Lecture 8

ELE 301: Signals and Systems

Prof. Paul Cuff

Princeton University

Fall 2011-12

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 1 / 37

Properties of the Fourier Transform

Properties of the Fourier Transform

I Linearity
I Time-shift
I Time Scaling
I Conjugation
I Duality
I Parseval

Convolution and Modulation

Periodic Signals

Constant-Coefficient Differential Equations

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 2 / 37

Linearity
Linear combination of two signals x1 (t) and x2 (t) is a signal of the form
ax1 (t) + bx2 (t).
Linearity Theorem: The Fourier transform is linear; that is, given two
signals x1 (t) and x2 (t) and two complex numbers a and b, then

ax1 (t) + bx2 (t) ⇔ aX1 (jω) + bX2 (jω).

This follows from linearity of integrals:

Z ∞
(ax1 (t) + bx2 (t))e −j2πft dt
−∞
Z ∞ Z ∞
= a x1 (t)e −j2πft dt + b x2 (t)e −j2πft dt
−∞ −∞
= aX1 (f ) + bX2 (f )

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 3 / 37

Finite Sums

This easily extends to finite combinations. Given signals xk (t) with Fourier
transforms Xk (f ) and complex constants ak , k = 1, 2, . . . K , then
K
X K
X
ak xk (t) ⇔ ak Xk (f ).
k=1 k=1

If you consider a system which has a signal x(t) as its input and the
Fourier transform X (f ) as its output, the system is linear!

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 4 / 37

Linearity Example

Find the Fourier transform of the signal

 1 1
x(t) = 2 2 ≤ |t| < 1
1 |t| ≤ 12

This signal can be recognized as

1 t  1
x(t) = rect + rect (t)
2 2 2
and hence from linearity we have
 
1 1 1
X (f ) = 2 sinc(2f ) + sinc(f ) = sinc(2f ) + sinc(f )
2 2 2

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 5 / 37

1.2
1 1
1
rect(t/2) + rect(t)
0.8 2 2
0.6
0.4
0.2
0
!0.2
!2.5 !2 !1.5 !1 !0.5
−2 −1 00 0.5
11 1.5
22 2.5

2
1
sinc(ω/π) + sinc(ω/(2π))
1.5 2
1

0.5

!0.5
!10 !8 !6 !4 !2 0 2 4 6 8 10
−4π −2π 0 2π 4π
ω L

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 6 / 37

Scaling Theorem
Stretch (Scaling) Theorem: Given a transform pair x(t) ⇔ X (f ), and a
real-valued nonzero constant a,
 
1 f
x(at) ⇔ X
|a| a

Proof: Here consider only a > 0. (negative a left as an exercise) Change

variables τ = at
Z ∞ Z ∞  
dτ 1 f
x(at)e −j2πft dt = x(τ )e −j2πf τ /a = X .
−∞ −∞ a a a

If a = −1 ⇒ “time reversal theorem:”

X (−t) ⇔ X (−f )

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 7 / 37

Scaling Examples

We have already seen that rect(t/T ) ⇔ T sinc(Tf ) by brute force

integration. The scaling theorem provides a shortcut proof given the
simpler result rect(t) ⇔ sinc(f ).
This is a good point to illustrate a property of transform pairs. Consider
this Fourier transform pair for a small T and large T , say T = 1 and
T = 5. The resulting transform pairs are shown below to a common
horizontal scale:

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 8 / 37

Compress in time - Expand in frequency
1.2 6

1
rect(t) sinc(ω/2π)
4
0.8
0.6
2
0.4
0.2
0
0
!0.2 !2
!20
−10 !10
−5 0
0
10
5
20
10 −10π
!10 −5π
!5 0
0 5
5π 10
10π
t ω
1.2 5

1
rect(t/5) 4

0.8 3

0.6 2

0.4 1
0.2 0
0 !1
!0.2 !2
5sinc(5ω/2π)
−10
!20 −5
!10
00 10
5 1020 −10π −5π
!10 !5
00 5
5π 10
10π
t ω
Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 9 / 37
Narrower pulse means higher bandwidth.
Scaling Example 2

As another example, find the transform of the time-reversed exponential

x(t) = e at u(−t).

This is the exponential signal y (t) = e −at u(t) with time scaled by -1, so
the Fourier transform is
1
X (f ) = Y (−f ) = .
a − j2πf

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 10 / 37

Scaling Example 3

As a final example which brings two Fourier theorems into use, find the
transform of
x(t) = e −a|t| .
This signal can be written as e −at u(t) + e at u(−t). Linearity and
time-reversal yield
1 1
X (f ) = +
a + j2πf a − j2πf
2a
=
a2 − (j2πf )2
2a
=
a2 + (2πf )2

Much easier than direct integration!

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 11 / 37

Complex Conjugation Theorem

Complex Conjugation Theorem: If x(t) ⇔ X (f ), then

x ∗ (t) ⇔ X ∗ (−f )

Proof: The Fourier transform of x ∗ (t) is

Z ∞ Z ∞ ∗
x ∗ (t)e −j2πft dt = x(t)e j2πft dt
−∞ −∞
Z ∞ ∗
= x(t)e −(−j2πf )t dt = X ∗ (−f )
−∞

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 12 / 37

Duality Theorem

We discussed duality in a previous lecture.

Duality Theorem: If x(t) ⇔ X (f ), then X (t) ⇔ x(−f ).
This result effectively gives us two transform pairs for every transform we
find.

Exercise What signal x(t) has a Fourier transform e −|f | ?

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 13 / 37

Shift Theorem

The Shift Theorem:

x(t − τ ) ⇔ e −j2πf τ X (f )

Proof:

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 14 / 37

Example: square pulse

Consider a causal square pulse p(t) = 1 for t ∈ [0, T ) and 0 otherwise.

We can write this as !
t − T2
p(t) = rect
T
From shift and scaling theorems

P(f ) = Te −jπfT sinc(Tf ).

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 15 / 37

The Derivative Theorem

The Derivative Theorem: Given a signal x(t) that is differentiable almost

everywhere with Fourier transform X (f ),

x 0 (t) ⇔ j2πfX (f )

Similarly, if x(t) is n times differentiable, then

d n x(t)
⇔ (j2πf )n X (f )
dt n

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 16 / 37

Dual Derivative Formula

There is a dual to the derivative theorem, i.e., a result interchanging the

role of t and f . Multiplying a signal by t is related to differentiating the
spectrum with respect to f .

(−j2πt)x(t) ⇔ X 0 (f )

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 17 / 37

The Integral Theorem

Recall that we can represent integration by a convolution with a unit step

Z t
x(τ )dτ = (x ∗ u)(t).
−∞

Using the Fourier transform of the unit step function we can solve for the
Fourier transform of the integral using the convolution theorem,
Z t 
F x(τ )dτ = F [x(t)] F [u(t)]
−∞
 
1 1
= X (f ) δ(f ) +
2 j2πf
X (0) X (f )
= δ(f ) + .
2 j2πf

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 18 / 37

Fourier Transform of the Unit Step Function

How do we know the derivative of the unit step function?

The unit step function does not converge under the Fourier transform.
But just as we use the delta function to accommodate periodic signals, we
can handle the unit step function with some sleight-of-hand.
Use the approximation that u(t) ≈ e −at u(t) for small a.

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 19 / 37

A symmetric construction for approximating u(t)

Example: Find the Fourier transform of the signum or sign signal


 1 t>0
f (t) = sgn(t) = 0 t=0 .
−1 t < 0


We can approximate f (t) by the signal

fa (t) = e −at u(t) − e at u(−t)

as a → 0.

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 20 / 37

This looks like
1.5

1 sgn(t)
0.5 e−t/5
0 e−t
!0.5

!1

!1.5
!2 !1.5 !1 !0.5 0 0.5 1 1.5 2
t

As a → 0, fa (t) → sgn(t).
The Fourier transform of fa (t) is
Fa (f ) = F [fa (t)]
= F e −at u(t) − e at u(−t)
 
 −at   at 
= F e u(t) − F e u(−t)
1 1
= −
a + j2πf a − j2πf
−j4πf
=
a2 + (2πf )2
Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 21 / 37

Therefore,
−j4πf
lim Fa (f ) = lim
a→0 a→0 a2 + (2πf )2
−j4πf
=
(2πf )2
1
= .
jπf

This suggests we define the Fourier transform of sgn(t) as

 2
j2πf f 6= 0
sgn(t) ⇔ .
0 f =0

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 22 / 37

With this, we can find the Fourier transform of the unit step,
1 1
u(t) = + sgn(t)
2 2
as can be seen from the plots

sgn(t) u(t)
1 1

0 t 0 t
−1 −1

The Fourier transform of the unit step is then

 
1 1
F [u(t)] = F + sgn(t)
2 2
 
1 1 1
= δ(f ) + .
2 2 jπf

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 23 / 37

The transform pair is then

1 1
u(t) ⇔ δ(f ) + .
2 j2πf

1
πδ(ω) + π
jω

ω
1
jω

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 24 / 37

Parseval’s Theorem
(Parseval proved for Fourier series, Rayleigh for Fourier transforms. Also
called Plancherel’s theorem)
Recall signal energy of x(t) is
Z ∞
Ex = |x(t)|2 dt
−∞

Interpretation: energy dissipated in a one ohm resistor if x(t) is a voltage.

Can also be viewed as a measure of the size of a signal.
Theorem: Z ∞ Z ∞
Ex = |x(t)|2 dt = |X (f )|2 df
−∞ −∞

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 25 / 37

Example of Parseval’s Theorem

Parseval’s theorem provides many simple integral evaluations. For
example, evaluate Z ∞
sinc2 (t) dt
−∞

We have seen that sinc(t) ⇔ rect(f ).

Parseval’s theorem yields
Z ∞ Z ∞
sinc2 (t) dt = rect2 (f ) df
−∞ −∞
Z 1/2
= 1 df
−1/2
= 1.

Try to evaluate this integral directly and you will appreciate Parseval’s
shortcut.
Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 26 / 37
? The Convolution Theorem ?

Convolution in the time domain ⇔ multiplication in the frequency domain

This can simplify evaluating convolutions, especially when cascaded.
This is how most simulation programs (e.g., Matlab) compute
convolutions, using the FFT.
The Convolution Theorem: Given two signals x1 (t) and x2 (t) with Fourier
transforms X1 (f ) and X2 (f ),

(x1 ∗ x2 )(t) ⇔ X1 (f )X2 (f )

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 27 / 37

Proof: The Fourier transform of (x1 ∗ x2 )(t) is

Z∞ Z∞
 
 x1 (τ )x2 (t − τ ) dτ  e −j2πft dt
−∞ −∞
Z∞ Z∞
 

= x1 (τ )  x2 (t − τ )e −j2πft dt  dτ.
−∞ −∞

Using the shift theorem, this is

Z∞  
= x1 (τ ) e −j2πf τ X2 (f ) dτ
−∞
Z∞
= X2 (f ) x1 (τ )e −j2πf τ dτ
−∞
= X2 (f )X1 (f ).

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 28 / 37

Examples of Convolution Theorem

Unit Triangle Signal ∆(t)


1 − |t| if |t| < 1
0 otherwise.

1
Δ(t)

t
-1 0 1

Easy to show ∆(t) = rect(t) ∗ rect(t).

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 29 / 37

Since
rect(t) ⇔ sinc(f )
then
∆(t) ⇔ sinc2 (f )

1.0
0.25 sinc2(ω/2π)
0.2

0.15

0.1

0.05

0 !10
0
!8 !6 !4 !2 0 2 4 6 8 10
−4π −2π 0 2π 4π
ω Transform

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 30 / 37

Multiplication Property

If x1 (t) ⇔ X1 (f ) and x2 (t) ⇔ X2 (f ),

x1 (t)x2 (t) ⇔ (X1 ∗ X2 )(f ).

This is the dual property of the convolution property.

Note: If ω is used instead of f , then a 1/2π term must be included.

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 31 / 37

Multiplication Example - Bandpass Filter

A bandpass filter can be implemented using a low-pass filter and

multiplication by a complex exponential.

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 32 / 37

Modulation
The Modulation Theorem: Given a signal x(t) with spectrum x(f ), then

x(t)e j2πf0 t ⇔ X (f − f0 ),

1
x(t) cos(2πf0 t) ⇔ (X (f − f0 ) + X (f + f0 )) ,
2
1
x(t) sin(2πf0 t) ⇔ (X (f − f0 ) − X (f + f0 )) .
2j

Modulating a signal by an exponential shifts the spectrum in the frequency

domain. This is a dual to the shift theorem. It results from interchanging
the roles of t and f .
Modulation by a cosine causes replicas of X (f ) to be placed at plus and
minus the carrier frequency.
Replicas are called sidebands.

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 33 / 37

Amplitude Modulation (AM)

Modulation of complex exponential (carrier) by signal x(t):

xm (t) = x(t)e j2πf0 t

Variations:

fc (t) = f (t) cos(ω0 t) (DSB-SC)

fs (t) = f (t) sin(ω0 t) (DSB-SC)
fa (t) = A[1 + mf (t)] cos(ω0 t) (DSB, commercial AM radio)
I m is the modulation index
I Typically m and f (t) are chosen so that |mf (t)| < 1 for all t

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 34 / 37

Examples of Modulation Theorem

rect(t) sinc(ω/2π)
1.2 1.2
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
!0.2 !0.2
−2
!2
−1
!1 0
0 11 22 −20π −10π
!20 !10 00 10π
10 20π
20

t ω
1.2
1
1
0.5 0.8
0.6
0
0.4
!0.5 0.2
0
!1
!0.2
!2
−2 !1
−1 0
0 1 2
1 2
−20π −10π
!20 !10 0
0 10
10π 20
20π
t ! "ω ! "
1 ω − 10π 1 ω + 10π
rect(t) cos(10πt) sinc + sinc
2 2π 2 2π

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 35 / 37

Periodic Signals
Suppose x(t) is periodic with fundamental period T and frequency
f0 = 1/T . Then the Fourier series representation is,
∞
X
x(t) = ak e j2πkf0 t .
k=−∞

Let’s substitute in some δ functions using the sifting property:

∞
X Z ∞
x(t) = ak δ(f − kf0 )e j2πft df ,
k=−∞ −∞
∞
!
Z ∞ X
= ak δ(f − kf0 ) e j2πft df .
−∞ k=−∞

P∞
This implies the Fourier transform: x(t) ⇔ k=−∞ ak δ(f − kf0 ).

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 36 / 37

Constant-Coefficient Differential Equations

n M
X d k y (t) X d k x(t)
ak k
= bk .
dt dt k
k=0 k=0

Find the Fourier Transform of the impulse response (the transfer function
of the system, H(f )) in the frequency domain.

Cuff (Lecture 7) ELE 301: Signals and Systems Fall 2011-12 37 / 37