THE DISCRETE-TIME FOURIER TRANSFORM

1. Lecture 4 44 minutes

4.1

Sam ler
XA(t)

C/D

x(n) = xA(nT)

Relationship in the time domain for a con tinous-time signal, its samples, and the re sulting sequence

T=Tj

t
T=T 2 =2T,
t
0 t

012345

012345

T=T

T=T 2 =2T

oh
0 a

XA(j CL) 0 TXA(j n) Ti TX(ejw)

7W

Relationship among the Fourier transforms of a continous-time signal, its samples, and the resulting se quence.

1 C0 ~
T, e.
e.

0
-T
0

0
T 2 h-CLT
V71

T2
2 b
V

2. Comments
In the previous lecture we developed the notion of the frequency
response of LSI systems and discussed the determination of the
frequency response from the unit sample response. We begin this
lecture by developing an inverse relation by interpreting the
expression for the frequency response as a Fourier series expansion,
the coefficients for which are the unit sample response values. We then
consider the generalization of the frequency response representation of
LSI systems to the Fourier transform representation of sequences.
Throughout this set of lectures we will exploit various Fourier
transform properties. Many of these properties are derived in a similar
manner and it is useful to understand the style in which they are

4.2

derived rather than simply memorizing the properties. In the lecture

the convolution property and some symmetry properties were discussed.
A number of other important properties are presented in the text and
should be carefully reviewed.
The lecture concludes with a discussion of the relationships between
continuous-time and discrete-time Fourier transforms. In particular
you should be aware from your background in continuous-time linear
system theory of the form of the Fourier transform of a sampled time
function. Converting a sampled time function to a sequence introduces
in essence a "time" normalization since the spacing of sequence values
is always interpreted to be unity. Equivalently this time normalization
corresponds in the Fourier domain to a frequency normalization.
3. Reading
Read sections 2.7, 2.8, 2.9 and

Text: Review section 2.6 (page 39). sections 3.1 through 3.5.
4. Problems

Problem 4.1
Determine the Fourier transform of each of the sequences below:
(a)
x(n) = 6(n
-

3)

1 1

Figure P4.1-1
(b)=11 x(n) =

6(n + 1) + 6(n) +

6(n

1)

0
Figure P4.1-2 (c) x(n)
=

n
a u(n) 0<a<l

0 Figure P4.1-3

4.3

(d)

x(n)

= u(n +

3)

u(n

4)

-3 Figure P4.1-4
Problem 4.2

-2

-1

Consider a linear shift-invariant system with unit-sample response = n h(n) = a u(n), where a is real and 0 < a < 1. If the input is x(n) = y(n) = sum. (b)

(a)

Sn

u(n),

0 <

|6|

< 1, determine the output y(n) in the form

(k1 an + k2 n) u(n) by explicitly evaluating the convolution H(e 3W) and Y(eJ")
show that

By explicitly evaluating the transforms X(e ]W),

corresponding to x(n),
Y(ejW)

h(n), and y(n) specified in part (a),

= H(ejW)

X(e j)

Problem 4.3
Let x(n) and X(e j) represent a sequence and its transform. Do not

assume that x(n) is real or that x(n) is zero for n < 0. terms of X(ejW) the transform of each of the following: (a) (b) (c) k x(n)
x(n - n ) n x(n)
where n 0 is an integer

Determine in

Problem 4.4
Let ha(t) denote the impulse response of a linear time-invariant
continuous-time filter and hd (n) the unit-sample response of a
linear shift-invariant discrete-time filter. h(t) = a
(a) ae-at 0 t > 0 t < 0
a > 0 ha (t) is given by

Determine the analog filter frequency response and sketch its

magnitude.
If h (n) = c h (nT), determine the digital filter frequency a d response and determine the value of c so that the digital filter has unity gain at w = 0. Sketch the magnitude of Hd (ejW) (b)

4.4

Problem 4.5
One context in which digital filters are frequently used is in filtering
bandlimited analog data as shown below:
digital filter
A(t)

SamplerXA(t) T

C/D

x (n)

H(e

3W

y (n)

D/C

yA (t)

LPF

yA(t)

Figure P4.5-1
Assume that the sampling period T is sufficiently small to prevent
aliasing. The system labelled D/C is a discrete to continuous time
converter, i.e. the inverse of a continuous to discrete time converter.
The system labelled LPF is an ideal lowpass filter with a gain of T in
the passband and cutoff frequency of rad/sec. The overall system is
equivalent to a continous-time filter. Indicated below are two choices
for T and the digital filter frequency response H(eJW). For each of
these, sketch the frequency response of the overall continuous-time system.

(a) 1r

H(e )W
1-=10 khz
T

Figure P4.5-2
(b)
1

H(ej)
T = 20 khz

7T

Tr

iT

4 Figure P4.5-3

Problem 4.6*
In Sec. 1.6 of the text, a number of symmetry properties of the Fourier
transform are stated. All these properties follow in a relatively straight
forward way from the transform pair. Following is a list of some of the

4.5

properties stated.

Prove that each is true.

In carrying out the proof

For

you may use the definition of the transform pair as given in Eqs. (1.19)
and (1.20) of the text and any previous property in the list. example, in proving property 3 you may use properties 1 and 2.
Sequence 1. 2. 3. 4. 5. x*(n) x*(-n) Re x(n) Im [x(n)] x (n) x 0 (n)
*

Fourier Transform
X*(e~ j) X*(e )W
X (e ") e
X0 (eW Re[X(eJW)]

6.

j Im [X(eJ))

Problem 4.7

f(n) and g(n) are causal and stable sequences with Fourier
transforms F(ejW) and G(ejW), 1f
7

respectively. F(ej) dw

Show that
dw

r F(e

) G(e

) dw =1 J

j!1G(e)

Problem 4.8
In the design of either analog or digital filters, we often approximate
a specified magnitude characteristic, without particular regard to the
phase. For example, standard design techniques for lowpass and bandpass
filters are derived from consideration of the magnitude characteristics
only.
In many filtering problems, one would ideally like the phase characteris tics to be zero or linear. For causal filters it is impossible to have zero phase. However, for many digital filtering applications, it is not necessary that the unit-sample response of the filter be zero for n < 0 if the processing is not to be carried out in real time. One technique commonly used in digital filtering when the data to be
filtered are of finite duration and stored, for example, on a disc or
magnetic tape, is to process the data forward and then backward through
the same filter.
Let h(n) be the unit-sample response of a causal filter with an arbitrary phase characteristic. Assume that h(n) is real and denote its Fourier transform by H(ej ). Let x(n) be the data we want to filter. The filtering operation is performed as follows.

4.6

(a) Method A:
x(n)
1.
,

g(n)
h(n)

g(-n)
11. , h(n)

r(n)

iii.

s(n) = r(-n)

Figure P4.8-1 Determine the overall unit-sample response h1 (n) that relates
x(n) and s(n), and show that it has a zero-phase characteristic.
(2) Determine (1)

|H1(e3W)|

and express it in terms of

H(e ])

and

arg[H(e W)I. (b) Method B: Process x(n) through the filter h(n) to get g(n). Also

process x(n) backward through h(n) to get r(n). then taken as the sum g(n) plus r(-n).

The output y(n) is

x(n)
i. h(n)

g(n)

..

x(-n)

h(n)

r(n)

iii

y(n)

= g(n)

+ r(-n)

Figure P4.8-2
This composite set of operations can be represented by a filter, with
input x(n), output y(n), and unit-sample response h 2 (n).
(1) (2) Show that the composite filter h2 (n) has a zero-phase characteristic. Determine

|H (e3)W| 2

and express it in terms of IH(eJw)j and

arg [H(ejW ). c. Suppose that we are given a sequence of finite duration, on which

we would like to perform a bandpass, zero-phase filtering operation.

Furthermore, assume that we are given the bandpass filter h(n), with
frequency response as specified in Figure P4.8-3, which has the
magnitude characteristic that we desire but linear phase. To achieve
zero phase, we could use either method (A) or (B). sketch|H 1(e3W)Iand |H2 (elw)|. Determine and

From these results which method would you

4.7

use to achieve the desired bandpass filtering operation? Explain why.

More generally, if h(n) has the desired magnitude but a nonlinear phase
characteristic which method is preferable to achieve a zero phase
characteristic?

|H(eW ) I

l ... ---

Tr

3Tr

7T

arg [H(e W)]

Figure P4.8-3

4.8

MIT OpenCourseWare http://ocw.mit.edu

Resource: Digital Signal Processing

Prof. Alan V. Oppenheim

The following may not correspond to a particular course on MIT OpenCourseWare, but has been provided by the author as an individual learning resource.

For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.