Chapter 2

Chapter 2
CT and DT Signal Representations
• 1 Fourier Series
• 2 Fourier Transform
• 3 Discrete Fourier Series
• 4 Discrete-Time Fourier Transform
• 5 Discrete Fourier Tranform
• 6 Applications
GOALS
1. Development of representations of CT and DT signals in the fre-
quency domain.
2. Familiarization with Fourier Series and transform representations

for CT and DT signals.
c A.C Singer and D.C. Munson, Jr. January 23, 2011

10 CT and DT Signal Representations
Since the signals we encounter in engineering, science, and everyday life are as varied as the applications
in which we engage them, it is often helpful to rst study these applications in the presence of simplied
versions of these signals. Much like a child learning to play an instrument for the rst time, it is easier to
start by attempting to play a single note before an entire musical score. Then, after learning many notes,
the child becomes a musician and can synthesize a much broader class of music, building up from many
notes. This approach of building-up our understanding of complex concepts by rst understanding their
basic building blocks is a fundamental precept of engineering and one that we will use frequently throughout
this book.
In this chapter, we will explore signals in both continuous time and discrete time, together with a number
of ways in which these signals can be built-up from simpler signals. Simplicity is in the eye of the beholder
and what makes a signal appear simple in one context may not shed much light in another context. Many
of the concepts we will develop throughout this text arise from studying large classes of signals, one building
block at a time, and extrapolating system (or application) level behavior by considering the whole as a sum
of its parts. In this chapter, we will focus specically on sinusoidal signals as our basic building blocks as
we consider both periodic and aperiodic signals in continuous and discrete time. Along this path, we will
encounter the Fourier series representations of periodic signals as well as Fourier transform representations of
aperiodic, innite-length signals. In later chapters, we will nd that so-called time-domain representations
of signals sometimes prove more fruitful, and for discrete-time signals there is a natural way to construct
signals one sample at a time.
2.1 Fourier Series representation of nite-length and periodic CT

signals
In many applications in science and engineering, we often work with signals that are periodic in time. That
is, the signal repeats itself over and over again with a given period of repetition. Examples of periodic signals
might include the acoustic signal that emenates from a musical instrument, such as a trumpet when a single
sustained note is played, or the vertical displacement of a mass in a frictionless spring-mass oscillator set
into motion, or the horizontal displacement of a pendulum swaying to and fro in the absence of friction.
Mathematically, we represent a periodic signal, x(t), as one whose value repeats at a xed interval of
time from the present. This interval, denoted T below, is called the period of the signal, and we express
this relationship
x(t) = x(t + T ), for all t. (2.1)
Equation (2.1) will, in general, be satised for a countably innite number of possible values of T when x(t)
is periodic. The smallest, positive value of T for which Eq. (2.1) is satised, is called the fundamental
period of the signal x(t). For sinusoidal signals, such as
x(t) = sin(ω0 t + φ), (2.2)
we can relate the frequency of oscillation, ω0 to the fundamental period, T. This can be computed by noting
that sinusoidal functions are equal when their arguments are either equal or dier only through a multiple
of 2π , i.e.
x(t) = x(t + T )
sin(ω0 t + φ) = sin(ω0 (t + T ) + φ)
sin(ω0 t + φ + 2kπ) = sin(ω0 (t + T ) + φ)
sin(ω0 (t + 2kπ/ω0 ) + φ) = sin(ω0 (t + T ) + φ) (2.3)
which, for k = 1, yields the relationship

T = 2π/ω0 , (2.4)
between the fundamental period, T , and the fundamental frequency ω0 . By analogy to sinusoidal signals,
we refer to the value of ω0 = 2π/T as the fundamental frequency of any signal that is periodic with a
fundamental period T.

2.1 Fourier Series representation of nite-length and periodic CT signals 11
x(t)=sin(4π/3 t)
1
0.8
0.6
0.4
0.2
x(t)
−0.2
−0.4
−0.6
−0.8
−1
0 2 4 6 8 10
t (sec)
Figure 2.1: The periodic sinusoidal signal x(t) = sin((4π/3)t).
Here, we will provide a number of examples of periodic signals in continuous-time, including sinusoidal,
square wave, traingular wave and complex exponential signals. By noting that any two periodic signals, x(t)
and y(t) with the same period T can be added together to produce a new periodic signal of the same period,
i.e.,
s(t) =x(t) + y(t)

s(t + T ) =x(t + T ) + y(t + T ) = s(t + T ),
in 1807 Jean Baptiste Fourier (1807) considered the notion of building a large set of periodic signals from
sinusoidal signals sharing the same period. Ignoring the phase,φ, for now, note that from (2.4), sinusoidal
signals that share the same period must have fundamental frequencies given by kω0 = 2kπ/T for dierent
values of k. If two sinusoidal signals shared the same fundamental frequency, then they would be the same
sinusoidal signal (recall that, for now, we are neglecting the phase, φ). We call such sinusoidal signals
whose fundamental frequencies kω0 are integer multiples of one fundamental frequency, harmonically-related
sinusoids. Such harmonically-related sinusoids could indeed share the period, 2π/ω0 while they would have
dierent fundamental freqeuncies and hence dierent fundamental periods.
We now consider how we might build-up a larger class of periodic signals from the basic building blocks of
harmonically-related sinusoids. To extend our discussion to include complex-valued signals, we will employ
Euler's relation to construct complex exponential signals of the form
x(t) =ej(ω0 t+φ) (2.5)
= cos(ω0 t + φ) + j sin(ω0 t + φ)
and in doing so, we can push the phase out of the picture so that it can be absorbed in a complex scalar
constant out front, i.e.
x(t) = cejω0 t ,
where, c = ejφ is simply a complex constant whose eects on the sinusoidal nature of the signal have been
conveniently parked outside the discussion. Complex-exponential signals of the form (2.5) are periodic with
fundamental frequency ω0 = 2π/T since they are simply constructed by pairing the real-valued periodic
signal cos(ω0 t) with the purely imaginary signal j sin(ω0 t).
By simply adding together harmonically-related sinusoidal signals, we can construct a large class of
periodic waveforms of amazing variety. For example, in Figure 2.2, note how by taking odd-valued harmonics
(sinusoids with harmonically-related fundamental frequencies that are odd multiples of a single frequency,
ω0 = 2π ), we obtain an increasingly improving approximation to a square wave with unit period.

0.8
0.6
0.4
0.2
x(t)
−0.2
−0.4
−0.6
−0.8
−1
0 0.2 0.4 0.6 0.8 1
t
PN 1
Figure 2.2: The periodic sinusoidal signal x(t) = k=1 k sin(2kπt), for k = 1, 3, 9 and 99.
Generalizing this idea, we can explore the class of signals that can be constructed by such harmonically-
related complex exponentials of the form
∞
X
x(t) = X[k]ejkω0 t . (2.6)
k=−∞
To bring the period of the periodic signal x(t) into the equation, (2.6) is often written
∞
X
x(t) = X[k]ej2πkt/T , (2.7)
k=−∞
where T = 2π/ω0 is the fundamental period and ω0 is the fundamental frequency of the periodic signal x(t).
The construction in (2.7) is referred to as the continuous-time Fourier series (CTFS) representation of x(t)
and (2.7) is often called the continuous-time Fourier series synthesis equation.
The Fourier series coecients X[k] can be obtained by multiplying (2.7) by e−j2πkt/T and integrating
over a period of duration T to obtain
Z T
x(t)e−j2πkt/T dt
0
∞
!
Z T X
j2π(m−k)t/T
= X[m]e dt,
0 m=−∞
where the limits of integration indicate that the we have chosen to evaluate the integral over the period
0 ≤ t ≤ T . Note the use of the dummy variable m in the summation for the CTFS, since the variable
k was already in use. To use k again would invite disaster into our derivation. Interchanging the order of
integration and summation (which can be done under suitable conditions on the summation), we obtain,
Z T
x(t)e−j2πkt/T dt
0
∞
X Z T
= X[m]ej2π(m−k)t/T dt. (2.8)
m=−∞ 0

To proceed, we need to evaluate the integral
T
Z T
j2π(m−k)t/T T
e dt = ej2π(m−k)t/T
0 j2π(m − k)
0
T
= [ej2π(m−k) − 1]
j2π(m − k)
=T δ[m − k],
where the second line arises from simple integration of an exponential function. The second line is readily
seen to be equal to zero when m 6= k and though one might be tempted to evaluate this line for m=k (using
a formula bearing the name of a famous 17th-century French mathematician), our eorts will be better spent
setting m=k into the integrand on the left hand side of the rst line, from which we obtain
Z T
1dt = T.
0
An interpretation of this result is that integration of a periodic complex exponential over an integer multiple,
(m − k), of its fundamental period, in this case T /2π(m − k) , is zero. The only periodic complex exponential
that survives integration over the period T is the DC, i.e. m = k , term.
We can now return to (2.8) and apply this result, to obtain
Z T ∞
X
x(t)e−j2πkt/T dt = X[m]T δ[m − k]
0 m=−∞
=T X[k], (2.9)
by the sifting property of the Kronocker delta function. We can now turn (2.9) around to obtain the
continuous-time Fourier series analysis equation,
Z T
1
X[k] = x(t)e−j2πkt/T dt. (2.10)
T 0
Putting the synthesis and analysis equations together, we have the continuous-time Fourier series represen-
tation of a periodic signal x(t) as
CT Fourier Series Representation of a Periodic Signal

Z T
1 2πkt
X[k] = x(t)e−j T dt (2.11)
T 0
∞
2πkt
X
x(t) = X[k]ej T (2.12)
k=−∞
Example: CTFS of a Square Wave

Let us return to the square wave signal that we visited in Figure 2.2. In the gure, we appeared
to have a method for constructing the periodic signal that, in the interval [0, 1], satises

1, 0 ≤ t ≤ 0.5
x(t) = (2.13)
−1 else.
Using (2.10), we obtain,

Z 1
X[k] = x(t)e−j2πkt dt (2.14)
0
Z 0.5 Z 1
= e−j2πktdt− e−j2πktdt
0 0.5
−1
[e−jπk −1]−[1−e−jπk ]

=
j2πk
−1
= 2[(−1)k − 1]
j2πk

0, k even
= 2
jπk k odd.
Note that the k=0 case can be readily evaluated by considering the integral in (2.14) for which
the integral can be easily seen to vanish by the antisymmetry of x(t) over the unit interval.
2.1.1 CT Fourier Series Properties
We have now been properly introduced to a method for building-up continuous-time periodic signals from a
class of simple sinusoidal signals in (2.11)and a method for analysing the make-up of such periodic signals
in terms of their constituent sinusoidal components in (2.12). Now that introductions are out of the way,
we can explore some of the many useful properties of the CTFS representation. As we shall see, it is often
helpful to consider the properties of a whole signal by virtue of the properties of its parts, and the relations
we develop next will often prove useful in this process.
2.1.1.1 Linearity
The CTFS can be viewed as a linear operation, in the following manner. When two signals x(t) and y(t)
are each constructed from their constituent sinusoidal signals according to the CTFS synthesis equation
(2.12), the linear combination of these signals, z(t) = ax(t) + by(t), for a, b real-valued constants, can be
readily obtained by combining the constituent sinusoidal signals through the same linear combination. More
specically, when x(t) is a periodic signal with CTFS coecients X[k] and y(t) is a periodic signal with CTFS
coecients Y [k] then the signal z(t) = ax(t) + by(t) has CTFS coecients given by Z[k] = aX[k] + bY [k].
The linearity property of the CTFS can be compactly represented as follows
CT F S CT F S CT F S
x(t) ←→ X[k], y(t) ←→ Y [k] =⇒ z(t) = ax(t) + by(t) ←→ aX[k] + bY [k].
This result can be readily shown by substituting z(t) = ax(t)+by(t) into the integral in (2.11) and expanding
the integral into the two separate terms, one for X[k] and one for Y [k].
2.1.1.2 Time Shift

When a sinusoidal signal x(t) = sin(ω0 t) is shifted in time, the resulting signal x(t − t0 ) can be represented
in terms of a simple phase shift of the origional sinusoidal signal, i.e. x(t − t0 ) = sin(ω0 (t − t0 )) = sin(ω0 t −
φ),where φ = ω0 t0 = 2πt0 /T. Periodic signals that can be represented using the CTFS contain many, possibly
innitely many, sinusoidal (or complex exponential) signals. When such periodic signals are delayed in time,
each of the constituent sinusoidal components of the signal are delayed by the same amount, however this
translates into a dierent phase shift for each component. This can be readily seen from the CTFS analysis
equation (2.11), as follows. For the signal y(t) = x(t − t0 ), we have

Z T
1 2πk
Y [k] = x(t − t0 )e−j T t dt
T t=0
Z T −t0
1 2πk
= x(s)e−j T (s+t0 ) dt
T s=−t0
Z 0 Z T −t0
1 2πk 1 2πk
= x(s)e−j T (s+t0 ) ds + x(s)e−j T (s+t0 ) ds
T s=−t0 T s=0
Z 0 Z T −t0
1 2πk 1 2πk
= x(s + T )e−j T (s+T +t0 ) ds + x(s)e−j T (s+t0 ) ds
T s=−t0 T s=0
Z T Z T −t0
1 2πk 1 2πk
= x(τ )e−j T (τ +t0 ) dτ + x(s)e−j T (s+t0 ) ds
T τ =T −t0 T s=0
Z T
1 2πkt0 2πk
= x(t)e−j T e−j T t dt
T t=0
2πkt0
= X[k]e−j T ,
where, the second line follows from the change of variable, s = t − t0 , the fourth line follows from the
periodicity of both the signal x(t) and the signal e−j2πkt/T with period T , the fth line follows from the
change of variable τ = s + T, and the last line follows from the denition of X[k] after factoring the linear
phase term e−j2πkt0 /T out of the integral. The time shift property of the CTFS can be compactly represented
as follows
CT F S CT F S 2πk
x(t) ←→ X[k] =⇒ y(t) = x(t − t0 ) ←→ X[k]e−j T t0 .
We see that a shift in time of a periodic signal corresponds to a modulation in frequency by a phase term that
is linear with frequency with a slope that is proportional to the delay. This can be made easier if we adopt
the convenient, but conceptually more challenging concept of integration over a period for the denition of
the CTFS.
2.1.1.3 Frequency Shift

When a periodic signal x(t) has a CTFS representation given by X[k], a natural question that might arise
is the what happens when the shifting that was discussed in section 2.1.1.2 is applied to the CTFS repre-
sentation,X[k]. Specically, if a periodic signal y(t) were known to have a CTFS representation given by
Y [k] = X[k − k0 ], it is interesting to understand the relationship in the time-domain between y(t) and x(t).
This can be readily seen through examination of the CTFS analysis equation,
Y [k] = X[k − k0 ]
1 T
Z
2π
= x(t)e−j T (k−k0 )t dt
T t=0
1 T
Z
2πk0
j t −j 2π
= x(t)e T e T kt dt
T t=0
1 T
Z
2πk0 2π
j t
= x(t)e T e−j T kt dt,
T t=0
which leads to the relation
CT F S CT F S
x(t) ←→ X[k] =⇒ y(t) = x(t)ejk0 ω0 t ←→ X[k − k0 ],
2π
where ω0 =T . We observe that a shift in the continuous time Fourier series coecients by an integer amount
k0 corresponds to a modulation in the time domain signal x(t) by a term whose frequency is proportional to
the shift amount.

2.1.1.4 Time Reversal

When a periodic signal x(t) = ej2πt/T is time-reversed, i.e. y(t) = x(−t), the eect on its CTFS representa-
tion can be simply observed
Z T
1 2π 2πk
X[k] = ej T t e−j T t dt
T t=0
(
1, for k=1
=
0 otherwise
and
1 T −j 2π t −j 2πk t
Z
Y [k] = e T e T dt
T t=0
(
1, for k = −1
=
0 otherwise.
More generally, from the CTFS synthesis equation,
∞
2πk
X
x(t) = X[k]ej T t ,
k=−∞
we see that by simply changing the sign of the time variable t, we obtain the general relation
∞
2πk
X
y(t) = x(−t) = X[k]e−j T t
k=−∞
∞
X 2π(−k)
= X[k]ej T t
k=−∞
∞
2πk
X
= X[−m]ej T t ,
m=−∞
yielding the relation
CT F S CT F S
x(t) ←→ X[k] =⇒ y(t) = x(−t) ←→ X[−k],
i.e., changing the sign of the time axis corresponds to changing the sign of the CTFS frequency index.
2.1.1.5 Time Scaling

When a periodic signal undergoes a time-scale change, such as one that compresses the time axes, y(t) =
x(at), where a>1 is a real-valued constant, the resulting signal y(t) would remain periodic, however the
period would change correspondingly, such that y(t + Ty ) = y(t) would be satised for a dierent period Ty .
This can be easily seen by substituting in forx(t) in the relation, y(t) = x(at) = y(t + Ty ) = x(a(t + Ty )) and
the noting that x(at) = x(at + T ), due to the periodicity of x(t) with period T. This leads to the relation
x(a(t + Ty )) = x(at + T ) or Ty = T /a. This makes intuitive sense, since the time-axis in the signal y(t) has
been compressed by a factor of a, therefore the time at which it will repeat must also have compressed by
the same factor. Now, even though the period of the signal y(t) has changed, we also are interested in the
full CTFS representaiton of y(t). This is given by

∞
2πk
X
y(t) = X[k]ej T at
k=−∞
∞
j 2πk
Ty t
X
= X[k]e ,
k=−∞
where the second line follows from the denition of Ty . Note that although we have that
CT F S CT F S
x(t) ←→ X[k] =⇒ y(t) = x(at) ←→ X[k],
that is the sequence of CTFS coecients Y [k] is identical to X[k], the CTFS representation for x(t) and
y(t) dier substantially, since they are dened for completely dierent periods, T 6= Ty . As a result, the
fundamental frequency for the periodic signal x(t) is 2π/T, which is dierent from that of y(t), which is
2πa/T . Hence, the frequency content of the signals dier substantially.
2.1.1.6 Conjugate Symmetry

The eect of conjugating a complex-valued signal on its CTFS representation can be seen by simply conju-
gating the CTFS synthesis relation,
∞
2πk
X
x(t) = X[k]ej T t
k=−∞
∞
X ∗
∗ j 2πk
T t
x (t) = X[k]e
k=−∞
∞
2πk
X
= X ∗ [k]e−j T t
k=−∞
∞
X 2π(−k)
= X ∗ [k]ej T t
k=−∞
∞
2πm
X
= X ∗ [−m]ej T t
m=−∞
yielding that
CT F S CT F S
x(t) ←→ X[k] =⇒ x∗ (t) ←→ X ∗ [−k].
When the periodic signal x(t) is real valued, i.e. x(t) only takes on values that are real numbers, then the
CTFS exhibits a symmetry property. This arises directly from the denintion of the CTFS, and that real
numbers equal their conjugates, i.e. x(t) = x∗ (t), such that
CT F S
x(t) = x∗ (t) ←→ X[k] =⇒ X[k] = X ∗ [−k].
Note that when the signal is real-valued and is an even function of time, such that x(t) = x(−t), then its
CTFS is also real-valued and even, i.e. X[k] = X ∗ [k] = X[−k]. It can be shown by similar reasoning that
when the signal is periodic, real-valued, and an odd function of time, that the CTFS coecients are purely
imaginary and odd, i.e. X[k] = −X ∗ [k] = −X[−k].
2.1.1.7 Products of Signals

When two periodic signals of the same period are multiplied in time, such that z(t) = x(t)y(t), the resulting
signal remains periodic with the same period, such that z(t) = x(t)y(t) = x(t + T )y(t + T ) = z(t + T ). Hence,

each of the three signals admit CTFS representations using the same set of harmonically related signals. We
can observe the eect on the resulting CTFS representation through the analysis equation,
Z T
1 2πk
Z[k] = (x(t)y(t))e−j T t dt
T t=0
∞
Z T !
1 X 2πm 2πk
= (y(t) X[m]ej T t )e−j T t dt
T t=0 m=−∞
∞ Z T
X 1 2π(k−m)
= X[m] y(t)e−j T t
dt
m=−∞
T t=0
∞
X
= X[m]Y [k − m].
m=−∞
The relationship between the CTFS coecients for z(t) and those of x(t) and y(t) is called a discrete
convolution between the two sequences X[k] and Y [k],
∞
CT F S CT F S CT F S
X
x(t) ←→ X[k], y(t) ←→ Y [k] =⇒ z(t) = x(t)y(t) ←→ X[m]Y [k − m].
m=−∞
2.1.1.8 Convolution
A dual relationship to that of multiplication in time, is multiplication of CTFS coecients. Specically,
when the two signals x(t) and y(t) are each periodic with period T, the periodic signal z(t) of period T,
whose CTFS representation is given by Z[k] = X[k]Y [k] corresponds to a periodic convolution of the signals
x(t) and y(t). This can be seen as follows,
∞
2πk
X
z(t) = (X[k]Y [k]) ej T t
k=−∞
∞
!
Z T
X 1 −j 2πk 2πk
= x(τ )e T τ dτ Y [k]ej T t
T τ =0
k=−∞
∞
!
Z T
1 X
j 2πk
= x(τ ) Y [k]e T (t−τ ) dτ
T τ =0 k=−∞
Z T
1
= x(τ )y(t − τ )dτ
T τ =0
where the integral relationship in the last line is called periodic convolultion. This leads to the following
property of the CTFS,
Z T
CT F S CT F S 1 CT F S
x(t) ←→ X[k], y(t) ←→ Y [k] =⇒ z(t) = x(τ )y(t − τ )dτ ←→ Z[k] = X[k]Y [k].
T τ =0
2.1.1.9 Integration
Rt
When the signal y(t) and x(t) are related through a running integral, i.e. y(t) = τ =0
x(τ )dτ , we can relate
their CTFS as follows,

Z t Z t ∞
d d X 2πk
x(t) = x(τ )dτ = ( X[k]ej T τ + X[0])dτ
dt τ =0 dt τ =0 k=−∞,k6=0
∞ Z t
d X 2πk
= ( X[k] ej T τ dτ + X[0]t)
dt τ =0
k=−∞,k6=0
∞ t
d X T j 2πk τ
= X[k] e T + X[0]
dt j2πk τ =0
k=−∞,k6=0
∞
d X T j 2πk t
= X[k] e T − 1 + X[0]
dt j2πk
k=−∞,k6=0
∞ ∞
d X T j 2πk t d X T
= e T −
X[k] X[k] + X[0]
dt j2πk dt j2πk
k=−∞,k6=0 k=−∞,k6=0
∞
d X T 2πk
= X[k] ej T t + X[0]
dt j2πk
k=−∞,k6=0
∞
d X 2πk
= ( Y [k]ej T t + X[0]t)
dt
k=−∞,k6=0
∞
X T 2πk
From this, if we let y(t) = X[k] ej T t + X[0]t
j2πk
k=−∞,k6=0
∞
d X 2πk
y(t) = X[k]ej T t + X[0]
dt
k=−∞,k6=0
= x(t).
This yields the property,
(
t T
6 0
Z
CT F S CT F S j2πk X[k] k=
x(t) ←→ X[k] =⇒ y(t) = x(τ )dτ ←→ ,
τ =0 0 k=0
where we must only consider x(t) such that X[0] = 0, or else y(t) would not be periodic.
2.1.1.10 Dierentiation
d
Similarly, we can consider the relationship between y(t) = dt x(t) and their corresponding CTFT represen-
tations. From the denition of the CTFS, we have
d
y(t) = x(t)
dt
∞
d X 2πk
= X[k]ej T t
dt
k=−∞
∞
X d j 2πk t
= X[k]
e T
dt
k=−∞
∞
X j2πk 2πk
= X[k] ej T t
T
k=−∞
from which we obtain the relation


CT F S d CT F S j2πk
x(t) ←→ X[k] =⇒ y(t) = x(t) ←→ X[k].
dt T
2.1.1.11 Parseval's relation

The energy containted within a period of a periodic signal can also be computed in terms of its CTFS
representation using Parseval's relation,
Z T ∞
CT F S 1 2
X
x(t) ←→ X[k] =⇒ |x(t)| dt = |X[k]|2 .
T t=0 k=−∞
This relation can be derived using the denition of the CTFS as follows,
Z T Z T
1 1
|x(t)|2 dt = x(t)x∗ (t)dt
T t=0 T t=0
∞
!
ZT
1 X
∗ j 2πk
= x(t) X [−k]e T t dt
T t=0 k=−∞
∞
!
Z T
X
∗ 1 j 2πk
= X [−k] x(t)e T t dt
T t=0
k=−∞
∞
!
Z T
X 1 2π(−k)
= X ∗ [−k] x(t)e−j T t
dt
T t=0
k=−∞
X∞
= X ∗ [−k]X[−k]
k=−∞
X∞
= |X[m]|2 .
m=−∞
Parseval's relation shows that the energy in a period of a periodic signal is equal to the sum of the energies
contained within each of the harmonic components that make up the signal through the CTFS representation.
2.2 Fourier transform representation of CT signals

Now that we have seen how we may build-up a large class of continuous-time periodic signals from the
set of simpler complex exponential periodic signals, we return to apply this line of thinking to the more
general class of continuous-time aperiodic (not periodic) signals. Just as was the case for periodic signals,
a remakably rich class of aperiodic signals can also be constructed from linear combinations of complex
exponentials. In the case of periodic continuous-time signals, since the signals of interest were periodic, the
CTFS was restricted to contruct such signals through combinations of harmonically related exponentials.
However for more general aperiodic signals, we may consider building an even larger class of signals by
removing this restriction on the ingredients used to makeup a given signal. Since harmonically related
complex exponentials can be enumerated, the CTFS took the form of a summation over the countably
innite set of all harmoically related exponentials of a given fundamental frequency. However, removing the
restriction to only using harmonically related terms, the class of all possible complex exponentials arises from
a continuum of possible frequenecy components and the form used with which to contruct linear combinations
will take the form of an integral, rather than an innite summation. Just as with the continuous-time Fourier
series, where the CTFS analysis equation provided a method for calculating the frequency components that
makeup a given periodic signal, the continuous-time Fourier transform provides a method for calculating the
spectrum of frequency components that makup an aperiodic signal from this class. The resulting integral
used to contruct this large class of signals using this specic spectrum of frequency components is called the
Fourier integral, or the continuous-time Fourier synthesis equation.

2.2 Fourier transform representation of CT signals 21
One method for introducing the continous-time Fourier transform is through the CTFS. By considering
continuous-time aperiodic signals as the result of taking continous-time periodic signals to the limit of an
innite period, we may observe how the CTFS transitions from a countable sum of harmonically-related
complex exponentials, into a continuous integral over the continuum of possible frequencies. Let us return
to the square wave signal that we visited in Figure 2.2. In this case, however, we will alter the signal to take
the form

1, 0 ≤ t ≤ 1
x(t) =
0 else
over the unit interval, t ∈ [0, 1]. Using (2.10), we once again obtain its CTFS representation, however this
time, we consider the period of repetition of the on period of the square wave to be given by the variable
T, i.e. we have

1, 0 ≤ t ≤ 1
x(t) =
0 else
for t ∈ [0, T ], and then repeating every T seconds. This yields the following CTFS representation
Z T
2πk
X[k] = x(t)e−j T t dt
0
Z 1
2πk
= e−j T t dt
0
−T −j 2πk
= e T −1
j2πk
−T −j πk −j πk πk

= e T e T − ej T
j2πk

T −j πk πk
= e T 2j sin
j2πk T

 sin( T ) −j πk
πk
πk e T k 6= 0
= T (2.15)
1 k = 0,
where the k = 0 term is once again determined by closer examination of the rst line of the derivation,
rather than attempting further analysis on the expression at containing vanishing terms in the numerator
and demoninator. We consider the expression in (2.15) for various values of T in Figure 2.3. By plotting
2πk
the magnitude of the CTFS coecients |X[k]| versus the harmonicaly related frequency components
T
for various values of T, ranging from T = 4, up to T = 32, we see that the envelope containing the CTFS
coecients remains constant, while the CTFS coecients move closer and closer to one another in absolute
frequency.
The envelope that is observed in the gure, can be viewed as the value that the CTFS representation
would take on as the period of the signal is made larger and larger. Recognizing this process, Fourier dened
this envelope as
Z ∞
X(ω) = x(t)e−jωt dt, (2.16)
t=−∞
where the frequency variable ω takes on all values on the real line, and for which (2.16) is known as the
continuous-time Fourier transform (CTFT). For this example, the continuous-time Fourier transform would
evaluate to

CTFS representation for T = 4 CTFS representation for T = 8

1 1
0.8 0.8
0.6 0.6
|X[k]|
|X[k]|
0.4 0.4
0.2 0.2
0 0
-30 -20 -10 0 10 20 30 -30 -20 -10 0 10 20 30
2πk/T 2πk/T
CTFS representation for T = 16 CTFS representation for T = 32

1 1
0.8 0.8
0.6 0.6
|X[k]|
|X[k]|
0.4 0.4
0.2 0.2
0 0
-30 -20 -10 0 10 20 30 -30 -20 -10 0 10 20 30
2πk/T 2πk/T
Figure 2.3: CTFS representation of the periodic signal in 2.17for T = 4, 8, 16, 32.
Z ∞
X(ω) = x(t)e−jωt dt
−∞
Z 1
= e−jωt dt
0
−1 −jω
= e −1
jω
−1 −jω/2 −jω/2
= e e − ejω/2
jω
1
= e−jω/2 2j sin (ω/2)
jω

 sin( ω2 ) −j ω2
ω e ω 6= 0
= 2 (2.17)
1 ω = 0.
While the CTFT analysis equation (2.16) provides the composition of any of a large class of signals
through a linear superposition of complex exponential signals of the form ejωt , the CTFT synthesis equation
provides the recipe for constructing such signals from their constituent set, as
Z ∞
1
x(t) = X(ω)ejωt dω.
2π ω=−∞
Together, the two expressions make up the CTFT representation for aperiodic signals,
Z ∞
1
x(t) = X(ω)ejωt dω
2π ω=−∞
Z ∞
X(ω) = x(t)e−jωt dt
t=−∞

CT Fourier Transform Representation of Aperiodic Signals

Z ∞
1
x(t) = X(ω)ejωt dω (2.18)
2π ω=−∞
Z ∞
X(ω) = x(t)e−jωt dt (2.19)
t=−∞
2.2.1 CT Fourier Transform Properties

We have now been properly introduced to a method for building-up continuous-time aperiodic signals from
a class of complex exponential signals in (2.18) and a method for analysing the make-up of such periodic
signals in terms of their constituent sinusoidal components in (2.19). Once again, now that introductions
are out of the way, we can explore some of the many useful properties of the CTFT representation. Many
of the properties of the CTFT follow directly, or along similar lines, of those of the CTFS.
2.2.1.1 Linearity
The CTFT can be viewed as a linear operation, in the following manner. When two signals x(t) and y(t)
are each constructed from their constituent complex exponential signals according to the CTFT synthesis
equation, the linear combination of these signals, z(t) = ax(t) + by(t), for a, b real-valued constants, can
be readily obtained by combining the constituent complex exponential signals through the same linear
combination. More specically, when x(t) is an aperiodic signal with CTFT coecients X(ω) and y(t) is an
aperiodic signal with CTFT Y (ω) then the signal z(t) = ax(t) + by(t) has a CTFT representation given by
Z(ω) = aX(ω) + bY (ω). The linearity property of the CTFT can be compactly represented as follows
CT F T CT F T CT F T
x(t) ←→ X(ω), y(t) ←→ Y (ω) =⇒ z(t) = ax(t) + by(t) ←→ aX(ω) + bY (ω).
2.2.1.2 Time Shift

For the signal y(t) = x(t − t0 ), we have
Z ∞
Y (ω) = x(t − t0 )e−jωt dt
t=−∞
Z ∞
= x(s)e−jω(s+t0 ) ds
s=−∞
Z ∞
= x(s)e−jωt0 e−jωs ds
s=−∞
= X(ω)e−jωt0 ,
where, the second line follows from the change of variable, s = t − t0 . The time shift property of the CTFT
can be compactly represented as follows
CT F T CT F T
x(t) ←→ X(ω) =⇒ y(t) = x(t − t0 ) ←→ X(ω)e−jωt0 .
We see that a shift in time of an aperiodic signal corresponds to a modulation in frequency by a phase term
that is linear with frequency with a slope that is proportional to the delay.

When a signal x(t) has a CTFT representation given by X(ω), a natural question that might arise is the
what happens when the shifting that was discussed in section 2.2.1.2 is applied to the CTFT representation,
X(ω). Specically, if a signal y(t) were known to have a CTFT representation given by Y (ω) = X(ω − ω0 ), it

is interesting to understand the relationship in the time-domain between y(t) and x(t). This can be readily
seen through examination of the CTFT analysis equation,
Y (ω) = X(ω − ω0 )
Z ∞
= x(t)e−j(ω−ω0 )t dt
t=−∞
Z ∞
x(t)ejω0 t e−jωt dt,

=
t=−∞
CT F T CT F T
x(t) ←→ X(ω) =⇒ y(t) = x(t)ejω0 t ←→ X(ω − ω0 ).
We observe that a shift in the frequency of the continuous time Fourier transform by an amount ω0 corre-
sponds to a modulation in the time domain signal x(t) by a term whose frequency is proportional to the
shift amount.

Analogous to the result for the CTFS, we have from the CTFT synthesis equation,
Z ∞
1
x(t) = X(ω)ejωt dω,
2π ω=−∞
we see that by simply changing the sign of the time variable t, we obtain the general relation
Z ∞
1
y(t) = x(−t) = X(ω)e−jωt dω
2π ω=−∞
Z ∞
1
= X(ω)ej(−ω)t dω
2π ω=−∞
Z ∞
1
= X(−ω)ejωt dω,
2π ω=−∞
CT F T CT F T
x(t) ←→ X(ω) =⇒ y(t) = x(−t) ←→ X(−ω),
i.e., changing the sign of the time axis corresponds to changing the sign of the CTFT frequency index.
2.2.1.5 Time Scaling

When signal undergoes a time-scale change, such as one that compresses the time axes, y(t) = x(at), where
a>1 is a real-valued constant, the resulting signal y(t) is given by
Z∞
y(t) = X(ω)ejωat dω
ω=−∞
Z∞
1
= X(ν/a)ejνt dν,
|a|
ν=−∞
where the second line follows from the substitution ν = aω. This yields the following relation for y(t) = x(at),
CT F T CT F T 1
x(t) ←→ X(ω) =⇒ y(t) = x(at) ←→ X(ω/a).
|a|


The eect of conjugating a complex-valued signal on its CTFT representation can be seen by simply conju-
gating the CTFT synthesis relation,
Z ∞
1
x(t) = X(ω)ejωt dω
2π ω=−∞
Z ∞ ∗
1
x∗ (t) = X(ω)e dωjωt
2π ω=−∞
Z ∞
1
= X ∗ (ω)e−jωt dω
2π ω=−∞
Z ∞
1
= X ∗ (ω)ej(−ω)t dω
2π ω=−∞
Z ∞
1
= X ∗ (−ω)ejωt dω
2π ω=−∞
yielding that
CT F T CT F T
x(t) ←→ X(ω) =⇒ x∗ (t) ←→ X ∗ (−ω).
When the signal x(t) is real valued, then the CTFT exhibits a symmetry property. This arises directly from
the denintion of the CTFT, and that real numbers equal their conjugates, i.e. x(t) = x∗ (t), such that
CT F T
x(t) = x∗ (t) ←→ X(ω) =⇒ X(ω) = X ∗ (−ω).
Note that when the signal is real-valued and is an even function of time, such that x(t) = x(−t), then its
CTFT is also real-valued and even, i.e. X(ω) = X ∗ (ω) = X(−ω). It can be shown by similar reasoning that
when the signal real-valued, and an odd function of time, that the CTFT is purely imaginary and odd, i.e.
X(ω) = −X ∗ (ω) = −X(−ω).

When signals are multiplied in time, such that z(t) = x(t)y(t), the resulting signal has a CTFS representation
that can be obtained through the analysis equation,
Z ∞
Z(ω) = (x(t)y(t))e−jωt dt
t=−∞
Z ∞ Z ∞
1
= (y(t) X(ν)e dν )e−jωt dt
jνt
t=−∞ 2π ν=−∞
Z ∞ Z ∞
1 −j(ω−ν)t
= X(ν) y(t)e dt dν
2π ν=−∞ t=−∞
Z ∞
1
= X(ν)Y (ω − ν)dν.
2π ν=−∞
The relationship between the CTFT representation for z(t) and those of x(t) and y(t) is seen to be a
convolution between the two CTFTs X(ω) and Y (ω),
Z ∞
CT F T CT F T CT F T 1
x(t) ←→ X(ω), y(t) ←→ Y (ω) =⇒ z(t) = x(t)y(t) ←→ X(ν)Y (ω − ν)dν.
2π ν=−∞
2.2.1.8 Convolution
A dual relationship to that of multiplication in time, is multiplication of CTFT representations. Specically,
the signal whose CTFT representation is given by Z(ω) = X(ω)Y (ω) corresponds to a convolution of the
signals x(t) and y(t). This can be seen as follows,

Z ∞
1
z(t) = (X(ω)Y (ω)) ejωt dω
2π ω=−∞
Z ∞ Z ∞
1
= x(τ )e−jωτ dτ Y (ω)ejωt dω
2π ω=−∞ τ =−∞
Z ∞ Z ∞
1
= x(τ ) Y (ω)ejω(t−τ ) dτ
τ =−∞ 2π ω=−∞
Z ∞
= x(τ )y(t − τ )dτ
τ =−∞
where the integral relationship in the last line is recognized as a convolultion. This leads to the following
property of the CTFT,
Z ∞
CT F T CT F T CT F T
x(t) ←→ X(ω), y(t) ←→ Y (ω) =⇒ z(t) = x(τ )y(t − τ )dτ ←→ Z(ω) = X(ω)Y (ω).
τ =−∞
2.2.1.9 Integration
Rt
When the signal y(t) and x(t) are related through a running integral, i.e. y(t) = τ =−∞
x(τ )dτ , we can
relate their CTFTs as follows,
Z t
CT F T CT F T 1
x(t) ←→ X(ω) =⇒ y(t) = x(τ )dτ ←→ X(ω) + πX(0)δ(ω),
τ =−∞ jω
where the relation is easiest shown using the dierentiation property derived next together with the following
observation. When ω = 0, Y (ω) is unbounded if X(0) is nonzero.
2.2.1.10 Dierentiation
d
Similarly, we can consider the relationship between y(t) = dt x(t) and their corresponding CTFT represen-
tations. From the denition of the CTFT, we have
d
y(t) = x(t)
dt Z ∞
d 1
= X(ω)ejωt dt
dt 2π ω=−∞
Z ∞
1 d
= X(ω) ejωt dt
2π ω=−∞ dt
Z ∞
1
= (jωX(ω)) ejωt dt
2π ω=−∞
from which we obtain the relation
CT F T d CT F T
x(t) ←→ X(ω) =⇒ y(t) = x(t) ←→ jωX(ω).
dt

The energy containted in a nite-energy signal (note that the CTFT exists in the case of nite energy signals,
i.e. signals that can be square integrated) can also be computed in terms of its CTFT representation using
Parseval's relation,
Z ∞ Z ∞
CT F T 1
x(t) ←→ X(ω) =⇒ |x(t)|2 dt = |X(ω)|2 dω.
t=−∞ 2π ω=−∞

2.3 Discrete-Fourier Series representation of DT periodic signals 27
Section CTFT Property Continuous Time Signal Continuous Time Fourier Transform
R∞
Denition x(t) X(ω) = t=−∞ x(t)e−jωt dt
2.2.1.1 Linearity z(t) = ax(t) + by(t) Z(ω) = aX(ω) + bY (ω)
2.2.1.2 Time Shift y(t) = x(t − T ) Y (ω) = X(ω)e−jωT
2.2.1.3 Modulation y(t) = x(t)ejω0 t Y (ω) = X(ω − ω0 )
2.2.1.4 Time Reversal y(t) = x(−t) Y (ω) = X(−ω)
1
2.2.1.5 Time Scaling y(t) = x(at) Y (ω) = |a| X(ω/a)
2.2.1.6 Conjugate Symmetry x(t) = x∗ (t) X(ω)
R∞ = X ∗
(−ω)
1
2.2.1.7 Products of Signals z(t) = x(t)y(t) Z(ω) = 2π ν=−∞ X(ν)Y (ω − ν)dν
R∞
2.2.1.8 Convolution z(t) = τ =−∞ x(τ )y(t − τ )dτ Z(ω) = X(ω)Y (ω)
Rt 1
2.2.1.9 Integration y(t) = τ =−∞ x(τ )dτ Y (ω) = jω X(ω) + πX(0)δ(ω).
d
2.2.1.10 Dierentiation y(t) = dt x(t) R∞ Y (ω) = jωX(ω)
R∞
2 1 2
2.2.1.11 Parseval's Relation x(t) t=−∞
|x(t)| dt = 2π −∞ |X(ω)| dω.
Other properties? tx(t), even part, odd part
conjsym part, conjasym part
Table 2.1: Properties of the Continuous Time Fourier Transform
This relation can be derived using the denition of the CTFS as follows,
Z ∞ Z ∞
2
|x(t)| dt = x(t)x∗ (t)dt
t=−∞ t=−∞
Z ∞ Z ∞
1 ∗ −jωt
= x(t) X (ω)e dω dt
t=−∞ 2π ω=−∞
Z ∞ Z ∞
1 ∗ −jωt
= X (ω) x(t)e dt dω
2π ω=−∞ t=−∞
Z ∞
1
= X ∗ (ω) (X(ω)) dω
2π ω=−∞
Z ∞
1
= X ∗ (ω)X(ω)dω
2π ω=−∞
Z ∞
1
= |X(ω)|2 dω.
2π ω=−∞
Parseval's relation shows that the energy measured in the time-domain of a nite-energy signal is equal to
the energy measured in the frequency domain through its CTFT representation.
2.2.2 CTFT Examples

Derivations of some of the signals in the Table 2.2.
2.3 Discrete-Fourier Series representation of DT periodic signals

In Section 2.1 we discussed the Fourier series representation as a means of building a large class of continuous
time periodic signals from a set of simpler, harmonically related complex exponential signals. In this section,
we consider the analogous notion of building a large class of periodic signals in discrete time from a set of
simpler, harminically related complex exponential discrete time signals. An important dierence between
the continuous time Fourier series and what we will develop in this section as the discrete time Fourier series
(DTFS), is that while the series used to construct periodic signals in continuous time is innite, the series
used to construct discrete time periodic signals is in fact a nite sum. This dierence simplies a number

Continuous Time Signal Continuous Time Fourier Transform

R∞
x(t) X(ω) = t=−∞
x(t)e−jωt dt
−at 1
e u(t), Real{a} > 0 jω+a
te−at u(t), Real{a} > 0 1
(jω+a)2
ejω0 t 2πδ(ω − ω0 )
1 2πδ(ω)
δ(t − T0 ) e−jωT0
cos(ω0 t) π[δ(ω − ω0 ) + δ(ω + ω0 )]
sin(ω
( 0 t) −jπ[δ(ω( − ω0 ) − δ(ω + ω0 )]
sin(W t)
W Wt πt t 6= 0 1, |ω| < W
sinc( ) =
π π W
π t=0 0, |ω| > W
( (
2sin(ωT )
1, |t| < T ωT ω ω 6= 0
2Tsinc( π ) =
0, |t| > T 2T ω=0
more more
more more
more more
more more
more more
Table 2.2: Continuous Time Fourier Transform Pairs
of issues that were delicate in the continuous case, such as notions of convergence, and existence of certain
limits.
Mathematically, we represent a periodic discrete time signal, x[n], as a signal whose value repeats at
a xed number of samples from the present. This interval, denoted N below, is called the period of the
signal, and we express this relationship
x[n] = x[n + N ], for all n. (2.20)
Equation (2.20) will, in general, be satised for a countably innite number of possible values of N. The
smallest, positive value of N for which Eq. (2.20) is satised, is called the fundamental period of the signal
x[n]. Discrete time sinusoidal signals, such as
x[n] = sin(ω0 n + φ), (2.21)
often enable us to relate the frequency of oscillation, ω0 to a fundamental period, N . While analogous to their
continuous time cousins, discrete time sinusoids need not always be periodic. While this may require a more
careful notion of what is meant by discrete time frequency, we will place this issue aside for the moment
and consider how the period of a periodic sinusoid relates to the arguments of the sinusoidal function. This
can again be computed by noting that sinusoidal functions are equal when their arguments are either equal
or dier only through a multiple of 2π , i.e.
x[n] = x[n + N ]
sin(ω0 n + φ) = sin(ω0 (n + N ) + φ)
sin(ω0 n + φ + 2kπ) = sin(ω0 (n + N ) + φ)
sin(ω0 (n + 2kπ/ω0 ) + φ) = sin(ω0 (n + N ) + φ) (2.22)
which yields the relationship

N = 2πk/ω0 . (2.23)
Depending on the value of ω0 , (2.23) may not provide an integer solution for N for any value of k. Note
that only if ω0 /π is rational, will there be an integral solution to (2.23), for which the smallest integer value
of N is the fundamental period associated with the discrete time frequency ω0 . In Figure (2.4), the two

sin(π n /4) sin(3 n /4)

1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6
-0.8 -0.8
-1 -1
0 5 10 15 20 25 0 5 10 15 20 25
samples, n samples, n
Figure 2.4: Examples of periodic and aperiodic sinusoidal signals x[n] = sin(πn/4) and x[n] = sin(3n/4).
sinusoidal signals x[n] = sin(πn/4) and x[n] = sin(3n/4) are shown. Note that the fundamental period of
N = 2π/(π/4) = 8 can be readily seen from gure for the periodic signal x[n] = sin(πn/4). However, the
aperiodic signal x[n] = sin(3n/4) does not exhibit periodicity for any value of n seen in the gure, and since
the frequency argument of the sinusoid is not a rational multiple of π, we are guaranteed that no such integer
period exists.
As in continuous time, any two periodic signals, x[n] and y[n] with the same period N can be added
together to produce a new periodic signal of the same period, i.e.,
s[n] =x[n] + y[n]

s[n + N ] =x[n + N ] + y[n + N ] = s[n + N ].
We again consider how we might build-up a larger class of periodic signals from the basic building blocks
of harmonically-related discrete time sinusoids. To extend our discussion to include complex-valued signals,
we again employ Euler's relation to construct complex exponential signals of the form
x[n] =ej(ω0 n+φ) (2.24)
= cos(ω0 n + φ) + j sin(ω0 n + φ)
enabling us to write
x[n] = cejω0 n ,
where, c = ejφ is simply a complex constant whose eects on the sinusoidal nature of the signal have again
been conveniently parked in front of the discussion. Complex-exponential signals of the form (2.24) may be
periodic or aperiodic depending on whether or not ω0 /π is rational.
Analogous to the CTFS, we can explore the class of signals that can be constructed by such harmonically-
related complex exponentials of the form
N −1
1 X
x[n] = X[k]ejkω0 n , (2.25)
N
k=0
where ω0 = 2πk/N. Note that the summation in (2.25) only covers N terms, rather than the innite sum
in (2.6) for the CTFS. This is due to the ninte number of harmonically related complex exponentials that
can be constructed with period N. Note that since the independent (time) variable in discrete time signals
only takes on integer values, complex exponentials of frequency ω0 are indistinguishable from those with
frequency ω0 + k2π for any k, i.e.
ejω0 n = ej(ω0 +k2π)n .

This result together with the fundamental period of N yields,
2πk 2π(k+N )
ej N n = ej N n
.

As a result, there are only N distinct complex exponential signals of period N. The resulting DTFS synthesis
equation, written in terms of the fundamental period of the signal set becomes
N −1
1 X
x[n] = X[k]ej2πkn/N , (2.26)
N
k=0
where ω0 = 2π/N is the fundamental frequency of the periodic signal x[n]. The construction in (2.26) is
referred to as the discrete-time Fourier series (DTFS) representation of x[n] and (2.26) is often called the
discrete-time Fourier series synthesis equation.
The Fourier series coecients X[k] can be obtained by multiplying (2.26) by e−j2πkn/N and summing
over a period of duration N to obtain
−1 −1 N −1
N N
!
X
−j2πkn/N
X 1 X
x[n]e = X[m]ej2π(m−k)n/N
n=0 n=0
N m=0
−1 N −1 −1
N N
!
X
−j2πkn/N 1 X X
j2π(m−k)n/N
x[n]e = X[m] e .
n=0
N m=0 n=0
To proceed, we need to evaluate the sum
−1
N
(
1−ej2π(m−k)N/N
X
j2π(m−k)n/N 1−ej2π(m−k)/N
m 6= k
e =
n=0
N m=k
(
1−ej2π(m−k)
1−ej2π(m−k)/N
m 6= k
=
N m=k
(
0 m 6= k
=
N m = k,
which leads to the result
N −1 N −1
X 1 X
x[n]e−j2πkn/N = X[m]N δ[m − k],
n=0
N m=0
= X[k],
by the sifting property of the Kronocker delta function. We can now return obtain the discretre-time Fourier
series analysis equation,
N
X −1
X[k] = x[n]e−j2πkn/N . (2.27)
n=0
Putting the synethesis and analysis equations together, we have the discrete-time Fourier series representa-
tion of a periodic signal x[n] as
DT Fourier Series Representation of a Periodic Signal

N −1
2πkn
X
X[k] = x[n]e−j N , all k (2.28)
n=0
N −1
1 X 2πkn
x[n] = X[k]ej N , all n, (2.29)
N
k=0
note that by convention, we dene the signal X[k] over all values of k, noting that due to the periodicity of
the sequence x[n] and of the signals ej2πkn/N , the sequence X[k]will also be periodic with period N. In this

derivation we used the following useful result for summation of a nite-length geometric series, which holds
for any r 6= 1,
n
X rm − rn+1
rk = .
1−r
k=m
Example: DTFS of a Square Wave

Consider the periodic discrete time sequence of period N =8 that satises

1, 0 ≤ n < 4
x[n] = (2.30)
0 4≤n<8
Using (2.27), we obtain,
7
X
X[k] = x[n]e−j2πkn/8
n=0
3
X
= e−j2πkn/8
n=0
(
1−e−j2πk4/8
1−e−j2πk/8
, 6 0
k=
=
4, k=0
1−e−jπk
(
1−e−jπk/4
, 6 0
k=
=
4, k=0
( −jπk/2
e (ejπk/2 −e−jπk/2 )
e−jπk/8 (ejπk/8 −e−jπk/8 )
, k 6= 0
=
4, k=0
(
e−j3πk/8 sin(πk/2)
sin(πk/8) , k 6= 0
=
4, k = 0.
2.3.1 DT Fourier Series Properties

We have now been properly introduced to a method for building-up discrete-time periodic signals from a
class of simple sinusoidal signals in (2.38) and a method for analysing the make-up of such periodic signals
in terms of their constituent sinusoidal components in (2.37). Now that introductions are once again out of
the way, we can explore some of the many useful properties of the DTFS representation.
2.3.1.1 Linearity
The DTFS can be viewed as a linear operation, in the following manner. When two signals x[n] and y[n]
are each constructed from their constituent sinusoidal signals according to the DTFS synthesis equation
(2.38), the linear combination of these signals, z[n] = ax[n] + by[n], for a, b real-valued constants, can be
readily obtained by combining the constituent sinusoidal signals through the same linear combination. More
specically, when x[n] is a periodic signal with DTFS coecients X[k] and y[n] is a periodic signal with CTFS
coecients Y [k] then the signal z[n] = ax[n] + by[n] has DTFS coecients given by Z[k] = aX[k] + bY [k].
The linearity property of the DTFS can be compactly represented as follows
DT F S DT F S DT F S
x[n] ←→ X[k], y[n] ←→ Y [k] =⇒ z[n] = ax[n] + by[n] ←→ aX[k] + bY [k].
This result can be readily shown by substituting z[n] = ax[n] + by[n] into the summation in (2.37) and
expanding the summation into the two separate terms, one for X[k] and one for Y [k].

2.3.1.2 Time Shift

When a sinusoidal signal x[n] = sin[ω0 n] is shifted in time, the resulting signal x[n − n0 ] can be represented
in terms of a simple phase shift of the origional sinusoidal signal, i.e. x[n − n0 ] = sin(ω0 (n − n0 )) =
sin(ω0 n + φ),where φ = −ω0 n0 . Periodic signals that can be represented using the DTFS contain many
sinusoidal (or complex exponential) signals. When such periodic signals are delayed in time, each of the
constituent sinusoidal components of the signal are delayed by the same amount, however this translates
into a dierent phase shift for each component. This can be readily seen from the DTFS analysis equation
2.37, as follows. For the signal y[n] = x[n − n0 ], we have
N −1
2πk
X
Y [k] = x[n − n0 ]e−j N n
n=0
N −1 N −1−n
X 0
2πk 2πk
X
= x[m]e−j N (m+n0 ) + x[m]e−j N (m+n0 )
m=N −n0 m=0

N −1
2πk
X
= x[m]e−j N (m+n0 )
m=0
N −1
2πk 2πk
X
= x[n]e−j N n0 e−j N m
m=0
2πk
= X[k]e−j N n0 ,
where, the second line follows from the change of variable, m = n − n0 , and the third line follows from the
periodicity of both the signal x[n] and the signal e−j2πkn/N with period N , and the last line follows from the
−j2πkn0 /N
denition of X[k] after factoring the linear phase term e out of the sum. The time shift property
of the DTFS can be compactly represented as follows
DT F S DT F S 2πk
x[n] ←→ X[k] =⇒ y[n] = x[n − n0 ] ←→ X[k]e−j N n0 .
We see that a shift in time of a periodic signal corresponds to a modulation in frequency by a phase term
that is linear with frequency with a slope that is proportional to the delay.

When a periodic signal x[n] has a DTFS representation given by X[k], a natural question that might arise
is the what happens when the shifting that was discussed in section2.3.1.2 is applied to the DTFS repre-
sentation,X[k]. Specically, if a periodic signal y[n] were known to have a DTFS representation given by
Y [k] = X[k − k0 ], it is interesting to understand the relationship in the time-domain between y[n] and x[n].
This can be readily seen through examination of the CTFS analysis equation,
Y [k] = X[k − k0 ]
N −1
2π
X
= x[n]e−j N (k−k0 )n
n=0
N −1
2π 2π
X
x[n]ej N k0 n e−j N kn

=
n=0
DT F S DT F S
x[n] ←→ X[k] =⇒ y[n] = x[n]ejk0 ω0 n ←→ X[k − k0 ],
2π
where ω0 =N . We observe that a shift in the discrete time Fourier series coecients by an integer amount
k0 corresponds to a modulation in the time domain signal x[n] by a term whose frequency is proportional
to the shift amount.


From the DTFS synthesis equation,
N −1
1 X 2πk
x[n] = X[k]ej N n ,
N
k=0
we see that by simply changing the sign of the time variable n, we obtain the relation
N −1
1 X 2πk
y[n] = x[−n] = X[k]ej N n
N
k=0
N −1
1 X 2π(−k)
= X[k]e−j N n
N
k=0
N −1
1 X 2π(N −k)
= X[k]e−j N n
N
k=0
N −1
1 X 2πm
= X[N − m]e−j N n ,
N m=0
DT F S DT F S
x[n] ←→ X[k] =⇒ y[n] = x[−n] ←→ X[N − k],
i.e., changing the sign of the time axis corresponds to changing the sign of the DTFS frequency index, where,
to keep the terms within the range from 0 to N, we add N to the index, which has no impact on their values,
owing to the periodicity of the DTFS coecients X[k] with period N as a function of k.

The eect of conjugating a complex-valued signal on its DTFS representation can be seen by simply conju-
gating the DTFS synthesis relation,
N −1
1 X 2πk
x[n] = X[k]ej N n
N
k=0
NX−1 ∗
1 2πk
x∗ [n] = X[k]ej N n
N
k=0
N −1
1 X 2πk
= X ∗ [k]e−j N n
N
k=0
N −1
1 X 2π(−k)
= X ∗ [k]ej N n
N
k=0
N −1
1 X 2π(N −k)
= X ∗ [k]ej N n
N
k=0
N −1
1 X 2πm
= X ∗ [N − m]ej N n
N m=0
yielding that
DT F S DT F S
x[n] ←→ X[k] =⇒ x∗ [n] ←→ X ∗ [N − k].

When the periodic signal x[n] is real valued, i.e. x[n] only takes on values that are real numbers, then the
DTFS exhibits a symmetry property. This arises directly from the denintion of the DTFS, and that real
numbers equal their conjugates, i.e. x[n] = x∗ [n], such that
DT F S
x[n] = x∗ [n] ←→ X[k] =⇒ X[k] = X ∗ [N − k].
Note that when the signal is real-valued and is an even function of time, such that x[n] = x[−n], then
its DTFS is also real-valued and even, i.e. X[k] = X ∗ [k] = X[−k] = X[N − k]. It can be shown by
similar reasoning that when the signal is periodic, real-valued, and an odd function of time, that the DTFS
coecients are purely imaginary and odd, i.e. X[k] = −X ∗ [k] = −X[−k] = −X[N − k].

When two periodic signals of the same period are multiplied in time, such that z[n] = x[n]y[n], the resulting
signal remains periodic with the same period, such that z[n] = x[n]y[n] = x[n + N ]y[n + N ] = z[n + N ].
Hence, each of the three signals admit DTFS representations using the same set of harmonically related
signals. We can observe the eect on the resulting DTFS representation through the analysis equation,
N −1
2πk
X
Z[k] = (x[n]y[n])e−j N n
n=0
N −1 N −1
X 1 X j 2πm 2πk
= X[m]e N n
y[n]e−j N n
n=0
N m=0
N −1 NX−1
1 X −j
2π(k−m)
n
= X[m] y[n]e N
N m=0 n=0
N −1
1 X
= X[m]Y [k − m],
N m=0
where the periodicity of Y [k] is used to determine values of Y [k − m] for terms k−m that fall outside the
range of 0 to N − 1. The relationship between the DTFS coecients for z[n] and those of x[n] andy[n] is
seen to be a form of discrete convolution, called a periodic convolution, between the two sequences X[k] and
Y [k],
N −1
DT F S DT F S DT F S 1 X
x[n] ←→ X[k], y[n] ←→ Y [k] =⇒ z[n] = x[n]y[n] ←→ X[m]Y [k − m].
N m=0
2.3.1.7 Convolution
A dual relationship to that of multiplication in time, is multiplication of DTFS coecients. Specically,
when the two signals x[n] and y[n] are each periodic with period N, the periodic signal z[n] N,
of period
whose DTFS representation is given by Z[k] = X[k]Y [k] corresponds to a periodic convolution of the signals
x[n] and y[n]. This can be seen as follows,

2.4 Discrete-time Fourier transform representation of DT signals 35
N −1
1 X 2πk
z[n] = (X[k]Y [k]) ej N n
N
k=0
−1 −1
N N
!
1 X X
−j 2πk 2πk
= x[m]e N m Y [k]ej N n
N m=0
k=0
−1 N −1
N
!
X 1 X 2πk
= x[m] Y [k]ej N (n−m)
m=0
N
k=0
N
X −1
= x[m]y[n − m]
m=0
where the summationin the last line is called periodic convolultion. This leads to the following property of
the DTFS,
N −1
DT F S DT F S DT F S
X
x[n] ←→ X[k], y[n] ←→ Y [k] =⇒ z[n] = x[m]y[n − m] ←→ Z[k] = X[k]Y [k].
m=0

The energy containted within a period of a periodic signal can also be computed in terms of its CTFS
representation using Parseval's relation,
N −1 N −1
DT F S
X 1 X
x[n] ←→ X[k] =⇒ |x[n]|2 = |X[k]|2 .
n=0
N
k=0
This relation can be derived using the denition of the DTFS as follows,
N
X −1 N
X −1
|x[n]|2 = x[n]x∗ [n]
n=0 n=0
−1 N −1
N
!
X 1 X ∗ 2πk
= x[n] X [N − k]ej N n
n=0
N
k=0
−1 −1
N N
!
1 X
∗
X
j 2πk
= X [N − k] x[n]e N n
N n=0
k=0
N −1 −1
N
!
1 X ∗ X 2π(N −k)
= X [N − k] x[n]e−j N n
N n=0
k=0
N −1
1 X
= X ∗ [N − k]X[N − k]
N
k=0
N −1
1 X
= |X[m]|2 .
N m=0
Parseval's relation shows that the energy in a period of a periodic signal is equal to the sum of the energies
contained within each of the harmonic components that make up the signal through the DTFS representation.
2.4 Discrete-time Fourier transform representation of DT signals

As with continuous-time signals, it is often convenient to represent discrete-time signals as a linear combi-
nation of simpler signals, or basis signals . From continuous-time system theory, we know that complex

exponential signals of the form est are a special class of signals called, eigensignals in that when placed
as the input to a linear, time-invariant system, the output of the system will be of the form est scaled by a
complex constant. As a result, such signals played an important role in the development of signal analysis
and synthesis methods through the CT Fourier transform and Laplace transform. For discrete-time systems,
we have that eigensignals of discrete-time linear-shift invariant systems include all signals that can be written
in the form of a discrete-time complex exponential sequence, or zn for all n and for any, possibly complex,
z. By restricting the class of such signals to have unity magnitude, we arrive at the class of complex expo-
nentials of the form ejωn for all n and for real-valued ω. These signals play a particularly important role
in the analysis of discrete-time systems due to this eigenfunction property, which implies that the response
of a linear shift-invariant system to a complex exponential input will be a complex exponential output of
the same frequency with amplitude and phase determined by the system. For real-valued systems, i.e. sys-
tems with real-valued impulse responses, when the input is sinusoidal of a given frequency, the output will
remain sinusoidal of the same frequency, again with amplitude and phase determined by the system. This
important property of linear shift invariant systems makes the representation of signals in terms of complex
exponentials extremely useful for studying linear system theory.
The discrete-time Fourier transform enables the construction of a wide class of signals from a superposition
of complex exponentials. Through the eigenfunction property, the response of a linear shift invariant system
to any signal in this class, i.e. any signal with a discrete-time Fourier transform, can be constructed by
adding up the responses to each of the eigenfunctions that make up the original signal. By linearity of the
system, the response of the system to a linear combination of complex exponentials will be given by the
same linear combination of the responses to the complex exponentials. The eigenfunction property of LSI
systems enables us to express very simply the response of the system to each of these complex exponentials.
The discrete-time Fourier transform, or DTFT, enables the representation of discrete-time sequences by
a superposition of complex exponentials. Many sequences of interest can be represented by the following
Fourier integral
Z π
1
x[n] = Xd (ω)ejωn dω, (2.31)
2π −π
where,
∞
X
Xd (ω) = x[n]e−jωn (2.32)
n=−∞
is the discrete-time Fourier transform of the sequence x[n]. These two expressions comprise the Fourier
representation of the sequence x[n]. Note that the DTFT, Xd (ω), is a complex-valued function of the real-
valued variable ω, when the sum (2.32) exists. The integral corresponds to the inverse DTFT and represents
the synthesis of the signal x[n] from a superposition of signals of the form
1
Xd (ω)ejωn dω,
2π
where we interpret the integral as the limit of a Riemann sum, i.e.
Z π 2π/∆ω
1 X
Xd (ω)ejωn dω = lim Xd (−π + k∆ω)ejωn ∆ω.
2π −π ∆ω→0
k=0
The value of the DTFT, Xd (ω), determines the relative amount of each of the complex exponentials ejωn that
is required to construct x[n]. The DTFT is referred to as Fourier analysis, as we analyze the composition of
the signal in terms of the complex exponentials that make it up. The inverse DTFT is referred to as Fourier
synthesis, as it can be viewed as synthesizing the signal from these basic components that make it up.
There is a strong similarity between the discrete-time Fourier transform and the z-transform for discrete-
time signals that we will study in Chapter 5. This relationship is similar to that between the continuous-
time Fourier transform and the Laplace transform for continuous-time signals. The more general Laplace
transform of a continuous-time signal can be written

e jω
angle = ω
Xd(ω) = X(ejω) 1
Figure 2.5: The DTFT viewed as evaluating the z-transform along the unit circle z = ejω in the z -plane.
Z ∞
XL (s) = x(t)e−st dt, (2.33)
t=−∞
when the integral exists. Substituting s = jω into (2.33), yields the CTFT. So, for signals for which the
CTFT exists, we can view the CTFT as a slice of the Laplace transform through the complex s-plane, along
the imaginary axis. Just as the Fourier transform for continuous-time signals can be viewed as evaluating the
Laplace transform along a specic curve, namely the imaginary axis in the s-plane, the DTFT can be viewed
as evaluating the more general z-transform along a specic curve in the complex z -plane. The z -transform
of a discrete-time sequence, given by,
∞
X
X(z) = x[n]z −n , (2.34)
n=−∞
is the same as the DTFT for values of z evaluated for a particular slice of the complex z -plane. Specically,
the DTFT can be seen to be the same as the z-transform evaluated along a curve in the z-plane corresponding
to the unit-circle, i.e.,
The DTFT exists as a regular function if and only if the region of convergence of the z-transform, the
values of z for which the summation in (2.34) converges, includes the unit circle, i.e. |z| = 1. For the case of
sinusoidal sequences, where X(z) contains poles on the unit circle, Xd (ω) can be dened in terms of impulse
distributions.
While for continuous-time signals, the notion of angular frequency is relatively well-dened, for discrete-
time signals, we also refer to the variable ω in Xd (ω) as the digital frequency. Angular frequency in
continuous-time is measured in Hz (cycles per second) or radians/sec. For discrete-time, angular frequency
is measured in cycles per sample or radians per sample. In some textbooks the variable ω is used to represent
analog frequency in the continuous-time Fourier transform. Here, we will use the variable ω to denote both
continuous-time frequency and discrete-time frequency and the specic meaning will be clear by the context.
For example, we will always refer to discrete-time Fourier transforms using the subscript d as in Xd (ω). Of
course, we could use any variable for the DTFT and the continuous-time Fourier transform. When necessary,
as in an expression relating a continuous-time frequency variable to an equivalent discrete-time frequency
variable through sampling, we may use Ω to represent analog frequenecy and ω to represent digital frequency.
This enables us to maintain clarity in our discussion and consistency with a number of other texts on the
topic. It is important to recall that while continuous-time sinusoids have a xed relationship between their
frequency of oscillation and the period of the periodic time-domain waveform, discrete-time sinusoids may
not be periodic at all. Recall that a signal of the form
x[n] = ejωn
is only periodic if the following relation holds
x[n] = x[n + P ].
Specally, we must have that
ej(ω0 n+k2π) = ej(ω0 (n+P ))

which corresponds to requiring that

2π P
= ,
ω0 k
i.e., the digital frequency must be a rational multiple of π. This relationship will certainly only hold for a
subset of all possible digital frequencies. Since the rational numbers are countable, and the real numbers are
uncountable, this relationship does not hold almost everywhere in ω. That is, for practical purposes, almost
any digital frequency you come up with, say by spinning a wheel and selecting the angle of the resulting
position with respect to its starting position, will correspond to a complex exponential sequence that is not
periodic. As a result, we rarely discuss the period of discrete-time complex exponentials and refer only to
their digital frequency instead.
To demonstrate that the Fourier transform synthesis equation, or inverse DTFT, in fact inverts the
DTFT, we can simply plug the denition of the DTFT into the synthesis equation as follows. From the
DTFT synthesis equation, we have
Z π
1
x[n] = Xd (ω)ejωn dω
2π −π
Z π ∞
1 X
−jωm
= x[m]e ejωn dω
2π −π m=−∞
∞ Z π
X 1
= x[m] ejω(n−m) dω
m=−∞
2π −π
∞
X
= x[m]δ[n − m]
m=−∞
= x[n],
where we have used that
(
π
1, n=m
Z
1 jω(n−m)
e dω = ejπ(n−m) −e−jπ(n−m)
2π −π 2πj(n−m) , n 6= m,
(
1, n=m
= (−1)−(−1)
2πj(n−m) , n 6= m
(
1, n = m
=
6 m
0, n =
= δ[n − m].
2.4.1 Properties of the DTFT

A number of important properties of the DTFT can be derived in a manner similar to those for the DTFS.
These are summarized at the end of this section in Table (2.5). Table (2.4) includes a number of DTFT
pairs.
2.4.1.1 Linearity
The DTFT can be viewed as a linear operation, in the following manner. When two signals x[n] and y[n]
satsify
DT F T
x[n] ←→ Xd (ω)
and

DT F T
y[n] ←→ Yd (ω),
the linear combination of these signals, z[n] = ax[n] + by[n], for a, b real-valued constants, can be readily
obtained by combining the constituent complex exponential signals through the same linear combination.
This is easily shown from the denition of the DTFT as follows
∞
X
Zd (ω) = (ax[n] + by[n])e−jωn
n=−∞
X∞ ∞
X
= a x[n]e−jωn + b y[n]e−jωn
n=−∞ n=−∞
= aXd (ω) + bYd (ω).
The linearity property of the DTFT can be compactly represented as follows
DT F T DT F T DT F T
x[n] ←→ Xd (ω), y[n] ←→ Yd (ω) =⇒ z[n] = ax[n] + by[n] ←→ aXd (ω) + bYd (ω).
2.4.1.2 Periodicity
The DTFT of every sequence is always periodic in that the following relation holds
Xd (ω) = Xd (ω + k2π),
for all integers k. The proof of this property lies in the periodicity of the complex exponential sequences ejωn
that are used to construct each sequence with a DTFT as follows,
∞
X
Xd (ω) = x[n]e−jωn
n=−∞
X∞
= x[n]e−j(ω+k2π)n
n=−∞
= Xd (ω + k2π).
This is dierent from the continuous-time Fourier transform, where we were interested in frequencies
spanning an innite range of real values. In contrast, in discrete-time, all digital frequencies can be captured
in a single interval of length 2π. The reason for this periodicity stems directly from the observation that
complex exponentials of the form ejωn are only unique over an interval of this range. That is the sequence
jωn
e is identical to the sequence ej(ω+2π)n . Since these two sequences have identical values, for all n, then
the composition of x[n] in terms of these sequences, i.e. the DTFT, must only require a single interval
containing them. Since the DTFT is periodic with period 2π , the DTFT only needs to be specied over an
interval of that length. It is often convenient to use the interval −π ≤ ω ≤ π so that the low frequencies are
centered around ω = 0. Note that since all frequencies that are multiples of 2π are indistinguishable, the
low frequencies are also those centered around any multiple of 2π . Similarly, the highest digital frequency
corresponds to ω=π as well as all odd multiples of π. We will return to this issue again when we discuss
the discrete-time frequency response of linear shift-invariant systems.
2.4.1.3 Real and Imaginary Part Symmetries

For real-valued sequences x[n], we have that the real-part of the DTFT is even, and the imaginary part is
odd, i.e.,
The proof follows from trigonometric properties of the real and imaginary parts. Specically, for real
valued x[n],we have


|Xd(ω)|
1
0.8
0.6
0.4
0.2
0
-2π -π 0 π 2π
frequency, ω
∠ Xd(ω)
4
-2
-4
-2π -π 0 π 2π
frequency, ω
Figure 2.6: Magnitude and phase of an example DTFT.
∞
X
<{Xd (ω)} = <{ x[n]e−jωn }
n=−∞
X∞
= <{ x[n](cos(ωn) + j sin(ωn))}
n=−∞
X∞
= <{ x[n](cos(ωn) + j sin(ωn))}
n=−∞
∞
X
= x[n] cos(ωn)
n=−∞
X∞
= x[n] cos(−ωn)
n=−∞
= <{Xd (−ω)}.
That the imaginary part of the DTFT is an odd function of ω, similarly follows from the antisymmetry of
the sine function.
2.4.2 Magnitude and Phase Symmetries

For real-valued sequences x[n], we have that the magnitude of the DTFT is an even function and the phase
of the DTFT is an odd function, i.e.,
|X(ω)| = |Xd (−ω)|

∠X(ω) = −∠Xd (−ω).
For example, |Xd (ω)| and ∠Xd (ω) might look as shown in Figure (2.6)

The proof of the magnitude and phase symmetries follows from the denition of the DTFT as below.
From the denition of the DTFT we have

∞
!2 ∞
!2 1/2
X X
|Xd (ω)| =  x[n] cos(ωn) + − x[n] sin(ωn) 
n=−∞ n=−∞

∞
!2 ∞
!2 1/2
X X
=  x[n] cos(ωn) + x[n] sin(ωn) 
n=−∞ n=−∞

∞
!2 ∞
!2 1/2
X X
=  x[n] cos(−ωn) + − x[n] sin(−ωn) 
n=−∞ n=−∞
= |Xd (−ω)|,
and for the phase of the DTFT we have that
P∞
− x[n] sin(ωn)
∠Xd (ω) = arctan P∞n=−∞
n=−∞ x[n] cos(ωn)
P∞
n=−∞ x[n] sin(−ωn)
= arctan P∞
n=−∞ x[n] cos(−ωn)
P∞
− n=−∞ x[n] sin(−ωn)
= − arctan P ∞
n=−∞ x[n] cos(−ωn)
= −∠Xd (ω),
as desired. The last line above follows since arctan is an odd function of its argument.
2.4.2.1 Time Shift

As we have seen for both continuous-time and discrete-time periodic signals, when a sinusoidal signal is
shifted in time, the resulting signal can be represented in terms of a simple phase shift of the origional
sinusoidal signal. A discrete-time signal x[n] that can be represented using the DTFT as a superposition of
possibly innitely many complex exponential signals of the form ejωn ,would necessarily have each of these
constituent complex exponential signals delayed by the same xed amount, which whould correspond to each
of the complex exponential signals undergoing a dierent shift in the phase of their exponent. The resulting
change in the DTFT of a discrete time signal x[n] that is delayed by a xed amount, i.e, y[n] = x[n − n0 ]
can be derived as follows
∞
X
Yd (ω) = x[n − n0 ]e−jωn
n=−∞
X∞
= x[m]e−jω(m+n0 )
n=−∞
X∞
= x[m]e−jωn0 e−jωm
m=−∞
= Xd (ω)e−jωn0 ,
where, the second line follows from the change of variable, m = n − n0 . The time shift property of the DTFT
can be compactly represented as follows

Figure 2.7: Sample DTFT Xd (ω) for the signal x[n].
DT F T DT F T
x[n] ←→ Xd (ω) =⇒ y[n] = x[n − n0 ] ←→ Xd (ω)e−jωn0 .
We see that a shift in time corresponds to a delay of each of the complex exponential components that
make up the signal and that this delay, in turn corresponds to a shift in the phase of each of the frequency
components by an amount that is linear with frequency with a slope that is proportional to the delay.
2.4.2.2 Modulation
When a signal x[n] has a DTFT representation given by Xd (ω), we again are interested in how a shift in
frequency would manifest itself in the time domain representation of the original signal. Specically, if a
signal y[n] were known to have a DTFT representation given by Yd (ω) = Xd (ω − ω0 ), it is interesting to
understand the relationship in the time-domain between y[n] and x[n]. This can be readily seen through
examination of the DTFT analysis equation,
Yd (ω) = Xd (ω − ω0 )
∞
X
= x[n]e−j(ω−ω0 )n
n=−∞
X∞
x[n]ejω0 n e−jωn

=
n=−∞
DT F T DT F T
x[n] ←→ Xd (ω) =⇒ y[n] = x[n]ejω0 n ←→ Xd (ω − ω0 ).
We observe that a shift in the discrete time Fourier transform by an amount ω0 corresponds to a modulation
in the time domain signal x[n] by a term whose frequency is proportional to the shift amount. This property
can be used together with linearity to determine the eect of modulation of a signal by a sinusoidal signal,

1 jω0 n
y[n] = cos(ω0 n)x[n] = e + e−jω0 n x[n]
2
resulting in

DT F T DT F T 1 1
x[n] ←→ Xd (ω) =⇒ y[n] = x[n] cos(ω0 n) ←→ Xd (ω − ω0 ) + Xd (ω + ω0 ) .
2 2
Example:
If x[n] has a DTFT as shown in Figure (2.7)
then y[n] = ejω0 n x[n] would have the DTFT as shown in Figure (2.8).
Example
If Xd (ω) has the form shown in Figure (2.9),
then Yd (ω), the DTFT of y[n] = cos(ω0 n)x[n] has the form shown in Figure (2.10)

Figure 2.8: Resulting DTFT of y[n] = ejω0 n x[n].
Figure 2.9: Example DTFT for x[n].

From the DTFT synthesis equation,
Z π
1
x[n] = Xd (ω)ejωn dω,
2π −π
we see that by simply changing the sign of the time variable n, we obtain the relation
Z π
1
y[n] = x[−n] = Xd (ω)ejω(−n) dω
2π −π
Z π
1
= Xd (ω)ej(−ω)n dω
2π −π
Z π
1
= Xd (−ω)ejωn dω,
2π −π
DT F T DT F T
x[n] ←→ Xd (ω) =⇒ y[n] = x[−n] ←→ Xd (−ω),
i.e., changing the sign of the time axis corresponds to changing the sign of the DTFT frequency index, ω.

The eect of conjugating a complex-valued signal on its DTFT representation can be seen by simply conju-
gating the DTFT synthesis relation,
Figure 2.10: Resulting DTFT for y[n].

Z π
x[n] = Xd (ω)ejωn dω
ω=−π
Z π ∗
x∗ [n] = Xd (ω)ejωn dω
ω=−π
Z π
= Xd∗ (ω)e−jωn dω
ω=−π
Z π
= Xd∗ (ω)ej(−ω)n dω
ω=−π
Z π
= Xd∗ (−ω)ejωn dω
ω=−π
yielding that
DT F T DT F T
x[n] ←→ Xd (ω) =⇒ x∗ [n] ←→ Xd∗ (−ω).
DTFT exhibits additional symmetry. This arises directly from the denintion of the DTFT, and that real
DT F T
x[n] = x∗ [n] ←→ Xd (ω) =⇒ Xd (ω) = Xd∗ (−ω).
Note that when the signal is real-valued and is an even function of time, such that x[n] = x[−n], then its
DTFT is also real-valued and even, i.e. Xd (ω) = Xd∗ (−ω) = Xd∗ (ω) = Xd (−ω). It can be shown by similar
reasoning that when the signal is real-valued and an odd function of time, that the DTFT is purely imaginary
and odd, i.e. Xd (ω) = Xd∗ (−ω) = −Xd∗ (ω) = −Xd (−ω).

When two discrete-time signals that can each be represented by a DTFT are multiplied in time, such that
z[n] = x[n]y[n], the resulting signal also has a DTFT representation. We can observe the eect on the
resulting DTFT representation through the analysis equation,
∞
X
Zd (ω) = (x[n]y[n])e−jωn
n=−∞
∞ Z π
X 1
= Xd (ν)e jνn
dν y[n]e−jωn
n=−∞
2π ν=−π
Z π ∞
X
1
= Xd (ν) y[n]e−j(ω−ν)n dν
2π ν=−π n=−∞
Z π
1
= Xd (ν)Yd (ω − ν)dν,
2π ν=−π
where the periodicity of Yd (ω) is used to determine values of Yd (ω − ν) for terms ω − ν outside the range of
[−π, π]. The relationship between the DTFTs of z[n] and of x[n] and y[n] is seen to be a form of convolution,
called a periodic convolution, between the two functionsXd (ω) and Yd (ω),
Z π
DT F T DT F T DT F T 1
x[n] ←→ Xd (ω), y[n] ←→ Y (ω) =⇒ z[n] = x[n]y[n] ←→ Xd (ν)Yd (ω − ν)dν.
2π −π

2.4.2.6 Convolution
A dual relationship to that of multiplication of discrete-time signals in time, is the multiplication of their
DTFT representations. Specically, when two signals x[n] and y[n] have corresponding DTFT representations
Xd (ω) and Yd (ω), the signal that corresponds to the DTFT Zd (ω) = Xd (ω)Yd (ω) corresponds to a discrete-
time convolution of the signals x[n] and y[n]. This can be seen as follows,
Z π
1
z[n] = (Xd (ω)Yd (ω)) ejωn dω
2π−π
∞
Z π !
1 X
−jωm
= x[m]e Yd (ω)ejωn dω
2π −π m=−∞
∞ Z π
X 1 jω(n−m)
= x[m] Yd (ω)e dω
m=−∞
2π −π
∞
X
= x[m]y[n − m].
m=−∞
This leads to the following property of the DTFT,
∞
DT F T DT F T DT F T
X
x[n] ←→ Xd (ω), y[n] ←→ Yd (ω) =⇒ z[n] = x[m]y[n − m] ←→ Zd (ω) = Xd (ω)Yd (ω).
m=−∞

A particularly useful relationship between the energy of a sequence in the time-domain and the energy
contained in its Fourier transform is captured by Parseval's relation. Since sequences with convergent
DTFTs are square summable, they have nite energy and we have that
∞ Z π
DT F T
X 1
x[n] ←→ Xd (ω) =⇒ |x[n]|2 = |Xd (ω)|2 dω.
n=−∞
2π −π
This relation can be derived using the denition of the DTFT as follows,
∞
X ∞
X
|x[n]|2 = x[n]x∗ [n]
n=−∞ n=−∞
∞ Z π ∗
X 1
= x[n] Xd (ω)ejωn dω
n=−∞
2π −π
∞ Z π
X 1
= x[n] Xd∗ (ω)e−jωn dω
n=−∞
2π −π
∞
!
Z π
1 X
= Xd∗ (ω) x[n]e −jωn
dω
2π −π n=−∞
Z π
1
= X ∗ (ω)Xd (ω)dω
2π −π d
Z π
1
= |Xd (ω)|2 dω.
2π −π
Example:
Parseval's relation can be used to compute the energy of a signal in the time domain or the
frequency domain. As a result, one of these is often simpler than the other. For example,

consider the sequence x[n] = u[n] − u[n − 10]. A rather complicated integral can be reduced to
a simple sum by noting the transform pair
DT F T sin(5ω/2)
u[n] − u[n − 10] ←→
sin(ω/2)
and using Parseval's relation as follows:
Z π 2 9
1 sin(5ω/2) X
dω = (1)2
2π −π sin(ω/2) n=0
= 10
Examples of DTFT
We continue our discussion of the discrete-time Fourier transform by considering a few simple
examples.
Example
Consider the following sequence containing two non-zero samples,
x[n] = δ[n] + δ[n − 1].
The discrete-time Fourier transform of this sequence can be computed directly from the denition
of the DTFT as
Xd (ω) = 1 + e−jω .
The magnitude of the discrete-time Fourier transform can be easily computed as
|Xd (ω)|2 = |1 + e−jω |2

= |1 + cos(ω) − j sin(ω)|2
= |1 + cos(ω)|2 + | sin(ω)|2
= 1 + 2 cos(ω) + cos2 (ω) + sin2 (ω)
= 2 + 2 cos(ω).
This result could also have been obtained by noting that when two exponential terms (or one
exponential term and one constant term) have the same magnitude, by factoring out a common
phase factor, a sinusoid can be constructed as follows,
|Xd (ω)|2 = |1 + e−jω |2

= |e−jω/2 (ejω/2 + e−jω/2 )|2
2 2
= e−jω/2 ejω/2 + e−jω/2
2
= ejω/2 + e−jω/2
2
= |2 cos(ω/2)|
= 4 cos2 (ω/2)
= 2 + 2 cos(ω),

3.5
2.5
|Xd(ω)|2
1.5
0.5
0
-1.5 -1 -0.5 0 0.5 1 1.5
ω/π
Figure 2.11: Discrete-time Fourier transform magnitude squared, |Xd (ω)|2 = 2 + 2 cos(ω).
where, in the last line, the trigonometric identity that cos2 (x) = (1 + cos(2x))/2. The magnitude
squared of the discrete-time Fourier transform is shown in Figure (2.11).
Since |Xd (ω)|2 is both periodic with period 2π |Xd (ω)|2 is

and symmetric around the origin,
completely determined by its values on the interval 0 ≤ ω ≤ π. This is similarly true for ∠Xd (ω).
Because of this, when x[n] is real (so that |Xd (ω)| and ∠Xd (ω) have even and odd symmetry,
respectively) we will often plot them on just the interval 0 ≤ ω ≤ π.
To nd ∠Xd (ω) in this example, we write
Xd (ω) = 1 + e−jω
= e−jω/2 (ejω/2 + e−jω/2 )
= e−jω/2 2 cos(ω/2), (2.35)
Now, since cos(ω/2) ≥ 0 for −π < ω < π , (2.35) expresses Xd (ω) in polar form, so that
∠Xd (ω) = −ω/2, −π < ω < π . The phase is plotted in Figure (2.12).
Notice that the phase of Xd (ω)

is an odd function. Also note that for ω > π , the expression
∠Xd (ω) = −ω/2 is not valid, however we simply use our knowledge that the DTFT is periodic
with period 2π. While it is clear from this example that the discrete-time Fourier transform is
periodic in the variable ω with period 2π, the DTFT is, in fact, always periodic with period 2π,
as shown in Section 2.4.1.2.
Example
Consider the sequence x[n] = δ[n − 1] − δ[n + 1]. For this sequence, we will plot the magnitude
|Xd (ω)| and phase ∠Xd (ω). For the magnitude, we have
|Xd (ω)| = e−jω − ejω

= |−2j sin(ω)|
= |2 sin(ω)| ,

Figure 2.12: Example DTFT phase for x[n] = δ[n] + δ[n − 1].
Figure 2.13: Magnitude of the DTFT for x[n] = δ[n − 1] − δ[n + 1].
which can be plotted as shown in Figure (2.13)
Note that |Xd (ω)| is again an even function of ω , and its appearance is completely specied by
on just the interval 0 ≤ ω ≤ π. The phase of Xd (ω) is found by noting
(
−2j sin(ω), {ω : sin(ω) > 0}
Xd (ω) =
2j sin(ω), {ω : sin(ω) < 0}
(
e−jπ/2 2 sin(ω), 0<ω<π
=
ejπ/2 |2 sin(ω)|, −π < ω < 0.
Both the top and bottom lines within the bracket are written in polar form, since sin(ω) > 0 for
0 < ω < π, and | sin(ω)| > 0. Thus,
(
− π2 , 0 < ω < π
∠Xd (ω) = π
2, −π < ω < 0
which can be plotted as in Figure (2.14).
Notice that ∠Xd (ω) is an odd function of the variable ω. This is again due to the symmetry
properties of the DTFT for real-valued x[n]. It is important to recall that these symmetry
properties of and ∠Xd (ω) hold only for real-valued sequences x[n]. If x[n] were not real-valued,
then this symmetry will not be present, as shown in the next example.
Example
Consider the sequence x[n] = δ[n] + jδ[n − 1]. For this sequence, we have the following discrete-
time Fourier transform,

Figure 2.14: DTFT phase example.
Figure 2.15: DTFT magnitude squared for the sequence x[n] = δ[n] + jδ[n − 1].
Xd (ω) = 1 + je−jω ,
which has a corresponding DTFT magnitude squared
2
|Xd (ω)| = (1 + je−jω )(1 − jejω )
= 1 − jejω + je−jω + 1
= 2 + 2 sin(ω).
The DTFT magnitude squared is shown in Figure (2.15).
Here, we see that |Xd (ω)| = |Xd (−ω)| does not hold.
Example
Consider the following sequence x[n] = an u[n], with a real-valued and |a| < 1. For this signal,
we have

Figure 2.16: DTFT Magnitude Squared for x[n] = an u[n].
∞
X
Xd (ω) = an e−jωn
n=−∞
X∞
= (ae−jω )n
n=−∞
1
=
1 − ae−jω
The DTFT magnitude squared is given by
1
|Xd (ω)| =
(1 − ae−jω )(1 − aejω )
1
=
1 + a2 − 2a cos(ω)
which for 0 < a < 1, the magnitude squared would appear as depicted in Figure (2.16).
Example
We next consider a sinusoidal input, x[n] = cos(ω0 n) for all n. As we will see in Chapter 5,
the z-transform for this sequence is undened for all z , including z on the unit circle, where it
coincides with the discrete-time Fourier transform. However, we are willing to extend the notion
of existence of the discrete-time Fourier transform to include such signals with the aid of impulse
distributions. Similar to their continuous-time counterparts, we can dene the DTFT of discrete-
time sinusoidal signals in terms of impulses. As such, we can dene the DTFT of such a signal
to satisfy
∞
X
Xd (ω) = cos(ω0 n)e−jωn
n=−∞
= π(δ(ω − ω0 ) + δ(ω + ω0 )), |ω| < π,
in the sense that the inverse discrete-time Fourier transform of this distribution would yield the
original signal x[n]. Here, Xd (ω) is a distribution, not a function, so the DTFT does not really
exists in the normal sense, and the summation dening the DTFT does not converge in any
meaningful sense to any function. However, if we use this distribution as the operational DTFT
of the sequence, then taking its inverse DTFT would yield

Figure 2.17: Depiction of the impulse distribution for the DTFT of x[n] = cos(ω0 n).
Z π Z π
1 π
Xd (ω)ejωn dω = (δ(ω − ω0 ) + δ(ω + ω0 ))ejωn dω
2π −π 2π −π
1 jω0 n
= (e + e−jω0 n )
2
= cos(ω0 n).
If we graphically depict the distribution Xd (ω), using our notation for representing impulse
distributions, we would obtain the sketch in Figure (2.17).
While an impulse distribution cannot be considered a function, nor can it be considered a proper
limit of a sequence of functions, we could approximate the above gure using tall, narrow rect-
angles around the frequencies of interest, i.e. in place of δ(ω − ω0 ) and δ(ω + ω0 ). This would
correspond to the discrete-time Fourier transform of an approximation to cos(ω0 n). Additionally,
the quality of the approximation improves as the rectangles get narrower and taller. This is
explored further in the next Example.
Example
Consider the nite-length sequence x[n] described below,
(
cos(ω0 n), 0 ≤ n ≤ N − 1
x[n] =
0 otherwise.
The sequence is nite-length, in that it has a nite number, N , of non-zero samples. This sequence
can be constructed as a windowed version of the original innite-length sequence cos(ω0 n). The
DTFT of the sequence can be written,

sin(N ω/2)
Figure 2.18: Periodic sinc function,
sin(ω/2) for N = 25. Note that the rst zero-crossing occurs at ω =
2π/N .
∞
X
Xd (ω) = x[n]e−jωn
n=−∞
N
X −1
= cos(ω0 n)e−jωn
n=0
N −1
X 1 jω0 n
= [e + e−jω0 n ]e−jωn
n=0
2
N −1 N −1
X 1 jω0 n −jωn X 1 −jω0 n −jωn
= e e + e e
n=0
2 n=0
2
1 1 − e−j(ω−ω0 )N 1 1 − e−j(ω+ω0 )N
= −j(ω−ω
+
2 1−e 0 ) 2 1 − e−j(ω+ω0 )
−j(ω−ω0 )N/2
− e−j(ω−ω0 )N/2 1 e−j(ω+ω0 )N/2 ej(ω+ω0 )N/2 − e−j(ω+ω0 )N/2
j(ω−ω0 )N/2

1e e
= +
2 e−j(ω−ω0 )/2 ej(ω−ω0 )/2 − e−j(ω−ω0 )/2 2 e−j(ω+ω0 )/2 ej(ω+ω0 )/2 − e−j(ω+ω0 )/2
1 −j(ω−ω0 )(N −1)/2 sin(N (ω − ω0 )/2) 1 sin(N (ω + ω0 )/2)
= e + e−j(ω+ω0 )(N −1)/2
2 sin((ω − ω0 )/2) 2 sin((ω + ω0 )/2)
| {z } | {z }
periodic sinc centered atω=ω0 periodic sinc centered atω=−ω0
Note that this expression contains two terms; one term corresponding to a ratio of sin expressions,
multiplied by a linear phase term, and another corresponding to a similar ratio of sin expressions
and a similar linear phase term. Each of these terms, corresponds to a periodic sinc function,
centered at the corresponding positive and negative frequencies of the original cosine expression.
The periodic sinc function is simply the DTFT of a length-N sequence of one's, and is depicted
in Figure (2.18).
For large N, the main lobe in the periodic sinc becomes narrow and large in amplitude so that

2.5 Discrete Fourier Transform representation of nite-length DT signals 53
Figure 2.19: The magnitude of the DTFT of the sequence x[n] = cos(ω0 n), for 0 ≤ n ≤ 24 and x[n] =
0,elsewhere.
the two terms in the DTFT of the windowed cosine sequence do not overlap much and we have
that
1 sin(N (ω − ω0 )/2) 1 sin(N (ω + ω0 )/2)

|Xd (ω)| ≈ +
2 sin((ω − ω0 )/2) 2 sin((ω + ω0 )/2)
This relation is expressed in Figure (2.19), again for the case of N = 25.
Now, as N becomes large, this gure begins to resemble, in some sense, the gure containing two
impulses. Similarly, as N becomes large, the windowed (truncated) cosine sequence becomes a
better approximation of the innite-length cosine sequence.
2.5 Discrete Fourier Transform representation of nite-length DT

signals
In Section 2.3 we discussed the discrete Fourier series representation as a means of building a large class of
discrete-time periodic signals from a set of simpler, harminically related discrete-time complex exponential
signals. In this section, we introduce the discrete Fourier transform (DFT) as an analogous notion of
building a large class of nite-length signals from a set of simpler, harminically related nite-length complex
exponential discrete time signals. An important dierence between the discrete-time Fourier series and what
we will develop in this section as the discrete Fourier transform, is that while the periodic signals and the
complex exponential signals used to construct them in the case of the DTFS were dened for all n, the
signals for which we consider a DFT representation are nite in length and are therefore only dened for
a nite interval of the time axis, n. Since this is a subtle dierence, we are able to capitalize on all of the
development of the DTFS. By considering a nite-length signal dened only on the interval 0 ≤ n ≤ N − 1,
as one period of an innite-length periodic signal dened for all n, we can directly map the DTFS into the
DFT for our purposes. Mathematically, if a signal x[n] is dened only on the interval 0 ≤ n ≤ N − 1, then
by considering the periodic signal x̃[n], dened as follows

Section DTFT Property Discrete Time Signal Discrete Time Fourier Transform
P∞
Denition x[n] Xd (ω) = n=−∞ x[n]e−jωn
2.2.1.1 Linearity z[n] = ax[n] + by[n] Zd (ω) = aXd (ω) + bYd (ω)
2.4.1.2 Periodicity x[n] Xd (ω) = Xd (ω + k2π)
2.4.1.3,2.4.2 Real Part Symmetry x[n] real valued <{Xd (ω)} = <{Xd (−ω)}
2.4.1.3,2.4.2 Imaginary Part Symmetry x[n] real valued ={Xd (ω)} = −=Xd (−ω)
2.4.1.3,2.4.2 Magnitude Symmetry x[n] real valued |Xd (ω)| = |Xd (−ω)|
2.4.1.3,2.4.2 Phase Symmetry x[n] real valued ∠Xd (ω) = −∠Xd (−ω)
2.4.1.3,2.4.2 Conjugate Symmetry x[n] real valued Xd (ω) = Xd∗ (−ω)
2.4.2.1 Time Shift y[n] = x[n − d] Yd (ω) = Xd (ω)e−jωd
2.4.2.2 Modulation y[n] = x[n]ejω0 n Yd (ω) = Xd (ω − ω0 )
2.4.2.3 Time-Reversal y[n] = x[−n] Yd (ω) = Xd (−ω)
2.4.2.4 Conjugation y[n] = x∗ [n] YdR(ω) = X ∗ (−ω)
1 π
2.4.2.5 Product of Signals z[n] = x[n]y[n] Zd (ω) = 2π ν=−π
Xd (ν)Yd (ω − ν)dν
P∞
2.4.2.6 Convolution z[n] = m=−∞ x[m]y[n − m] Zd (ω) = Xd (ω)Y
R π d (ω)
P∞ 2 1 2
2.4.2.7 Parseval's Relation x[n] n=−∞ |x[n]| = 2π ω=−π |Xd (ω)| .
Table 2.3: Properties of the Discrete Time Fourier Transform
Discrete Time Signal Discrete Time Fourier Transform

P∞ −jωn
x[n] Xd (ω) = n=−∞ x[n]e
1
an u[n], |a| < 1 1−ae−jω
ae−jω
nan u[n], |a| < 1 (1−ae−jω )2
P∞
ejω0 n k=−∞ 2πδ(ω − ω0 + 2kπ)
P∞
1 k=−∞ 2πδ(ω + 2kπ)
δ[n] P∞ 1
cos(ω0 n) k=−∞ π[δ(ω − ω0 + 2kπ) + δ(ω + ω0 + 2kπ)]
P∞
sin(ω
( 0 n) −jπ k=−∞ [δ(ω( − ω0 + 2kπ) − δ(ω + ω0 + 2kπ)]
sin(ω0 n)
ω0 ω0 n
πn n 6= 0 1, |ω| ≤ ω0
π sinc π = ω0
π n=0 0, ω0 < |ω| ≤ π
( (
sin(N ω/2)
1, 0 ≤ n < N sin(ω/2) ω 6= 0
0, otherwise N ω=0
Table 2.4: Discrete Time Fourier Transform Pairs

∞
X
x̃[n] = x[n + rN ], (2.36)
k=−∞
then over the interval from 0 ≤ n ≤ N − 1, we have that x[n] = x̃[n].

The Fourier series coecients X̃[k] for this periodic signal can be obtained by
N
X −1
X̃[k] = x̃[n]e−j2πkn/N .
n=0
While this expression is valid for all k, we only require one period of X̃[k], i.e. 0 ≤ k ≤ N − 1, for the
inverse DTFS relation to reconstruct x̃[n]. Putting the synethesis and analysis equations together, we have
the discrete-time Fourier series representation of a periodic signal x[n] as
N −1
1 X 2πkn
x̃[n] = X̃[k]ej N ,
N
k=0
which is again valid for all n, since the signal x̃[n] is periodic and dened for all n. However, if we are only
interested in nite-length signals, then viewing them as a single period of an innite-length periodic signal as
in (2.36), we can use the DTFS to both analyze and reconstruct nite-length signals from the rst period of
the underlying innite-length periodic signals x̃[n] and X̃[k]. Specically, we can dene the discrete Fourier
transform of a nite-length signal, dened only over an interval of length N samples as
Discrete Fourier Transform Representation of a Finite-Length Signal

N −1
2πkn
X
X[k] = x[n]e−j N ,0 ≤ k ≤ N − 1, (2.37)
n=0
N −1
1 X 2πkn
x[n] = X[k]ej N ,0 ≤ n ≤ N − 1. (2.38)
N
k=0
2.5.1 Discrete Fourier Transform Properties

We can now explore some of the many useful properties of the DFT representation, noting that these follow
directly from the properties of the DTFS.
2.5.1.1 Sampling Property

P∞
While DFT is related to the DTFS of the periodic signal x̃[n] = k=−∞ x[n + kN ], it can also be shown to
be related to the DTFT of the innite length signal xzp [n] = x[n], 0 ≤ n ≤ N − 1,and xzp [n] = 0, outside
this region. The subscript zp stands for zero-padding, where the innite length signal xzp [n], can be
viewed as padding the nite length signal x[n] with zeros outside the interval 0 ≤ n ≤ N − 1, over which it
is dened. By observing the similarity between the DTFT and DFT representations,
N −1 ∞
2πk
X X
X[k] = x[n]e−j N n = xzp [n]e−jωn = Xd (ω) ,
n=0 n=−∞ ω= 2πk
N ω= 2πk
N
where Xd (ω) here refers to the DTFT of the innite-length sequence xzp [n]. We see that the DFT of the
nite-length signal x[n] can be viewed as a set of N evenly-spaced samples of the DTFT of the zero-padded
innite-length signal xzp [n] taken at samples ωk = 2πk/N, for 0 ≤ k ≤ N − 1. This can be seen pictorially
in Figure (2.20).
Note that the last DFT sample, X[N − 1] does not correspond to a sample taken at ω = 2π, but rather
to the left of 2π, at ω = 2π(N − 1)/N.

Figure 2.20: The DFT as samples of the DTFT, i.e X[k] = Xd ( 2πk
N ).
2.5.1.2 Linearity
The DFT can be viewed as a linear operation as follows
DF T DF T DF T
x[n] ←→ X[k], y[n] ←→ Y [k] =⇒ z[n] = ax[n] + by[n] ←→ aX[k] + bY [k].
This result can be readily shown by substituting z[n] = ax[n] + by[n] into the summation in (2.37) and
expanding the summation into the two separate terms, one for X[k] and one for Y [k].
2.5.1.3 Circular Time Shift

P∞
When a periodic signal x̃[n] = k=−∞ x[n + kN ] is shifted in time, the eect on the DTFS of the resulting
−j2πkn0 /N
periodic signal x̃[n − n0 ] can be viewed as applying a linear phase term e to the original DTFS
coecients. For a nite-length signal x[n], we cannot dene a time-shift in the same manner, by simply
shifting the time index, as this would require evaluating the signal x[n] outside the range over which it is
dened, namely 0 ≤ n ≤ N − 1. However, we can use the time shift property of the DTFS to relate a
slightly dierent notion for nite length signals. By considering a time shift of n0 samples of the underlying
innite-length periodic signal x̃[n − n0 ] as a circular shift of the nite-length sequence x[n], we can study
apply the time-shifting property of the DTFS to nite-length signals. To this end, we dene a circular shift
of a nite-length sequence, x[n] as follows. First, we dene the modulo operator, k N = k mod N, as,
k N = r, where k = `N + r,
for any integer `. For example 4 7 = 7, 7 4 = 3, 4 4 = 0, −5 4 = 3, −2 4 = 2. Note that
it is sometimes helpful to use −k N = N − k N . Now we can dene the nite-length signal y[n] as
a circular shift of the nite-length sequence x[n] as
y[n] = x̃[n − n0 ] = x[ n − n0 N ],

i.e. a circular shift of the nite-length signal x[n] by n0 samples, is written x[ n − n0 N ] but can be
viewed as the result of taking a single period of the periodic signal x̃[n − n0 ], 0 ≤ n ≤ N − 1.
over the range
The corresponding eect on the DFT of the sequence can be compactly represented as follows
DF T DF T 2πk
x[n] ←→ X[k] =⇒ y[n] = x[ n − n0 N ] ←→ X[k]e−j N n0 .
We see that a circular shift in time of a nite length signal corresponds to a modulation in frequency by
a phase term that is linear with frequency with a slope that is proportional to the delay. Note that the
resulting DFT is exactly the same as that which we would have obtained by rst periodically extending the
sequence x[n] to the signal x̃[n], and taking the DTFS representation of the resulting time-shifted signal
x̃[n − n0 ].


From the analogous property of the DTFS, we can obtain the relation
DF T DF T
x[n] ←→ X[k] =⇒ y[n] = x[n]ejk0 ω0 n ←→ X[ k − k0 N ],
2π
where ω0 = N , and where we have used the modulo notation to enable the resulting DFT to remain
nite-length.

From the DTFS time-reversal property, we obtain,
DF T DF T
x[n] ←→ X[k] =⇒ y[n] = x[ −n N ] = x[ N − n N ] ←→ X[ −k N ] = X[ N − k N ],
where the modulo operator in N − n N and N − k N only comes into use for n = 0 and k = 0
since for all other values, the terms within the modulo operator are within the range of 0, ...N − 1. This
corresponds in the DFT representation to changing the sign of the DFT frequency index, where, to keep the
terms within the range from 0 to N, we add N to the index, and take the result modulo N, which results in
a reversal of the order of the DFT coecients X[ N − k N ] = X[N − k N ].

The eect of conjugating a complex-valued signal on its DFT representation can be seen from the DTFS
property as
DF T DF T
x[n] ←→ X[k] =⇒ x∗ [n] ←→ X ∗ [ N − k N ].
DFT exhibits a symmetry property. This arises directly from the denintion of the DFT, and that real
DF T
x[n] = x∗ [n] ←→ X[k] =⇒ X[k] = X ∗ [ N − k N ].

When two nite-length signals of the same length are multiplied in time, such that z[n] = x[n]y[n], the
resulting signal remains nite length with the same length by denintion. Hence, each of the three signals
admit DFT representations using the same set of harmonically related signals. We can observe the eect on
the resulting DFT from the analogous DTFS property,
N −1
DF T DF T DF T 1 X
x[n] ←→ X[k], y[n] ←→ Y [k] =⇒ z[n] = x[n]y[n] ←→ X[m]Y [ k − m N ].
N m=0
2.5.1.8 Circular Convolution

A dual relationship to that of multiplication in time, is multiplication of DFT coecients. Specically, when
the two signals x[n] and y[n] are each of nite length N, the nite length signal z[n] of length N, whose DFT
representation is given by Z[k] = X[k]Y [k] corresponds to a circular convolution of the signals x[n] and y[n].
This leads to the following property of the DFT,
N −1
DF T DF T DF T
X
x[n] ←→ X[k], y[n] ←→ Y [k] =⇒ z[n] = x[m]y[ n − m N ] ←→ Z[k] = X[k]Y [k].
m=0

Length-N Discrete Time Signal Discrete Fourier Transform

PN −1
x[n], 0 ≤ n ≤ N − 1 X[k] = n=0 x[n]e−j2πkn/N , 0 ≤ k ≤ N − 1
ax[n] + by[n] aX[k] + bY [k]
x[ n − d N ] X[k]e−j2πkd/N
x[n]ej2π`n/N X[ k − ` N ]
x[ −n N ] X[ N − k N ]
x∗ [n] X ∗ [ N − k N ]
1
P N −1
PN −1
x[n]y[n] N m=0 X[m]Y [ k − m N ]
m=0 x[m]y[ n − m N ] X[k]Y [k]
PN −1 2 1
PN −1 2
Parseval's relation: n=0 |x[n]| = N k=0 |X[k]| .
x[n] real-valued X[k] = X ∗ [ N − k N ]
x[n] real-valued <{X[k]} = <{X[ N − k N ]}
x[n] real-valued ={X[k]} = −={X[ N − k N ]}
x[n] real-valued |X[k]| = |X[ N − k N ]|
x[n] real-valued ∠X[k] = −∠X[ N − k N ]
Table 2.5: Discrete Fourier Transform Properties

The energy containted within the nite duration of a nite length signal can also be computed in terms of
its DFT representation using Parseval's relation,
N −1 N −1
DF T
X 1 X
x[n] ←→ X[k] =⇒ |x[n]|2 = |X[k]|2 .
n=0
N
k=0
The properties of the DFT discussed in this section are summarized in Table (2.5).
2.6 Applications of spectral representations and signal analysis,

DFT spectral analysis
In Section2.5 the DFT was introduced as a method for representing nite-length signals using a linear
combination of simpler, nite-length segments of complex exponential signals, in a manner analogous to
the construction of periodic signals from the periodic extentions of the same complex exponential family of
signals. In Chapter 8 we will explore a number of fast algorithms for explicitly computing the DFT. One of
the primary drivers of tremendous growth in the application and use of digital signal processing was, in fact,
this discovery. There are two important aspects of the DFT that make its use in digital signal processing
systems so pervasive. First, because the DFT is represented by a nite-length sum over quantities that are
available, namely the signal x[n],whose DFT is desired, and complex numbers of the form ej2πkn/N ,which can
be easily represented using a pair of real numbers in the digital signal processor, one of the real part and one
for the imaginary part, the DFT can be exactly computed. This is in contrast to the CTFT and the DTFT
which contain either innite-length integrals or innite-length summations that cannot be exactly computed.
Second, and perhaps more importantly, as we will see in Chapter 8, the DFT can be computed so eciently
that we can make use of many of the DFT properties developed in this and later chapters for performing a
wide variety of operations on the signals of interest to our application. For example, we will see how linear
ltering can be very eciently accomplished in the frequency domain through the multiplication of the
DFT representation of signals. We will also explore how this can be accomplished for innite-length signals
by stitching together the outputs generated through the use of the nite-length representation generated
through the DFT.
Another equally desireable application that can be well-handled using the DFT is the process of spectral
analysis. In many applications, including radio communications, image restoration, video compression, even

2.6 Applications of spectral representations and signal analysis, DFT spectral analysis 59
N −1 −1
xa (t) {x[n]}n=0 {X[k]}N
- C/D - - DFT
- k=0
Figure 2.21: System for computing approximate samples of the CTFT of the analog signal xa (t).
analog circuit analysis, it is important to have a direct measure of the frequency content of signals of interest.
This can be accomplished in the laboratory using equipment known as a spectrum analyzer, in which the
signal of interest is decomposed into its Fourier components, in real-time, and an image of the frequency
composition of the signal is displayed on a screen or an oscilloscope. Such measures are extremely useful
and require real-time processing of analog signals in order to privide an accurate picture of the frequency
composition of a signal as either the signal or its environment is changing. While for analog signals, it is
the continuous-time Fourier transform that provides information about the frequency composition of the
signal, as we will explore in more detail in Chapter 4, many of the frequency-domain properties of signals
and systems can accurately be measured by discrete-time processing of sampled versions of analog signals.
To make the process more concrete, suppose that we are interested in the frequency content of an analog
signal xa (t), that is known to be approximately bandelimited to B Hz. That is, suppose that we know that
CT F T
xa (t) ←→ Xa (Ω) ≈ 0, |Ω| > 2πB
i.e., that the CTFT of the signal xa (t), given by Xa (Ω) is known to be approximately zero outside of the
frequency range |Ω| < 2πB. However, we don't know the precise frequency composition of xa (t), and would
like to obtain a real-time measurement of Xa (Ω) based on processing xa (t). One way to accomplish this is
using DFT. Recall that the DFT of the nite-length signal x[n] enables us to compute samples of the DTFT
of the innite-length zero-padded signal xzf [n]. We also know from Chapter 3and will explore in greater
detail in Chapter 4, how the CTFT of an analog signal xa (t) is related to the DTFT of a sampled-version of
that signal, x[n] = xa (nT ). We will now explore how these notions can be put together to provide a means
for constructing a real-time spetrum analyzer using the DFT.
We will approach this problem as follows. Given access to the analog signal xa (t),we want to compute
approximate samples of Xa (Ω). Here is the proposed method for accomplishing this task. We will assume
that the signal of interest xa (t) has nite support on the interval [0, (N − 1)T ] and is nearly bandlimited
as described above. That is, we are interested in the signal in the range t ∈ [0, (N − 1)T ], where T is a
parameter of our sampling process that we will be able to control. The signal xa (t) to be processed by the
processor depicted in Figure 2.21to yeild our samples,
is sampled at a rate of one sample every T seconds producing the discrete-time signals x[n] = xa (nT ).
N −1
Given the duration of interest, we obtain N samples, which we compactly represent as x[n] n=0 . These
N −1
samples are then processed with the DFT analysis equation (2.37) producing the values X[k] k=0 . In the
frequency domain we know that the following relationship between the continuous-time Fourier transform
(CTFT) of xa (t) and the discrete-time Fourier transform (DTFT) of x[n] holds,
∞
1 X ω + 2kπ
Xd (ω) = Xa .
T T
k=−∞
That is, if xa (t) is nearly bandlimited and T is small enough, there is little detrimental aliasing so that then
Xd (ω) will contain copies of Xa (Ω) scaled to t within the digital frequency range |ω| < π and scaled in
amplitude by a factor of 1/T. This is depicted in Figure (2.22).
−1
The DFT, {X[k]}N
k=0 is a set of samples of Xd (ω) on the interval [0, 2π] such that X[k] = Xd ( 2πk
N ),
0 ≤ k ≤ N − 1. Thus, we have (for N odd):

Xd(ω)
k = 0 term k = –1 term
 ω  ω – 2π 
 T
1
= Xa 
T 
1
= Xa
T T
ω
–π π 2π
Figure 2.22: Spectrum of xa (t) scaled in P

amplitude and frequency and replicated on the digital frequency
∞
Xd (ω) = T1 k=−∞ Xa ω+2kπ

axis according to the relation
T .
N −1
(1
Xa 2πk

T NT , 0≤k≤ 2
X[k] ≈ 1 2π(k−N )

N −1
T Xa NT , 2 < k ≤ N − 1.
−1
So, we have that {X[k]}N
k=0 actually contain approximate, scaled samples of Xa (Ω). It is important to note
that the rst half of the DFT samples provide samples of Xa (Ω) for Ω > 0, while the second half of the
DFT gives samples of Xa (Ω) for Ω < 0. This peculiarity arises due to the denition of the DFT, whereby
−1
{X[k]}N
k=0 is a set of samples of Xd (ω) on the interval [0, 2π] rather than on [−π, π].
2.6.1 DFT Spectral Analysis of Sinusoids
Suppose that we wish to determine the frequency content of an analog signal that containes a number of
sinusoidal components, of the form
M
X
xa (t) = Ai cos(Ωi t),
i=1
and we have available only the following samples of the signal, x[n] = xa (nT ), for 0 ≤ n ≤ N − 1. From this
N −1
set of samples, x[n]
n=0
, we wish to determine as much information about xa (t) as possible. Namely, we
M M
would like to determine the values of M, Ω and Ai .
i=1 i=1
One approach to this problem is to use DFT spectral analysis as depicted in Figure 2.21. Variations on this
type of problem arise in numerous applications, including such wide-ranging topics as digital communications,
radio astronomy, and, in particular, applications involving rotating machinery. For example, an acoustic
transducer coupled to a piece of rotating machinery will output a periodic signal (sum of sinusoids) plus,
perhaps, a smaller nonperiodic component. A frequency (i.e., DFT) analysis of this signal can indicate
whether the machinery requires maintenance or replacement. Similarly, in an underwater setting, ships can
be identied through DFT analysis of the acoustic signals they emit, which are collected by hydrophones.
Consider a single sinusoidal component, captured within x[n] as
x[n] = A cos(Ω0 nT ),
for 0 ≤ n ≤ N − 1.The DFT of x[n] is the set of samples of the DTFT:


AN
2
ω
–Ω0T π
Ω0T
2π 2π
Ω0T – Ω 0T +
N N
Figure 2.23: Schematic depiction of two periodic sinc functions at ω = ±Ω0 T.
N
X −1
X (ω) = A cos(Ω0 nT )e−jωn ,
n=0
N −1
X A −j(ω−Ω0 T )n
= e + e−j(ω+Ω0 T )n
n=0
2
A N A N

−j(ω−Ω0 T ) N 2−1 2 sin (ω − Ω0 T ) 2 −j(ω+Ω T ) N −1
2 sin (ω + Ω 0 T ) 2
= e +e 0 2 .
sin (ω − Ω0 T ) 12 sin (ω + Ω0 T ) 21
| {z } | {z }
T1 (ω) T2 (ω)
The two periodic sincs T1 (ω) and T2 (ω) above have peaks at ω = ±Ω0 T and give rise to a DTFT that
appears as in Figure 2.23.
4π
The width of each main lobe is
N based on the zero crossings of the periodic sinc function. If N is large
enough, the main contributions from these pulses do not substantially overlap in frequency much and we
obtain that
|Xd (ω)| = |T1 (ω) + T2 (ω)|

≈ |T1 (ω)| + |T2 (ω)|
A sin (ω − Ω0 T ) N2 A sin (ω + Ω0 T ) N2

= + .
2 sin (ω − Ω0 T ) 12 2 sin (ω + Ω0 T ) 12
−1
Therefore, {X[k]}N
k=0 will provide approximate samples of the magnitude of the above plot,

2πk
|X[k]| = Xd
N
A sin ( 2πk N
A sin ( 2πk N

N − Ω0 T ) 2 N + Ω0 T ) 2
≈ + ,
2 sin ( 2πk 1 2 sin ( 2πk 1

N − Ω0 T ) 2 N + Ω0 T ) 2
from which we can estimate Ω0 and A. If x[n] is obtained from samples of a signal containing a sum of
several sinusoids, we will have multiple peaks and we can estimate all of the parameters M, Ωi , and Ai using
one of a number of possible methods, depending on what other information we have about the environment
in which the signals have been obtained. For example, we might use the following algorithm:
M = Number of peaks detected on the interval [0, π]

Ωi T = Location of the i − th peak
N Ai
= Amplitude of the i − th peak
2
Ω0T Ω1T ω
2π
Ω0T + Ω1T – 2π
N N
Figure 2.24: Two closely-spaced peaks in the spectral analysis of a signal containing sinusoidal components.
ω
Figure 2.25: Two closely spaced peaks that overlap by 50% of their maximum amplitude.
Actually, sinusoidal spectral analysis, via the DFT, is not usually this straightforward. Some of the problems
that frequently arise include:
1. Two Ωi are so close together that the two peaks add together into a single peak, or do not easly resolve
into two distinct peaks. One possible solution is to use a larger value of N, i.e., collect data over a
longer observation interval so that peaks will be narrower and will not overlap. Suppose that the signal
contains the two frequencies Ω0 and Ω1 . Then the main lobes of the periodic sincs may look like that
shown in Figure2.24. To clearly distinguish the two peaks, we might require that the main lobes not
overlap, i.e.,
2π 2π
Ω1 T − < Ω0 T +
N N
4π
(Ω1 − Ω0 ) T >
N
4π
NT > .
(Ω1 − Ω0 )
Therefore, the necessary observation interval NT increases as the frequency separation (Ω1 − Ω0 )
decreases. So, the closer two sinusoidal components are to one another, the longer you will need to
wait, in order to collect sucient data to resolve the two peaks. In practice the above condition on
NT is often too conservative, since we can still discern two peaks even if there is some overlap. If we
instead require that there be no more than 50% overlap as shown in Figure 2.25, then the required
condition becomes m
2π
Ω1 T − > Ω0 T
N
2π
NT > .
(Ω1 − Ω0 )
2. In the previous analysis, we ignored the eect of the amplitude of the two frequency components,
A0 and A1 . If two of the frequencies, Ωi are fairly close together and one of the Ai is much smaller than
the other, this may cause the small peak to be buried in the sidelobes of the larger peak. One possible
solution to this problem is to consider applying a window to the data prior to computing the DFT,
to reduce the resulting sidelobes. Since the DTFT of the signal x[n] = w[n]xa (nT ) implicitly contains
the window function w[n] multiplying the signal xa (nT ), then the eect of the wundown w[n] on the

resulting DTFT, and therefore the resulting DFT, must be considered. In the analysis so far, we have
assumed that a rectangular window, w[n] = 1,for 0 ≤ n ≤ N − 1,was used. By selecting a window w[n]
that has lower side-lobe behavior than the rectangular window, which has a periodic sinc function as
its DTFT, we may be able to aleviate some of the side lobe issues. This is a matter that is discussed in
more detail in chapter 6.Unfortunately, however, this is not a cure-all. While windowing reduces the
size of the resulting side lobes near the peak, it also has the eect of widening the main lobe so that
closely spaced sinusoids will be harder to distinguish.
3. The most important diculty in spectral analysis is the presence of noise in the signal aquisition
environment and in the signal itself. There are a host of methods that could have been used for
sinusoidal spectral analysis that would work perfectly (i.e. perfect estimation of all parameters) if
there were no noise. Properly handling the spectral estimation problem in the presence of noise is
a topic of great interest in the digital spectral analysis eld and a number of methods exist, whose
particular implementation and ecacy depends strongly on the type of noise that is present and on
the signal to noise ratio, i.e. the relative strength of the signal components Ai and the noise power in
the frequency range of interest. The achievable resolution in spectral analysis ultimately depends on
properties of the noise, and on N T, the length of observation interval of xa (t).
2.6.2 Zero-Padding in DFT Spectral Analysis

Consider a spectral estimation problem as described in the previous section, in which we have obtained a
−1
set of samples {x[n]}N
n=0 , from an analog signal, such that x[n] = xa (nT ),for 0 ≤ n ≤ N − 1. Let xzp [n] be
the innite length signal obtained by letting xzp [n] = x[n], for 0 ≤ n ≤ N − 1, and set to zero outside of this
interval. We then have that the DFT of x[n]is related to the DTFT of xzp [n] through

2πk
X[k] = Xd , 0 ≤ k ≤ N − 1.
N
We now consider another nite-length signal x2 [n], of length N + M, such that
(
x[n], 0 ≤ n ≤ N − 1
x2 [n] =
0 N ≤ n ≤ N + M − 1,
so that the resulting nite-length signal contains both the original set of samples x[n] = xa (nT ), as well as
the additional M zero values from xzp [n] that follow these samples in the signal xzp [n]. Let us consider how
the length N + M DFT of the signal x2 [n] relates to the DTFT of the signalxzp [n]. From the denition of
the DFT, we have
X−1
N +M
2πk
X2 [k] = x2 [n]e−j N +M n
n=0
N −1
X 2πk
= xzp [n]e−j N +M n
n=0
N
X −1
= xzp [n]e−jωn
n=0 ω= N2πk
+M

2πk
= Xd ,
N +M
i.e., X2 [k] provides a set of N +M samples of the DTFT of xzp [n] equally spaced over the interval ω ∈ [0, 2π].
That is, zero padding of the signal prior to taking the DFT provides a more densely spaced set of samples
of precisely the same DTFT as the original, non-zero padded length N signal, x[n]. One reason that zero-
padding is often used in DFT spectral analysis is the desire for a more densely spaced set of samples of
Xd (ω), for example, to make a visual plot of the samples as an approximation to the true Xd (ω), which
is a function of the continuous variable, ω. A second, and often more compelling reason for zero padding,

Xd (ω)
ω
Ω0T π 2π–Ω0T 2π
2π 2π
Ω0T – Ω0T +
N N
Figure 2.26: DTFT Magnitude for the sequence x[n] = cos(Ω0 nT ).
arises when the sequence length, N, is not a length for which a fast DFT implementation is available. For
example, we will explore a number of ecient implementations of the DFT, known collectively as the fast
Fourier transform, or FFT, algorithms in chapter 8, for which N must be a power of two, i.e., N = 2ν .
As a result, it may be desireable to pad the length of the sequence out to the nearest length for which an
FFT implementation is available. This is a consideration of particular importance if N is large, or if many
length-N DFTs are needed.
Example: DFT Spectral Analysis of Sinusoids

N −1
If xa (t) = cos(Ω0 t), and we have available the set of samples, {x[n]}n=0 , where x[n] = cos(Ω0 nT ),
then the magnitude of the
DTFT of this sequence looks as shown in Figure 2.26.

N −1
Computing the DFT of the sequence {x[n]}n=0 provides samples of Xd (ω). Depending on the
length, N, of the DFT and the value of Ω0 T , there may or may not be samples of the DFT at or
near the peaks of the mainlobes and sidelobes of the DTFT. Zero-padding prior to the DFT will
yield a denser collection of samples of Xd (ω) and will therefore provide a better representation
of Xd (ω), especially when plotted graphically.
N −1
The use of one of a variety of suitably designed windows, {w[n]}n=0 , can aid in DFT spec-
tral analysis when the window is applied to the data prior to taking the DFT, i.e. Y [k] =
PN −1 −j2πkn/N N −1
n=0 w[n]x[n]e . Note the sequence {x[n]} n=0 , already implicitly contains a rectangu-
lar window, w[n] = 1, 0 ≤ n ≤ N − 1. A dierent window could be applied by simply computing
PN −1
Y [k] = n=0 w[n]x[n]e−j2πkn/N , for a suitable window sequence, such as Hamming window,
prior to computing the DFT. This will widen main lobe (by a factor of two) but will also greatly
reduce the sidelobe behavior.
Short Time Fourier Transform (or Windowed Fourier Transform)

In practice, for long signals, such as speech or radio transmissions, it is informative to examine
the frequency content of the signal changes over time. To enable this, we can parse the signal
into shorter blocks of time, which are often overlapped with one-another, and compute the DFT
of each of these blocks. This process is often termed the short-time Fourier transform (STFT),
and a graphical display of the magnitude of the STFT is called the spectrogram of the signal.
Figure 2.27illustrates the process in computing the STFT.
When the signal of interest is a speech signal, as shown in Figure2.28, the STFT of the signal is
often shown as an image, as it appears in the gure, right.

Figure 2.27: Graphical depiction of the process of computing the short-time Fourier transform of a long
signal by applying a sequence of overlapping windows. These windowed segments of the longer signal
are then processed using the fast Fourier transform (FFT) algorithm for computing the DFT, whose log
magnitude is then shown as a function of time.
Figure 2.28: A segment of a signal generated from samples of a speech signal and its spectrogram computed
using the code shown in table 2.6.

% LOAD DATA
load mtlb;
x = mtlb; % x holds the speech signal
figure(1), clf
plot(0:4000,x)
xlabel('n')
ylabel('x(n)')
% SET PARAMETERS
R = 256; % R: block length
window = hamming(R); % window function of length R
N = 512; % N: frequency discretization
L = 35; % L: time lapse between blocks
fs = 7418; % fs: sampling frequency
overlap = R - L;
% COMPUTE SPECTROGRAM
[B,f,t] = specgram(x,N,fs,window,overlap);
% MAKE PLOT
figure(2), clf
imagesc(t,f,log10(abs(B)));
colormap('jet')
axis xy
xlabel('time')
ylabel('frequency')
title('SPECTROGRAM, R = 256')
The key command
[B,f,t] = specgram(x,N,fs,window,overlap);
computes the short time Fourier transform of a signal using a sliding window. The description of
its input and output are as follows.
Input:
x - a vector that holds the input signal
N - species the FFT (or DFT) length that specgram uses
fs - a scalar that species the sampling frequency
window - species a windowing function
overlap - number of samples by which the sections overlap
Output:
B - matrix contains the spectrogram which is the magnitude of the STFT of the input signal
f - a column vector contains the frequencies at which specgram computes the discrete-time Fourier
transform. The length of f is equal to the number of rows of B.
t - a column vector of scaled times, with length equal to the number of columns of B. t(j) is the
earliest time at which the j-th window intersects x. t(1) is always equal to 0.
Table 2.6: M-le code used to generate the spectrogram shown in Figure 2.28.

Chapter 2

Uploaded by

Copyright:

Available Formats

Chapter 2

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Chapter 2

Uploaded by

Copyright:

Available Formats

Chapter 2

CT and DT Signal Representations

• 3 Discrete Fourier Series

• 4 Discrete-Time Fourier Transform

• 5 Discrete Fourier Tranform

2. Familiarization with Fourier Series and transform representations

c A.C Singer and D.C. Munson, Jr. January 23, 2011

2.1 Fourier Series representation of nite-length and periodic CT

x(t) = x(t + T ), for all t. (2.1)

x(t) = sin(ω0 t + φ), (2.2)

which, for k = 1, yields the relationship

c A.C Singer and D.C. Munson, Jr. January 23, 2011

Figure 2.1: The periodic sinusoidal signal x(t) = sin((4π/3)t).

s(t) =x(t) + y(t)

x(t) =ej(ω0 t+φ) (2.5)

c A.C Singer and D.C. Munson, Jr. January 23, 2011

c A.C Singer and D.C. Munson, Jr. January 23, 2011

To proceed, we need to evaluate the integral

CT Fourier Series Representation of a Periodic Signal

Example: CTFS of a Square Wave

Using (2.10), we obtain,

c A.C Singer and D.C. Munson, Jr. January 23, 2011

2.1.1 CT Fourier Series Properties

2.1.1.2 Time Shift

c A.C Singer and D.C. Munson, Jr. January 23, 2011

2.1.1.3 Frequency Shift

c A.C Singer and D.C. Munson, Jr. January 23, 2011

2.1.1.4 Time Reversal

More generally, from the CTFS synthesis equation,

yielding the relation

2.1.1.5 Time Scaling

c A.C Singer and D.C. Munson, Jr. January 23, 2011

2.1.1.6 Conjugate Symmetry

2.1.1.7 Products of Signals

c A.C Singer and D.C. Munson, Jr. January 23, 2011

c A.C Singer and D.C. Munson, Jr. January 23, 2011

This yields the property,

from which we obtain the relation

c A.C Singer and D.C. Munson, Jr. January 23, 2011

2.1.1.11 Parseval's relation

2.2 Fourier transform representation of CT signals

c A.C Singer and D.C. Munson, Jr. January 23, 2011

c A.C Singer and D.C. Munson, Jr. January 23, 2011

CTFS representation for T = 4 CTFS representation for T = 8

CTFS representation for T = 16 CTFS representation for T = 32

c A.C Singer and D.C. Munson, Jr. January 23, 2011

CT Fourier Transform Representation of Aperiodic Signals

2.2.1 CT Fourier Transform Properties

2.2.1.2 Time Shift

2.2.1.3 Frequency Shift

c A.C Singer and D.C. Munson, Jr. January 23, 2011

which leads to the relation

2.2.1.4 Time Reversal

2.2.1.5 Time Scaling

c A.C Singer and D.C. Munson, Jr. January 23, 2011

2.2.1.6 Conjugate Symmetry

2.2.1.7 Products of Signals

c A.C Singer and D.C. Munson, Jr. January 23, 2011

from which we obtain the relation

2.1 Fourier Series representation of nite-length and periodic CT