4

THE DISCRETE-TIME FOURIER TRANSFORM

Solution 4.1

The Fourier transform relation is given by

+C0

X(ejW) = 1 x(n) e-jWn thus:

n=-o +O0

(a) X(e") =W 6(n - 3) e-jwn = e-jw3

n=-o

2 'ejW + 1
(b) X(e j) = 1 + e- = 1 + cosw
2f

(c) X(e W) =2 an e-jn =2 (ae jw)n 1 a

1 - ae

n=O n=O
+3 6 7w
) =E e-jon = ej3 e- wn s

(d) X(e
n=-3 n=O sin(w)

Solution 4.2

(a) In problem 2.4(c) we determined that the convolution of an u(n) and

n u(n) was given by

y(n) = n+1 _ n+1 u(n)

y(n) = n+l n+l u(n)

= [n a + n _ (n)

thus, k a B and k a

(b) From problem 4.1 (c) it follows that

. 1
H(e 3 ) = 1 a-j

1 - a

and

X(e j) = 1
1 -e

S4.1

The Fourier transform of y(n) as obtained in (a)

00

Y(eS) =2 n+l _ n+l e-jwn

n=0

=n
- jon n -jwn

n= C,

6 a- 1 - Se -3w -a 1 - ae- j

=(1 - Be~5" (1 - ae - )

Solution 4.3

+00 +0

(a) Xa(e j) = kx(n)e-jon = k x(n)e-jwn

n=-o n=-o

= k X(ej)

(b) Xb(e) x(n - n0 ) e-jn

n=-co

Making the substitution of variables

m = n -
or n = m + n0
n0

Xb (e) = x(m) e- j(m+n0 e-jon0 x(m) e-jom

m=- 0 m=-00

=e-jon 0 X(e jw)

(c) The transform of x(n) is given by

+0

3 x(n) e-jwn

X (e w) =

n=-o

dX (ej w) - j ton
thus de) = (-jn) x(n) e
dw

n=-o
or +C0

dX(e j = n x(n) e-jon

dw

= Xc (e

S4.2

Solution 4.4

(a) H = h (t) e jQ t dt= aeat e-jQ t t a

-a 0+

IHa (j)1 = Thus |II(jQ)| is as sketched below:

2 + a2

IH(jQ) I

Figure S4.4-1

(b) hd (n) = cae-anT u(n) = ca(e-aT )n u(n)

ca jO 1 -e-aT
thus Hd (e -j
- T e-3 For Hd(e ) = 1, c = a
1 e-aT a

With this choice of c

(1- eaT 2
IH (eJ ) 12 2
d 1 - 2e-aT cos w + e aT

thus lHd(ejW)I is:

Figure S4.4-2

S4.3

Note in particular that while the frequency response of the

continuous-time filter asymptomatically approaches zero the frequency
response of the digital filter doesn't. However as the sampling period
T decreases, the value of IHd(ejW)| at w = 7 decreases toward zero.
The difference in the minimum values of |Ha(jP)| and IHd(e W)I is

of course due to aliasing.

Solution 4.5

From the discussion in the lecture, we know that

x A (jQ2) 2:(j
1

T A
+ 2'rr
T

r=-co
and

X(e JW) = XA()

W e
X XA (j) IT = w

Let us assume that XA(j) has some arbitrary shape as indicated below.
Since we are assuming that T is sufficiently small to prevent aliasing,
XA(jQ) must be zero for J|I > . Then XA(jQ) and X(e3W) are as

shown in figure S4.5-1. Y(ejW) corresponding to the output of the

filter and YA(jQ) and YA(jQ) follow in a straightforward way and are
as indicated in figure S4.5-1. Thus yA(n) could be obtained directly

by passing x(n) through an ideal lowpass filter with unity gain in

the passband and a cutoff frequency of 7T rad/sec. For the case in

part (a) the cutoff frequency of the overall continuous-time filter

is x 10 rad/sec and for the case in part (b) the cutoff frequency
4 4
is x 10 rad/sec.

S4.4

 XA (j 0)
1

XA
1

I -10
7T 7T

XA (e)

Y(e )

K K
100000
-2 r Tr
27T
4

4
T
YA Q

K> 27T
K
7T Tr
K> 2
T 4 T 4T -p

YA (j )

7T 7T

Figure S4.5-1

S4.5
 *
Solution 4.6

(1) and (2) can be verified by direct substitution into the inverse

Fourier transform relation. (3) and (4) follow from (1) since

Re [x(n)] = [x(n) + x (n)j and jIm [x(n)) = Ix(n) - x (n).

(5) and (6) follow from (2) since Re[X(ew)] = X(elw) + X (eW)]

and jIm [X(e")] = (X (e) - X (e~)3 .

Solution 4.7*

If X(ejW) denotes the Fourier transform of x(n), then

x(O) = X(e j) dw

-Jr

Thus, with y(n) denoting the convolution of f(n) and g(n) and since

Y(e ) = F(e W) G(e j), we wish to show that

y(O) = f(O) g(Q)

+CO

But yfn) = f(k) g(n - k)

k=-co

so y(O) = f(k) g(-k)

k=-o

Since f(k) is zero for k < 0 and g(-k) is zero for k > 0,

f(k) g(-k) = f(0) g(O)

k=-o

Solution 4.8*

(a) Method A:

Consider x(n) as a unit-sample 6(n). Then

g(n)=h(n) and r(n) = h(n) * g(-n) = h(k) h(-n+k)

+0 k=-o

Finally, s(n) = r(-n) = h(k) h(k + n)

k=-o

Consequently, h1 (n) = h(k) h(k + n)

k=- 0

S4.6
To show that this corresponds to zero phase, we wish to show that
h (n) = h1 (-n) since from Table 2.1 of the text with h (n), if
h (n) = h1 (-n) then

H (e )W = H (ejW) and hence the frequency response is real.

h (-n) = h(k) h(k - n)

k=-o

letting k - n = r,

h1 (-n) = h(n + r) h(r)

k=-o

which is -identical to h (n).

Alternatively we can show that h1 (n) corresponds to a zero-phase

filter by arguing in the frequency domain.

Specifically,

Let g(n) = g(-n). Then Ge) - G, (e ) = X (e ) H (e)

Also, R(e ) = X (e ) H (e W) H(e) = X (ejW) IH(e j)12

and S(ejW) = R (e )

= X(e W) |H(ejW )12

Thus, H (e ) = IH(e 2.
)| Since H (e )W is real, it has a zero phase

characteristic.

(b) Method B:

G (e ) - X(e W) H(e )W

R(e W) X (ej) H(e )W

Y(e )-=) G (ej) + R(e )W

- X(e j) [H(ejW) + H*(e )]

= X(e j) [2 Re H(ejW)]

Therefore H 2 (ejW) = 2 Re H(e ) = 2 IH(e )I cos[(arg H(ejW)]) and

consequently is also zero phase.

(c) H (ejW) and H2(e ) are sketched below. Clearly method A is the

preferable method.

S4.7
 H (eJ])

Figure S4.8-1

H (e"')

7­
3-7

Figure S4.8-2

S4.8

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.