to negative feedback.

a. How much can the the gain of the controller or process be increased
without making the closed loop system unstable?

b. How much additional negative phase shift can be introduced at the

cross-over frequency without making the closed loop system unstable?

4.19 A Bode plot of the open loop transfer function of the controlled lower
tank in the double tank process is shown in Figure 4.6. What is the
delay margin of the system?

26
 Chapter 4. Feedback Systems

1
10

Magnitude
0
10

−1
10
−1 0
10 10

−90

−135
Phase

−180

−225

−270
−1 0
10 10
Frequency [rad/s]

Figure 4.5 Bode plot of the open loop system in Figure 4.18.

2
10

1
10
Magnitude

0
10

−1
10

−2
10
−2 −1 0
10 10 10

−90

−135
Phase

−180

−225

−270
−2 −1 0
10 10 10
Frequency [rad/s]

Figure 4.6 Bode plot of the open loop transfer function of the controlled lower tank
in the double tank process in problem 4.19.

27
5. State Feedback and Kalman Filtering

5.1 A linear system is described by the matrices

   
 −1 1   
β 
A=
 
 B=
  C = 0 γ  D=0
0 −2 1

a. For which values of β is the system controllable?

b. For which values of γ is the system observable?

5.2 A linear system is described by the matrices

   
 −2 0 
  4   
A=
  B=
 
 C =  −1 1 
1 0 −2

Find the set of controllable states.

5.3 Consider the system

   
dx  −2 −1  1
=
 x+
   u
dt 1 0 2
 
y = 1 1 x

Is it observable? If not, find the set of unobservable states.

5.4 Consider the system

     
dx  −1 0  1 1
=
 x+
    u, x(0) = 
 

dt 0 −2 0 1

Which of the states ( 3 0.5 )T , ( 5 5 )T , ( 0 0 )T , ( 10 0.1 )T or

( 1 −0.5 )T can be reached in finite time?

5.5 Consider the following system:

   
dx  −2 3  1
=
 x+
   u
dt 1 −4 2
 
y = 3 7 x

Is it controllable?

5.6 A dynamic system is described by the state space model below

   
 −2 2  5
ẋ = 
 x+
   u
0 −3 0
 
y = 1 0 x

28
 Chapter 5. State Feedback and Kalman Filtering

a. Is the system controllable? Which states can be reached in finite time

from the initial state x(0) = ( 0 0 )T ?

b. Calculate the transfer function of the system.

c. Can the same input-output relation be described with fewer states?

Write down such a representation, if possible.

5.7 A linear dynamical system with transfer function G(s) is given. The
system is controllable. Which of the following statements are unques-
tionably true?

a. The poles of the closed loop system’s transfer function can be arbitrar-
ily placed by means of feedback from all states.

b. The zeros of the closed loop system’s transfer function can be arbi-
trarily placed by means of feedback from all states.

c. If the state variables are not available for measurements, they can
always be estimated by diffrentiating the system output.

d. If the state vector is estimated by a Kalman filter

x̂˙ = A x̂ + Bu + K ( y − C x̂)

one can obtain an arbitrarily fast convergence of the estimate x̂ to-

wards the actual state vector x, by choice of the matrix K .

5.8 Determine a control law u = l r r − Lx for the system

   
dx  −1 0  1
=
 x+
   u
dt 0 −2 2
 
y = 1 1 x

such that the poles of the closed loop system are placed in −4 and the
stationary gain is 1.

5.9 The position of a hard drive head is described by the state space model
   
dx  −0.5 0 
 3
=
 x+
  u
dt 1 0 0
 
y = 0 1 x

a. Determine a state feedback

u = − Lx + l r r

which places the poles of the closed loop system in s = −4 ± 4i and

results in static gain 1 from reference to output.

29
Chapter 5. State Feedback and Kalman Filtering

θ
steer rockets

direction of motion

Figure 5.1 The lunar lander in assignment 5.10.

b. Determine a Kalman filter

d x̂
= A x̂ + Bu + K ( y − C x̂)
dt
for the system. Briefly motivate necessary design choices.

5.10 Figure 5.1 shows the lunar lander LEM of the Apollo project. We will
study a possible system for controlling its horizontal movement above
the moon surface.
Assume that the lander floats some distance above the moon surface
by means of the rocket engine. If the angle of attack (the angle of
the craft in relation to the normal of the moon surface) is nonzero,
a horizontal force component appears, yielding an acceleration along
the moon surface.
Study the block diagram in Figure 5.2 showing the relation between
the control signal u of the rocket engine, the angle of attack, θ , and
the position z. The craft obeys Newton’s law of motion in both the θ
and z directions. The transfer function from the astronaut’s control
signal u to the position z is
k1 k2
G z (s) =
s4
and it is practically impossible to manually maneuver the craft. To
facilitate the astronaut’s maneuvering task, we alter the craft dy-
namics by introducing internal feedback loops. This means that the
astronaut’s control lever is not directly connected to the motors, but
works as a joystick that decides the desired velocity of the craft.
A controller should then convert the movement of the control lever to
a control signal to the steer rockets. We are in possession the following
measurement signals:

r P u θ¨ θ˙ θ z̈ ż z
K1 1/ s 1/ s K2 1/ s 1/ s

feedback x1 x2 x3

Figure 5.2 Block diagram of the lander dynamics along the z-axis.

30
 Chapter 5. State Feedback and Kalman Filtering

• The time derivative of the attack angle, θ˙, measured by a rate

gyro.
• The acceleration in the z direction, z̈, measured by accelerometers
mounted on a gyro-stabilized platform.
• The speed in the z direction, ż, measured by Doppler radar.

a. Introduce the states

x1 = θ˙
x2 = z̈
x3 = ż

and write the system on state-space form. Let the velocity in the z
direction be the output signal of the system.

b. Determine a feedback controller which utilizes the three measure-

ments, and results in a closed-loop system with three poles in s =
−0.5, and lets the control signal of the astronaut be the speed refer-
ence in the z direction. You don’t have to calculate the gain l r .

5.11 A conventional state feedback law does note guarantee integral action.
The following procedure is a way of introducing integral action. Let
the nominal system be

dx
= Ax + Bu
dt
y = Cx

Augment the state vector with an extra component

Zt Zt
xn+1 = e(s) ds = (r(s) − y(s)) ds

The obtained system is described by

     
d xe  A 0  B 0
=
  xe +   u + 
   r
 
dt −C 0 0 1

where  
 x 
xe = 
 

xn+1
A state feedback law for this system results in a control law of the
form
u = − Lx − l n+1 xn+1 = − L e x e
This controller, which steers y towards r, obviously has integral action.
Use this methodology in order to determine a state feedback controller
with integral action for the system
   
dx 0 1 0
=
 x+
   u
dt 0 0 1
 
y = 1 0 x

31
Chapter 5. State Feedback and Kalman Filtering

such that the closed loop system obtains the characteristic polynomial

(s + α) s2 + 2ζωs + ω 2 = 0
! "

5.12 Consider the system

   
dx  −2 1  1
=
 x+
   u
dt 1 −2 2
 
y = 0 1 x

One wishes to estimate the state variables by means of the model

d x̂
= A x̂ + Bu + K ( y − C x̂)
dt

Determine K such that the poles of the Kalman filter are placed in
s = −4.

5.13 Consider the dynamical system

   
dx  −4 −3  1
=
 x+
   u
dt 1 0 0
 
y = 1 3 x

One desires a closed loop system with all poles in −4.

a. Assign feedback gains to all states such that the closed loop system
obtains the desired feature.

b. Assume that only the output y is available for measurement. In order

to use state feedback, the state x must be first be estimated by means
of e.g. a Kalman filter, yielding the estimate x̂. Subsequently, the
control law u = − L x̂ can be applied.
Is it possible to determine a Kalman filter for which the estimation
error decreases according to the characteristic polynomial (s + 6)2 ?

c. Is it possible to determine a Kalman filter for which the estimation

error decreases according to the characteristic polynomial (s + 3)2 ?
Briefly comment the obtained results.

32
6. Design methods

6.1 A PID controller has the transfer function

1
3 4
G R (s) = K 1 + + Td s
Ti s

a. Determine the gain and phase shift of the controller at an arbitrary

frequency ω.

b. At which frequency does the controller have its minimal gain? What
is the gain and phase shift for this frequency?

6.2 The process

1
G(s) =
(s + 1)3
is controlled by a PID controller with K = 2, Ti = 2 and Td = 0.5.
In order to investigate the effect of changing the PID parameters, we
will change K , TI and Td by a certain factor, one at a time. We will
observe how this affects both the step response (from reference and
load disturbance) and the Bode plot of the controlled open loop system.
The reference is a unit step at t = 0 whereas the load disturbance is
a negative unit step.

a. We start by studying what happens when the parameters are quadru-

pled, one at a time. Figure 6.1 shows the nominal case ( K , Ti , Td ) =
(2, 2, 0.5) (solid curves) together with the cases (8, 2, 0.5), (2, 8, 0.5)
and (2, 2, 2). Pair the three Bode plots and the step responses of Fig-
ure 6.1 with the three cases.

b. We now study what happens when each parameter is decreased by

a factor 2. The nominal case ( K , Ti , Td ) = (2, 2, 0.5) (solid curves) is
shown in Figure 6.2 together with the cases (1, 2, 0.5), (2, 1, 0.5) and
(2, 2, 0.25). Pair the three Bode plots and the three step responses in
Figure 6.2 with these three cases.

6.3 The steer dynamics of a ship are approximately described by

dr
J + Dr = Cδ
dt

where r is the yaw rate [rad/s] and δ is the rudder angle [rad]. Fur-
ther, J [kgm2 ] is the momentum of inertia wrt the yaw axis of the
boat, D [Nms] is the damping constant and C [Nm/rad] is a constant
describing the rudder efficiency. Let the rudder angle δ be the control
signal. Give a PI controller for control of the yaw rate, such that the
closed loop system obtains the characteristic equation

s2 + 2ζωs + ω 2 = 0

33
Chapter 6. Design methods

2
10
1

Magnitude
10
0
10
−1
10
−2
10
−3
10
−2 −1 0 1 2
10 10 10 10 10

0
−45
Phase

−90
−135
−180
−225
−2 −1 0 1 2
10 10 10 10 10
Frequency [rad/s]
1.5

1
y

0.5

0
0 5 10 15 20 25 30 35 40
t [s]

Figure 6.1 Bode plot and step response for the case when the PID parameters in
sub-assignment 6.2b have been multiplied by four. The solid curves correspond to the
nominal case.

2
10
1
Magnitude

10
0
10
−1
10
−2
10
−3
10
−2 −1 0 1 2
10 10 10 10 10

0
−45
Phase

−90
−135
−180
−225
−2 −1 0 1 2
10 10 10 10 10
Frequency [rad/s]
1.5

1
y

0.5

0
0 5 10 15 20 25 30 35 40
t [s]

Figure 6.2 Bode plot and step response for the case when the PID parameters in
sub-assignment 6.2b have been divided by two. The solid curves correspond to the
nominal case.

34
 Chapter 6. Design methods

6.4 An electric motor can approximately be described by the differential

equation
d 2θ dθ
J 2 +D = ki I
dt dt
where J is the moment of inertia, D is a damping constant and k i
is the current constant of the motor. Further, θ denotes the turning
angle and I the current through the motor. Let θ be the measurement
signal and I the control signal. Determine the parameters of a PID
controller such that the closed loop system obtains the characteristic
equation
(s + a)(s2 + 2ζωs + ω 2 ) = 0
Discuss how the parameters of the controller depend on the desired
specifications on a, ζ and ω.

6.5 a. Draw the Bode plot of a PI controller (let K = 1 and Ti = 1).

b. Draw the Bode plot of a PD controller (let K = 1 and Td = 15).

6.6 A cement oven consists of a long, inclined, rotating cylinder. Sediment

is supplied into its upper end and clinkers emerge from its lower end.
The cylinder is heated from beneath by an oil burner. It is essential
that the combustion zone temperature is kept constant, in order to
obtain an even product quality. This is achieved by measuring the
combustion zone temperature and controlling the fuel flow with a PI
controller. A block diagram of the system is shown in Figure 6.3.

reference fuel burner zone

temperature flow temperature

Σ GR GP

-1

Figure 6.3 Block diagram of a cement oven with temperature controller.

The transfer function from fuel flow to combustion zone temperature

is given by
e−9s
G P (s) =
(1 + 20s)2
and the transfer function of the controller is

1
G R (s) = K (1 + )
sTi

Use Ziegler-Nichol’s frequency method to determine the parameters

of the controller.

35
Chapter 6. Design methods

y PI

6.7 Martin has heard that the optimal effect from training is obtained
when the pulse is 160 beats per minute (bpm). By feeding back the
signal from his heart rate monitor to a treadmill, he wants to control
the speed such that the pulse is exactly at the optimal value.

a. Suppose the dynamics in Martins body can approximately be described

by the linear system
1 1
ẋ = − x + u
30 15 (6.1)
y=x
where u is the speed of the treadmill, and x is the pulse in bpm.
Design a PI controller such that both closed loop system poles are in
−0.1.
b. The model in (6.1) is not available to Martin. Thus, he decides to tune
his PI controller using Ziegler-Nichols frequency method. The Bode
diagram for a more accurate model is shown in Figure 6.4. What
controller parameters does Martin obtain? Use the Bode diagram.
Bode Diagram
1
10
Magnitude (abs)

0
10

−1
10

−2
10 0
−45
−90
−135
Phase (deg)

−180
−225
−270
−315
−360
−405
−450
−2 −1 0
10 10 10
Frequency (rad/sec)

Figure 6.4 Bodediagram i uppgift 6.7

6.8 Use Ziegler-Nichol’s step response and frequency method, to determine

the parameters of a PID controller for a system with the step response

36
 Chapter 6. Design methods

and Nyquist curve given in Figure 6.5. Also, determine a PI and a

PID controller using the Lambda method with λ = T.

Step Response
1.2

0.8
Amplitude

0.6

0.4

0.2

0
0 2 4 6 8 10 12
Time [s]
Nyquist Curve

ω=1
0
Im

−0.5

−1

−1 −0.5 0 0.5 1
Re

Figure 6.5 Step response and Nyquist curve of the system in assignment 6.8.

6.9 Consider a system with the transfer function

1
G(s) = e−s
s+1
a. Draw the step response of the system and use Ziegler-Nichol’s step
response method to determine the parameters of a PID controller.
Write down the values of the obtained controller parameters K , Ti
and Td .
b. Use Ziegler-Nichol’s frequency method to determine the parameters
of a PID controller.

37
Chapter 6. Design methods

c. Use the Lambda method with λ = T to determine the parameters of

a PID controller.

6.10 A process is to be controlled by a PID controller obtained through

Ziegler-Nichol’s methods.

a. Use the step response method for the process with the solid step
response curve in Figure 6.6.
1.2

0.8

0.6

0.4

0.2

−0.2
0 5 10 15 20 25 30 35 40

Figure 6.6

b. The Nyquist curve of the same system is shown in Figure 6.7. The
point marked ’o’ corresponds to the frequency ω = 0.429 rad/s. Apply
the frequency method to the process.
0.4

0.2

0 ο

−0.2

−0.4

−0.6

−0.8

−1
−0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1

Figure 6.7

c. Unfortunately the step response method results in an unstable closed

loop system. The frequency method yields a stable but poorly damped
system. The reason why the step response method works so badly,

38
 Chapter 6. Design methods

is that it tries to approximate the process with a delayed first order

system (the dashed step response above). By exploiting the Nyquist
curve, one can obtain PID parameters yielding the solid curve step
response in Figure 6.8. The dashed and dotted curves were obtained
through the step response method.
How do you think K has changed in the third method, as compared
to the Ziegler-Nichol’s methods (increase or decrease)?

2.5

1.5

0.5

−0.5
0 10 20 30 40 50 60 70 80

Figure 6.8

6.11 A second order system has the Bode plot shown in Figure 6.9. We
would like to connect a link G K in series with the system, in order to
increase the speed of the closed loop system. The cross-over frequency,
ω c , (the angle for which pG0 p = 1) is used as a measure of the system’s
speed. Which of the following G K -candidates yield a faster system?

A GK = K , K > 1
1
B GK =
s+1
s+1
C GK =
s+2
D G K = e−sT , T > 0

6.12 A system has the transfer function

1
G P (s) =
s(s + 1)(s + 2)

The system is part of a feedback loop together with a proportional

controller with gain K = 1. The control error of the resulting closed
loop system exhibits the following behavior: e( t) → 0, t → ∞ when
the setpoint is a step and e( t) → 2, t → ∞ when the setpoint is a
ramp.

39
Chapter 6. Design methods

Figure 6.9 Bode plot of the system in assignment 6.11.

Design a compensation link G k (s) which together with the propor-

tional controller decreases the ramp error to a value less than 0.2.
Also, the phase margin must not decrease by more than 6○ .

6.13 Consider a system with the following transfer function

1.1
G P (s) =
s(s + 1)

A proportional controller with gain K = 1 is used to close the loop.

However, the closed loop system becomes too slow. Design a compen-
sation link, G k (s), that roughly doubles the speed of the closed loop
system without any change in robustness, i.e. the crossover frequency
ω c should be doubled and the phase margin φ m should not decrease.

6.14 Consider the system

1
G1 (s) =
s(s + 1)(s + 2)

If controlled by a proportional controller with gain K = 1, the sta-

tionary error of the closed loop system is e = 0 for a step input (r = 1,
t > 0) and e = 2 for a ramp input (r = t, t > 0). One wants to in-
crease the speed of the system by a factor 3, without compromising
its phase margin or the ability to eliminate stationary errors. Device
a compensation link G k (s) that fulfils the above criteria.

40
 Chapter 6. Design methods

6.15 A servo system has the open loop transfer function

2.0
Go (s) =
s(s + 0.5)(s + 3)

The system is subject to simple negative feedback and has a step

response according to Figure 6.10. As seen from the figure, the system
is poorly damped and has a significant overshoot. The speed however,
is satisfactory. The stationary error of the closed loop system with a
ramp input is e 1 = 0.75.

2 y(t)

1.5

0.5

0
0 10 20 30

Figure 6.10 Step response of the closed loop servo system in assignment 6.15.

Design a compensation link that increases the phase margin to φ m =

50○ without affecting the speed of the system. (φ m = 50○ yields a
relative damping ζ ( 0.5 which corresponds to an overshoot M (
17%.) The stationary ramp error of the compensated system must not
be greater than e 1 = 1.5.

6.16 Consider a system with the open loop transfer function

1.5
G1 (s) =
s(s2 + 2s + 2)

The system is subject to simple negative feedback. The settling time

(5%) is Ts = 8.0 s, the overshoot is Mo = 27%, and the stationary
ramp error (r( t) = t) is e 1 = 1.33.
Device a phase lag compensation link

s+a
G k (s) = K
s + a/ M

such that the stationary ramp error of the closed loop system is de-
creased to e 1 = 0.1, while speed and damping (robustness) are virtu-
ally sustained.

41
7. Controller Structures

7.1 Figure 7.1 shows a block diagram of the temperature control system
in a house. The reference temperature (the thermostat set point) is
given by r, the output y is the indoor temperature and the disturbance
d is due to the outdoor temperature. The transfer function G1 (s)
represents the dynamics of the heating system and G2 (s) represents
the dynamics of the air inside the house. The controller G R is a P
controller with gain K = 1.
Assume that the influence d of the outdoor temperature can be exactly
measured. Determine a feedforward link H, such that the indoor
temperature becomes independent of the outdoor temperature. What
is required in order to obtain a good result from the feedforward?

d
H

r u y
Σ GR Σ G1 Σ G2

−1

Figure 7.1 Block diagram of the temperature control system in a house.

7.2 Figure 7.2 shows a block diagram of a level control system for a tank.
The inflow x( t) of the tank is determined by the valve position and
the outflow v( t) is governed by a pump. The cross section of the tank
is A = 1 m2 . The assignment is to control the system so that the level
h in the tank is held approximately constant despite variations in the
flow v. This is done by adjusting the valve at the outflow from the
buffer tank.

framkoppling
v
G
F

−1

P-reg ventil tank

h h
ref
Σ K Σ G Σ G
V T

−1

Figure 7.2 Block diagram of the level control system in assignment 7.2.

42
 Chapter 7. Controller Structures

The transfer function of the valve from position to flow is

1
Gv (s) =
1 + 0.5s
The tank dynamics can be determined through a simple mass balance.

a. Assume that G F = 0, i.e. that we don’t have any feedforward. Design a

P controller such that the closed loop system obtains the characteristic
polynomial (s + ω )2 . How large does ω become? What stationary level
error is obtained after a 0.1 step in v( t)?

b. Design a PI controller which eliminates the stationary control error

otherwise caused by load disturbances. Determine the controller pa-
rameters so that the closed loop system obtains the characteristic
polynomial (s + ω )3 . How large does ω become?

c. To further decrease the influence of load disturbances, we introduce

feedforward based on measurements of v( t). Design a feedforward
controller G F that eliminates the influence of outflow variations by
making corrections to x( t).
Comment. As all variables describe deviations from the operation point, the
reference value for the level h can be set to zero.

7.3 Consider the system in Figure 7.3. The transfer function of the process
is given by
1
G P (s) =
s+3
and G R (s) is a PI controller with transfer function
1
G R (s) = K (1 + )
STi
K f is a constant feedforward from the reference signal r.

Kf

r u y
Σ G R (s) Σ G P (s)

−1

Figure 7.3 Block diagram showing assignment 7.3.

a. Let K f = 0 and determine K and Ti such that the poles of the

closed loop system are placed in −2 ± 2i, which is assessed to supress
disturbances well.

b. Discuss the influence of the feedforward on the system’s response to

reference changes.
The closed loop transfer function of the system has one zero. Eliminate
it by choosing an appropriate constant feedforward K f .

43
Chapter 7. Controller Structures

7.4 The system in assignment 7.3 can be described by an equivalent block

diagram, according to Figure 7.4. Write down the transfer functions

y y
r
H Σ G
ff

H
fb

Figure 7.4 Equivalent block diagram in assignment 7.3.

Hff (s) and Hfb (s). Discuss the result and consider the effect of the
feedforward when the controller contains a D term.

7.5 The block diagram in Figure 7.5 shows cascade control of a tank. The

v1 v2

e y
1
yr Σ GR2 Σ G R1 Σ G1 G2 y
Σ

−1

−1

Figure 7.5 The cascade in assignment 7.5.

transfer function G1 describes a valve whereas the transfer function

G2 describes the dynamics of the tank. The objective is to control the
tank level y. This is done by controlling the valve G1 in an inner
control loop, whereas y is controlled by an outer control loop. Both
the control loops are cascaded so that the reference of the inner loop
is the output of the controller in the outer loop.
There are two disturbances in the system, namely the disturbance flow
v2 , which is added to the controlled flow y1 and pressure variations v1
in the flow before the valve. Discuss the choice of controller (P or PI)
in the inner and outer loop, respectively, with respect to elimination
of stationary control errors at step changes in disturbances v1 and v2 .

2
7.6 Consider Figure 7.5 and assume that G1 (s) = s+2 describes a valve
whereas G2 (s) = 1s describes a tank.

a. Determine a P controller G R1 (s) = K 1 such that the inner control

loop becomes 5 times faster than the uncontrolled valve.

44
 Chapter 7. Controller Structures

b. Design a PI controllerG R2 (s) = K 2 (1 + T1i s ) for the outer loop, which

gives closed loop poles a factor 10 closer to the origin than the pole
for the inner control loop. Approximate the inner loop by Ginner (s) (
Ginner (0).

7.7 In a certain type of steam boiler, a dome is used to separate the steam
from the water (see Figure 7.6). It is essential to keep the dome level
constant after load changes. The dome can be described by the model

F(s)
−3 s − 0.01
10
s(s + 0.1)
Y(s)
Σ
M(s) −3
10
s

Figure 7.6 Block diagram of steam boiler with dome.

10−3 s − 0.01
Y (s) = M (s) + 10−3 F (s)
s s(s + 0.1)

where Y is the dome level [m], M is the feed water flow [kg/s] and F
is the steam flow [kg/s].

a. Assume a constant steam flow. Design a P controller, controlling the

feed water flow by measuring the dome level. Choose the controller
parameters such that the control error caused by a step in the dome
level goes down to 10 % of its initial value after 10 seconds.

b. Consider the closed loop system. Write down the stationary level error
Y caused by a step disturbance of 1 kg/s in the steam flow F.

c. Consider the initial system. Determine a feedforward link H (s) from

steam flow F (s) to feed water flow M (s), such that the level Y becomes
independent of changes in the steam flow.

7.8 Assume that a servo motor

1
G P (s) =
s(s + 1)

is controlled by the P controller G R (s) = 2. What is the delay margin

of the system?

7.9 Consider the same process and controller as in the previous assign-
ment. Now the process is controlled over a very slow network which
introduces a one second delay in the control loop. In order to deal with
this problem an Smith predictor is utilized, see Figure 7.7.

45
Chapter 7. Controller Structures

a. Assume that the model and the process are identical. What are the
transfer functions for the blocks (Controller, Process, Model, Model
with no delay) in our example?
b. The block diagram of the Smith predictor can be redrawn according to
Figure 7.8. What is the transfer function of the Smith predictor (from
e to u) in our example?
c. Use the approximation e x ( 1 + x in order to simplify the transfer
function of the controller. Compare the controller to compensation
links.
r u y
Σ controller process
−
y1 −
model Σ
y2
model with
no delay

Figure 7.7 Working principle of the Smith predictor.

Smith predictor

r e u y
Σ Σ GR GP

Ĝ P − Ĝ0P

−1

Figure 7.8 Block diagram equivalent to Figure 7.7.

7.10 Figure 7.9 shows the result of a frequency analysis carried out on the
beam (a part of the ’ball on the beam’ process). One sees that the
process dynamics can be well approximated by an integrator, for low
frequencies. One also sees that for high frequencies, the phase curve
diverges in a way which resembles a delay. Consequently, it would be
possible to describe the process by
k −sL
G(s) = e
s
Use the Bode plot in order to determine approximate values of the
gain k and delay L.

46
 Chapter 7. Controller Structures

1
10

0
10
Magnitude

−1
10

−2
10

−3
10
0 1 2 3
10 10 10 10

−100
Phase

−200

−300

−400
0 1 2 3
10 10 10 10
Frequency [rad/s]

Figure 7.9 Measured Bode plot of the beam.

47
8. Design Examples

8.1 Depth Control of Submarine

Purpose
This assignment deals with depth control of a submarine from the
forties. Two control methods are tested – PD and state feedback. The
latter method was used in reality.

Background
Depth control of submarines can be achieved by means of varying the
rudder angle β according to Figure 8.1. The depth h is measured by

h
β

Figure 8.1 Depth control of the submarine in assignment 8.1.

means of a manometer. By manually generating a sinusoidal rudder

angle β (by means of a table and watch — don’t forget that this was
the end of the forties) one can use frequency analysis to estimate
the transfer function G(s) from β to h (for a constant speed v). The
resulting Bode plots for three different speeds are shown in Figure 8.2.

Specifications
In this case no specifications were given except "Make it as good as
possible".

Problem Formulation
Assume that the speed is v = 3 knots. The problem lies in computing
a control law which gives a satisfactory settling of the depth h for the
given speed. This does not guarantee equally satisfactory results at
other speeds.
In an initial approach one wanted to control the depth h of the sub-
marine, solely based on measurements of h.

48
 Chapter 8. Design Examples

10 3

10 2

Magnitude
10 1

10 0

10 −1

10 −2
10 −3 10 −2 10 −1 10 0

−50

−100
Phase [deg]

−150

−200

−250

−300
10 −3 10 −2 10 −1 10 0
Frequency [rad/s]

Figure 8.2 Bode plot of the estimated transfer function G(s) from β [deg] to h [m]
in assignment 8.1 for the speeds v = 3 (solid curves), 5 (dashed curves) and 7 knots
(dotted curves).

a. What is the maximal allowed gain K in order to achieve a stable

closed loop system with a P controller β = K ( h ref − h). Use the Bode
plot in Figure 8.2?
b. One wants to obtain a cross-over frequency ω c = 0.03 rad/s, using a
PD controller Gr (s) = K (1 + TD s). How shall K and TD be chosen in
order to obtain a 60○ phase margin φ m ?
c. How is the stability of the closed loop system in (b) affected if the
speed is increased from 3 to 7 knots? Suggest different ways in which
speed variations can be taken into consideration.

For angular frequencies above 0.05 rad/s one can use the approxima-
tion 
k
 Gαβ (s) = v

s2 (8.1)

 G hα (s) = v
s
where Gαβ (s) and G hα (s) are the transfer functions from β to α and
from α to h, respectively (see Figure 8.3). The constant k v depends
on the speed v.
d. Determine k v by means of the Bode plot in Figure 8.2. (1 knot ( 1.852
km/h = 1.852/3.6 ( 0.514 m/s.)

49
Chapter 8. Design Examples

β kv α h
v
2 s
s

Figure 8.3 Block diagram of a submarine model which is valid for ω > 0.05 rad/s.

e. Assume that the approximate model

kv v
G hβ (s) =
s3

is under P control β = K ( h re f − h). Determine which values of K

that yield an asymptotically stable system. Does this concur with the
results obtained in sub-assignment a?

One can improve the performance of the control system by utilizing

additional feedback form the trim angle α and its derivative dα/dt.
f. Introduce the states x1 = dα/dt, x2 = α and x3 = h together with the
input u = β . Use the control law u = ur − l 1 x1 − l 2 x2 − l 3 x3 = ur − Lx
and determine L such that the characteristic equation of the closed
loop system becomes

(s + γω 0 )(s2 + 2ζω 0 s + ω 20 ) = 0

g. The reference h ref for the depth h is introduced according to

ur = L r h ref

How shall L r be chosen in order to obtain h = h ref in stationarity?

One decided to choose ζ = 0.5 and γ = 0.2 which was considered to

give an adequately damped step response. However, the choice of ω 0
requires some further thought. It should not be chosen too low, since
the approximate model (8.1) is only valid for ω > 0.05 rad/s. On the
other hand, choosing ω 0 too high would result in large rudder angles
caused by the large values of the coefficients l j , j = 1, 2, 3.
h. How large can ω 0 be chosen if a step disturbance in the manometer
signal corresponding to ∆h = 0.1m should not give rise to larger
rudder angles than 5○ ?

In the actual case ω 0 = 0.1 rad/s was chosen. A semi-automatic system

was evaluated first. The signal u − ur − l 1 x1 − l 2 x2 − l 3 x3 was displayed
to an operator, who manually tried to keep the signal zero by means
of the ordinary rudder servo. The control action was very satisfactory.
Settling times of 30-60 s were obtained throughout the speed range.
The complete automatic system was then evaluated on the Swedish
submarine ’Sjöborren’ (The Sea Urchin). The accuracy during march
in calm weather was ±0.05 m.

8.2

50
 Chapter 8. Design Examples

Control of Elastic Servo

Purpose
The aim of the assignment is to control the angular speed of a flywheel
which is connected to another flywheel by a weak axis. The second
flywheel is driven by a motor. Different control strategies are evaluated
and compared with respect to performance.

Background
Figure 8.4 shows a simplified model of an elastic servo. It could also
constitute a model of a weak robot arm or an elastic antenna system
mounted on a satellite. The turn angles of the flywheels are denoted

Figure 8.4 Model of the elastic servo in assignment 8.2.

φ 1 and φ 2 , respectively, whereas ω 1 = φ̇ 1 and ω 2 = φ̇ 2 denote the

corresponding angular speeds. The flywheels have moments of inertia
J1 and J2 , respectively. They are connected by an axle with spring
constant k f and damping constant d f . The system is subject to bearing
friction, which is represented by the damping constants d 1 and d 2 .
One of the flywheels is driven by a DC motor, which is itself driven by
a current-feedback amplifier. The motor and amplifier dynamics are
neglected. The momentum of the motor is proportional to the input
voltage u of the amplifier, according to

M = km · I = km ki u

where I is the current through the rotor coils. Momentum equilibrium

about the flywheel yields the following equations
I
J1ω̇ 1 = − k f (φ 1 − φ 2 ) − d 1ω 1 − d f (ω 1 − ω 2 ) + k m k i u
J2ω̇ 2 = + k f (φ 1 − φ 2 ) − d 2ω 2 + d f (ω 1 − ω 2 )

We introduce the state variables


x = ω1
 1


x2 = ω 2



x3 = φ 1 − φ 2

and consider the angular speed ω 2 as the output, i.e.

y = kω 2 · ω 2

51
Chapter 8. Design Examples

This gives us the following state space model of the servo.

 d1+d f df k   km ki 

 − J1 J1 − J1f 
  J1 

   
ẋ = Ax + Bu =  d d +d kf  x +

0 

u
− f J2 2
 f 
 
 
   
 J2
 J2    
1 −1 0 0
 
y = C x =  0 kω 2 0  x

The following values of constants and coefficients have been measured

and estimated for a real lab process.

J1 = 22 · 10−6 kgm2
J2 = 65 · 10−6 kgm2
k f = 11.7 · 10−3 Nm/rad
d f = 2e − 5
d 1 = 1 · 10−5 Nm/rad/s
d 2 = 1 · 10−5 Nm/rad/s
k m = 0.1 Nm/A
k i = 0.027 A/V
kω 1 = kω 2 = 0.0167 V/rad/s

Problem Formulation
The input is the voltage u over the motor and we want to control the
angular speed ω 2 of the outer flywheel.
It is desired to quickly be able to change ω c , while limiting the con-
trol system’s sensitivity against load disturbances and measurement
noise. The system also requires active damping, in order to avoid an
excessively oscillative settling phase.

Specifications
1. The step response of the closed loop system should be fairly well
damped and have a rise time of 0.1-0.3 s. The settling time to
±2% shall be at most 0.5 s. A graphical specification of the step
response is given in Figure 8.5.
2. Load disturbances must not give rise to any static errors.
3. Noise sensitivity should not be excessive.

Ziegler-Nichols Method
The Bode plot of the transfer function from u to ω 2 is shown in
Figure 8.6.
a. Use Ziegler-Nichols frequency method in order to determine suitable
PID parameters.
Ziegler-Nichols method often gives a rather oscillative closed loop
system. However, the obtained parameters are often a reasonable
starting point for manual tuning.

52
 Chapter 8. Design Examples

1.4

1.2

0.8

0.6

0.4

0.2

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Figure 8.5 The step response of the closed loop system shall lie between the dashed
lines.

10 0

10 −1

10 −2

10 −3

10 −4

10 −5
10 0 10 1 10 2 10 3

−50

−100

−150

−200

−250

−300
10 0 10 1 10 2 10 3
Frequency [rad/s]

Figure 8.6 Bode plot of the servo process.

State Feedback and Kalman Filtering

If it is possible to measure all states, the poles of the closed loop
system can be arbitrarily placed through the feedback control law

u( t) = − Lx( t) + L r yr ( t)

if the system is also controllable.

53
Chapter 8. Design Examples

b. Determine the gain L r so that the stationary gain of the closed loop
system becomes 1, i.e. y = yr in stationarity.

In order to meet specification 2, one must introduce integral action in

the controller. One way to achieve this it thorough the control law
Zt
u( t) = − Lx( t) + L r yr ( t) − L i ( y(s) − yr (s))ds
−∞

This can be interpreted as feedback from an ’extra’ state xi according

to I
ẋi = y − yr
u = − Lx + L r yr − L i xi
Figure 8.7 shows a block diagram of the entire system

Lr

y −x u y
r 1
Σ
i
L Σ process
s i

x
−L

−1

Figure 8.7 Block diagram of the state feedback control system in assignment 8.2.

c. How does the augmented state space model look like? Introduce the
notion x e for the augmented state vector.
Since the states are not directly measurable, they must be recon-
structed in some way. A usual way is to introduce a Kalman filter

x̂˙ = A x̂ + Bu + K ( y − C x̂)

and then close the loop from the estimated states x̂

u = − L x̂ + L r yr − L i xi

It is, however, unnecessary to estimate xi since we have direct access

to this state. The block diagram of the entire system is shown in
Figure 8.8. Let L′ denote the augmented row matrix ( L L i ) and call
the augmented system matrices A′ and B′ , respectively. The problem
consists in finding suitable L, L i and K by placing the eigenvalues of
A′ − B′ L′ and A − KC. Since the both eigenvalue problems are of a
bit too high dimension for enjoyable hand calculations, we use Matlab
to investigate a few choices of pole placements.

54
 Chapter 8. Design Examples

Lr

y −x u y
r 1
Σ
i
L Σ process
s i

^
x
Kalman
−L
filter

−1

Figure 8.8 Block diagram showing the Kalman filter and state feedback in assign-
ment 8.2.

In order not to end up with too many free parameters, we place the
poles in a Butterworth pattern. I.e. the poles are equally distributed
on a half circle in the left half plane. We place the eigenvalues of
A′ − B′ L′ on a half circle with radius ω m , whereas the eigenvalues of
A − KC are placed on a half circle with radius ω o (see Figure 8.9). A

15

ωo
10

5 ωm

−5

−10

−15
−25 −20 −15 −10 −5 0 5

Figure 8.9 The pole placement in assignment 8.2.

suitable ω m can be obtained from Specification 1, i.e. that the settling

time Ts to reach within 2% of the stationary value must be less than
0.5 s. A coarse estimation of Ts for a second order system with relative
damping ζ and natural frequency ω is given by

ln ε
Ts ( −
ζω

where ε is the maximal deviation from the final value. Since we have
a 4th degree system, we cannot use this approximation directly.

55
Chapter 8. Design Examples

If we, however, only consider the least damped pole pair (ζ = 0.38 and
ω = ω m ) in Figure 8.9 we obtain

ln ε
ωm ( − (8.2)
Tsζ

d. Which value of ω m is obtained from the formula (8.2)?

We let ω m = 20 which yields Ts < 0.5 s. We can let L r = 0 since we
have integral action in the controller and thus stationary closed loop
gain 1. Figure 8.10 shows the step response of the closed loop system
for L r = 0 and L r chosen according to sub-assignment b, respectively.
By letting L r = 0 the step response overshoot is sufficiently decreased
to fulfill the specification.
We now fix ω m and vary ω o . The following test shall be used to
evaluate the control performance. At time t = 0 there is a unit step
in the reference value yr followed by a load disturbance d = −1 in the
control signal at t = 1 and the introduction of measurement noise (in
y) at t = 3. The variance of the noise is 0.01. The result is shown in
Figure 8.11.
e. Which value of ω o seems to be best when it comes to elimination of
load disturbances? Which ω o is best when it comes to suppressing
measurement noise?

1.6

1.4

1.2

0.8

0.6

0.4

0.2

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Figure 8.10 Step response of the closed loop system for ω m = 20 rad/s. Choosing
L r according to sub-assignment b yields the system with the larger overshoot. The
other curve is the step response corresponding to L r = 0.

56
 Chapter 8. Design Examples

1.5

1
r and y

0.5

0
Time [s]
0 0.5 1 1.5 2 2.5 3 3.5 4

30

20

10
u

−10

−20
Time [s]
0 0.5 1 1.5 2 2.5 3 3.5 4

Figure 8.11 Evaluation of control with ω m = 20 and ω o = 10 (solid curves), 20

(dashed curves) and 40 (dotted curves).

57
9. Interactive Comparison Between Model
Descriptions

There are many ways to describe the dynamics of processes and control
systems, e.g., step responses, transfer functions, state-space descriptions,
pole-zero diagrams, Bode diagrams, and Nyquist diagrams. A good way of
learning the correspondence between different descriptions is to use interac-
tive tools. On the web page
http://aer.ual.es/ilm/
there are several interactive programs that may be downloaded for free.
The module Modeling is convenient to use to study model descriptions. The
interface of this module is shown in the figure below.
The model structure that you want to study is entered by "dragging" poles
and zeros in the pole-zero diagram. Parameters and dead time may then be
changed by dragging points or lines in the different diagrams, or by entering
numerical values for the transfer function. The best way of learning to use
the tool and examine its possibilities is to try things out by experimenting
with the different menus. More information is available on the web page.
The gradation of the axes in the different diagrams can be changed by
clicking at the small triangles by the gradations. It is also possible to zoom
in and out.
Under the tab File you find the useful command “Reset data”, which
resets all values to default.

58
 Chapter 9. Interactive Comparison Between Model Descriptions

9.1 Study the transfer function

K
G(s) =
1 + sT
Let the starting poing be K = T = 1.

a. First, vary the gain K and note how the pole, step response, Nyquist
diagram and Bode diagram are affected. How can K be determined
from the step response, Nyquist diagram, and Bode diagram?

b. Set K = 1 and vary T. How are the different representations affected?

Why is the shape of the Nyquist diagram not affected?

c. Set K = T = 1 and add a dead time L such that the transfer function
becomes
K
G(s) = e− Ls
1 + sT
Vary L and obseve how that affects the representations. Explain what
happens with the step response. Why does the Nyquist curve look like
it does? Why is the gain curve of the Bode diagram not affected?

9.2 Study the transfer function

K
G(s) =
(1 + sT1 )(1 + sT2 )

Let the starting point be K = T1 = T2 = 1.

a. First, vary the gain K and note how poles, step response, Nyquist
diagram, and Bode diagram are affected.

b. Set K = 1 and vary T1 and T2 . Note the difference between the cases
T1 ( T2 and T1 ≫ T2 . How can G(s) be approximated in the case
T1 ≫ T2 ?

c. Set K = T1 = T2 = 1 and add a zero, such that the transfer function

becomes
K (1 + sT3 )
G(s) =
(1 + sT1 )(1 + sT2 )
Vary T3 and observe how that affects the representations. What hap-
pens when T3 < 0? What does this mean if you want to control such
a process? Try to explain the phenomenon.

9.3 Study the transfer function

ω2
G(s) =
s2 + 2ζωs + ω 2

Let the starting point be ζ = 0.7 och ω = 1.

a. First, vary the frequency ω and note how poles, step response, Nyquist
diagram, and Bode diagram are affected.

b. Set ω = 1 and vary ζ . Note how the representations change.

59
Chapter 9. Interactive Comparison Between Model Descriptions

60
Solutions to Chapter 1. Model Building and
Linearization

1.1 a. A block diagram of the system is shown in Figure 1.1. The person that
takes a shower senses the temperature and the flow (measurement
signals), and adjusts the shower handle (control signal) to get desired
temperature and flow. Feedback is primarily used. Disturbances may
be variations in water pressure and temperature in the water pipes.

Temp Hot
Person Shower
Cold
Flow

Figure 1.1 Block diagram of person taking a shower.

b. A car driver uses several control signals: the gas pedal, the brake,
the steering wheel. The driver wants to control the car such that it
keeps on the road with desired velocity, and keeps safe distance to
other vehicles. Measurement signals are the speedometer, and visual
feedback of how the car turns, the distance to the car in front, and
other road conditions. The uses feedforward, e.g., to adjust the veloc-
ity in advance before a curve, but also feedback, by looking at the
speedometer to keep desired velocity. A block diagram of the system
is shown in Figure 1.2.

Road Conditions Steering

Road Curvature Accelerator
Driver Car
Distance Brakes
Velocity

Figure 1.2 Block diagram of car driving.

c. The control signal is the heat from the stove plate. Measurement
signals from the systems are obtained by observing how intensively
the water boils, and sensing how soft the potatoes are. Feedback is
used, e.g., when adjusting the power of the stove plate when the
water boils too intensively. Feedforward is used when following a
given recipe, e.g., ”the potatoes are done after 20 minutes” or ”when
the water boils, decrease the heat of the stove plate to half of full
power”. A block diagram is shown in Figure 1.3.

1.2 a.
v̇ = ku
The system is linear.

61
Solutions to chapter 1. Model Building and Linearization

Heat
Stove Heat
Stove Pot & Water Potatoes
handle

Audio / Visual
Human Sensing (softness)

Figure 1.3 Block diagram of potato boiling.

b. Now, we get an additional differential equation

ṗ = v

Choosing the states as x1 = v, x2 = p, we get

ẋ1 = ku
ẋ2 = x1
y = x2

One can also write the system on matrix form

   
 0 0  k
ẋ = 
 x+
   u
1 0 0
 
y = 0 1 x

c. The term mx2 makes the system nonlinear. The stationary point is
given by
−0.001( x0 )2 + u0 = 0
which for u0 = 0.1 gives x0 = ±10. The stationary velocity thus
becomes y0 = 10 m/s.
d. The system is given by

ẋ = f ( x, u) = −mx2 + ku
y = ˆ( x, u) = x

We differentiate f and ˆ with respect to x and u and get

€f
= −2mx
€x
€f
=k
€u
€ˆ
=1
€x
€ˆ
=0
€u
We insert the stationary point ( x0 , u0 , y0 ) = (10, 0.1, 10) in the expres-
sions for the derivatives, and introduce the new variables ∆ x = x − x0 ,
∆ y = y − y0 , ∆u = u − u0 . We get the linear system
d∆ x
= −0.02∆ x + ∆u
dt
∆y = ∆x

62
 Solutions to chapter 1. Model Building and Linearization

1.3 a. With x1 = y and x2 = ẏ the system is given by

   0   
1   x1   0 
 ẋ1       
=  1 
+

   k c 
 
   f
ẋ2  − −  x2 
m m m
  x 
 1
y = 1 0
   
x2

b. which has the solution

1
3 4
σ −σ t
y( t) = 1− e sin ω d t − e−σ t cos ω d t
k ωd
p
where σ = c/2m and ω d = k/m − c2 /4m2 .

1
k

1.4 With states x1 = vout and x2 = v̇out , the system is given by

      
 ẋ1   0 1   x1   0 
 = 1
     +  1 
     vin
ẋ2 − LC − R
L 
x2 LC
 
 x
 1
vout =  1 0  
  
x2

1.5 a. We can choose e.g. the height h as state variable. The volume change
in the tank is given by

A ḣ = q in − q out
√
and from Torricelli’s law we obtain q out = a 2 ˆ h. The sought differ-
ential equation becomes

ap 1
ḣ + 2 ˆ h = q in
A A

b.

ap 1
ḣ = − 2 ˆ h + q in ( = f ( h, q in ))
A A
y=h ( = ˆ( h, q in ))

63
Solutions to chapter 1. Model Building and Linearization

c. The outlflow must equal the inflow q0out = q0in . The level is calculated
by letting ḣ = 0, which yields
42
1 q0in
3
0
h =
2ˆ a

We determine the partial derivatives

r
€f a ˆ €f 1
=− =
€h A 2h € q in A
€ˆ €ˆ
=1 =0
€h € q in

By inserting h = h0 above and introducing variables which denote

deviations from the operating point: ∆h = h − h0 , ∆q in = q in − q0in ,
∆ y = ∆h the linearized system is
r
˙ =−a ˆ 1
∆h ∆h + ∆q in
A 2h0 A
∆ y = ∆h

1.6
      
 ẋ1   0 1 0   x1   0 

 
   
  
   


 ẋ2 

 =
 0 0 1 



 x2 
 0
+ u
   
      
ẋ3 −1 −2 −3 x3 1
 
  x1 

 

y = 1 0 0
 x2 

  
x3

1.7 a.

ẋ1 = x2
√
ẋ2 = − x1 − x1 x2 + u2
y = x1

b. A stationary point implies ẋ1 = ẋ2 = 0. From the first equation

we directly obtain x2 = 0. Subsequently, the second equation yields
√
x1 = u2 . Hence there are infinitely many stationary points and they
can be parametrized through t as ( x01 , x02 , u0 ) = ( t4 , 0, t).

c. u0 = 1 gives the stationary point ( x01 , x02 , u0 ) = (1, 0, 1). We let

f1 ( x1 , x2 , u) = x2
√
f2 ( x1 , x2 , u) = − x1 − x1 x2 + u2
ˆ( x1 , x2 , u) = x1

64
 Solutions to chapter 1. Model Building and Linearization

and compute the partial derivatives

€ f1 € f1 € f1
=0 =1 =0
€ x1 € x2 €u
€ f2 1 € f2 € f2
= − √ − x2 = − x1 = 2u
€ x1 2 x1 € x2 €u
€ˆ €ˆ €ˆ
=1 =0 =0
€ x1 € x2 €u

At the stationary point we have

€ f1 € f1 € f1
=0 =1 =0
€ x1 € x2 €u
€ f2 1 € f2 € f2
=− = −1 =2
€ x1 2 € x2 €u
€ˆ €ˆ €ˆ
=1 =0 =0
€ x1 € x2 €u

After a variable substitution, the linearized system can be written

∆˙x1   0
      
 1   ∆ x1   0 
 =
 
 +
    ∆u
∆˙x2
  1 
− 2 −1 ∆ x2 2
   ∆x 
 1
∆y =  1 0  
 

∆ x2

1.8 At the sought operating point it holds that

0 = x21 x2 + 1
0 = x1 x22 + 1
x2 π 2
y = arctan +
x1 8
2
which yields x01 = −1, x02 = −1 and y0 = π
4
+ π8 . Computation of the
partial derivatives now yields

€ f1 € f1 € f1 √
= 2x1 x2 = x21 = 2 cos u
€ x1 € x2 €u
€ f2 € f2 € f2 √
= x22 = 2x1 x2 = − 2 sin u
€ x1 € x2 €u
€ˆ − x2 €ˆ x1 €ˆ
= 2 = 2 = 4u
€ x1 x1 + x22 € x2 x1 + x22 €u

With the variable substitution

π
∆u = u −
4
∆ x1 = x1 + 1
∆ x2 = x2 + 1
π π2
∆y = y − − .
4 8

65
Solutions to chapter 1. Model Building and Linearization

the linearized system becomes

∆˙x1   2 1   ∆ x1   1 
      

 =
    +
   ∆u

∆˙x2
 
1 2 ∆ x2 −1
   ∆x 
1 1  1
∆y = 
2
−2    + π ∆u.

∆ x2

1.9 a. The nonlinear state space equations are

ẋ1 = x2 = f1 ( x1 , x2 , u)
β
ẋ2 = ω 2 x1 − +u = f2 ( x1 , x2 , u)
x21
y = x1 = ˆ( x1 , x2 , u)

b. At stationarity it holds that

β
r̈( t) = ω 2 r0 − +0 =0
r20

i.e. r30 = β /ω 2 . We now compute the partial derivatives

 € f1 € f1 € f1     

 € x1 € x2 €u 
  0 1 0  0 1 0

 
  
  
€ f2 € f2 € f2
ω + 2β /r30
2  3ω 2
  


 € x1 € x2 €u


 =

 0 1
=
  0 1



 
    
 €ˆ €ˆ €ˆ  1 0 0 1 0 0
€ x1 € x2 €u

The linear system hence becomes

   
d∆ x  0 1 0
=
 2  ∆x + 
  ∆u
 
dt 3ω 0 1
 
∆y =  1 0  ∆x

66
Solutions to Chapter 2. Dynamical Systems

2.1 a. The transfer function for a linear system on state-space form is given
by

G(s) = C (s I − A)−1 B + D
  3 s 0   10 1 4−1  0 
     
= 1 0  −
  
  
 
0 s −1 −1 1
1
=
(s − 10)(s + 1) + 1

The transfer function gives the relation

1
Y (s) = U (s)
(s − 10)(s + 1) + 1

which may be written as

s2 Y (s) − 9sY (s) − 9Y (s) = U (s)

Inverse Laplace transform gives the differential equation

ÿ − 9 ẏ − 9y = u

b. Laplace transform of the differential equation gives

Js2 Y (s) + DsY (s) = U (s)

We get
1
Y (s) = U (s)
Js2 + Ds
and thus the transfer function is

1
G(s) =
Js2 + Ds

When writing the system on state-space form, the states may be

chosen as x1 = y, x2 = ẏ. This gives the equations

ẋ1 = ẏ = x2
1 1
ẋ2 = ÿ = (− D ẏ + u) = (− Dx2 + u)
J J

The system on state-space form is given by

   
 0 1  0
ẋ = 
 
D 
x+
1 u
0 −J J
 
y = 1 0 x

67
Solutions to chapter 2. Dynamical Systems

c. The transfer function is given by

1
G(s) = C (s I − A)−1 B + D = 1(s − −1/ k)−1 1/ k =
1 + sk

d. The transfer function gives the following relation between Y (s) and
U (s)
(s3 + αs2 + β s) Y (s) = γU (s)
Inverse Laplace transform gives
...
y + α ÿ + β ẏ = γ u

Choosing the states

x1 = y
x2 = ẏ
x3 = ÿ

gives the state-space description

   
 0 1 0   0

 
 
 

ẋ = 

 0 0 1 

 x+

 0u
    
0 −β −α γ
 
y = 1 0 0 x

We may also use any of the "standard forms" for state-space descrip-
tions given in the collection of formulae (diagonal form, observable
canonical form, controllable canonical form). To write the system on
diagonal form we must factorize the transfer function, which is hard
when we don’t know the values of α and β . Thus, the controllable or
observable canonical forms are more convenient to choose. We com-
pare the coefficients in our transfer function to the structure in the
collection of formulae and get

a 1 = α, a2 = β , a 3 = 0, b1 = 0, b2 = 0, b3 = γ

A state-space description on observable canonical form is thus given

by
   
 −α 1 0   0

 
 
 

ẋ = 

 −β 0 1 

 x+

 0u
    
0 0 0 γ
 
y = 1 0 0 x

The model is presented and analyzed in the paper Complex dynam-

ics in low-dimensional continuous-time business cycle models: the
Sil’nikov case, H.W. Lorenz, System Dynamics Review, vol.8 no.3, 1992.

68
 Solutions to chapter 2. Dynamical Systems

2.2 a. The transfer function is

G(s) = C (s I − A)−1 B + D
4−1 3 4
s+2 0 5
3
= ( −1 1 ) +2
0 s+3 2
2s2 + 7s + 1
= .
s2 + 5s + 6

From the transfer function it is easy to determine the differential

equation

Y (s) = G(s)U (s)

(s2 + 5s + 6) Y (s) = (2s2 + 7s + 1)U (s)
ÿ + 5 ẏ + 6y = 2ü + 7u̇ + u

b. The transfer function is

G(s) = C (s I − A)−1 B + D
4−1 3 4
s+7 −2 3
3
= ( −2 1 ) =
15 s−4 8
2s + 3
= .
s2 + 3s + 2

The differential equation becomes

Y (s) = G(s)U (s)

(s2 + 3s + 2) Y (s) = (2s + 3)U (s)
ÿ + 3 ẏ + 2y = 2u̇ + 3u

5s + 8
c. G(s) = , ẏ + y = 5u̇ + 8u
s+1
3s2 + 7s + 18
d. G(s) = , ÿ + 2 ẏ + 5y = 3ü + 7u̇ + 18u
s2 + 2s + 5

2.3 a. Partial fraction expansion of the transfer function yields

2 5
G(s) = 2 + −
s+3 s+2

and by applying the inverse Laplace transform, one obtains the im-
pulse response

h( t) = L−1 G(s) = 2δ ( t) + 2e−3t − 5e−2t , t ≥ 0.

Comment. Because the system matrix was given in diagonal form, another
possibility would have been to compute the impulse response as
   e−2t  
0 5
h( t) = Ce At B + Dδ ( t) =  −1

1
    + 2δ ( t),
−3t   
t ≥ 0.
0 e 2

69
Solutions to chapter 2. Dynamical Systems

The step response is computed by e.g. integrating the impulse response

Zt Zt
2δ (τ ) + 2e−3τ − 5e−2τ dτ
! "
y( t) = h(τ ) dτ =
0 0
6t
5 −2τ 2
5
=2+ e − e−3τ
2 3 0
1 5 −2t 2 −3t
= + e − e , t ≥ 0.
6 2 3

b. The transfer function has the partial fraction expansion

1 1
G(s) = +
s+1 s+2

and the impulse response becomes

h( t) = L−1 G(s) = e− t + e−2t , t ≥ 0.

The step response is thus given by

Zt
3 1
y( t) = h(τ ) dτ = − e− t − e−2t , t ≥ 0.
0 2 2

c. h( t) = 5δ ( t) + 3e− t , y( t) = 8 − 3e− t , t≥0

√
d. h( t) = 3δ ( t) + e− t sin 2t + e− t cos 2t = 3δ ( t) + 2e− t sin 2t + π4
! "

y( t) = 3 + 51 e− t (3 + sin 2t − 3 cos 2t) , t ≥ 0

2.4 After the Laplace transform, one obtains

s X = A X + BU
Y = C X + DU

Solve for X

(s I − A) X = BU
X = (s I − A)−1 BU

This gives
1 2
Y = C (s I − A)−1 BU + DU = C (s I − A)−1 B + D U

2.5 a. The poles are the solutions of the characteristic equation s2 +4s+3 = 0,
i.e. s = −1 and s = −3. The system lacks zeros. Since all poles have
negative real part, the system is stable.

b. The static gain is G(0) = 31 .

70
 Solutions to chapter 2. Dynamical Systems

c. At a step response, the input signal u( t) is a step, that has the Laplace
transform U (s) = 1s . The output becomes

1 1
Y (s) = G(s)U (s) =
s2 + 4s + 3 s

The final value can be calculated using the final value theorem

1 1 1
lim y( t) = lim sY (s) = lim s =
t→∞ s→0 s→0 s2 + 4s + 3 s 3

The final value theorem may be used only if we know that the final
value exists (i.e., that y( t) does not go to infinity). Since we have
shown that the system is stable, we know that the final value exists.
The initial value may, in the same manner, be computed using the
initial value theorem

1 1
lim y( t) = lim sY (s) = lim s =0
t→0 s→∞ s→∞ s2 + 4s + 3 s

d. At an impulse response, the Laplace transform of the input signal is

U (s) = 1. The output signal becomes

1
Y (s) =
s2 + 4s + 3

We use the final and initial value theorems

1
lim y( t) = lim sY (s) = lim s =0
t→∞ s→0 s→0 s2 + 4s + 3
1
lim y( t) = lim sY (s) = lim s =0
t→0 s→∞ s→∞ s2 + 4s + 3
e. The step response of the system is given by

1 1
Y (s) =
s2 + 4s + 3 s

from a previous subproblem. The derivative of a signal is in the

Laplace domain given by multiplication of s. We may denote the
derivative of the step response z( t) and get

1
Z (s) = sY (s) =
s2 + 4s + 3

We see that the derivative of the step response is the same as the
impulse response, and from the previous subproblem we thus get

lim z( t) = 0
t→0

2.6 a. The poles are the solutions the characteristic equation s2 + 0.6s +
0.25 = 0, i.e. s = −0.3 ± 0.4i. The system lacks zeros.

71
Solutions to chapter 2. Dynamical Systems

b. The static gain is G(0) = 1.

1
c. The input (a step) has the Laplace transform U (s) = s. The output
becomes
0.25
Y (s) = G(s)U (s) =
s(s2 + 0.6s + 0.25)
Because this system has complex poles, we first rewrite it as

ω2
Y (s) =
s(s2 + 2ζωs + ω 2 )

where ω = 0.5 and ζ = 0.6. We then utilize the inverse Laplace

transformation (transform no. 28) and obtain

y( t) = 1 − 1.25e−0.3t sin(0.4t + 0.9273)

The step response is shown below.

Step Response
1.2

0.8
Amplitude

0.6

0.4

0.2

0
0 5 10 15 20
Time [s]

2.7 Laplace transformation of the differential equation m ÿ + c ẏ + ky = f

yields
(ms2 + cs + k) Y = F
and the transfer function is hence
1
G(s) = .
ms2 + cs + k
p
The poles are s = −c/2m ± i k/m − c2 /4m2 . A change in k implies a
change of the imaginary part of the poles. A change in c affects both
the real and imaginary parts.
The poles cannot end up in the right half plane due to physical reasons,
since c ≥ 0 and m > 0.

1
2.8 a. G(s) =
LCs2 + RCs + 1
s
1 A 2h0
b. G(s) = , T=
Ts + 1 a ˆ

72
 Solutions to chapter 2. Dynamical Systems

2.9 a. To be asymptotically stable, all eigenvalues of the system matrix A

must lie strictly within the left half plane (LHP). I.e. Re(λi ) < 0 ∀ i.
The eigenvalues of A are given by the characteristic equation

det(λ I − A) = 0

which in this case has two solutions, λ1 = −i and λ2 = i. Since

the eigenvalues do not lie strictly within the LHP, the system is not
asymptotically stable.

b. If all the eigenvalues of A lie strictly within the LHP, we are guaran-
teed stability. If any eigenvalue lies strictly in the RHP we have an
unstable system. If, on the other hand, there are eigenvalues on the
imaginary axis, the system can be either stable or unstable.
In this example there are no eigenvalues in the RHP. Additionally, all
eigenvalues on the imaginary axis are distinct. This tells us that the
system is stable.

2.10 The characteristic equation is

s3 + 2s2 + 3s + 7 = 0

The transfer function is stable if all coefficients are positive, which is

the case, and if the product of the s2 - and s1 coefficients is greater
than the s0 coefficient. The transfer function is therefore not stable,
since 2 · 3 < 7.

2.11 First and second order systems are stable if and only if the coefficients
in the denominator polynomial of the transfer function are positive.
All transfer functions except for G3 are stable. All step responses
except for E corresponds to stable systems. Thus, G3 = E.
All transfer functions except G3 , G4 , and G7 have a static gain of
Gi (0) = 1, which means that the final value of the step response is
one. G7 has a static gain of G7 (0) = 2/3, which gives G7 = C.
Step response D has a derivative that is not equal to 0 at t = 0.
Checking the initial derivative of the transfer functions gives that
this only is true for G1 and G5 . G1 has a time constant of 10s while
G5 has a time constant of 1s. Thus, G5 = D.
Now two step responses remain, A and B, which corresponds to second
order systems with complex poles and a static gain of one. The transfer
functions that fulfills these specifications are G2 and G6 . The relative
damping ζ is less for G6 than for G2 , which gives G2 = A and G6 = B.

2.12 1 The system has the poles in −1/4 ± i and a zero in −1. The transfer
function is thus

s+1 s+1
G(s) = K 1 2
(K .
(s + 4) +1 s2 + 12 s + 1

73
Solutions to chapter 2. Dynamical Systems

The initial value, initial derivative and final value become

y(0) = lim G(s) = 0

s→+∞
ẏ(0) = lim sG(s) = K ,= 0
s→+∞
lim y( t) = G(0) = K ,= 0
t→+∞

The step response is oscillating with period T = 2π /1 ( 6. This must

be step response D.
2 The system has poles in −1 and −2 and a zero in 1. The transfer
function is
s−1
G(s) = K
(s + 1)(s + 2)
The initial value, initial derivative and final value become

y(0) = lim G(s) = 0

s→+∞
ẏ(0) = lim sG(s) = K ,= 0
s→+∞
K
lim y( t) = G(0) = − ,= 0
t→+∞ 2

We see that the initial derivative and the final value have different
signs. This fits step response F.
3 The system has poles in −1/4 ± i and a zero in 0. The transfer
function is
s s
G(s) = K 1 2
(K 2 1
(s + 4 ) + 1 s + 2s + 1
The initial value, initial derivative and final value become

y(0) = lim G(s) = 0

s→+∞
ẏ(0) = lim sG(s) = K ,= 0
s→+∞
lim y( t) = G(0) = 0
t→+∞

The step response is oscillating with period T = 2π /1 ( 6. This is

step response G.
4 The system has poles in −1 and −2 and a zero in −3. The transfer
function is
s+3
G(s) = K .
(s + 1)(s + 2)
The initial value, initial derivative and final value become

y(0) = lim G(s) = 0

s→+∞
ẏ(0) = lim sG(s) = K ,= 0
s→+∞
3K
lim y( t) = G(0) = ,= 0
t→+∞ 2

The initial derivative and final value have the same sign. The only
nonoscillative step response which suits these criteria is C.

74
 Solutions to chapter 2. Dynamical Systems

2.13a. For the impulse response, u( t) = δ ( t) (Dirac function). We have

U (s) = 1
Y (s) = G(s)U (s) = G(s)

Initial value:
lim y( t) = lim sY (s) = lim sG(s)
t→0 s→∞ s→∞

Final value:
lim y( t) = lim sY (s) = lim sG(s)
t→∞ s→0 s→0

All transfer functions have stable poles, and thus we may use the final
and initial value theorems.
y(0) lim t→∞ y( t)
G1 (s) 1 0
G2 (s) 3 0
G3 (s) 0 0
G4 (s) 0 0
G5 (s) 0 1
G6 (s) 0 1

Both impulse responses in the figure have initial value 0 and final
value 0. Thus, they correspond to G3 (s) and G4 (s). The poles of G3 (s)
are located in s = −1, and the poles of G4 (s) in s = −3. G4 (s) is
thus a faster system, which corresponds to the impulse response for
lemonade. G3 (s) is slower, which corresponds to the impulse response
for whole grain pasta.

b. Normally, you eat a certain amount of food in a relatively short time,

and then do not eat for a longer period of time. Thus, you may model
food intake as an impulse, that occurs instantaneously compared to
the time it takes for the body to digest the food.
A step response in food intake would correspond to eating continuously
during a longer time. Feeding through intravenous therapy could be
described by a step response.

2.14a.

Y = G1 (U + G2 Y )
Y (1 − G1 G2 ) = G1U
G1
Y = U
1 − G1 G2

b.

Y = G2 ( H1U + G1U + H2 Y )
Y (1 − G2 H2 ) = (G2 H1 + G2 G1 )U
G2 H1 + G2 G1
Y = U
1 − G2 H2

75
Solutions to chapter 2. Dynamical Systems

c. Introduce the auxiliary variable Z, being the output of G1

Z = G1 (U + G3 ( Z + G2 Z ))
Z (1 − G1 G3 − G1 G3 G2 ) = G1U
G1
Z= U
1 − G1 G3 − G1 G3 G2
G2 G1
Y = U
1 − G1 G3 − G1 G3 G2

d.

Y = G2 (− H2 Y + G1 (U − H1 Y ))
Y (1 + G2 H2 + G2 G1 H1 ) = G2 G1U
G2 G1
Y = U
1 + G2 H2 + G2 G1 H1

2.15a. From the block diagram we get

Y (s) = G P (s)(U (s) + D(s))

U (s) = G R (s) E (s)
E (s) = R(s) − Y (s)

Solve for Y (s):

G P (s)G R (s) G P (s)

Y (s) = R(s) + D(s)
1 + G P (s)G R (s) 1 + G P (s)G R (s)

The system has two input signals, R(s) and D(s). The transfer func-
tion from R(s) to Y (s) is

G P (s)G R (s)
G yr (s) =
1 + G P (s)G R (s)

b. The transfer function from D(s) to Y (s) is

G P (s)
G yd (s) =
1 + G P (s)G R (s)

c. Solve for E (s):

1 G P (s)
E (s) = R(s) − D(s)
1 + G P (s)G R (s) 1 + G P (s)G R (s)

The transfer function from R(s) to E (s) becomes

1
G er (s) =
1 + G P (s)G R (s)

76
 Solutions to chapter 2. Dynamical Systems

d. Solve for U (s):

G R (s) G P (s)G R (s)

U (s) = R(s) − D(s)
1 + G P (s)G R (s) 1 + G P (s)G R (s)

The transfer function from D(s) to U (s) becomes

G P (s)G R (s)
Gud (s) = −
1 + G P (s)G R (s)

2.16a. Partial fraction expansion yields

s2 + 6s + 7 s+1 −1 2
G(s) = 2
= 2 +1= + +1
s + 5s + 6 s + 5s + 6 s+2 s+3

One has a certain freedom when choosing the coefficients of the B

and C matrices. However, the products bi c i remain constant. Let e.g.
b1 = b2 = 1. This enables us to immediately write the system in
diagonal form:
   
dx  −2 0  1
=
  x +
   u
dt 0 −3 1
   
y = −1 2  x
 +1u

b. First rewrite the system as

b0 s + b1 s+1
G(s) = +d= 2 +1
s2 + a 1 s + a 2 s + 5s + 6

The controllable canonical form can be directly read from the transfer
function    
dx  −5 −6  1
= x+
   u
dt 1 0 0
   
y = 1 1 x +1u
 

c. The observable canonical form is obtained in the same manner

   
dx  −5 1  1
=
 x+
   u
dt −6 0 1
   
y = 1 0 x + 1u

77
Solutions to Chapter 3. Frequency Analysis

3.1 a. The output is given by

1 2
y( t) = pG(3i)p sin 3t + arg G(3i)

where √
0.01 1 + 100ω 2
pG(iω )p = √ √
1 + ω 2 1 + 0.01ω 2
and
arg G(iω ) = arctan 10ω − arctan ω − arctan 0.1ω

For ω = 3 one obtains pG(iω )p = 0.0909 and arg G(iω ) = −0.003

which gives
y( t) = 0.0909 sin(3t − 0.003)

b. Reading from the plot yields pG(3i)p ( 0.09 and arg G(3i) ( 0. We
obtain
y( t) = 0.09 sin 3t

3.2 a. The air temperature will affect the water temperature much faster in
the small garden pool than in the sea. Faster influence means that the
gain is greater for higher frequencies. Thus, the solid line represents
the garden pool – G2 (s), and the dashed line the sea water – G1 (s).

b. The time scale in the Bode diagram is hours. Thus, we convert the
period time to hours:

1 year = 365 · 24 h = 8760 h

The angular frequency of the oscillation thus becomes

2π 2π
ω= = rad/h = 7 · 10−4 rad/h.
T 8760

We read off the dashed amplitude curve at this frequency and get

pG(iω )p ( 0.5

This means that the oscillation in the output (the water temperature)
has half as big amplitude as the oscillation in the input (the air tem-
perature) at this frequency. The difference between the maximum and
minimum temperature in the air is ∆Tlu f t = 19○ C−(−5○ C ) = 24○ C.
The difference between the maximum and minimum temperature in
the water thus becomes ∆Tvatten = 0.5 · 24○ C = 12○ C.

c. The angular frequency for the oscillation is now

2π 2π
ω= = rad/h = 0.26 rad/h.
T 24

78
 Solutions to chapter 3. Frequency Analysis

We read off the solid phase curve at this frequency and get

arg(G(iω )) ( −30○ .

30 ○
This means that the peak of the output signal will be 360 ○ ( 0.08
period = 0.08 · 24 h = 1.92 h later than the peak of the input signal.
Thus, the pool is as warmest around 15.00.

3.3 a. The output is given by

1 2
y( t) = pG(iω )p sin ω t + arg G(iω )

where

10 10
pG(iω )p = =p
(iω )2 + 0.5iω + 1 (1 − ω 2 )2 + (0.5ω )2

and

10
arg G(iω ) = arg = − arg (1 − ω 2 ) + 0.5ωi
! "
(iω )2
+ 0.5iω + 1

0.5ω

 − arctan , ω<1



 1 − ω2
= −π /2, ω=1




 0.5ω
 − arctan − π, ω>1
1 − ω2

The output becomes

10.4 sin (0.2t − 5.9○ ), 20.0 sin ( t − 90.0○ ), 0.011 sin (30t − 179.0○ )

b. For ω = 0.2 one reads pG(iω )p ( 10 and arg G(iω ) ( −5○ . For ω = 1
one reads pG(iω )p ( 20 and arg G(iω ) ( −90○ . For ω = 30 one reads
pG(iω )p ( 0.01 and arg G(iω ) ( −180○ . The output is approximately

10 sin (0.2t − 5○ ), 20 sin ( t − 90○ ), 0.01 sin (30t − 180○ )

3.4 We use the following general approach to draw Bode plots

• Factor the transfer function of the system.
• Determine the low frequency asymptote (small s).
• Determine the corner frequencies (i.e. the magnitude of the poles
and zeros of the system.)
• Draw the asymptotes of the gain curve from low to high frequen-
cies, aided by the following rules of thumb
– A pole decreases the slope of the gain curve by 1 at the corner
frequency.

79
Solutions to chapter 3. Frequency Analysis

– A zero increases the slope of the gain curve by 1 at the corner

frequency.
• Draw the asymptotes of the phase curve from low to high fre-
quencies, aided by the following rules of thumb
– A (stable) pole decreases the value of the phase curve by 90○
at the corner frequency.
– A (stable) zero increases the value of the phase curve by 90○
at the corner frequency.
• Draw the real gain- and phase curves, aided by the asymptotes
and sample curves in the collection of formulae.

a. The transfer function can be written

1
G(s) = 3 ·
1 + s/10

Low frequency asymptote: G(s) ( 3.

Corner frequency: ω = 10 rad/s (pole).
The gain curve starts with slope 0 and value 3. The slope decreases
by 1 at ω = 10 rad/s, due to the pole, and thus ends being −1.
The phase curve starts at 0○ . The phase is decreased by 90○ at ω =
10 rad/s, due to the pole, and thus ends being −90○ .
The asymptotes and the finished Bode plots are shown in Figure 3.1.

1
10
Magnitude

0
10

−1
10
0 1 2
10 10 10

−30
Phase

−60

−90
0 1 2
10 10 10
Frequency [rad/s]

3
Figure 3.1 The Bode plot of G(s) = 1+s/10
.

b. The transfer function can be written

1 1
G(s) = 10 · ·
1 + 10s 1 + s

80
 Solutions to chapter 3. Frequency Analysis

Low frequency asymptote: G(s) ( 10.

Corner frequencies: ω = 0.1 rad/s (pole), ω = 1 rad/s (pole).
The gain curve starts with slope 0 and value 10. The slope is decreased
by 1 at ω = 0.1 rad/s, due to the first pole, and by 1 at ω = 1 rad/s,
due to the second pole. Thus, the final slope becomes −2.
The phase curve starts at 0○ . The phase is decreased by 90○ at ω =
0.1 rad/s, due to the first pole, and by 90○ at ω = 1 rad/s, due to the
second pole. Thus, the final phase is −180○ .
The asymptotes and the finished Bode plot are shown in Figure 3.2.

1
10
Magnitude

0
10

−1
10

−2
10
−2 −1 0 1
10 10 10 10

−45
Phase

−90

−135

−180
−2 −1 0 1
10 10 10 10
Frequency [rad/s]

10
Figure 3.2 Bode plot of G(s) = (1+10s)(1+s)
.

c. The transfer function can be written

1
G(s) = e−s ·
1+s

Low frequency asymptote: G(s) ( 1.

Corner frequency: ω = 1 rad/s (pole).
The delay (e−s ) does not affect the gain curve, which starts with slope
0 and value 1. The slope is decreased by 1 at ω = 1 rad/s, due to the
pole, and the final slope is thus −1.
The phase curve is harder to sketch. One approach is to draw the
asymptotes of the system without the delay and superposition it with
the phase curve of e−s , which can be obtained from the collection
of formulae or by computing some points and interpolating between
these.
Anyway, we see that the phase curve starts at 0○ and that the phase
then decreases both due to the pole (at ω = 1 rad/s) and the delay.
The delay causes the phase to approach −∞ for large ω.
The finished plot is shown in Figure 3.3.

81
Solutions to chapter 3. Frequency Analysis

0
10
Magnitude

−1
10
−1 0 1
10 10 10

−90
Phase

−180

−270

−360
−1 0 1
10 10 10
Frequency [rad/s]

e−s
Figure 3.3 Bode plot of G(s) = 1+s
.

d. The transfer function can be written

1 1
G(s) = · (1 + s) ·
s 1 + s/10

1
Low frequency asymptote: G(s) ( .
s
Corner frequencies: ω = 1 rad/s (zero), ω = 10 rad/s (pole).
The gain curve starts with slope −1. The slope increases by 1 at
ω = 1 rad/s, due to the zero, and at ω = 10 rad/s the slope decreases
by 1, due to the pole. Consequently, the final slope is −1.
The phase curve starts at −90○ . The phase increases by 90○ at ω =
1 rad/s, due to the zero, and decreases by 90○ at ω = 10 rad/s, due to
the pole. Consequently, the final phase is −90○ .
The finished plot is shown in Figure 3.4.

e. The transfer function can be written

1 1
G(s) = 2 · · (1 + 5s) ·
s 1 + 2ζ (s/2) + (s/2)2

where ζ = 0.2.
2
Low frequency asymptote: G(s) ( .
s
Corner frequencies: ω = 0.2 rad/s (zero), ω = 2 rad/s (complex conju-
gated pole pair).
The gain curve starts with slope −1. The slope is increased by 1 at
ω = 0.2 rad/s, due to the zero, and decreased by 2 at ω = 2 rad/s, due
to the pole pair. Consequently, the final slope is −2.

82
 Solutions to chapter 3. Frequency Analysis

1
10

Magnitude
0
10

−1
10
−1 0 1 2
10 10 10 10

−30
Phase

−60

−90
−1 0 1 2
10 10 10 10
Frequency [rad/s]

1+s
Figure 3.4 Bode plot of G(s) = s(1+s/10)
.

The phase curve starts at −90○ . The phase is increased by 90○ at

ω = 0.2 rad/s, due to the zero, and decreased by 180○ at ω = 2 rad/s,
due to the pole pair. Consequently, the final phase is −180○ .
The low damping (ζ = 0.2) of the complex conjugated pole pair gives
the gain curve a resonance peak at ω = 2 rad/s. Additionally, the
phase decreases rapidly at this frequency, cf. the sample curves in the
collection of formulae. The finished plot is shown in Figure 3.5.

3.5 a. The Nyquist curve start in 3 (the static gain) for ω = 0 rad/s. Both
the gain and phase are strictly decreasing, which makes the curve
turn clockwise while its distance to the origin decreases. The gain
and phase approach 0 and −90○ , respectively, for large values of ω.
The curve is thus bound to the fourth quadrant and approaches the
origin along the negative imaginary axis as ω → ∞.

83
Solutions to chapter 3. Frequency Analysis

2
10
Magnitude
1
10

0
10

−2 −1 0 1
10 10 10 10

−45
Phase

−90

−135

−180
−2 −1 0 1
10 10 10 10
Frequency [rad/s]
2(1+5s)
Figure 3.5 Bode plot of G(s) = s(1+0.2s+0.25s2 )
.

Aided by this analysis, one can now sketch the Nyquist curve by
choosing a few frequencies (e.g. ω = 1, 10 and 100 rad/s) and drawing
the corresponding points in the complex plane. The finished curve is
shown in Figure 3.6.

0.5

−0.5
Im

−1

−1.5

−2
−0.5 0 0.5 1 1.5 2 2.5 3 3.5
Re
3
Figure 3.6 Nyquist curve of G(s) = 1+s/10
.

b. The Nyquist curve starts in 10 (the static gain) for ω = 0 rad/s.

Both the gain and phase are strictly decreasing, which makes the
curve turn clockwise while its distance to the origin decreases. The
gain and phase approach 0 and −180○ , respectively, for large values
of ω. The curve will thus go from the fourth to the third quadrant,
approaching the origin along the negative real axis as ω → ∞.

84
 Solutions to chapter 3. Frequency Analysis

The intersection with the negative imaginary axis can be drawn by

reading off the magnitude when the phase is −90○ . One can now
sketch the Nyquist curve by choosing a few additional frequencies
(e.g. ω = 0.1, 1 rad/s) and drawing the corresponding points in the
complex plane. The finished curve is shown in Figure 3.7.

−2
Im

−4

−6

−2 0 2 4 6 8 10 12
Re
10
Figure 3.7 Nyquist curve of G(s) = (1+10s)(1+s)
.

c. The Nyquist curve starts in 1 (the static gain) for ω = 0 rad/s. Both
the gain and phase are strictly decreasing, which makes the curve
turn clockwise while its distance to the origin decreases. The gain
and phase approach 0 and −∞, respectively, for large values of ω.
The curve will thus rotate infinitely many times as it approaches the
origin. The first intersections with the axis can be drawn by reading
off the magnitude when the phase is −90○ , −180○ , −270○ and −360○ ,
respectively. The finished curve is shown in Figure 3.8.

0.5

0
Im

−0.5

−1

−1 −0.5 0 0.5 1 1.5

Re
e−s
Figure 3.8 Nyquist curve of G(s) = 1+s
.

85
Solutions to chapter 3. Frequency Analysis

3.6 Let the sought transfer function be G(s). The gain curve starts with
slope −1, which indicates that G(s) contains a factor 1s (an integrator).
We observe that there are two corner frequencies: ω 1 = 1 and ω 2 =
100 rad/s. The gain curve breaks upwards once at ω 1 and downward
once at ω 2 . Hence, the nominator hosts a factor 1 + s, whereas the
denominator contains a factor 1 + s/100. In addition, G(s) contains a
constant gain K . We thus have

K (1 + s)
G(s) =
s(1 + s/100)

We evaluate the low frequency asymptote of the gain curve at e.g.

ω = 0.01 rad/s, in order to determine K . This yields

K
pG(0.01i)p = 0.01 =1 [ K = 0.01

Finally we verify that the phase curve matches this system.

3.7 Let the sought transfer function be G(s). The gain curve has two
corner frequencies: ω 1 = 2 and ω 2 = 100 rad/s. The gain curve breaks
downwards once at ω 1 and three times at ω 2 . Thus the denominator
s 3
of G(s) contains the factors (1 + 2s ) and (1 + 100 ) . The slope of the low
frequency asymptote is 1. Thus G(s) has a factor s in the nominator.
Additionally, G(s) contains a constant gain K . We have

Ks
G(s) = s 3
(1 + 2s )(1 + 100 )

The factor K is computed by determining a point on the LF asymptote,

e.g. G L F (s) = K s

pG L F (iω )p = Kω = 1

for ω = 0.5 rad/s. This gives

K =2

Finally we verify by checking that the phase curve matches this sys-
tem.

86
Solutions to Chapter 4. Feedback Systems

4.1 a. Laplace transformation of the differential equation yields

sY (s) + 0.01Y (s) = 0.01U (s)

The transfer function G P (s) is thus given by

0.01
Y (s) = G P (s)U (s) = U (s)
s + 0.01

b. The transfer function of the closed loop system becomes

0.01
G P (s)G R (s) K 0.01 K
G(s) = = s+0.01
0.01
=
1 + G P (s)G R (s) 1 + s+0.01 K s + 0.01 + 0.01 K

c. The desired and actual characteristic polynomials are the same if all
their coefficients match. Identification of coefficients yields

0.1 = 0.01 + 0.01 K \ K=9

4.2 Since r( t) = 0, the control error becomes e( t) = − y( t).

G P (s)
Y (s) = G P (s)( F (s) − G R (s) Y (s)) \ Y (s) = F (s)
1 + G R (s)G P (s)

If f ( t) is a unit step, we have F (s) = 1s .

a. Seek y(∞) for G R = K

1 1 1
y(∞) = lim sY (s) = lim s =
s→0 s→0 (ms2 + ds + K ) s K

The function sY (s) has all poles in the left-half plane when the pa-
rameters m, d and K are positive.

b. The same assignment, but with G R (s) = K 1 + K 2 /s. This yields

1 1
y(∞) = lim s K2
s→0 (ms2 + ds + K 1 + s )
s
s
= lim =0
s→0 ms3 + ds2 + K1s + K2

under the assumption of stability, which is the case for m > 0, d > 0
and K 1 > md K 2 > 0. Rule: If the disturbance is a step, one needs at
least one integrator before the point in the block diagram where the
disturbance is introduced, in order to make the stationary error zero.

87
Solutions to chapter 4. Feedback Systems

4.3 a. For the closed loop system it holds, when R = 0, that

U (s) = K (0 − Y (s)) = − K (G P (s)U (s) + N (s))

from which one obtains

−K
U (s) = N (s)
1 + KG P (s)
(4.1)
1
Y (s) = G P (s)U (s) + N (s) = N (s)
1 + KG P (s)

1
b. Inserting G P (s) = s+1 into (4.1) yields the relations

− K (s + 1)
U (s) = N (s)
s+1+ K
s+1
Y (s) = G P (s)U (s) + N (s) = N (s) =: G yn (s) N (s)
s+1+ K
In stationarity it holds that

y( t) = ApG yn (iω )p sin(ω t + arg G yn (iω ))

√
1 + ω2
3 4
ω
= Ap sin ω t + arctan ω − arctan
( K + 1)2 + ω 2 K +1
u( t) = − Ky( t)
√
1 + ω2
3 4
ω
= −K Ap sin ω t + arctan ω − arctan
( K + 1)2 + ω 2 K +1

c. With A = 1 and K = 1 the amplitudes of the oscillations in u and y

become s
1 + ω2
Au =
4 + ω2
s
1 + ω2
Ay =
4 + ω2
For ω = 0.1 rad/s the amplitudes become

Au ( 0.5
A y ( 0.5

while ω = 10 rad/s yields

Au ( 1
Ay ( 1

4.4 With G P (s) = 1/( Js2 ) we obtain

E (s) = θ re f (s) − θ (s)

= θ re f (s) − G P (s)( M (s) + KG R (s) E (s))
1 G P (s)
[ E (s) = θ re f (s) − M (s)
1 + KG P (s)G R (s) 1 + KG P (s)G R (s)

88
 Solutions to chapter 4. Feedback Systems

Assume step changes in disturbance momentum M d0 and reference

0
θ re f . We postulate G R (s) = Q(s)/ P ( S ), which gives

1
0
1 θ re f Js2 M0
E (s) = · − · d
Q(s) s Q(s) s
1+ K 1+ K 2
Js2 P (s) Js P (s)
0
s2 J P (s) θ re f P (s) M d0
= · − ·
s2 J P (s) + K Q(s) s s2 J P (s) + K Q(s) s

The stationary error becomes

e ∞ = lim e( t) = lim s E (s)

t→∞ s→0

P (0) P (0)
=0− M d0 = − M0
K Q(0) K Q(0) d

where we have assumed that Q(0) ,= 0 and that the conditions for the
final value theorem are fulfilled. We see that P (0) = 0 yields e ∞ = 0.
In order to eliminate persistent angular errors caused by disturbance
momenta, it is consequently required to utilize a controller G R (s) with
at least one pole in the origin ( P (0) = 0).

4.5 The input of the thermocouple is the temperature u( t) of the bath,

which gives
1
u( t) = t [ U (s) = 2
s
The output y( t) is the reading of the temperature sensor. Thus

1 1
Y (s) = G(s)U (s) = ·
1 + sT s2

For the error e( t) = u( t) − y( t) it holds that

1 1 1
5 6
sT
E (s) = U (s) − Y (s) = 2 1 − = · 2
s 1 + sT 1 + sT s

The stationary error is obtained by means of the final value theorem

s2 T 1
e(∞) = lim s E (s) = lim = T = 10
s→0 s→0 1 + sT s2

The thermocouple measurement is hence 10○ less than the actual

temperature. I.e. the actual temperature of the bath is 102.6○ C +
10○ C = 112.6○ C.
Observe that the error in this has a bounded limit, despite the fact
that both u( t) and y( t) lack (bounded) limits as t → ∞. It is the
difference between u and y which converges to a constant value.

89
Solutions to chapter 4. Feedback Systems

4.6 The low frequency asymptote is

K
G L F (s) =
s2
where the constant K is given by

K
pG L F (iω )p = ; pG L F (i)p = 1 [ K =1
ω2
At the corner frequency ω 1 = 1 rad/s the slope changes from −2 to 0,
and at ω 2 = 5 rad/s it changes from 0 to −1. The transfer function for
the open loop system is thus

(1 + sT1 )2
Go (s) =
s2 (1 + sT2 )

where T1 = 1/ω 1 = 1 and T2 = 1/ω 2 = 0.2.

The transfer function of the closed loop system becomes

Go (s)
G(s) =
1 + Go (s)

The output is Y (s) = G(s) R(s) and the error E (s) becomes

1 s2 (1 + 0.2s)
E (s) = R(s) − Y (s) = R(s) = R(s)
1 + Go s2 (1 + 0.2s) + (1 + s)2

a.
a
R(s) = [ e ∞ = lim e( t) = lim s E (s)
s t→∞ s→0

as2 (1 + 0.2s)
= lim =0
s→0 s2 (1 + 0.2s) + (1 + s)2
The system can thus track inputs r( t) = a without a stationary error.

b.
b
R(s) = [ e ∞ = lim e( t) = lim s E (s)
s2 t→∞ s→0
−
bs(1 + 0.2s)
= lim 2 =0
s→0 s (1 + 0.2s) + (1 + s)2

The system can also track inputs r( t) = bt without a stationary error.

c.
2c
R(s) = [ e ∞ = lim e( t) = lim s E (s)
s3 t→∞ s→0

2c(1 + 0.2s)
= lim = 2c ,= 0
s→0 s2 (1 + 0.2s) + (1 + s)2
The input r( t) = ct2 , however, yields a stationary error.

d. Superposition can be used, since the closed loop system is linear and
time invariant (LTI). Here, the input is the sum of the inputs in sub-
assignments a and b. The total (superpositioned) stationary error thus
becomes e ∞ = 0 + 0 = 0.

90
 Solutions to chapter 4. Feedback Systems

e. The input r( t) = sin( t) yields

1
R(s) = [ lim s E (s)
1 + s2 s→0

s2 (1 + 0.2s)
= lim s =0
s→0 (s2 (1 + 0.2s) + (1 + s)2 )(1 + s2 )

but the input r( t) = sin( t) yields the output y( t) = yo sin( t + φ ), where

yo = pG(i)p , φ = ar ˆ G(i)

once transients have decayed. The error e( t) = r( t) − y( t) is thus also

a sinusoid and the limit
lim e( t)
t→∞

does not exist. This shows that the final value theorem should not be
used without caution. It is only valid for cases where a limit really
exists. The criterion is that all poles of s E (s) must have negative real
parts. (The factor s2 + 1 in the denominator yields two poles on the
imaginary axis.)

4.7 a. The sensitivity function is given by

1 1 s3 + 3s2 + 3s + 1
S (s) = = 6.5
=
1 + G P (s)G R (s) 1 + (s+ 1)3
s3 + 3s2 + 3s + 7.5

b. For ω = 0 rad/s we have p S (iω )p = 1/7.5. Low-frequency load distur-

bances are thus damped by a factor 7.5.

c. The sensitivity functions has its maximum value p S (iω )p =( 10 at

ω ( 1.6 rad/s.

4.8 a. The left curve shows the complementary sensitivity function, whereas
the sensitivity function is given by the right curve.

b. The disturbances at various frequencies are amplified according to the

gain curve of the sensitivity function. Disturbances below 2 rad/s are
hence reduced, disturbances in the range 0.2 to 2 rad/s are amplified
and disturbances above 2 rad/s pass straight through. The worst case
gain, 2, is obtained at the frequency 0.55 rad/s.

c. The complementary sensitivity function, corresponding to the closed

loop transfer function from r to y, lies close to 1 up to approximately
0.7 rad/s.

d. The maximal magnitude of the sensitivity function equals the inverse

of the minimal distance between the Nyquist curve and the point −1.
The minimal distance is thus 1/2 = 0.5. The distance to −1, as the
Nyqist curve intersects the negative real axis, must hence be at least
0.5. This implies that the gain margin is at least 2.

91
Solutions to chapter 4. Feedback Systems

4.9 The closed loop transfer function is given by

Go K
G(s) = = 2
1 + Go s + 2s + K

The poles of the closed loop system are given by the characteristic
equation √
s2 + 2s + K = 0 [ s = −1 ± 1 − K
For K = 0 the roots s 1,2 = 0, −2, i.e. the poles of the open loop system,
are obtained. The closed loop system G(s) has a double pole in s = −1
for K = 1. And as K → ∞ the roots become

s 1,2 = −1 ± i∞

The root locus, i.e. the roots of the characteristic equation as K varies,
is shown in Figure 4.1 .

Figure 4.1 Root locus of the system in assignment 4.9.

4.10 The open loop transfer function of the system is

K (s + 10)(s + 11) Q(s)

Go (s) = =K
s(s + 1)(s + 2) P (s)

The closed loop system becomes

Go (s) K Q(s)
G(s) = =
1 + Go (s) P (s) + K Q(s)

92
 Solutions to chapter 4. Feedback Systems

The characteristic equation is thus

P (s) + K Q(s) = 0
\s(s + 1)(s + 2) + K (s + 10)(s + 11) = 0
\s3 + (3 + K )s2 + (2 + 21 K )s + 110 K = 0

a. The criterion for stability is that all coefficients of the characteristic

polynomial
s3 + (3 + K )s2 + (2 + 21 K )s + 110 K
are positive and that

(3 + K )(2 + 21 K ) > 110 K

The inequality yields

15 2
K2 − K+ >0
7 7

It is fulfilled for K > 2 and K < 1/7. The closed loop system is hence
stable for
1
0<K <
7
and
K>2

b. Find the root locus for the characteristic equation, P (s) + K Q(s) = 0

s(s + 1)(s + 2) + K (s + 10)(s + 11) = 0 (4.2)

Let n = the degree of P (s) and m = the degree of Q(s). The root locus
has a maximum of max(n, m) = 3 branches.
Starting points:

P (s) = 0 [ s = 0, −1, −2

Ending points:
Q(s) = 0 [ s = −10, −11
The third branch will approach infinity.
To the right of each real point of the root locus, there must exists an
odd number of zeros of P (s) and Q(s). The points x, which fulfill this
are
x < −11 − 10 < x < −2 − 1 < x < 0
The root locus has pn − mp = 1 asymptote. This is the negative real
axis, since the range x < −11 on the real axis belongs to the root
locus.
The intersection with the imaginary axis is obtained by introducing
s = iω (4) above. This yields

−(3 + K )ω 2 + 110 K + i(−ω 3 + (2 + 21 K )ω ) = 0

93
Solutions to chapter 4. Feedback Systems

The resulting equation has a solution ω = K = 0 and

−(3 + K )ω 2 + 110 K = 0
I

ω 2 − (2 + 21 K ) = 0
√ √
gives K = 1/7, ω = ± 5 or K = 2, ω = ± 44.
We know from sub-assignment a that the closed loop system is unsta-
ble for 1/7 < K < 2. Consequently, the root locus lies in the right half
plane for these values of K . The principal shape of the root locus is
shown in Figure 4.2.

Figure 4.2 Root locus of the system in assignment 4.10.

4.11a. The open loop transfer function of the system is

K
G0 (s) =
s(s + 1)(s + 2)

The closed loop transfer function is thus

G0 (s) K
Gcl (s) = =
1 + G0 (s) s(s + 1)(s + 2) + K

The system is asymptotically stable if all zeros of the characteristic

polynomial

s(s + 1)(s + 2) + K = s3 + 3s2 + 2s + K

have negative real parts. This is the case if all coefficients are positive
and if
3·2> K
The system is thus asymptotically stable if 0 < K < 6.

94
 Solutions to chapter 4. Feedback Systems

b. Now we want to study the dependence of K on the stationary error,

as the reference increases as a linear function of time. The Laplace
transform of the control error is given by

1 s(s + 1)(s + 2)
E (s) = R(s) = R(s)
1 + G0 s(s + 1)(s + 2) + K

With r( t) = 0.1t, i.e. R(s) = 0.1/s2 , we obtain

0.1(s + 1)(s + 2)
E (s) =
s(s(s + 1)(s + 2) + K )

The signal s E (s) has all poles in the left-half plane when 0 < K < 6,
according to sub-assignment a. For this case we can utilize the final
value theorem
0.1(s + 1)(s + 2) 0.2
e(∞) = lim s =
s→0 s(s(s + 1)(s + 2) + K ) K

In order to obtain a stationary error less than 5 mV for the given

reference, it is required that K > 40. For such large values of K
the system is, however, not stable. It it hence impossible to meet the
specification.

4.12 According to the Nyquist theorem, the closed loop system is stable
exactly for those K > 0, which are also

a. K < 2

b. K < 1/1.5 = 2/3

c. K < 1/1.5 = 2/3

d. K < 1/(2/3) = 1.5

4.13 The Nyquist curve intersects the negative real axis when arg(G0 (iω )) =
−π, i.e. when
−3 arctan(ω ) = −π
This is fulfilled when √
π
ω = tan = 3
3
The intersection point is given by
√ 1
pG0 (i 3)p =
8
This means that the system is stable for K < 8.

4.14 The system is stable for

1
0< K < \ 0 < K < 0.29
3.5
as well as
1
1< K < \ 1< K <2
0.5

95
Solutions to chapter 4. Feedback Systems

4.15 The easiest way to solve the problem is through the Nyquist theorem.
The transfer function of the process is

e−9s
G P (s) =
(1 + 20s)2

The phase shift of the process is

arg G P (iω ) = −9ω − 2 arctan(20ω )

We seek the frequency for which the phase shift is −180○ . It is obtained
by solving the equation

−9ω − 2 arctan(20ω ) = −π

The equation lacks an analytic solution. However, it can be solved

numerically in several ways. The solution is

ω 0 ( 0.1

The next step is to determine the process gain for the given frequency.

1
pG(iω 0 )p = = 0.2
1 + 400ω 20

This gives us the gain margin

1
Am = =5
0.2
The gain K = 5 is thus the maximal admissible gain.

4.16 The loop transfer function is

1
10(1 + 2s
) 5(1 + 2s)
G P (s)G R (s) = e−sL · = e−sL ·
(1 + 10s) s(1 + 10s)

The cross-over frequency is the frequency where the magnitude of the

loop transfer function is equal to one.
p
5 1 + 4ω 2c
pG0 (iω c )p = p =1
ωc 1 + 100ω 2c

The equation can be solved numerically or analytically

25(1 + 4ω 2c ) = ω 2c (1 + 100ω 2c )
ω 4c − 0.99ω 2c − 0.25 = 0
ω 2c ( 1.199
ω c ( 1.1 rad/min

The phase at this frequency is

arg G0 (iω c ) = arctan 2ω c − arctan 10ω c − 90○ − ω c L

96
 Solutions to chapter 4. Feedback Systems

The requirement of the phase margin, φ m ≥ 10○ , gives

φ m = 180○ + arg G0 (iω c )

= 180○ + arctan 2ω c − arctan 10ω c − 90○ − ω c L
( 70○ − ω c L ≥ 10○

This gives the following limit for the time delay L:

60 π
L≤ · = 1 min
ω c 180

The time delay must therefore be less than one minute.

4.17a. True. Am = 1/p KG P (iω 0 )p, where ω 0 is the frequency for which the
Nyquist curve intersects the negative real axis.

b. True. φ m = π + arg G P (iω ) for pG P (iω )p = 1.

c. False. As K is decreased, all points on the Nyquist curve move closer

to the origin. Thus the phase margin increases as K is decreased.

d. True. The system is stable for K = 1 and all poles of G P (s) lie in
the left half plane. Consequently, the simplified Nyquist criterion can
be applied. For K = 2, the point −1 lies to the right of the Nyquist
curve, when it is traversed as ω increases. The closed loop system is
thus stable.

4.18a. This is the definition of the gain margin. From the plot one sees that
the phase −180○ corresponds to the gain ∼ 0.4. This yields the gain
margin 1/0.4 = 2.5.

b. This is the definition of the phase margin. From the plot one sees
that for gain 1, the phase is approximately −140○ . This yields a phase
margin of approximately 180○ − 140○ = 40○ .

4.19 The cross-over frequency and phase margin are read to be ω c = 0.07
and φ m = 40○ , respectively. The delay margin becomes

φm 40○ · 180
π
○
Lm = = = 10
ωc 0.07

97
Solutions to Chapter 5. State Feedback and
Kalman Filtering

5.1 a. From the controllability matrix

  β 1−β


Ws =  B AB  = 
 

1 −2

we obtain det Ws = −β − 1, i.e. controllability for all β ,= −1.

b. The observability matrix

   
 C  0 γ 
Wo = 
 =
  

CA 0 −2γ

has zero determinant independent of γ , i.e. the system is not observ-

able for any value of γ .

5.2 The controllability matrix is given by

   4 −8 
 
Ws =  B AB  = 
 

−2 4

The controllable states are determined by the columns of Ws and are

given as α(2, −1)T , where α is a scalar.

5.3    
 C   1 1 
Wo = 
 =
  

CA −1 −1
We see that Wo is singular (det Wo = 0). The state x is non-observable
if and only if (iff)
Wo x = 0
We obtain a non-observable x iff x1 + x2 = 0. The non-observable
states are thus given by
 
 1 
x = α
 

−1

where α is a number ,= 0.

5.4 The controllability matrix

   1 −1 
 
Ws =  B AB  = 
 

0 0

is singular, yielding an uncontrollable system. We can, however, con-

duct a more detailed investigation. The system can be written

d x1

 = − x1 + u, x1 (0) = 1
dt
 x2 = −2x ,
 d
x2 (0) = 1
2
dt

98
 Solutions to chapter 5. State Feedback and Kalman Filtering

Hence x2 ( t) = x2 (0) e−2t = e−2t , independent of the applied control

signal u. On the contrary, x1 can be controlled by means of u, to take
on any desired value. Thus, x2 → 0 as t → ∞. The states ( x1 , x2 ) which
can be reached in finite time t < ∞ make up the band 0 < x2 < 1 in
Figure 5.1.

x2

x1

Figure 5.1 Reachable states in assignment 5.4.

   
 3   10 
As a consequence, only the states 
  and 
   can be reached

0.5 0.1
in finite time.

5.5 The system is controllable, since

  1 4 


Ws =  B AB  = 
 

2 −7

has full rank.

5.6 a.    5 −10 
 
Ws =  B AB  = 
 

0 0
Ws has rank 1, i.e. the system is not controllable. The states which
can be reached in finite time from the origin are determined by the
columns of Ws . The controllable states can be parametrized by t as
x = t · (1 0)T .

b. G(s) = C (s I − A)− 1B + D = 5/(s + 2).

c. The following is a minimal state space representation of G(s)

I
ẋ = −2x + 5u
y=x

5.7 a. True. Since the system is controllable, one can place the poles of the
closed loop system arbitrarily by means of linear feedback from all
state variables.

99
Solutions to chapter 5. State Feedback and Kalman Filtering

b. False. A linear state feedback does not affect the zeros of the closed
loop system.

c. True if the system is observable.

d. True if the system is observable.

5.8 The closed loop system becomes

I
ẋ = ( A − BL) x + Bl r r
y = Cx

The characteristic equation is thus

det(s I − A + BL) = s2 + (3 + l 1 + 2l 2 )s + 2(1 + l 1 + l 2 ) = 0

We need (s + 4)2 = s2 + 8s + 16 = 0. Identification of coefficients

yields l 1 = 9, l 2 = −2. The closed loop transfer function is G(s) =
C (s I − A + BL)−1 Bl r . The stationary gain is G(0) is unity if

lr
G(0) = C (− A + BL)−1 Bl r = =1
4

yielding l r = 4.

5.9 a. The characteristic polynomial of the closed loop system is given by

s + 0.5 + 3l 1 3l 2
det(s I − ( A − BL)) = = s2 + (0.5 + 3l 1 )s + 3l 2
−1 s

The desired characteristic polynomial is

(s + 4 + 4i)(s + 4 − 4i) = s2 + 8s + 32

Identification of coefficients yields

   
L =  5/2 32/3  =  2.5 10.7 

The closed loop system transfer function is G yr (s) = C (s I − A +

BL)−1 Bl r . The stationary gain is unity if

3
G yr (0) = C (− A + BL)−1 Bl r = lr = 1
32

which yields l r = 32/3.

b. According to a rule of thumb, the observer poles shall be chosen 1.5–2

times faster than the state feedback. Place the poles of the Kalman
filter so that the distance from the origin is twice as large, leading to
the following characteristic polynomial

(s + 8 + 8i)(s + 8 − 8i) = s2 + 16s + 128

100
 Solutions to chapter 5. State Feedback and Kalman Filtering

The characteristic polynomial of the Kalman filter is given by

s + 0.5 k1
det(s I − ( A − KC )) = = s2 + (0.5 + k 2 )s + 0.5k 2 + k 1
−1 s + k2

Identification of coefficients yields

 
 120.25 
K=
 

15.5

5.10a. From the block diagram we see that

1
X1 = K 1U
s
1
X2 = K2 X1
s
1
X3 = X2
s

which gives

ẋ1 = K 1 u
ẋ2 = K 2 x1
ẋ3 = x2
y = x3

In matrix form we get

   
 0 0 0  K1 

 
 
 

ẋ = 

 K2 0 0 

 x+

 0 
 u
   
0 1 0 0
 
y = 0 0 1 x

b. The feedback law is given by

u = l r r − Lx

The closed-loop system becomes

ẋ = Ax + B(l r r − Lx) = ( A − BL) x + Bl r r

The poles of the closed-loop system are given by the eigenvalues of

A − BL, i.e., the roots of the closed loop characteristic equation
 
 s + K1l 1 K1l 2 K1l 3 

 

det(s I − ( A − BL)) = det 

 − K2 s 0 


 
0 −1 s
3 2
= s + K 1 l 1s + K 1 K 2 l 2 s + K 1 K 2 l 3 = 0

101
Solutions to chapter 5. State Feedback and Kalman Filtering

The poles of the closed loop system are given by the eigenvalues of
A − BL, i.e. the roots of the closed loop characteristic equation

(s + 0.5)3 = s3 + 1.5s2 + 0.75s + 0.125 = 0

One immediately obtains the solution


 1.5

 l1 =


 K1

 0.75
l2 =

 K1 K2



 0.125
l 3 =

K1 K2

5.11 The augmented system becomes

   
x1     x1     
d 
 
  A 0  
  B 0

 x2 
 =
 
 x2 
 +  u +
    r
dt 
 
 −C 0 
 
 0 1
x3 x3
      
 0 1 0   x1   0   0

 
 
 
 
 
 
 

=

 0 0 0


 x2

 + 
 1 
 u + 
 0 r

         
 
−1 0 0 x3 0 1
| {z } | {z } | {z } | {z }
Ae xe Be Br
 
We seek L e =  l 1 l2 l 3  such that

det(s I − ( A e − B e L e )) = (s + α)(s2 + 2ζωs + ω 2 )

Insertion of A e , B e and L e into the above expression yields

s3 + l 2 s2 + l 1 s − l 3 " s3 + (α + 2ζω )s2 + (ω 2 + 2ζωα)s + αω 2

Identifications of coefficients now yields

l 1 = ω 2 + 2ζωα
l 2 = α + 2ζω
l 3 = −αω 2

 T
5.12 The estimation error x̃ fulfills x̃˙ = ( A− KC ) x̃. where K =  k 1 k2  .
The characteristic equation of the estimation error becomes

det(s I − ( A − KC )) = s2 + (4 + k 2 )s + k 1 + 2k 2 + 3 = 0

The desired characteristic equation is

(s + 4)2 = s2 + 8s + 16 = 0

Identification of coefficients yields k 1 = 5, k 2 = 4.

102
 Solutions to chapter 5. State Feedback and Kalman Filtering

5.13a. The characteristic equation of the closed loop system is given by

s + 4 + l1 3 + l2
det s I − ( A − BL) = = s2 + (4 + l 1)s + 3 + l 2 = 0
! "
−1 s

The desired characteristic equation is

(s + 4)2 = s2 + 8s + 16 = 0

Which yields l 1 = 4 and l 2 = 13. The control law becomes

u = −l 1 x1 − l 2 x2 = −4x1 − 13x2

b. The states shall be estimated by means of a Kalman filter, i.e.

d x̂
= A x̂ + Bu + K ( y − C x̂)
dt

For x̃ we have
d x̃
= ( A − KC ) x̃
dt
Determine K such that all eigenvalues of the matrix A − KC are
placed in λ = −6.

det(λ I − A + KC ) = λ2 + (4 + k 1 + 3k 2 )λ + 3 + 3k 1 + 9k 2
= (λ + 6)2 = λ2 + 12λ + 36

Identify the coefficients and solve for k 1 and k 2 :

I I
4 + k 1 + 3k 2 = 12 k 1 + 3k 2 = 8
[
3 + 3k 1 + 9k 2 = 36 k 1 + 3k 2 = 11

The system of equations lacks solution, see the comment below.

c. The states are to be estimated by a Kalman filter, for which the

eigenvalues of A − KC shall be chosen such that

λ2 + (4 + k 1 + 3k 2 )λ + 3 + 3k 1 + 9k 2 = (λ + 3)2 = λ2 + 6λ + 9

Identifying coefficients and solving for k 1 and k 2 yields

I I
4 + k 1 + 3k 2 = 6 k 1 + 3k 2 = 2
[
3 + 3k 1 + 9k 2 = 9 k 1 + 3k 2 = 2

This leaves only one equation, which implies that there exists infinitely
many solutions, e.g. k 1 = 2, k 2 = 0 or k 1 = 0, k 2 = 32 etc.
The drawback of the proposed observer pole placement is that it yields
an estimation slower than the closed loop system. This does not affect

103
Solutions to chapter 5. State Feedback and Kalman Filtering

the response to reference changes, which is governed by the poles of

the closed loop system. However, the handling of process disturbances
becomes slower.

Comment
An inspection of the system’s observability shows that

C 1 3
det Wo = = =0
CA −1 −3

I.e. the system is not observable. The transfer function of the system
is given by

s+3 1
G(s) = C (s I − A)−1 B = =
s2 + 4s + 3 s+1

The eigenvalue −3 corresponds to a non-observable mode. The mode

is, however, controllable, which follows from the canonical controllable
form realization of the system. The characteristic equation of A − KC
can be written

det(λ I − A + KC ) = (λ + 3)(λ + k 1 + 3k 2 + 1)

This means that the Kalman filter has to estimate the non-observable
mode with its own speed. I.e. (at least) one of the eigenvalues of
A − KC must be placed in −3. This explains the failure to compute
a Kalman filter when the eigenvalues were to be placed in −6 and a
success when they were to be placed in −3. Note that in cases such as
this one, the result is either that there does not exist a solution K to
the Kalman filter problem, or that it exists infinitely many solutions.
When the system is observable, there exists a unique solution K to
the Kalman filter.

104
Solutions to Chapter 6. Design methods

6.1 a. The frequency function of the controller is given by

1
3 4
G R (iω ) = K 1 + i(ω Td − )
ω Ti

The gain and phase shift for a frequency ω are directly obtained
from the gain function A(ω ) and phase function φω of the controller,
respectively.
r
1 2
A(ω ) = pG R (iω )p = K 1 + (ω Td − )
ω Ti
1
φ (ω ) = arg G R (iω ) = arctan(ω Td − )
ω Ti

b. We immediately realize that the gain function A(ω ) has a unique

minimum for Im A(iω ) = 0, which means that

1
ω min = √
Ti Td

At this frequency the gain and phase shift are given by

A(ω min ) = K
φ (ω min ) = 0

Note that the phase shift is negative for ω < ω min (phase lag) and
positive for ω > ω min (phase lead).

6.2 a. The dashed gain curve is identical to the nominal one, except that it is
raised by a factor 4. This is thus the case where K has been multiplied
by 4. Observe that the dashed phase curve is not visible in the plot
since it coincides with the solid phase curve. The dotted gain curve
differs from the nominal (solid) curve at low frequencies, for which it is
lower. This indicates that Ti has been increased, resulting in decreased
low frequency gain. Also note that the phase curve has been raised for
low frequencies. The last (dash-dotted) curve apparently corresponds
to the case where Td has been increased. This is further indicated
by the factor 4 raise of the gain curve for high frequencies. Also for
this case, one can notice a certain increase in the phase, although for
somewhat higher frequencies.
The dashed step response is faster and less damped than the nominal
(solid) one. This is a characteristic sign of an increased gain K . The
corresponding Bode plot confirmingly shows that the cutoff frequency,
ω c , has increased (faster), while the phase margin has decreased (less
damped. The dotted step response features a slow mode both in the
reference- and load disturbance responses. Observe the relatively fast
increase in the reference response to approximately 0.8, followed by a
slow convergence to 1. This must be due to decreased integral action.

105
Solutions to chapter 6. Design methods

The integral time Ti has thus increased in this case. The corresponding
Bode plot shows that ω c is virtually unchanged. This is seen in the
step response by the fact that the first part has approximately the
same speed as the nominal case, whereas the following slow settling
is due to the decreased low frequency gain. The last (dash-dotted)
step response obviously corresponds to an increase of the derivative
time Td . The reference response is subject to an fast initial increase,
followed by a somewhat slower settling. This is seen in the Bode plot
by the fact that the high frequency gain has increased, while the
low frequency gain has remained unchanged. The load response is
somewhat slower and more damped than in the nominal case.

b. The dashed gain curve is lowered by a factor 2 in comparison to the

nominal one, corresponding to a decreased value of K . The dashed
phase curve consequently coincides with the nominal (solid) case.
The dotted gain curve has been increased for low frequencies, i.e. Ti
has decreased. The dash-dotted gain curve has been lowered at high
frequencies, i.e. Td has decreased.
The dashed step response is slower and more damped than the nomi-
nal one. This indicates that K has decreased, since neither a decrease
in Ti nor Td would yield a more damped step response. This is further
confirmed by the Bode plot. The only case where ω c has decreased is
when K has decreased. It is also the only case for which the phase
margin has increased. The two remaining step responses are both less
damped than the nominal one. In order to determine which of these
corresponds to a decrease of Ti , we look at the corresponding Bode plot
(the dotted one). This shows that the cutoff frequency ω c has increased
somewhat compared to the other nominal case. The dash-dotted Bode
plot, however, shows that the decrease of Td has not changed ω c . The
dotted step response is initially somewhat faster than the nominal
(solid) one, whereas the dash-dotted one is initially approximately as
fast as the nominal one. This implies that the dotted step response
corresponds to a decrease in Ti , while the dash-dotted one corresponds
to a decrease in Td .

6.3 The transfer function of the process is given by

C
GP =
Js + D
and the transfer function of the PI controller is given by

1
3 4
GR = K 1 +
sTi

We can now write down the closed loop transfer function Gcl as

GRGP
Gcl =
1 + GRGP

The characteristic polynomial is the denominator of Gcl

D + CK CK
s2 + s+
J JTi

106
 Solutions to chapter 6. Design methods

and the desired characteristic polynomial is

s2 + 2ζωs + ω 2

Identification of coefficients yields


2ζω J − D
K =

C
T = 2ζω J − D

i 2 ω J

6.4 The transfer function of the process, Gp , is given by

ki
Θ = GP I = I
Js2 + Ds
and the transfer function of the PID controller, G R , is given by

1
3 4
I = G R ( Θ re f − Θ) = K 1 + + sTd ( Θ re f − Θ )
sTi

where Θ re f is the Laplace transform of the reference value of θ . The

transfer function of the closed loop system, G, is thus given by

GRGP
Θ = GΘ re f = Θ re f
1 + GRGP

The characteristic polynomial is given by the denominator of G and is

D + K k i Td 2 K ki K ki
s3 + s + s+
J J JTi

We hence arrive at the polynomial equation

(s + α)(s2 + 2ζωs + ω 2 ) = s3 + (α + 2ζω )s2 + (2αζω + ω 2 )s + αω 2

Identification of coefficients yields the equations


D + K k i Td


 = α + 2ζω


 J
 Kk
i
= 2αζω + ω 2

 J

 K ki
= αω 2



JTi

from which one can calculate the sought controller parameters


J


 K = (2αζω + ω 2 )


 ki
2ζ 1

Ti = +

 ω α


 α + 2ζω − D/ J
Td =

2αζω + ω 2

107
Solutions to chapter 6. Design methods

6.5 a. The transfer function of the controller is

1 1 s+1
Gr (s) = K (1 + ) = (1 + ) =
sTi s s

The low frequency asymptote becomes

1
Gr (s) (
s

I.e. the gain curve is a straight line with slope = -1 and arg Gr (iω ) =
−90○ . The slope of the gain curve increases to 0 at the corner frequency
ω 1 = 1.
The high frequency asymptote is Gr (s) ( 1 with pGr (iω )p = 1, i.e.
slope = 0 and arg Gr (iω ) = 0. The corresponding Bode plot is shown
in Figure 6.1.

Figure 6.1 Bode plot of a PI controller with K = 1 and Ti = 1.

b. The transfer function of the controller is

Gr (s) = K (1 + Td s) = 1 + s

The low frequency asymptote becomes Gr (s) ( 1, i.e. the gain curve is
a straight line with magnitude 1 and slope = 0 and the phase curve
is described by arg Gr (iω ) = 0○ . The slope of the gain curve increased
to +1 at the corner frequency ω 1 = 1.
The high frequency asymptote is given by Gr (s) ( s, i.e. the slope of
the gain curve is +1 and the phase curve is described by arg Gr (iω ) =
+90○ . The corresponding Bode plot is shown in Figure 6.2.

6.6 With the Ziegler-Nichols frequency method, an oscillation is generated

at the frequency ω 0 , where the phase shift of the process is −180○ .
This frequency is given by

−2 arctan 20ω 0 − 9ω 0 = −π

Numerical solution gives ω 0 ( 0.1 rad/s.

108
 Solutions to chapter 6. Design methods

Figure 6.2 Bode plot of a PD controller with K = 1 and Td = 1.

The process gain at ω 0 is

1
= 0.2
1 + 202ω 20

The critical gain K c becomes

1
Kc = =5
0.2
and the period time is
2π
To = = 63
ωo
The controller parameters become
I
K = 0.45 K c = 2.25
Ti = To /1.2 = 53

6.7 a. Laplace transform of Martins dynamics give

1 1
sX = − X+ U
30 15
Y =X

and the transfer function Y = G P U becomes

2
GP =
30s + 1

The process is controlled by a PI controller, U = G R ( R − Y ) where

1 Ti s + 1
G R = K (1 + )= K
Ti s Ti s

109
Solutions to chapter 6. Design methods

The controller parameters K and Ti should be chosen such that the

closed loop transfer function Y = Gcl R has its poles in −0.1. The
transfer function from R to Y is

GP GR K TiTsi+s 1 30s2+1 K (Ti s + 1)2

Gcl = = Ti s+1 2
=
1 + GPGR 1 + K Ti s 30s+1 Ti s(30s + 1) + K (Ti s + 1)2

where the denominator polynomial has the same roots as

2K + 1 K
s2 + s+
30 Ti 15

These roots should be −0.1, i.e., the same as for

(s + 0.1)2 = s2 + 0.2s + 0.01

This is true if the coefficients for the polynomials are equal, i.e.,

2K + 1
= 0.2
30
K
= 0.01
Ti 15

The controller parameters thus become

K = 2.5
Ti = 16.7

b. When using Ziegler-Nichols frequency method, an experiment is per-

formed using feedback with a P controller. The gain is increased until
the system starts oscillating (stability boundary). The Bode diagram
of the process can be thought of as the Bode diagram for the loop
transfer function, when the controller is a P controller with a gain of
1. When the gain is increased, the gain curve in the Bode diagram will
move upwards, whereas the phase will remain unaffected. The system
will start to oscillate when the phase margin becomes 0. In the figure
it can be seen that this will happen when the cutoff frequency is 0.3
rad/s, since the phase curve crosses −180○ at this frequency. This is
the resonant frequency, with which the system will oscillate. For this
frequency, the gain is equal to 0.2. Thus, a gain of 5 is needed in
order for the system to start oscillating with the resonant frequency.
According to the collection of formulae, we have

K = 0.45 K 0
Ti = T0 /1.2

where K 0 is the gain and T0 the period time of the oscillation. With

K0 = 5
2π
T0 = = 20.9
0.3
we get the controller parameters K = 2.25 and Ti = 17.4.

110
 Solutions to chapter 6. Design methods

6.8 Ziegler-Nichols step response method: With the customary notion we

obtain a = 0.3 and b = 0.8. The controller parameters become K =
1.2/ a = 4, Ti = 2b = 1.6 and Td = b/2 = 0.4.
Ziegler-Nichols frequency method: The Nyquist curve intersects the
negative real axis in −1/3 for ω = 1 rad/s, which yields To = 2π /ω =
2π and K c = 3. The controller parameters become K = 0.6 K c = 1.8,
Ti = To /2 = 3.1 and Td = To /8 = 0.8.
The Lambda method: By drawing the tangent to the step response,
an approximation of the dead time is obtained, L ( 0.8s. The step
response has reached 63% of its final value after approximately 2.6s.
The time constant thus becomes T = 2.6 − 0.8 = 1.8s. The static gain
is K P = 1. With λ = T, we get the PI controller parameters

1 T
K= ( 0.7
Kp L + λ
Ti = T = 1.8s

For a PID controller, we get the parameters

1 L/2 + T
K= =1
K p L/2 + λ
Ti = T + L/2 = 2.2s
TL
Td = = 0.33s
L + 2T
6.9 a. The step response of the system is shown in Figure 6.3.

Figure 6.3 Step response of G(s) = e−s /(s + 1).

From the figure we obtain (with the customary notion) a = b = 1.

This yields the controller parameters K = 1.2/ a = 1.2, Ti = 2b = 2
and Td = b/2 = 0.5.

b. The resonance frequency is determined by arg G(iω 0 ) = − arctan ω 0 −

ω 0 = −π. Numerical solution yields ω 0 ( 2.03, resulting in T =
2π /ω 0 = 3.1.
Further, K 0 = 1/pG(iω 0 )p = 2.26, which gives K = 1.4, Ti = 1.5 and
Td = 0.39.

111
Solutions to chapter 6. Design methods

c. From the step response in subproblem a, we get the process parame-

ters K P = 1, L = 1 and T = 1. This gives the controller parameters

1 L/2 + T
K= =1
K p L/2 + λ
Ti = T + L/2 = 1.5
TL
Td = ( 0.33
L + 2T

6.10a. The figure does not allow for any greater precision. Draw the tangent
of the step response where the derivative attains a maximum and
study the intersection of the tangent and the two coordinate axis. The
parameter a is given by the distance between 0 and the intersection
with the vertical axis, whereas b is given by the distance between 0
and the intersection with the horizontal axis. In our example we have
a = 0.65 and b = 4. From the table we obtain the following controller
parameters: K = 1.9, Ti = 8 and Td = 2.

b. The critical gain K c is the gain which causes the Nyquist curve to
pass through -1. In our case we have K c = 1/0.55 = 1.8. The critical
period T0 corresponds to the frequency at ’o’, i.e. T0 = 2π /ω = 14.6.
This yields the controller parameters: K = 1.1, Ti = 7.3 and Td = 1.8.

c. The value of K obtained from the last method is smaller than the
values obtained through Ziegler-Nichol’s methods.

6.11 Generally it is required that pG K (iω c )p > 1 in order for ω c to increase.

A The speed of the system increases, but simultaneously its robust-
ness is reduced since the phase margin decreases.
B pG K p < 1 for all ω, resulting in decreased cross-over frequency and
speed.
C Cf. B.
D pG K p = 1 for all ω, leaving the cross-over frequency unaffected.

6.12 The process is connected in a feedback loop with a proportional con-

troller. By adding a compensation link one wants to decrease the
ramp error of the compensated system by a factor 10. Simultaneously,
a small decrease in robustness (phase margin) is accepted, resulting
in a certain decrease of the system’s transient behavior.
We can affect the ramp error by introducing a phase lag compensation
link
s+a
G k (s) = M
sM + a
Assuming that the system is stable, the resulting ramp error becomes

1 1
lim e( t) = lim s E (s) = lim s · ·
t→∞ s→0 s→01 + G k (s)G P (s) s2
(sM + a)(s + 1)(s + 2) 2
= lim =
s→0 s(sM + a)(s + 1)(s + 2) + K M (s + a) KM

112
 Solutions to chapter 6. Design methods

By choosing M = 10 ( K = 1) the ramp error is reduced to 0.2.

Now it remains to decide a value for a. The phase lag link contributes
to a phase lag in √ the open loop. The phase lag is largest around the
frequency ω = a/ M. In order not to compound the transient behav-
ior of the closed loop system excessively, a must be chosen such that
the phase around the cross-over frequency is left unaffected. This can
be achieved by choosing a adequately small. However, a overly small
value of a results in a long time before the ramp error decreases to 0.2.
Let ω c denote the cross-over frequency of the uncompensated system.
At this frequency, the compensation link has a phase contribution

ωc Mω c
arg G k (iω c ) = arctan − arctan
a a

A simple rule of thumb is to choose a = 0.1ω c . In our example it

means that the compensation link contributes with a phase shift of

arg G k (iω c ) = arctan 10 − arctan 100 ( −5.1○

The crossover frequency may be determined numerically according to

1
pG P (iω c )p = p p =1
ωc 1 + ω 2c 4 + ω 2c

which gives ω c ( 0.45-

The compensation link thus becomes

s + 0.045 s + 0.045
G k (s) = 10 =
10s + 0.045 s + 0.0045

Figure 6.4 Bode plot of the uncompensated open loop system (solid line) and com-
pensated open loop system (dash-dotted line) in assignment 6.12. K = 1 for both
cases.

113
Solutions to chapter 6. Design methods

In Figure 6.4 Bode plots for both the uncompensated open loop sys-
tem KG P (s) and the compensated open loop system KG k (s)G P (s) are
shown.
The compensation link alters the transient behavior of the system.
Figure 6.5 shows how the overshoot of the step response has increased,
compared to the uncompensated system. Also the settling time has
increased, partly due to the slow mode in the compensation link.
The purpose of introducing the compensation link was to decrease
1.5

0.5

0
0 10 20 30 40 50

Figure 6.5 Step responses of the uncompensated closed loop system (solid line) and
the compensated closed loop system (dashed line) in assignment 6.12.

the ramp error. Figure 6.6 shows the error e = r − y for both the
uncompensated and compensated systems, with r = t.
As seen from the figure, the compensated system fulfills the criterion
of a ramp error less than 0.2.
3

0
0 20 40 60 80 100

Figure 6.6 Ramp error of the uncompensated system (solid line) as well as the
compensated system (dashed line) in assignment 6.12.

6.13 Use a phase lag compensation link

s+b
G k (s) = K N
s + bN

114
 Solutions to chapter 6. Design methods

The crossover frequency ω c of the uncompensated system can be read

from the Bode plot in Figure 6.7. One can also determine ω c from the

Figure 6.7 Bode plot of the uncompensated open loop system (solid line) as well as
the compensated open loop system (dash-dotted line) in assignment 6.13.

equation
1.1
pG P (iω c )p = p =1
ωc ω 2c + 1
This yields ω c = 0.84. The new cross-over frequency is chosen to
be ω ∗c = 1.68. The phase shift of the uncompensated system at the
frequency ω c is

arg G P (iω c ) = −90o − arctan(0.84) = −130o

In order not to decrease the phase margin, it must hold that

arg(G k (iω ∗c )G P (iω ∗c )) ≥ arg G P (iω c )

We have

arg G P (iω ∗c ) = −90o − arctan(1.68) = −149o

For the compensation link it must hence hold that

arg G k (iω ∗c ) ≥ 19o

From the collection of formulae we find that N = 2 is adequate.√The

compensation link has its maximal phase shift at the frequency b N.
This shall occur at the new cross-over frequency, i.e.

√ ω∗
ω ∗c = b N [ b = √ c = 1.2
N

115
Solutions to chapter 6. Design methods

Now choose K such that ω ∗c √ becomes the actual cross-over frequency

(observe that pG k (iω ∗c )p = K N)

pG k (iω ∗c )G P (iω ∗c )p = 1 [ K = 2.1

We thus obtain the compensation link

s + 1.2
G k (s) = 4.2
s + 2.4

Figure 6.7 shows the Bode plot of the uncompensated open loop sys-
tem G P (s) as well as the compensated open loop system G k (s)G P (s).
Figure 6.8 shows the step responses of the uncompensated and com-
pensated systems.
1.5

0.5

0
0 5 10 15 20

Figure 6.8 Step response of the uncompensated closed loop system (solid line) as
well as the compensated closed loop system (dashed line) in assignment 6.13.

6.14 We choose a phase lead link

s+b
G k (s) = K K · N
s + bN

The specification implies that the low frequency gain shall not de-
crease (which would increase the stationary error). The cross-over
frequency shall increase by a factor 3 and the phase margin shall
remain unchanged.
The open loop transfer function is

s+b 1
G0 (s) = G k (s)G1 (s) = K K · N ·
s + bN s(s + 1)(s + 2)

The Bode plot of G1 is presented in Figure 6.9. From this, or from

numerical calculations, the cross-over frequency is determined to ω c =
0.45 rad/s and the phase margin is φ m = 53○ . The new cross-over
frequency shall thus be ω ∗c = 3 · ω c = 1.35 rad/s with unchanged
phase margin. Since arg G1 (iω ∗c ) ( −180○ , the phase curve must be
raised approximately 50○ by G k .

116
 Solutions to chapter 6. Design methods

Figure 6.9 Bode plot of the uncompensated system G1 (solid line) and compensated
system G k G1 (dash-dotted line) in assignment 6.14.

From the collection of formulae it is obtained that N = 8 gives a

maximal phase lead√of approximately 50○ . The phase lead is maximal
at the frequency b N = ω ∗c , yielding b = 0.48. The gain shall be
unity at the new cross-over frequency ω ∗c , i.e.

pG k (iω ∗c )p · pG1 (iω ∗c )p = 1

√
The magnitude of the compensator is pG k (iω ∗c )p = K K N. Numerical
calculations give pG1 (iω ∗c )p = 0.183. Hence

1
KK = √ = 1.9
N · 0.183

The step response of the uncompensated and compensated systems,

respectively, are shown in Figure 6.10 and the ramp response is shown
in Figure 6.11. Since K K > 1 the stationary errors become smaller
than before, thus fulfilling the specifications.

6.15 From the Bode plot of Go (s) (see Figure 6.12) we obtain φ m = 20○
and ω c = 0.7 rad/s. Unchanged speed necessitates a compensation
link which does not affect the cross-over frequency. We hence need a
phase lead of ∆φ = 30○ at ω = ω c = 0.7 rad/s. We utilize a phase lead
compensation link
s+b
G k (s) = K N
s + bN
1. The sample curves in the collection of formulae yield N = 3.
√ 0.7
2. b N = ω c [ b= √
3
= 0.40
√ 1
3. pG k (iω c )Go (iω c )p = K N · 1 gives K = √ = 0.58
N

117
Solutions to chapter 6. Design methods

1.5

0.5

0
0 5 10 15 20

Figure 6.10 The step response of the uncompensated closed loop system (solid line)
as well as the compensated closed loop system (dashed line) in assignment 6.14.

10

0
0 2 4 6 8 10

Figure 6.11 The ramp response of the uncompensated system (solid line) as well
as the compensated system (dashed line) in assignment 6.14.

The compensation link thus becomes

s + 0.4
G k (s) = 0.58 · 3
s + 1.2

The system is stable, so the resulting stationary error becomes

1 s(s + 0.5)(s + 3)(s + bN )

E (s) = R(s) = R(s)
1 + G k Go s(s + 0.5)(s + 3)(s + bN ) + 2 K N (s + b)

With R(s) = 1/s2 the stationary ramp error becomes

1.5
lim s E (s) = = 1.29
s→0 2K

which fulfills the specification. Figure 6.13 shows the step response
of the system before and after the compensation. The ramp error is
shown in Figure 6.14. The fact that the ramp error is increased by the
compensation is due to K < 1.

118
 Solutions to chapter 6. Design methods

Figure 6.12 Bode plot of Go (s) in assignment 6.15.

2 y(t)

1.5

0.5

0
0 10 20 30

Figure 6.13 Step response of the uncompensated closed loop system (solid line) and
compensated system (dashed line) in assignment 6.15.

6.16 We know that a phase lag compensation link dimensioned according to

the rules of thumb will decrease the phase margin by approximately
6○ , which yields a certain decrease of robustness. In order not to obtain
an excessive overshoot, we start by decreasing the gain of the process
in order to increase the phase margin.
From the Bode plot of the process (see Figure 6.15) we find that G − 1
has a phase shift of −133○ at the cross-over frequency ω c = 0.7. At
the frequency ω ∗c = 0.6 we have the phase shift −133○ + 6○ = −127○
and the gain pG1 (ω ∗c )p = 1.2.
By decreasing the open loop gain by a factor 1.2 we obtain the new
cross-over frequency ω ∗c and a phase margin increase of 6○ . Since
we cannot affect the process gain directly, we equivalently let K =

119
Solutions to chapter 6. Design methods

2 ramp error e(t)

1.5

0.5

0
0 10 20 30

Figure 6.14 Ramp error of the uncompensated closed loop system (solid line) as
well as the compensated closed loop system (dashed line) in assignment 6.15.

Figure 6.15 Bode plot of the uncompensated open loop system (solid line) as well
as the compensated open loop system (dash-dotted line) in assignment 6.16.

1/1.2 = 0.83 in the compensation link.

The main problem is to decrease the stationary ramp error to e 1 ≤ 0.1.
The final value theorem gives
1
e(∞) = lim sU (s) =
s→0 1 + G k (s)G1 (s)
1 (s + a/ M )s(s2 + 2s + 2) 2
= lim s 2 2
= ≤ 0.1
s→0 s (s + a/ M )s(s + 2s + 2) + 1.5 K (s + a) 1.5 K M

which yields M ≥ 16. Choose M = 16. According to the rule of thumb

we let a = 0.1ω ∗c = 0.06. The chosen compensation link thus becomes

s + 0.06
G k (s) = 0.83
s + 0.00375

120
 Solutions to chapter 6. Design methods

Figure 6.16 shows the step response before and after the compensa-
tion. The ramp errors of the uncompensated closed loop system and
1.5

0.5

0
0 10 20 30

Figure 6.16 Step response of the uncompensated closed loop system (solid line) as
well as the compensated closed loop system (dashed line) in assignment 6.16.

the compensated closed loop system are shown in Figure 6.17.

3

0
0 20 40 60 80

Figure 6.17 Ramp error of the uncompensated closed loop system (solid line) as
well as the compensated closed loop system (dashed line) in assignment 6.16.

Comment:
Since we have decreased the open loop gain we obtain a decreased
cross-over frequency and hence a somewhat slower system. In Fig-
ure 6.16 one especially notes the slow mode which appears as the
process settles. It is caused by the slow pole of the controller in com-
bination with the low gain. The rise time and damping are, however,
virtually unaffected. An alternative to decreasing the open loop gain,
in order to maintain the desired phase margin, would be to introduce
a phase lead compensation link.

121
Solutions to Chapter 7. Controller
Structures

7.1 The disturbance d does obviously lack influence if

1
G1 (s) H (s) + 1 = 0 \ H (s) = −
G1 (s)

To be a practically useful control law it is required that the distur-

bance can be measured, that the model G1 (s) of the heating system
is a "good" description of reality and that the inverse transfer func-
tion 1/G1 (s) is practically realizable. This means that H (s) must not
contain derivatives of the signal d. The realization of H (s) can also
be problematic if G1 (s) lacks a stable inverse (i.e. if G1 (s) has right
half plane zeros, which is equivalent to being a non-minimum phase
system). Further, we cannot invert processes with low pass character-
istics more than at low frequencies and delays can obviously not be
inverted.

7.2 A block diagram for the system is shown in Figure 7.1. Mass balance

feedforward
v
G
F

−1

h P controller valve tank h

ref x
Σ K Σ G Σ G
V T

−1

Figure 7.1 Block diagram of the level controlling system in assignment 7.2.

for the tank yields

dh
A = x( t) − v( t)
dt
Laplace transformation gives ( A = 1 m2 )

1
H (s) = ( X (s) − V (s))
s

The transfer function of the tank is thus

1
GT (s) =
s

a. The closed loop transfer function becomes

GT G V K K
G(s) = =
1 + GT G V K 0.5s2 + s + K

122
 Solutions to chapter 7. Controller Structures

The characteristic polynomial is hence

s2 + 2s + 2 K

The desired characteristic polynomial is

(s + ω )2 = s2 + 2ωs + ω 2

Identification of coefficients yields


ω = 1
K = 1
2

The transfer function form v( t) to h( t) is given by

GT 1 + 0.5s
H (s) = − V (s) = − V (s)
1 + GT G V K s(1 + 0.5s) + K

If v( t) is a step of amplitude 0.1 we obtain V (s) = 0.1/s. The final

value theorem gives

0.1
h(∞) = lim s H (s) = −
s→0 K

The theorem may be used since s H (s) is of second order with positive
coefficients in the denominator.

b. A PI controller has the transfer function

1
G R (s) = K (1 + )
sTi

The closed loop transfer function becomes

GT G V G R K (1 + sTi )
G(s) = =
1 + GT G V G R s(1 + 0.5s)sTi + K (1 + sTi )

The characteristic polynomial becomes

2K
s3 + 2s2 + 2 K s +
Ti

The desired characteristic polynomial is

(s + ω )3 = s3 + 3ωs2 + 3ω 2 s + ω 3

Identification of coefficients yields


2

ω=


 3


2
K=

 3


T = 9


i
2

123
Solutions to chapter 7. Controller Structures

c. The relation between the flow disturbance v and the level h is given
by
GT (G V G F − 1)
H (s) = V (s)
1 + GT G V G R
To eliminate the influence of v, we choose

1
G F (s) = = 1 + 0.5s
GV

Note that G F is not realizable. One can either cancel the derivative
term or add a low-pass filter so that the infinite gain at high frequen-
cies is avoided.

7.3 The closed loop system has the transfer function

(G R + K f )G P ( K + K f )s + K /Ti
= 2
1 + GP GR s + (3 + K )s + K /Ti

a. The characteristic equation of the closed loop system is

s2 + (3 + K )s + K /Ti = 0

The desired characteristic equation is

(s + 2 − 2i)(s + 2 + 2i) = s2 + 4s + 8 = 0

Identification of coefficients yields K = 1 and Ti = 1/8.

b. The feedforward K f affects the zeros of the closed loop system, but
leaves the poles unaffected. The poles can be placed by means of the
controller H in order to obtain adequate supression of disturbances,
cf. sub-assignment a above. One can subsequently translate the zeros
by means of K f in order to e.g. reach a desired overshoot in the refer-
ence step responses. The zero of the closed loop system is eliminated
by choosing K f = − K . With the pole placement in√sub-assignment a,
which corresponds to a relative damping ζ = 1/ 2 ( 0.7, the over-
shoot of the closed loop system becomes approximately 5%.

7.4 The block diagram in assignment 7.3 can be re-drawn according to

Figure 7.2. By comparing to the block diagram in assignment 7.3 we
see that H f f = H + K f and H f b = H. Observe that manipulation of
K f offers the possibility to neutralize the derivation in H, i.e. achieve
a controller which derivates the output, but not the reference value.

7.5 The system has three inputs: the reference yr and the two disturbances
v1 and v2 . The transfer functions between these three signals and the
output y are given by

G1 G2 G R1 G R2 G1 G2
Y = Yr + V1
1 + G1 G R1 + G1 G2 G R1 G R2 1 + G1 G R1 + G1 G2 G R1 G R2
(1 + G1 G R1 )G2
+ V2
1 + G1 G R1 + G1 G2 G R1 G R2

124
 Solutions to chapter 7. Controller Structures

Kf

Yr H + + G(s)

H –1

Figure 7.2 Modified block diagram in assignment 7.3.

Let us call the three transfer functions G yr , Gv1 and Gv2 , respectively.
Ideally we would have G yr = 1 and Gv1 = Gv2 = 0 for all frequencies.
This is, however, not achievable. Nonetheless, we can assure that
it holds in stationarity, i.e. for s = 0. For a P controller we have
G R (0) = K , where K is the gain of the controller. For a PI controller
it holds that G R (0) = ∞.
The transfer function G yr becomes unity if G R2 is a PI controller. The
transfer function Gv1 becomes 0 if any of the controllers are PI. The
transfer function Gv2 , however, is only zero if G R2 is a PI controller.
Consequently G R2 must contain an integral part in order to guarantee
0 stationary control error. The controller G R1 can then be chosen to
be a P controller. (If we furthermore want the internal signal y1 to
coincide with its reference, also this controller would need an integral
part.)

7.6 a. The closed loop transfer function is given by

K 1 G1 (s) 2 K1
Ginner (s) = =
1 + K 1 G1 (s) s + 2 + 2 K1

In order to make the system 5 times as fast, the pole of the closed
loop system must be placed in s = −10, calling for K 1 = 4.

b. The approximation Ginner(s) ( Ginner(0) = 0.8 yields

G R2 (s)G2 (s)Ginner (0) ( K 2 s + KTi2 )0.8

Gouter (s) = =
1 + G R2 (s)G2 (s)Ginner (0) s2 + 0.8 K 2 s + 0.8 KTi2

The specification of a system 10 times slower than the inner loop calls
for a pole in s = −1. Since we deal with a second order system, we
choose to locate both poles in s = −1 (somewhat slower than the single
pole case). This yields K 2 = 2.5 and Ti = 2.

125
Solutions to chapter 7. Controller Structures

Comment. A general rule when cascading controllers is to make the inner

loop 5–10 times faster than the outer loop in order to enable separation of
the controller calculations for the two loops. The actual closed loop system
(without approximations) becomes

10(2s + 1)
Gouter (s) =
s3 + 10s2 + 20s + 10

and has poles in approximately −7.516, −1.702 and −0.7815 where the
slower pole (s = −0.7815) will be the one essentially determining the speed
of the system. This corresponds fairly well to the specified speed.

7.7 a. Since the steam flow is assumed to be constant, we can let F = 0,

which yields the following description of the dome

10−3
Y (s) = M (s)
s

Since the controller is of P type we have M (s) = K ( Yr − Y ), where Yr

denotes the reference dome level. This yields

K
Y (s) = Yr (s)
K + 103 s

Since the system is linear and subject to negative feedback, a step

disturbance in the level gives rise to the same transient behavior
as a step disturbance in the reference. Hence let Yr (s) = 1s . Inverse
transformation of Y (s) yields
−3 t
y( t) = 1 − e− K10

The specification on the settling time of the system now yields

−2
y(10) = 1 − e− K10 = 0.9 [ K = 230

b. The dome and P controller are described by

K s − 0.01
Y (s) = Yr (s) + F (s)
K + 1000s (s + 0.1)(1000s + K )

1
Let Yr (s) = 0. A step disturbance in the steam flow F (s) = s thus
gives
s − 0.01 1
Y (s) =
(s + 0.1)(1000s + K ) s
The final value theorem yields

s − 0.01 1 −0.1
lim y( t) = lim sY (s) = lim s =
t→∞ s→0 s→0 (s + 0.1)(1000s + K ) s K

and stationary error becomes

0.1
e = yr − y = − y =
K

126
 Solutions to chapter 7. Controller Structures

c. Determine a feedforward link H (s) from steam flow F (s) to feed water
flow M (s) for the initial system, such that the level Y (s) becomes
independent of changes in the steam flow.
The system with feedforward H (s) F (s) is described by

10−3 s − 0.01
Y (s) = ( M (s) + H (s) F (s)) + 10−3 F (s)
s s(s + 0.1)
10−3 10−3 s − 0.01
3 4
= M (s) + + H (s) F (s)
s s s + 0.01

We want the influence from F (s) to be zero. Therefore choose H (s)

so that the expression in front of F (s) becomes zero. This criterion is
fulfilled when
s − 0.01
H (s) = −
s + 0.1
which gives the desired feedforward.

7.8 The delay margin is given by

φm
Lm =
ωc

First we compute the cross-over frequency ω c as

2 2
pG0 (iω c )p = pG P (iω c )G R (iω c )p = = p "1
iω c (iω c + 1) ω c ω 2c + 1
s
√
−1 + 17
\ ω 4c + ω 2c −4=0 \ ωc = = 1.25
2
Then we calculate the the phase margin φ m

π
φ m = π + arg G0 (iω c ) = π − − arctan ω c = 0.675
2

We thus obtain L m = φ m /ω c = 0.54.

7.9 a. The one second delay e−s is considered part of the process.

Controller G R (s) = K
1
Process G P (s) = e−s
s(s + 1)
1
Model Ĝ P (s) = G P (s) = e−s
s(s + 1)
1
Model without delay Ĝ0P (s) =
s(s + 1)

b. According to the block diagram the control signal is given by

1 2
U (s) = G R (s) E (s) + Ĝ P (s)U (s) − Ĝ0P (s)U (s)

127
Solutions to chapter 7. Controller Structures

1
10

Magnitude 0
10

−1
10
−2 −1 0 1 2
10 10 10 10 10

60

50

40
Phase

30

20

10

−10
−2 −1 0 1 2
10 10 10 10 10
Frequency [rad/s]

Figure 7.3 Bode plot of the Smith predictor.

The transfer function of the controller becomes

G R (s)
U (s) = E (s)
1 − G R (s)Ĝ P (s) + G R (s)Ĝ0P (s)
2 2s(s + 1)
= E (s) = E (s)
2 2 s(s + 1) + 2 − 2e−s
1− −
e +s
s(s + 1) s(s + 1)

The Bode plot of the controller is shown in Figure 7.3. One notes
that the Smith predictor gives a large phase lead at the cross-over
frequency of the initial system.

c.

2s(s + 1) 2s(s + 1)
U (s) = E (s) ( E (s)
s(s + 1) + 2 − 2e− s s(s + 1) + 2 − 2(1 − s)
2(s + 1)
= E (s)
s+3

This is a phase lead link with N = 3.

7.10 The gain curve of the system is given by

k
pG(iω )p =
ω

It is sufficient to read the value of the gain curve at a single frequency

in order to determine k. The gain is e.g. 1 at approximately ω = 4.5.
This yields
k
1= \ k = 4.5
4.5

128
 Solutions to chapter 7. Controller Structures

The phase curve of the system is given by

arg G(iω ) = −π /2 − ω L

Now, it is sufficient to read the value of the phase curve at a single fre-
quency in order to determine L. The phase is e.g. −π at approximately
ω = 120. This yields

−π = −π /2 − 120L \ L = 0.013

129
Solutions to Chapter 8. Design Examples

8.1 a. The phase curve for v = 3 knots cuts −180○ at ω o ( 0.03 rad/s. At this
frequency we have pG(i0.03)p ( 2. The gain K must hence be smaller
than 0.5 in order to yield a stable closed loop system.

b. In order to acquire the cross-over frequency ω c and phase margin φ m

it is required that
I
pGr (iω c )G(iω c )p = 1
arg Gr (iω c )G(iω c ) = φ m − 180○

where Gr (s) = K (1 + TD s). This leads to the equations

 q
 2 2
K pG(iω c )p 1 + TDωc = 1
arg G(iω ) + arctan T ω = φ − 180○
c D c m

With ω c = 0.03 rad/s, φ m = 60○ , pG(iω c )p ( 2 and arg G(iω c ) ( −180○

we obtain
 √
 tan 60○ 3
Td = 0.03 = 0.03 ( 57.7


1 1

 K= q ( = 0.25
 2 ·2
pG(iω c )p 1 + Td2ω 2c


c. If the speed suddenly increases from 3 to 7 knots, we have to turn

to the dotted Bode plots in Figure 8.2. The most drastic change is
that the gain curve has been raised by a factor 20. Additionally,
the phase curve has decreased for frequencies above 0.03 rad/s. This
results in heavily reduced phase- and gain margins. A more thorough
examination shows that this in fact leads to instability of the closed
loop system. This can be seen in the Bode plot in Figure 8.2, which
shows both the nominal case v = 3 knots and the case v = 7 knots.
One way to avoid this problem is to instead choose v = 7 knots as the
nominal case for the calculation of the PD controller. This, however,
means that one has to accept a slower settling time for the slowest
speed v = 3 knots. A better way is to let K and Td depend on the
speed v. This method is known as gain scheduling.

d. The transfer function from β to h can be approximated by

kv v
G hβ (s) =
s3

From the Bode plot one sees that pG hβ (i · 0.1)p ( 0.04 for v = 3 knots
= 3 · 1.852/3.6 ( 0.5144 · 3 m/s, which yields

0.13 · 0.04
kv ( ( 2.6 · 10−5
3 · 0.5144

130
 Solutions to chapter 8. Design Examples

10 2

Magnitude
10 1

10 0

10 −1

10 −2

10 −3
10 −3 10 −2 10 −1 10 0

Phase [deg]
−50
-50

−100
-100

−150
-150

−200
-200
10 −3 10 −2 10 −1 10 0
Frequency [rad/s]

Figure 8.1 Bode plot of the PD compensated open loop system in assignment 8.1.
The solid curves show the case v = 7 knots. Note that the latter case yields an
unstable closed loop system.

e. The characteristic equation of the closed-loop is given by

s3 + K k v v = 0

Since not all coefficients are positive, the closed-loop system is not
asymptotically stable for any value of K . In sub-assignment a it was
concluded through the measured frequency response that the closed
loop system was stable for K < 0.5. The explanation to this ap-
parent contradiction is found in the Bode plot which was used in
sub-assignment a: The approximation only holds for high frequencies
(ω > 0.05). For low gains, such as K < 0.5, the cross-over frequency
ω c < 0.03 lies outside the valid range of the model.
For e.g. ω < 0.03 the Bode plot shows a phase above −180○ while the
simplified model features the phase −270○ for all frequencies.

f. If x = (α̇, α, h)T and u = β we obtain the state space equations

   
 0 0 0  kv 

 
 
 

ẋ = 

 1 0 0

 x+

 0 u
    
0 v 0 0

With the state feedback u = − Lx + ur the characteristic polynomial

of the closed loop system becomes

p(s) = det(s I − A + BL) = s3 + k v l 1 s2 + k v l 2 s + k v vl 3

the desired characteristic polynomial is

p(s) = (s+γω 0 )(s2 +2ζω 0 s+ω 20 ) = s3 +(γ +2ζ )ω 0 s2 +(2γζ +1)ω 20 s+γω 30

131
Solutions to chapter 8. Design Examples

Direct comparison gives


 (γ + 2ζ )ω 0

 l1 =


 kv


(2γζ + 1)ω 20

l2 =

 kv


γω 30



l =

3
kv v

g. Here stationarity means constant height, h = h ref . This in turns mean

that all derivatives of h must be zero, i.e. α = 0 and α̇ = 0. When
the height has reached its correct value the control signal u = β must
also be zero since the submarine would otherwise continue to rise.
Thus L r is obtained from the equation

0 = L r h ref − l 1 · 0 − l 2 · 0 − l 3 h ref

For v = 3 knots, we end up with the following result

γω 30 γω 30 · γω 30 ·
Lr = l3 = ( (
kv v 2.6 · 10−5 · 3 · 0.5144 4.0 · 10−5

h. At a momentary disturbance ∆h = 0.1 m the rudder angle becomes

γω 30 0.2ω 30
∆β = l 3 · ∆h = · 0.1 (
vk v 3 · 0.5144 · 2.6 · 10−5
Since ∆β ≤ 5○ , we must have
4 31
5 · 3 · 0.5144 · 2.6 · 10−5
3
ω0 ≤ ( 0.1
0.2

8.2 a. The oscillation frequency ω o ( 27 rad/s and critical gain K c ( 3.6

can be read from the Bode plot. The oscillation period is hence To =
2π /ω o ( 0.23. This yields the PID parameters K = 0.6 K c ( 2.2,
Ti = To /2 ( 0.12 and Td = To /8 ( 0.03. The step response of the
closed loop system is shown in Figure 8.2. The specifications are
apparently not fulfilled. A PID controller (with filter factor) can be
considered a second order controller with integral action. As a matter
of fact, the specifications can be met, using a more general second
order controller with integral action (see Figure 8.3). If one tries to
interpret it as a PID controller, one would end up with a negative
derivative time Td .

b. In stationarity all derivatives of the states must be zero ẋ = 0. It

hence holds that
0 = Axo + Buo = ( A − BL) xo + BL r yr
I

yr = yo = C xo

This yields
1
Lr = −
C ( A − BL)−1 B

132
 Solutions to chapter 8. Design Examples

1.6

1.4

1.2

0.8

0.6

0.4

0.2

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Figure 8.2 The step response with PID control according to Ziegler-Nichols.

1.6

1.4

1.2

0.8

0.6

0.4

0.2

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Figure 8.3 The step response of the closed loop system with a second order inte-
grating controller.

c. With x augmented to x e = ( x1 , x2 , x3 , xi )T we obtain

 d1 +d f df k 

 − J1 J1
− J1f 0

 km ki 
 J1 



0 

 
 
 
 
 

 df
 d f +d2
− J2
kf
0
 
 0 
 
 0 

ẋ e = 

 J2 J2 

 xe + 

 

 u+

 

 yr


 1 −1 0 0




 0 




 0 



 
    
0 kω 2 0 0 0 −1
 
y =  0 kω 2
0 0  xe

where the reference yr has been introduces as an extra input.

d. The approximate value of ω m becomes

ln 0.02
ωm ( − ( 20
0.5 · 0.38

133
Solutions to chapter 8. Design Examples

e. When it comes to load disturbances, the fast Kalman filter (ω o = 40)

has the best performance. It is also best when it comes to suppressing
the influence of of measurement noise. However, it is the worst choice
when it comes to suppressing the influence of noise in the control
signal. The two cases ω o = 10 and ω o = 20 feature approximately
the same noise sensitivity, while ω o = 10 is slower when it comes to
eliminating load disturbances. A satiable choice is thus ω o = 20.

134
Solutions to Chapter 9. Interactive
Comparison Between Model
Descriptions

9.1

a. The amplitude of the step response is affected, but not the time con-
stant. The pole is not affected. The Nyquist diagram keeps its shape,
but each point on the curve moves radially from the origin. In the
Bode diagram, the gain curve changes in the y direction while the
phase curve is unchanged.
Since the control signal is a unit step, K is given by the stationary
value of the measurement signal in the step response. In the Nyquist
diagram, K is given from the starting point on the positive real axis.
In the Bode diagram, K is given from the gain curve where ω → 0.

b. The amplitude of the step response is not affected, but the time con-
stant is given by T. The pole is in s = −1/T, which means that a large
time constant gives a pole close to the origin, whereas a short time
constant gives a pole far away from the origin. In the Bode diagram,
the corner frequencty is 1/T, and it thereby varies when T varies.
No change is visible in the Nyquist diagram, but the frequency varies
along the curve.
Suppose that we have two processes with different values of T. Then,
we can always find two frequencies such that

K K
G(iω 1 T1 ) = = G(iω 2 T2 ) =
1 + iω 1 T1 1 + iω 2 T2
i.e., all points that are on the first Nyquist curve are also on the
second one, but at another frequency.

c. A variation in L corresponds to a translational movement of the step

response. We can not represent a dead time in a singularity diagram.
The gain curve in the Bode diagram is not affected, since p e−iω L p = 1,
but the phase is reduced. For each frequency point in the Nyquist dia-
gram, the distance to the origin remains unchanged, but the phase de-
creases. Since the phase goes towards −∞ when ω → ∞, the Nyquist
curve has the spiral shape.

9.2

a. The changes are the same as in problem 1a.

b. The changes are analogous to the ones in problem 1b. When T1 ≫ T2 ,

the step response is similar to the one in problem 1a, with T ( T1 . The
Bode and Nyquist diagrams are also similar to the ones in problem
1a for low frequencies. Thus, you can in many cases approximate the
transfer function by

K K
G(s) = (
(1 + sT1 )(1 + sT2 ) (1 + sT1 )

135
Solutions to chapter 9. Interactive Comparison Between Model Descriptions

c. If the zero is far away from the origin, the representations are not
significantly affected. If the zero is negative and is close to the origin,
there is a large overshoot in the step response. If the zero is positive
and close to the origin, the step response will initially go in the wrong
direction. If the zero is positive it will give a positive contribution to
the phase.
When T3 < 0, e.g., when the zero is in the right half plane, the process
is hard to control. You can imagine that it is hard for a controller to act
in the right way when a control signal change makes the measurement
signal go in the wrong direction initially. The phenomenon could be
understood by writing the transfer function in the following way

K (1 + sT3 ) K s K T3
G(s) = = +
(1 + sT1 )(1 + sT2 ) (1 + sT1 )(1 + sT2 ) (1 + sT1 )(1 + sT2 )

Thus, the transfer function consists of two terms, one that is the
transfer function that we had in problem 2a, and one that is the same
transfer function, but multiplied with sT3 . Thus, the second term is
proportional to the derivative of the measurement signal we would
have obtained if we did not have any zero. If T3 < 0, this term will
give a negative contribution, which explains that the step response
initially goes in the wrong direction.

9.3

a. The frequency ω affects the speed of the system, but not the shape of
the step response. Variations in ω moves the poles radially from the
origin. In the Bode diagram, ω does not affect the shape, but only the
location of the corner frequency. The shape of the Nyquist curve is not
affected, but the frequencies are moved along the curve.

b. The relative damping ζ does not affect the speed of the step response,
but the shape. A small value of ζ gives an oscillatory and poorly
damped response. A small value of ζ gives a large resonance peak in
the Bode diagram. In the Nyquist diagram, you get a big increase of
gain and fast phase shift around ω.

136