Lecture 6 - Frequency Response Introduction

K. J. Åström • Dynamics is central to automatic control

• The special case of linear time-invariant systems
1. Introduction
• Two views on dynamics
2. The input-output view of Dynamical Systems
• State models, white boxes
3. Frequency response
dx dx
4. Nyquist curves and Bode plots = f ( x, u), = Ax + Bu,
dt dt
5. Summary
• Linear LTI systems
Theme: Review of dynamics. The input-output view of dy-
namical systems. The Table. Fourier’s idea: sinusoidal in- dx
= Ax + Bu, y = Cx + Du
puts. Graphical representations of frequency response. Nyquist dt
and Bode plots. The concepts of minimum and non-minimum
phase. • Input-output models, black boxes Y (s) = G (s) U (s)

Recall State Models Input-Output Models

• The Great Modelers: Tycho Brahe, Kepler and Newton • The heritage of electrical engineering
• To predict the future motion of the planets it is enough to • Fits the needs of block diagrams
know their current positions and velocities. • Makes it possible to deal with systems having a large
• The state is the least information about a system that is number of states.
required for the prediction of its future development. Bode: Electronic feedback amplifiers are much more
• Philosophical consequences, causality, predestination, complex than steam engines, systems have orders 50-
(chaos). 100 rather than 2-4. (A lot of capacitors!)
• A detailed description of the physics of a system, balances • Synonyms: input-output models, external descriptions,
of mass, momentum and energy. black boxes.
• The state is the least number of variables required to • Experimental determination of dynamics
describe storage of mass, momentum and energy.
• Synonyms: state models, internal descriptions, white
boxes.

c K. J. Åström September, 2000 1
 The Idea of Black Boxes Linear Time Invariant Systems
Input
System
Output Linearity: Let (u 1 , y1) och (u2 , y2) be input-output pairs.
Then (au1 + bu2, a y1 + a y2) is also an input-output pair,
superposition.
Consider a system as a black box. Forget about the internal
details and focus on the input-output behavior of the system. Time-invariance: Let δ t be an operator that shifts a signal
t time units forward and let (u, y) be an input-output pair. A
• Make a Giant Table over all pairs of inputs and outputs system is time-invariant if (δ t u, δ t y) is also an input-output
• A stroke of luck: A few entries suffice for linear time- pair.
invariant systems Consequences: The table can be simplified drastically for
• Steps (step response) linear time-invariant systems. It is enough to give one pair only.
• Impulses (impulse response) If all initial conditions are zero the output for all inputs is given
by the transfer function of the system.
• Sinusoids (Fourier, frequency responses)

t
y( t) = g( t − s)u( s) ds, Y ( s) = G ( s) U ( s)
0

Frequency Response Interpretation of Frequency Responses

• Fourier’s idea: An LTI system is completely determined by The complex number G (iω ) tells how a sinusoid propagates
its response to sinusoidal signals. through the system in steady state. If the input is u(t) = sin ω t,
• Implications for the Table then the output is
 
• Transmission of sinusoids given by G (iω ) y( t) =  G ( iω ) sin ω t + arg G ( iω )
• The jω -method
The number  G (iω ) is called gain ratio or simply gain and the
• Analytic continuation number arg G (iω ) is called phase of the transfer function.
• The transfer function G (s) uniquely given by its values on
the imaginary axis
• Experimental determination of the frequency response


c K. J. Åström September, 2000 2
 Notice Steady State Responses Proof
Output y Consider a system with transfer function G (s) having distinct
0.25

0.2 stable poles α k . Let the input be u(t) = e iω 0 t = cos ω 0 t +

0.15 1
i sin ω 0 t. The Laplace transform of the input is U (s) = .
s − iω 0
0.1

0.05

0
The Laplace transform of the output is
 Rk
−0.05
1 1 1
−0.1
0 5 10 15 Y ( s) = G ( s) = G ( iω 0 ) +
s − iω 0 s − iω 0 s − α k α k − iω 0
Input u
1
This corresponds to the time function
0.5
 Rk
y( t) = G ( iω 0 ) eiω 0 t + eα k t
0
α k − iω 0

Since all α k are negative the first terms go to zero and y(t) →
−0.5

−1
0 5 10 15 G (iω 0 ) eiω 0 t for large t.

The Nyquist Curve The Bode Plot

H. Nyquist was born in Sweden. Emigrated to the US and Bode was a researcher at Bell Laboratories. The complex
made his career at Bell Laboratories. The Nyquist curve function G (s) can also be represented by two graphs, one
represents the transfer function by showing a graph of the for the gain curve  G (iω ) and one for the phase curve. It is
complex number G (iω ) as a function of frequency. tradition to use logarithmic scales for frequency and gain and
linear scales for the phase. arg G (iω ). A nice consequence of
Physical interpretation!
this is that the curves have asymptotes that are very easy to
Im G(iω )
obtain!
Ultimate point

−1 ϕ Re G(iω )
a


c K. J. Åström September, 2000 3
 10
5
Gain Matlab
10
4
There are excellent tools in Matlab for drawing Nyquist curves
3
and Bode plots in the Control System Toolbox in Matlab.
10

10
2 NYQUIST Nyquist frequency response of LTI models.
BODE Bode frequency response of LTI models.
1
10
−2 −1 0 1 2
10 10 10 10 10
w Use the help function to figure out how the commands work.
Phase Try a few examples.
50 Notice that Matlab uses decibel (dB) as a standard unit for
0 amplitude, where 20 dB corresponds to a factor 10. If you think
−50 dB is an unnecessary complication it is easy to make your own
−100
plots with other units.
−150

−2 −1 0 1 2
10 10 10 10 10
w

Sketching Bode Plots Differentiator

Bode Diagrams

It is easy to sketch Bode plots because with the right scales From: U(1)
20

they have linear asymptotes. This is useful in order to get a 15

10

quick estimate of the behavior of a system. It is also a good 5

G ( s) = s
0

way to check numerical calculations.

Phase (deg); Magnitude (dB)

−5

−10

−15

Consider first a transfer function which is a polynomial G (s) = We have G (iω ) = iω −20

91

B (s)/ A(s). We have

log  G ( iω ) = log ω
90.5

To: Y(1)
log G ( s) = log B ( s) − log A( s)
90

arg G ( iω ) = π /2
89.5

Since a polynomial is a product of terms:

89
−1 0 1
10 10 10

Frequency (rad/sec)

Matlab:
s, s + a, s2 + 2ζ as + a2
sys=tf([1 0],1)
it suffices to be able to sketch Bode diagrams for these terms. bode(sys,{0.1,1})
The Bode plot of a complex system is then obtained by com-
position.

c K. J. Åström September, 2000 4
 Integrator First Order System G (s) = s + a
√
We have G (iω ) = a + iω , hence  G (iω ) = ω 2 + a2 and
Bode Diagrams

From: U(1)

arg G (iω ) = arctan ω /a, hence

20

15

10
1
G ( s) =
5

s
0
1
log  G ( iω ) = log (ω 2 + a2 ), arg G ( iω ) = arctan ω / a

Phase (deg); Magnitude (dB)

−5

−10
2
We have G (iω ) =
−15

−20

−i ω1 −89


−89.5

 log a if ω << a,
To: Y(1)
 √
log  G ( iω ) = − log ω −90
log  G ( iω ) log a + log 2 if ω = a, ,


arg G ( iω ) = −π /2
−90.5

log ω if ω >> a

−91
−1 0 1
10 10 10

Frequency (rad/sec) 
 0 if ω << a,
Matlab: 


π 1 ω
arg G ( iω ) + log if ω a, ,
sys=tf(1,[1 0])  4 2 a


bode(sys,{0.1,1}) 
 π
 if ω >> a
2

10
2
Gain Second Order System G (s) = s2 + 2ζ a + a2

1
G ( iω ) = a2 − ω 2 + 2iζ aω
10
1
log  G ( iω ) = log (ω 4 + 2a2ω 2 (2ζ 2 − 1) + a4 )
2
arg G ( iω ) = arctan 2ζ aω /( a 2 − ω 2 )
0
10
−2 −1 0 1 2

ω /a
10 10 10 10 10

Phase 

2 log a if ω << a,

2 log a + log 2ζ if ω = a, ,
80
log  G ( iω )



2 log ω if ω >> a
60


40
0 if ω << a,


π ω −a
if ω = a, ,
20
arg G ( iω ) +

 2 aζ
0
−2 −1 0 1 2


ω /a π if ω >> a
10 10 10 10 10


c K. J. Åström September, 2000 5
 2
Gain Sketching Bode Plots
10

1
10
System
200( s + 1) 1+s
G ( s) = =
s( s + 10)( s + 200) 10s(1 + 0.1s)(1 + 0.01s)
0
10

−1
10
−1 0 1
• Determine break points (poles and zeros) sort them in
ω /a
10 10 10
increasing frequency
Phase 1
• Start with low frequencies ( G (s) )
150 10s
• Draw the low frequency asymptote
100

• Go over all break points and note the slope changes

50
• A crude sketch of the phase curve is obtained by using the
0
−1 0 1
relation that, for systems with no RHP poles or zeros, one
ω /a unit slope corresponds to a phase of 90 ○
10 10 10

0
Gain The Concept of Minimum Phase
10

10
−1
A system is called a minimum phase system if all its poles
and zeros are in the left half plane. Minimum phase systems
10
−2 are easy to control.
For minimum phase systems the phase curve is given by the
−3
10
−1 0 1 2 3 gain curve and vice versa. An approximate relation is
ω
10 10 10 10 10

Phase π d log  G ( iω )
0
arg G ( iω ) ,
2 d log ω
−50
A slope of one for the gain curve corresponds to 90 ○ phase.
−100 The exact relations are called Bodes relations. Systems that
are not minimum phase are called non-minimum phase. The
−150
property of non-minimum phase imposes severe limitations to
10
−1 0
10
ω
10
1 2
10 10
3
what can be achieved by control.


c K. J. Åström September, 2000 6
 Bode’s Relations between Amplitude and Phase The Smoothing Kernel
let G (s) be a transfer function with all poles and zeros in the We have
left half plane. Introduce
∞
ω + ω0 π2
log dω =
0 ω − ω0 2

∞
2ω 0
log  G ( iω ) − log  G ( iω 0 )
arg G ( iω 0 ) = dω
Weight

0 π ω 2 − ω 20 6


1 ∞ d log  G ( iω ) ω + ω0 π d log  G ( iω ) 5
= log dω
π 0 d log ω ω − ω0 2 d log ω 4

y

∞
log  G ( iω ) 2ω 2 ω −1 arg G ( iω ) − ω 0−1 arg G ( iω 0 ) 2
=− 0 dω
log  G ( iω 0 ) π 0 ω 2 − ω 20 1

 
2ω 2 ∞ d ω −1 arg G ( iω ) ω + ω0 0
−3 −2 −1 0 1 2 3
=− 0 log dω 10 10 10 10
ω /a
10 10 10
π 0 dω ω − ω0

a−s Step Response for System with RHP Zero

RHP Zero G (s) =
a+s
Laplace transform of the step response h is
ω
We have  G (iω ) = 1 and arg G (iω ) = −2 arctan G ( s)

∞
a = e− sth( t) dt
Gain s 0
1
10

If G has a RHP zero at s = α > 0 we have

0

∞
G (α )
10

0= = e−α th( t) dt
α 0
−1
10

ω /a
−1 0 1
10 10 10

Phase Since the integral is zero the step response must assume both
0 positive and negative values
−50
1

−100 0.5

0
−150

ω /a
−1 0 1 −0.5
10 10 10

0 0.5 1 1.5 2 2.5 3


c K. J. Åström September, 2000 7
 s−a
Time Delay G (s) = e−sL RHP Pole G (s) = s+a

We have  G (iω ) = 1 and arg G (iω ) = ω L We have  G (iω ) = 1 and arg G (iω ) = −2 arctan ωb =
1
Gain −π + 2 arctan ωb
10

10
1
Gain
0
10

0
10

−1
10

ωL
−1 0 1
10 10 10

Phase 10
−1

ω /a
−1 0 1
10 10 10
0

−100
Phase
0

−200
−50

−300
−100

−400

ωL
−1 0 1
10 10 10 −150

ω /a
−1 0 1
10 10 10

Airplanes Power Systems

The transfer function from elevon to height of an airplane with • Level dynamics in boilers is non-minimum phase because
elevons in the rear are always non-minimum phase. The Wright of the shrink and swell phenomena
brothers avoided this by an elevon in front. • The transfer function from tube opening to power is for a
Modern fighter planes have canards in the front and even jet hydro electric power system
thrusters to avoid the problem. P( s) P0 1 − 2sT
=
X-29 is an experimental aircraft. In one operating condition the A( s) A0 1 + sT
system is approximately described by the transfer function
s − 26
Gnmp ( s) =
s−6

One pole and one zero in the right half plane. This plane is
difficult to control well.


c K. J. Åström September, 2000 8
 Rear Wheel Steering Summary
• Backing with a car is non-minimum phase • The input-output view of dynamical systems
• A bicycle with rear wheel steering has the transfer function • Describe a system by making a table of all input-output
pairs
θ ( s) mhV0 V0 − x f s
= • Fourier’s idea: look at steady state propagation of sinu-
δ ( s) J p c s2 − mg h/ J p
soids
Both poles and zeros in the right half plane! • Frequency response G (iω )
• An airplane with rudders in the rear. • Graphical representations, very useful for intuition
• A boat. • Bode and Nyquist plots
• The concept of non-minimum phase
• Severe performance limitation because of non-minimum
phase


c K. J. Åström September, 2000 9