Digital Signal Processing

Fourier Analysis
 PERIODICITY
PERIODIC NON-PERIODIC
Continuous-Time Continuous-Time
CONTINUOUS-
Fourier Series Fourier Transform
TIME
(CTFS) (CTFT)
FORM

Discrete-Time
Fourier Series or Discrete-Time
DISCRETE-
Discrete Fourier Fourier Transform
TIME
Transform (DTFT)
(DFT)
 Introduction

Fourier representation of signals plays an extremely
important role in both continuous-time and discrete-
time signal processing

It provides a method for mapping signals into another
“domain” in which to manipulate them.

It provides a different way to interpret signals and
systems
 Frequency Response

Eigenfunctions of linear shift-invariant
systems are sequences that, when input to the
system, pass through with only a change in
(complex) amplitude.
 Frequency Response
Signals of the form
x ( n ) = e jnω −∞ < n<∞
where ω is a constant, are eigenfunctions of LTI
systems. This may be shown from the convolution
∞
sum: y (n ) = h (n )* x (n ) = ∑ h (k ) x (n − k )
k = −∞
∞
= ∑ h ( k ) e jω ( n − k )
k = −∞
∞
=e jnω
∑ h (k )e
k = −∞
− jk ω

. = H ( e jω ) e jnω
 Frequency Response
Thus, the eigenvalue, which we denote by H(e jω), is
∞
H ( e jnω ) = ∑ h ( k ) e − jk ω
k = −∞
Note that H(ejω)is, in general, complex-valued and
depends on the frequency ω of the complex
exponential. Thus, it may be written in terms of
its real and imaginary parts
H ( e jω ) = H R ( e jω ) + jH I ( e jω )
or in terms of its magnitude and phase
. H ( ) ( )
e jω
= H e jω
e jφ (ω ) n
 Frequency Response
H (e ) = H ( e jω ) H * ( e jω ) = H R2 ( e jω ) + H I2 ( e jω )
2
where jω

and H I ( e jω )
φh (ω ) = tan −1
H R ( e jω )
A graphical representation that is often used instead
of the phase is the group delay
dφh (ω )
τ h (ω ) = −
dω
where the phase is taken to be a continuous and
differentiable function of ω by adding integer
multiples of π to the principal value of the phase
(unwrapping the phase).
 Frequency Response

The function H(ejω) is very useful and
important in the characterization of LTI
systems and is called the frequency response.

The frequency response defines how a
complex exponential is changed in (complex)
amplitude when it is filtered by the system.

The frequency response is particularly useful if
an input signal can be decomposed into a sum
of complex exponentials.
 Example
Let x(n) = cos(nω0) be the input to a linear time-
invariant system with a real-valued unit
impulse response h(n). If x(n) is decomposed
into a sum of two complex exponentials.
x ( n ) = 12 e jnω0 + 12 e − jnω0
The response of the system may be written as
( ) ( )
y ( n ) = 12 H e jω0 e jnω0 + 12 H e − jω0 e − jnω0
 Example
Because h(n) is real-valued, H(ejω) is conjugate
symmetric
H ( e − jω ) = H * ( e jω )

Therefore,
( ) ( )
y ( n ) = 12 H e jω0 e jnω0 + 12 H * e jω0 e − jnω0
and it follows that
{ ( ) } ( )
y ( n ) = Re H e jω0 e jnω0 = H e jω0 cos ( nω0 + φ (ω0 ) )
 Periodicity

The frequency response is a complex-valued
function of ω and is periodic with a period of π.

The frequency response of a linear time-invariant
continuous-time system is not periodic.

A discrete-time complex exponential of frequency
ω0 is the same as a complex exponential of
frequency ω0 + 2π; that is
.
( )
x n =e jnω0
=e
jn (ω0 + 2π )
 Periodicity

Therefore, if the input to a linear time-
invariant system is x ( n ) = e
jnω0
, the response
must be the same as the response to the signal
x ( n) = e
jn (ω0 + 2π )
. This, in turn, requires that

H e( jω0
) = H (e j (ω0 + 2π )
)
 Symmetry

If h(n) is real-valued, the frequency response is a conjugate
symmetric function of frequency:
H (e − jω
)) implies
= H *(e ) jω

Conjugate symmetry of H(e jω that the real part is an ω, and

that the imaginary part is odd,
H R ( e jω ) = H R ( e − jω ) H I ( e jω ) = − H I ( e − jω )

Conjugate symmetry also implies that the magnitude is even,
H (e jω
) = H (e ) − jω

and that the phase and group delay are odd.

φh (ω ) = −φh ( −ω ) τ h (ω ) = −τ h ( −ω )
.
 Example
Consider the LTI system with unit impulse
response h ( ) u ( n)
n = a n

where α is a real number with |α| < 1. The

frequency response is
∞ ∞
H ( e jω ) = ∑ h ( n ) e− jnω = ∑ α n e − jnω
n =−∞ n=0
∞
= ∑ (α e )
− jω n 1
=
n =0 1 − α e − jω
 Example
The squared magnitude of the frequency response is
H (e ) = H (e ) H *(e ) =
jω
2
jω jω 1 1 1
⋅ =
1−αe − jω
1−αe jω
1 + α 2 − 2α cos ω
and the phase is
I ( ) = tan −1
jω
H e −α sin ω
φh (ω ) = tan −1 jω
H R (e ) 1 − α cos ω
Finally, the group delay is found by differentiating
the phase. The result is
α 2 − α cos ω
τ h (ω ) = −
. 1 + α 2
− 2α cos ω
 Inverting the Frequency Response
Given the frequency response of a linear time-
invariant system ∞
H ( e jω ) = ∑ h ( n ) e − jnω
n =−∞
the unit sample response may be recovered by
integration:
H ( e jω ) e jnω d ω
1 π
h (n) = ∫π
2π −

The integral may be taken over any period of

length 2π.
 Example
For a system with a frequency response given by
1 ω ≤ ωc
H (e ) = 
jω

0 ωc < ω ≤ π
(this system is referred to as an ideal low-pass
filter), the unit impulse response is
1 ωc 1 sin nωc
h (n) = ∫ω e
jnω
dω = 

jnωc
e −e − jnωc

 =
2π − c 2 jπ n πn
Note that this system is noncausal (it is also
unstable) and, therefore, unrealizable.
 Filters

The term digital filter, is often used to refer to
a discrete-time system.

A digital filter is defined by J. F. Kaiser as a
“computational process or algorithm by which
a sampled signal or sequence of numbers
(acting as the input) is transformed into a
second sequence of numbers termed as the
output signal.”
 Filters

The computational process maybe that of a
lowpass filtering (smoothing), bandpass
filtering, interpolation, the generation of
derivatives, etc.

Filters may be characterized in terms of their
system properties, such as linearity, time-
invariance, causality, stability, etc., and may be
classified in terms of their frequency response.
 Linear Phase
A linear time-invariant system is said to have a linear phase if
its frequency response is of the form
H ( e jω ) = A ( e jω ) e − jαω
where α is a real number and A(ejω) is a real-valued function of
ω. Note that the phase of H(ejω) is
 −αω
 when A ( e
jω
)≥0
φh (ω ) = 
−αω + π when A ( e ) < 0 jω

Similarly, a filter is said to have generalized linear phase if the

frequency response has the form
H ( e jω ) = A ( e jω ) e
− j (αω − β )

Thus, filters with linear phase or generalized linear phase has a

constant group delay.
 All-Pass Filter
A system is said to be all-pass filter if the frequency
response magnitude is constant:
H ( e jω ) = c
An example of an all-pass filter is the system that
has a frequency responsee− jω − α
H ( e jω ) =
1 − α e − jω
where α is any real number with |α| < 1. The unit
sample of this all-pass filter is
. h ( n ) = −αδ ( n ) + (1 − α 2 ) α n −1u ( n − 1)
Frequency Selective Filters
 Interconnection of Systems

Filters are often interconnected to create
systems that have desirable properties.

Two common types of connections are series
(cascade) and parallel.
 Cascade

A cascade is equivalent to a single linear time-invariant system

with a unit impulse response.
h ( n ) = h1 ( n ) ∗ h2 ( n )
and a frequency response H ( e jω ) = H1 ( e jω ) H 2 ( e jω )
Note that the log magnitude of the cascade is the sum of the
log magnitudes of the individual systems.
20 log H ( e jω ) = 20 log H1 ( e jω ) + 20 log H 2 ( e jω )
and the phase and group delay are additive.
φ (ω ) = φ1 (ω ) + φ2 (ω )
. τ (ω ) = τ (ω ) + τ (ω )
1 2
 Parallel

A parallel network is equivalent to a single linear

time-invariant system with a unit impulse
response h ( n ) = h1 ( n ) + h2 ( n )
Therefore, the frequency response of the parallel
network is H e ( )jω
( ) jω
= H1 e + H 2 e ( )
jω
 The Discrete-Time Fourier Transform
The frequency response of a linear time-invariant
system is found by multiplying h(n) by a
complex exponential e –jnω and summing over
n. The discrete-time Fourier transform (DTFT)
of a sequence x(n) is defined
∞
in the same way
X ( e jω ) = ∑ x ( n ) e − jnω
n =−∞
Thus, the frequency response of a linear time-
invariant system , H(ejω) is the DTFT of the
unit impulse response h(n).
 The Discrete-Time Fourier Transform
In order for the DTFT to exist, the summation of
the sequence must converge. This, is turn,
requires that x(n) be absolutely summable:
∞

∑ x ( n) = S < ∞
n =−∞
 Example
The DTFT of the sequence
x1 ( n ) = α nu ( n ) α <1
∞ ∞

is X1 ( e ) = ∑α e
jω
= ∑ (α e )
n − jnω − jω n

n =0 n=0
Using the geometric series, this sum is
X1 ( e ) =
jω 1
1 − α e − jω
provided |α| < 1. Similarly, for the sequence
x2 ( n ) = −α nu ( −n − 1) α >1
∞ −1

The DTFT is 2 ( ) = ∑ x2 ( n ) e − jnω = − ∑ α n e − jnω

jω
X e
n =−∞ n =−∞
 Example
Changing the limits on the sum, we have
∞ ∞
X 2 (e jω
) = −∑ α −n
e jnω
= − ∑ (α e )
−1 jω n
+1
n =1 n =0

If |α| > 1, this sum is

X 2 (e ) = − 1−α
jω 1 1
+1 =
−1 jω
e 1 − α e − jω
Therefore, x1(n) = αnu(n) & x2(n) = –αnu(–n – 1)
both have the same DTFT.
 Inverse DTFT
Given X(ejω), the sequence x(n) may be
recovered using the inverse DTFT.

X ( e jω ) e jnω d ω
1 π
x (n) = ∫π
2π −
Some Common DTFT Pairs
 Example
Suppose X(ejω) consists of an impulse at frequency ω = ω0.
( )
X e jω = δ (ω − ω0 )
( )
Using the inverse DTFT, 1 π 1 jnω0
x (n) = ∫ X e e dω =
jω jnω
e
2π − π 2π
Note that although x(n) is not absolutely summable, by
allowing the DTFT to contain impulses, consider the DTFT
of sequences that contain complex exponentials. As another
X (e ) = πδ (ω − ω ) + πδ (ω + ω )
example. If jω
0 0
computing the inverse DTFT, it is found that
. ( ) 1 jnω0
x n = e + e2
1 − jnω0
2 = cos ( nω0 )
 DTFT Properties
Periodicity
The discrete-time Fourier transform is periodic in ω
with a period of 2π:
(
X ( e jω ) = X e j (ω + 2π ) )
This property follows directly from the definition of
the DTFT and the periodicity of the complex
exponentialX ( e ( ) ) = ∑ x ( n ) e ( ) = ∑ x ( n ) e− jnω e− j 2πω
∞ ∞
j ω + 2 π − jn ω + 2 π

n =−∞ n =−∞
∞

. = ∑ x ( n ) e − jnω = X ( e jω )
n =−∞
DTFT Properties
 DTFT Properties
Symmetry
The DTFT often has some symmetries that may be
exploited to simplify the evaluation of the DTFT or
the inverse DTFT.
 DTFT Properties
Linearity
The discrete-time Fourier transform is a linear
operator. If X1(ejω) is the DTFT of x1(n), and
X2(ejω) is the DTFT of x2(n).
ax1 ( n ) + bx2 ( n ) ⇔ aX 1 ( e jω ) + bX 2 ( e jω )
DTFT

Shifting Property
Shifting a sequence in time results in the
multiplication of the DTFT by a complex
exponential (linear phaseDTFT term):
. x ( n − n0 ) ⇔ e − jn ω X ( e jω )
0
 DTFT Properties
Time-Reversal
Time-reversing a sequence results in a frequency reversal
of the DTFT:
x ( −n ) ⇔ X ( e − jω )
DTFT

Modulation
Multiplying a sequence by a complex exponential results
in a shift in frequency of the DTFT
( )
DTFT
x ( n) ⇔ X e
jnω0 j (ω −ω0 )
e
Thus, modulating a sequence by a cosine of frequency ω0:
( ) ( )
DTFT
x ( n ) cos nω0 ⇔ 12 X e ( ) + 12 X e ( )
j ω −ω j ω +ω
. 0 0
 DTFT Properties
Convolution Theorem
The DTFT of sequence that is formed by convolving two
sequences, x(n) and h(n), is the product of the DTFTs of
x(n) and h(n):
h ( n ) * x ( n ) ⇔ H ( e jω ) X ( e jω )
DTFT

Multiplication (Periodic Convolution) Theorem

As with the time-shift and modulation properties, there is
a dual to the convolution theorem that states that
multiplication in the time domain corresponds to
(periodic) convolution in the frequency domain:
( ( ) )
DTFT 1 π
. x ( n) y ( n) ⇔ ∫ X e jθ
Y e j (ω −θ )
dθ
2π −π
 DTFT Properties
Parseval’s Theorem
A corollary to the multiplication theorem is the
Parseval’s theorem,
∞
X ( e ) dω
1 π
x ( n) =
2
∑
2 jω

n =−∞ 2π ∫− π

Paseval’s theorem is referred to as the

conservation of energy theorem, because it
states that the DTFT operator preserves energy
when going from the time domain into the
frequency domain.
 APPLICATIONS – LTI Systems
and LCCDEs
An important subclass of LTI systems contains those whose
input, x(n), and output y(n), are related by a linear constant
coefficient difference equation (LCCDE):
p q
y ( n ) = −∑ a ( k ) y ( n − k ) + ∑ b ( k ) x ( n − k )
k =1 k =0

The linearity and shift properties of the DTFT may be used to

express this difference equation in the frequency domain
Y ( e jω ) = −∑ a ( k ) e − jkωY ( e jω ) + ∑ b ( k ) e − jkω X ( e jω )
p q

k =1 k =0
or  
Y ( e jω ) 1 + ∑ a ( k ) e− jkω  = X ( e jω ) ∑ b ( k ) e − jkω
p q

 k =1  k =0
 APPLICATIONS – LTI Systems
and LCCDEs
Therefore, the frequency response of this system is
q

Y (e) jω ∑ b ( k ) e − jkω

H (e ) =
jω
= k =0

X (e ) jω p
1+ ∑ a (k ) e − jkω

k =1
 Example
Consider the linear time-invariant system
characterized by the second-order linear constant
coefficient difference equation
y ( n) =1.3433y ( n −1) − 0.9025y ( n − 2) + x ( n) −1.4142x ( n −1) + x ( n − 2)
Solution:
The frequency response may be found by inspection
without solving the difference equation for h(n) as
follows:
− jω −2 jω
− +
H ( e jω ) =
1 1.4142 e e
. 1 − 1.3433e − jω + 0.9025e −2 jω
 Example
Given a frequency response function
ω
such as
1 + e −2 j
H (e ) =jω

2 − e− jω + 0.5e−2 jω
A difference equation may easily be found that
will implement this system. First, dividing
numerator and denominator by 2 and rewriting
the frequency response as follows:
ω
0.5 + 0.5e−2 j
H (e ) =
jω

1 − 0.5e − jω + 0.25e−2 jω
Therefore, the difference equation for this system
is y ( n ) = 0.5 y ( n − 1) − 0.25 y ( n − 2 ) + 0.5 x ( n ) + 0.5 x ( n − 2 )
 APPLICATIONS – Performing
Convolutions
Example:
If the unit sample response of an LTI system is
h ( n ) = α nu ( n )
Find the response of the system to the input x(n)
= βnu(n), where |α| < 1, and |β| < 1, and α ≠ β
because the output of the system is the
convolution of x(n) with h(n).
y (n) = h (n) * x (n)
 Example
The DTFT of y(n) is
Y (e ) = H (e ) X (e ) =
jω jω 1
jω 1
1 − α e − jω 1 − β e − jω
Therefore, all that is required is to find the
inverse DTFT of Y(ejω). This may be done
easily by expanding Y(ejω) as follows:
Y (e ) =
jω 1 A B
= +
(1 − α e − jω
)(1 − β e − jω
) 1 − α e − jω
1 − β e − jω

where A and B are constants that are to be

determined.
 Example
Expressing the right-hand side of this expansion
over a common denominator
1
=
( A + B ) − ( Aβ + Bα ) e− jω
(1 − α e )(1 − β e ) (1 − α e )(1 − β e )
− jω − jω − jω − jω

and equating the coefficients, the constants A and

B may be found by solving the pair of
equations A+ B =1
Aβ + Bα = 0
α β
A= B=−
The result is α −β α −β
 Example
Therefore, α / (α − β ) β / (α − β )
Y (e ) =
jω
−
1−αe − jω
1 − β e − jω

and it follows that the inverse DTFT is

 α β n
y (n) =  α −
n
β  u (n)
α − β α −β 
APPLICATIONS – Solving Difference
Equations
Example:
Solve the following LCCDE for y(n) assuming
zero initial conditions
y ( n ) − 0.25 y ( n − 1) = x ( n ) − x ( n − 2 )
for x(n) = δ(n).
Solution:
Begin by taking the DTFT of each term:
Y ( e jω ) − 0.25e − jωY ( e jω ) = X ( e jω ) − e −2 jω X ( e jω )
 Example
Because the DTFT of ωx(n) is X(ejω) = 1, ω
1 − e −2 j e −2 j
Y (e ) =
jω 1
− jω
= − jω
−
1 − 0.25e 1 − 0.25e 1 − 0.25e− jω
Using the DTFT pair
DTFT 1
( 0.25) u ( n ) ⇔
n

1 − 0.25e − jω
The inverse DTFT of Y(ejω) may be easily found
using the linearity and shift properties
y ( n ) = ( 0.25 ) u ( n ) − ( 0.25 ) u ( n − 2)
n n−2
 APPLICATIONS – Inverse Systems
The inverse of a system with unit impulse h(n) is
a system that has a unit impulse response g(n)
such that
h ( n) * g (n) = δ (n)

In terms of the frequency response, it is easy to

see that, if the inverse of H(ejω) exists, it is
H (e ) =
equal to −1 jω 1
H ( e jω )
 APPLICATIONS – Inverse Systems
Care must be exercised, however, because not all
systems are invertible or, if the inverse exists, it
may be noncausal. For example, the ideal low-
pass filter does not have an inverse. The inverse
of H ( e jω ) = 1 − 2e − jω

H (e ) = ( )
−1 jω 1 jω
is − jω
= G e
1 − 2e
which corresponds to a system that has a noncausal
unit impulse response
. g ( n ) = −2− n u ( −n − 1)
 Example
If the frequency response of an LTI system is
− jω
1− e
H (e ) =
1
jω 4
− jω
1+ e 1
2

the inverse system is − jω

1+ e
G (e ) = H (e ) =
1
jω −1 jω 1
= 2

H (e ) 1− e
jω 1 − jω
4

which has a unit sample response

g ( n ) = ( 0.25 ) u ( n ) + 0.5 ( 0.25 ) u ( n − 1)
n n −1