Ae2235 Topic I 5 2020
Ae2235 Topic I 5 2020
Ae2235 Topic I 5 2020
Max Mulder
Version 2019-2020
Tuesday May 1, 2020
Delft
University of
Technology
. 2 / 47
2-1 Introduction
In this lecture we will discuss continuous-time (CT) system models,
and ask ourselves an important question: given a system, and given
an input signal, what would the output signal look like? This is
known as the system analysis problem.
We will introduce the important class of linear, time-invariant (LTI)
systems. These LTI systems are used very extensively because:
1. . . . we have powerful analysis techniques to study them,
2. . . . many real-world systems can be approximated by them, and
3. . . . the techniques used already suggest approaches to deal
with non-linear systems.
. 3 / 47
2-1 Introduction (2)
Generally, there are three basic approaches for finding the response
of a fixed, linear system to a given input:
1. Obtain a solution through solving the differential equation that
governs the system dynamics,
2. Carry out a solution in the time domain, using the superposition
integral,
3. Use frequency-domain analysis tools such as Fourier/Laplace
transforms.
The first approach is dealt with in courses like Differential Equations and, to some
extent, Vibrations. The second and third approaches are discussed in Chapters 2
and 4 of Ziemer, respectively
. 4 / 47
System Models
. 5 / 47
2-2 System Modelling Concepts
System models represent the interaction of signals and systems,
and the relationship of causes and effects for that system.
Usually, the causes of interest are referred to as inputs and the
effects as outputs of the system. Remember the Example 1-1 of the accelerometer, where the
acceleration of the case was considered the ‘input’, and the movement of the proof mass the ‘output’.
x1 (t)
x2 (t) system y1 (t)
.. ..
. y = Hx .
xm (t) yp (t)
. 6 / 47
2-2 System Modelling Concepts (2)
Systems can have many inputs and many ouputs; here we focus on
single-input, single-output (SISO) systems: m = p = 1.
A symbolical way to express the dependency of the system output
on the input as:
y(t) = H [x(t)] (2-1a)
or y = Hx for short (2-1b).
Eqs (2-1a,b) are to be read: y(t) is the response of H to x(t), where
initial conditions are included, if needed.
The symbol H serves therefore as identifying the system, and
specifying the operation on x(t) to produce y(t).
. 7 / 47
2-2 System Modelling Concepts (3) VIS
Systems can be categorized in a number of ways:
1. instantaneous dynamic
2. time-invariant time-varying
3. linear nonlinear
4. causal non-causal
5. continuous-time discrete-time
. 8 / 47
2-2 System Modelling Concepts (4)
. 9 / 47
2-2 System Modelling Concepts (5)
For dynamic systems the output depends on past (or future, for a
non-causal system) values of the input in addition to present time.
These systems can often be modelled by relationships such as:
dn y(t) dn−1 y(t) dm x(t)
an + an−1 n−1 + . . . + a0 y(t) = bm dtm + . . . + b0 x(t), (2-3)
dtn dt
where an , . . . a0 , bm , . . . b0 are constants. Note that non-zero initial
conditions may be specified.
Clearly then, the system response at time t, y(t) depends not only
on the value of the input at that particular time, x(t), but also on its
values in the past.
Note that the order of the system is defined as the order of the
highest derivative of y(t) (in the case of (2-3), n), when all integrals
in (2-3) are removed. !!differs from Ziemer!!
Introduction System Models Impulse Response Frequency Response
. 10 / 47
2-2 System Modelling Concepts (6)
dy(t)
@home: Is + t2 y(t) = 5x(t) time-invariant?
dt
dy(t)
@home: Is + y(t)y(t) = 5x(t) time-invariant?
dt
. 11 / 47
2-2 System Modelling Concepts (7)
. 12 / 47
2-2 System Modelling Concepts (8)
. 13 / 47
2-2 System Modelling Concepts (9)
. 14 / 47
2-2 System Modelling Concepts (10)
. 15 / 47
Impulse Response
. 16 / 47
2-2 Impulse Response VIS
The input-output relation of a system can be described in terms of
the integral relation:
R∞
y(t) = h(λ)x(t − λ)dλ , (2-5a)
−∞
. 17 / 47
2-2 Impulse Response (2)
Note that the superposition integral is nasty, and being just simple engineers we prefer to
consider systems in the frequency domain, where the convolution becomes a multiplication!
Introduction System Models Impulse Response Frequency Response
. 18 / 47
2-2 Impulse Response (3)
. 19 / 47
2-2 Impulse Response (4)
. 20 / 47
2-2 Impulse Response (5)
We wish to obtain:
1. the ODE relating input x(t) to output y(t), and
2. convert this ODE to integral form.
We can assume that x(t) is applied at t0 , and that y(t0 ) = y0 .
. 21 / 47
2-2 Impulse Response (6)
Example 2-1 (continued) Solution (1) Using Kirchhoff’s voltage law for the loop
indicated by i(t) we obtain:
x(t) = Ri(t) + y(t), (2-7)
where y(t), the voltage across the capacitor, is related to the current i(t) by:
dy(t)
i(t) = C (2-8)
dt
Substituting (2-8) in (2-7) yields:
dy(t)
RC + y(t) = x(t), (2-9)
dt
the linear, time-invariant ODE with order 1 that relates the input x(t) to the output y(t).
Solution (2) The integral equation relating x(t) to y(t) is obtained by solving the ODE, as
discussed in detail in Ziemer on pages 52-53. The result is:
Rt 1 (t−λ)
− RC
y(t) = x(λ) RC e dλ (2-18)
−∞
1 − t
So the impulse response of the RC circuit is: h(t) = RC e RC u(t) (2-19)
Introduction System Models Impulse Response Frequency Response
. 22 / 47
2-2 Impulse Response (7)
Example 2-1 (continued) The response of this RC network to a unit step input
signal u(t):
1 : t>0
u(t) = ,
0 : t<0
can be computed as:
Rt Rt 1 e−λ/RC u(λ)dλ
s(t) = u(t − λ)h(λ)dλ = u(t − λ) RC
−∞ −∞
Rt 1 −λ/RC
= RC e dλ, t ≥ 0 (2-46)
0
= 1 − e−t/RC , t ≥ 0.
Clearly then, since s(t) = 0 for t < 0, this so-called step-response of the system can be
written as:
−t/RC
s(t) = 1 − e u(t) (2-47)
. 23 / 47
2-2 Impulse Response (8)
Example 2-1 (continued) The impulse response function h(t) of the system, as
well as its response s(t) to a unit step function u(t) is shown in the following figure, for two
values of RC (RC = 1 and RC = 3 seconds).
1.2
0.8
We see that, when RC is larger, the
0.6 RC = 1 s impulse response is broader, i.e., it
h(t)
0.4
reacts ‘slower’ to an input unit im-
0.2 RC = 3 s
0
pulse.
−0.2
−2 0 2 4
t, s
6 8 10 When t = RC in (2-47) we obtain
that the system response to the input
1
step reaches 63% of its final value.
RC = 1 s
0.8
0.6 63%
RC = 3 s So, for RC = 3 s it takes longer
s(t)
−0.2
−2 0 2 4 6 8 10
t, s
. 24 / 47
Frequency Response
. 25 / 47
2-7 LTI Frequency Response
In this section we will study the response of an LTI system to an
input that equals a sinusoidal function.
We assume that the system is ‘at rest’ at t = 0 and that the input
signal x(t) is defined as:
0 : t < t0
x(t) =
A sin(2πf0 t) : t > t0
We take a heuristic approach and just start to look at what the
simple RC network system response looks like with this particular
input signal.
Let’s assume that RC = 3 s, and that the input signal amplitude
A = 1, and its frequency f0 = 0.2 Hz; t0 is set at 3 s.
. 26 / 47
2-7 LTI Frequency Response (2)
0
magnitude and phase, but with the
−0.5
same frequency.
−1
0 5 10 15 20 25 30 The output signal has a ‘bump’ in the
t, s
beginning, but when time increases
0.4 (t ≤ 12 s) we get a pure harmonic
0.2
with constant amplitude.
In fact, the ‘bump’ is caused by
y(t)
0
the system’s transient response,
−0.2
discussed in somewhat more detail
−0.4
0 5 10 15 20 25 30
later (although we need Laplace to
t, s discuss it properly...).
. 27 / 47
2-7 LTI Frequency Response (3)
When plotting the input signal (blue) and output signal (green)
together in one plot, we can better see how the amplitude of the
output is different (≈ 0.26), as well as its phase, relative to the input.
The output signal appears to ‘lag’ the input with ≈ 1.0 s.
1
0.5 δA τ
y(t), x(t)
−0.5
−1
0 5 10 15 20 25 30
t, s
Now, when the input signal x(t) would have had another frequency,
the effects on the amplitude and phase would be different.
Introduction System Models Impulse Response Frequency Response
. 28 / 47
2-7 LTI Frequency Response (4)
. 29 / 47
2-7 LTI Frequency Response (5) VIS
In the following it will be shown that the amplitude of the sinusoidal
output signal y(t) equals the amplitude of the input sinusoid x(t),
multiplied with A(ω).
The phase of the sinusoidal output signal y(t) equals the phase of
the input sinusoid x(t), added by θ(ω).
In fact, the complex exponential function of frequency H(ω):
H(ω) = A(ω)ejθ(ω) , (2-97)
. 30 / 47
2-7 LTI Frequency Response (6) VIS
Proof
Consider the response of an LTI system H to input signal x(t) = ejωt .
Using the superposition integral (2-5a) we obtain:
R∞ jω(t−λ)
y(t) = e h(λ)dλ (2-98)
−∞
jωt
R∞ . jωt
y(t) = e h(λ)e −jωλ
dλ = e H(ω), (2-99)
−∞
where H(ω) is the frequency response function, which is simply
the Fourier transform of the impulse response function (with λ = t):
R∞
H(ω) = h(t)e−jωt dt (2-100)
−∞
. 31 / 47
2-7 LTI Frequency Response (7)
Example 2-16 Again let’s consider the simple RC network of Example 2-1. Its
impulse response h(t) was found to be:
So: 1
H(ω) = 1 + jωRC (2-102)
. 32 / 47
2-7 LTI Frequency Response (8)
Example 2-16 (continued) The frequency response function H(ω) can also be
written as follows:
1 1 − jωRC 1 −ωRC
H(ω) = 1 + jωRC · 1 − jωRC = 2 +j
1 + (ωRC) 1 + (ωRC)2
= Re(H(ω)) + Im(H(ω))
p 2
Then we obtain for the amplitude response A(ω) = |H(ω)| = Re (H(ω)) + Im2 (H(ω)):
A(ω) = p 1 (2-104a)
1 + (ωRC)2
Im(H(ω))
and for the phase response θ(ω) = 6 H(ω) = arctan( ):
Re(H(ω))
θ(ω) = − arctan(ωRC) (2-104b)
The system amplitude and phase responses are a function of frequency ω (or, f , it doesn’t
matter) and can therefore be shown as a function of frequency.
These plots are known as Bode plots. An example is shown on the next slide, for two
values of RC (RC = 1 and RC = 3 s).
. 33 / 47
2-7 LTI Frequency Response (9)
0.8
0.6
A(ω)
0.4
0.2 RC = 1 s
0
RC = 3 s
0 10 20 30 40 50 60 70 80 90 100
ω, rad/s
−0.5
θ(ω)
−1
RC = 1 s
−1.5
RC = 3 s
−2
Introduction 0 10 Models
System 20 30 40
Impulse 50 60
Response 70 80
Frequency 90 100
Response
ω, rad/s
. 34 / 47
2-7 LTI Frequency Response (10)
Example 2-16 (continued) It can hardly be seen what happens, so let’s plot it
on logarithmic (amplitude) and semi-log (phase) scales:
10
1 Bode
0
10
A(ω)
−1 RC = 1 s
10
RC = 3 s
−2
10
−3
10
−2 −1 0 1 2
10 10 10 10 10
ω, rad/s
0
−0.5
θ(ω)
−1
RC = 1 s
RC = 3 s
−1.5
−2
−2 −1 0 1 2
10 10 10 10 10
. 35 / 47
2-7 LTI Frequency Response (11)
Example 2-16 (continued) Now let’s look at our earlier example: the response
at 0.2 Hz (= 1.26 rad/s), and RC = 3 s:
10
1 Bode
0
10
A(ω) 0.26
−1 RC = 1 s
10
−2
10
RC = 3 s
−3
2πf0
10
−2 −1 0 1 2
10 10 10 10 10
ω, rad/s
0
−0.5 RC = 3 s
θ(ω)
−1
RC = 1 s
−1.31 rad
−1.5
2πf0
−2
−2 −1 0 1 2
10 10 10 10 10
. 36 / 47
2-7 LTI Frequency Response (12)
. 37 / 47
2-7 LTI Frequency Response (13)
Example 2-16 (continued) Of course these values could have been calculated
analytically, using (2-104a) and (2-104b), by substituting RC = 3 s, and ω = 1.26 rad/s.
@home: Do this at home, and also try another frequency, in fact, why not try to make an
m-file that draws the Bode plot?
From the Bode plot, we see that for RC = 3 s the RC circuit amplitude response equals ‘1’
for frequencies up until ≈ 0.1 rad/s. The phase response is rather small for these
frequencies. In other words, when a sinusoidal input signal would be used with a frequency
lower than this 0.1 rad/s, the output signal would be ‘passed through’ completely. It would
be relatively unchanged!
We also see from the Bode plot that the higher the frequency, the smaller the amplitude
response and the larger the phase response. For instance, at ω ≈ 30 rad/s, the amplitude
1
response is ≈ 100 . Then, when a sinusoidal input signal would be used with a frequency
higher than this 30 rad/s, the output signal would be ‘blocked’ completely. It would be
‘filtered out’!
. 38 / 47
2-7 LTI Frequency Response (14)
Example 2-16 (continued) Here we show the time response of the RC circuit
(RC = 3 s), to an input sinusoid with frequency ω = 0.05 rad/s (top) and ω = 30 rad/s
(bottom).
1
−0.5
−1
0 50 100 150
t, s
0.5
y(t), x(t)
−0.5
−1
0 1 2 3 4 5 6
t, s
Introduction System Models Impulse Response Frequency Response
. 39 / 47
2-7 LTI Frequency Response (15)
. 40 / 47
2-7 LTI Frequency Response (16)
That is, when the input signal can be constructed from a sum of
sinusoidal signals (like in the Fourier Series, for periodic signals, but
the same holds for the Fourier transform!), and we are dealing with a
linear time-invariant system, for which the superposition principle
holds, . . .
. . . then we can compute the LTI system response as being the sum
of the responses of the system to each of the individual input signal
components that together construct the input signal:
∞ ∞
P j2πnf0 t
P j2πnf t
y(t) = H[x(t)] = H Xn e = Xn H e 0
n=−∞ n=−∞
. 41 / 47
2-7 LTI Frequency Response (17)
Example 2-16 (continued) Consider an input signal x(t) that consists of a sum
of 10 sinusoids with frequencies an integer multiple of ω0 = 0.01 rad/s, so:
m=10
P
x(t) = Am sin(mω0 t + φm ).
m=1
The frequency integers m are (1, 3, 7, 11, 31, 51, 117, 217, 417, 1117), the amplitudes Am are
all one, the phases φm are all zero.
In the time domain, this signal looks like:
4
x(t)
−2
−4
0 10 20 30 40 50 60
t, s
. 42 / 47
2-7 LTI Frequency Response (18)
Example 2-16 (continued) Passing this signal through the RC circuit (RC = 3
s) yields the output signal y(t) below (in green). The input signal x(t) is shown in blue.
4
x(t), y(t)
−2
−4
0 10 20 30 40 50 60
t, s
Clearly, the RC circuit filters out most of the high-frequency components of the input signal
x(t). Just look at the Bode plot to see how each of the frequency components is affected!!
Introduction System Models Impulse Response Frequency Response
. 43 / 47
2-7 LTI Frequency Response (19)
Example 2-16 (continued) When we look into the frequency domain, we first
plot the power spectral density of the input signal x(t), |X(ω)|2 .
1
PSD x(t)
10
A2
0 2
10
|X(ω)|2
−1
10
−2
10
−3
10
−2 −1 0 1 2
10 10 10 10 10
ω, rad/s
A2
The PSD of x(t) equals 2
(we show the one-sided spectrum) at all 10 frequency
components that together constitute x(t).
. 44 / 47
2-7 LTI Frequency Response (20)
1
PSD x(t), and FRF squared
10
A2
0 2
|X(ω)|2 , |H(ω)|2
10
−1
10
−2
10 |H(ω)|2
−3
10
−2 −1 0 1 2
10 10 10 10 10
ω, rad/s
Then the PSD of y(t) follows by multiplying the PSD of x(t) with |H(ω)|2 .
. 45 / 47
2-7 LTI Frequency Response (21)
Example 2-16 (continued) Which gives us our end result, the PSD of y(t),
obtained by simply multiplying the PSD of x(t) with the FRF squared.
1
PSD x(t), FRF squared, and PSD y(t)
10
|X(ω)|2 , |H(ω)|2 , |Y (ω)|2
A2
0 2
10
−1
10
−2
10 A2 |H(ω)|2 |H(ω)|2
2
−3
10
−2 −1 0 1 2
10 10 10 10 10
ω, rad/s
Clearly, working in the frequency domain makes our lives so much easier, as we obtain
simple multiplications of Fourier-transformed entities.
. 46 / 47
Recommended home work
@home: Study Example 3-9
Note that we have skipped sections 2-5 and 2-6, don’t study them ;)
. 47 / 47