SDOF linear oscillator

Response to Periodic and Non‐periodic Loadings

Giacomo Boffi

http://intranet.dica.polimi.it/people/boffi‐giacomo

Dipartimento di Ingegneria Civile Ambientale e Territoriale

Politecnico di Milano

March 17, 2020

Outline

SDOF linear
oscillator
Giacomo Boffi
Response to Periodic Loading
Response

Fourier
Transform
Fourier Transform
The DFT

General Load
The Discrete Fourier Transform H(ω) vs h(t)

Response to General Dynamic Loadings

Relationship between Analysis in Time and Frequency Domains

Response to Periodic Loading

SDOF linear
oscillator
Response to Periodic Loading Giacomo Boffi
Introduction
Response
Fourier Series Representation Intro

Fourier Series of the Response Fourier Series

Response’s FS

An example Example

Fourier
Transform

Fourier Transform The DFT

General Load

H(ω) vs h(t)
The Discrete Fourier Transform

Response to General Dynamic Loadings

Relationship between Analysis in Time and Frequency Domains

Introduction

A periodic loading is characterized by the identity SDOF linear

oscillator
Giacomo Boffi
p(t) = p(t + T )
Response
Intro
2π
where T is the period of the loading, and ω1 = T is its principal frequency. Fourier Series
Response’s FS
Example

Fourier
p Transform

The DFT

General Load

H(ω) vs h(t)

p(t) p(t + T )

t
T
Introduction

SDOF linear
oscillator
Giacomo Boffi

Response
Intro

Fourier theorem asserts that periodic loadings can be represented by an infinite Fourier Series
Response’s FS

series of harmonic functions. E.g., for an antisymmetric periodic loading of period Example

Fourier
T we have a series composed of antisymmetric harmonic functions Transform
∑∞ ∑∞ The DFT
p(t) = p(−t) = j=1 pj sin jω1 t = j=1 pj sin ωj t (with ωj = j 2π
T ). General Load

H(ω) vs h(t)
Introduction

SDOF linear
oscillator
Giacomo Boffi
The steady‐state response of a SDOF system for a harmonic loading
Response
∆pj (t) = pj sin ωj t is known; with βj = ωj /ωn the s‐s response is: Intro
Fourier Series
pj
xj,s‐s = k D(βj , ζ) sin(ωj t − θ(βj , ζ)) = aj cos ωj t + bj sin ωj t. Response’s FS
Example

Fourier
Transform

The DFT

General Load

H(ω) vs h(t)
Introduction

SDOF linear
oscillator
Giacomo Boffi
The steady‐state response of a SDOF system for a harmonic loading
Response
∆pj (t) = pj sin ωj t is known; with βj = ωj /ωn the s‐s response is: Intro
Fourier Series
pj
xj,s‐s = k D(βj , ζ) sin(ωj t − θ(βj , ζ)) = aj cos ωj t + bj sin ωj t. Response’s FS
Example

Fourier
The response to an individual harmonic component can be interpreted as a term of Transform

another different Fourier series, that possibly represents the steady‐state response The DFT

General Load
of the dynamic system to p(t).
H(ω) vs h(t)
Introduction

SDOF linear
oscillator
Giacomo Boffi
The steady‐state response of a SDOF system for a harmonic loading
Response
∆pj (t) = pj sin ωj t is known; with βj = ωj /ωn the s‐s response is: Intro
Fourier Series
pj
xj,s‐s = k D(βj , ζ) sin(ωj t − θ(βj , ζ)) = aj cos ωj t + bj sin ωj t. Response’s FS
Example

Fourier
The response to an individual harmonic component can be interpreted as a term of Transform

another different Fourier series, that possibly represents the steady‐state response The DFT

General Load
of the dynamic system to p(t).
H(ω) vs h(t)
It can be shown that, under very wide assumptions, the infinite series whose terms
are the s‐s responses to the harmonic components of p(t) is indeed the Fourier
series representation of the SDOF steady‐state response to p(t).
Introduction

SDOF linear
oscillator
Giacomo Boffi

Due to the asymptotic behaviour of D(β; ζ) (D Response

Intro
Fourier Series

goes to zero as β −2 for β ≫ 1) it is apparent that a good

Response’s FS
Example

Fourier
approximation to the steady‐state response can Transform

The DFT

be obtained using a limited number of General Load

H(ω) vs h(t)
low‐frequency terms.
Fourier Series

SDOF linear
oscillator
Giacomo Boffi

Response
Using Fourier theorem any practical periodic loading can be expressed as a series Intro
Fourier Series

of harmonic loading terms. Response’s FS

Example

Fourier
Transform

The DFT

General Load

H(ω) vs h(t)
Fourier Series

SDOF linear
oscillator
Giacomo Boffi

Response
Using Fourier theorem any practical periodic loading can be expressed as a series Intro
Fourier Series

of harmonic loading terms. Response’s FS

Example
We can represent a loading p(t), of period Tp , using its Fourier series expansion Fourier
Transform
∞
∑ ∞
∑ 2π The DFT
p(t) = aj cos ωj t + bj sin ωj t, ωj = j ω 1 = j .
j=0 j=0
Tp General Load

H(ω) vs h(t)
Fourier Series

Considering the orthogonality relationship over the period Tp , SDOF linear

oscillator

∫ Tp ∫ Tp
Giacomo Boffi
Tp
cos ωi t cos ωj t dt = sin ωi t sin ωj t dt = δij , Response
0 0 2 Intro

∫ Tp
Fourier Series
Response’s FS

cos ωi t sin ωj t dt = 0, for i, j = 0, . . . , ∞ Example

0 Fourier
Transform

The DFT

General Load

H(ω) vs h(t)
Fourier Series

Considering the orthogonality relationship over the period Tp , SDOF linear

oscillator

∫ Tp ∫ Tp
Giacomo Boffi
Tp
cos ωi t cos ωj t dt = sin ωi t sin ωj t dt = δij , Response
0 0 2 Intro

∫ Tp
Fourier Series
Response’s FS

cos ωi t sin ωj t dt = 0, for i, j = 0, . . . , ∞ Example

0 Fourier
Transform

the amplitudes of the harmonic components are The DFT

General Load
∫ Tp ∫ Tp
1 2 H(ω) vs h(t)
a0 = p(t) dt, aj = p(t) cos ωj t dt,
Tp 0 Tp 0
∫ Tp
2
b0 = 0 bj = p(t) sin ωj t dt.
Tp 0

Note that the case i = j = 0 is a special case...

Fourier Coefficients

SDOF linear
oscillator
If p(t) has not an analytical representation and must be measured experimentally or Giacomo Boffi
computed numerically, we may assume that it is possible
Response

(a) to divide the period in N equal parts ∆t = Tp /N , Intro

Fourier Series
Response’s FS
(b) measure or compute p(t) at a discrete set of instants t1 , t2 , . . . , tN , with tm = m∆t, Example

Fourier
obtaining a discrete set of values pm , m = 1, . . . , N (note that p0 = pN by periodicity). Transform

The DFT

General Load

H(ω) vs h(t)
Fourier Coefficients

SDOF linear
oscillator
If p(t) has not an analytical representation and must be measured experimentally or Giacomo Boffi
computed numerically, we may assume that it is possible
Response

(a) to divide the period in N equal parts ∆t = Tp /N , Intro

Fourier Series
Response’s FS
(b) measure or compute p(t) at a discrete set of instants t1 , t2 , . . . , tN , with tm = m∆t, Example

Fourier
obtaining a discrete set of values pm , m = 1, . . . , N (note that p0 = pN by periodicity). Transform

The DFT
Under these assumptions the, e.g., cosine‐wave amplitude coefficients can be
General Load
approximated using the trapezoidal rule of integration:
H(ω) vs h(t)

2∆t ∑ 2 ∑ 2 ∑
N N N
jm 2π
aj ≊ pm cos ωj tm = pm cos(jω1 m∆t) = pm cos .
Tp m=1 N m=1 N m=1 N
Periodicity

SDOF linear
oscillator
The coefficients of the Discrete Fourier Transform are periodic Giacomo Boffi

with period N . Response

Intro

Or, in other words, aj+N = aj for ∀j.

Fourier Series
Response’s FS
Example

Fourier
Transform

The DFT

General Load

H(ω) vs h(t)
Periodicity

SDOF linear
oscillator
The coefficients of the Discrete Fourier Transform are periodic Giacomo Boffi

with period N . Response

Intro

Or, in other words, aj+N = aj for ∀j.

Fourier Series
Response’s FS
Example

To prove the previous statement we’ll compute aj+N according to its definition: Fourier
Transform

The DFT
2 ∑ 2 ∑
N N
2(j + N )mπ 2(jm + N m)π General Load
aj+N = pm cos = pm cos
N m=1 N N m=1 N H(ω) vs h(t)

( )
2 ∑ 2 ∑
N N
2jmπ 2 jm π
= pm cos + 2mπ = pm cos = aj
N m=1 N N m=1 N
Exponential Form

The Fourier series can also be written in terms of exponentials of imaginary argument, SDOF linear
oscillator
∞
∑ Giacomo Boffi
p(t) = Pj exp iωj t
Response
j=−∞ Intro
Fourier Series

where the complex amplitude coefficients are given by Response’s FS

Example

∫ Tp Fourier
1 Transform
Pj = p(t) exp −iωj t dt, j = −∞, . . . , +∞.
Tp 0 The DFT

General Load

H(ω) vs h(t)
Exponential Form

The Fourier series can also be written in terms of exponentials of imaginary argument, SDOF linear
oscillator
∞
∑ Giacomo Boffi
p(t) = Pj exp iωj t
Response
j=−∞ Intro
Fourier Series

where the complex amplitude coefficients are given by Response’s FS

Example

∫ Tp Fourier
1 Transform
Pj = p(t) exp −iωj t dt, j = −∞, . . . , +∞.
Tp 0 The DFT

General Load
For a sampled pm we can write, using the trapezoidal integration rule and substituting H(ω) vs h(t)
tm = m∆t = m Tp /N , ωj = j 2π/Tp :

1 ∑
N
2π j m
Pj ≊ pm exp(−i ).
N m=1 N

For sampled input also the coefficients of the exponential series are periodic, Pj+N = Pj .
Undamped Response

We have seen that the steady‐state response to the jth sine‐wave harmonic can be SDOF linear
oscillator
written as [ ] Giacomo Boffi
bj 1
xj = sin ωj t, βj = ωj /ωn , Response
k 1 − βj2 Intro
Fourier Series
Response’s FS
analogously, for the jth cosine‐wave harmonic, Example

[ ] Fourier
Transform
aj 1
xj = cos ωj t. The DFT
k 1 − βj2 General Load

H(ω) vs h(t)
Undamped Response

We have seen that the steady‐state response to the jth sine‐wave harmonic can be SDOF linear
oscillator
written as [ ] Giacomo Boffi
bj 1
xj = sin ωj t, βj = ωj /ωn , Response
k 1 − βj2 Intro
Fourier Series
Response’s FS
analogously, for the jth cosine‐wave harmonic, Example

[ ] Fourier
Transform
aj 1
xj = cos ωj t. The DFT
k 1 − βj2 General Load

H(ω) vs h(t)
Finally, we write
 [ ] 
1  ∑∞
1 
x(t) = a0 + (aj cos ωj t + bj sin ωj t) .
k 1 − βj2 
j=1
Damped Response

SDOF linear
In the case of a damped oscillator, we must substitute the steady state response oscillator

for both the jth sine‐ and cosine‐wave harmonic, Giacomo Boffi

Response
∞
a0 1 ∑ +(1 − βj2 ) aj − 2ζβj bj Intro

x(t) = + cos ωj t+ Fourier Series

k k (1 − βj2 )2 + (2ζβj )2
j=1
Response’s FS
Example

∞
1 ∑ +2ζβj aj + (1 − βj2 ) bj
Fourier
Transform
+ sin ωj t.
k
j=1
(1 − βj2 )2 + (2ζβj )2 The DFT

General Load

H(ω) vs h(t)
Damped Response

SDOF linear
In the case of a damped oscillator, we must substitute the steady state response oscillator

for both the jth sine‐ and cosine‐wave harmonic, Giacomo Boffi

Response
∞
a0 1 ∑ +(1 − βj2 ) aj − 2ζβj bj Intro

x(t) = + cos ωj t+ Fourier Series

k k (1 − βj2 )2 + (2ζβj )2
j=1
Response’s FS
Example

∞
1 ∑ +2ζβj aj + (1 − βj2 ) bj
Fourier
Transform
+ sin ωj t.
k
j=1
(1 − βj2 )2 + (2ζβj )2 The DFT

General Load

H(ω) vs h(t)
As usual, the exponential notation is neater,

∑∞
Pj exp iωj t
x(t) = .
k (1 − βj2 ) + i (2ζβj )
j=−∞
Example

SDOF linear
As an example, consider the loading p(t) = max{p0 sin 2πt
Tp , 0} oscillator
Giacomo Boffi

p0 max[sin(2 π t/Tp),0.0]
Response
Intro
p0
Fourier Series
Response’s FS
Example

Fourier
Transform

The DFT
0.5 p0
General Load

H(ω) vs h(t)

0.0 0.5 Tp T 1.5 Tp 2Tp

Example

SDOF linear
As an example, consider the loading p(t) = max{p0 sin 2πt
Tp , 0} oscillator
Giacomo Boffi

∫ Tp /2 Response
1 2πt p0 Intro
a0 = po sin dt = , Fourier Series
Tp 0 Tp π Response’s FS

∫ { Example

2 Tp /2
2πt 2πjt 0 for j odd
dt = p0 [ 2 ]
Fourier
aj = po sin cos Transform
Tp 0 Tp Tp π 1−j 2 for j even, The DFT

∫ Tp /2 { General Load
p0
2 2πt 2πjt for j = 1
bj = po sin sin dt = 2 H(ω) vs h(t)
Tp 0 Tp Tp 0 for n > 1.
Example cont.

SDOF linear
oscillator
( )
sin ω1 t − 23 cos 2ω1 t − cos 4ω2 t − . . . with the
p0 π 2 Giacomo Boffi
Assuming β1 = 3/4, from p = π 1+ 2 15
dynamic amplifiction factors Response
Intro

1 16 1 4 1 1 Fourier Series

D1 = = , D2 = =− , D4 = =− , D6 = . . . Response’s FS

1 − (1 34 )2 7 1 − (2 34 )2 5 1 − (4 34 )2 8 Example

Fourier
Transform
etc, we have
The DFT
( )
p0 8π 8 1 General Load
x(t) = 1+ sin ω1 t + cos 2ω1 t + cos 4ω1 t + . . .
kπ 7 15 60 H(ω) vs h(t)
Example cont.

SDOF linear
oscillator
( )
sin ω1 t − 23 cos 2ω1 t − cos 4ω2 t − . . . with the
p0 π 2 Giacomo Boffi
Assuming β1 = 3/4, from p = π 1+ 2 15
dynamic amplifiction factors Response
Intro

1 16 1 4 1 1 Fourier Series

D1 = = , D2 = =− , D4 = =− , D6 = . . . Response’s FS

1 − (1 34 )2 7 1 − (2 34 )2 5 1 − (4 34 )2 8 Example

Fourier
Transform
etc, we have
The DFT
( )
p0 8π 8 1 General Load
x(t) = 1+ sin ω1 t + cos 2ω1 t + cos 4ω1 t + . . .
kπ 7 15 60 H(ω) vs h(t)

Take note, these solutions are particular solutions! If your solution has to respect given
initial conditions, you must consider also the homogeneous solution.
Example cont.

SDOF linear
oscillator
xi = Σj=1,..,i aj cosωjt + bj sinωjt Giacomo Boffi
5
x0 Response
4 x1 Intro

x2 Fourier Series
3 x4 Response’s FS
Example
2
x(t) k π / po

Fourier
1 Transform

The DFT
0
General Load
-1
H(ω) vs h(t)
-2

-3

-4
0 1 2 3
t/Tp
The Fourier Transform

SDOF linear
oscillator
Giacomo Boffi
Response to Periodic Loading
Response

Fourier
Fourier Transform Transform
Extension of Fourier Series to non periodic functions Non Periodic
Functions

Response in the Frequency Domain H (ω )

The DFT

General Load
The Discrete Fourier Transform H(ω) vs h(t)

Response to General Dynamic Loadings

Relationship between Analysis in Time and Frequency Domains

Non periodic loadings

It is possible to extend the Fourier analysis to non periodic loading. Let’s start from SDOF linear
oscillator
the Fourier series representation of the load p(t), Giacomo Boffi
∑
+∞
2π
p(t) = Pr exp(iωr t), ωr = r∆ω, ∆ω = , Response

−∞
Tp Fourier
Transform
Non Periodic
Functions
H (ω )

The DFT

General Load

H(ω) vs h(t)
Non periodic loadings

It is possible to extend the Fourier analysis to non periodic loading. Let’s start from SDOF linear
oscillator
the Fourier series representation of the load p(t), Giacomo Boffi
∑
+∞
2π
p(t) = Pr exp(iωr t), ωr = r∆ω, ∆ω = , Response

−∞
Tp Fourier
Transform
Non Periodic
introducing P (iωr ) = Pr Tp and substituting, Functions
H (ω )

1 ∑ ∆ω ∑
+∞ +∞
The DFT
p(t) = P (iωr ) exp(iωr t) = P (iωr ) exp(iωr t). General Load
Tp −∞ 2π −∞
H(ω) vs h(t)
Non periodic loadings

It is possible to extend the Fourier analysis to non periodic loading. Let’s start from SDOF linear
oscillator
the Fourier series representation of the load p(t), Giacomo Boffi
∑
+∞
2π
p(t) = Pr exp(iωr t), ωr = r∆ω, ∆ω = , Response

−∞
Tp Fourier
Transform
Non Periodic
introducing P (iωr ) = Pr Tp and substituting, Functions
H (ω )

1 ∑ ∆ω ∑
+∞ +∞
The DFT
p(t) = P (iωr ) exp(iωr t) = P (iωr ) exp(iωr t). General Load
Tp −∞ 2π −∞
H(ω) vs h(t)

Due to periodicity, we can modify the extremes of integration in the expression for
the complex amplitudes,
∫ +Tp /2
P (iωr ) = p(t) exp(−iωr t) dt.
−Tp /2
Non periodic loadings (2)

SDOF linear
oscillator
Giacomo Boffi
If the loading period is extended to infinity to represent the non‐periodicity of the loading (Tp → ∞)
then (a) the frequency increment becomes infinitesimal (∆ω = T2πp → dω) and (b) the discrete Response

Fourier
frequency ωr becomes a continuous variable, ω. Transform
In the limit, for Tp → ∞ we can then write Non Periodic
Functions
∫ +∞ H (ω )
1
p(t) = P (iω) exp(iωt) dω The DFT
2π −∞
∫ +∞ General Load

P (iω) = p(t) exp(−iωt) dt, H(ω) vs h(t)

−∞

which are known as the inverse and the direct Fourier Transforms, respectively, and are collectively
known as the Fourier transform pair.
SDOF Response

SDOF linear
oscillator
Giacomo Boffi

In analogy to what we have seen for periodic loads, the response of a damped SDOF system can be Response
written in terms of H(iω), the complex frequency response function,
Fourier
∫ +∞ Transform
1 Non Periodic
x(t) = H(iω) P (iω) exp iωt dt, where Functions
2π −∞ H (ω )

[ ] [ ] The DFT
1 1 1 (1 − β 2 ) − i(2ζβ) ω
H(iω) = = , β= . General Load
k (1 − β 2 ) + i(2ζβ) k (1 − β 2 )2 + (2ζβ)2 ωn
H(ω) vs h(t)
To obtain the response through frequency domain, you should evaluate the above integral, but
analytical integration is not always possible, and when it is possible, it is usually very difficult, implying
contour integration in the complex plane (for an example, see Example E6‐3 in Clough Penzien).
The Discrete Fourier Transform

SDOF linear
oscillator
Response to Periodic Loading Giacomo Boffi

Response

Fourier Transform Fourier

Transform

The DFT
The Discrete Fourier Transform The DFT
Aliasing

The Discrete Fourier Transform The FFT

Aliasing General Load

H(ω) vs h(t)
The Fast Fourier Transform

Response to General Dynamic Loadings

Relationship between Analysis in Time and Frequency Domains

Discrete Fourier Transform

SDOF linear
oscillator
Giacomo Boffi

Response

Fourier
Transform
To overcome the analytical difficulties associated with the inverse Fourier The DFT

transform, one can use appropriate numerical methods, leading to good The DFT
Aliasing

approximations. The FFT

General Load

H(ω) vs h(t)
Discrete Fourier Transform

Consider a loading of finite period Tp , p(t) = p(t + Tp ), it is SDOF linear

oscillator
∫ Tp Giacomo Boffi
1
Pr = p(t) exp(−iωr t) dt, r = −∞, . . . , ∞.
Tp 0 Response

Fourier
Transform

The DFT
The DFT
Aliasing
The FFT

General Load

H(ω) vs h(t)
Discrete Fourier Transform

Consider a loading of finite period Tp , p(t) = p(t + Tp ), it is SDOF linear

oscillator
∫ Tp Giacomo Boffi
1
Pr = p(t) exp(−iωr t) dt, r = −∞, . . . , ∞.
Tp 0 Response

Fourier
Transform
Now, think that Tp has been divided in N equal intervals ∆t = Tp /N and the only thing we
The DFT
know about p(t) is the set of N values ps = p(ts ) = p(s∆t). The DFT
Aliasing
It is still possible to compute the approximate value of every complex amplitude The FFT

coefficients using the trapezoidal rule of integration. General Load

H(ω) vs h(t)
Discrete Fourier Transform

Consider a loading of finite period Tp , p(t) = p(t + Tp ), it is SDOF linear

oscillator
∫ Tp Giacomo Boffi
1
Pr = p(t) exp(−iωr t) dt, r = −∞, . . . , ∞.
Tp 0 Response

Fourier
Transform
Now, think that Tp has been divided in N equal intervals ∆t = Tp /N and the only thing we
The DFT
know about p(t) is the set of N values ps = p(ts ) = p(s∆t). The DFT
Aliasing
It is still possible to compute the approximate value of every complex amplitude The FFT

coefficients using the trapezoidal rule of integration. General Load

Taking into account that, by peridiocity, p0 ≡ pN , it is H(ω) vs h(t)
∫ Tp
1
Pr = p(t) exp(−iωr t) dt
Tp 0
( N −1 ) N −1
1 ∑ 1 ∑ 2πrs
≊ ∆t ps exp(−iωr ts ) = ps exp(−i ).
N ∆t s=0
N s=0 N
Discrete Fourier Transform

SDOF linear
In the last two passages we have used the following identities oscillator
Giacomo Boffi
pN = p0 , exp(iωr tN ) = exp(iωr t0 ),
Response
2π Tp 2π rs
ωr ts = r∆ω s∆t = rs = . Fourier
Tp N N Transform

The DFT

Take note that the discrete function exp(−i 2πrs

The DFT

N ), defined for integer r, s is Aliasing

periodic with period N , implying that the complex amplitude coefficients are The FFT

themselves periodic with period N . General Load

H(ω) vs h(t)

Pr+N = Pr
Starting in the time domain with N distinct complex numbers, ps , we have found
that in the frequency domain our load is described by N distinct complex numbers,
Pr , so that we can say that our function is described by the same amount of
information in both domains.
Aliasing

Only N/2 distinct frequencies sin(21 * (2π)/Tp * s Tp/N), N=20, s=0,..,20 SDOF linear
∑ −1 ∑+N/2 sin(22 * (2π)/Tp * s Tp/N), N=20, s=0,..,20
oscillator
( N 0 = −N/2 ) contribute
1
Giacomo Boffi
to the load representation, what
Response
if the frequency content of the 0.5
Fourier
loading has contributions from Transform

frequencies higher than ωN/2 ? 0 The DFT

The DFT
What happens is aliasing, i.e., the Aliasing
The FFT
upper frequencies contributions -0.5
General Load
are mapped to contributions of H(ω) vs h(t)
lesser frequency. -1

0 1/4 Tp
Aliasing

Only N/2 distinct frequencies sin(21 * (2π)/Tp * s Tp/N), N=20, s=0,..,20 SDOF linear
∑ −1 ∑+N/2 sin(22 * (2π)/Tp * s Tp/N), N=20, s=0,..,20
oscillator
( N 0 = −N/2 ) contribute
1
Giacomo Boffi
to the load representation, what
Response
if the frequency content of the 0.5
Fourier
loading has contributions from Transform

frequencies higher than ωN/2 ? 0 The DFT

The DFT
What happens is aliasing, i.e., the Aliasing
The FFT
upper frequencies contributions -0.5
General Load
are mapped to contributions of H(ω) vs h(t)
lesser frequency. -1

0 1/4 Tp

See the plot above: the contributions from the high frequency sines, when
sampled, are indistinguishable from the contributions from lower frequency
components, i.e., are aliased to lower frequencies!
Aliasing (2)

SDOF linear
oscillator
Giacomo Boffi

Response
The maximum frequency that can be described in the DFT is called the Nyquist Fourier
2π
frequency, ωNy = 12 ∆t . Transform

The DFT
It is usual in signal analysis to remove the signal’s higher frequency The DFT

components preprocessing the signal with a filter or a digital filter. Aliasing

The FFT

It is worth noting that the resolution of the DFT in the frequency domain for a General Load

given sampling rate is proportional to the number of samples, i.e., to the H(ω) vs h(t)

duration of the sample.

The Fast Fourier Transform

SDOF linear
oscillator
Giacomo Boffi

The operation count in a DFT is in the order of N2 . Response

A Fast Fourier Transform is an algorithm that reduces the number of arithmetic Fourier
Transform
operations needed to compute a DFT.
The DFT
The first and simpler FFT algorithm is the Decimation in Time algorithm by Cooley The DFT
Aliasing
and Tukey (1965). The FFT

General Load

H(ω) vs h(t)
The Fast Fourier Transform

SDOF linear
oscillator
Giacomo Boffi

The operation count in a DFT is in the order of N2 . Response

A Fast Fourier Transform is an algorithm that reduces the number of arithmetic Fourier
Transform
operations needed to compute a DFT.
The DFT
The first and simpler FFT algorithm is the Decimation in Time algorithm by Cooley The DFT
Aliasing
and Tukey (1965). The FFT

The algorithm introduced by Cooley and Tukey is quite complex because it allows General Load

to proceed without additional memory, we will describe a different algorithm, that H(ω) vs h(t)

is based on the same principles but requires additional memory and it’s rather
simpler than the original one.
Decimation in Time DFT Algorithm

For simplicity, assume that N is even and split the DFT summation in two separate sums, SDOF linear
oscillator
with even and odd indices
Giacomo Boffi

∑
N −1
xs e−
2πi Response
Xr = N sr , r = 0, . . . , N − 1
Fourier
s=0 Transform

∑
N/2−1
∑
N/2−1 The DFT
x2q e− x2q+1 e−
2πi 2πi
= N (2q)r + N (2q+1)r . The DFT
Aliasing
q=0 q=0 The FFT

− 2πi
N r
2q q General Load
Collecting e in the second term and letting N = N/2 , we have
H(ω) vs h(t)
Decimation in Time DFT Algorithm

For simplicity, assume that N is even and split the DFT summation in two separate sums, SDOF linear
oscillator
with even and odd indices
Giacomo Boffi

∑
N −1
xs e−
2πi Response
Xr = N sr , r = 0, . . . , N − 1
Fourier
s=0 Transform

∑
N/2−1
∑
N/2−1 The DFT
x2q e− x2q+1 e−
2πi 2πi
= N (2q)r + N (2q+1)r . The DFT
Aliasing
q=0 q=0 The FFT

− 2πi
N r
2q q General Load
Collecting e in the second term and letting N = N/2 , we have
H(ω) vs h(t)

∑
N/2−1
∑
N/2−1
x2q e− N/2 qr + e− x2q+1 e− N/2 qr ,
2πi 2πi 2πi
Xr = N r

q=0 q=0

i.e., we have two DFT’s of length N/2. The operations count is just 2(N/2)2 = N 2 /2, but
we have to combine these two halves in the full DFT.
Decimation in Time DFT
Say that
SDOF linear
Xr = Er + e−
2πi
N r Or oscillator
Giacomo Boffi
where Er and Or are the even and odd half‐DFT’s, of which we computed only coefficients
from 0 to N/2 − 1. Response

To get the full sequence we have to note that Fourier

Transform

1 the E and O DFT’s are periodic with period N/2, and The DFT
The DFT
−πi
2 exp(−2πi(r + N/2)/N ) = e exp(−2πir/N ) = − exp(−2πir/N ), Aliasing
The FFT

so that we can write General Load

{ H(ω) vs h(t)
Er + exp(−2πir/N )Or if r < N/2,
Xr =
Er−N/2 − exp(−2πir/N )Or−N/2 if r ≥ N/2.

The algorithm that was outlined can be applied to the computation of each of the
half‐DFT’s when N/2 were even, so that the operation count goes to N 2 /4. If N/4 were
even ...
Pseudocode for CT algorithm

SDOF linear
oscillator

def fft2(X, N): Giacomo Boffi

if N = 1 then Response

Y = X Fourier
Transform
else
The DFT
Y0 = fft2(X0, N/2) The DFT
Aliasing
Y1 = fft2(X1, N/2) The FFT

for k = 0 to N/2‐1 General Load

Y_k = Y0_k + exp(2 pi i k/N) Y1_k H(ω) vs h(t)

Y_(k+N/2) = Y0_k ‐ exp(2 pi i k/N) Y1_k

endfor
endif
return Y
from cmath import exp, pi

def d_fft(x,n):
”Direct fft of x, a list of n=2**m complex values”
return _fft(x,n,[exp(‐2*pi*1j*k/n) for k in range(n/2)])

def i_fft(x,n):
”Inverse fft of x, a list of n=2**m complex values”
transform = _fft(x,n,[exp(+2*pi*1j*k/n) for k in range(n/2)])]
return [x/n for x in transform]
def _fft(x, n, tw):
”””Decimation in Time FFT, to be called by d_fft and i_fft.
x is the signal to transform , a list of complex values
n is its length , results are undefined if n is not a power of 2
tw is a list of twiddle factors , precomputed by the caller

returns a list of complex values ,not normalized if inverse transform”””

if n == 1: return x # bottom reached , DFT of a length 1 vec x is x

# call fft with the even and the odd coefficients in x

# the results are the so called even and odd DFT’s
e, o = _fft(x[0::2], n/2, tw[::2]), _fft(x[1::2], n/2, tw[::2])

# assemble the partial results:

# 1st half of full DFT is put in even DFT, 2nd half in odd DFT
for k in range(n/2):
e[k], o[k] = e[k]+tw[k]*o[k], e[k]‐tw[k]*o[k]

# concatenate the two halves of the DFT and return to caller

return e + o
If we strip all comments, our FFT function becomes
def _fft(x, n, tw):
if n==1:return x
e,o=_fft(x[0::2],n/2,tw[::2]),_fft(x[1::2],n/2,tw[::2])
for k in range(n/2):e[k],o[k]=e[k]+tw[k]*o[k],e[k]‐tw[k]*o[k]
return e+o
Dynamic Response (1)

SDOF linear
To evaluate the dynamic response of a linear SDOF system in the frequency domain, use oscillator
the inverse DFT, Giacomo Boffi
∑
N −1
2π rs
xs = Vr exp(i ), s = 0, 1, . . . , N − 1 Response

r=0
N Fourier
Transform
where Vr = Hr Pr . Pr are the discrete complex amplitude coefficients computed using the
The DFT
direct DFT, and Hr is the discretization of the complex frequency response function, that The DFT

for viscous damping is Aliasing

[ ] [ ]
The FFT

1 1 1 (1 − βr2 ) − i(2ζβr ) ωr General Load

Hr = = , βr = .
k (1 − βr2 ) + i(2ζβr ) k (1 − βr2 )2 + (2ζβr )2 ωn H(ω) vs h(t)

while for hysteretic damping it is

[ ] [ ]
1 1 1 (1 − βr2 ) − i(2ζ)
Hr = = .
k (1 − βr2 ) + i(2ζ) k (1 − βr2 )2 + (2ζ)2
Dynamic Response (2)

SDOF linear
oscillator

Some word of caution... Giacomo Boffi

If you’re going to approach the application of the complex frequency response function Response
without proper concern, you’re likely to be hurt. Fourier
Transform

The DFT
The DFT
Aliasing
The FFT

General Load

H(ω) vs h(t)
Dynamic Response (2)

SDOF linear
oscillator

Some word of caution... Giacomo Boffi

If you’re going to approach the application of the complex frequency response function Response
without proper concern, you’re likely to be hurt. Fourier
Transform
Let’s say ∆ω = 1.0, N = 32, ωn = 3.5 and r = 30, what do you think it is the value of β30 ?
The DFT
The DFT
Aliasing
The FFT

General Load

H(ω) vs h(t)
Dynamic Response (2)

SDOF linear
oscillator

Some word of caution... Giacomo Boffi

If you’re going to approach the application of the complex frequency response function Response
without proper concern, you’re likely to be hurt. Fourier
Transform
Let’s say ∆ω = 1.0, N = 32, ωn = 3.5 and r = 30, what do you think it is the value of β30 ?
If you are thinking β30 = 30 ∆ω/ωn = 30/3.5 ≈ 8.57 you’re wrong! The DFT
The DFT
Aliasing
The FFT

General Load

H(ω) vs h(t)
Dynamic Response (2)

SDOF linear
oscillator

Some word of caution... Giacomo Boffi

If you’re going to approach the application of the complex frequency response function Response
without proper concern, you’re likely to be hurt. Fourier
Transform
Let’s say ∆ω = 1.0, N = 32, ωn = 3.5 and r = 30, what do you think it is the value of β30 ?
If you are thinking β30{= 30 ∆ω/ωn = 30/3.5 ≈ 8.57 you’re wrong! The DFT
The DFT

r∆ω r ≤ N/2 Aliasing

Due to aliasing, ωr = , The FFT

(r − N )∆ω r > N/2 General Load

note that in the upper part of the DFT the coefficients correspond to negative frequencies H(ω) vs h(t)
and, staying within our example, it is β30 = (30 − 32) × 1/3.5 ≈ −0.571.
If N is even, PN/2 is the coefficient corresponding to the Nyquist frequency, if N is odd
P N −1 corresponds to the largest positive frequency, while P N +1 corresponds to the
2 2
largest negative frequency.
Response to General Dynamic Loading

SDOF linear
oscillator
Response to Periodic Loading Giacomo Boffi

Response

Fourier Transform Fourier

Transform

The DFT
The Discrete Fourier Transform General Load
h(t)
Duhamel
Numerical Duhamel
Response to General Dynamic Loadings
H(ω) vs h(t)
Response to infinitesimal impulse
Duhamel Integral
Numerical integration of Duhamel integral

Relationship between Analysis in Time and Frequency Domains

Response to a short duration load

SDOF linear
oscillator
Giacomo Boffi
An approximate procedure to evaluate the maximum displacement for a short
impulse loading is based on the impulse‐momentum relationship, Response

∫ t0
Fourier
Transform

m∆ẋ = [p(t) − kx(t)] dt. The DFT

0 General Load
h(t)

When one notes that, for small t0 , the displacement is of the order of t20 while the Duhamel
Numerical Duhamel

velocity is in the order of t0 , it is apparent that the kx term may be dropped from H(ω) vs h(t)
the above expression, i.e., ∫ t0
m∆ẋ ≊ p(t) dt.
0
Response to a short duration load

SDOF linear
Using the previous approximation, the velocity at time t0 is oscillator

∫ t0
Giacomo Boffi
1
ẋ(t0 ) = p(t) dt, Response
m 0 Fourier
Transform
and considering again a negligibly small displacement at the end of the loading, The DFT
x(t0 ) ≊ 0, one has General Load
h(t)
∫ t0 Duhamel
1 Numerical Duhamel
x(t − t0 ) ≊ p(t) dt sin ωn (t − t0 ).
mωn 0
H(ω) vs h(t)

Please note that the above equation is exact for an infinitesimal impulse loading.
Response to a short duration load

SDOF linear
Using the previous approximation, the velocity at time t0 is oscillator

∫ t0
Giacomo Boffi
1
ẋ(t0 ) = p(t) dt, Response
m 0 Fourier
Transform
and considering again a negligibly small displacement at the end of the loading, The DFT
x(t0 ) ≊ 0, one has General Load
h(t)
∫ t0 Duhamel
1 Numerical Duhamel
x(t − t0 ) ≊ p(t) dt sin ωn (t − t0 ).
mωn 0
H(ω) vs h(t)

Please note that the above equation is exact for an infinitesimal impulse loading.

p(τ ) dτ
dx(t − τ ) = sin ωn (t − τ ), t > τ,
mωn
Undamped SDOF

SDOF linear
oscillator
For an infinitesimal impulse, the impulse‐momentum is exactly p(τ ) dτ and the
Giacomo Boffi
response is
p(τ ) dτ Response
dx(t − τ ) = sin ωn (t − τ ), t > τ , Fourier
mωn Transform

and to evaluate the response at time t one has simply to sum all the infinitesimal The DFT

contributions for τ < t, General Load

h(t)
∫ t Duhamel

1 Numerical Duhamel

x(t) = p(τ ) sin ωn (t − τ ) dτ, t > 0. H(ω) vs h(t)

mωn 0

This relation is known as the Duhamel integral, and tacitly depends on initial rest
conditions for the system.
Undamped SDOF

SDOF linear
oscillator
For an infinitesimal impulse, the impulse‐momentum is exactly p(τ ) dτ and the
Giacomo Boffi
response is
p(τ ) dτ Response
dx(t − τ ) = sin ωn (t − τ ), t > τ , Fourier
mωn Transform

and to evaluate the response at time t one has simply to sum all the infinitesimal The DFT

contributions for τ < t, General Load

h(t)
∫ t Duhamel

1 Numerical Duhamel

x(t) = p(τ ) sin ωn (t − τ ) dτ, t > 0. H(ω) vs h(t)

mωn 0

This relation is known as the Duhamel integral, and tacitly depends on initial rest
conditions for the system.
Jean‐Marie Constant Duhamel (Saint‐Malo, 5 February 1797 — Paris, 29 April 1872)
Damped SDOF

SDOF linear
oscillator
Giacomo Boffi
The derivation of the equation of motion for a generic load is analogous to what
we have seen for undamped SDOF, the infinitesimal contribution to the response Response

at time t of the load at time τ is Fourier

Transform

The DFT
p(τ )
dx(t) = dτ sin ωD (t − τ ) exp(−ζωn (t − τ )) t ≥ τ General Load
mωD h(t)
Duhamel
Numerical Duhamel
and integrating all infinitesimal contributions one has
H(ω) vs h(t)
∫ t
1
x(t) = p(τ ) sin ωD (t − τ ) exp(−ζωn (t − τ )) dτ, t ≥ 0.
mωD 0
Evaluation of Duhamel integral, undamped

SDOF linear
oscillator
Using the trigonomtric identity
Giacomo Boffi

sin(ωn t − ωn τ ) = sin ωn t cos ωn τ − cos ωn t sin ωn τ Response

Fourier
Transform
the Duhamel integral is rewritten as
The DFT
∫t ∫t General Load
p(τ ) cos ωn τ dτ p(τ ) sin ωn τ dτ
x(t) = 0
sin ωn t − 0
cos ωn t h(t)
Duhamel
mωn mωn Numerical Duhamel

= A(t) sin ωn t − B(t) cos ωn t

Undamped SDOF
systems
Damped SDOF
systems

where { ∫t H(ω) vs h(t)

A(t) = mω
1
n ∫0
p(τ ) cos ωn τ dτ
t
B(t) = mωn 0 p(τ ) sin ωn τ dτ
1
Numerical evaluation of Duhamel integral, undamped

SDOF linear
oscillator
Giacomo Boffi

Usual numerical procedures can be applied to the evaluation of A and B, e.g., Response

using the trapezoidal rule, one can have, with An = A(n∆τ ), Fourier
Transform
yn = p(n∆τ ) cos(n∆τ ) and zn = p(n∆τ ) sin(n∆τ ) we can write The DFT

General Load
∆τ
An+1 = An + (yn + yn+1 ) , h(t)
Duhamel
2mωn Numerical Duhamel

∆τ Undamped SDOF

Bn+1 = Bn +
systems
(zn + zn+1 ) . Damped SDOF
2mωn systems

H(ω) vs h(t)
Evaluation of Duhamel integral, damped

SDOF linear
oscillator
Giacomo Boffi
For a damped system, it can be shown that
Response

x(t) = A(t) sin ωD t − B(t) cos ωD t Fourier

Transform

The DFT
with General Load
∫ h(t)
exp −ζωn t t Duhamel

A(t) = p(τ ) exp ζωn τ cos ωD τ dτ , Numerical Duhamel

mωD 0
Undamped SDOF

∫
systems

exp −ζωn t t
Damped SDOF
systems

B(t) = p(τ ) exp ζωn τ sin ωD τ dτ . H(ω) vs h(t)

mωD 0
Numerical evaluation of Duhamel integral, damped

SDOF linear
oscillator
Giacomo Boffi

Numerically, using e.g. Simpson integration rule and yn = p(n∆τ ) cos ωD τ , Response

Fourier
Transform
An+2 = An exp(−2ζωn ∆τ )+ The DFT

∆τ General Load
[yn exp(−2ζωn ∆τ ) + 4yn+1 exp(−ζωn ∆τ ) + yn+2 ] h(t)
3mωD Duhamel
Numerical Duhamel

n = 0, 2, 4, · · · Undamped SDOF
systems
Damped SDOF
systems

(You can write a similar relationship for Bn+2 ) H(ω) vs h(t)

The Time and the Frequency Domains

SDOF linear
oscillator
Giacomo Boffi
Response to Periodic Loading
Response

Fourier
Transform
Fourier Transform
The DFT

General Load
The Discrete Fourier Transform H(ω) vs h(t)

Response to General Dynamic Loadings

Relationship between Analysis in Time and Frequency Domains

Transfer Functions

SDOF linear
oscillator
The response of a linear SDOF system to arbitrary loading can be evaluated by a Giacomo Boffi

convolution integral in the time domain, Response

∫ t Fourier
Transform
x(t) = p(τ ) h(t − τ ) dτ,
The DFT
0
General Load

with the unit impulse response function h(t) = mω1 D exp(−ζωn t) sin(ωD t), or H(ω) vs h(t)

through the frequency domain using the Fourier integral

∫ +∞
x(t) = H(ω)P (ω) exp(iωt) dω,
−∞

where H(ω) is the complex frequency response function.

Transfer Functions

SDOF linear
oscillator
Giacomo Boffi

Response
These response functions, or transfer functions, are connected by the direct and
Fourier
inverse Fourier transforms: Transform

∫ +∞ The DFT

H(ω) = h(t) exp(−iωt) dt, General Load

−∞ H(ω) vs h(t)
∫ +∞
1
h(t) = H(ω) exp(iωt) dω.
2π −∞
Relationship of transfer functions

SDOF linear
We write the response and its Fourier transform: oscillator

∫ t ∫ t Giacomo Boffi

x(t) = p(τ )h(t − τ ) dτ = p(τ )h(t − τ ) dτ Response

0 −∞
∫ +∞ ∫ +∞ [∫ t ] Fourier
Transform
X(ω) = x(t) exp(−iωt) dt = p(τ )h(t − τ ) dτ exp(−iωt) dt The DFT
−∞ −∞ −∞ General Load

H(ω) vs h(t)
where we changed the lower limit of integration, in the first equation, from 0 to
−∞ because p(τ ) = 0 for τ < 0.
Relationship of transfer functions

SDOF linear
We write the response and its Fourier transform: oscillator

∫ t ∫ t Giacomo Boffi

x(t) = p(τ )h(t − τ ) dτ = p(τ )h(t − τ ) dτ Response

0 −∞
∫ +∞ ∫ +∞ [∫ t ] Fourier
Transform
X(ω) = x(t) exp(−iωt) dt = p(τ )h(t − τ ) dτ exp(−iωt) dt The DFT
−∞ −∞ −∞ General Load

H(ω) vs h(t)
where we changed the lower limit of integration, in the first equation, from 0 to
−∞ because p(τ ) = 0 for τ < 0.
Since h(t − τ ) = 0 for τ > t, the upper limit of the second integral in the second
equation can be changed from t to +∞,
∫ +s ∫ +s
X(ω) = lim p(τ )h(t − τ ) exp(−iωt) dt dτ
s→∞ −s −s
Relationship of transfer functions

SDOF linear
oscillator
Introducing a new variable θ = t − τ we have Giacomo Boffi
∫ +s ∫ +s−τ Response
X(ω) = lim p(τ ) exp(−iωτ ) dτ h(θ) exp(−iωθ) dθ Fourier
s→∞ −s −s−τ Transform

The DFT
with lim s − τ = ∞, we finally have General Load
s→∞
∫ +∞ ∫ +∞
H(ω) vs h(t)

X(ω) = p(τ ) exp(−iωτ ) dτ h(θ) exp(−iωθ) dθ

−∞ −∞
∫ +∞
= P (ω) h(θ) exp(−iωθ) dθ = P (ω) H(ω)
−∞

where we have recognized that the first integral is the Fourier transform of p(t).
Relationship of transfer functions

SDOF linear
oscillator
Giacomo Boffi

Response
Our last relation was Fourier
∫ +∞
Transform

The DFT
X(ω) = P (ω) h(θ) exp(−iωθ) dθ
−∞ General Load

H(ω) vs h(t)
but X(ω) = H(ω)P (ω), so that, noting that in the above equation the last integral
is just the Fourier transform of h(θ), we may conclude that, effectively, H(ω) and
h(t) form a Fourier transform pair.