Digital Signal Processing Reference
In-Depth Information
an analog signal and which produces a finite set of discrete-frequency spectrum val-
ues. This Fourier transform approximation is well suited for calculation by a digital
computer. It is called the
discrete Fourier transform
(DFT).
To use a digital computer to calculate the Fourier transform of a continuous-
time signal, we must use sampled values of in the form of a discrete-time signal
x
[
n
]. From Section 12.1, we have the discrete-time Fourier transform (DTFT):
x(t)
(x[n]) =
q
n=-
q
x[n]e
-jnÆ
.
[eq(12.1)]
X(Æ) =
The DTFT is computed from discrete-time samples, but is a continuous
function of the frequency variable and cannot be represented exactly in a digital
computer. However, using digital computations, we can approximate
X(Æ)
X(Æ)
by cal-
culating discrete-frequency samples of the continuous-frequency function.
To generate the discrete-frequency samples, we must, for practical reasons,
limit ourselves to a finite set of discrete-time samples. For the purpose of this de-
velopment, we will let the symbol
N
represent the number of samples chosen to
represent the discrete-time signal. We choose the value of
N
sufficiently large so
that our set of samples adequately represents all of
x[n].
We can select our finite
set of samples by multiplying the infinite set
x[n]
by a rectangular windowing
function [5]:
1, n = 0, 1, 2, Á , N - 1
0,
b
w
R
[n] =
.
otherwise
Then the set of samples used to calculate the frequency spectrum is
x[n], n = 0, 1, 2, Á , N - 1
0,
b
x
N
[n] = x[n]w
R
[n] =
.
(12.26)
otherwise
The frequency spectrum of the signal
x
N
[n]
is given by (12.1):
q
N- 1
x
N
[n]e
-jnÆ
=
a
x[n]e
-jnÆ
.
X
N
(Æ) =
(x
N
[n]) =
a
(12.27)
n=-
q
n= 0
For the remainder of the development, we consider the set of
N
samples of
to be the complete signal. We will drop the subscript on and refer to the
finite discrete-time sequence simply as This is justified by the assumption that
we have chosen the finite set of samples so that they adequately represent the entire
signal.
We now select
N
samples of to represent the frequency spectrum. We
could choose more than
N
samples, but we must choose at least
N
to avoid creating
x
N
[n]
x
N
[n]
x[n].
X
N
(Æ)