Chapter 6

Transfer Functions

As a matter of idle curiosity, I once counted to find out what the order of the
set of equations in an amplifier I had just designed would have been, if I had
worked with the differential equations directly. It turned out to be 55.
Henrik Bode, 1960

This chapter introduces the concept of transfer function which is a com-

pact description of the input-output relation for a linear system. Combining
transfer functions with block diagrams gives a powerful method of dealing
with complex systems. The relations between transfer functions and other
system descriptions of dynamics is also discussed.

6.1 Introduction
The transfer function is a convenient representation of a linear time invari-
ant dynamical system. Mathematically the transfer function is a function
of complex variables. For finite dimensional systems the transfer function
is simply a rational function of a complex variable. The transfer function
can be obtained by inspection or by by simple algebraic manipulations of
the differential equations that describe the systems. Transfer functions can
describe systems of very high order, even infinite dimensional systems gov-
erned by partial differential equations. The transfer function of a system
can be determined from experiments on a system.

6.2 The Transfer Function

An input-output description of a system is essentially a table of all possible
input-output pairs. For linear systems the table can be characterized by one

139
140 CHAPTER 6. TRANSFER FUNCTIONS

input pair only, for example the impulse response or the step response. In
this section we will consider another interesting pairs of signals.

Transmission of Exponential Signals

Exponential signals play an important role in linear systems. They appear
in the solution of the differential equation (6.5) and in the impulse response
of a linear systems. Many signals can be represented as exponentials or as
sum of exponentials. A constant signal is simply eαt with α = 0. Damped
sine and cosine signals can be represented by

e(σ+iω)t = eσt eiωt = eσt (sin ωt + i cos ωt)

Many other signals can be represented by linear combination of exponentials.

To investigate how a linear system responds to the exponential input u(t) =
est we consider the state space system
dx
= Ax + Bu
dt (6.1)
y = Cx + Du.

Let the input signal be u(t) = est and assume that s 6= λi (A), i = 1, . . . , n,
where λi (A) is the ith eigenvalue of A. The state is then given by
Z t Z t
At A(t−τ ) sτ At At
x(t) = e x(0) + e Be dτ = e x(0) + e e(sI−A)τ ) B dτ
0 0

Since s 6= λ(A) the integral can be evaluated and we get

¯t
x(t) = eAt x(0) + eAt (sI − A)−1 ¯ e(sI−A)τ )) B
¯
³τ =0 ´
At At
= e x(0) + e (sI − A) −1
e(sI−A)t − I B
³ ´
= eAt x(0) − (sI − A)−1 B + (sI − A)−1 Best

The output of (6.1) is thus

y(t) = Cx(t) + Du(t)

³ ´
= CeAt x(0) − (sI − A)−1 B + [D + C(sI − A)−1 B]est ,

a linear combination of exponential functions with exponents est and eλi t ,

where λi are the eigenvalues of A. One term of the output is proportional
6.2. THE TRANSFER FUNCTION 141

to the input u(t) = est . This term is called the pure exponential response.
If the initial state is chosen as

x(0) = (sI − A)−1 B

the output only consists of the pure exponential response and both the state
and the output are proportional to the input

x(t) = (sI − A)−1 Best = (sI − A)−1 Bu(t)

y(t) = [C(sI − A)−1 B + D]est = [C(sI − A)−1 B + D]u(t).

The ratio of the output and the input

G(s) = C(sI − A)−1 B + D (6.2)

is the transfer function of the system. (6.1) The function

Gxu (s) = C(sI − A)−1

is the transfer function from input to state.

Using transfer functions the response of the system (6.1) to an exponen-
tial input is thus
³ ´
y(t) = Cx(t) + Du(t) = CeAt x(0) − (sI − A)−1 B + G(s)est , (6.3)

Coordinate Changes and Invariants

The matrices A, B and C in (6.1) depend on the coordinate system but not
the matrix D which directly relates inputs and outputs. Since transfer func-
tion relates input to outputs the transfer function should also be invariant
to coordinate changes in the state space. To show this consider the model
(6.1) and introduce new coordinates z by the transformation z = T x, where
T is a regular matrix. The system is then described by
dz
= T (Ax + Bu) = T AT −1 z + T Bu = Ãz + B̃u
dt
y = Cx + DU = CT −1 z + Du = C̃z + Du

This system has the same form as (6.1) but the matrices A, B and C are
different

à = T AT −1 , B̃ = T B, C̃ = CT −1 , D̃ = D (6.4)
142 CHAPTER 6. TRANSFER FUNCTIONS

Computing the transfer function of the transformed model we get

G̃(s) = C̃(sI − Ã)−1 B̃ = CT −1 T (sI − A)−1 T −1 T B

= CT −1 (sI − T AT −1 )−1 T B = C(sI − A)−1 B = G(s)

which is identical to the transfer function (6.2) computed from the system
description (6.1). The transfer function is thus invariant to changes of the
coordinates in the state space.

Transfer Function of a Linear ODE

Consider a linear input/output system described by the differential equation

dn y dn−1 y dm u dm−1 u
+ a 1 + . . . + a n y = b 0 + b 1 + . . . + bm u, (6.5)
dtn dtn−1 dtm dtm−1
where u is the input and y is the output. Note that here we have generalized
our previous system description to allow both the input and its derivatives
to appear. The differential equation is completely described by two polyno-
mials
a(s) = sn + a1 sn−1 + a2 sn−2 + . . . + an−1 s + an
(6.6)
b(s) = b0 sm + b1 sm−1 + . . . + bm−1 s + bm ,
where the polynomial a(s) is the characteristic polynomial of the system.
To determine the transfer function of the system (6.5), let the input be
u(t) = est . Then there is an output of the system that also is an exponential
function y(t) = y0 est . Inserting the signals in (6.5) we find

(sn + a1 sn−1 + · · · + an )y0 est = (b0 sm + b1 sm−1 · · · + bm )e−st

If a(α) 6= 0 it follows that

b(s) st
y(t) = y0 est = e = G(s)u(t). (6.7)
a(s)

The transfer function of the system (6.5) is thus the rational function

b(s)
G(s) = , (6.8)
a(s)

where the polynomials a(s) and b(s) are given by (6.6). Notice that the
transfer function for the system (6.5) can be obtained by inspection.
6.2. THE TRANSFER FUNCTION 143

Example 6.1 (Transfer functions of integrator and differentiator). The trans-

fer function G(s) = 1/s corresponds to the diffrential equation
dy
= u,
dt
which represents an integrator and a differentiator which has the transfer
function G(s) = s corresponds to the differential equation
du
y= .
dt
Example 6.2 (Transfer Function of a Time Delay). Time delays appear in
many systems, typical examples are delays in nerve propagation, communi-
cation and mass transport. A system with a time delay has the input output
relation
y(t) = u(t − T ) (6.9)
Let the input be u(t) = est . Assuming that there is an output of the form
y(t) = y0 est and inserting into (6.9) we get

y(t) = y0 est = es(t−T ) = e−sT est = e−sT u(t)

The transfer function of a time delay is thus G(s) = e−sT which is not a
rational function.

Steady State Gain

The transfer function has many useful physical interpretations. The steady
state gain of a system is simply the ratio of the output and the input in
steady state. Assuming that the the input and the output of the system
(6.5) are constants y0 and u0 we find that an y0 = bn u0 . The steady state
gain is
y0 bn
= = G(0). (6.10)
u0 an
The result can also be obtained by observing that a unit step input can be
represented as u(t) = est with s = 0 and the above relation then follows
from Equation (6.7).

Poles and Zeros

Consider a linear system with the rational transfer function
b(s)
G(s) =
a(s)
 144 CHAPTER 6. TRANSFER FUNCTIONS

The roots of the polynomial a(s) are called poles of the system and the
roots of b(s) are called the zeros of the system. If p is a pole it follows
that y(t) = ept is a solution to the (6.5) with u = 0 (the homogeneous
equation). The function ept is called a mode of the system. The free motion
of the system after an arbitrary excitation is a weighted sum of the modes.
Since the pure exponential output corresponding to the input u(t) = est
with a(s) 6= 0 is G(s)est it follows that the pure exponential output is zero
if b(s) = 0. Zeros of the transfer function thus blocks the transmission of
the corresponding exponential signals.
The poles of the transfer function are the eigenvalues of the system ma-
trix A in the state space model. They depend only on the the dynamics
matrix A, which represents the intrinsic dynamics of the system. The zeros
depend on how inputs and outputs are coupled to the states. The zeros thus
depend on all matrices A, B, C and D in the state space description.
To find the zeros of the state space system (6.1) we observe that the
zeros are complex numbers s such that the input u(t) = est gives zero
output. Inserting the pure exponential response x(t) = x0 est and y(t) = 0
in (6.1) gives.

sest x0 = Ax0 est + Bu0 est

0 = Cest x0 + Dest u0 ,

which can be written as

· ¸· ¸
sI − A B x0
= 0.
C D u0

This equation has a solution with nonzero x0 , u0 only if the matrix on the
left does not have full rank. The zeros are thus the values s such that
· ¸
sI − A B
det =0 (6.11)
C D

Notice in particular that if the matrix B has full rank the matrix has n
linearly independent rows for all values of s. Similarly there are n linearly
independent columns if the matrix C has full rank. This implies that systems
where the matrices B or C are of full rank do not have zeros. In particular
it means that a system has no zeros if it is fully actuated or of the full state
Ä is measured.
Example 6.3 (Transfer Function for Heat Propagation). Consider the one
dimensional heat propagation in a semi-infinite metal rod. Assume that the
6.3. FREQUENCY RESPONSE 145

input is the temperature at one end and that the output is the temperature
at a point on the rod. Let θ be the temperature at time t and position x.
With proper choice of scales heat propagation is described by the partial
differential equation
∂θ ∂2θ
= 2 , (6.12)
∂t ∂ x
and the point of interest can be assumed to have x = 1. The boundary
condition for the partial differential equation is

θ(0, t) = u(t)

To determine the transfer function we assume that the input is u(t) = est .
Assume that there is a solution to the partial differential equation of the
form θ(x, t) = ψ(x)est , and insert this into (6.12) gives

d2 ψ
sψ(x) = ,
dx2
with boundary condition ψ(0) = est . This ordinary differential equation has
the solution √ √
ψ(x) = Aex s + Be−x s .
Matching the boundary conditions gives A = 0 and that B = est and the
solution is
√ √
y(t) = x(1, t) = θ(1, t) = ψ(1)est = e− s st
e = e− s
u(t)
√
The system thus has the transfer function G(s) = e− s.

6.3 Frequency Response

Frequency response is a method where the behavior of a system is character-
ized by its response to sine and cosine signals. The idea goes back to Fourier,
who introduced the method to investigated heat propagation in metals. He
observed that a periodic signal can be approximated by a Fourier series.
Since
eiωt = sin ωt + i cos ωt
it follows that sine and cosine functions are special cases of exponential
functions. The response to sinusoids is thus a special case of the response
to exponential functions.
Consider the linear time-invariant system (6.1). Assume that all eigen-
values of the matrix A have negative real parts. Let the input be u(t) = eiωt
 146 CHAPTER 6. TRANSFER FUNCTIONS

0.2

0.1

y
0

−0.1
0 5 10 15

0.5
u

0
PSfrag replacements
−0.5

−1
0 5 10 15
t

Figure 6.1: Response of a linear time-invariant system with transfer function

G(s) = (s + 1)−2 to a sinusoidal input (full lines). The dashed line shows
the steady state output calculated from (6.13).

and let G(s) be the transfer function of the system. It follows from (6.3)
that the output is
³ ´
y(t) = Cx(t) + Du(t) = CeAt x(0) − (sI − A)−1 B + G(iω)eiωt .

Since the matrix all eigenvalues of A have negative real parts the first term
decays exponentially and the solution converges to the steady state response
y(t) = Im G(iω)eiωt .
¡ ¢

Since u(t) = sin ωt = Im eiωt we can obtain the response to a sinusoid by

¡ ¢

taking the imaginary parts, hence

y(t) = Im G(iω)eiωt = Im |G(iω)|ei arg G(iω) eiωt
¡ ¢ ¡ ¢

= |G(iω)|Im ei(arg G(iω)+ωt) = |G(iω)| sin ωt + arg G(iω) .

¡ ¢ ¡ ¢

The steady state output generated by the input u(t) = sin (ωt) is thus
y(t) = |G(iω)| sin (ωt + arg G(iω)), (6.13)
where |G(iω)| is called the gain of the system and arg G(iω) is called the
phase of the system. This is illustrated in Figure 6.1 which shows the re-
sponse of a linear time-invariant system to a sinusoidal input. The figure
shows the output of the system when it is initially at rest and the steady
state output given by (6.13). The figure shows that after a transient the
output is indeed a sinusoid with the same frequency as the input.
 6.4. THE BODE PLOT 147

3
10
|G(iω)|

2
10

1
10
−2 −1 0 1 2
10 10 10 10 10

90
arg G(iω)

0
PSfrag replacements
−90
−2 −1 0 1 2
10 10 10 10 10
ω

Figure 6.2: Bode plot of the transfer function C(s) = 20 + 10/s + 10s of an
ideal PID controller. The top plot is the gain curve and bottom plot is the
phase curve. The dashed lines show straight line approximations of the gain
curve and the corresponding phase curve.

6.4 The Bode Plot

The frequency response G(iω) can be represented by two curves, the gain
curve and the phase curve. The gain curve gives gain |G(iω)| as a function
of frequency ω and the phase curve gives phase arg G(iω) as a function of
frequency ω. The curves obtained when logarithmic scales for frequency
and gain and linear scale for phase are used is called the B ode plot, see
Figure 6.2 An useful feature of the Bode plot of a rational transfer function
is that the gain curve can be approximated by piecewise by straight lines
with integer slopes. The lines change slope at the poles and zeros of the
transfer function. Tracing the curve for increasing frequencies the slope
increases with one unit at a zero and it decreases with one unit at a pole.
These straight lines correspond to phase curves that are horizontal at values
that are integer multiples of 90◦ . This is illustrated in Figure 6.2 which gives
the Bode plot of an ideal PID controller with the transfer function

10 10(s + 1)2
C(s) = 20 + + 10s = .
s s
The Bode plot is shown in full lines and the straight line approximation
in dashed lines. For ω < 0.1 we have G(s) ≈ 10/s, the approximation of
148 CHAPTER 6. TRANSFER FUNCTIONS

the gain curve is a line with slope -1, and the phase curve is horizontal
arg G(iω) = −90◦ . For ω > 10 we have G(s) ≈ 10/s, the approxima-
tion of the gain curve is a straight line, and the phase curve is horizontal,
arg G(iω) = 90◦ .
It is easy to sketch Bode plots because they have linear asymptotes.
This is useful in order to get a quick estimate of the behavior of a system.
It is also a good way to check numerical calculations. Consider a transfer
function which is a polynomial G(s) = b(s)/a(s). We have

log G(s) = log b(s) − log a(s)

Since a polynomial is a product of terms of the type :

k, s, s + a, s2 + 2ζas + a2

it suffices to be able to sketch Bode diagrams for these terms. The Bode
plot of a complex system is then obtained by adding the gains and phases
of the terms.
Example 6.4 (Bode Plot of an Integrator). Consider the transfer function

k
G(s) =
s

We have G(iω) = k/iω which implies

log |G(iω)| = log k − log ω, arg G(iω) = −π/2

The gain curve is thus a straight line with slope -1 and the phase curve is a
constant at −90◦ . Bode plots of a differentiator and an integrator are shown
in Figure 6.3
Example 6.5 (Bode Plot of a Differentiator). Consider the transfer function

G(s) = ks

We have G(iω) = ikω which implies

log |G(iω)| = log k + log ω, arg G(iω) = π/2

The gain curve is thus a straight line with slope 1 and the phase curve is a
constant at 90◦ . The Bode plot is shown in Figure 6.3.
 6.4. THE BODE PLOT 149

1 1
10 10
|G(iω)|

|G(iω)|
0 0
10 10

−1 −1
10 10
−1 0 1 −1 0 1
10 10 10 10 10 10

90 90
arg G(iω)

arg G(iω)
0 0

PSfrag replacements
−90 −90
−1 0 1 −1 0 1
10 10 10 10 10 10
ω ω

Figure 6.3: Bode plot of the transfer functions G(s) = 1/s (left) and G(s) =
s (right).

Compare the Bode plots for the differentiator in and the integrator in
Figure 6.3. The plot for the differentiator is obtained by mirror imaging the
gain and phase curves for the integrator in the horizontal axis. This follows
from the following property of the logarithm.

1
log = − log G = − log |G| − i arg G
G

Example 6.6 (Bode Plot of a First Order System). Consider the transfer
function
a
G(s) =
s+a
We have
log G(s) = log a − log s + a

Hence

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

The Bode Plot is shown in Figure 6.4. Both the gain curve and the phase
 150 CHAPTER 6. TRANSFER FUNCTIONS

1 1
10 10

|G(iω)|/a

|G(iω)|/a
0 0
10 10

−1 −1
10 10
−1 0 1 −1 0 1
10 10 10 10 10 10
0 0
arg G(iω)

arg G(iω)
−90 −90

PSfrag replacements PSfrag replacements

−180 −180
−1 0 1 −1 0 1
10 10 10 10 10 10
ω/a ω/a

Figure 6.4: Bode plots of the systems G(s) = a/(s + a) (left) and G(s) =
a2 /(s2 + 2ζas + a2 ) (right). The full lines show the Bode plot and the
dashed lines show the straight line approximations to the gain curves and the
corresponding phase curves. The plot for second order system has ζ = 0.1,
0.2, 0.5 and 1.0.

curve can be approximated by the following straight lines

(
log a if ω << a,
log |G(iω)| ≈ ,
− log ω if ω >> a
(
0 if ω << a,
arg G(iω) ≈ .
− π2 if ω >> a

Notice that a first order system behaves like a constant for low frequencies
and like an integrator for high frequencies. Compare with the Bode plot in
Figure 6.3.
Example 6.7 (Bode Plot of a Second Order System). Consider the transfer
function
a2
G(s) = 2
s + 2aζs + a2
We have
log G(iω) = 2 log a − log (−ω 2 + 2iaζω + a2 ).
6.5. FREQUENCY RESPONSES FROM EXPERIMENTS 151

Hence
1
log ω 4 + 2a2 ω 2 (2ζ 2 − 1) + a4
¡ ¢
log |G(iω)| = 2 log a −
2
2ζaω
arg G(iω) = − arctan 2
a − ω2
The gain curve has an asymptote with zero slope for ω << a. For large
values of ω the gain curve has an asymptote with slope -2. The largest
gain Q = maxω |G(iω)| = 1/(2ζ), called the Q value, is obtained for ω =
a. The phase is zero for low frequencies and approaches 180◦ for large
frequencies. The curves can be approximated with the following piece-wise
linear expressions
(
0 if ω << a,
log |G(iω)| ≈ ,
−2 log ω if ω >> a
(
0 if ω << a,
arg G(iω) ≈ ,
−π if ω >> a

The Bode plot is shown in Figure 6.4.

Interpretations of Bode Plots

The Bode plot gives a quick overview of a system. Many properties can
be read from the plot. Because logarithmic scales are used the plot gives
the properties over a wide range of frequencies. Since any signal can be
decomposed into a sum of sinusoids it is possible to visualize the behavior of
a system for different frequency ranges. Furthermore when the gain curves
are close to the asymptotes the system can be approximated by integrators
of differentiators. Consider for example the Bode plot in Figure 6.2. For low
frequencies the gain curve of the Bode plot has the slope -1 which means
that the system acts like an integrator. For high frequencies the gain curve
has slope +1 which means that the system acts like a differentiator.

6.5 Frequency Responses from Experiments

Modeling can be difficult and time consuming. One reason why control has
been so successful is that the frequency response of a system can be deter-
mined experimentally by perturbing the input sinusoidally. When steady
152 CHAPTER 6. TRANSFER FUNCTIONS

Figure 6.5: The Hewlett Packard signal analyzer can be used to determine
frequency response experimentally.

state is reached the amplitude ratio and the phase lag gives the frequency
response for the excitation frequency. The complete frequency response is
obtained by sweeping over frequency. By using correlation techniques it is
possible to determine the frequency response very accurately. An analytic
transfer function can be obtained from the frequency response by curve fit-
ting. Nice instruments are commercially available, see Figure 6.5.
Example 6.8 (A Piezoelectric Drive). Experimental determination of fre-
quency responses is particularly attractive for systems with fast dynamics.
A typical example is given if Figure 6.6. In this case the frequency was
obtained in less than a second. The full line shows the measured frequency
response. The transfer function

kω22 ω32 ω52 (s2 + 2ζ1 ω1 + ω12 )(s2 + 2ζ4 ω4 + ω42 )

G(s) = e−sT
ω12 ω42 (s2 + 2ζ2 ω2 + ω22 )(s2 + 2ζ3 ω3 + ω32 )(s2 + 2ζ5 ω5 + ω52 )

with ω1 = 2420, ζ1 = 0.03, ω2 = 2550, ζ2 = 0.03, ω3 = 6450, ζ3 = 0.042,

ω4 = 8250, ζ4 = 0.025, ω5 = 9300, ζ5 = 0.032, T = 10−4 , and k = 5. was
fitted to the data. The frequencies associated with the zeros are located
where the gain curve has minima and the frequencies associated with the
poles are located where the gain curve has local maxima. The relative
damping are adjusted to give a good fit to maxima and minima. When a
good fit to the gain curve is obtained the time delay is adjusted to give a good
fit to the phase curve. The fitted curve is shown i dashed lines. Experimental
determination of frequency response is less attractive for systems with slow
dynamics because the experiment takes a long time.
 6.5. FREQUENCY RESPONSES FROM EXPERIMENTS 153

1
10
Gain

0
10

2 3 4
10 10 10

0
Phase

−90

PSfrag replacements −180

−270
2 3 4
10 10 10
Frequency [Hz]

Figure 6.6: Frequency response of a piezoelectric drive for an atomic force

microscope. The input is the voltage to the drive amplifier and the output
is the output of the amplifier that measures beam deflection.

The human eye is an organ that is easily accessible for experiments. It

has a control system which adjust the pupil opening to regulate the light in-
tensity at the retina. This control system was explored extensively by Stark
in the late 1960s. To determine the dynamics light intensity on the eye
was varied sinusoidally and the pupil opening was measured. A fundamen-
tal difficulty is that the closed loop system is insensitive to internal system
parameters. Analysis of a closed loop system thus gives little information
about the internal properties of the system. Stark used a clever experimen-
tal technique which allowed him to investigate both open and closed loop
dynamics. He excited the system by varying the intensity of a light beam
focused on the eye and he measured pupil area, see Figure 6.7. By using
a wide light beam that covers the whole pupil the measurement gives the
closed loop dynamics. The open loop dynamics was obtained by using a
narrow beam. By focusing a narrow beam on the edge of the pupil opening
the gain of the system can be increased so that the pupil oscillates. The
result of one experiment for determining open loop dynamics is given in
 154 CHAPTER 6. TRANSFER FUNCTIONS

Example 6.9 (Pupillary Light Reflex Dynamics).

Figure 6.7: Light stimulation of the eye. In A the light beam is so large
that it always covers the whole pupil. This experiment gives the closed loop
dynamics. In B the light is focused into a beam which is so narrow that it
is not influenced by the pupil opening. This experiment gives the open loop
dynamics. In C the light beam is focused on the edge of the pupil opening.
The pupil will oscillate if the beam is sufficiently small. From [?].

Figure 6.8. Fitting a transfer function to the gain curves gives a good fit for
G(s) = 0.17/(1 + 0.08s)3 . This curve gives a poor fit to the phase curve as
as shown by the dashed curve in Figure 6.8. The fit to the phase curve is
improved by adding a time delay. Notice that a time delay does not change
the gain curve. The final fit gives the model
0.17
G(s) = e−0.2s .
(1 + 0.08)3

Ä
The Bode plot of this is shown with dashed curves in Figure 6.8.
Example 6.10 (Determination of Thermal Diffusivity). The Swedish Physi-
cist Ångström used frequency response to determine thermal conductivity
in 1861. A long metal rod with small cross section was used. A heat-wave is
generate by periodically varying the temperature at one end of the sample.
Thermal diffusivity is then determined by analyzing the attenuation and
phase shift of the heat wave. A schematic diagram of Ångström’s apparatus
is shown in Figure 6.9. The input was generated by switching from steam
to cold water periodically. Switching was done manually because of the low
frequencies used. Heat propagation is described by the one-dimensional heat
equation
∂T ∂2T
= κ 2 − µT, (6.14)
∂t ∂x
λ
where κ = ρC , and λ the thermal conductivity, ρ density, and C specific
heat. The term µT is an approximation of heat losses due to radiation, and
convection. The transfer function relating temperatures at points with the
distance a is √
G(s) = e−a (s+µ)/κ , (6.15)
 6.5. FREQUENCY RESPONSES FROM EXPERIMENTS 155

0.2

|G(iω)|
0.1
0.05

0.02
0.01
2 5 10 20

arg G(iω)
PSfrag replacements PSfrag replacements −180

ω
|G(iω)| −360
arg G(iω) 2 5 10 20
ω

Figure 6.8: Sample curves from open loop frequency response of the eye
(left) and Bode plot for the open loop dynamics (right). Redrawn from the
data of [?]. The dashed curve in the Bode plot is the minimum phase curve
corresponding to the gain curve.Perhaps both Bode and Nyquist plots?

and the frequency response is given by

s p
µ+ω 2 + µ2
log |G(iω)| = −a
2κ
s p
−µ + ω 2 + µ2
arg G(iω) = −a .
2κ

Multiplication of the equations give

a2 ω
log |G(iω)| arg G(iω) = . (6.16)
2κ
Notice that the parameter µ which represents the thermal losses disappears
in this formula. Ångström remarked that (6.16) is indeed a remarkable
simple formula. Earlier values of thermal conductivity for copper obtained
by Peclet was 79 W/mK. Ångström obtained 382 W/mK which is very
close to modern data. Since the curves shown in Figure 6.9 are far from
sinusoidal Fourier analysis was used to find the sinusoidal components. The
procedure developed by Ångström quickly became the standard method for
determining thermal conductivity.
156 CHAPTER 6. TRANSFER FUNCTIONS

Figure 6.9: Schematic diagram of AAngström’s equipment (left) and sample

of his data(right). The copper rod had length 0.57 m, and diameter of 23.75
mm. Holes were drilled at distances of 0.05 m.

Notice that both for the piezo drive and for the pupillary dynamics it is
not easy to derive appropriate models from first principle. It is often fruitful
to use a combination.

6.6 Block Diagrams and Transfer Functions

The combination of block diagrams and transfer functions is a powerful way
to represent control systems. Transfer functions relating different signals
in the system can be derived by pure algebraic manipulations of the the
transfer functions of the blocks using block diagram algebra. To show how
this can be done we will begin with simple combination of systems.
Since the transfer function is characterized by transmission of exponen-
tial signals we let the input be u(t) = est . Recall from Section 6.2 that the
pure exponential output a linear system with transfer function G(s) is

y(t) = G(s)est = G(s)u(t)

which we simply write as y = Gu.

Now consider a system which is a cascade combination of systems with
the transfer functions G1 (s) and G2 (s), see Figure 6.10A. Let the input of
the system be u = est . The pure exponential output of the first block is the
exponential signal G1 u, which is also the input to the second system. The
pure exponential output of the second system is

y = G2 (G1 u) = G1 G2 u
PSfrag replacements PSfrag replacements PSfrag replacements

e e
6.6. BLOCK DIAGRAMS
G1 AND TRANSFER FUNCTIONS
G1 157
−G2 −G2
G1 G2 u e y
Σ G1
G1 u y
Σ
Σu y Σ
G1 G2 G 2 −G2
Σ Σ

Figure 6.10: Series (left), parallel (middle) and feedback (right) connections
of two linear systems.

The transfer function of the system is thus G = G1 G2 , i.e. the product of

the transfer functions.
Consider a parallel connection of systems with the transfer functions G1
and G2 , see Figure 6.10b. Let u = est be the input to the system. The pure
exponential output of the first system is then y1 = G1 u and the output of
the second system is y2 = G2 u. The pure exponential output of the parallel
connection is thus

y = G1 u + G2 u = (G1 + G2 )u.

The transfer function of the parallel connection is thus G = G1 + G2 .

Consider a feedback connection of systems with the transfer functions
G1 and G2 , as shown in Figure 6.10c. Let r = est be the input to the system,
y the pure exponential output, and e be the pure exponential part of the
error. Writing the relations for the different blocks and the summation unit
we find

y = G1 e
e = r − G2 y.

Elimination of e gives
y = G1 (r − G2 y),
hence
(1 + G1 G2 )y = G1 r,
which implies
G1
y= r.
1 + G 1 G2
The transfer function of the feedback connection is thus
G1 (s)
G(s) = .
G1 (s) + G2 (s)
PSfrag replacements

158 − CHAPTER 6. TRANSFER FUNCTIONS

d n

r e u x y
Σ C Σ P Σ

−1

Figure 6.11: Block diagram of a feedback system.

With a little practice the equation relating the signals of interest can be
written by inspection. We illustrate this by an example
Example 6.11 (Relations Between Signals in a Block Diagram). Consider the
system in Figure 6.11. The system has two blocks representing a process P
and a controller C. There are three external signals, the reference r, the load
disturbance d and the measurement noise n. A typical problem is to find
out how the error e related to the signals r d and n? To derive the transfer
function we simply assume that all signals are exponential functions and we
drop the arguments of signals and transfer functions.
To obtain the desired relation we simply trace the signals around the
loop. We begin with the signal we are interested in, i.e. e. The error e is
given by
e = r − y.
The signal y is the sum of n and x

y = n + x,

where x is the output of the process, i.e.

x = P v = P (d + u),

where u is the output of the controller, i.e.

u = Ce.

Combining the equations gives

¡ ¢
e = r − y = r − (n + x) = r − n + P (d + u)
¡ ¢
= r − n + P (d + Ce) .
6.6. BLOCK DIAGRAMS AND TRANSFER FUNCTIONS 159

Hence
¡ ¢
e = r − n + P (d + Ce) = r − n − P d − P Ce.
Solving this equation for e gives

1 1 P
e= r− n− d = Ger r + Gen n + Ged d (6.17)
1 + PC 1 + PC 1 + PC

The error is thus the sum of three terms, depending on the reference r, the
measurement noise n and the load disturbance d. The function
1
Ger =
1 + PC

is the transfer function from reference r to error e, Gen is the transfer func-
tions from measurement noise n to error e and Ged is the transfer functions
from load disturbance d to error e.
The example illustrates an effective way to manipulate the equations to
obtain the relations between inputs and outputs in a feedback system. The
general idea is to start with the signal of interest and to trace signals around
the feedback loop until coming back to the signal we started with. With a
some practice the equation (6.17) can be written directly by inspection of
the block diagram. Notice that all terms in Equation (6.17) formally have
the same denominators, there may, however, be factors that cancel.
The combination of block diagrams and transfer functions is a powerful
tool because it is possible both to obtain an overview of a system and find
details of the behavior of the system. When manipulating block diagrams
and transfer functions it may happen that poles and zeros are canceled.
Some care must be exercised as is illustrated by the following example.
Example 6.12 (Cancellation of Poles and Zeros). Consider a system de-
scribed by the equation.
dy du
−y = −u (6.18)
dt dt
Integrating the equation gives

y(t) = u(t) + cet

where c is a constant. The transfer function of the system is

Y (s) s−1
= =1
U (s) s−1
 160 CHAPTER 6. TRANSFER FUNCTIONS

Since s is a complex variable the cancellation is clearly permissible and we

find that the transfer function is G(s) = 1 and we have seemingly obtained
a contradiction because the system is not equivalent to the system

y(t) = u(t)

In the next section we will give a fundamental results that gives insight
into the cancellation problem.
Ä

6.7 The Kalman Decomposition

Equation (6.2) gives a formula for the transfer function of a linear system
on state space form. It appears at first sight that the denominator of the
transfer function has the characteristic polynomial det (sIA ) as a numerator.
This is not necessarily true because there may be cancellations as illustrated
in Example 6.12. A very good insight into what happens can be obtained by
using the concepts of reachability and observability discussed in Chapter 5.
The key result is Kalman’s decomposition theorem, which says that a linear
system can be divided into four subsystems. A linear system can thus be
decomposed into four subsystems: Sro which is reachable and observable,
Srō which is reachable not observable, Sr̄o which is not reachable observable,
and Sr̄ō which is not reachable not observable.

Diagonal Systems
We will show this in the special case of systems where the matrix A has
distinct eigenvalues. Such a system can be represented as
   
Λro 0 0 0 βro
dz  0 Λrō 0
 0 
βro  u
 
= z +
dt 0 0 Λr̄o 0   0 
0 0 0 Λr̄ō 0
£ ¤
y = γro 0 γr̄o 0 z + Du.

All states zk such that βk 6= 0 are controllable and all states such that γk 6= 0
are observable. The system can be represented as The transfer function of
the system is
G(s) = γro (sI − Aro )−1 βro + D.
It is uniquely given by the subsystem which is reachable and observable. A
block diagram of the system is given in Figure 6.12.
6.7. THE KALMAN DECOMPOSITION 161

Figure 6.12: Kalman’s decomposition of a linear system with distinct eigen-

values.

The Cancellation Problem

The Kalman decomposition gives a good insight into what happens when
poles and zeros are canceled in a transfer function as illustrated in Exam-
ple 6.12. In this particular case the system has two subsystems Sro and Sr̄o .
The system Sro is a static system with transfer function G(s) = 1 and the
subsystem Sr̄o which is observable but non reachable has the dynamics.

dx
=x
dt
This system is unstable and the unstable state will appear in the output.

The General Case

Some linear algebra is required to solve the general case. Introduce the
reachable subspace Xr which is the linear subspace spanned by the columns
of the reachability matrix Wr . The state space is the direct sum of Xr and
another linear subspace Xr̄ . Notice that Xr is unique but that Xr̄ can be
chosen in many different ways. Choosing a coordinates with xr ∈ Xr and
xbarr ∈ Xr̄ the system equations can be written as
µ ¶ µ ¶µ ¶ µ ¶
d xc A11 A12 xc B1
= + u (6.19)
dt xc̄ 0 A22 xc̄ 0

where the states xc are reachable and xc̄ are non-reachable.

 162 CHAPTER 6. TRANSFER FUNCTIONS

Introduce the unique subspace X¯o of non-observable states. This is right

the null space of the observability matrix Wo . The state space is the direct
sum of X¯o and the non-unique subspace Xo . Choosing a coordinates with
xo ∈ Xo and xbaro ∈ Xō the system equations can be written as
µ ¶ µ ¶µ ¶
d xo A11 0 xo
=
dt xō A21 A22 x0̄
µ ¶ (6.20)
¡ ¢ xo
y = C1 0
xō
where the states xo are observable and xō not observable (quiet)
The intersection of two linear subspaces is also a linear subspace. Intro-
duce Xrō as the intersection of Xr and Xō and the linear subspace Xro which
together with Xrō spans Xr . Finally we introduce the linear subspace Xr̄o
which together with Xrō , Xrō and Xrō spans the full state space. Notice that
the decomposition is not unique because only the subspace Xrō is unique.
Combining the representation (6.19) and (6.20) we find that a linear
system can be transformed to the form
   
A11 0 A13 0 B1
dx  A 21 A 22 A 23 A 24 
  B2 
= x+ u
dt  0 0 A33 0   0 (6.21)
0 0 A43 A44 0
¡ ¢
y = C1 0 C 2 0 x
where the state vector has been partitioned as
 T
xro
xrō 
x= xr̄o 


xr̄ō
A block diagram of the system is shown in Figure 6.13. By tracing the
arrows in the diagram we find that the input influences the systems Soc and
Sōc and that the output is influenced by Soc and Soc̄ . The system Sōc̄ is
neither connected to the input nor the output. The transfer function of the
system is
G(s) = C1 (sI − A11 )−1 B1 (6.22)
It thus follows that the transfer function is uniquely given by the reachable
and observable subsystem Soc . When cancellations occur it means that there
ares subsystems that are not reachable and observable.
Ä
6.8. LAPLACE TRANSFORMS 163

Figure 6.13: Kalman’s decomposition of a linear system with distinct eigen-

values.

6.8 Laplace Transforms

Transfer functions are traditionally introduced using Laplace transforms.
Students who are familiar with Laplace transform will get additional in-
sight from this. Traditionally Laplace transforms were also used to compute
responses of linear system to different stimuli. Today we can easily gener-
ate the responses using computers. Only a few elementary properties are
needed for basic control applications. There is a however a beautiful theory
for Laplace transforms which makes it possible to use many powerful tools
of the theory of functions of a complex variable to get deep insights into the
behavior of systems.
Consider a time function f : R+ → R which is integrable and grows no
faster than es0 t for large t. The Laplace transform maps f to a function
F = Lf : C → C of a complex variable. It is defined by
Z ∞
F (s) = e−st f (t)dt, Res > s0 (6.23)
0

The transform has some properties that makes it very well suited to deal
with linear systems. First we observe that the transform is linear because
Z ∞
L(af + bg) = e−st (af (t) + bg(t))dt
0
Z ∞ Z ∞ (6.24)
−st
=a e f (t)dt + b e−st g(t)dt = aLf + bLg
0 0
164 CHAPTER 6. TRANSFER FUNCTIONS

Next we will calculate the Laplace transform of the derivative of a function.

We have
Z ∞ Z ∞
df −st 0 −st
¯∞
L = e f (t)dt = e f (t)¯ + s e−st f (t)dt = −f (0) + sLf
¯
dt 0 0 0
where the second equality is obtained by integration by parts. We thus
obtain the following important formula for the transform of a derivative
df
L = sLf − f (0) = sF (s) − f (0) (6.25)
dt
This formula is particularly simple if the initial conditions are zero because
it follows that differentiation of a function corresponds to multiplication of
the transform with s.
Since differentiation corresponds to multiplication with s we can expect
that integration corresponds to division by s. This is true as can be seen by
calculating the Laplace transform of an integral. We have
Z t Z ∞³ Z t ´
−st
L f (τ )dτ = e f (τ )dτ dt
0 0 0
Z t Z ∞ −sτ
e −st e 1 ∞ −sτ
Z
¯∞
¯
=− e−sτ f (τ )dτ ¯ + f (τ )dτ = e f (τ )dτ
s 0 0 0 s s 0
hence
t
1 1
Z
L f (τ )dτ =
Lf = F (s) (6.26)
0 s s
Integration of a time function thus corresponds to dividing the Laplace trans-
form by s.

The Laplace Transform of a Convolution

Consider a linear time-invariant system with zero initial state. The relation
between the input u and the output y is given by the convolution integral
Z ∞
y(t) = g(t − τ )u(τ )dτ.
0
We will now consider the Laplace transform of such an expression. We have
Z ∞ Z ∞ Z ∞
−st −st
Y (s) = e y(t)dt = e g(t − τ )u(τ )dτ dt
0 0 0
Z ∞Z t
= e−s(t−τ ) e−sτ g(t − τ )u(τ )dτ dt
Z0 ∞ 0 Z ∞
−sτ
= e u(τ )dτ e−st g(t)dt = G(s)U (s)
0 0
6.9. FURTHER READING 165

The result can be written as Y (s) = G(s)U (s) where G, U and Y are the
Laplace transforms of g, u and y. The system theoretic interpretation is
that the Laplace transform of the output of a linear system is a product
of two terms, the Laplace transform of the input U (s) and the Laplace
transform of the impulse response of the system G(s). A mathematical
interpretation is that the Laplace transform of a convolution is the product
of the transforms of the functions that are convoluted. The fact that the
formula Y (s) = G(s)U (s) is much simpler than a convolution is one reason
why Laplace transforms have become popular in control.

The Transfer Function

The properties (6.24) and (6.25) makes the Laplace transform ideally suited
for dealing with linear differential equations. The relations are particularly
simple if all initial conditions are zero.
Consider for example a system described by (6.1). Taking Laplace trans-
forms under the assumption that all initial values are zero gives
sX(s) = AX(s) + BU (s)
Y (s) = CX(s) + DU (s)
Elimination of X(s) gives
³ ´
Y (s) = C[sI − A]−1 B + D U (s) (6.27)

The transfer function is thus G(s) = C[sI − A]−1 B + D. Compare with

(6.2).
The formula (6.27) has a strong intuitive interpretation because it tells
that the Laplace transform of the output is the product of the transfer
function of the system and the transform of the input. In the transform
domain the action of a linear system on the input is simply a multiplication
with the transfer function. The transfer function is a natural generalization
of the concept of gain of a system.

6.9 Further Reading

The concept of transfer function was an important part of classical con-
trol theory, see James Nichols Phillips. It was introduced via the Laplace
transform Gardner Barnes which also was used to calculate response of lin-
ear systems. The Laplace transform is of less importance nowadays when
responses to linear systems can easily be generated using computers.
166 CHAPTER 6. TRANSFER FUNCTIONS

6.10 Exercises
1. The linearized model of the pendulum in the upright position is char-
acterized by the matrices
· ¸ · ¸
0 1 0 £ ¤
A= , B= , C = 1 0 , D = 0.
1 0 1
Determine the transfer function of the system.
2. Consider the differential equation
dn y dn−1 y dn−2 y
+ a 1 + a 2 + . . . + an y = 0
dtn dtn−1 dtn−2
Let λ be a root of the polynomial
sn + a1 sn−1 + · · · + an = 0.
Show that the differential equation has the solution y(t) = eλt .
3. Consider the system
dn y dn−1 y dn−1 u dn−2 u
+ a 1 + . . . + a n y = b 1 + b 2 + . . . + bn u,
dtn dtn−1 dtn−1 dtn−2
Let λ be a zero of the polynomial
b(s) = b1 sn−1 + b2 sn−2 + · · · + bn
Show that if the input is u(t) = eλt then there is a solution to the
differential equation that is identically zero.
4. Consider the linear state space system
ẋ = Ax + Bu
y = Cx.
Show that the transfer function is
b1 sn−1 + b2 sn−2 + · · · + bn
G(s) =
sn + a1 sn−1 + · · · + an
where
b1 = CB
b2 = CAB + a1 CB
b3 = CA2 B + a1 CAB + a2 CB
..
.
bn = CAn−1 B + a1 CAn−1 B + . . . + an−1 CB