I Introduction

Solutions to
Recommended Problems
S1.1

(a) Using Euler's formula,

,r4
r 1r F
\IE S
e cos + j sin- 4 2
Since z = iej"/4
Re{z}= Re '2 + 2

(b) Similarly,

Im{z}= Im 2 2 4
(c) The magnitude of z is the product of the magnitudes of 2 and eJT/4. However,
I| = i, while Iei"| = 1 for all 0. Thus,
IzI = e =I
- ie 4
| - 1 =

(d) The argument of z is the sum of the arguments of i and e'"/4 . Since 4 = 0 and
4 ei" = 0 for all 0,

= 4 e = + +es"=

(e) The complex conjugate of z is the product of the complex conjugates of 1 and
ei"/4 . Since * = and (ej")* = e ~i' for all 0,
z* =* * 4 =i -i'/4

(f) z + z* is given by
1r/4 1 . 4 ej"/4 + e /4 Vr2\/
z + z* =e + e-/4 = = cos ­
2 2 2 4 2
Alternatively,
z+ z*
Refz} 2, or z + z* = 2Re{z}= 2 4 2

S1.2
(a) Express z as z = a + jQ, where Re{z} = a and Im{z} = Q. Recall that z* is the
complex conjugate of z, or z* = a - jQ. Then
z + z* (a+jQ)+ (o -jQ) 2a + 0
2 2 2
(b) Similarly,
z - z* (a +fj) - (a -jQ) _ + 2jQ _.
2 2 2

S1-1
Signals and Systems
S1-2

S1.3

(a) Euler's relation states that e'" = cos 0 + j sin 0. Therefore, e-o = cos(-0) +
j sin (-0). But, cos (-0) = cos 0 and sin (-0) = -sin 0. Thus, e-j" = cos 0 ­
j sin 0. Substituting,
e" + e - _ (cos 0 + j sin0) + (cos 0 - j sin0) 2 cos 0
2 2 2
(b) Similarly,
ej"- e -_ (cos 0 + j sin 0) - (cos 0 - j sin0) 2j sin0 s
= =-sin 6
2j 2j 2j

S1.4
(a) (i) We first find the complex conjugate of z = re'". From Euler's relation,
rej" = r cos 0 + Jr sin 0 = z. Thus,
z* = r cos 0 - jr sin 0 = r cos 0 + jr (-sin 0)
But cos 0 = cos (-0) and -sin 0 = sin (-0). Thus,
z* = r cos (-0) + jr sin (-0) = re -j"
2 20
z 2 = (re'")2 = r 2 (ejo) = r ei
2
(ii)
(iii) jz = er/ 2reo = reO+<,/2)
(iv) zz* = (re")(re-jo) = r 2em- 0 )>= r2 1
z 0
rej" e o+)e20
() z* re ~j"

(vi) = j__= e - e
z re 0 r
(b) From part (a), we directly plot the result in Figure S1.4-1, noting that for z =
re", r is the radial distance to the origin and 0 is the angle counterclockwise
subtended by the vector with the positive real axis.

Im{z}

3 jr/6
3

Re(z}

Figure S1.4-1
 Introduction / Solutions
S1-3

(i)

(ii)
Im z)
z2 = 4 ej/3
4 9
9

Rez)

Figure S1.4-3

(iii)
Im Iz
jz= 2 ej27/3
3

2 21r
33

Refz)

Figure S1.4-4

(iv)
Indz)

4
9

Figure S1.4-5
Signals and Systems
S1-4

(v)

z*

Itnzl l=

z Rejzj

Figure S1.4-6

(vi)

7T ReWjz
-- 6

3
2
1 3 -jn/6
z 2

Figure S1.4-7

S1.5
This problem shows a useful manipulation. Multiply by e +a/ 2
e-ja/2 = 1, yielding
e+ja/2e -ja/2( 1 - eja) ja/
2
(e -ja/ 2
_ eia/2 )
Now we note that 2j sin (-x) = - 2j sin x = e -x - ex. Therefore,

1 - el" = ei" 2 (-2j sin 2)

Finally, we convert -j to complex exponential notation, -j = e -jr/2. Thus,

1 - eia = ea/2 2e -j/2 sin = 2 sin ejt(a-7r)/21

S1.6
There are three things a linear scaling of the form x(at + b) can do: (i) reverse
direction => a is negative; (ii) stretch or compress the time axis = a I # 1; (iii) time
shifting => b # 0.
 Introduction / Solutions
S1-5

(a) This is just a time reversal.

Note: Amplitude remains the same. Also, reversal occurs about t = 0.

(b) This is a shift in time. At t = -2, the vertical portion occurs.

x(t + 2)

t
-2

Figure S1.6-2

(c) A scaling by a factor of 2 occurs as well as a time shift.

x(2t + 2)

t
-1 1
2

Figure S1.6-3

Note: a > 1 induces a compression.

(d) All three effects are combined in this linear scaling.

x(1 -3t)

t
2 1
3 3
Figure S1.6-4
Signals and Systems
S1-6

S1.7

This should be a review of calculus.

(a) e 2tdt = -e -2 -2a [-e 2(0)

= 1 -2a

(b) e -3' dt = -i = lim (-e 31

) ie-3(2 >
Jf2 12 t-o0 3

Therefore,

fe 3- dt = 0 + 1e 6 56
MIT OpenCourseWare
http://ocw.mit.edu

Resource: Signals and Systems

Professor 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.
2 Signals and Systems: Part I
Solutions to
Recommended Problems
S2.1
(a) We need to use the relations w = 21rf, where f is frequency in hertz, and
T = 2w/w, where T is the fundamental period. Thus, T = 1/f.
w /3 1 1
(i) f=-= =-Hz, T- 6s
2r 21 6 f
3r/4 3 8
(ii) f = =-Hz, T=-s
2w 8 3
3/4 3 8-x
(iii) f = = Hz, T = -s
3
2r 87r
Note that the frequency and period are independent of the delay r, and the
phase 0_.
(b) We first simplify:
cos(w(t + r) + 0) = cos(wt + wr + 0)
Note that wT + 0 could also be considered a phase term for a delay of zero.
Thus, if w, = w, and wX, + 0, = wor, + 6, + 2xk for any integer k, y(t) = x(t)
for all t.

(i) Wx = o ~ mr+Ox=
W,
2
W, Wyr, +O ,= 3 1 - 3 =O0+ 2wk

Thus, x(t) = y(t) for all t.

(ii) Since wx # w,, we conclude that x(t) # y(t).
(iii) COX = coy, omr, + 6X = ((i) + i #34 3(1) + a + 2,7k

Thus, x(t) # y(t).

S2.2
(a) To find the period of a discrete-time signal is more complicated. We need the
smallest N such that UN = 21k for some integer k > 0.

(i) N = 2wk =* N = 6, k = 1
3

(ii) N = 2rk =o N = 8, k = 2
4
(iii) 2N = 2wk => There is no N such that aN = 2wk, so x[n] is not periodic.
(b) For discrete-time signals, if Ox = Q, + 2rk and Qxrx + Ox = Qr, + O, + 2k, then
x[n] = y[n].

(i) ' 81r + 2wk (the closest is k = -1), so x[n] # y[n]

3 3

(ii) Ox = I3, (2) + =r - r + 2k, k = 1, so x[n] = y[n]

(iii) Ox= 1,,((1) + i = ((0) + 1 + 27rk, k = 0, x[n] = y[n]

S2-1
Signals and Systems
S2-2

S2.3

(a) (i) This is just a shift to the right by two units.

x[n-2]

-1 0 1 2 3 4 5 6
Figure S2.3-1

(ii) x[4 - n] = x[-(n - 4)], so we flip about the n = 0 axis and then shift
to the right by 4.

x[4-n]

0
0 2~l
-1 0 1
111
2 3 4 5 6
n
Figure S2.3-2

(iii) x[2n] generates a new signal with x[n] for even values of n.

x [2n]

0123
Figure S2.3-3
2 0 n

(b) The difficulty arises when we try to evaluate x[n/2] at n = 1, for example (or
generally for n an odd integer). Since x[i] is not defined, the signal x[n/2] does
not exist.

S2.4
By definition a signal is even if and only if x(t) = x(-t) or x[n] = x[-n], while a
signal is odd if and only if x(t) = -x(- t) or x[n] = -x[-n].
(a) Since x(t) is symmetric about t = 0, x(t) is even.
(b) It is readily seen that x(t) # x(- t) for all t, and x(t) K -x(- t) for all t; thus
x(t) is neither even nor odd.
(c) Since x(t) = -x(- t), x(t) is odd in this case.
 Signals and Systems: Part I / Solutions
S2-3

(d) Here x[n] seems like an odd signal at first glance. However, note that x[n] =
-x[-n] evaluated at n = 0 implies that x[O] = -x[O] or x[O] = 0. The analo­
gous result applies to continuous-time signals. The signal is therefore neither
even nor odd.
(e) In similar manner to part (a), we deduce that x[n] is even.
(f) x[n] is odd.

S2.5
(a) Let Ev{x[n]} = x[n] and Od{x[n]} = x[n]. Since xe[n] = y[n] for n >_ 0 and
xe[n] = x[ -fn], x,[n] must be as shown in Figure S2.5-1.

Xe[n] 2

41 4

n
-5 -4 -3 -2 -1 0 1 2 3 4 5

Figure S2.5-1

Since x[n] = y[n] for n < 0 and x[n] = -x,[-n], along with the property that
x0 [O] = 0, x[n] is as shown in Figure S2.5-2.

Xo [n]

1 2 3 4
0 n
-4 -3 -2 -1 0

Figure S2.5-2

Finally, from the definition of Ev{x[n]} and Od{x[n]}, we see that x[n] = x,[n] +
x[n]. Thus, x[n] is as shown in Figure S2.5-3.
Signals and Systems
S2-4

(b) In order for w[n] to equal 0 for n < 0, Od{w[n]} must be given as in Figure
S2.5-4.

Odfw[n]} 14

-4 - 3 --2 --
n
0 1 2 3 4

Figure S2.5-4

Thus, w[n] is as in Figure S2.5-5.

w[n] 21

p e - 0 n
-3 -2 -1 0 1 2 3 4
Figure S2.5-5

S2.6

(a) For a = -ia"is as shown in Figure S2.6-1.

x[n] 1

Figure S2.6-1

(b) We need to find a # such that e#' = (-e-')". Expressing -1 as ei", we find
e on = (ej'e -)T or 0 = -1 + jr
Note that any # = -1 + jfr + j27rk for k an integer will also satisfy the preced­
ing equation.
 Signals and Systems: Part I / Solutions
S2-5

(c) Re{e(-1 +')t ) = e -Re{ej'"} = e -" cos rn,

Im~e (- 1+j} n = e -"Im{e'"} = e~" sin in

Since cos 7rn = (- 1)' and sin 7rn = 0, Re{x(t)) and Im{y(t)} for t an integer are
shown in Figures S2.6-2 and S2.6-3, respectively.

1
Refe (- + ir)n

-1 1In n
0 -e­

-e

Figure S2.6-2

Imfe (- +j7T)n I
0 0 0 0 0 -nf
-2 -1 0 1 2
Figure S2.6-3

S2.7
First we use the relation (1 + j) = \/Tej"!' to yield
x(t ) = \/ \/ej'4eJ"/4e(-I+j2*)t = 2eir/2e(-1+j2 )t

(a) Re{x(t)} = 2e-'Re{ew"!ej'i'} = 2e-' cos( 2t + 2)

Re{x (t)}

envelope is 2e-r

2- -­

Figure S2.7-1
Signals and Systems
S2-6

(b) Im{x(t)) = 2e-'Im{ejr/2e 2 1rt } = 2e-' sin 27rt

ImIx (t)}
envelope is 2et

2 -­

Figure S2.7-2

(c) Note that x(t + 2) + x*(t + 2) = 2Re{x(t + 2)}. So the signal is a shifted
version of the signal in part (a).

x(t + 2) + x*(t + 2)

, 4e 2

Figure S2.7-3

S2.8
(a) We just need to recognize that a = 3/a and C = 2 and use the formula for SN,
N = 6.
6

= 2 =32
a /3\ _
-a)
(3)a

(b) This requires a little manipulation. Let m = n - 2. Then

5
6 4 4 1-
­
nb= nb= =0 bm""2=b2(b=2 =
n=2 m=O M=0 1 -b
 Signals and Systems: Part I / Solutions
S2-7

(c) We need to recognize that (2)2n = (1)'. Thus,

since - 1
- = o-4

S2.9
(a) The sum x(t) + y(t) will be periodic if there exist integers n and k such that
nT1 = kT 2, that is, if x(t) and y(t) have a common (possibly not fundamental)
period. The fundamental period of the combined signal will be nT1 for the small­
est allowable n.
(b) Similarly, x[n] + y[n] will be periodic if there exist integers n and k such that
nN = kN 2. But such integers always exist, a trivial example being n = N 2 and
k = N1 . So the sum is always periodic with period nN, for n the smallest allow­
able integer.
(c) We first decompose x(t) and y(t) into sums of exponentials. Thus,
13 1 ei(1 6Tt/ 3) -j(16irt/3)
x(t) = 1 ej( 1t/ ) + -e j(21rt/3) + e
2
e
__A6r3 ____

2 2 j
Y( jrt -e-jrt
23 2j
Multiplying x(t) and y(t), we get
3
z(t) - e ( -+/3 - e-j5w/s)
7 )t( + / )t -j 1/3
4j 4j 4j 4j
1 ej(19r/3 )t + 1 ej(13 r/3) t + 1 e j(13r/3)t - -j(19/3)t
2 2 2 2
We see that all complex exponentials are powers of e j(/3). Thus, the funda­
mental period is 2 7r/(7r/3) = 6 s.

S2.10

(a) Let 7 x[n] = S. Define m = -n and substitute

n = -oo

M=Z-QO
x[mI
M= Sx
-OO
[-m] = -

since x[m] is odd. But the preceding sum equals -S. Thus, S = -S, or S = 0.
(b) Let y[n] = x 1[n]x2[n]. Then y[-n] = x1[-n]x 2[-n]. But x1[-nj = -xl[n] and
x2[-n] = x 2[n]. Thus, y[-n] = -x1[n]x2[n] = -y[n]. So y[n] is odd.
(c) Recall that x[n] = x[fn] + x[n]. Then

E x2[n] = E (xe[nl + x.[n])2

= x[n] + 2 E Xe[flx]Xfl + >7 x 2[n]

n= -co n= o n= -0

But from part (b), x,[nxo[n] is an odd signal. Thus, using part (a) we find that
the second sum is zero, proving the assertion.
Signals and Systems
S2-8

(d) The steps are analogous to parts (a)-(c). Briefly,

(i) S =f x0(t)dt = x,(-r) dr

=0­ x 0 (r) dr = -S, or S = 0, where r = -t

(ii) y(t) = Xo(t)xe(lt),

y(-t) = xo(-t)x(-t) = -xo(t)xo(t)
- y(t), y(t) is odd

(iii) x 2 (t) dt = f (X(t) + x 0 (t)) 2 dt

= x(t ) dt + 2 Xe(t)xo(t)dt + x 2(t)dt,

while 2r x(t)xo(t) dt = 0

S2.11
(a) x[n] = ewonT - ei2 nTTo. For x[n] = x[n + N], we need
x[n + N] = ej 2x(n +N)T/To eji[ 2 n(T/To) + 2rN(T/To)] = ej2nT/To

The two sides of the equation will be equal only if 27rN(T/TO) = 27rk for some
integer k. Therefore, TITO must be a rational number.
(b) The fundamental period of x[n] is the smallest N such that N(T/TO) = N(p/q)
= k. The smallest N such that Np has a divisor q is the least common multiple
(LCM) of p and q, divided by p. Thus,

N LCM(p, q);
note that k = LCM(p, q)
p q
The fundamental frequency is 2ir/N, but n = (kT 0)/T. Thus,

Q= 2 = = WT = q T
N kT,, k LCM(p, q) 0

(c) We need to find a value of m such that x[n + N] = x(nT + mT). Therefore,
N = m(T./T), where m(T./T) must be an integer, or m(q/p) must be an integer.
Thus, mq = LCM(p, q), m = LCM(p, q)/q.
MIT OpenCourseWare
http://ocw.mit.edu

Resource: Signals and Systems

Professor 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.
3 Signals and Systems: Part II
Solutions to
Recommended Problems
S3.1

(a)
x[n]= 8[n] + 8 [n - 3]

n
0 1 2 3

Figure S3.1-1

(b)
x[n] = u[n]-u[n-- 5]

1111
0041T
-1 0 1 2 3 4 5
Figure S3.1-2

(c)
x [n]=S[n] + -1 [n -- +(_1)2S[n] +(_1)3 5[n -3]

-3 -2 -1
1
0 1 2 3 4 5 6 7
n

Figure S3.1-3

(d)
x(t)= u(t +3) - u(t - 3)

t
-3 0 3

Figure S3.1-4

S3-1
Signals and Systems
S3-2

(e)
x(t) =6(t + 2)

t
-2 0

Figure S3.1-5

(f)

S3.2
(1) h
(2) d
(3) b
(4) e
(5) a, f
(6) None

S3.3
(a) x[n] = b[n - 1] - 26[n - 2] + 36[n - 3] - 26[n ­ 4] + b[n - 5]
(b) s[n] = -u[n + 3] + 4u[n + 1] - 4u[n ­ 2] + u[n - 4]

S3.4

We are given Figure S3.4-1.

 Signals and Systems: Part II / Solutions
S3-3

x(- t) and x(1 - t) are as shown in Figures S3.4-2 and S3.4-3.

x (-t)

-12

Figure S3.4-2

x(1--t)
x1-0

-11 1
Figure S3.4-3

(a) u(t + 1) - u(t - 2) is as shown in Figure S3.4-4.

-1 0 1 2
t
Figure S3.4-4

Hence, x(1 - t)[u(t + 1) - u(t - 2)]1 looks as in Figure S3.4-5.

6
t
-1 1

Figure S3.4-5
Signals and Systems
S3-4

(b) -u(2 - 3t) looks as in Figure S3.4-6.

2
3
t

Figure S3.4-6

Hence, u(t + 1) - u(2 - 3t) is given as in Figure S3.4-7.

2
3

Figure S3.4-7

So x(1 ­ t)[u(t + 1) - u(2 - 3t)] is given as in Figure S3.4-8.

S3.5

(a) y[n] = x2[n] + x[n] - x[n - 1]

(b) y[n] = x2[n] + x[n] - x[n - 1]
(c) y[n] = H[x[n] ­ x[n - 1]]
x 2[n] + x 2 [n - 1] - 2x[nJx[n ­ 1]
(d) y[n] = G[x 2[n]]
x 2[n] - X 2 [n - 1
 Signals and Systems: Part 11/ Solutions
S3-5

(e) y[n] = F[x[n] - x[n - 1]]

= 2(x[n] - x[n - 1]) + (x[n - 1] - x[n - 2])
y[n] = 2x[n] - x[n - 1] - x[n - 2]
(f) y[n] = G[2x[n] + x[n - 11]
= 2x[n] + x[n - 11 - 2x[n - 1] - x[n - 21
= 2x[n] - x[n - 1] - x[n - 2]

(a) and (b) are equivalent. (e) and (f) are equivalent.

S3.6
Memoryless:
(a) y(t) = (2 + sin t)x(t) is memoryless because y(t) depends only on x(t) and not
on prior values of x(t).
(d) y[n] = Ek=. x[n] is not memoryless because y[n] does depend on values of
x[-] before the time instant n.
(f) y[n] = max{x[n], x[n - 1], ... , x[-oo]} is clearly not memoryless.
Linear:
(a) y(t) = (2 + sin t)x(t) = T[x(t),
T[ax 1 (t) + bx 2 (t)] = (2 + sin t)[axi(t) + bx 2 (tt)
= a(2 + sin t)x 1(t) + b(2 + sin t)x 2 (t)
= aT[x 1 (t)] + bT[x 2 (t)]
Therefore, y(t) = (2 + sin t)x(t) is linear.
(b) y(t) = x(2t) = T[x(t)],
T[ax 1(t) + bx 2 (t)] = ax1 (2t) + bx 2 (t)
= aT[x1 (t)) + bT[x 2 (t)]
Therefore, y(t) = x(2t) is linear.

(c) y[n] = ( x[k] = T[x[n]],

k=­

T[ax1 [n] + bx2[n]] = a T x1 [k] + b L x 2[k]

k= -x k=-w
= aT[x1[n]] + bT[x2[n]]
Therefore, yin] = E=_ x[k] is linear.

(d) y[n] = > x[k] is linear (see part c).

k= -o

dxt
d
T[ax 1(t) + bx 2(t)] = -[ax 1 (t)+±bx
2 (t)]

dx 1 (t) dx (t)
= a dt + b dt2 = aT[x 1 (t)] + bT[x 2 (t)]

Therefore, y(t) = dx(t)/dt is linear.

(f) y[n] = max{x[n], . . . , x[-oo]} = T1x[n]],
T[ax 1[n] + bx2[n]] = max{ax 1 [n] + bx 2[n], . . . , ax 1[-oo] + bx2[ - o]}
# a max{x[n], . . . , x1[-oo]} + b max{x 2[n], . . . , X2[-00])
Therefore, y[n] = max{x[n], .. . , x[-oo]} is not linear.
Signals and Systems
S3-6

Time-invariant:
(a) y(t) = (2 + sin t)x(t) = T[x(t)],
T[x(t - T 0 )] = (2 + sin t)x(t - T0 )
9 y(t - T0 ) = (2 + sin (t - T0 ))x(t - T0 )
Therefore, y(t) = (2 + sin t)x(t) is not time-invariant.
(b) y(t) = x(2t) = T[x(t)],
T[x(t - # x(2t - TO) = y(t - T 0 )
T0 )] = x(2t - 2T0)
Therefore, y(t) = x(2t) is not time-invariant.

(c) y[n] = ( x[kJ = T[x[n]],

T[x[n - NO]J = ( x[k - NO] = y[n - N 0 ]

Therefore, y[n] = Ek'= _.x[k] is time-invariant.

(d) y[n] = E
k=­
x[k] = T[x[n]],
n n-NO

T[x[n - NO]] = E x[k - N] = x[l] = y[n - N

0 ]
k=- =-w0

Therefore, y[n] = E" _.x[k] is time-invariant.

=dx(t)
(e) y(t) dt T[x(t)],
d
T[x(t - To)] = x(t - To) = y(t - To)
dt
Therefore, y(t) = dx(t)/dt is time-invariant.
Causal:
(b) y(t) = x(2t),
y(1) = x(2)
The value of y(-) at time = 1 depends on x(-) at a future time = 2. Therefore,
y(t) = x(2t) is not causal.

(d) y[n] = ( x[k]

k=­

Yes, y[n] = E .x[k] is causal because the value of y[-] at any instant n
depends only on the previous (past) values of x[-].
Invertible:
(b) y(t) = x(2t) is invertible; x(t) = y(t/2).
(c) y[n] = E _.x[k] is not invertible. Summation is not generally an invertible
operation.
(e) y(t) = dx(t)/dt is invertible to within a constant.
Stable:
(a) If Ix(t) I < M, Iy(t) I < (2 + sin t)M. Therefore, y(t) = (2 + sin t)x(t) is stable.
(b) If |x(t)| < M, |x(2t)I < M and ly(t)| < M. Therefore, y(t) = x(2t) is stable.
(d) If |x[k]| 5 M, ly[n]j 5 M - E_,, which is unbounded. Therefore, y[n] =
E"Lx[k] is not stable.
 Signals and Systems: Part II / Solutions
S3-7

S3.7

(a) Since H is an integrator, H-1 must be a differentiator.

H~':
dx(t)
y(t) = d
dt
G: y(t) = x(2t)
G 1: y(t) = x(t/2)

(b)

Solutions to
Optional Problems
S3.8
(a) x 2(t) = xi(t) - xi(t - 2)

y)(t - 2)
2(t) = 1 (t)

Figure S3.8-1
Signals and Systems
S3-8

(b) xA(t) = Xi( t) + xI(t + 1)

y 3(t)=y (t)+y 1 (t + 1)

-1 10 1 2

Figure S3.8-2

u(t
-2)
-u(1
-t)
y(t)=e-(t-1)u
-1)+u(-t)+
(c) x(t) = u(t - 1) - u(t - 2)

Figure S3.8-3

(d) y[n] = 3y 1 [n] ­ 2y2[n] + 2y 3 [n]

3 3

p p14
p n

-3 -2 -1 0 1 2 3

-4

Figure S3.8-4
 Signals and Systems: Part 11/ Solutions
S3-9

(e) y2[n] = y,[n] + yi[n - 11

y2 [n] 2 __

0 1 2 3 4 5
Figure S3.8-5

Y3[n] = y 1 [n + 1]

-1 0 234
Figure S3.8-6

(f) From linearity,

y1(t) = 1r + 6 cos(2t) - 47 cos(5t) + '/e cos(6t),
1 + 0t 4"
2
x 2(t) = 1 + t2 = (-t )".
n=O

So y 2(t) = 1 - cos(2t) + cos(4t) - cos(6t) + cos(8t).

S3.9
(a) (i) The system is linear because

Tlaxi(t) + bx 2 (t)] = 1
n=­
23 [ax (t) + bx (t )](t - nT)

= a T3 x 1 (t)b(t - nT) + b ( xst - nT)

= aT[xi(t)] + bT[x 2 (t)]

(ii) The system is not time-invariant. For example, let xi(t) = sin(22rt/T).
The corresponding output yi(t) = 0. Now let us shift the input xi(t) by
r/2 to get

x2 (t) = sin (
+r =
cos (2)
Now the output
+00

Y2(0 >7 b(t - nT) =Ay, + 2= 0

n = -oo
Signals and Systems
S3-10

(b) y(t) = E x(t)b(t - nT)

= E cos(2,rt)b(t - nT)

cos(27rt)

0< t

Figure S3.9-1

T=1

0 tt
I ~11
2

Figure S3.9-2
 Signals and Systems: Part II / Solutions
S3-11

y (t )

T=
T~__ t t

-1
Figure S3.9-5

y (t)

Figure S3.9-6

(c) y(t) = e t cos(27rt)b(t - nT)

etcos(2nrt)
et

t
22

Figure S3.9-7

y (t)

T= I
T=1 tt
1 3
2

Figure S3.9-8
Signals and Systems
S3-12

y (t)
T= Ie3
2 2
2 I 2- t
-e1/2 I I 3/2

Figure S3.9-9

y (t)
T= YWe3
4t
F r2 2
-el2 -e 3 12

Figure S3.9-10

T= Y3 1 y~t)
12 2
1 2

Figure S3.9-12

S3.10

(a) True. To see that the system is linear, write

y2 (t) = T 2 [T1 [x(t)]] T[x(t)],

T 1[ax 1 (t) + bx 2(t)] = aT 1 [x1(t)] + bT[x 2(t)]
T2[T1[ax 1(t) + bx 2 (t)]] = T2[aT1 [x1(t)] + bT1[x 2(t)]I
= aT2[T1 [x1 (t)]] + bT2[(T[x2t)]]
= aT[x 1 (t)] + bT[x 2(t)J
 Signals and Systems: Part 11/ Solutions
S3-13

We see that the system is time-invariant from

T 2[T 1[x(t - T)]] = T 2[y 1(t - T)l
= y 2(t -T),
Tx(t - T)] = y 2(t - T)
(b) False. Two nonlinear systems in cascade can be linear, as shown in Figure S3.10.
The overall system is identity, which is a linear system.

x(t) i Reciprocal - Reciprocal 0 y(t)=x(t)

1
x(t)
Figure S3.10

(c) y[n] = z[2n] = w[2n] + {w[2n - 1] + {w[2n - 21

= x[n] + {x[n - 11

The system is linear and time-invariant.

(d) y[n] = z[-nl = aw[-n ­ 11 + bw[-n] + cw[-n + 1]
= ax[n + 11 + bx[nl + cx[n - 1]
(i) The overall system is linear and time-invariant for any choice of a, b,
and c.
(ii) a= c
(iii) a= 0

S3.11
(a) y[n] = x[n] + x[n - 11 = T[x[n]]. The system is linear because
T[ax 1 [n] + bx 2 [n]| = ax1[n] + ax1[n - 1] + bx 2[n] + bx 2[n - 1]
= aT[x1 [n]] + bT[x 2[n - 1]]
The system is time-invariant because
y[n] = x[n] + x[n - 1] = Tjx[n]],
T[x[n - N]] = x[n - N] + x[n ­ 1 - N]
= y[n - N]

(b) The system is linear, shown by similar steps to those in part (a). It is not
time-invariant because
T[x[n ­ N]] = x[n - N] + x[n - N - 1] + x[O]
# y[n - N] = x[n ­ N] + x[n - N - 1] + x[-NJ

S3.12
(a) To show that causality implies the statement, suppose
x1 (t) - yl(t) (input x 1(t) results in output y 1(t)),
x 2(t) - y2),
where y 1(t) and y 2(t) depend on x 1(t) and x 2(t) for t < to. By linearity,
xI(t) - x 2(t) -+ y 1 (t) - y 2(t)
Signals and Systems
S3-14

If Xi(t) = x 2 (t) for t < to, then y 1 (t) = y 2 (t) for t < to. Hence, if x(t) = 0 for
t < to, y(t) = 0 for t < to.
(b) y(t) = x(t)x(t + 1),
x(t) = 0 for t < to =* y(t) = 0, for t < to
This is a nonlinear, noncausal system.
(c) y(t) = x(t) + 1 is a nonlinear, causal system.
(d) We want to show the equivalence of the following two statements:
Statement 1 (S1): The system is invertible.
Statement 2 (S2): The only input that produces the output y[n] = 0 for all n is
x[n] = 0 for all n.
To show the equivalence, we will show that
S2 false S1 false and
S1 false S2 false
S2 false == S1 false: Let x[n] # 0 produce y[n] = 0. Then cx[n] == y[n] = 0.
S1 false S2 false: Let xi => yi and x 2 =* Y2. If x 1 # X 2 but y1 = Y2, then
X1 - X2 0 0 but yi - yi = 0.
(e) y[n] = x 2 [n] is nonlinear and not invertible.
MIT OpenCourseWare
http://ocw.mit.edu

Resource: Signals and Systems

Professor 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.
4 Convolution
Solutions to
Recommended Problems
S4.1
The given input in Figure S4.1-1 can be expressed as linear combinations of xi[n],
x 2[n], X3 [n].

x,[ n]

0 2
Figure S4.1-1

(a) = 2x 1 [n] - 2x 2 [n] + x3[n]

x 4 [n]

(b) Using superposition, y 4[n] = 2yi[n] - 2y 2[n] + y3 [n], shown in Figure S4.1-2.

-1 0 1

Figure S4.1-2
(c) The system is not time-invariant because an input xi[n] + xi[n - 1] does not
produce an output yi[n] + yi[n - 1]. The input x,[n] + xi[n - 11 is xi[n] +
xi[n - 1] = x2[n] (shown in Figure S4.1-3), which we are told produces y 2[n].
Since y 2[n] # yi[n] + yi[n - 1], this system is not time-invariant.

x 1 [n] +x 1 [n-1] =x2[n]

n
0 1
Figure S4.1-3

S4-1
Signals and Systems
S4-2

S4.2
The required convolutions are most easily done graphically by reflecting x[n] about
the origin and shifting the reflected signal.
(a) By reflecting x[n] about the origin, shifting, multiplying, and adding, we see
that y[n] = x[n] * h[n] is as shown in Figure S4.2-1.

(b) By reflecting x[n] about the origin, shifting, multiplying, and adding, we see
that y[n] = x[n] * h[n] is as shown in Figure S4.2-2.

y[n]

3
2

0 1 2 3 4 5 6

Figure S4.2-2

Notice that y[n] is a shifted and scaled version of h[n].

S4.3
(a) It is easiest to perform this convolution graphically. The result is shown in Fig­
ure S4.3-1.
 Convolution / Solutions
S4-3

y(t) = x(t) * h(t)

4-­

| t
4 8

Figure S4.3-1

(b) The convolution can be evaluated by using the convolution formula. The limits
can be verified by graphically visualizing the convolution.

y(t) = 7x(r)h (t - r)dr

= e-'-Ou(r - 1)u(t - r + 1)dr

1
t+

e (- dr, t> 0,

-0, t < 0,
Let r' = T - 1. Then

y( ) e- d r -e t > 0,
0 0 , t < 0

(c) The convolution can be evaluated graphically or by using the convolution

formula.

y(t) = x(r)6(t - , - 2) dr = x(t - 2)

So y(t) is a shifted version of x(t).

y(t)

It- I / \, t

1 3 5

Figure S4.3-2
Signals and Systems
S4-4

S4.4

(a) Since y[n] = E=-oox[m]h[n - m],

y[n] = b[m
6 - no]h[n - m] = h[n - no]
m= -oO

We note that this is merely a shifted version of h[n].

y [n] = h1[12­ I

ae|41 8 n
(n 1) no (n1+ 1)

Figure S4.4-1

(b) y[n] = E =_c(!)'u[m]u[n ­ m]

For n > 0: 1 +
y[n] = = 2( 1
,
y[n] = 2 - (i)"

Forn < 0: y[n] = 0

Here the identity
N-i
_ N
T am 1­ a
Mr=O

has been used.

y[n]

2--­

140

0 1 2

Figure S4.4-2

(c) Reversing the role of the system and the input has no effect on the output
because

y[n] = E x[m]h[n ­ m] = L h[m]x[n ­ m]

m=-oo

The output and sketch are identical to those in part (b).

 Convolution / Solutions
S4-5

S4.5
(a) (i) Using the formula for convolution, we have

y 1 (t) = x(r)h(t - r) dr

= r(-)-2 u(t - r) dr
t

e -( 2
= - dr, t > 0,

2e 10 = 2(1 e t > 0,

y(t) = 0, t< 0

y 1 (t)

-
2 -

t
0

Figure S4.5-1

(ii) Using the formula for convolution, we have

y2(t) = 2e-(t-r)/2 dr, 3 t>- 0,

=4(1 - e-t/2), 3 t : 0,

y 2 (t) = { 2e-(-­ /2

3
d-, t >_3,

4(e -(t-3)/2 _ e-t/2

S4e (t-r)/2 0
0

= 4e- t/ 2 (e'/ 2 t 3,
- 1),
y 2(t) = 0, t 0

y 2 (t)

0 1 2 3 4 5 6

Figure S4.5-2
Signals and Systems
S4-6

(b) Since x 2 (t) = 2[xl(t) - xl(t - 3)] and the system is linear and time-invariant,
y 2(t) = 2[yi(t) - y1(t - 3)].

For 0 s t s 3: y 2(t) = 2yi(t) = 4(1 - e-'/2)

For 3 t y 2(t) = 2y,(t) - 2yi(t - 3)
= 4(1 - t
e-1/2 4(1 - e- -3)2)
3
= 4e- t/ e /
2 2
_ 1
Fort< 0: y 2(t) = 0
We see that this result is identical to the result obtained in part (a)(ii).

Solutions to
Optional Problems
S4.6
(a)
x(T)

T
0 1

Figure S4.6-1

h(-1 -r)
2,

-2 T

--1 0

--1 '
Figure S4.6-2
 Convolution / Solutions
S4-7

h(0-r)

Figure S4.6-3

h(1 -- )

Figure S4.6-4
Signals and Systems
S4-8

Using these curves, we see that since y(t) = x(t) * h(t), y(t) is as shown in
Figure S4.6-6.

y(t)

1 W
--1 2
t
0 1

--- 4
Figure S4.6-6

(b) Consider y(t) = x(t) * h(t) =

f0 x(t - r)h(r)dr.

h(r)

0 1 2
Figure S4.6-7

For 0 < t < 1, only one impulse contributes.

x(t -r)

Figure S4.6-8

For 1 < t < 2, two impulses contribute.

x(t ­ )

Figure S4.6-9
 Convolution / Solutions
S4-9

For 2 < t < 3, two impulses contribute.

x(t- r)

Figure S4.6-10

For 3 < t < 4, one impulse contributes.

x(t ­ t)
Figure S4.6-11

For t < 0 or t > 4, there is no contribution, so y(t) is as shown in Figure

S4.6-12.

y(t)

3 -­

I-­
U' 1 2 3 4
Figure S4.6-12

S4.7
y[n] = x[n] * h[n]

=1 x[n - m]h[m]

nO
=- anmm,
- n > 0,
M=0
Signals and Systems
S4-10

y[n] = a" = a" L (la)

a n+1 _ n+1
n > 0,
a - #

y[n] = 0, n < 0

S4.8
(a) x(t) = E_= - kT) is a series of impulses spaced T apart.

x(t)

t
-2T -T 0 T 2T
Figure S4.8-1

(b) Using the result x(t) *(t - to) = x(to), we have

y(t)

.. t
-2 3 -1 0 1 1 3 2
2 2 2 2

Figure S4.8-2

So y(t) = x(t) * h(t) is as shown in Figure S4.8-3.

y(t)

-2 -3 -1 -j 0 1 1 3 2
22 2 2
Figure S4.8-3
 Convolution / Solutions
S4-11

S4.9
(a) False. Counterexample: Let g[n] = b[n]. Then
x[n] * {h[n]g[n]} = x[n] h[0],
{x[n]* h[n]}g[n] = b[n] [x[n] *h[n]]
n=0

and x[n] may in general differ from b[n].

(b) True.

y(2t) = fx(2t - r)h(r)dr

2
Let r' = T/ . Then

y(2t) = f x(2t - 2r')h(2r')2dr'

= 2x(2t)* h(2t)

(c) True.
y(t) = x(t) * h(t)
y(- t) = x(-t) * h(-t)

-f x(-t + r)h(-r)dr = f [-x(t - r)][ -h(r)] dT

=f x(t - r)h(r) dr since x(-) and h(-) are odd fu ictions

- y(t)
Hence y( t) = y(-t), and y(t) is even.
(d) False. Le t
x(t) = b(t - 1)
h(t) = b(t + 1)
y(t) = b(t), Ev{ y (t)} = 6(t)
Then
x(t) * Ev{h(t)} =b(t - 1) * i[b(t + 1) + b(t - 1)]
= i[b(t) + b(t - 2)],
Ev{x(t)} * h(t) =1[6t - 1) + b(t + 1)] * b(t + 1)
= 1[b(t) + b(t + 2)]
But since 1[6(t - 2) + b(t + 2)] # 0,
Ev{y(t)} # x(t) *Ev{h(t)} + Ev{x(t)} *h(t)

S4.10
(a)
9(t) = Jro TO2 1 (r)22(t - r) dr,
O

D(t + T0 ) = J 2 1 (T)- 2 (t + To - r) d-r

= TO 1 (r) 2 (t - r)dr = (t)

Signals and Systems
S4-12

(b) a+TO

9a(t) = T 2 1 (i)2 2 (t - -) dr,

a = kTo + b, where 0 b - To,

(k+1)T0 +b
9a(t) = 2 1 (r)A 2 (t - i) dr,
kT0+b

TO+b

Pa(t) = fb 2 1 (i)± 2 (t - r) di, i' = i - b

=
FTo
T1()- 2 t - r) dr +
T7 0 +b
1&)2(t - r) di
b TO

r) d1 2- ) di
=
=
T)TO Tb

21
t(-r
()A

)12( t ­ T )
-
)

dr =q t)

(c) For 0 s t - I

9(t ) = ft
0e
e- di +
+
Ti±e-'di I

e/2+t1/2+1

=(-e-' t + (-e-T
0 11/2+t)

(t) = 1 - e~' + e-(*±1/2) - e-1 = 1 - e-1 + (e 1/2 - 1)e-

For s t < 1:

= ft
t(t) e-' di = e- ( 1/2) - e­

- 1/2 )e

(d) Performing the periodic convolution graphically, we obtain the solution as

shown in Figure S4.10-1.

2 X1[n] *x2[n]

19

0 1 3 4 5

-16 (one period)

Figure S4.10-1
 Convolution / Solutions
S4-13

(e)

S4.11
(a) Since y(t) = x(t) *h (t) and x(t) = g(t) *y(t), then g(t) *h(t) = 6(t). But

g(t )*h (t) = grob t - kT hboTr - lT d,

wk=0 1=0

= T gkhlo(t - (1+ k)T)

k=O 1=0

Let n = 1 + k. Then = n - k and

g(t)* h (t) = Ygkhn t- n

n=0 \k=0

n F, n = 0,
k=0 n O
Therefore,
go = 1/h 0 ,
gi = -- hi/ho,
1 (-hl h2)
2 o ho ho

(b) We are given that h0 = 1, hi = I, hi = 0. So

= 1,
g1 = -i)
g2=+(1)2,
g2 = -()
Signals and Systems
S4-14

Therefore,

g(t) = ( (-) (tkT)

k=O

(c) (i) Each impulse is delayed by T and scaled by a, so

h(t) = (t -kT )
k=O

(ii) If 0 < a < 1, a bounded input produces a bounded output because

y(t) = x (t)* h (t),

| y(t)| < Zak 6(r ­ kT)x(t - r) di

k=0 -w

<
k=O
a f -w
_(T - kT)Ix(t - T)I dr

Let M = maxlx(t)|. Then

1
I y(t) < M ak= M , al < 1
k=O

If a > 1, a bounded input will no longer produce a bounded output. For

example, consider x(t) = u(t). Then
t
00
yt) = Ta f 6(,r - kT) dr
k=O -w

Since f 6(r - kD di = u (t-kT ),

y(t) = ( aku(t-kT)
k=O

Consider, for example, t equal to (or slightly greater than) NT:

N

y(NT) = Z ak
k=O

If a > 1, this grows without bound as N (or t) increases.

(iii) Now we want the inverse system. Recognize that we have actually solved
this in part (b) of this problem.
gi = 1,
g2 = -a
gi = 0, i # 0, 1
So the system appears as in Figure S4.11.

y(t) _0_ x(t)

Delay T

Figure S4.11
 Convolution / Solutions
S4-15

(d) If x[n] = 6[n], then y[n] = h[n]. If

x[n] = go[n] + iS[n-N],
then
y[n] = -h[n] + -h[n],
y[n] = h[n]

S4.12

(a) b[n] = #[n] - -4[n - 1],

x[n] = ( x[k][n - k] = ( x[k]Q([n - k] - -p[n - k - 1]),

k=-- k=--w

x[n] = E (x[k] - ix[k - 1])4[n- k]

k= -w

So ak = x[k] - lx[k - 1].

(b) If r[n] is the response to #[n], we can use superposition to note that if

x[n] = ( akp[n - k],

k=­

then

y[n] = Z akr[n - k]
k= -w

and, from part (a),

y[n] = ( (x [k] - fx[k - 1])r[n - k]

k= ­

(c) y[n] = i/[n] * x[n] * r[n] when

[n] = b[n] - in- 1]
and, from above,
3[n] 4[n] - -[- 1]
So
/[n] = #[n] - -#[n - 11 - 1(#[n - 11 - -$[n - 2]),
0[n] = *[n] - *[n - 1] + {1[n - 2]
(d) 4[n] - r[n],
#[n - 1] - r[n - 1],
b[n] = 4[n] - -1[n- 1] - r[n] - fr[n -1]
So
h[n] = r[n] - tr[n -1],
where h[n] is the impulse response. Also, from part (c) we know that
y[n] = Q[n] *x[n] *r[n]
and if x[n] = #[n] produces r[n], it is apparent that #[n] * 4[n] = 6[n].
MIT OpenCourseWare
http://ocw.mit.edu

Resource: Signals and Systems

Professor 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.
5 Properties of Linear,
Time-Invariant Systems
Solutions to
Recommended Problems
S5.1
The inverse system for a continuous-time accumulation (or integration) is a differ­
entiator. This can be verified because
d[ x(r) dr =x(t)

Therefore, the input-output relation for the inverse system in Figure S5.1 is
dy(t)
x(t ) = dt

y(t F h(t) x x(t)

Figure S5.1

S5.2
(a) We want to show that
h[n] - ah[n - 11 = b[n]
Substituting h[n] = anu[n], we have
anu[n] - aa" -u[n - 1] = a"(u[n] - u[n - 1])
But
u[n] - u[n - 1] = b[n] and an6[n] = ab[n] = b[n]
(b) (i) The system is not memoryless since h[n] # kb[n].
(ii) The system is causal since h[n] = 0 for n < 0.
(iii) The system is stable for Ia I < 1 since
1
|al" 1 - |a
is bounded.
(c) The system is not stable for |a l > 1 since E"o 1a|" is not finite.

S5.3
(a) Consider x(t) = 6(t) -+ y(t) = h(t). We want to verify that h(t) = e -2u(t), so
dy(t) - -2e- 2
u(t) + e -' 6(t), or
dt
dy(t) + 2y(t) =e -21 6(t),
dt

S5-1
Signals and Systems
S5-2

but ed' b(t) = 6(t) because both functions have the same effect on a test func­
tion within an integral. Therefore, the impulse response is verified to be correct.
(b) (i) The system is not memoryless since h(t) # kb(t).
(ii) The system is causal since h(t) = 0 for t < 0.
(iii) The system is stable since h(t) is absolutely integrable.

2t
IhtI dt = e dt = -le -2t

S5.4

By using the commutative property of convolution we can exchange the two systems
to yield the system in Figure S5.4.

u(t) - d/dt L y(t)

Figure S5.4

Now we note that the input to system L is

du(t)
= 6(t ),
dt
so y(t) is the impulse response of system L. From the original diagram,
ds(t)
dt = yt)
Therefore,

h(t) = ds(t)
dt

S5.5
(a) By definition, an inverse system cascaded with the original system is the iden­
tity system, which has an impulse response h(t) = 6(t). Therefore, if the cas­
caded system has an input of b(t), the output w(t) = h(t) = 6(t).
(b) Because the system is an identity system, an input of x(t) produces an output
w(t) = x(t).

Solutions to
Optional Problems
S5.6
(a) If y(t) = ayi(t) + by 2(t), we know that since system A is linear, x(t) = ax,(t)
+ bx 2 (t). Since the cascaded system is an identity system, the output w(t) =
ax 1(t) + bx 2 (t).
 Properties of Linear, Time-Invariant Systems / Solutions
S5-3

ay 1(t + by 2 (t) B axI(t) + bx 2(t)

Figure S5.6-1

(b) If y(t) = y 1 (t - r), then since system A is time-invariant, x(t) = x,(t - -) and
also w(t) = xi(t - r).

yI(t -r) B - xI(t- r)

Figure S5.6-2

(c) From the solutions to parts (a) and (b), we see that system B is linear and time-
invariant.

S5.7
(a) The following signals are obtained by addition and graphical convolution:
(x[n] + w[n]) * y[n] (see Figure S5.7-1)
x[n] * y[n] + w[n] * y[n] (see Figure S5.7-2)

x[n] +w[n]

yInI

1?
0 n -

0 1 -1 6
-1
(x[n] +w[n]) *y[n]
39

1 2
-l 0

-3 6
Figure S5.7-1
Signals and Systems
S5-4

x[n] *y[n]

2 2
1 0 n -0 1 9 n
-1 0 -1 0

-2

3 x[n] *y[n] +w[n] *y[n]

1 2
-I 0

-2<

-3
Figure S5.7-2

Therefore, the distributive property (x + w) * y = x * y + w * y is verified.

(b) Figure S5.7-3 shows the required convolutions and multiplications.

2 (x[n] *y[n])-w[n]

x[n] *(y[n] -w[n])

1 1 2
n 1 0 n
-1 0 0 3
-1 -1

-2

Figure S5.7-3

Note, therefore, that (x[n] * y[n]) - w[n] # x~n] * (y[n] - w[n]).

 Properties of Linear, Time-Invariant Systems / Solutions
S5-5

S5.8
Consider

y(t) = x(t) * h(t) =

J-
x(t - r)h(r)dr

= x(r)h(t - r) dr

(a) y'(t) = x'(t - r)h(r) d-r = x'(t) * h(t)

= x(r)h'(t - r) dr = x(t) * h'(t),

where the primes denote d/dt.

(b) y(t) = x(t) * h(t),
y(t) = x(t) * u 1(t) * ui(t) * h(t),

y(t) = x(r) dr * h'(t)

(c) y(t) = x(t) * h(t),

y(t) = x(t) * ui(t) * h(t) * u _ 1(t),

y(t) = x'(r) * h(r) dr

(d) y(t) = x(t) *h(t)

= x(t) * u(t) * h(t) * n 1(t),

y(t) = Xt) * h(r) dr

S5.9
(a) True.

f0 |h(t)| dt =
f T |h(t )|

(b) False. If h(t) = 3(t - to) for to > 0, then the inverse system impulse response
dt = oo

is b(t + to), which is noncausal.

(c) False. Suppose h[n] = u[n]. Then

Z
n= -oo
Ih[n]| = [
n= -oo
u[n] = c

(d) True, assuming h[n] is finite-amplitude.

0o L

n=-co
Zh[n]I = E
n=-K
Ih[n] = M (a number)

(e) False. h(t) = u(t) implies causality, but J u(t) dt = oo implies that the sys­
tem is not stable.
Signals and Systems
S5-6

(f) False.
hi(t) = 6(t - ti), ti > 0 Causal
h 2(t) = b(t + 2 ), t2 > 0 Noncausal
h(t) = hI(t) * h 2(t) = 6(t + t 2 - ti), t2 !- t 1 Causal
(g) False. Suppose h(t) = e-'u(t). Then

e - tu(t)dt = -e-' -1 Stable

0

The step response is

u(t - r)e - T u(r) d-r e -T dr

= (
(1 - e~')u(t),

(1 - e -') dt =t + e-'1 = oo

(h) True. We know that u[n] = E=O b[n - k] and, from superposition, s[n] =
Ef=0 h[n - k]. If s[n] # 0 for some n < 0, there exists some value of h[k] # 0
for some k < 0. If s[n] = 0 for all n < 0, h[k] = 0 for all k < 0.

S5.10

(a) J g(r)u1 (r) dr = -g'(0),

g(r) = x(t - r), t fixed,
- dg(r) _ _ dx(t - -r)
x(t - )u1(r) dr=
Jodr r=o dr t=0

dx(t - r) dx(t)
dt r=O dt

(b) - [gt)ft)]_
g(t)f(t)uj(t) dt dt =
= -[g'(t)ftt) ± g(t)f'(t)]__
- IgMAO+ Xt 1(l It =0
= -[g'(O)f(O) + g(0)f'(0)],

T g(t)[f(O)ui(t) - f'(0)S(t)] dt = -f(0)g'(0) - f'(0)g(0)

So when we use a test function g(t), f(t)ui(t) and f(O)ui(t) - f'(0)6(t) both
produce the same operational effect.

(c)
-- o
x(r)u 2(r) dr = x(r))
_Y6
- f _.dr
-u (r)dr
*~dx dx/ d2 X
-J -uir) dr = 7 ) + d2U0(r) dr
2
d x
dr2 r=0

(d) f g(r)fr)u (r) dr = g"(r)f(r) +

2 2g'(r)f'(r) + g(r)f"(r)
 Properties of Linear, Time-Invariant Systems / Solutions
S5-7

Noting that 2g'(7r)f'(r) ,= = - 2f'(O) fg(r)ui(r)dr, we have an equivalent oper­

ational definition:
f(r)u 2 (r) = f(O)u 2(r) - 2f'(0)ui(r) + f"(O)b(r)

S5.11
(a) h(t) * g(t) = J'. h(t - r)g(r) dr = f' h(t - )g(r) dr since h(t) = 0 for t < 0
and g(t) = 0 for t < 0. But if t < 0, this integral is obviously zero. Therefore,
the cascaded system is causal.
(b) By the definition of stability we know that for any bounded input to H, the out­
put of H is also bounded. This output is also the input to system G. Since the
input to G is bounded and G is stable, the output of G is bounded. Therefore, a
bounded input to the cascaded system produces a bounded output. Hence, this
system is stable.

S5.12
We have a total system response of
h = {[(h, * h 2 ) + (h 2 * h 2 ) - (h 2 * hl)] * hi + h-i} * hl
h = (h 2 * h) + (hi 1 * h 2 ')

S5.13
We are given that y[n] = x[n] * h[n].

y[n] = ( x[n - k]h[k]

k= -w

Iy[n]I = E
k=-Oo
x[n - k]h[k]

max {y[n]|} = max x[n - k]h[k]

k=-o

<E max {Ix[n- k]|}h[k]|

k= -o

=maxm{x n]l} z:
k= -o
lh[k]|

We can see from the inequality

max{Iy[n]|) max{|x[n]|} Z Ih[k]|

k= -o0

that E Ih[k]| 5 1 max {Iy[n]|} s max {Ix[n]|}. This means that E Ih[k]I
- 1 is a sufficient condition. It is necessary because some x[n] always exists that
yields y[n] = E=_ Ih[k]1. (x[n] consists of a sequence of +1's and -l's.) There­
fore, since max {x[n]} = 1, it is necessary that E= h[k] 1 5 1 to ensure that y[n]
S max {Ix[n]I} = 1.
MIT OpenCourseWare
http://ocw.mit.edu

Resource: Signals and Systems

Professor 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.
6 Systems Represented by
Differential and Difference
Equations
Solutions to
Recommended Problems
S6.1
We substitute ya(t) = ay1(t) + Oy2(t) into the homogeneous differential equation
dya(t)
dt + ay3(t)= dd [ay 1(t) + #y2(t)] + a[ayi(t) + #y 2(t)]
dt dt
Since differentiation is distributive, we can express the preceding equation as
ady t) +0dy2t) + aay1(t) + afly2(t)
ddty ddty
aa _ di(
dt + ay1(t) I + dt
dt + ay2(t)
ay()
However, since both y 1(t) and y 2(t) satisfy the homogeneous differential equation,
the right side of the equation is zero. Therefore,
dy3 (t) + ay3(t)
= 0
cit

S6.2
(a) We are assuming that y(t) = es'. Substituting in the differential equation yields
d2 d
d (es') + 3 (e t ) + 2e" = 0

so that
s2e"t + 3se' + 2e" = es'(s 2 + 3s + 2) = 0
For any finite s, e" is not zero. Therefore, s must satisfy
0 = s2 + 3s + 2 = (s + 1)(s + 2), s = -1, -2
(b) From the answer to part (a), we know that both yi(t) = e-' and y 2 (t) = e-2
satisfy the homogeneous LCCDE. Therefore,
ys(t) = Kie~' + K 2e 2,

for any constants K 1, K 2, will also satisfy the equation.

S6.3
(a) Assuming y(t) of the form
y(t) = Kest ,
we substitute into the LCCDE, setting x[n] = 0:
dy(t) 1 1 s
0= + y(t) =Ksel+K-7es = Ket±
dt 2 22

S6-1
Signals and Systems
S6-2

Since K # 0 and e" # 0, s must equal --. K then becomes arbitrary, so the
family of y(t) that satisfies the homogeneous equation is
y(t) =Ke-12
(b) Substituting into eq. (P6.3-1) y 1 (t) = Ae' for t > 0, we find
dy1 (t) 1 1 ~ t
t >0
dy I 1y t)= -Ae-'+ Ae'=e-'
Since e-' never equals zero, we can divide it out. This gives us an equation
for A,
A
asA = -2
2
(c) For y 1(t) = (2e -t/ 2 - 2e- t)u(t),

dyi(t) [2 (- -) e -1/2
- 2(-1)e -'], t > 0
dt t0, t <0,
dyi(t) 1 F(-e 1
t 2
+ 2e -t) + 1 (2e-' 2
- 2e- t) = e-', t> 0
dt 2 0, t < 0

= x(t)

S6.4
(a) Note that since y[n] is delayed by one sample by the delay element, we can label
the block diagram as shown in Figure S6.4.

x [n] - 1y [n[-1

_1
2

Figure S6.4
Thus y[n] = x[n] - 1],
- +y[n or y[n] + ly[n - 1] = x[n].
(b) Since the system is assumed to be causal, y[n] must be zero before a nonzero
input is applied. Therefore, x[n] = 0 for n < 0, and consequently y[n] must be
zero for n <0. Thus, y[-5] = 0.
(c) Since x[n] = [n] = 0 for n < 0, y[n] must also equal zero for n <0. For n = 0,
we have y[0] + ly[-1] = 1 or, substituting for y[n],
Ka~u[0] + Ka-lu[-1] = 1,
K + i- 0 = 1, or K=1
For n > 0, we have
y[n] + ly[n - 1] = 0 or an + ia"-1 = 0
 Systems Represented by Differential and Difference Equations / Solutions
S6-3

since K = 1. Thus, a must equal -- for a" + -a1 to equal 0 for all n > 0.
Therefore, y[n] = (-1)"u[n]. Substituting into the left side of the difference
equation, we have
(-2)" u[n] + (1)"-1 u[n-1] = (-2)" u[n] - (-2)" u[n-1]

n=0
1,
-0, otherwise
(d) We can successively calculate y[n] by noting that y[ -1] = 0 and that
y[n] = -1y[n - 1] + b[n]

n=0, y[0] = -- - 0 + 1 = 1
n = 1, y[1] = - 1 + 0 = -1
n = 2, y[2] = -i (-)+0 =i

We see that these correspond to the answer to part (c).

S6.5
(a) Performing the manipulations in inverse order to that done in the lecture (see
Figure S6.5-1) yields the system shown in Figure S6.5-2.

x[r - [n] y [n]

D
11
3
Figure S6.5-1

x [n] y [n]

-l__
_ - - - - --- L _ -­
A B
Figure S6.5-2
Signals and Systems
S6-4

Since the system is linear and time-invariant, we can exchange the order of the
two boxes A and B, yielding the direct form I shown in Figure S6.5-3.

x [n] + y [n]

D | |D

x [n -1 -2 1y[n-1

B A
Figure S6.5-3

(b) From the direct form I, we see that the intermediate variable q[n] is related to
x[n] by
q[n] = x[n] - 2x[n - 1]
The signal y[n] can be described in terms of q[n] and y[n - 11 as
y[n] = q[n] + iy[n - 1]
Combining the two equations yields
y[n] = iy[n - 1] + x[n] - 2x[n - 11, or
y[n] - iy[n - 11 = x[n] - 2x[n - 1]
(c) (i) Figure S6.5-4 shows that if we concentrate on the right half of the dia­
gram of direct form II given in Figure P6.5, we see the relation
y~n] = r[n] - 2r[n - 11

r [n] y[n]

r[n- 1] -2

Figure S6.5-4

(ii) Similarly, Figure S6.5-5 shows that if we concentrate on the first half of
the diagram, we obtain the relation
r[n] = x[n] + ir[n - 1], or x[n] = r[n] - ir[n - 1]
 Systems Represented by Differential and Difference Equations / Solutions
S6-5

r [n]
x [n] - +

r[n- 1]

Figure S6.5-5

(iii) From the two equations obtained in parts (i) and (ii),
x[n] = r[n] - ir[n - 11 (S6.5-1)
and
y[n] = r[n] - 2r[n - 1], (S6.5-2)
we solve for r[n],obtaining
r[n] = fx[nJ - ky[n]
Substituting r[n] into eq. (S6.5-1), we have
x[n] = ,x[n] - y[n] - }{5xfn - 1] - iy[n - 1]},
which simplifies to
y[n] - iy[n - 1] = x[n] - 2x[n - 1]

S6.6
(a) Integrating both sides of eq. (P6.6-1) yields
y(t) + afy(t) dt = bx(t) + c fx(t) dt, or
y(t) = -a fy(t) dt + bx(t) + c fx(t) dt
Thus, we set up the direct form I in Figure S6.6-1.

x(t) P. y(t)
+ +

C -a

Figure S6.6-1
Signals and Systems
S6-6

(b) Since we are told that the system is linear and time-invariant, we can inter­
change boxes A and B, as shown in Figure S6.6-2.

-'-4­
b

-a c

L-_- -- ___._­
B
Figure S6.6-2

Combining the two integrators yields the final answer, shown in Figure S6.6-3.

x(t) b + -1y t)

-a C

Figure S6.6-3

Solutions to
Optional Problems
S6.7

(a) In Figure S6.7 we convert the block diagram from Figure P6.7 to direct form I.

x [n] + q [n] + P y [n]

D D

-4
Figure S6.7
 Systems Represented by Differential and Difference Equations / Solutions
S6-7

q[n] is given by
q[n] = x[n] + x[n - 1]
while
y[n] = q[n] - 4y[n - 1]
Substituting for q[n] yields
y[n] + 4y[n - 11 = x[n] + x[n - 1]
(b) The relation between x[n] and r[n] is r[n] = -4r[n 1] + x[n]. For such a -
simple equation, we solve it recursively when a[n] = x[n].
n b[n] r[n - 1] r[n]
<0 0 0 0
0 1 0 1
1 0 1 -4
2 0 -4 16
3 0 16 -64
We see that r[n] = (-4)"u[n].
(c) y[n] is related to r[n] by
y[n] = r[n] + r[n - 1]
Now y[n] = h[n], the impulse response, when x[n] = 6[n], and
h[n] = (-4)"u[n] + (-4)"-Iu[n - 1]
This expression for h[n] can be further simplified:
h[n] = (-4)"u[n] + (-4)"-Iu[n - 1]
or

0 0'
h[n] =
1, n =0

For n > 0,
h[n] = (-4)" + (-4)"-1
-3(-4)"-1

Thus,
h[n] = b[n] - 3(-4)"1 u[n - 1]

S6.8

Note that the system in Figure P6.8 is not in any standard form. Relating r(t) to x(t)
first, we have

f a[x(t) + r(t)] dt = r(t), or (S6.8-1)

dr(t) - ar(t) =
dt
represented in the system shown in Figure S6.8.
Signals and Systems
S6-8

x(t) r(t)

x(t)+r(t) f

Figure S6.8

The signal y(t) is related to r(t) as follows:

r(t) + bf r(t) dt = y(t), or (S6.8-2)

dr(t) dy(t)
+ br(t) =
dt dt
Solving for dr(t)/dt in eqs. (S6.8-1) and (S6.8-2) and equating, we obtain
dy(t)
ar(t) + ax(t) = -br(t) + dt
dt
Therefore,
-a 1 dy(t) (S6.8-3)
r(t) = a~x(t) +a bd86-3
a +b a +b dt
We now substitute eq. (S6.8-3) into eq. (S6.8-1) (or eq. S6.8-2), which, after simpli­
fication, yields
dy 2(t) dy(t) dx(t)
dt =dta d ±t+abx(t)
dt 2 dt-a

S6.9
(a) Substituting y[n] = Az" into the homogeneous LCCDE, we have
Az" - -Azo ' = 0

Dividing by Azo" yields

zo -i= 0, or z=
(b) For the moment, assume that the input is I[n] = Kej"0"u[n] and the resulting
output is P[n] = Yei"Ou[n]. Thus,
P4n] - iPn - 11 = t[n]
Substituting for P[n] and .{n] yields
YeJ" - eKe"0" for n > 1
Dividing by e "o", we get
Y e-j" - Y = K
 Systems Represented by Differential and Difference Equations / Solutions
S6-9

Thus
K K
Y= 1 =\/ - cos 0 e
1
j tan- 1(sin no)/(2-cos go)1 or

y n e -j tan-n](sin e)e(2"cos no)­

cos 0 \/ o
KK
Therefore,

0
y[n] = Re[Ye' u[n]] = a_ Ko~ KK 'On
Retei tan l[(sin Oo)/(2-o
CoSjU~f

K
= B cos( 0 n + 0), where B = ­

sin g0
0 = -tan-'( o o

S6.10
The important observation to make is that if [dtr(t)]/dti is the input to the system
H, then [dis(t)]/dt' will be the output. Suppose that we construct a signal
M d'r(t)
q(t) = (a
i= 1

The response of H to the excitation q(t) is

M
p(t) = ( a dit
i=1

However, q(t) = 0 for all t. Therefore, p(t) = 0 for all t. Thus,

M
ad's(t) =0
aidt

S6.11
(a) Substituting y(t) = Aeso' into the homogeneous LCCDE, we have
N dky(t) N dk
ak d =(a k(Ae)=0
k=0 k=0

= aks Aet =0
\k= 0)/

Since A # 0 and esot # 0, we get

N

p(so) = ( aks = 0
k=0

(b) Here we need to use a rather subtle trick. Note that

d
Ate" = -- (Ae')
ds
Signals and Systems
S6-10

Using this alternative form for Ate", we obtain

N dk/d~ [N Ak 1
t dt (Aest
ak dt d Ae = aI
k=O d cis cis k=

d dp(s)
= - [p(s)Ae"] = Atp(s) + A est
ds ds
For s = so, p(so) = 0. Also, since p(s) is of the form
p(s) = (s - so) 2q(s),
we have
dp(s) 0
ds =S=
Therefore, AteO' satisfies the homogeneous LCCDE.
(c) Substituting y(t) = est , we get the characteristic equation
s 2 + 2s + 1 = 0, or sO = -1
Thus, y(t) = Ke~' + K 2te~'. For y(0) = 1 and y'(0) = 1, we need K, = 1 and
K 2 - Ki = 1, or K2 = 2. Thus,
y(t) = e~' + 2te­
MIT OpenCourseWare
http://ocw.mit.edu

Resource: Signals and Systems

Professor 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.
7 Continuous-Time Fourier Series
Solutions to
Recommended Problems
S7.1
(a) For the LTI system indicated in Figure S7.1, the output y(t) is expressed as

y(t) =f h(r)x(t - r) dr,

where h(t) is the impulse response and x(t) is the input.

LTI

x(t) ON h(t) y (t)

Figure S7.1

For x(t) = ew',

y(t) = f h()ew(- T) dr

= ejwt { h(r)e-j' dr
= e+iwtH(w)
(b) We are given that the first-order differential equation is of the form
dy(t) + ay(t) = x(t)
dt
From part (a), when x(t) = eiw', then y(t) = ej"'H(w). Also, by differentiating
y(t), we have
dy(t) = jcoej"'H(co)
dt)

Substituting, we get
jwewtH(w) + aejw'H(w) ewt
Hence,
jwH(o) + aH(w) = 1, or
1
H(w) =
a + jw.

S7.2

(a) The output of a discrete-time LTI system is given by the discrete-time convo­
lution sum
y[n] = ( h[k]x[n - k]
k

S7-1
Signals and Systems
S7-2

If x[n] = z", then

y[n] = ( h[k]z"-k
k
= z" h[k]z -k
k
= z"H(z)
(b) We are given that the first-order difference equation is of the form
y[n] + ay[n - 1] = x[n]
From part (a), if x[n] = z", then y[n] = z"H(z). Hence,
y[n - 1] = z" -1H(z).
By substitution,
z"H(z) + az"- H(z) =z"
which implies
(1 + az-')H(z) = 1,
1
H(z) = _
1 + az-1

S7.3

(a) x(t) = sin (lort + 6

= j2xt5 _ -j21t5
2j 2j
We choose wo, the fundamental frequency, to be 21r.
x(t) = a eik-ot,
k
where
ejr/6 __e -jr/6
a5 =a_5
2j 2j
Otherwise ak = 0.
(b) x(t) = 1 + cos(27rt)
ej2xt e-j2wt
=1+ +ejit
2 2
2
For wo = 7r, a_ = ai = -, and ao = 1. All other ak's = 0.

(c) x(t) = [1 + cos(27rt)] [sin(lort +

= sin(10it + + cos(2irt)sin( 10irt +

= e /6 ej2rt5 e 2 e -j2t5) j2+rt + le -j2+ t eji( 6 e 2 t5

r __e 2* e -j27rt5

2
-_ . _}15_ e -j2xts + ej t6 _ e-j2xt4
2j 2j 4j 4j
+ e j21rt4 ee -j2t6

4j 4j
 Continuous-Time Fourier Series / Solutions
S7-3

Therefore,
x(t) = ( ake*4'

2
where wo = 1r.
ejr/6
a4 =- . a-4
43j 4j.
/6
e jr/6 _ e-ir
a5 = -, a-5, - I
2j 2j
e jr/6 _e-jr16
a6 - a_6 = .
4j 4j
All other ak's = 0.

S7.4
(a)

Note that the period is To = 6. Fourier coefficients are given by

ak = - x(t)e -jk" 0 t dt

We take wo = 27r/To = r/3. Choosing the period of integration as -3 to 3, we

have
1 1 F2
ak = e jk( /3)t - 1 -jk(r/3)t dt
6 2 6 f1
_ _ 1 e -jk(r/3)t 1 1 -jk(ir/3)t 2

6 -jk(7r/3) 2 6 -jk(7r/3) 1

1 [e +j(7/3)k _ e +j(2r/3)k - e-j(21/3)k + e j(-/3)kI

-j27rk
cos(2ir/3)k cos(7r/3)k
jrk jrk
Therefore,
7r
x(t) = akejkwot
3
k

and
cos(27r/3)k - cos(r/3)k
ak = jrk
Signals and Systems
S7-4

Note that ao = 0, as can be determined either by applying L'H6pital's rule or by

noting that

ao = (1|T0 )
f x(t)dt.
(b)
x(t)

Figure S7.4-2

The period is To = 2, with wo = 2-x/2 = r. The Fourier coefficients are

ak = - fTO x(t)e -iko'dt

To o
Choosing the period of integration as -i2 t o -2, we have
a 3/2
k= 2 x(t)e ikot
-­ dt
-1/2

3/2
[6(t) 26(t - 1)]eikO' dt
2 1/2 -

1 . 1.
e-k-o = - (e-i-)k
2 2
Therefore,
a o - 2, ak = - ()k
It is instructive to plot ak, which we have done in Figure S7.4-3.
 Continuous-Time Fourier Series / Solutions
S7-5

S7.5
(a) (i) and (ii)
From Problem 4.12 of the text (page 260), we have

x {t - T)= -x(t),

which means odd harmonics. Since x(t) is real and even, the waveform has real
coefficients.
(b) (i) and (iii)

-x(t) = x t ­

which means odd harmonics. Since x(t) is real and odd, the waveform has imag­
inary coefficients.
(c) (i)
W~ T)
-x(t) = x t 4 )'
which means odd harmonics. Also, x(t) is neither even nor odd.

Solutions to
Optional Problems
S7.6
x(t) is specified in the interval 0 < t < T/4, as shown in Figure S7.6-1.

x(t)

| t
T T
8 4

Figure S7.6-1

(a) Since x(t) is even, we can extend Figure S7.6-1 as indicated in Figure S7.6-2.

x (t)

t
_T _T T T
4 8 8 4
Figure S7.6-2
Signals and Systems
S7-6

Since x(t) has only odd harmonics, it must have the property that x(t - T/2)
= -x(t), as shown in Figure S7.6-3.

x(t) in
x(t - T) = -x(t)
2 Figure S7.6-2 shifted

T T T 37
4 4 2 4

-x(t) in
Figure S7.6-2
/
Figure S7.6-3

So we have x(t) as in Figure S7.6-4.

x(t)

_T 0 T T 3T T
4 4 2 4

-1-­

Figure S7.6-4

(b) In the interval from t = 0 to t = T/4, x(t) is given as in Figure S7.6-5.

 Continuous-Time Fourier Series / Solutions
S7-7

Since x(t) is odd, for - T/4 < t < T/4 it must be as indicated in Figure S7.6-6.

x(t)

T T T
4 8 4

Figure S7.6-6

Since x(t) has odd harmonics, x[t - (T/2)] = -x(t). Consequently x(t) is as
shown in Figure S7.6-7.

x(t)

T T T
8 4
-1

Figure S7.6-7

S7.7
1
ak = - x(t)e -jkw 0t dt
TO TO

(a) alk = - x(t - to)e -jk" 0 t dt

T0 fTO

Substituting r = t - to, we obtain

1F
ak = - I x(r)eikwor dr - e -jkoto
OTO
= ake -jk"0'0o
1
(b) ak = -
T0Of17O x(- t)e jkwot dt

Substituting r = - t, we have
1
ak = - = a-k
T0OfT 0 x(T)e kwOdr
Signals and Systems
S7-8

x*(t)e -jk 0t dt
50TO 0
(c) = -

= T0 x(t)ekwot dt = ak,

6k = ak

(d) a(d) = fi x(at)eik( 2 ra/T)t dt

0 To la
Let r = at. Then

ak = - x(r)e -jk(2"/To)' dr = ak
TO To

Therefore,

S7.8
(a) Since O4(t) are eigenfunctions and the system is linear, the output is

y(t) = L
k=-00
XkCk'k(t).

2
d 2x(t)
2
dx(t)
(b) y(t)=t dt +t
dt2 dt
kk(t) = tk,
d0k(t) ktk1_
dt
d2dtk(t) "k-Jt2 '
2 ktk 1)tk-2
dt = -

So if k(t) = x(t), then

y(t) = t 2k(k -
2
1)tk- + tktk I
= k(k - 1)tk + ktk
= kotk = k 4k(t)

The eigenfunction 4k(t) has eigenvalue Xk = k2

S7.9
(a) 9(t) = 2 1(t) * 2 2(t)

= fT I 1(r)A2 (t - r) dr

The Fourier coefficients for P(t) are given by

Ck = -- I 1(r) 2 (t - r) dre -jk(2wTot

)t
50TO TOL

= --
1
J 1 (r)e -<k(2./T 0)Tr
F
- r)e jk(2r T 0)(t-) dt
Toakbk
 Continuous-Time Fourier Series / Solutions
S7-9

(b) Since z(t) * z(t) =x(t), as shown in Figure S7.9-1, then 2(t) is shown in Figure
S7.9-2.

x ( t)

2
z(t) z(t)

t t t
-1 1 -1 1 -2 2

Figure S7.9-1

In Figure S7.9-2, To = 5. Hence,

T(t ) - Toz' = 4 sinc 5

(c) Without explicitly carrying out the convolutions, we can argue that the aperi­
odic convolution of x 1(t) and x 2 (t) will be symmetric about the origin and is
nonzero from t = -2T to t = 2T. Now, if 1 1(t) and 2 2(t) are periodic with period
To, then the periodic convolution, P(t), will be periodic with period To. If To is
large enough, then q(t) is the periodic version of y(t) with period To. Hence, to
recover y(t) from P(t) we should extract only one period of P(t) from t =
-TO/2 to t = TO/2 and set y(t) = 0 for ItI > TO/2, where TO/2 2T, or To -
4T.

S7.10
(a) The approximation is
N

N(t)= akk(t
k=-N

with the corresponding error signal

eN(t) = x(t) ­ Nt)
N
= x(t) - ( akk(t)
k=-N
Signals and Systems
S7-10

Hence,

|eN(t) I = x(t) - kk x*() a*t)

= |x(t)| 2 - ( akx(t)4k(t) -
k
z k
akx(04k() + ( (I
k
aka14k(M 1()

If we integrate, f eN(t) 12 dt, and use the property that

k = 1,
a 1,
otherwise,
we get b=

E = fb Ix(t) I'dt - Za* x(t)kp(t) cit

a k

- T ak X*(tk(t) dt + T |akI
k a k

Since ai = bi + jc,

=E r OC x*(t)Oi(t) dt + 2bi
an d x(t)k7(t) dt ­

and

-_ = j x(t)(t )dt - j f x*(t)i(t) dt + 2c,

aci
Setting

-= 0 and = 0,

we can multiply the second equation byj and add the two equations to get
(E aE
ab + - = 0

By substitution, we get
bi + jci = ax(t)O(t) dt
= a,

(b) If {4(t)} are orthogonal but not orthonormal, then the only thing that changes
from the result of part (a) is

Tab a
k1
qk*~c(t) dt Z Iak IAkk 2 k
ab

It is easy to see that we will now get

a = fbx(t)4it) dt
 Continuous-Time Fourier Series / Solutions
S7-11

(c) Since

fTo+a einwot e-jnwot dt =

for all values of a, using parts (a) and (b) we can write

ai = a x(t)e-"nwo' dt

_ 1
=- x(t)e-nwot dt
T0 fTo
MIT OpenCourseWare
http://ocw.mit.edu

Resource: Signals and Systems

Professor 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.
8 Continuous-Time
Fourier Transform
Solutions to
Recommended Problems
S8.1
(a)
x(t)

t
Tj Tj
2 2

Figure S8.1-1

Note that the total width is T,.

(b)

i(t)

t
3T 1 -- T1 To T1 T1 To Tl 3
T 1 =O
2 2 2 2 2 2

Figure S8.1-2

(c) Using the definition of the Fourier transform, we have

X(w) = ox(t)e -j dt =
Ti/2 le-j" dt since x(t) = 0 for ItI>

Til 12 sin wTi

2
= e - (e -jwTI1/2 _ eT1/ ) = 2
JW -T1/2 (e

See Figure S8.1-3.

S8-1
Signals and Systems
S8-2

(d) Using the analysis formula, we have

ak = T f X(t)e ~jk 0
t dt,

where we integrate over any period.

_ 1
ak -T
f
-T
To12 (t)e -jk(2w/T)t dt = TO f -T
T1 /2
1
/2
e -jk(2/T O)t dt
1
To

sin kr(T1 /T 0 ) _ sin ,r(2k/3)

ak = (e jkirToT 0 -- eikir1/TO) =
TO ,rk 7rk
-jk 2,

Note that ak = 0 whenever (27rk)/3 = irm for m a nonzero integer.

(e) Substituting (21rk)/To for o, we obtain
1 2 sin(7rkT1 /T 0 ) sin rk(T 1 /T 0 )
T_1X( =2)T
= ak
TO I"=(21rk)/ To To 2,rk/To ,rk
(f) From the result of part (e), we sample the Fourier transform of x( t), X(w), at
w = 2irk/To and then scale by 1/To to get ak.
 Continuous-Time Fourier Transform / Solutions
S8-3

S8.2

(a) X(w) = fx(t)e -j4t dt = (t - 5)e -j' dt = e ~j = cos 5w - j sin 5w,

by the sifting property of the unit impulse.

IX(w)| = |ej 5wI = 1 for all w,

<tX(w) = tan-' LIM{X(xc)}

Re{X(co)}
tan -sin5wo)
cos 5w
= -5w

(b) X(w) = e -a t u(t)e -i"' dt = e -ate -j't dt

0 - (a+jw)t dt -1 e- (a+jo)t

a+jw o

Since Re{a} > 0, e -at goes to zero as t goes to infinity. Therefore,

-1 1
X(W) = 1(0 - 1) =
a + jw a +jw
2 1
IX(W)I = [X(W)X*(w)]l/ = 2
+o2
a +ja -jw)] \/ >

X(w) + X*(O) a
Re{X(w)} = 2 a +
2
(2 i

= X(W) - X*(W) -W
Im{X(W)} 2 a2 + w 2 '>

Im{X(W)) _ w
-X(O) = tan- = -tan a

The magnitude and angle of X(w) are shown in Figure S8.2-2.

Signals and Systems
S8-4

IX(w)I

-a a

4X(W)
2
Tro

-aa

2T

Figure S8.2-2

(c) X(o) = e 1+12)tU(t )e -''' dt = e(-1+ 2)te -jct dt

1 0
[ 1-+j(2-)t
1 + j(2 - )0

Since Re{-1 + j(2 - w)} < 0, lim, e[- 1+j(2-w) = 0. Therefore,

1
1 + j(o - 2)
1
IX(wO)| = [X(x)X*(W)]11 2 = + (ow-
2
\/1 2)

X(w) + X*(W) 1
Re{X()} =
2 1 + (w - 2)2
X(w) - X*(co) -(o - 2)
Im{X(W)} =
+ ((A -2)2
2 1

X() = tan-' ImX(W) -tan-'(w - 2)

LRe{X(w)}
The magnitude and angle of X(w) are shown in Figure S8.2-3.
 Continuous-Time Fourier Transform / Solutions
S8-5

X(w)I

1 2 3

4X(W)

Figure S8.2-3

Note that there is no symmetry about w = 0 since x(t) is not real.

S8.3

(a) X3(w) = X3(t)e -' dt

Substituting for x 3(t), we obtain

X 3(() = J [ax 1 (t) + bx 2(t0)e 1-'dt

=f ax 1(t)e -wt dt + bx 2 (t)e -jwt dt

= af x 1(t)e -j dt + b f x 2(t)e -- ' dt = aX1 (w) + bX 2 (w)

(b) Recall the sifting property of the unit impulse function:

f h(t)b(t - to) dt = h(t o)

Therefore,

21rb(w - wo)ej-' do = 2rewot

Signals and Systems
S8-6

Thus,

2
1- 7rb(w - wo)ei-' dw = ei-o'
2,r -"oo

Note that the integral relating 2-7rS(w - wo) and e-o' is exactly of the form

x(t) = - X(w)ej- t dw,

where x(t) = e jot and X(w) = 2-7rb(co - wo). Thus, we can think of ei-o' as the
inverse Fourier transform of 2 7rb(w - wo). Therefore, 27rb(o - wo) is the Fourier
transform of eiwo'.
(c) Using the result of part (a), we have

X(M) = 7{x(t )} = 51 1 ake 2r j =

T, a, 5T {eik( 2,rs/t
From part (b),
-21rk)
5{eik(2,rT)t = 2,irb ( W

Therefore,

Z(w) = 27rab (W
k = -00

(d)
X(w)

2 sin 3 2 sin 2

To

3 3

Figure S8.3

S8.4

(a) We see that the new transform is

Xa(f) = X(w)
W=2,wf

We know that

x(t) =- f X(w)e'j t do
 Continuous-Time Fourier Transform / Solutions
S8-7

Let w = 2irf. Then dw = 21rdf, and

1
x(t) = X(21rf)ej 2"2ir df = Xa(f )ei2 ,"f df
27 f= -o

Thus, there is no factor of 21r in the inverse relation.

IXa(f)I
TI

2 1 1 2
TI TI TF T,

Figure S8.4

(b) Comparing

Xb(V) = -1 x(t)e jv' dt and X(w) = x(t -j~" dt,

we see that

Xb(V) = X(W) or X(w) = \7 X(w)

V2_
7r -V

The inverse transform relation for X(w) is

1 *0 j
x(t) - X(w)e dw = 1 X2r(w)ej- dw

=1
Xb(v)e jvt dv,

where we have substituted v for w. Thus, the factor of 1/2-x has been distributed
among the forward and inverse transforms.

S8.5
(a) By inspection, To = 6.
2
(b) ak = 1 TO )e -k( /To)t dt
TO J T0
We integrate from -3 to 3:
1 1 1
ak= 2 (ti + 1)+ 6(t) + -6(t
2 - + c 6
J-

= 1 +l 1 + 'je (1
-Jlk6.. Cos 2k
Signals and Systems
S8-8

(t) = 6( + Cos 27rk eik(2 / 6 )t

(c) (i) X 1(w) = x 1 (t)e -"' dt - [6(t + 1) + b(t) + -1(t - 1)]e -j't dt

- ie * + 1 + ie -"1 + cos w

(ii) X 2(w) = x 2(t)e -''' dt = [3( t) + it(t - 1) + -3(t - 5)]e - dt

= 1 + ie -j + le -i5 w

(d) We see that by periodically repeating x 1 ( t) with period Ti = 6, we get t(t), as

shown in Figure S8.5-1.

x 1 (t)

2
t
-1 0 1
x 1 (t - 6)

0 5 6 7

xl(t + 6)

2
-7
t
-6
t
-5
|tt)
0
t

x(t)

-7 -6 -5 -1 0 1 5 6 7
Figure S8.5-1

Similarly, we can periodically repeat x 2(t) to get 1(t). Thus T2 = 6. See Figure
S8.5-2.
 Continuous-Time Fourier Transform / Solutions
S8-9

x 2 (t)

tilt
t

01

X2 (t -6)

2
t
t 0

t2~ x 2 (t ± 6)
6
1 7
1
2
t

-6-5 0 7

X(t)

-6 -5
IL -1 0
t2 1
t It
5 6 7
t
Figure S8.5-2

(e) Since I(t) is a periodic repetition of x 1 (t) or X 2 (0), the Fourier series coefficients
of t(t) should be expressible as scaled samples of X,(co). Evaluate X 1 ( 0 ) at w=
21rk/6. Then
6
co-= k*k2rk 1 (27rk\

Similarly, we can get ak as a scaled sample of X 2(w). Consider X 2(21rk/6):

X2 - 1 5e +
j2k/6 e
X1(x) =irk1
+2o =6,rk n X x
62 2
2 6
But e = e j(lOrk/62rk)
njlOrk6 e rkI . Thus,

6
X2 = 1 + cos - = a.
A X6
Although X,(w) # X 2(w), they are equal for w =27rk/6.
Signals and Systems
S8-10

S8.6

(a) By inspection,
7 1
e -atu(t) 9 .
a + jw
Thus,
7I 1
e- 7tu(t) 1
7 + jw
Direct inversion using the inverse Fourier transform formula is very difficult.
(b) Xb(w) = 26(w + 7) + 26(o - 7),

Xb(t) = - Xb(w)ej do = - 2
[6(w + 7) + 6(w - 7)]ei-' dw
1 1 . 2
e-­
It + e = cos 7t

(c) From Example 4.8 of the text (page 191), we see that
37 2a
e alti 9 _2a
a 2 + W2
However, note that

ax(t) aX(w)
since

ax(t)e -j" dt = a x(t)e -jw dt = aX(w)

Thus,
1 - 1 1 7 1
-e -" l or - e -31t
2a a2 + (02 9 + .2 6
(d) Xa(w)Xb(W) = Xa(w)[26(o + 7) + 26(w - 7)]
= 2X(-7)(co + 7) + 2Xa(7)6(w - 7)
2 2
XdCO)= - (w +7) + b(o7)
7 - j7 7 +j7
1f*[2 ±7+ 2
b(co - 7) ej" do
xaOt = f- .7 6(o + 7) +
27r-- 7 -37 7 +j7
1 1 . 1 1 .
XA(t) = - e ­ 7' + - ej_ '
ir7-j7 xr7+j7
Note that

=_ 1=( ,
1 -jr/4 1 +jir/4
7 +j7 7 2 7 -j7 7 2
Thus

= - [e -j(7t + ei( 7 t-/4)] = -- cos 7t - -)

r7 -
XA(t) - -/4)
2 77r 4
 Continuous-Time Fourier Transform / Solutions
S8-11

we ~j3 w, 0 5 w < 1,
3
(e) Xe(w) = we -j w, -1 ! (O 0,

t0, elsewhere,
x(t) = f X(w)e-' dw = i [J' we -jamei' do - we ~i3-ejwl dw

Note that
e ax
f dx
xe" = , -(ax
ax -1)

Substituting a = j(t - 3) into the integrals, we obtain

x(t) = -
21r
1
L(
(j(t
j(t - 3)w

- 3))
(j(t - 3)w - 1)
I

0
0
egj(t -

(j(t - 3))­ 2
3)w
(j )) (j(t - 3)w - 1)
0

J
which can be simplified to yield
1 [ cos (t - 3) - 1 sin (t - 3)
x(t) = 7r (t - 3)2 + (t - 3)_

Solutions to
Optional Problems
S8.7

(a) Y(w) = y(t)e -j dt x(r)h( t - -) dr e -j'' dt

S ft x(,r)h(t - 1)e -j"' dr dt

(b) Let r t - T and integrate for all r and r. Then

Y(w) = TT=- .fro x(r)h(r)e -jw( r-+*T) dr dr

= x(r)e -' d-r {r=-0 h(r)e ~'j dr

= X(w)H(w)

S8.8
(a) Using the analysis equation, we obtain
1 rT/2_ 1
Ilk = I T/ e(t)e -jk(21r/T) t dt-T
T -T/2 T
Thus all the Fourier series coefficients are equal to 1/T.
(b) For periodic signals, the Fourier transform can be calculated from ak as

X(w) = 21 ak w-T
k=-00
Signals and Systems
S8-12

In this case,
21 2rk
PM=T 6 T

P(wj)
21Tr
T

_47 _ 2n 0 2_ 4
T T T T

Figure S8.8

(c) We are required to show that

2(t) = x(t )* p(t)

Substituting for p(t), we have

x(t)* p(t) = xt * kT)

k Ot ­

Using the associative property of convolution, we obtain

x(t) * p(t) = T, [x(t) * b(t - kT)]

k= -o

From the sifting property of b(t), it follows that

x(t)* pX)O x(t - kT) =t(t)

k=­

Thus, x(t) * p(t) is a periodic repetition of x(t) with period T.

(d) From Problem P8.7, we have
X(w) = X()P(x)
27rk
= X(-) - -T­
k=

2xk
=jI -- X(w)5(c
Since each summation term is nonzero only at w = 27rk/T,

*0 27r 2rk 2xk

i = y=) T X( T B

From this expression we see that the Fourier series coefficients of t(t) are

ak = - X ,
Ti T
which is consistent with our previous discussions.
MIT OpenCourseWare
http://ocw.mit.edu

Resource: Signals and Systems

Professor 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.
9 Fourier Transform Properties
Solutions to
Recommended Problems
S9.1
The Fourier transform of x(t) is

X(w) = x(t)e -jw dt = fe-t/2 u(t)e dt (S9.1-1)

Since u(t) = 0 for t < 0, eq. (S9.1-1) can be rewritten as

X(w) = e-(/ 2+w)t dt

+2
1 + j2w
It is convenient to write X(o) in terms of its real and imaginary parts:
X(w) 2 1-j2 2 -j4w
1 + j2w 1 -j2wJ 1 + 4W2
2 . 4w
1 + 4W 2
1 + 4W2
2
Magnitude of X(w) = /V1 ++4w 2

X(w) = tan-'(-2w) = -tan-' (2w)

+2 -4w
Re{X()} = 1+4w , 2
Im{X(w)} = 1 +4W 2

(a)
IX(w)I
2

Figure S9.1-1

(b)
4X(O)

1T

____________________4

21
Figure S9.1-2

S9-1
Signals and Systems
S9-2

(c)
RelX(w)t

Figure S9.1-3

(d)
Im X(o)

Figure S9.1-4

S9.2

(a) The magnitude of X(w) is given by

IX~)|I = /X2(w) + XI(W),
where XR(w) is the real part of X(w) and X(w) is the imaginary part of X(w). It
follows that
IwI < W,
|X(W)| =
IwI >W

IX(co)|

w i W
Figure S9.2-1

The phase of X(w) is given by

<X(w) = tan-' X
'\XRGA))/
=wtan-'(1), IWI <W
 Fourier Transform Properties / Solutions
S9-3

4X(w)
iT

| Co

-W W
Figure S9.2-2

(b) +
1M j wIl<W
X(- C) = 0,L l otherwise
I0

0,
1 + j, otherwise
|w l < W

0, otherwise

Hence, the signal is not real.

S9.3
For x(t) to be real-valued, X(w) is conjugate symmetric:
X(-w) = X*(W)
(a) X(w) = X(w)|eiX(W)
= IX(w) Icos(4IX(w)) + j IX(w) Isin(4X(x))
Therefore,
X(-o) = |X(-w)|cos(4X(-w)) + j|X(-w)|sin(4X(-w))
= |X(w) Icos(4X(w)) - j IX(w) Isin(4X(w))
= X*(W)

Hence, x(t) is real-valued.

(b) X(w) = XR(w) + jXw)
X(-w) = XR(-W) +jX,(-W)
= XR(w) + j[ -X(w) + 2,r] for w > 0
X*(w) = XR(w) - jX 1(
Therefore,
X*(w) X(-W)
Hence, x(t) is not real-valued.

S9.4

(a) (i) X(,w) = x(t)e i* dt

We take the complex conjugate of both sides to get

X*(w) = f-x*(t)ej-' dt
Signals and Systems
S9-4

Since x(t) is real-valued,

X*(W) = x(t)ei-t dt

Therefore,

X*(-o)= f x(t)e - dt

= X(W)

(ii) x(t) = - X(w)ej" dw

Taking the complex conjugate of both sides, we have

x*(t) = - X*(w)e -w"dw

Therefore,

x*(-) =0 X*(w)ej"" do

Since x(t) = x*(-t), we have

- X(co)ei" dw = -f X*(w)ejw" dw

This shows that X(w) must be real-valued.

(b) (i) Since x(t) is real, X(w) = X*(-w). Since x(t) is real and even, it satisfies
x(t) = x*(-t) and, therefore, X(w) is real. Hence, X(w) = X*(-w) =
X(-w). It follows that X(o) is real and even.
(ii) If x(t) is real, X(w) = X*(-w). Since x(t) is real and odd, x(t) =
-x*(-t); an analysis similar to part (a)(ii) proves that X(w) must be
imaginary. Hence, X(w) = X*(- w) = -X(- c). It follows that X(w) is also
odd.

S9.5
(a) 5{e~""'} = 5{e-"'u(t) + e"'u(-t)}
1_ 1
a +jw a- jo
2a

a2 + W2
(b) Duality states that

g(t) -G(w)

G(t) -21rg(-w)
Since
ea 2a
e a2 + W2,
 Fourier Transform Properties / Solutions
S9-5

we have
1 c7
1 + t2

1 s
-
1 X ­
x
(C) 1 + (3t) 2 - -e since x(at)
3 |al a
(d) We are given Figure S9.5-1.

x (t)

t
-T T

Figure S9.5-1

rT A
X(w) = A e--'' dt - . (e -jwT - e )wT
-r -Jw
- 2j sin coT
=A w
sin(wT)
T
= 2TA
eT
Sketches of y(t), Y(w), and X(w) are given in Figure S9.5-2.

X(<o)

/ N 10---.--CO

T T

y(t) 4TA

2T 22T

Figure S9.5-2
Signals and Systems
S9-6

Substituting 2T for T in X(w), we have

Y(w) = 2(2T) sin(w2T)

The zero crossings are at

w,2T = nr, or w = n

S9.6

(a) x(t) = - X(w)e- t dw

Substituting t = 0 in the preceding equation, we get

x(0) = 1 X( ) de
27r --

(b) X(w) = f x(t)e -jwt dt

Substituting o = 0 in the preceding equation, we get

X(0) =x(t) dt

S9.7
(a) We are given the differential equation
dy(t) + 2y(t) =
x(t) (S9.7-1)
dt

Taking the Fourier transform of eq. (S9.7-1), we have

jwY(w) + 2Y(w) = X(w)
Hence,
Y(w)[2 + jw] = X(w)
and
Y(w) = _1
H(w) =
X(W) 2 + jw'
2 - jw
H(w) = 1 = 1 c 2 - jo
2 + jw 2+j 2-jo) 4+ W2
2
4 + (,)2 -1 3 + x2 '
 Fourier Transform Properties / Solutions
S9-7

4 S2 ) 4 +2
IH(W)1 2
= (4 + c2)2 +
(4 + W2)2 (4 + W2)2>

IH(w)I =
\4 + W2

(b) We are given x(t) = e~'u(t). Taking the Fourier transform, we obtain
1
X(W)= ., Hx)= 2 +jW
1+j

Hence,

( 1 1 1 1
-(1 + j)(2 + j) 1 + jo 2 + jo

(c) Taking the inverse transform of Y(w), we get

y(t) = e-u(t) - e -2u(t)

Signals and Systems
S9-8

S9.8

A triangular signal can be represented as the convolution of two rectangular pulses,

as indicated in Figure S9.8.

Since each of the rectangular pulses on the right has a Fourier transform given
by (2 sin w)/w, the convolution property tells us that the triangular function will
have a Fourier transform given by the square of (2 sin w)/w:
2
sin w
X(()) = 4 (0).)2

Solutions to
Optional Problems
S9.9
We can compute the function x(t) by taking the inverse Fourier transform of X(w)

x(t) = ±27r f-. ire'' dw

(ei-
2 t~(j) e
sin oot
t
Therefore,
[sin (wot) 1
y(t) = cos (wet) s
I t
II
 Fourier Transform Properties / Solutions
S9-9

From the multiplicative property, we have

Y(w) = X(W) * [r(o - W) - 7r(WO + (0,]
Y(o) is sketched in Figure S9.9.

I T2C Iw
0 0
cj - W() Wc Wc ± W()ic C Cj C u cj( ± (W)

Figure S9.9

S9.10
(a) x(t) = e - cos wotu(t), a > 0
= e -"'u(t)cos(wot)
Therefore,

X(w) = . * [7r(w - wo) + 76(w + WO)]

2r a + Jw
1/2 1/2
a + j(W - WO) a + j(w + wo)
(b) x(t) = e -3 1 ' sin 2t
1 6
e- 3itiI
9 + w2

sin 2t - [6(- wo) - 6(w + wo)], wo = 2

Therefore,

= - (6
X(w)X()=1 2
2)*
7
-[S(o -
.
wo) - b(W + Wo)]
21r 9 + w2 3
j3 j3
9 + (w + 2)2 9 + (w - 2)2

(c) x(t) = sin wt (sin 2wt

1
X(W) = - XI) * X2(W),
21
where
IwI < 7,
Xi(w)XM=1,
=
0, otherwise

= l
IwI < 27,
X2(W)
10, otherwise
Signals and Systems
S9-10

Hence, X(w) is given by the convolution shown in Figure S9.10.

XI) X 2 (W)

-w­- o---I
T 7T -21T 27T

X(W)

-3 -2r -7 F r 2r 3
Figure S9.10

S9.11
We are given the LCCDE

dy(t)
=Acos oot
dt +±2ytt)

We can view the LCCDE as

dy(t) + 2y(t)
= x(t),
dt

the transfer function of which is given by

1
H(w) = . and x(t) = A cos oot
2 +j
We have already seen that for LTI systems,

y(t) = |H(wo)I A cos(oot + 4), where # = <H(wo)

1
= \4 -+w
A cos(wot + $)
 Fourier Transform Properties / Solutions
S9-11

For the maximum value of y(t) to be A/3, we require

1 _1

4 0+w
9
Therefore, wo = ± 5

S9.12

d2y(t) 2dy(t)
(a) (a
51 dt2 + dt + 3y(t) = -w 2Y(W) + 2jwY(w) + 3Y(w)
= (-W2 + j2w + 3)Y(o),
A(w) = -o 2 +j2w + 3
(4dx(t) x(t) I = 4jwX(w) - X(w)
(b)~ 3dt

= (j4w - 1)X(w),
B(o) = j4w - 1,
A(w)Y(w) = B(w)X(o),
B(o)
Y(W) =
A(w) X(x)
= H(o)X(w)
Therefore,
-1 + j4w
H(o) - B(w)
A(w) -W + 3 +j2w
2

1 - j4w
2 _ 3 -j2w

S9.13

sin W t

-W W
Figure S9.13-1

X(w)

sin 0t
it

-W0 0

Figure S9.13-2
Signals and Systems
S9-12

H(w)

sin (2o 0t)

h(t)=i r t

-2w0 2w0
Figure S9.13-3

Y(w)
2
7r

Y(w) = X(W)H(w) =
-wo ww
-o0 "0

Figure S9.13-4

Therefore, y(t) = sin(ot)

t

S9.14

(a) Energy |X(W)|2 dw

27r-
=
-o

Area = (4)(2) + (2)(1)(1)

= 10
5
Energy = -
Tr
 Fourier Transform Properties / Solutions
S9-13

(b)

1 1

-1 1 -2 2

Figure S9.14-2

sin t sin 2t
irt rt

S9.15

Given that
yi(t) = 2rX(-w)|,
we have
2
yi(t) = 1 x(u)ejtu du

Similarly, let y 2 (t) be the output due to passing x(t) through F twice.

Y2) = 21-0 21f x(u)evu du eitv dv

= (21r)2 fw -G x(u) fv eft+ u)vdv du

= (2_)2 f -0 x(u)(27r)b(t + u) du
= (21)' X(-t )
Finally, let y(t) be the output due to passing x(t) through F three times.

yAt = w(t) = 271- (2 7r)x(-u)ej'"du

= (21r)'f e -'u x(u) du

= (2.)'X(t)

S9.16
We are given

x(t) = n- e -'u(t), a > 0

(n - 1)!
Signals and Systems
S9-14

Let n = 1:
x(t) = e~"'u(t), a > 0,
1
X(w) = .
a + jw
Let n = 2:
x(t) = te~"u(t),

X(w) = j -dwa +jw. since tx(t) j-X(w)

d-
dw
__ 1
(a + jw) 2
Assume it is true for n:

x(t) = e~"'
1-atu(t),
(n - 1)!
1
X(w) =
(a + jw)"
We consider the case for n + 1:

x(t) = - e- at)
n!
j d 1 1
X(w) = d (a +jw)
n do [(a + j,)"n
_j d
n dw
= (-n)(a + jo)-"-Ij
n
1
(a + jo)"+I
Therefore, it is true for all n.
MIT OpenCourseWare
http://ocw.mit.edu

Resource: Signals and Systems

Professor 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.
10 Discrete-Time Fourier Series
Solutions to
Recommended Problems
S10.1
The output of a discrete-time linear, time-invariant system is given by

y[n] = j h[kjx[n - k],

k= ­

where h[n] is the impulse response and x[n] is the input. By substitution, we have
the following.

(a) y[n] =
)k k=0
e jwn 2
1 - i- 3
k 2 ]
(b) y[n] = (1ke j,(n-
k)4
ej Cr 4
k=0 k=0

1 - fjir/4)

(c) y[n] = ()k[

e 8
) i(n -)/4) 1e
+±2e ''r8e - '''"- k)/41 where we have used
0 -
2 Euler's relation

e (/8) e(rn/4) [e 3 (w/4 ) 1 k +21-e -j(/ 8)e -j(rn/4) e 2 ]

k=0 2 k=0
Iej[(1/ 8 )+(in/4)] S -j[(/8)+(rn/4)]
1 -j(r/4) + 2e ­ 1 (,r/4)

Cos ( n + -) 2Cos (4n

-o
- O
4 s "4

S1O.2

(a) 1 1[n] = 1 + sin( 10f)

To find the period of t 1[n], we set 2 1 [n] = 1 1[n + N] and determine N. Thus
27r
1 + sin 21rn 1 + sin [1 0 ( N

= 1 + sin n + N)

Since
2s 27n
sin (0n + 2r) =sin 10n

S10-1
Signals and Systems
S10-2

the period of t1n] is 10. Similarly, setting t 2[n]

12
207
1+ sin (2 n +-2
ns
= 1+ sin
20ir (i
= t

2(n +N)
2 [n + N], we have

+-2
7+ 20-)
-1+ sin 20-x n + - + -N
(12 2 12
Hence, for -irN to be an integer multiple of 21r, N must be 6.
u rn
(b) 21[n] = 1 + sin (21

Using Euler's relation, we have

J(2r/10)n
x 1 [n] = 1 ±
+ e27non _ e (S10.2-1)
2j 2j
Note that the Fourier synthesis equation is given by
21l[n] =. T ake jk(2x/N)n
k= (N)

where N = 10. Hence, by inspection of eq. (S10.2-1), we see that

-1
ao =
2j
1
all = and
2j
alk = 0, 2 s k - 8,
-8 : k - -2
Similarly,
1 e e-j(20Y/12>
t2[n] = 1 + -j(-/2>

2j
Therefore, N = 12.
1 1
a 2 0 = 1, a 2 -1 e -fir/2) 1 a 21 - - ear/2> = -, and
2j 2' 23 2
a2 ± 2, . . . , a2±= 0

(c) The sequence alk is periodic with period 10 and a 2k is periodic with period 12.

S10.3
The Fourier series coefficients can be expressed as the samples of the envelope
1 sin[(2Ni + 1)9/2]
ak - where N, = 1 (see Example 5.3 on
N sin (9/2) 0=2wk/N
page 302 of the text)
sin (39/2)
N sin (0/2) 0=2irk/N
(a) For N = 6,

ak =-
sin 3(2k)]
1 sin [26/)
6
1
sin (2)
1 2r i (k
sin sin -- )
26
 Discrete-Time Fourier Series / Solutions
S10-3

(b) For N = 12,

.i [3 (27rk{
1 sin 2 / 1
ak = - ___________
12
[112
12
(c) For N = 60,
(27rk
.i [ sin. -grk
k
1
60
2n 60)] 1 s 20)
[1 (2ck 60
sin 60 sin

S10.4

(a) The discrete-time Fourier transform of the given sequence is

X(Q) = ( x[n]e -'""

n= -o
= leit + 1 + ie ­
= 1 + cos Q
X(Q) is sketched in Figure S10.4.

X(E2)

IT IT

Figure S10.4

(b) The first sequence can be thought of as

9,[n] = x[n] * T, 6[n - 3k]I

Lk =0-Go

Hence
21 0" 2,rk
YI(Q) = X(Q) - 3 ,

Therefore, the Fourier series of y 1 [n] is given by

1 (2 k 1( 2 k\
ak=--Y1--3 kJ=-1+cos , for all k
Signals and Systems
S1O-4

The second sequence is given by

y 2[n] = x[n]* [n - 5k]

Similarly, the Fourier series of this sequence is given by

ak = 5- 1 1 + cos (5
, for all k
This result can also be obtained by using the fact that the Fourier series coeffi­
cients are proportional to equally spaced samples of the discrete-time Fourier
transform of one period (see Section 5.4.1 of the text, page 314).

S10.5
(a) The given relation
3
x[n] = ake k(2r'')"
k=o

results in the following set of equations

ao + ai + a2 + a3 = x[O] = 1,
ao + aie^,fir/2> + a 2e'" + aej(3 / 2) = x[1] = 0,
ao + alei" + a2e 2 ' + a3e 3 v = x[2] = 2,
ao + aie(3 / 2) + a2e 3' + asej( 9/ 2) = x[3] = - 1
The preceding set of linear equations can be reduced to the form
ao + a1 + a 2 + a3 = 1,
ao + jai - a 2 -ja 3 = 0,
ao - a1 + a 2 - a 3 = 2,
ao - jai - a 2 + ja 3 = -1
Solving the resulting equations, we get
1 __1+j 1-j
ao = , a- , a 2 = +1, a= - 4 (S10.5-1)
2 4 4
By the discrete-time Fourier series analysis equation, we obtain
ak = 1 + 2e -ik - e -j(3rk/2)j,
which is the same as eq. (S10.5-1) for 0 k - 3.

S10.6
(a) ak = ak+10 for all k is true since t[n] is periodic with period 10.
(b) ak = a_, for all k is false since I[n] is not even.
(c) ake ik(21/) is real. This statement is true because it would correspond to the Four­
ier series of t[n + 2], which is a purely real and even sequence.
(d) ao = 0 is true since the sum of the values of I[n] over one period is zero.
 Discrete-Time Fourier Series / Solutions
S10-5

Solutions to
Optional Problems
S10.7
The Fourier series coefficients of x[n], which is periodic with period N, are given by

ak = - T x[n]e -jk(2w/N)n
NI j(wNf
n=O

ForN = 8,

ak = E x[n]e-k(/
8n=0
4
)n (S10.7-1)

(a) We are given that

ak = cos (-)+ sin

4 (4
3k,
(S10.7-2)
ak = .1ej(wk/4) + -e -j(ik/4) + 1 e-t3rk/4) _ - e -(3rk/4)
2 2 2j 2j
Hence, by comparing eqs. (S10.7-1) and (S10.7-2) we can immediately write
3 ] + 4jb[n
x[n] = 46[n - 1] + 46[n - 7] - 4jb[n - - 51, 0 n !5 7
7 7
(b) x[n] k(2r/8) k jk(/4)n
k=O k=O

ei(kr/3)
- 1e-j(kr/3)] jk(/4)n

k=O II

k=O
7
1 [e(wn/ 4)+(7r/3) 1 1 [(7n/4)--(7r/3)
2j 1 -
4
)+(-/ )]
[eJ(rn/
3
2 1 j[(rn/4)-(/3)
7 4
S[(7n/4)+(7r/3)] {( -n/ )-(7-/3)j

2j 1 ee[(rn/4 )+(v/ 3
) 1 j[(n/4)-(r/3)j

7 7
Sx[n] k jk(2/8) akjk(w/4)n
k=O k=O
+ ejrn + ej(51/ 4 )n + ej( r/ )n
3 4 7 4
1 + ej(/ 4 )n + ej( -/ )n

=1++ (-) +2cos (n + 2 cos n 0 n 7

(d) Using an analysis similar to that in part (c), we find

x[n] = 2 + 2 cos4 n+ co + Cos n,

S10.8
The impulse response of the LTI system is
h[n] = (i)'"
Signals and Systems
S10-6

The discrete-time Fourier transform of h[n] is

H(Q) =
n=O
(- e -j"" +
n=-o
( ~"
~) -- 1
1 1
e~"
1+ 1 -i -1
3
5 - 4 cos 0

(a) (i) x[n] = sin n 3r= -e3r!4)n

1j
-
1
e -j(3,/4)n
4 2j 2j
The period of x[n] is

sin ( n) = sin - (n + N)
Thus
i 3
sin 3r n + -rN)
3)
si 4r
(4
We set 37rN/4 = 21rm to get N = 8 (m = 3). Hence, the period is 8.
7
x[n] T akejk(2w/8)n
k=O

Therefore,
1
a3 = =- a5

All other coefficients a, are zero. By the convolution property, the Fourier
series representation of y[n] is given by bk, where

bk = aJH(C
T s0=(2k)/8
Thus
1 3
b 2j 5 - 4 cos(3r/4)
=b*

All other bk are zero in the range 0 k - 7.

(ii) I[n] = 6[n - 4k]

k= -o

The Fourier series of t[n] is

3
ak = t4n~e -jk<2/4> = 1, f r all k
n=O

And the Fourier series of D[n] is

bk akH(Q)I
1 3 3
f or all k
4 -=-
=
5 - 4 cos[(2r/2)k] 20
 Discrete-Time Fourier Series / Solutions
S10-7

(iii) The Fourier series of f[n] is

ak = 1 + 2 cos (k), 0 s k - 5

and the Fourier series of 9[n] is

bk = akH(Q)

1 + 2 cos Irk 3
(3)5 - 4 cos[(7r/3)k]
(iv) x[n] = j" + (-1)"
The period of k[n] is 4. x[n] can be rewritten as
2
x[n] = [ej(r/ )n + (e'")"
3

T k jk(2/4)n
k=O

Hence,
ao = 0, ai = 1,
a2 = 1, a3 = 0
Therefore, bo = b = 0 and
3
bi
5 - 4 cos(7r/2) E
3 3
b2 =
5 - 4 cos r 9
(b) h[n] is sketched in Figure S10.8.

h [n ]

-2 -1
012
Figure S10.8

H(Q) = E h[n]e-"a" -_,;2 - e'o + 1 + e-~" + e-j2Q

n = -00

H(Q) = 1 - 2j sin 0 - 2j sin 20

It follows from part (a):

1 1 3-7r
b3 = --j H(Q) - sin - -sin -- = b*
=31r/4 2j 4 2
All other coefficients bk are zero, in the range 0 k 5 7.
Signals and Systems
S10-8

(ii) bk = H)
1 j. rk-j 1 j .7rk
sin- - sin k =- - -sin­
42 2 2 4 2 2

(iii) bk = - 1 + 2 cos ( k H(Q) f=rk/3

(iv) b0 = 0,
b 1 = H(Q) =1 - 2j,

b 2 = H(Q) =1,
b= 0

S10.9

x[n] ak

(a) x[n -no] -E2=

50 k jk(2w/N)n0
ake-j(/Nn

(b) x[n] - x[n - 11 5 ak[l - e -j(2rk/N)

N
(c) x[n] - x [n ­ 2 . ak(l - ei-*),
-jr N even
0, k even,
2ak, k odd

(d) x[n] + x n + -], period N

( 2

2 (NI x[nI + x n + - e -jk(4/N)n

Nn=O
2
= a2k

) L =*[- e -jk( 2 /N)n

Nn=O

2
a* = N1 N-i1 jk( /N)n

1-N+1
=- ~Lx[n]e -jk(2/N)n ak

Therefore, ak = atk.

S10.10
(a) i'i[n] = I[n] + 9[n],
i[n + NM]= .[n + NM] + P[n + NM]
= t[n] + 9[n]
= f>[n]
Hence, ?ID[n] is periodic with period NM.
 Discrete-Time Fourier Series / Solutions
S10-9

(b) Ck = Z
NMn=0
NM-I
n jk(2W/NM)n
NM-1
E [N[n] + P[n]]e jk(2T/NM)n
=

1 NM-1 NM=1

= ZNMn4N
[nje n=0
-jk(2/NM)n + 1 N
>3
n=0 :P4nje-j(/N
jk(2/NM)n

N-1 M-1 1 M-1 N-1

NM [n] E
n 1=0
e -jk(2./NM)(n+LN) + N
NMnO 1=0
ik(2/NM)(n+IM)

'1 1
N akIM + 1bk/N, for k a multiple of M and N,
1
akIM, for k a multiple of M,
1
1 bk/N, for k a multiple of N,
0, otherwise

S10.11

(a) t[n] = sin [(4

To find the period, we set g[n] = T[n + N]. Thus,

sin [(n - 1)] i(n + N - 1) sin [(n - 1)+ N

Let (7rN)/4 = 21i, when i is an integer. Then N = 8 and
41
1 e~ir(n-
1
1[n] )/ - 1 e -il(n -- 1)/41
2j 2j
1 e j(w/4) j(rn/4) j(/4) -j(,rn/4)

2j 2j
Therefore,
e -j(i/4)
2j 2j
All other coefficients ak are zero, in the range 0 : k - 7. The magnitude and
phase of ak are plotted in Figure S10.11-1.

lak I

0 1 2 3 4 5 6 7

4ak 3

I k
0 2 3 4 5 6 7
- 31r1 1
Figure S10.11-1
Signals and Systems
S10-10

(b) The period N = 21 and the Fourier series coefficients are

1 1
a 7 = a 14 = a 3 = a, 8 = ­
2 2j
The rest of the coefficients ak are zero. The magnitude and phase of ak are given
in Figure S10.11-2.

|ak

k
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

4ak

2
3
-k
0 1 2 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
2

Figure S10.11-2

(c) The period N = 8.

a 3 = a* = e -j(-/3)
The rest of the coefficients ak are zero. The magnitude and phase of ak are given
in Figure S10.11-3.

1 1
akk

0 1234567
4ak
3

3
0 k
0 1 2 4 5 6 7
Fr
3

Figure S10.11-3
MIT OpenCourseWare
http://ocw.mit.edu

Resource: Signals and Systems

Professor 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.
11 Discrete-Time Fourier Transform
Solutions to
Recommended Problems
S11.1

(a) X(Q) = 7 x[n]e -jn"

= (i" [nje -a

= (le -j")n
n=O
1
1 -ie-
Here we have used the fact that
1
a 1-a for |a l < 1

(b) x[n] = (a" sin Qon)u[n]

We can use the modulation property to evaluate this signal. Since

sin gon -- [4( - GO) - 4( + 00)],

2j
periodically repeated, then

X(Q) = 21 - ae- 1- ae -1Z+0)

periodically repeated.

(c) X(Q) = e -/"

n=O

1 -e-4
1 - e~i"

using the identity

N-1 1 N
a'= ­
T 1- a
Alternatively, we can use the fact that x[n] = u[n] - u[n - 4], so
1 e -j41 1 - e4
1 -e ~ 1- e 1 - e­
(d) x[n] = ({)"u[n + 2]
= (I)n+ 2()-2U[n + 2]
= 16({)n+ 2 u[n + 2]
We know that

16 ( "u[n] )" 7 16
16 ,j

2
16 16)1\+2 u[n~+ 21 7 16e' 0
1- je'

S11-1
Signals and Systems
S11-2

S11.2
(a) The difference equation y[n] - ly[n - 1] = x[n], which is initially at rest, has
a system transfer function that can be obtained by taking the Fourier transform
of both sides of the equation. This yields
Y(g)(1 - le -ia) =X(Q),
so
H(Q) = Y(Q) 1

X(Q) 1 (2)-il
(b) (i) If x[n] = b[n], then X(Q) = 1 and

1
Y(Q) = H(Q)X(Q) =

so
y[n] = (I)"U[n]

(ii) X(Q) = e -sano, so

Y(Q) = e _ln
12e

and, using the delay property of the Fourier transform,

y[n] = (i)""Ou[n no]

(iii) If x[n] = [n], then

1
X(Q) = 2 -j.2
1 4e

AMi= (1 1 ) ( ) = -2 3
so
y[n] = -2(i)"u[nj + 3(2)"u[n]

S11.3
(a) We are given a system with impulse response

h[n] = (1)cos ] u[n]

The signal h,[n] = (2)"u[n] has the Fourier transform

1
H,(Q) = ­

1 2
Using the modulation theorem, we have
1[ 1 1
H(Q) = 2 _ - -r/2 > + 1 -

(b) We expect the system output to be a sinusoid modified in amplitude and phase.
Using the results in part (a) and the fact that
2
x[n] = le(in/ ) + le -j(in/2)
 Discrete-Time Fourier Transform / Solutions
S11-3

we have

H(Q)
H()
=
('1 1 1
Q=r/2 2 1-i + 3

2 2 4jin2

so

y[n] = 2ej(1n/ 2) + ­
3 3
4 -ir
= - -2n
3 2

S11.4
(a) The use of the Fourier transform simplifies the analysis of the difference
equation.
1 1
y[n] + 1y[n - 1]- y[n - 2] = x[n] - x[n - 1],
1 1
Y(Q)(1 + 1e ~i" 8 -j21) = X(Q)(1 - e -jf),

Y(H) 1 - e ~iu
= H(Q) =.
X(Q) (1 + le -o)(1 - le -i")
We want to put this in a form that is easily invertible to get the impulse
response h[n]. Using a partial fraction expansion, we see that
2 -1
H() 1 + ie j + 1 - 1 e jfl)
-" I
so
h[n] = 2(-i)"u[n] -()"u[n]
(b) At R = 0, H(Q) = 0. At 9 w/4, H(Q) = 0.65ex1 22 >.Since h[n] is real, H(Q) =
H*(-Q), so H(-Q) = H*(Q) and H(-7r/4) = 0.65e -j122>. Since H(Q) is periodic
in 21,

H(-) = H (r)= 0.65e2

S11.5
(a) x[n] is an aperiodic signal with extent [0, N - 1]. The periodic signal

9[n] = E x[n + rN]

,"=-00
is periodic with period N. To get the Fourier series coefficients for 9[n], we sum
over one period of 9[n] to get
N-1

ak = T x[n]e -jk(21/N)n
n =0
Signals and Systems
S11-4

(b) The Fourier transform of x[n] is

X(Q) = x[n]e -"

N-1

= Z x[nje -"
n=0

since x[n] = 0 for n < 0, n > N - 1.

We can now easily see the relation between ak and X(Q) since
1 )1 N-1 x[n]e -jk(2i/N)n
SX Q)I = - T
N 9= (2xk)/N N n=O

Therefore,
1 (27rk
-Xi,-=ak
N N

S11.6

(a) Signal Description Transform

Continuous time Infinite duration Periodic I, III

Continuous time Infinite duration Aperiodic III
Continuous time Finite duration Aperiodic III, I*
Discrete time Infinite duration Periodic II, IV
Discrete time Infinite duration Aperiodic IV
Discrete time Finite duration Aperiodic IV, II*
*Because these two signals are aperiodic, we know that they do not possess a
Fourier series. However, since they are both finite duration, the Fourier series
can be used to express a periodic signal that is formed by periodically replicat­
ing the finite-duration signal.
(b) The discrete-time Fourier series has time- and frequency-domain duality. Both
the analysis and synthesis equations are summations. The continuous-time
Fourier transform has time- and frequency-domain duality. Both the analysis
and synthesis equations are integrals.
(c) The discrete-time Fourier series and Fourier transform are periodic with peri­
ods N and 2-r respectively.

Solutions to
Optional Problems
S11.7

Because of the discrete nature of a discrete-time signal, the time/frequency scaling

property does not hold. A result that closely parallels this property but does hold
 Discrete-Time Fourier Transform / Solutions
S11-5

for discrete-time signals can be developed. Define

x[n/k], if n is a multiple of k,
0, otherwise
X(k)[n]is a "slowed-down" version of x[n] with zeros interspersed. By analysis in
the frequency domain,
X(k)() = X(kQ),

which indicates that X(k)(Q) is compressed in the frequency domain.

S11.8

(a) X(Q - QO) is a shift in frequency of the spectrum X(Q). We will see later that
this is the result of modulating x[n] with an exponential carrier. To derive the
modification xm[n], we use the synthesis equation:

x.[n] =-1 f X(Q - Qo)e'2" dQ

Changing variables so that 0 - 00 = 0', we have

Xm[n] = X(Q')ej('+Qo)n di = x[n]e"On

(b) Using the synthesis equation, we have

- Re{X($)}e "" dQ - f2,, - [X() + X*(Q)]e"Qn dQ

1 1(
= -x[n| + X()e -j"" dQ
2 2( 2 2
= -{x[n] + x*[-n]}
2

(c)(C --
1 f f mXgeidQ
Im{X(Q)}e'"" dQ 1 f2"[X(Q) - .X*(Q) ]ejnd
= -- ei" dG
1 1(1 *
-x[n = X(Q)e -"" dQ)
2j 2j \2 2 ,
1
= -{x[n] - x*[-n])
2j
(d) Since IX() 12 = X(Q)X*(Q), we see that the inverse transform will be in the form
of a convolution. Since

I X*(G)ein dQ = 21
X(Q)e -j"" d)
27 2, 7r

=x*[ -n],
then

21
7 2,
\X(Q)| 2
ein d2 = x[n] * x*[-n]
Signals and Systems
S11-6

S11.9
We are given an LTI system with impulse response

h[n] = sin(-xn/3)
irn
(a) We know from duality that H(Q) is a pulse sequence that is periodic with period
21r. Suppose we assume this and adjust the parameters of the pulse so that

1-
21r
J H()e'""dU = h[n]

Let a be the pulse amplitude and let 2W be the pulse width. Then
a
21r
fw
-w
gd.
2-7
a (ei"w-
jn
e iflW)

a 2 sin Wn
21r n '
so a = 1 and W = 7r/3, as indicated in Figure S11.9-1.

H(Q2)

-7T _IT A T

Figure S11.9-1

(b) We know that

31 n - 37 3r
cos- 4 n{ + r -4 r)+o +-4-)

periodically repeated, and that multiplication by (- 1)' shifts the periodic spec­
trum by 7r, so the spectrum Y(Q) is as shown in Figure S11.9-2.

Y(72)

IT IT

4 4

Figure S11.9-2

From Figures S11.9-1 and S11.9-2, we can see that

Y(Q) = H(Q)X(Q) = X(Q)
 Discrete-Time Fourier Transform / Solutions
S11-7

Therefore,

y[n] = x[n] = (-1) cos-n = cos­

4 4

S11.10
Here
dX(Q)
Y(Q) = 2X(Q) + e-i"X(Q)

(a) (i) The system is linear because if

x[n] = axl[n] + bx 2[n],
then
y[n] = ay1 [n] + by 2[n],
where y 1[n] is obtained from x1[n] via the given transfer function. The
similar result applies for Y2[n].
(ii) The system is time-varying by the following argument.
If x[n] - y[n], does x[n - 11 -+ y[n - 1]?

x[n - 1] e -i"X(Q)
The corresponding Y(Q) is

2ei"X($) + e -i"X(')e -i" + je j"X(Q) - e -i"dX(C)

dQ

# e -i" 2X(Q) + e -i"X(Q) -dX()

dQ
(iii) If x[n] = 6[n], X(Q) = 1. Then
Y(Q) = 2 + e
y[n] = 26[n] + bfn - 11

S11.11
2[n] = akejk(21/N)n
k=(N)

(a) If we multiply both sides of this equation by e -jl(2w/N)n and sum over (N), we
obtain
t! nje -jl(2./N)n =k j(k--)(2/Nn
n= (N) k=(N) n=(N)

If k is held fixed, the summation over (N) is zero unless k = 1,which yields Nal.
Thus

a, = - ne jl(21/N)n
n=(N)
Signals and Systems
S11-8

and therefore

- ( x[n]e -jk(2r/N)n
ak =
N n=(N)
(b) We are given that x[n] is an aperiodic signal

x[n] = - X(Q)e'j" d2

(i) By multiplying both sides by e -join and summing over all n, we have

x[n]e -jIn =
2
X(Q) e"(-")n di
n = oo 1r f2 _n 0

(ii) En ej' -"")n needs to be evaluated. We can recognize that this summa­
tion is a Fourier series representation

7 e'j -n =T an e j[(2(QQ-i))/T]n

where T = 27r and a. = 1. The periodic function represented by this series

is a periodic impulse train with period T = 2 r, so

exa-oon = 27 (( - Q, + 27rn)
n= -oo n =-oo

(iii) Only a single impulse in the train appears in the integration interval of
one period. So

2i J X(Q) e -Gl" = X(Q1 + 2irn)

= X(01)
Therefore, the analysis formula for aperiodic discrete signals has been
verified to be analogous to the analysis formula in part (a).

X(Q) = ( x[n]e "

n=-oo

S11.12
(a) The Fouriertransform of ejk(21/N)n can be performed by inspection using the syn­
thesis formula

e jk( 2w/N)n = - X( jI" dQ)

X(Q) =27rb Q , |7r | < ir

N
and since we know that X(Q) is periodic in Q = 2
7r, we have

e jk(2r/N)n L 6k 2-7rk +27rm

m= -- N!
(b) By using superposition and the result in part (a), we have

ake
aejk(21/N)n k( -o aN
k
Q -
2 + 2rm
k =(N) M= -0 k =(N) \
 Discrete-Time Fourier Transform / Solutions
S11-9

(c) We can change the double summation to a single summation since ak is periodic:
27k 027k
2,r1( akb Q N + 27rn =27r akb Q N
- k=(N) k=-w

So we have established the Fourier transform of a periodic signal via the use of
a Fourier series:

[n] = ake(21/N)n 1 k( 2)
k=(N) k=-w

(d) We have

.[n] = ( x[n - kN] ( X(Q)e -jfkN

k=-w k=-w

As in S11.11(b)(ii), we can show that

k=-x -jkN = kw 6(Q2.rk

Therefore,
1 2xk

=k = 1 X(-)
N 10=(2rk)/N
MIT OpenCourseWare
http://ocw.mit.edu

Resource: Signals and Systems

Professor 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.
12 Filtering
Solutions to
Recommended Problems
S12.1
(a) The impulse response is real because

h(t) = - H(w)ew' dw,

21­
h*(t) = f H*(w)e ~i"' dw = - H(w)e -i-t dw

1 H(-w)e-' do = --- H(w)eij' dw = h(t)

where we used the fact that H(co) = H*(o) = H(-w).

The impulse response is even because

h(t) = -- H(w)e'wt dw,

21 ao

h(-t) = - H(w)e -jwt dw

2,7

= ­ H(-w)e' t dw
21-o
Since H(-w) = H(w),

h(-t) = H(w)ej' dw
= h(t)

The impulse response is noncausal because h(- t) = h(t) # 0.

(b) x(t) = b(t - 9n),

x(t) = ( aejitsl2ir
k= -00

an = M~-j[(2"k*)ITI
ak = 9odt x(t)e

Here T = 9, s0
2rk
ak = - and 5J{ei(2*kt)!TI) =
9 T)
Consequently, the Fourier transform of the filter input is as shown in Figure
S12.1-1.

X(W)

27r
9

47T 27r 0 2In 4nf

97 9 9 9
Figure S12.1-1

S12-1
Signals and Systems
S12-2

Since Y(w) = H(w)X(w), the Fourier transform of the filter output is as shown
in Figure S12.1-2.

Y(W)

2ni
9

2n 0 2r
9 9
Figure S12.1-2

(c) We determine y(t) by performing an inverse Fourier transform on Y(W) as found

in part (b). Using superposition, we have
1 2 21rt
y(t)YO=9
= -+-co
+9 CO 9

S12.2

From the filter frequency response plots we can determine that

H(w) = 0.25e -j(r/8) at w = wi = 7,
4
H(w) = 0.5e -(T/) at w = W2 =2r

Using superposition, we easily determine y(t) to be

71
y(t) = 0.25 sin(irt + 1/8) + cos( 2rt ­

S12.3
dvo
(a) RC- +v = v,
dt
Taking the Fourier transform of this equation, we have
(RCjo + 1)Ve(w) = V,(w)
We now define
H, ( Ve(w) 1
V,(w) 1 + jwRC
We can see from this expression that vc(t) is a lowpass version of v,(t).
 Filtering / Solutions
S12-3

The magnitude and phase of Hi(w) are given in Figure S12.3-1.

OdB

-20

-40

-60
0.1/r 1/r 1O/T 100/r
r = RC

-37r/4 L
0.1/r 1/7- 10/k 100/r
r = RC

Figure S12.3-1

(b) RCd(v - Vr) + V, - Vr =Vs,

dt
RCjwV,(w) - RCjwV,(w) - Vr(W) = 0,
(jwRC)V,(w) = (1 + jwRC)Vr(w),
V,(w) jwRC
_

V,(w) 1 + jwRC
Signals and Systems
S12-4

The magnitude and phase of H 2(w) are given in Figure S12.3-2.

3
0

0
0
0
CN

1 10
R C RC

Figure S12.3-2

(c) The cutoff frequencies are we = 1/RC in both cases.

jwRC
(d) V(M) 1 - H1) = = 11=
H2(W)
V,(w) 1 + jwRC
This is the same frequency response as sketched in part (b). We have trans­
formed a lowpass into a highpass filter by a feed-forward system. The cutoff
frequency, as in part (c), is w, = 1/RC.

S12.4
Consider 0 go !5 r. In this range, the gain of the filter IH(Q) I is go. The phase shift
for the positive frequency component is +7r/2 and the shift for the negative fre­
quency component is -7r/2. Since
1
x[n] = cos (gon + 0) = -[e(On+0) + e -j6on+Oj,
2
2
y[n] = 0 [eI1On+O+(/ ) + e -j[Uon+O+(ir/2)
2
=j (on+") - e ~j(fl0 n+')
[e
2
y[n] = -9 0 sin (gon + 0)
It is apparent from this expression that H(Q) is a discrete-time differentiator. A sim­
ilar result holds for -7r ! go s 0.
 Filtering / Solutions
S12-5

If go is outside the range -7r :5 go r, we can express x[n] identically using a

go within this range. For example,

x[n] = cos (n + )
= cos - n + ,

y[n] = 7 sin - n + )

S12.5
(a) We see by examining yi[n] and y 2[n] that y[n] averages x[n] and thus tends to
suppress changes while y 2[n] tends to suppress components that have not var­
ied from x[n - 1] to x[n]. Therefore, the yi[n] system is lowpass and y 2[n] is
highpass.
(b) Taking the Fourier transforms yields
1+ e-
Yi(Q) = X(Q)
2'
Hi(9) = -(1 + e -ja
2

Figure S12.5-1

Y2(Q) = X(Q) 2e
H 2(0) = -(1 - e )
2
Signals and Systems
S12-6

S12.6
(a) By inspection we see that the impulse response is given by
[ NN
h,[n] = E b[n - k]
2N + 1 k= -N

sin Q2N + 1)
(b) H 2 () = 1 ­
2N + 1
1
Ssin(/2)
(c)

|H1 (2)l

2w 21r2 2T ­ 2' 21T 21r+ 2,

2N + 1 2N + 1 2N+ 1 2N +1

Figure S12.6-1

IH2WA

I I

2w 2tr2
2N+1 2N+1

Figure S12.6-2

Zero and one crossings are at

2,r k.
(2N + 1)
(d) H 2 (Q) is an approximation to a highpass filter.

S12.7
(a) From the specification that H(O) = 1, we know that

H(w) = a
a + jw

H()| = a
(b)
(h) I1~) 1 (a 2 + Wa2)1/2)

( a = H()I
(a2 + 4)'/2 1,2
 Filtering / Solutions
S12-7

The low end specification is satisfied for a > 4, as shown in Figure S12.7-1.

IH(w)I
w=2

Figure S12.7-1

The high end specification is met for a !: 6, as shown in Figure S12.7-2.

IH( o)l
w =6

-f2­

Figure S12.7-2

The range of a such that the total specification is met is 4 2! a 2 6.

Solutions to
Optional Problems
S12.8
The easiest method for solving this problem is to recognize that passing x(t) through
H(w) is equivalent to performing

-2 dx(t)
dt
This is easily seen since

x(t) = - X(w)eiw' dw,

27r -.

-2 dx(t) = 17- f -2jw X(w)e- t dw

dt 2,rJ
HCCQ)

dx(t )
-2 lt -2jwX(w)
dt
Signals and Systems
S12-8

(a) -2 dx(t) 2 = -2jej' = y(t)

dt ) dti o ut)
(b) -2
()2 dtdt -2 d[(sin dt
w2t)u(t)] o2w(cos wt)u(t)

1 _ - 6
(C) XMx = =6J + ". ,
jw(6 + jw) jw 6 +jw

x(t) = 1[u(t) - - 6t u(t)

-2 dx(t) -2 [16(t) + e -6 1u(t) 6 e -6t6(t)

=
dt [6
6
= -2e- -u(t)
Alternatively, for this part it is perhaps simpler to use the fact that
- 2jw
Y(w) = H(w)X(w) =
JM(6 + jw)
2
6 + jw
so that y(t) = -2e -6 1 u(t)

1
(d) X(M) =
2 + jo
x(t) = e -2 1u(t)
-2 dx(t) - 2[-2e 2
'u(t) + e -2'tt)]= 4e -2 1 u(t) - 26(t)
dt

S12.9
(a) H(Q) = H,(Q)e ~m"
(i) H,(Q) is real and even:
h4n] - H,.()
From Table 5.1 of the text (page 335), we see that the even part of h4n]
has a Fourier transform that is the real part of H,(Q). This result is easily
verified:

Sh4-nle-j = ),h[n]e(G

so
s(h,[n] + h,[-) - [H,(Q) + H*(n)],
Ev~h,[n]} Re{H,.(Q)}
Now since

Re{H,.(Q)} = H,(Q),
we have that Ev{hn]} = h,.[n], i.e., h,[n] is even, and therefore
h,[n] = h,-n]
 Filtering / Solutions
S12-9

(ii) From Table 5.1,

x[n - no] -e -iflno

so
H,(Q)e -j" h,[n -l ,
h[n] = hn - M]
(b) h4n] = h4-n]
Since h[n] = hAn - MI,
h[n + M] = hn],
h[M - n] = h,(M - n) - M] = h,[-n],

but
hn] = h,-n] =+ h[M - n| = h[M + n]
(c) h[M + n] = h[M - n] from part (b). Since h[n] is causal, h[M - n] = 0 for
n > M. But if h[M + n] = h[M - n], then
h[M+ n] = 0 for n > M,
so
h[n] = 0 for n > 2M
Summarizing, we have
h[n] = 0 for n < 0, n > 2M

S12.10
(a)
Hi (n)

1- ­

i --4- Q
- I _ ( T
c 2i( r + Qc 7T

Hv( N

Figure S12.10-1
Signals and Systems
S12-10

(b) If the cutoff frequency Q. = ir/N, the total system is an identity system.
N-1 N-1
(c) h[n] = L hk[n] = ej(2 7rnk/N)ho[n]
k=O k=O

[1- e j2rn

(2nN)Ih[n],
h[n] Nho[n], n = an integer multiple of N,
0, n # an integer multiple of N,
so r[n] is as shown in Figure S12.10-2.

(d) ho[n] = -, n = 0,
N)
ho[n] = 0, n
an integer multiple of N,
=
are the necessary and sufficient conditions.

S12.11

From the system diagram,

Y(w) = X(w)[a - G(w)],
H(w) = a - G(w)

(a) <t H(w) is 0 for all w.

 Filtering / Solutions
S12-11

(b)
|H(w)|

"1 -a­
I I

01 W 2

4H(w)

0 OI 2

Figure S12.11-2

(c) - H(w) is 7r for all o.

|H(w)I

CI i oj

Fg1 S2
Figure S12.11-3 2
MIT OpenCourseWare
http://ocw.mit.edu

Resource: Signals and Systems

Professor 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.
13 Continuous-Time Modulation
Solutions to
Recommended Problems
S13.1
(a) By the shifting property,

x(t)ejawc t - X(w - 3we) = Y(w)

The magnitude and phase of Y(w) are given in Figure S13.1-1.

IY(W)I

2w.. 3wc 4wc

4Y(W)

2 4w.

2w~ 3w~

Figure S13.1-1

(b) Since ej 3 wc+ji2 e r/2 e 3 wt, we are modulating the same carrier as in part (a)
except that we multiply the result by eij/ 2 . Thus
Y(w) = ej' 2X(w - 3c)
Note in Figure S13.1-2 that the magnitude of Y(w) is unaffected and that the
phase is shifted by ir/2.

|Y(w)|

I w()

4 Y(w)

7r -­

2wC 4wc

Figure S13.1-2

(c) Since
ej3wct g -j3wct
cos 3 xeot = 2 +
2 2'

S13-1
Signals and Systems
S13-2

we can think of modulation by cos 3wet as the sum of modulation by

Sj3eet , -j3wc t
2 and
2
Thus, the magnitude and phase of Y(o) are as shown in Figure S13.1-3. Note the
scaling in the magnitude.

jY(w)I

-3wc 3ic

4 Y(w)

-3w 3w
2
c
Figure S13.1-3

(d) We can think of modulation by sin 3wet as the sum of modulation by

ej3wct -jr/2 e -j3wct -fr/2

2 and 2
Thus, the magnitude and phase of Y(w) are as given in Figure S13.1-4. Note the
scaling by 1in the magnitude.

IY(w)I

-3oic 3oc
4 Y(W)

-4c -2wc 2c 4

Figure S13.1-4

(e) Since the phase terms are different in parts (c) and (d), we cannot just add spec­
tra. We need to convert cos 3wct + sin 3wet into the form A cos(3wet + 0). Note
 Continuous-Time Modulation / Solutions
S13-3

that
cos(a - #) = cos a cos # + sin a sin #
4
Let a = 3wet and # = 7r/ . Then

cos (3xet - = (cos 3cet + sin 3wet)

Thus
-r
cos 3Uet + sin 3wet = \/2 cos (3wet

Now we write c(t) as

2e
3
wct
- ("!) + 2 e j3wt -(r/4)J
2 2
Modulating by each exponential separately and then adding yields the magni­
tude and phase given in Figure S13.1-5. (Note the scaling in the magnitude.)

Y(W)|

3co~
-3co~
-3wc 3 Ct
4 Y(W)

2wc

Figure S13.1-5

S13.2
In Figure S13.2-1 we redraw the system with some auxiliary signals labeled.

r, t) r2(t) ir3(t) r4(t)

x(t) X X 1 y(t)

s(t) m(t) d(t)

Figure S13.2-1
Signals and Systems
S13-4

By the modulation property, Ri(w), the Fourier transform of ri(t), is

1
R1(w) = - [X(w) * S(w)]
27r
Since S(w) is composed of impulses, Ri(w) is a repetition of X(w) centered at -2c,
0, and 2 we, and scaled by 1/(27r). See Figure S13.2-2.

R 1 (w)

1
2ir

Figure S13.2-2

(a) Since m(t) = d(t) = 1, y(t) is ri(t) filtered twice by the same ideal lowpass filter
with cutoff at we. Thus, comparing the resulting Fourier transform of y(t),
shown in Figure S13.2-3, we see that y(t) = 1/(27r)x(t), which is nonzero.

Y(W)

1
2ir

WCo

Figure S13.2-3

(b) Modulating ri(t) by ei-c' yields R 1 (w - we) as shown in Figure S13.2-4.

R 1 (o - wc)
1
27

S.

Figure S13.2-4
 Continuous-Time Modulation / Solutions
S13-5

Similarly, modulating by e ~j'Wyields Ri(o + coc) as shown in Figure S13.2-5.

R I(w + wc)
1
2n

-2wc -c

Figure S13.2-5

Since cos wot = (ewct + e -iwct)/ 2 , modulating ri(t) by cos wct yields a Fourier
transform of r 2(t) given by
Ri(w - wc) + Ri(w + wo)
2
Thus, R 2 (W) is as given in Figure S13.2-6.

R 2(CJ)

2-t

-wc II 2
co
-2Fc -- oc Wc 23c.

Figure S13.2-6

After filtering, R3(o) is given as in Figure S13.2-7.

R 3(W)

-- c Wc

Figure S13.2-7
Signals and Systems
S13-6

R 4 (w) is given by shifting R 3 (W) up and down by we and dividing by 2. See Figure
S13.2-8.

R 4 (W)

1
4w

-2c -wc Wc 2.c

Figure S13.2-8

After filtering, Y(w) is as shown in Figure S13.2-9.

Y(W)
1
4w

Figure S13.2-9

Comparing Y(w) and X(w) yields

1
y(t) = -- x(t)
4-r
(c) Since
g jwct _ -jwct

sin wet = 2j
then

R 2 (w) = Ri(w - oc) - Ri(w + we)

2j
which is drawn in Figure S13.2-10.

R2(W)

I/

-4I
4w1

Figure S13.2-10

After filtering, R 3(x) = 0. Therefore, y(t) = 0.

 Continuous-Time Modulation / Solutions
S13-7

(d) In this case, it is not necessary to know r 3 (t) exactly. Suppose r 3 (t) is nonzero,
with RA(W) given as in Figure S13.2-11.

After modulating by d(t) = cos 2cot, R 4 (W) is given as in Figure S13.2-12.

R 4 (w)

-2wc -~oc c 2o

Figure S13.2-12

After filtering, y(t) = 0 since R 4(w) has no energy from -we, to we,.
(e) For this part, let us calculate R 2(w) explicitly.
RI(w - 2wc) + R,(w + 2wc)
R2(W) = n

which is drawn in Figure S13.2-13.

R 2 (W)

1
4ff

-2wc Wc Wc 2wc
Figure S13.2-13
Signals and Systems
S13-8

After filtering, R 3 (w) is as shown in Figure S13.2-14.

R 3 (w)

47r

-coc Wc

Figure S13.2-14

Modulating again yields R 4(w) as shown in Figure S13.2-15.

R 4 (W)

c S c
Figure S13.2-15

Finally, filtering R 4 (w) gives the Fourier transform of y(t), shown in Figure
S13.2-16.

Y(o)

8ir

Figure S13.2-16

Thus,

y(t) = 1- x(t)
87r

S13.3

(a) The demodulator signal w(t) is related to x(t) via

W(t) = (cos wdt) (cos Wt)x(t)
 Continuous-Time Modulation / Solutions
S13-9

Since cos A cos B = 1[cos(A - B) + cos(A + B)],

w(t) = i[cos(Aw)t + cos(Aw + 2we)t]x(t)
- i[cos(A)t]x(t) + -[cos(Aw + 2wc)t]x(t)
The first term is bandlimited to ± (wM + IAwl), while the second term is band-
limited from Aw + 2wc - wM to Aw + 2wc + wM. Thus after filtering, only the first
term remains. Therefore, the output of the demodulator lowpass filter is given
by .x(t)cos AWt.
(b) Consider first IAw I > wM Then for X(w) as given, ix(t)cos Awt has a Fourier
transform as shown in Figure S13.3-1.

_|
I

-lAwi Iol
Figure S13.3-1

For IAwl <wM, there is some overlap. See Figure S13.3-2.

O <IAWl<wJ M 1 M O<IAWK 1.TM

M

1
14

-awI lwl -JAWI AW

Figure S13.3-2

S13.4
(a) In this case,
y(t) = [A + cos wL)Mt cos(Wct + Oc)

But
Cos oMt cos(Wct + O') = 1[cos((WM ~ cjt ~ Oc)+ cOs((WM + Oc)t + cj]

Thus,
1 1
y(t) = A cos(wct + Oc) + 1cos((wM w)t O + 1cos((WM + W')t + Oc
2 2
= eiwct + e -Jwct + -e ijceJ(WM- C)t
2 2 4
t
+ 1 ejoce -i(WM-Wc)t + 1 ejOCe(WM+)dt + ! e -joce -j(wm±(cc)
4 4 4
Signals and Systems
S13-10

We recognize that the preceding expression is a Fourier series expansion. Using

Parseval's theorem for the Fourier series, we have

TO k=

Thus,

P,=2(A) + 42
A2
2
Since
max Ix(t)| 1
A A
then
1 1
9 2M 2
4,
as shown in Figure S13.4-1.

Py

1 4M

1 1

overmodulation
Figure S13.4-1
(b) The power in the sidebands is found from P, when A = 0. Thus, P,y = and the
efficiency is
E= =
2
1/(2M ) + } 2 + M2 >

which is sketched in Figure S13.4-2.

 Continuous-Time Modulation / Solutions
S13-11

Solutions to
Optional Problems
S13.5

(a) Using the identity for cos(A + B), we have

A(t)cos(oct + Oc) = A(t) (cos 0e cos oct - sin 0c sin oct)
Thus, we see that
x(t) = A(t) cos 0c,
y(t) = -A(t) sin 0,
Therefore,
z(t) = A(t)cos(oct + 0c)
= x(t)cos wct + y(t)sin wot

(b) Consider modulating z(t) by cos oc. Then

z(t)cos wot = x(t)cos2 wet + y(t)sin oct cos wct
Using trigonometric identities, we have

z(t)cos oct = + cos 2et + sin 2 wet

2 2 2
If we use an ideal lowpass filter with cutoff wc and if A(t), and thus x(t), is
bandlimited to then we recover the term x(t)/2. Thus the processing is as
_oc,
shown in Figure S13.5-1.

z (t) X2: x (t)

-WC C

Cos coCt

Figure S13.5-1

(c) Similarly, consider

z(t)sin oct = x(t)cos oct sin wct + y(t)sinswct

= 2 sin 2wet + y) y) cos 2wct

2 2 2
Filtering z(t) sin wct with the same filter as in part (b) yields y(t), as shown in
Figure S13.5-2.
Signals and Systems
S13-12

z(t) rX eyt

sin coc t

Figure S13.5-2

(d) We can readily see that

x 2(t) + y 2(t) = A 2 (t) (cos 20e + sin20') = A 2(t)

Therefore, A(t) = x 2 (t) + y 2(t). The block diagram in Figure S13.5-3 sum­
marizes how to recover A(t) from z(t).

z(t) A (t)

sin ct
Figure S13.5-3

Note that to be able to recover A(t) in this way, the Fourier transform of A(t)
must be zero for o > oc| I and A(t) > 0. Also note that we are implicitly assum­
ing that A(t) is a real signal.

S13.6
From Figures P13.6-1 to P13.6-3, we can relate the Fourier transforms of all the
signals concerned.

SI(M) = - [X (W 010--
2)
X(
2j

S2(W) = [ X
x '
-O + X o ­

Thus, Si(w) and S2(O) appear as in Figure S13.6-1.

 Continuous-Time Modulation / Solutions
S13-13

SI (w)

S2(w)
1
2

0 0
2 2

Figure S13.6-1

After filtering, S 3(w) and S 4(w) are given as in Figure S13.6-2.

S3 (w) S4 (Co)

Figure S13.6-2

S5(U)is as follows (see Figure S13.6-3):

[
S5(W)= 1 S3 ( W- We
-) - S3 (W + W, +2

Note that the amplitude is reversed since (1/2j)(1/2j) = -I.

Signals and Systems
S13-14

Ss (C)

1
4

-- oc c
4

Figure S13.6-3

S6 (w) is as follows and as shown in Figure S13.6-4.

S6(w)= S4 -cw- 2 + S4 x + We + 2)]

Finally, Y(w) = S5(w) + S6 (w), as shown in Figure S13.6-5.

Y(W)

2 -­

~-oc
I______
oc
CO
Figure S13.6-5

Thus, y(t) is a single-sideband modulation of x(t).

S13.7

Note that
qI(t) = [si(t)cos wot + s 2 (t)sin wot]cos wot
= s 1 (t)cos 2xot + s 2(t)sin wot cos wot

Using trigonometric identities, we have

qi(t) = isi(t) + isi(t)cos 2wet + is (t)sin
2 2wot
Thus, if s 1(t) is bandlimited to ± wo and we use the filter H(w) as given in Figure
S13.7, y 1(t) will then equal s 1(t).
 Continuous-Time Modulation / Solutions
S13-15

H(w)

Figure S13.7

Similarly,
q 2(t) = si(t)cos wet sin wot + s 2(t)sin 2Wot
SIM sin 2w t + s 2 (t) s 2 (t) 2wot
2 2 2
Using the same filter and imposing the same restrictions on s 2 (t), we obtain y 2(t) =
s2 (t).

S13.8
(a) X(w) is given as in Figure S13.8-1.

X(w)

-1

-ojM WM
Figure S13.8-1

For Y(w), the spectrum of the scrambled signal is as shown in Figure S13.8-2.

Y(W)

Figure S13.8-2

Thus, X(w) is reversed for w > 0 and w < 0.

Signals and Systems
S13-16

(b) Suppose we multiply x(t) by cos Wot. Denoting z(t) = x(t)cos wmt, we find that
Z(w) is composed of scaled versions of X(w) centered at ± wM. See Figure
S13.8-3.

Z(W)

-2wm M WM 2
WM

Figure S13.8-3

Filtering z(t) with an ideal lowpass filter with a gain of 2 yields y(t), as shown
in Figure S13.8-4.

x (t) MX 2. iy(t)
_M
- IjMJ
Cos WMt
Figure S13.8-4

(c) Suppose we use the same system to recover x(t). Let y(t)cos Wmt = r(t). Then
R(w) is as given in Figure S13.8-5.

R(o)
1

2
WM WM

-2wM -- "M ljM 2wM

Figure S13.8-5

Filtering with the same lowpass filter yields x(t).

MIT OpenCourseWare
http://ocw.mit.edu

Resource: Signals and Systems

Professor 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.
14 Demonstration of
Amplitude Modulation
Solutions to
Recommended Problems
S14.1
(a) We see in Figure S14.1-1 that the modulating cosine wave has a peak amplitude
of 2K = 2, so that K = 1. At the point in time when the modulating cosine wave
is zero, the total signal is A = 2, so K/A = 0.5. Therefore, the signal has 50%
modulation. See Figure S14.1-1.

(b) 2K = 2, K = 1, A = 1, so K/A = 1, and the signal has 100% modulation. See

Figure S14.1-2.

S14-1
Signals and Systems
S14-2

2K

.2 .4 .6 .8

Figure S14.1-2

(c) 2K = 2, K = 1, A = 0.5, so K/A = 2, and the signal has 200% modulation.

2K

T-0
(till 11 zm "V
irri~irur~v
UW TU
Figure S14.1-3
 Demonstration of Amplitude Modulation / Solutions
S14-3

S14.2
(a) (i)

y(t)

3-
2--

1--

10 12

Figure S14.2-1

(ii)
J, (t0

3\ /1

0-- t
1 2 3 4 5 6

-2

-3 -

Figure S14.2-2
Signals and Systems
S14-4

(iii)

(b) (i) Y(w) = y(t)e -j"' dt

= x (-t) e j' tdt, t' = , dt' = 1

2'
= x(t')e -jw"' 2 dt'

= 2 X(2w)

Therefore, Y(w) is a compressed version of X(w). See Figure S14.2-4.

(ii) From the convolution theorem,

Y(W) = 27r- X(Q)H(w - Q)dQ,

-.
 Demonstration of Amplitude Modulation / Solutions
S14-5

where cos irt 5 H(w), and H(w) is as shown in Figure S14.2-5. There-
fore, Y(w) is as given in Figure S14.2-6.

H(w)

7T iT

-7T 7T

Figure S14.2-5

Y(W)

37r -n _7 3n
2 2 2 2

Figure S14.2-6

(iii) Y(W) = -- X(Q)P(w - Q)dQ

P(w) is an impulsive spectrum, as shown in Figure S14.2-7, because the
corresponding p(t) is periodic. (Note that only odd harmonics are
present.)

P(w)

-31r I i
3

-5 7 -7r T S5.

Figure S14.2-7
Signals and Systems
S14-6

Therefore Y(w) is as shown in Figure S14.2-8.

S14.3
(a) ii
(b) i
(c) iii
(d) vi
(e) v
(f) iv
(g) vii
(h) x
(i) ix
(j) viii

S14.4
(a) We are considering
N-I
X(Q) = Z x[n]e'-j,
n=O

which is effectively the Fourier transform of a signal of infinite duration mul-

tiplied by a window of length N:

X(Q) = cos wonT(u[n] - u[n - N])e -jn

From the convolution theorem we can compute the Fourier transform of the
product of these two sequences:

cos wonT - x[b(Q - woT) + b(O + woT)], -w < 0 < i

9r 1 - e -jfN ei-j(N-1)/2 sin N/2
u[n]-u[n-] 1 - e- sin 0/2
 Demonstration of Amplitude Modulation / Solutions
S14-7

Therefore,

X( = e -j(--woT)(N-1)/2 sin[N(Q - woT)/21 + 1 e -j(g+w 0T)(N-1)/2 sin[N(O + woT)/2]

2 sin[(Q - wOT)/21 2 sin[(Q + woT)/2]
as shown in Figure S14.4-1. (Note that the spectrum is periodic with period
27r.)

IX(G2)I

0
-o 0 T 007 7

Figure S14.4-1

N-I

(b) X(Qk) = x[ne -j"k*

n=O

X (2;k cos wonTe -j(2rk/N)n

N)
N-1 N-1
- 1 jwonTe -j(2ik/N)n -JwonTe -j(2ik/N)n
2 2~
n=O

S( j(oT- 2k/N)N) 1 j( ~w1oT- 2k/N)N

2 1 - ej(woT- 2wk/N) 2 1 -ej(-woT- 2k/N)

(i) For woT = 27r(2) and N = 5, the first term is zero for
k =. . . -3, 2, 7, . . .
However, when k = 2 we have the ratio of
1 ei 21(2 / 5 -k/ 5 ) 5 0
2 1 e j21(2/5--k/5) 0
Signals and Systems
S14-8

and we treat the limit as k - 0. Using L'H6pital's rule, we have i(5) =

2.5. Similarly, the second term is zero except when k = ... -2, 3, 8, . . . .
Taking the limit yields 2.5. So X(27rk/5) is as shown in Figure S14.4-2.

X( 27rk)
5

2.5 --

k
0 1 2 3 4

Figure S14.4-2

Note that X(27rk/5) is periodic in k with period 5 since X(Q) is periodic in

9 with period 27r.

( ( 27k) 1 (1 - e j(WOT 2wk/N)N ( - eji(w0T -2xk/N)N

(i)N ) 2 1 - e itw0T-21rk/N) 2 j(- w0T- 2xk/N)

Now woT = 2r-3, and the numerator and denominator are nonzero for all
k. Evaluating the preceding expression yields X(k) as shown in Figure
S14.4-3.

IX(k)|

3--

k
0 1 2 3 4

Figure S14.4-3
MIT OpenCourseWare
http://ocw.mit.edu

Resource: Signals and Systems

Professor 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.
15 Discrete-Time Modulation
Solutions to
Recommended Problems
S15.1
Recall that the Fourier transform of a train of impulses p(t) is P(w), as shown in
Figure S15.1-1.

P(w)

4T 27 0 27 4in
T T T T

Figure S15.1-1

Since x,(t) = x(t)p(t),

X,(w) = -2 KX(O)P(w - 0) dO

by the modulation property. Thus, X,(w) is composed of repeated versions of X(W)

centered at 27rk/T for an integer k and scaled by 1/T, as shown in Figure S15.1-2.

X,(W)
1
T

27r 7T
_7' 27T
T 2T 2T T
Figure S15.1-2

Since Xr(w) = X,(xw)H(w), it is as indicated in Figure S15.1-3.

Thus
1 1
X,(w) = X(wo) or x, = x(t)

S15-1
Signals and Systems
S15-2

S15.2
For go = ir/2, C(Q) is given as in Figure S15.2-1.

C(n)

-27r li31
T 7lI r 1 7
2
2 2 -~ I y f

Figure S15.2-1

By the modulation theorem,

9{x[n]c[n]) = 5{y[n]} = = - C(O) X(Q - 0) dO

Thus, Y(Q) is X(Q) centered on each impulse in Figure S15.2-1 and scaled by J, as
shown in Figure S15.2-2.

Y(R)

II I I I I
3If _T int 3nt 7T 3'r 21r
2 2 4 2 4 2

Figure S15.2-2

For go, = ir/4, C(Q) is given as in Figure S15.2-3.

C(92)

7T T 7IT

7_it it itf7
4 4 4 4

Figure S15.2-3

Thus, Y(Q) in this case is as shown in Figure S15.2-4.

 Discrete-Time Modulation / Solutions
S15-3

Y(u)
1
2

2
-2-r 21T
4 4 4 2 4

Figure S15.2-4

S15.3
From the lecture we know that the system in Figure S15.3-1 is equivalent to a filter
with response centered at 9 = 1r, as shown in Figure S15.3-2.

x[n] H(9) yyn]

(-1)"

Figure S15.3-1

H'(92)

S1
-­

4 r 14

Figure S15.3-2
Therefore, the total response is the sum of H'(Q) and H(Q), shown in Figure
S15.3-3.

I
I
a
I
II I
I

1' 3iw _i
4 4
Figure S15.3-3
Signals and Systems
S15-4

As an example, consider x[n] with Fourier transform X(Q) as in Figure S15.3-4.

X(92)

-7r if

Figure S15.3-4

Then, after multiplication by (-1)", the resulting signal has the Fourier transform
given in Figure S15.3-5.

F eIi
Figure S15.3-5

After filtering by H(Q), the resulting signal has the spectrum given in Figure
S15.3-6.

Figure S15.3-6

Finally, multiplying by (- 1)' again yields the spectrum in Figure S15.3-7.

-IT
A i
E
Figure S15.3-7
 Discrete-Time Modulation / Solutions
S15-5

Thus, the spectrum of y[n] is given by the sum of the spectrum in Figure S15.3-8
and X(Q), as shown in Figure S15.3-8.

Y(n)

FI S

Figure S15.3-8

S15.4
(a) P(Q) is composed of impulses spaced at 21r/N, where N is the period of the
sequence. In this case N = 2. The amplitude is 2 1ra,:
1
ak 2 T p[n]e j2kn2
n=O
= 1[le -j(2wkO/2) + Oe-j(2 kl/ 2 )I 2

Thus, P(Q) is as shown in Figure S15.4-1.

We now perform the periodic convolution of X(Q) with P(Q) and scale by
1/(21r) to obtain the spectrum in Figure S15.4-2.
Signals and Systems
S15-6

(b) To recover x[n] from y[n], we can filter y[n] with H(Q) given as in Figure
S15.4-3.

(c) Using p[n] we can send only every other sample of x 1 [n]. Similarly, we can send
every other sample of x2[n] and interleave them over one channel. Note, how­
ever, that we can do this only because X(Q) is bandlimited to less than 7r/2.

S15.5

We note that s(t) is a periodic signal. Therefore, S(w) is composed of impulses cen­
tered at (21rk)/T for integer k. The impulse at w = 0 has area given by 2rao, where
ao is the zeroth Fourier series coefficient of s(t):

ao =
T r
s(t) dt = j 1dt
A/2

-sis
= ­
A
T
Thus, S(w) is as shown in Figure S15.5-1.

2na_1
2 irA
T
1 2ra

2 27
TrT

Figure S15.5-1

The Fourier transform of x(t)s(t), denoted by R(w), is given by

R(w) = f 1 X(O)S(w - 0) dO = T, a.X W - 27rn

 Discrete-Time Modulation / Solutions
S15-7

If X(w) = 0 for I I > w/T, then R(w) will equal (A/T)X(w) in the region I co I < w/T.
Therefore, for H(co) as in Figure S15.5-2, the signal y(t) = x(t).

H(w)

T T

Figure S15.5-2

Solutions to
Optional Problems
S15.6
(a) Consider the labeling of the system in Figure S15.6.

x [n] X r [n] h [n] Iv [n] X

Xo y [n]

[n] r1 $2 [n]
Figure S15.6

r[n] = $1[n]x[n]
v[n] = E r[kjh[n - k] = $[kjx[kjh[n
0 - k]
k=- k=

y[n] = v[n] 2[n] 2[n] h[n - k]4 1[k]x[k]

k=-w

Suppose x1[n] = ax[n]. Then

yi[n] = 02[n] j h[n - k]# 1[k]ax[k] = ay[n]

k= -w

Now let x 2[n] = x1[n] + x 0 [n]. Then

y2[n] = 02[n] E h[n - k]41[k](x 1 [k] + xo[k]) = y 1[n] + yo[n]

k= -w

and the system is linear.

Signals and Systems
S15-8

If 4 1[n] = b[n], then

y[n] = 0 2 [n] 3 h[n - k]b[k]x[k] = 02[n]h[n]x[O]

k=-wo

If x[n] is shifted so x 1 [n] = x[n - 11, then

y1n] = 4 2[n]h[n]x1[0] = 4 2[n]h[n]x[-1] # y[n - 1]
and the system is not time-invariant.
(b) From part (a),

y[n] = Z ( h[n - k]z-x[k]

k=-wT

Let x[n - m] = xijn]. Then

y 1 [n] = z" h[n - k]z-kx1[k] = z- h[n - k]z-kx[k - m]

k= -w k= -w

Letp = k - m, k = p + m. Then

y1[n] = z" h[(n - m) - p]z-'-x[p]

= z m h[(n - m) - p]z-x[p]
p=-w0
= y[n - m]

Therefore, the system is time-invariant.

S15.7
In general, w(t) is recoverable from w,(t) if W,(w) contains repeated versions of
W(w) that do not overlap, i.e., that have no aliasing, as shown in Figure S15.7.

W,(C)

27r c _r 27r
T T T

W(W)

Figure S15.7

Since W(c) is repeated with period 27r/T, the largest frequency component of W(W),
Co,must be less than or equal to ir/T. From the modulation property,
1
W(W) = - X(W) * X 2(W)
27r
 Discrete-Time Modulation / Solutions
S15-9

Thus, since the length of a convolution of two signals is the sum of the individual
lengths,
c= W1 ± W2

From the preceding observations,

ir
- > W1 + W2 or T<
T W1 + A2

S15.8
(a) If a, = -12/21r, then the portion of X(Q) around Qj will be modulated down to
about 0 = 0 and then filtered by H(Q). We now need to reshift the spectrum
back to its original position. Therefore, we need to modulate by e2'i', or # =
+Qj/27r.
(b) Consider i = 0, 1. Then the corresponding filters are as given in Figure S15.8.

-20 0 E20 2r . 2 2 + E2
0
N N N 0

Figure S15.8

For no overlap and complete coverage of the frequency band, we need

27r
0 - 0, or 90 = ­
N N

S15.9
(a) Since s(t) is periodic in T, S(w) will consist of impulses located at 27rk/T. See
Figure S15.9-1.
Signals and Systems
S15-10

If Jos(t) = 0, then the spectrum looks like Figure S15.9-2.

S(W)

_ ___ t wt
6n _4r _2n 0 27r 47r 67r
T T T T T T
Figure S15.9-2

Of course, other impulses may also be zero.

(b) Y(w) will be equal to a sum of the shifted and scaled versions of X(w). Spe­
cifically,
S(27)x 2 9n
Y(w) = f X()S(w - 0) d = 1
27r -n 1 T T /(S15.9-1)

an X

where an is the nth Fourier series coefficient of one period of s(t). For some
region Y(w) to be zero, successive terms in the sum in eq. (S15.9-1) cannot over­
lap. Thus, the maximum T is such that /irT= w,, or T = 7r/we.
(c) In general, we need to find some n such that an # 0. Then we use an ideal real
bandpass filter to isolate the nth term of the sum in eq. (S15.9-1). The resulting
signal r(t) has Fourier transform R(w) given by

R(c) = anX 2 Tr) + aX o + 2)

Let an = rne"". Then r(t) can be thought of as

r(t) = x(t) + On
2r COS
I,.o (T
(remember the effect of modulating by a cosine signal). Suppose we multiply
r(t) by

I cos + O

Then

q(t) = r(t) -cos 2T + 0i = x(t) 2 cos 2(27rT

T + O

= x(t) 1 + cos 42rnt+ 20.)

If we now use a lowpass filter with cutoff wi/T, we get x(t). If we had picked the
smallest n such that an # 0, we could have avoided the bandpass filtering
because higher harmonics are eliminated by the lowpass filter.
 Discrete-Time Modulation / Solutions
S15-11

S15.10
(a) Y(Q) will consist of repeated versions of X(Q) centered at (21r/5) + 21rk and
scaled by 1. Thus, Y(Q) is as shown in Figure S15.10-1.

Y(n)

1
2

-iT 2nr_ 2 7
5 5

Figure S15.10-1

(b) Z(Q) will consist, in turn, of repeated versions of Y(Q), centered at (47r/5) +
21rk and scaled by 2, as shown in Figure S15.10-2.

Z(W)

-ote thatthe 2i 2 4
5 5 5 5
Figure S15.10-2

Note that the version of Y(Q) centered at 67r/5 contributes to the spectrum
between -3.7r/5 and ir.
(c) Two possible choices are given in Figures S15.10-3 and S15.10-4.
MIT OpenCourseWare
http://ocw.mit.edu

Resource: Signals and Systems

Professor 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.
16 Sampling
Solutions to
Recommended Problems
S16.1
If wo = 7r X 10', then
cos(won X 10-3) = cos(irn) =

Similarly, for wo = 31 X 10-3 and wo = 57 X 10-3,

cos((on X 10-3) = (-1)"

S16.2
The sampling function

p(t) = (t - nT), T = 13,

has a spectrum given by

2r =o -2rk
P(co) a WYE~k

= 67r ( (w - 61rk),

shown in Figure S16.2-1.

P(W)

67]

-127r -67r 0 67T 121r

Figure S16.2-1

cos(wot) has a spectrum given by rb(o - wo) + rb(o + wo), shown in Figure
S16.2-2.

cos (CO0 t)

F r S 1. 2

Figure S16.2-2

S16-1
Signals and Systems
S16-2

From the convolution theorem

1
=
2wr P(w) * [irb(w ­ wo) + gr(w + wo)]
Hence, it is straightforward to find X,(x).
(a) (i) For wo = 7r:

37r

-7n -6n -5r - nT 57 67 77

Figure S16.2-3

(ii) For wo = 2r:

37r

6 4
-81r - 7T - 7r -2In 2
7T
4
1r 67r 87r

Figure S16.2-4

(iii) For wo = 31r:

 Sampling / Solutions
S16-3

(iv) For wo = 5r:

Figure S16.2-6

(b) From part (a), it is clear that (i) and (iv) are identical.

S16.3
The signal x(t) = cos(wet + 0), where wo = 2wfo, can be written as
0
x(t) = lei'eiot + le -' e -jWOt
and the spectrum of x(t) is given by
X(w) = rej'"(w - wo) + re ~j'0 (w + wo)
The spectrum of p(t) is given by
2rk)
P(W)= 2
T k -0 T,
Therefore, the spectrum of x,(t) is

X,(W) =
2Tk -
Wo+ e -job (W -
2 rk+ )
/ L e6e
and the spectrum of X,(W) is given by
X,(w) = H(w)X,(w)

wr
(a) wo = 2r X 250, 0 =-) T = 10- ,
4,

X,,(w) = Tr [e'b(w - 2w X 10'k - 21 X 250)

Tk= -

2
+ e -j"S(w -
1 X 103k + 27 X 250)]
Hence, only the k = 0 term is passed by the filter:
X,(w) = w[ej'5(w - 2w X 250) + e -i"S(w + 21 X 250)]
and
25
X,(t) = 1 ei'ei2 x ot + 1 e -'e -22rX25o
2 2
= cos (2w X 250t + 0)

= cos (21 X 250t +

Signals and Systems
S16-4

(b) wo = 2w X 750 Hz, T = 10-,

2
X,(w) = >" [eb(w - 2w X 103 k - r X 750)
2 2
+ e -j'b(w - r X 103 k + x X 750)]
Only the k = 1 term has nonzero contribution:

X,(w) = r [ej"b(w + 27 X 250) + e ~ib(w - 2

r X 250)]
T

Hence,
X,(t) = cos (2w X 250t - 0)
= cos (27 X 250t -

(c) wo 27 X 500, = , T = 10- ,

X,(w) = ~ [ej0b(w - 2 r X 10 3 k - 2 r X 500)

2
+ e -j"b(w - 7 X 10 3k + 2
r X 500)]
Since H(w) = 0 at w = 2r X 500, the output is zero: x,(t) = 0.

S16.4

(a) x,(t) = x(t) 6(t - 2An) - x(t) 6(t - A - 2An)

= x(t) :b(t - 2An) - - A - 2An)I

L n =- o n -o t

By the convolution theorem,

1 2wr
X,(w) = X(w) 6 (

1 2 * 2-7
X(w) * - T -n -- ) e
27 2An _o

= X(W)* -
w,,

12(An)fl)
- n7

X,(w) is sketched in Figure S16.4-1 and Y(w) is sketched in Figure S16.4-2.

 Sampling / Solutions
S16-5

X,(w)

_3ff -r 7r 3L7r
3n2n n + , 2n 3n

Figure S16.4-1

Y(W)

Fi S1.4ff­
Figure S16.4-2

(b)

xP(t) X x(t)

-WM CJM

cos("f)

Figure S16.4-3

(c)

2A

yt X x(t)

-CJM liM

cos(Z)

Figure S16.4-4

(d) A is maximum when ir/A is minimum. From part (a) we see that aliasing is
avoided in X,(w) if wm zr/A. Hence, Amax = */Wu.
Signals and Systems
S16-6

S16.5
(a) The transform of the sampled function appears as in Figure S16.5-1.

Passband

-- IW

2n __n_ 0 x21T
T 2T 2T T

Figure S16.5-1

Hence, (a) matches (i).

(b)

Passband

2" "- "- 0 I T 21r

T T 2T 2T T T

Figure S16.5-2

Hence, (b) does not match any.

(c) Matches (ii).
(d)

M MKY7 2T 2T
M

Figure S16.5-3

Hence, (d) does not match any.

S16.6
Since the input x,(t) cannot be distinguished for certain values of w,the output also
should not be distinguishable for certain values of w. Hence, Q(w) must be periodic
in w. Therefore, Figure P16.6-3 is a possible candidate, but Figure P16.6-2 is not.
 Sampling / Solutions
S16-7

Solutions to
Optional Problems
S16.7

P(w)

... ..
4-A _ 21T 0 2w 4r
T T T T

Xp(w)

4
NA 0
NA
NAa N A
T NA 1 W2

2n
2T

T W2

Figure S16.7-1

Note that as T increases, (21r/T) - W2 approaches w = 0. Also, there is aliasing when

2xo - C 2 < (2r/T) - w2 < C 2. If 2 w1 - W2 >---0 (as given in the problem), then it is
easy to see that there is no aliasing when 0 ! (21r/T) - W2 ! 2oI - C 2 . For maximum
T, we choose a minimum allowable value of 21r/T:

- = x2 Tmax = 21,
Tax W2

which is sampling at half the Nyquist rate. X,(w) for this case is given in Figure
S16.7-2.

XP(W)

_ 2 0 2T 2T
T -T 2 T

Figure S16.7-2

Hence,
A = T, wb = 27r/T, Wa = WI
Signals and Systems
S16-8

S16.8

We are given the system shown in Figure S16.8-1.

xix(t) X2 (t)
x(t) - )(- H, (c) X, (t)

eIwot p(t)

Figure S16.8-1

(a) X(o) and Xi(o) are as shown in Figure S16.8-2. X 2(w) is as shown in Figure
S16.8-3, and X,(o) is therefore as given in Figure S16.8-4.

X(W) X1(Go)

I W- ________W,

2 Wl
W w2 (W1 - W2) (W 2 - 1 )
2 2

Figure S16.8-2

X 2 (o)

1W­ 02) (W2 -- W1)

2 2

Figure S16.8-3

Xp(w)

1
T

T T'

Figure 516.8-4
 Sampling / Solutions
S16-9

(b) 2x/Tmax equals the Nyquist rate for X 2 ((o):

2(W2 - Wi
= W2 - WI
2

Hence,
27
Tmax
(W2 - wi)

(c)

xP(t) x Re g X(t)
x2TH1(o)

e -'Ot
Figure S16.8-5

S16.9

The composite waveform spectrum is given in Figure S16.9-1.

We can alias the noise region to get maximum T. This corresponds to the aliased
spectrum, shown in Figure S16.9-2.

-2W -W 0 W 2W 3W 4W 5W
Figure S16.9-2

The value of T is given by

w-= 3W- Tax =­

T "'' 3W
The value of A is Tmax for y(t) = x(t).
Signals and Systems
S16-10

S16.10
The spectra of x, 2(t), where T = ir/W, given in Figures S16.10-1 and S16.10-2, could
have generated x,(t):

X 1 (O)

T-­

-w 6
W W
2
W
6
W
2
6w
6

Figure S16.10-1

X 2 (O)

22 2 2
Figure S16.10-2

S16.11

(a) From the sampling theorem, 27r/T >_2W. Hence,

7Tr
7r Tmax
W W
Since

X,(W) = Tk X k ,

we require A = T for x,(t) = x(t).

The minimum value of We is W so that we do not lose any information, and
the maximum value of W is (27r/T) - W to avoid periodic spectral contribution.
(b) (i) X(w) = 0 for IwI > W Hence,

T.ax
TmT
= I , A = T, W< W < - W
W T
(ii) X(w) = 0 for coI > 2W. Hence,

T. = , A = T, 2W< W < - 2W
2W T
 Sampling / Solutions
S16-11

(iii) X(w) = 0 for |coI > 3W Hence,

Tmax = , A = T, 3W< We < - 3W

3W' T
(iv) X(w) = 0 for w I > W/10. Hence,
W 27r W
Tax - A=T, -
"W '10 T -1
MIT OpenCourseWare
http://ocw.mit.edu

Resource: Signals and Systems

Professor 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.
17 Interpolation
Solutions to
Recommended Problems

S17.1
It is more convenient to solve this problem in the time domain than in the frequency
domain. Since x,(t) = x(t)p(t) and p(t) is an impulse train, x,(t) is a sampled ver­
sion of x(t), as shown in Figure S17.1-1.

x,(t)

xP\ t

-4T 0 4T

Figure S17.1-1

x(t) = x(t) bt - nT)

n= -oo

= Z
n= -oO
x(nT) 5(t - nT)

Since y(t) = x,(t) * h(t) and x,(t) is impulsive, the convolution carried out in the
time domain is as shown in Figure S17.1-2.

y(t)

\t
-4T 0 5T
Figure S17.1-2

Here we have that x(t) is sampled by a rectangular pulse train as opposed to an

impulse train.

S17-1
Signals and Systems
S17-2

Since h(t) * (1/T)h(t), shown in Figure S17.1-3, is wider than the sampling
period T, the resultant w(t) is not a triangularly sampled version of x(t). w(t) con­
sists of the superposition of waveforms shown in Figure S17.1-4.

1- ­

\ t
0 T 2T
Figure S17.1-3

w(t)

--4T i 6T

Figure S17.1-4

We note that this superposition is actually a linear interpolation between the sam­
ples of x(t). For example, Figure S17.1-5 convolved with Figure S17.1-6 equals
Figure S17.1-7.

x,(t)

1 Ji1 t
0 1 2
Figure S17.1-5
 Interpolation / Solutions
S17-3

t t
1 1 0 1 2 3

Figure S17.1-6 Figure S17.1-7

Now adding the shifted and scaled triangles yields Figure S17.1-8, which we see is
the linear interpolation between samples of x,(t).

0 3

Figure S17.1-8

Now since (1/T)h(t) * h(t) is Figure S17.1-9, we expect that w(t) is the linear inter­
polation of x,(t) shifted right by T, shown in Figure S17.1-10.

w (t)

/000 1'0
-4T 6T
2T

Figure S17.1-9 Figure S17.1-10

Signals and Systems
S17-4

S17.2

y(t) in all cases is the superposition of two signals.

(a)
y 1 (t)

Figure S17.2-1

(b)
Y 2 (t)

- - - - -_ --
0 2 3

Figure S17.2-2

(c)
y 3 (t)

sinin(t 1)
T (t -1)

sin n (t - 2)
2 (t - 2)

Figure S17.2-3
 Interpolation / Solutions
S17-5

S17.3
(a) |
P(W)

4
XP(wj)

\ A'-Z
2Ir WCm
T TT
~+ WO

Figure S17.3-1

We have
22r
27 - COM> Win,

T
2
CO,> wm,

where w, is the sampling frequency. To ensure no aliasing we require that

T < 7r/om.
(b) To recover x(t) from x,(t), we must interpolate. We have previously shown that
by lowpass filtering, the spectrum is recovered, assuming that sampling has
been performed at a sufficiently high rate. The interpolation may be done in
many different ways, however, depending on the cutoff frequency we choose
for the lowpass filter. For example, any of the filters HI(o), H 2(w), and H3(w) in
Figures S17.3-3 to S17.3-5 may be used to interpolate x,(t), whose Fourier
transform is shown in Figure S17.3-2, to yield x(t).
Signals and Systems
S17-6

Xp(o)

2T 2T T7- T
Figure S17.3-2

HI(o)

2T 2T

Figure S17.3-3

H2(w)

7T 7 ITT7

T 2T 2T T

Figure S17.3-4

H3(w)

T 2T 2T T

Figure S17.3-5
 Interpolation / Solutions
S17-7

(c) If

x,(t) = ( x(nT) b(t - nT)

n= -00

and H(w) is an ideal lowpass filter whose Fourier transform is shown in Figure
S17.3-6, then

h(t) = T sin TWC sine c

==ct
irt 7r
and
Two
x(t) = xp(t) * h(t) = -- x(nT) sine (t - sr nT)
irn=-o

H(w)

-Coc 'oc

Figure S17.3-6

S17.4

xXt X X t g (t) No H(cj) - y (t)

p(t)

Figure S17.4

We want to choose H(w) so that the cascade of the two filters is an ideal lowpass
filter. In this example,

G(x) = 2 sin A/ e ~jwA2

so that

H(w) = e jA/2 IcoI < r/A,

2 sin wA/2
H(w) = 0 otherwise
Signals and Systems
S17-8

S17.5

X,(Go)

1
T

-41r X 104 0 rX 104 47X 104

X(E2)
1
T

-27r 0 7T 27r
Y(2)
1
T

-27r 4 0 2r

Yc()

0 X104

Figure S17.5

Solutions to
Optional Problems
S17.6

(a) We want ho[n] such that x,[n] * ho[n] = x 0[n]. We sketch ho[n] in Figure
S17.6-1.

ho [n]

0 N-I
Figure S17.6-1
 Interpolation / Solutions
S17-9

(b) xO[n] = x,[n] * ho[n]

If there is no aliasing, we can recover x[n] from xO[n] by proper filtering. Since

jQN . e jQ(N- 1)/2 sin N/2

Ho(Q) = 1 -- e -"
1 -u sin 0/2
we require

N/HO(Q), |A|
H(Q) =
0, 101 -<Iir

N(ejU(N -- 1)/2
sinW/2' ~N

0, N< |

(c)' hl[n] = (1/N) (ho[n] * ho[-n]), so hi[n] is a triangular discrete-time pulse, as

shown in Figure S17.6-2.

h, [n]

-N
..?TII ii 0
I it.. N

Figure S17.6-2

(d) We require that

N
forI|I <­
~N'
H(Q) =

0, for -7 < |Q I < ir

From part (c),

H () = N
N
)1
[
|HO(U)|2 =H( o
2
2
(sin 9/2
sin N/2)'2 N'

< I0I < r

N

S17.7

(a) Taking Fourier transforms of both sides of the LCCDE yields

jY(wO) + Ye(W) = 1,
1
Ye(o) = .,
1 + Jw'
Signals and Systems
S17-10

so yc(t) = e-'u(t). By examining the sampler followed by conversion to an

impulse train, we note that
y[n] = yc(nT) = e -" Tu[n]
(b) Since y[n] = e -nTu[n], we can take the Fourier transform to yield

Y(Q) = e -"T e -jfn

n=O

1 - e (T+jQ)

In order for w[n] = 6[n], we require W(R) = 1 for all Q. Thus,

Y(Q)H(Q) = W(Q) = 1,
H(Q) = 1 - e -(T+jQ)
Now since H(Q) = 2=_oo h[n]e -", we see by inspection that
h[O] = 1,
h[1] =-e
h[n]= 0, n # 0, 1

S17.8
(a) Since X,(w) = X(w)P(w), we conclude that X,(w) is as shown in Figure S17.8-1.

X,(w)

K s
HwH
Figure S17.8-1

X,(w) = X(w) 6(co - no,)

= Z X(nwo) 6(w - nwo)

(b) From the convolution theorem,

y(t) = x(t)*p(t)
and
2 rn
p(t) = - - Ws
 Interpolation / Solutions
S17-11

The fact that

{pxt)}= 6 - nw)
71= -OD

can be easily verified. Therefore,

1 02,7rn\
y(t) = x(t) *- T 5 Kt- n,
(s n =-OD
(c) We see from the sketch in Figure S17.8-2 that for no time-domain aliasing,
27r
T -- - T, so W_s ­
W, T

y (t)

I
WS

0
'// a
27 T
WS

Figure S17.8-2

(d) x(t) may be recovered from y(t), assuming that no time-domain aliasing
has occurred, by low-time filtering y(t) from t = - T to T and applying a gain
of oS.
MIT OpenCourseWare
http://ocw.mit.edu

Resource: Signals and Systems

Professor 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.
18 Discrete-Time Processing of
Continuous-Time Signals
Solutions to
Recommended Problems
S18.1
(a) Since x,(t) = xc(t)p(t), then X,(w) is just a replication of Xe(w) centered at mul­
tiples of the sampling frequency, namely 8 kHz or 27r8 X 10' rad/s. The sam­
pling period is T = 1/8000.

X,(W)

T
8000

-21r X 8000 21r X 4000 27n X 8000

Figure S18.1-1

(b) X(Q) is just a rescaling of the frequency axis, where 21r8 X 103 becomes 2 1r.
X(Q) is shown in Figure S18.1-2.

(c) Y(Q) is the product G(Q)X(Q). Therefore, Y(Q) appears as in Figure S18.1-3.

Y(2)
8000
6000

1 T
-2ff -- 4 u 48 2f

Figure S18.1-3

S18-1
Signals and Systems
S18-2

(d) Y,(w) is a frequency-scaled version of Y(w) but only in the range 0 = --w to 7,
as shown in Figure S18.1-4. Also note the gain of T.

Ye(W)

_ - 8000 -8000
r S 4

Figure S18.1-4

S18.2
(a) The maximum nonzero frequency component of H(w) is 5 00w. Therefore, this
frequency can correspond to, at most, the maximum digital frequency before
folding, i.e., Q = x. From the relation wT = Q,we get

T. = - =2 ms
500w
(b) Since w = 500w maps to Q = 7r, the discrete-time filter G(Q) is as shown in Figure
S18.2-1.

_' .- .04

Figure S18.2-1
(c) The complete system is given by Figure S18.2-2. Note the need for an anti-alias­
ing filter.
 Discrete-Time Processing of Continuous-Time Signals / Solutions
S18-3

S18.3
(a) Recall that Xc(w) is as given by Figure S18.3-1.

Xc(w)

7r IT

-10000r 100007T
Figure S18.3-1

X(Q) is given by eq. (S18.3-1) and Figure S18.3-2.

1 00 27rn)
X(Q) = -
Tn
E
= -oo
= 20000 Xc[20000(9 - 27rn)] (S18.3-1)

Ye(w) is given by eq. (S18.3-2) and Figure S18.3-3.

TX(wT), IwI <

T
cI(W) (S18.3-2)
elsewhere

Thus x(t) = y(t) in this case.

Signals and Systems
S18-4

(b) Xc(w) is as given in Figure S18.3-4.

Xc(w)

co
-54000r 540007r

Figure S18.3-4

We now use eq. (S18.3-1), shown in Figure S18.3-5.

2000OXc(20000(2 - 2n))
n= 0
200007r

54T
20

n= -1

n =+ 1

iT 27r

Figure S18.3-5

Thus, in the range ± r, X(Q) = 20000 E",,U X,[20000(9 - 2irn)] is given as in

Figure S18.3-6.

X(2 )

200007 -­

F41T
i4gr IS
20 20
Figure S18.3-6
 Discrete-Time Processing of Continuous-Time Signals / Solutions
S18-5

Using eq. (S18.3-2), we find Ye(w) as in Figure S18.3-7.

YJ(w)

-14000n 14000T
Figure S18.3-7

Note aliasing since 27000 Hz is above half the sampling rate of 20000 Hz.
(c) Xe(w) is as given in Figure S18.3-8.

tT74
Xc(w)

-34000ir 340007r

Figure S18.3-8

Again we use eq. (S18.3-1), shown in Figure S18.3-9.

Signals and Systems
S18-6

Thus X(Q) is given as in Figure S18.3-10.

Finally, from eq. (S18.3-2) we have Y,(w) shown in Figure S18.3-11.

Yc(w)

-6000i 600r

Figure S18.3-11

S18.4
It is required that we sample at a rate such that the discrete-time frequency 7r/ 2 will
correspond to c. The relation between 9, and c is Q, = ocTo. Thus, we require

= TO
W"C

As wc increases, demanding a wider filter, To decreases, and consequently the sam­

pling frequency must be increased. There are two ways to calculate Wa. First, since
we are sampling at a rate of

or = 4(c
(ir/2)/(Ac '

we need an anti-aliasing filter that will remove power at frequencies higher than
half the sampling rate; therefore wa = 2wc. Alternatively, we note that the "folding
frequency," or the frequency at which aliasing begins, is 0 = 7r. Since 9 = r/2 cor­
responds to wc, then r must correspond to 2w,.
 Discrete-Time Processing of Continuous-Time Signals / Solutions
S18-7

S18.5
2
(a) We sketch X(Q) by stretching the frequency axis so that 7 corresponds to the
sampling frequency with a gain of 1/TO. We then repeat the spectrum, as shown
in Figure S18.5-1.

X(O2)
1

7T 7T

Figure S18.5-1

After filtering, Y(Q) is given as in Figure S18.5-2.

Y(E2)

TO

3~ 3

Figure S18.5-2

(b) We see that Y(Q) looks like X(w) filtered and then sampled. The discrete-time
frequency is -/3. Again, 2-x corresponds to 27r/To, so 7/3 corresponds to ir/3T.
Thus, if x(t) is filtered by G(w) as given in Figure S18.5-3, then y[n] = z[n].

G(c)

3 To 3To

Figure S18.5-3
Signals and Systems
S18-8

Solutions to
Optional Problems
S18.6
(a) Since we are allowing all frequencies less than 1007r through the anti-aliasing
filter, we need to sample at least twice 100r, or 2 00r. Thus, 2 007r = 2-x/T or
To = 10 ms. To find K, recall that impulse sampling introduces a gain of 1/TO.
To account for this, K must equal To, or K = 0.01.
(b) (i) Since X(w) is bandlimited to 100-r, the anti-aliasing filter has no effect. The
Fourier transform of x,(t), the modulated pulse train, is given in Figure
S18.6-1.

Xp(2)
1 - 00
To

-4007 -1007T 1007 4007

Figure S18.6-1

Since To = 0.005, the sampling frequency is 4007r. After conversion to a

discrete-time signal, X(Q) appears as in Figure S18.6-2.

X(W)
200

-2w7r 2w
2 2

Figure S18.6-2

After filtering, Y(Q) is given by Figure S18.6-3.

Y(n)

200

3 3~

Figure S18.6-3
 Discrete-Time Processing of Continuous-Time Signals / Solutions
S18-9

(ii) There are three effects to note in D/C conversion: (1) a gain of To, (2) a
frequency scaling by a factor of T,, and (3) the removal of repeated spec­
tra. Thus, Y(w) is as shown in Figure S18.6-4.

Y(W)

IT IT
3To 3To

Figure S18.6-4

S18.7
After the initial shock, you should realize that this problem is not as difficult as it
seems. If instead of h[n] we had been given the frequency response H(Q), then
He(w) would be just a scaled version of H(Q) bandlimited to ir/T. Let us find, then,
H(Q). Using properties of the Fourier transform, we have

1
Y(Q) =- ei"YG) + X(Q)
2
Y(Q) 1
H(0) X(Q) 1 - ie-j"
Thus,
1
IH(Q)I =
\ - cos 0
( sin g
-<H() = -tan' 1 - cos

Therefore, the magnitude and phase of He(c) are as shown in Figure S18.7.

1
He(w)| = \/ -coswT' T'

0, elsewhere

-tan-'
( isinwT
I,icoswT
1 - T

0, elsewhere
Signals and Systems
S18-10

IHc(w) I

r iT
T T

4Hc(w)

/00000* I /.A

7T
T
V 7V
Figure S18.7

S18.8

The system under study is shown in Figure S18.8.

covrsot p(t) T[ ) x:c(t)W yc~t

W ypW ot) sont
x en
x [n] -- annipleConversion
impulse -- +ahst)uXnca to T[n e -qr y [n]

T T

( 8(t-nT)
n=­

Figure S18.8

From our previous study, we know that Xe(o) in the range ±ir/Tlooks just like
X(Q) in the range ±ir. Similarly, Y,(w) between -- r/T and +r/T looks like Y(Q) in
the range -ir to r. Although there is a factor of T, we can disregard it in analyzing
this system because it is accounted for in the H(w) filter. The transformation of xc(t)
to yc(t) will correspond to filtering x[n], yielding y[n]. In fact, the equivalent system
will have a system function H(Q) given by

H(Q) = He , |u| < r,

where He(w) is the Fourier transform of h(t). Thus, we need to find Hc(w). The rela­
tion between yc(t) and xc(t) is governed by the following differential equation:

d td + 4'(dyc
+y4t dyt ~ 3yc(tt)=t)
+ ) = xc(t )
 Discrete-Time Processing of Continuous-Time Signals / Solutions
S18-11

Using the properties of the Fourier transform, we have

2
(j) Ye(w) + 4(j)Ye(w) + 3Y0(w) = Xco),
1
He(w) =1 2
(jco) + 4jw + 3
Therefore,

H(Q) 2 + 4j + Jul <K

+ 4j- + 3

S18.9
(a) It is instructive to sketch a typical y,(t), which we have done in Figure S18.9-1.

Let us suppose that T is changed by being reduced. Then the envelope of y,(t)
seems to correspond to a higher-frequency cosine. At time kt,
2wk 2w(kT) 2wt
y,(t) = cos
N
- b(t - kT) = cos NT
NT 6(t - kT) = cos NT
6(t - kT),
-

where we use the sampling property of the impulse function. Thus,

y,(t) = Y cos (t - kT) = cos wot ( ot - kT),

k= -w0 k=-w0

where wo = 21/NT.
If the minimum wo is wi, and since T = 21/N(o,
2T
Tmax=
A) I
Similarly,

T.in =
NW2
Signals and Systems
S18-12

(b) Recall that sampling with an impulse train repeats the spectrum with a period
of 2r/T and a gain factor of 1/T. Since 5([cos(2t/NT)J is as given by Figure
S18.9-2, Y,(w) is then given by Figure S18.9-3.

7T 7T

2,n 27r
NT NT

Figure S18.9-2

T - NT T T

7rr 2n Nr i

2| r
27r 27/N - 1 ~-WO WO= 2rN - 1\ 21r
T T N T N T

Figure S18.9-3

(c) The minimum value of N is 2, corresponding to the impulses at o and (27r/T ­

wo) being superimposed at w/T. The lowpass filter cutoff frequency must be such
that the (superimposed) impulses at 7r/T are in the passband and those at 37r/T
are outside the passband. Consequently,
ir 3­
T T
(d) Comparing Y(w) and Y,(w) in Figures S18.9-2 and S18.9-3 respectively, we see
that for N > 2 the cosine output will have an amplitude of 1/T = w/21r. If N =
2, then the output amplitude will be 2/T = w/7r.

S18.10
(a) By sampling sc(t), we get
s[n] = setnT) = x(nT) + ax(nT - TO) = x(nT) + ax[(n - 1)T]
since T = To. Let x[n] = x(nT). Then
s[n] = x[n] + ax[n - 1]
Therefore
x[n] = -ax[n - 11 + s[n]
 Discrete-Time Processing of Continuous-Time Signals / Solutions
S18-13

This is a first-order difference equation, so given s[n], we can find x[n]. Since
x(t) is appropriately bandlimited, we can then set
y[n] = -ay[n - 1] + s[n]
which will make
A
yc(t) = -x(t)
T
(b) From part (a) we see that T = A will make y(t) = x(t).
(c) Since we do not want to alias, we still need T < 7r/wM. Now
s(t) = x(t) + ax(t - TO)

Taking the continuousFourier transform, we see that

S(w) = X(w) + ae -iwTox(w)
Thus, the continuous-time inverse system has frequency response
1
1 + ae ~-T0

We want to implement this in discrete time. Therefore, using the relation, we

obtain
1 -
(Q)
H(Q) = H, A =i1 + aaef1 jQ(T 0 /T)' ileM <M
s <
bM

Again, the filter should be A = T.

MIT OpenCourseWare
http://ocw.mit.edu

Resource: Signals and Systems

Professor 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.
19 Discrete-Time Sampling
Solutions to
Recommended Problems
S19.1
x[n] is given by
x[n] = (-1)" = e'"
Hence, the Fourier transform of x[n] is

X(Q) = 57 b(O - r - 2-xk)

k= -c

Now p[n] can be written as

1 + (-1)"
p[n] = 2
Hence, its Fourier transform is given by

P = 0 - 21rk) + 6(9 - r - 2x7rk)

2k= -oo

It is clear that x,[n] = p[n]. Hence

X,(Q) = P(Q)

S19.2
(a) x,[n] is x[n] "stretched" by interspersing with zeros, as indicated in Figure
S19.2-1.

x [n] x,[n ]

0
n 0
0
1.
Figure S19.2-1

X,()= ( xjnle -1""

n =-00

= x,[2nle -ji2n + x,[2n + l]e -j"l**+

nt= -o

= ( x[nje -j(21)n + 0
= = -X

=X(29)

S19-1
Signals and Systems
S19-2

(b) xd[n] = x[2n],

Xd(Q) = x[n]e -jfln

= ( x[2n]e-jn

= L (x[n] + (-1)"x[n])e -jln/2

-j(fl/2)n + __ 00 _xne j[(12/2)-In

= __ x[n]e
( + : x[n]e
(2n=
X +
X ( - ,r)

(c) X8,(Q) =X(2Q)

XS(92)

Figr S IT
-7r 8 -9 IT

Figure S19.2-2

Xd(Q) = X + x
X - ,r)

'X(Q/2) is indicated in Figure S19.2-3. Therefore, Xd(Q) is as shown in Figure

S19.2-4.

1 X(2)
2 2
1
2

7rI IT 7T

Figure S19.2-3

Xd(E2 )

-27r -I - 0 T 27r

Figure S19.2-4
 Discrete-Time Sampling / Solutions
S19-3

S19.3
(a) For N = 1, p[n] = 1. Hence

P(Q) = 2(r b(Q - 2rk),

k= -- o

as shown in Figure S19.3-1.

P(Q)
214

-27 0 2-n

Figure S19.3-1

For N = 2,

p[n] = 6[n - 2k]

k=­

Hence

P(Q) = ( O - 7rk),
k= -o

shown in Figure S19.3-2.

-IT 0 iT

Figure S19.3-2

ForN =L

p[n] = ( b[n - LkI

k = -w

Hence
-2rk
P(Q) = k=

shown in Figure S19.3-3.

Signals and Systems
S19-4

(b) X,(Q), the spectrum of x,[n], is proportional to the periodic convolution of P(Q)
and X(Q). Consequently, with P(Q) as indicated in Figure S19.3-3 and X(Q) as
indicated in Figure S19.3-4, X,(Q) is shown in Figure S19.3-5. In order that x[n]
be reconstructible from x,[n] using an ideal lowpass filter, aliasing must be
avoided, which requires that

9M <N -M, or UM <

N N

X(92)

-21 -M M 2T

Figure S19.3-4

P(92)

FiM u M 2.3-5
N

Figure S19.3-5

(i) Qm = 37r/10. Therefore, to avoid aliasing,

> 37 10
or <
N10
Since N must be an integer, we require that N s 3. For N = 3, the cutoff
frequency of the lowpass filter must be greater than 3r/10 and less than

3 10 3)10
5
(ii) Q, = 37r/ . To avoid aliasing,
> 3w
or N<
N 5 3
Since N must be a positive integer, this requires that N = 1, i.e., x[n]
cannot be sampled.
 Discrete-Time Sampling / Solutions
S19-5

S19.4
(a) The sampling period Ti is 3 ms for the system in Figure P19.4-2 to be equivalent
to the one in Figure P19.4-1.
(b) X(Q) is sketched in Figure S19.4-1.

IX(92)I

-2T2 2T

Figure S19.4-1

From the result of part (a), Y(Q) is as shown in Figure S19.4-2.

|Y(G)|

| l- ' e I ­

-27T 31T 37T 2n

2 2

Figure S19.4-2

S19.5
(a) Consider xd,[n] and Xdfn], and let
Xd3[n] = Xd,[n] + aXd in]
Then
± aXdn/N], n = 0, +N,
xP3 [n] = XdXnl/N]
0, otherwise
But

x,,[n] =
0, otherwise
And

=I d[/ ,
ax,2[n]
otherwise
Signals and Systems
S19-6

Hence,
Xd[f/N] + aXd2 [n/N], n = 0, ±N,...,
x,[n] + axP2[n] =
0, otherwise
and
xP3 [n] = xp,[n] + ax 2 [n]
So system A is linear.
(b) Take xd,[n] as shown in Figure S19.5-1, with N = 4.

-1 0 1 2
Figure S19.5-1

Then x,,[n] is as shown in Figure S19.5-2.

0 4 8
Figure S19.5-2

Take Xd2[n] = Xd,[n + 1]. Then x,,[n] is as shown in Figure S19.5-3.

-4 0 4
Figure S19.5-3

Hence, system A is not time-invariant.

(c) x,[n] =
xd[n/N], n = 0, N,... ,
0, otherwise
Hence

X,()= xd[n]e -jl(Nn) = Xd(NQ),

as shown in Figure S19.5-4.

 Discrete-Time Sampling / Solutions
S19-7

X (E2)

7T IT
3 3
Figure S19.5-4

(d) X(Q) is as shown in Figure S19.5-5 for exact bandlimited interpolation.

X(Q)

3 F-

Figure S19.5-5

S19.6

(a) x,(t) is sketched in Figure S19.6-1, and Y,(w) is sketched in Figure S19.6-2.

-- - .....x~(t)

-2To -TO To 2 To

Figure S19.6-1

Y,(w)
-H-
­

-2wo -- 0 CO 2coo

Figure S19.6-2
Signals and Systems
S19-8

(b) X,(w) is sketched in Figure S19.6-3, and y,(t) is sketched in Figure S19.6-4.

X,(o)
1
To

2
17 7T 7T 7r
2
TO 2TO To
Figure S19.6-3

(c) Yes, y,(t) is periodic and this is reflected in Y,(w), which contains impulses.

Solutions to
Optional Problems
S19.7
n =0, 2, +4,. . .,
(a) x,[n] =x[n,
0, n =±1, 35,...
This is sketched in Figure S19.7-1.

Ip[nn

0
Figure S19.7-1

Similarly, xd[n] = x[2n], as shown in Figure S19.7-2.

 Discrete-Time Sampling / Solutions
S19-9

Figure S19.7-2

(b) X,(Q) is obtained as follows:

x,[n] = x[n] + 2(- 1)"x[n]
XCQ) - iX(Q) + jX(Q - ir)

and
1 -
Xa(Q) = X- x- 1 ),

which are shown in Figures S19.7-3 and S19.7-4. See Problem P19.2(b).

Xp(E2)

4 4 4 4

Figure S19.7-3

Xd(n)= Xp ()

2 2

Figure S19.7-4

S19.8
(a) We know that the Fourier transform of p[n] is given by
27rk)
k= -0
Signals and Systems
S19-10

Aliasing will be just avoided when the sampled spectra will look as shown in
Figure S19.8-1.

Hence, we require that

2>r 67r
or N< 2= N - 3
N> 11' 6
Consequently, aliasing is avoided if 1 S N 3. X,(Q) for N = 1, 2, and 3 are
shown in Figure S19.8-2.

N ='1A

121

N1 XP(2)
=1227
1
N=2 2

-27r
1
I7
37r_ 31
i
F - I
27r

--2 _42 _2\ 3n 1 3l 2n 2

N=3

N 3
3
-21T 47r -27r - 7r 3In 2r 4_ 21T
3 3 11 0 11 3 3

Figure S19.8-2
 Discrete-Time Sampling / Solutions
S19-11

(b) An appropriate H(Q) is shown in Figure S19.8-3.

H(2)

3nr 37r

Figure S19.8-3

S19.9

(a) y] x[3n] + x[3n + 1] + x[3n + 2]

3
y[-4]= 0
y[-3] = 0
y[-2] =x[-4] =1

x[-3] + x[-2] + x[-1] 2

y[-1] =
3
[0] = x[0] + x[1] + x[2] _
3
x[3] + x[4] + x[5] 1
y[1] =
y[2] = 0
y[3] = 0
Hence, y[n] can be sketched as in Figure S19.9.

4
3

-4 -3 -2 -1 0 1 2 3

Figure S19.9

(b) If
z[n] = [x[nj + x[n + 11 + x[n + 2]], for all n
Signals and Systems
S19-12

and
y[n] = z[3n],
we have expressed the processing as a combination of filtering and decimati

S19.10
If h[O] = 1 and h[n] = 0 for n = kN, k # 0, it is easy to see that the samples
xo[n] that came from x[n] will be unaffected. Hence,
y[kN] = x[k], for all k
MIT OpenCourseWare
http://ocw.mit.edu

Resource: Signals and Systems

Professor 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.
20 The Laplace Transform
Solutions to
Recommended Problems
S20.1
(a) The Fourier transform of the signal does not exist because of the presence of
growing exponentials. In other words, x(t) is not absolutely integrable.
(b) (i) For the case a = 1, we have that
x(t)e -' = 3e'u(t) + 4e 2 u(t)

Although the growth rate has been slowed, the Fourier transform still
does not converge.
(ii) For the case a = 2.5, we have that
x(t)e-'' = 3e-0-5 t u(t) + 4e 0 5 tu(t)
The first term has now been sufficiently weighted that it decays to 0 as t
goes to infinity. However, since the second term is still growing exponen­
tially, the Fourier transform does not converge.
(iii) For the case a = 3.5, we have that
x(t)e -"' = 3e -' 5 t u(t) + 4e - 0 "'u(t)
Both terms do decay as t goes to infinity, and the Fourier transform con­
verges. We note that for any value of a > 3.0, the signal x(t)e -' decays
exponentially, and the Fourier transform converges.
(c) The Laplace transform of x(t) is
3 4 7(s -7)
X( s-2 s-3 (s - 2)(s - 3)'

and its pole-zero plot and ROC are as shown in Figure S20.1.

Im
s plane

--- -F/ Re
2 17 3
7

Figure S20.1

Note that if a > 3.0, s = a + jw is in the region of convergence because, as we

showed in part (b)(iii), the Fourier transform converges.

S20-1
Signals and Systems
S20-2

S20.2

(a) X(s) = e -at u(t)e 't dt =

T Ls + a
The Laplace transform converges for Re~s} + a > 0, so
o + a>0,0 or o> -a,
as shown in Figure S20.2-1.

Im

s plane

-4w 0 Re

Figure S20.2-1

(b) X(s) =
The Laplace transform converges for Re~s) + a > 0, so
o + a>0,0 or o>-a,
as shown in Figure S20.2-2.

Im
s plane

Re

Figure S20.2-2

-e (s+a)t dt = e(s+a) 0
(c) X(s) = -- e atu(-t)e - dt =
JW- s + a­

s + a
if Re~s) + a < 0, o + a < 0, , < -a.
 The Laplace Transform / Solutions
S20-3

Figure S20.2-3

S20.3
(a) (i) Since the Fourier transform of x(t)e ~'exists, a = 1 must be in the ROC.
Therefore only one possible ROC exists, shown in Figure S20.3-1.

(ii) We are specifying a left-sided signal. The corresponding ROC is as given

in Figure S20.3-2.

Im

s plane

XRe
-2 2

Figure S20.3-2
Signals and Systems
S20-4

(iii) We are specifying a right-sided signal. The corresponding ROC is as given

in Figure S20.3-3.

Im

s plane

)X Re
-2 2

Figure S20.3-3

(b) Since there are no poles present, the ROC exists everywhere in the s plane.
(c) (i) a = 1 must be in the ROC. Therefore, the only possible ROC is that shown
in Figure S20.3-4.

(ii) We are specifying a left-sided signal. The corresponding ROC is as shown

in Figure S20.3-5.

Im
s plane

0 Re
-2 2

Figure S20.3-5
 The Laplace Transform / Solutions
S20-5

(iii) We are specifying a right-sided signal. The corresponding ROC is as given

in Figure S20.3-6.

(d) (i) a = 1 must be in the ROC. Therefore, the only possible ROC is as shown
in Figure S20.3-7.

s plane

- Re

Figure S20.3-7

(ii) We are specifying a left-sided signal. The corresponding ROC is as shown

in Figure S20.3-8.

Im

2 s plane

Re

-2

Figure S20.3-8
Signals and Systems
S20-6

(iii) We are specifying a right-sided signal. The corresponding ROC is as

shown in Figure S20.3-9.

s plane

Figure S20.3-9

Constraint on ROC for Pole-Zero Pattern

x(t) (a) (b) (c) (d)
(i) Fourier
transfor_ -2 < < 2 Entire s plane a> -2 a > 0
converges

(ii) x(t) = 0, a< -2 Entire s plane o< -2 a < 0

t > 10

(iii)x(t) =0 o> 2 Entire s plane a> -2 a> 0

Table S20.3

S20.4
(a) For x(t) right-sided, the ROC is to the right of the rightmost pole, as shown in
Figure S20.4-1.

Im

s plane

Re

Figure S20.4-1
 The Laplace Transform / Solutions
S20-7

Using partial fractions,

1 = 1 1
(s+ 1)(s + 2) s+ 1 s + 2'
so, by inspection,
x(t) = e -t u(t) - e 2t
U(t)
(b) For x(t) left-sided, the ROC is to the left of the leftmost pole, as shown in Figure
S20.4-2.

Im
s plane

2u 1- Re

Figure S20.4-2

Since
- 1
X(S) =
s +1 s+ 2
we conclude that
x(t) = -e - t u(-t) - (-e 2- u(-t))
(c) For the two-sided assumption, we know that x(t) will have the form
fi(t)u(- t) + fA(t)u(t)
We know the inverse Laplace transforms of the following:
1 e ~'u(t), assuming right-sided,
S+ 1 I-e'u(-t), assuming left-sided,
1 e- u(t), assuming right-sided,
s + 2 -e -2'U(-t), assuming left-sided
Which of the combinations should we choose for the two-sided case? Suppose
we choose
x(t) = e -u(t) + (-e -2')u(-t)
We ask, For what values of a does x(t)e -0' have a Fourier transform? And we
see that there are no values. That is, suppose we choose a > -1, so that the
first term has a Fourier transform. For a > -1, e - 2 te -'' is a growing exponen­
tial as t goes to negative infinity, so the second term does not have a Fourier
transform. If we increase a, the first term decays faster as t goes to infinity, but
Signals and Systems
S20-8

the second term grows faster as t goes to negative infinity. Therefore, choosing
a > -1 will not yield a Fourier transform of x(t)e -'. If we choose a 5 -1, we
note that the first term will not have a Fourier transform. Therefore, we con­
clude that our choice of the two-sided sequence was wrong. It corresponds to
the invalid region of convergence shown in Figure S20.4-3.

Im

s plane

Re
-2 -- -0 '/

Figure S20.4-3

If we choose the other possibility,

x(t) = -e-u(-t) - e 2n(t),

we see that the valid region of convergence is as given in Figure S20.4-4.

Im

splane
-, Re
-2 1 0

Figure S20.4-4

S20.5
There are two ways to solve this problem.
Method 1
This method is based on recognizing that the system input is a superposition of
eigenfunctions. Specifically, the eigenfunction property follows from the convolu­
tion integral

y(t) = h(r)x(t - r) dr

Now suppose x(t) = e". Then

y(t) = f h(r)ea(" r)dr = e'" h(r)e -"dr

 The Laplace Transform / Solutions
S20-9

Now we recognize that

f h(r)e ~a' dr = H(s) ,s=a

so that if x(t) = e ", then

y(t) H(s) sale't,

i.e., e a' is an eigenfunction of the system.

Using linearity and superposition, we recognize that if
x(t) = e -t/2 + 2et/3
then

y(t) = e -t1 2H(s) + 2e-' 13H(s)

s= -1/2 s= -1/3

so that
y(t) = 2e -t/2 + 3e ~'/' for all t.

Method 2
We consider the solution of this problem as the superposition of the response
to two signals x 1 (t), x 2(t), where x 1 (t) is the noncausal part of x(t) and x 2 (t) is the
causal part of x(t). That is,
xi(t) = e -t/2U(-t) + 2e t /3 U( -),
x 2(t) = e - t 2u(t) + 2e - t 3u(t)
This allows us to use Laplace transforms, but we must be careful about the ROCs.
Now consider L{xi(t)}, where C{-} denotes the Laplace transform:
1 2 1
-Lxi(t)} = Xi(s) - _ - , Re{s) < -1
s+ s+ 3 2

Now since the response to x 1(t) is

y 1(t) = £ '{XI(s)H(s)},
then
1 2
Yi(s) -2 -1 < Res) <
(s+1)(s+i) (s+i)(s+1)' 2'
2 -2 -3 3
s+1 s+i s+I s+1'
5 2 3
s+1 s+i s+1'

y 1 (t) = 5e - tu(t) + 2e -t /2 U(- t) + 3e -t /3 u(- t)

Signals and Systems
S20-10

The pole-zero plot and associated ROC for Yi(s) is shown in Figure S20.5-1.

Im

s plane

Re
-1__u1

Figure S20.5-1

Next consider the response y 2 (t) to x 2(t):

x 2(t) = e -t/2U(t) + 2e ~'3 u(t),
1+1 2 i, 1
X2(s) s
S 1-IS-
+ Re{s} > - '
+

Y2 (s) = X 2(s)H(s) = 1 + 2
(s + 2)(s + 1) (s + 1)(s+ 1)'
+ 2 3 -3
Y2(s) =
s+i s+1 s+i s+1'

t 3
y 2 (t) = -5e -'u(t) + 2e ~1 /2U(t) + 3e - / u(t)

The pole-zero plot and associated ROC for Y2(s) is shown in Figure S20.5-2.

s plane
3
2

FY
Figure S20.5-2

Since y(t) = y 1 (t) + y 2(t), then

y(t) = 2e t~ / 2 + 3e ~'/3 for all t

S20.6

(a) Since

X(s) = x(t)e -s dt
 The Laplace Transform / Solutions
S20-11

and s = a + jw, then

X(s)
s=,+j
= J ­
x(t)e -"'e-" dt

We see that the Laplace transform is the Fourier transform of x(t)e -'' from the
definition of the Fourier analysis formula.

(b) x(t)e -" = X[s) e'f' dw

27r -_ ,+jo

This result is the inverse Fourier transform, or synthesis equation. So

x(t) = e 27r -oo ~

X(s) l,4j.] 1ei' dw

=-I X(s) (+jw)t dw,

and letting s = a + jw yields ds = j dw:

x(t) = -2 rj
X(s)e' ds
_-joo

Solutions to
Optional Problems
S20.7
1
(a) X(s) = , Refs) > -1
s + 1
Therefore, x(t) is right-sided, and specifically
x(t) = e -u(t)
1
(b) X(s) = , Re{s) < -1
+ 1s
Therefore,
x(t) = -e- t u(-t)
s
(c) X(s) = Refs) > 0
S2 + 4'
Since
. 1
eisot
S -j W
L 1
s + jw 0
1. 1(1 1=+
L{cos(wot )u(t)} = £-e*ot + - e-"=- . +
12 22 Jo s+jo
s
L{cos(wot)u(t)) 2 + W2

if X(s) = 2s then x(t) = cos(2t)u(t)

Signals and Systems
S20-12

s+1 _ s+1 -1 2
(d) X(s) = 2=- + ,so
s 2 + 5s + 6 (s + 2)(s + 3) s + 2 s + 3
x(t) = -e - 2 tu(t) + 2e - 3 u(t)
s+1 -1 2
(e) X(s)
(s + 2)(s + 3) s + 2 s + 3'
x(t) e - 2 tu(-t) - 2e - 3 u(-t)
s - s +1
(f) X(s) = 0 < Re{s} < 1
s2(s - 1)

s-i1 s(s -1) s2 s-_1)

+ 1 -1+ -1 +-1
s
s-s s s2 + s 1
S1 1

t
x(t) = -e u(-t) - tu(t)
S2 _ S + 1
(g) X(s) = (S + 1)2 > < Res}
S(s + 1) 2 - 3s 3s
(S + 1)2 (S + 1)2
3(s + 1) 3
(S + 1)2 (S + 1)2>
x(t) = b(t) - 3e -'u(t) + 3te-'u(t)

s + 1
(h) X(s) =
(s + 1)2 + 4
Consider

Y(s) = 2 s - y(t) = cos(2t)u(t) from part (c)

s + 4

Now

f(t)e ~ F(s + a),

x(t) = e cos (2t)u(t)

S20.8
The Laplace transform of an impulse ab(t) is a.Therefore, if we expand a rational
Laplace transform by dividing the denominator into the numerator, we require a
constant term in the expansion. This will occur only if the numerator has order
greater than or equal to the order of the denominator. Therefore, a necessary con­
dition on the number of zeros is that it be greater than or equal to the number of
poles.
This is only a necessary and not a sufficient condition as it is possible to con­
struct a rational Laplace transform that has a numerator order greater than the
 The Laplace Transform / Solutions
S20-13

denominator order and that does not yield a constant term in the expansion. For
example,
s2 + 1 1
X(s) - ,
s s

which does not have a constant term. Therefore a necessary condition is that the
number of zeros equal or exceed the number of poles.

S20.9
(a) x(t) = e -a'u(t), a < 0,

X(s) = ,
s + a'
and the ROC is shown in Figure S20.9-1.

(b) x(t) = -e a'u(-t), a > 0,

= 1,
X(S)
s -- a
and the ROC is shown in Figure S20.9-2.

Figure S20.9-2
Signals and Systems
S20-14

(c) x(t) = e a'u(t), a > 0,

1
X(s)- s- a'

and the ROC is shown in Figure S20.9-3.

s plane

Figure S20.9-3

t
(d) x(t) = e -a" I a > 0,
t
= e -a u(t) + e a'u(- t),
1 -1
X(S) = + -,
s+a s-a'
and the ROC is shown in Figure S20.9-4.

Im

s plane

-a

Figure S2 0.9-4

(e) x(t) = u(t),

X(s) = fe -S dt=-,
0 s
and the ROC is shown in Figure S20.9-5.
 The Laplace Transform / Solutions
S20-15

s plane

Figure S20.9-5

(f) x(t) = btt - to),

X(s) = fb(t - to)e -' dt = e ~sto

and the ROC is the entire s plane.

(g) x(t) = T0 a k t-kT ),

k=0

X(s) = ( ak t
6(t - kT)e -s dt
k=0

= Lake -skT = 1
k=0

with ROC such that ae-T < 1. Now

1
a2 -2sr < 1 - 2 log a - 2sT < 0 - s > - log a
T

(h) x(t) = cos (wot + b)u(t)

Using the identity
cos(a + b) = cos a cos b - sin a sin b

we have that
x(t) = cos b costwot)u(t) - sin b sin(wot)u(t)

Using linearity and the transform pairs

£ S
cos(W0 t) 2 +2 ,

Wo
sin(wot) 2 + W2 ,
s2 0
Signals and Systems
S20-16

we have

X(s) = cos b s - sin b 2o2

S 0x S 0o

[s - (tan b)wo]
X(s) =cosb
s2 + W2 '
and the ROC is shown in Figure S20.9-6.

s plane

- Re

--o0

Figure S20.9-6

(i) Consider

xi(t) = sin(wot + b)u(t)

= (sin wot cos b + cos wot sin b)u(t)

Using linearity and the preceding sin w0t, cos wot pairs, we have

X 1(s) = cos b + sin b S

s2 + W2 s2 + oC '

[s + (cot b)wo]
X 1 (s) = sin bW

Using the property that

f(t)e ~4'- F(s + a),

we have

X(s) = sin b [s + a + (cot b)wO]

(s + a)2 + o'
with the ROC as given in Figure S20.9-7.
 The Laplace Transform / Solutions
S20-17

Im

-a + jo
span~e

Re
-(a + wo cot b)'

-a -jcoo

Figure S20.9-7

S20.10

(a) X(s) = x(t)e-" dt

Consider

X 1(s) = t)e S" dt

Letting t = -t', we have

X1(s) = x(t')e"' dt'

=X(-s),
but X 1(s) = X(s) since x(t) = x(-t). Therefore, X(s) = X(-s).

(b) X(s) = x(t)e -" dt

Consider

X 1(s) = -x(-t)e - dt,

X 1(s) = -x(t'e"' dt'

=-X(S),
but X 1(s) = X(s) since x(t) = -x(--t). Therefore, X(s) = -X(-s).
(c) We note that if X(s) has poles, then it must be two-sided in order for x(t) =
x(-t).
Ks
(i) X(s) = ( s
(s + 1)(S - 1),
-Ks -Ks
X(-s) = = # X(s),
(-s + 1)(-s - 1) (s - 1)(s + 1)
so x(t) # x(- t).
Signals and Systems
S20-18

(ii) X(S) =K(s + 1)(s - 1)

S

X(-s) K(-s + 1)(-s - 1 ) 0 X(s)

-s
Also, this pole pattern cannot have a two-sided ROC.

(iii) X(S) = K(s + j)(s - j)

(s + 1)(s - 1)'
K(-s +j)(-s -j) _K(s - j)(s + j) S
X(-s) = = =-~s
(-s + 1)(-s - 1) (s - 1)(s + 1)
so this can correspond to an even x(t). The corresponding ROC must be
two-sided, as shown in Figure S20.10-1.

(iv) This does not have any possible two-sided ROCs.

(d) We see from the results in part (c)(i) that X(s) = -X(-s), so the result in part
(c)(i) corresponds to an odd x(t) with an ROC as given in Figure S20.10-2.

Im

s plane

Re

Figure S20.10-2

Parts (c)(ii) and (c)(iv) do not have any possible two-sided ROCs. Part (c)(iii) is
even, as previously shown, and therefore cannot be odd.
MIT OpenCourseWare
http://ocw.mit.edu

Resource: Signals and Systems

Professor 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.
21 Continuous-Time
Second-Order Systems
Solutions to
Recommended Problems
S21.1

(a) H 2(s) = -e-au(-t)e-" dt = - e"(a+s) dt

Following our previous arguments, we can integrate only if the function dies
out as t goes to minus infinity. e -' will die out as t goes to minus infinity only
if Re{x} is negative. Thus we need Re{a + s} < 0 or Re(s) < -a. For s in this
range,

a + s
(b) (i) hi(t) has a pole at -a and no zeros. Furthermore, since a > 0, the pole
must be in the left half-plane. Since hi(t) is causal, the ROC must be to
the right of the rightmost pole, as given in D, Figure P21.1-4.
(ii) h 2 (t) is left-sided; hence the ROC is to the left of the leftmost pole. Since
a is positive, the pole is in the left half-plane, as shown in A, Figure
P21.1-1.
(iii) h 3 (t) is right-sided and has a pole in the right half-plane, as given in E,
Figure P21.1-5.
(iv) h 4 (t) is left-sided and has a pole in the right half-plane, as shown in C,
Figure P21.1-3.
For a signal to be stable, its ROC must include the fw axis. Thus, C, D, and F
qualify. B is an ROC that includes a pole, which is impossible; hence it corre­
sponds to no signal.

S21.2
(a) By definition,

X(s) = fl x(t)e -t dt
= e -te -S dt

We limit the integral to (0, oo) because of u(t), so

-1* e -(+s)t
X(s) = e -(1+s>' dt =
fo'M 1 + s 0

If the real part of (1 + s) is positive, i.e., Re(s) > - 1, then

lim e -(+s)t = 0

S21-1
Signals and Systems
S21-2

Thus

= 0(-1) 1(-1) 1
X(s) Refs) > -1
1+s 1+s 1+s
The condition on Re{s} is the ROC and basically indicates the region for which
1/(1 + s) is equal to the integral defined originally. Similarly,

H(s) = e 2tu(t)e - dt = J e -( 2 ** dt = 1
Refs) > -2
0 s + 2'
(b) By the convolution property of the Laplace transform, Y(s) = H(s)X(s) in a
manner similar to the property of the Fourier transform. Thus,
1
Refs) > -1,
(s + 1)(s + 2)
where the ROC is the intersection of individual ROCs.
(c) Here we can use partial fractions:
1 A B
(s + 1)(s + 2) s + 1 s + 2'
A = Y(s)(s + 1) = 1,
s= -1

B = Y(s)(s + 2) = -1
S= -2

Thus,
1 1
Y(s)+-,2 Refs) > -1
s+1I s+ 2
Recognizing the individual Laplace transforms, we have
y(t) = e -u(t) - e ~2 t u(t)

S21.3
(a) The property to be derived is

x(t - to) e SoX(s),

with the same ROC as X(s).
Let y(t) = x(t - to). Then

Y(s) = y(t)e -' dt = x(t - to)e " dt

Let p = t - to. Then t = p + to and dp = dt. Substituting

Y(s) = x(p)e -"s>+to) dp

Since we are not integrating over s or to, we can remove the e S'o term,

Y(s) = e - o (p)e -svdp = e -stoX(s)

 Continuous-Time Second-Order Systems/Solutions
S21-3

Note that wherever X(s) converges, the integral defining Y(s) also converges;
thus the ROC of X(s) is the same as the ROC of Y(s).
(b) Now we study one of the most useful properties of the Laplace transform.

xi(t) * x 2(t) _ X 1(s)X 2(s),

with the ROC containing R1 n R 2. Let

y(t) = x 1 (ir)x 2 (t - r) dr

Then

Y(s) = f xI(i)x 2(t - 1)e dr dt

= f x 1 (r) f x2(t- r)e - dt di

Suppose we are in a region of the s plane where X 2(s) converges. Then using the
property shown in part (a), we have

J x 2(t - r)e -"t dt = e -STX(s)

Substituting, we have

Y(s) = x 1 (-)e -sX2(s) dT = X 2 (s) x(ir)e -" dr

We can associate this last integral with Xi(s) if we are also in the ROC of xi(t).
Thus Y(s) = X 2(s)Xi(s) for s inside at least the region R, n R 2 . It could happen
that the ROC is larger,but it must contain R, n R 2 ­

S21.4
(a) From the properties of the Laplace transform,
Y(s) = X(s)H(s)

A second relation occurs due to the differential equation. Since

dkx(t) L kXS
dtk rskX(s)

and using the linearity property of the Laplace transform, we can take the
Laplace transform of both sides of the differential equation, yielding
s2Y(s) - sY(s) - 2Y(s) = X(s).
Therefore,
Y(s) _ 1 1
H(s) S - =
X(s) S2 - s - 2 (s - 2)(s + 1)
Signals and Systems
S21-4

The pole-zero plot is shown in Figure S21.4-1.

Im

s plane

X Re
-1 2

Figure S21.4-1

(b) (i) For a stable system, the ROC must include thejw axis. Thus the ROC must
be as drawn in Figure S21.4-2.

Figure S21.4-2

(ii) For a causal system, the ROC must be to the right of the rightmost pole,
as shown in Figure S21.4-3.

s plane

Figure S21.4-3
 Continuous-Time Second-Order Systems/Solutions
S21-5

(iii) For a system that is not causal or stable, we are left with an ROC that is
to the left of s = -1, as shown in Figure S21.4-4.

Im
s plane

iX Re
--l 2

A
Figure S21.4-4

(c) To take the inverse Laplace transform, we use the partial fraction expansion:
1 1
H(s)
H'-= 1 =
A +
B _
= +
(s +1)(s - 2) s + 1 s - 2 s+ 1 s - 2
We now take the inverse Laplace transform of each term in the partial fraction
expansion. Since the system is causal, we choose right-sided signals in both
cases. Thus,
h(t) = -ie-u(t) + le 2
'u(t)

S21.5
o = 0: Since there is a zero at s = 0, 1H(jO)I = 0. You may think that the phase is
also zero, but if we move slightly on the jw axis, 4 H(jw) becomes

(Angle to s = 0) - (Angle to s = -1) = - - 0 = ­

2 2
= 1: The distance to s = 0 is 1 and the distance to s = -1 is /2. Thus
1 _/

IH(jl)I - - 2
/ZF 2
The phase is

(Angle to s = 0) - (Angle to s = -1)

4 =-
= - -= H(jl)
2 4

W= oo: The distance to s = 0 and s = -1 is infinite; however, the ratio tends to 1

as oincreases. Thus, IH(joo) I = 1. The phase is given by
T r
= 0
2 2
Signals and Systems
S21-6

The magnitude and phase of H(jw) are given in Figure S21.5.

(H(jw)l

4H(jo)

Figure S21.5

S21.6
The pole-zero plot is shown in Figure S21.6.

Im

X -2
s plane

Re
-5 -0.1

X -2

Figure S21.6

Because the zero at s = -5 is so far away from thejw axis, it will have virtually no
effect on IH(jw)I. Since there is a zero at o = 0 and poles near o = 2, we estimate a
valley (actually a null) at w = 0 and a peak at o i 2.
±
 Continuous-Time Second-Order Systems/Solutions
S21-7

Solutions to
Optional Problems
S21.7
(a) Let y(t) be the system response to the excitation x(t). Then the differential
equation relating y(t) to x(t) is
d yt) + 2 w. dy(t) + Woy(t) = w2x(t)
dt2 dt

Integrating twice, we have

T= w
y(t) + 2 0nf 0 y(i-) di- n f w j x(-) di d-T',

y(t) = -2co,
t
y(r) dr - f.
tr

y(r) dr dr' + o4 0 J. x(r) dr dr',

shown in Figure S21.7-1.

Signals and Systems
S21-8

Recall that Figure S21.7-1 can be simplified as given in Figure S21.7-2.

x(t) -- am +-- y(t)

f
Direct Form II

In
u .7

Figure S21.7-2

(b) (i) For a constant w,, and 0 s < 1, H(s) has a conjugate pole pair on a circle
centered at the origin of radius wn. As changes from 0 to 1, the poles
move from close to the jo axis to -CO, as shown in Figures S21.7-3,
S21.7-4, and S21.7-5.
Figure S21.7-3 shows that for s~ 0 the pole is close to thejw axis, so
|H(jw) I has a peak very near w..

s plane
f ~0 /
Re
\ AfWn

|H(ico)

I
-on II-2 2 (On d 1 -2[2

Figure S21.7-3
 Continuous-Time Second-Order Systems/Solutions
S21-9

Figure S21.7-4 shows that the peaks are closer together and more spread
out at { = 0.5.

s plane
0O.5/
Re

IH(jw)

-Cn -V 1 -2 2 Wn v1 -2 2
Figure S21.7-4

Figure S21.7-5 shows that at s~ 1 the poles are so close together and far
from the jw axis that IH(jw) I has a single peak.
Signals and Systems
S21-10

(ii) For constant between 0 and 1, the poles are located on two straight
lines. As w, increases, the peak frequency increases as well as the band­
width, as indicated in Figures S21.7-6 and S21.7-7.

Im
s plane

Wn2 0

Re

IH(jco)|

Figure S21.7-6

Im
s plane

wn > 0

IH(jo)|

-on V 1 -2 2 Wnon 1 -2 2

Figure S21.7-7
 Continuous-Time Second-Order Systems/Solutions
S21-11

S21.8
(a) (i) The parallel implementation of H(s), shown in Figure S21.8-1, can be
drawn directly from the form for H(s) given in the problem statement.
The corresponding differential equations for each section are as follows:
d 2y 1(t) dyi(t) dx(t)
+ + y i(t)
dt2 dt dt
d'y 2 (t) 2dy 2 (t)
+ + 2y(t) =xMt)
dt dt
y(t) = yi(t) + y 2(t)

Y1 (t)

y (t)

x(t)

Figure S21.8-1

(ii) To generate the cascade implementation, shown in Figure S21.8-2, w<

first express H(s) as a product of second-order sections. Thus,
s(s2 + 2s + 2) + (S2 + s + 1) s3 + 3S2 + 3s + 1
H(s) =
(S2 + S + 1)(S2 + 2s + 2) (S2 + s + 1)(s2 + 2s + 2)
Now we need to separate the numerator into two sections. In this case,
the numerator equals (s + 1)', so an obvious choice is
(s + 1)(s2 + 2s + 1)
Thus,
H~s)- S+ 1 s 2 + 2s + 1
H (S) s2 + S + 1 S2 + 2s + 2)
Signals and Systems
S21-12

The corresponding differential equations are as follows:

d2 r(t) dr(t) dx(t)
2
dt + dt + r(t) dt
d2y(t) 2dy(t) d2 r(t) 2dr(t)
dt 2 dt = dt
2
dt
+ r(t)

r(t)
y(t)

Figure S21.8-2

(b) We see that we could have decomposed H(s) as

H(s) = (s2 + 2s + 1 s+ 1
(S2 + S+1 )S 2+ 2s +2)
Thus, the cascade implementation is not unique.

S21.9
(a) Decompose sin wot as
gjwot _ -jwot

2j
Then

xI(t) = sin(wot)u(t) = e.U(t) - e u(t)

2j 2j
Using the transform pair given in the problem statement and the linearity prop­
erty of the Laplace transform, we have

Xi(s) = - ­
2js-jwO S + jAOo

1 2jwo _ CO
2j S2 + wo S2 + w'

with an ROC corresponding to Re(s) > 0.

(b) x 2(t) = e -2' sin(wot)u(t). Since

e 2' sin(wot)u(t) - Xi(s + 2),

 Continuous-Time Second-Order Systems/Solutions
S21-13

the ROC is shifted by 2. Therefore,

e 2 ' sin(wot)u(t) +­
(s + 2)2 + w0
and the ROC is Re{s) > -2. Here we have used our answer to part (a).
(c) Since
£ dX(s)
tx(t))

with the same ROC as X(s), then

te -2'u(t)
£
-1
d [ 1 1
ds (s + 2)'
Thus

te 2 1 u(t) -[(s + 2)2 (s + 2)2

with the ROC given by Re{s} > -2.

(d) Here we use partial fractions:
s+ 1 A B
(s + 2)(s + 3) s + 2 s + 3'
A [s+ 1 -1 B [s+11 -2
s + 3J s=-2 s +2 s=-3 -1
s+1 -1 2
(s + 2)(s + 3) s + 2 s+3 (S21.9-1)
The ROC associated with the first term of eq. (S21.9-1) is Re{s} > -2 and the
ROC associated with the second term is Re{s} > -3 to be consistent with the
given total ROC. Thus,
2t
x(t) = -e -- u(t) + 2e -3'u(t)
(e) From properties of the Laplace transform we know that
L
x(t - T) +-~ e ~TX(s),
with the same ROC as X(s). Since
-c 1
e --3'u(t)
s + 3'
with an ROC given by Re{s} > -3, (1 - e-2)/(s + 3) must correspond to
x(t) = e 3 u(t) - e -( -2)u(t - 2)

S21.10
(a) (1), (2): An impulse has a constant Fourier transform whose magnitude is unaf­
fected by a time shift. Hence, the Fourier transform magnitudes of (1) and (2)
are shown in (c).
(3), (5): A decaying exponential corresponds to a lowpass filter; hence, (3)
could be (a) or (d). By comparing it with (5), we see that (5) corresponds to
kte-"'u(t), which has a double pole at -a. Thus, (5) is a steeper lowpass filter
than (3). Hence, (3) corresponds to (d) and (5) corresponds to (a).
Signals and Systems
S21-14

(4), (7): These signals are of the form e-'" cos(wot)u(t). For larger a, the
poles are farther to the left. Hence H(jw) I for larger a is less peaky. Thus, (4)
corresponds to (f) and (7) corresponds to (g).
(6): If we convolve x(t) = 1 with h(t) given in (6), we find that the output
is zero. Thus (6) corresponds to a null at w = 0, either (b) or (h). Note that (6)
can be thought of as an h(t) given by (1) minus an h(t) given by (3). Thus, the
Fourier transform is the difference between a constant and a lowpass filter.
Therefore, (6) is a highpass filter, or (b).
(b) (a), (d): These are simple lowpass filters that correspond to (i) or (ii). Since (a)
is a steeper lowpass filter, we associate (a) with (ii) and (d) with (i).
(b), (h): These require a null at zero, and thus could correspond to (iii) or
(viii). In the case of (iii), as w increases, one pole-zero pair is canceled so that
for large w,H(s) looks like a lowpass filter. Hence, (b) corresponds to (viii) and
(h) corresponds to (iii).
(c): Here we need a pole-zero plot that is an all-pass system. The only pos­
sible pole-zero plot is (vi).
(e): Here we need a null on thejw axis, but not at w = 0. The only possibility
is (v).
(f), (g): These are resonant second-order systems that could correspond to
(iv) or (vii). Since poles closer to thejw axis lead to peakier Fourier transforms,
(f) must correspond to (iv) and (g) to (vii).
MIT OpenCourseWare
http://ocw.mit.edu

Resource: Signals and Systems

Professor 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.
22 The z-Transform
Solutions to
Recommended Problems
S22.1

(a) The z-transform H(z) can be written as

H(z) = z
z -2
Setting the numerator equal to zero to obtain the zeros, we find a zero at z = 0.
Setting the denominator equal to zero to get the poles, we find a pole at z = 1.
The pole-zero pattern is shown in Figure S22.1.

z plane

ROC: IzI > R

2

Figure S22.1

(b) Since H(z) is the eigenvalue of the input z' and the system is linear, the output
is given by

y[n] = 1 3) +3 1 (2)"
= 3()" + 4(2)"

S22.2
(a) To see if x[n] is absolutely summable, we form the sum
N-I N-i

SN = x[nhl 2' = 1 2
n=o n=O

Since limN-oSN diverges, x[n] is not absolutely summable.

(b) Since x[n] is not absolutely summable, the Fourier transform of x[n] does not
converge.
N

N(N- "
(C) SN
n=o(2

S22-1
Signals and Systems
S22-2

limN-oN is finite for Ir > 2. Therefore, the Fourier transform of r-"x[n] con­
verges for Ir| > 2.

(d) X(z) = 2"z~" = E (2z- 1)

n=O n.O

__ 1
= 2z- for 12z~'I < 1
Therefore, the ROC is IzI > 2.
1
(e) X 1(e") =
1 e -s
Therefore, x 1 [n] = (2)"u[n].

S22.3
(a) Since x[n] is right-sided, the ROC is given by Iz I > a. Since the ROC cannot
include poles, for this case the ROC is given by Iz I > 2.
(b) The statement implies that the ROC includes the unit circle Iz = 1. Since the
ROC is a connected region and bounded by poles, the ROC must be
5 < IzI < 2
(c) For this situation there are three possibilities:
(i) Izi < 3
(ii) -1 < I z| < 2

(iii) IzI > 2

(d) This statement implies that the ROC is given by |z < 1.

S22.4

(a) (i) X, (z) = x I~ -n T (-2 )f

n=-oon=O

1
1 - '

with an ROC of < 1, or IzI > 2.

-1 2z
(ii) X2 (z) = E (i)z~

Letting n = -m, we have

X 2(z) = - l()--m'
m= 1
GO 2z
= - L (2z) m
= - 1-2z
m1

1 - z '
with an ROC of 12z| < 1, or Iz < i.
 The z-Transform / Solutions
S22-3

(b) (i)
Im (z

ROC

Re lz

Figure S22.4-1

(ii)
ImZl

ROC
Re (z)

lzl= 1

Figure S22.4-2

2z
(c) (i) X 3(z) 2 z- = 2 (1 > 1, as shown
=
n=O
-1z) -Z-1 .The ROC is Izi
in Figure S22.4-3.

-1i

(ii) X 4(z) = - 2"z~ = - Z 2-"z

n=

z/2 z
n=Al
1 -(z/2) z -2'
with an ROC of Iz/21 < 1, or IzI < 2, shown in Figure S22.4-4.
Signals and Systems
S22-4

Im

/j

A. Re Iz)

Izl=2

Figure S22.4-4
(d) For the Fourier transform to converge, the ROC of the z-transform must include
the unit circle. Therefore, for x 1 [n] and x 4[n], the corresponding Fourier trans­
forms converge.

S22.5
Consider the pole-zero plot of H(z) given in Figure S22.5-1, where H(a/2) = 1.

z plane

K ' 1 to 5zeros
Figure S22.5-1

(a) When H(z) = z/(z - a), i.e., the number of zeros is 1, we have

H(ej) = cos 9 + j sin 0_

(cos 0 - a) + j sin 12
Therefore,
2
|H(e )| =
1 + a - 2a cos 0'

and we can plot IH(ejQ)I as in Figure S22.5-2.

 The z-Transform / Solutions
S22-5

IH(en2)I

Figure S22.5-2

When H(z) =z/(z - a), i.e., the number of zeros is 2, we have

- cos 20 + j sin 22
(cos 0 - a) + j sin 2
Therefore,

|H(es")| = 1 + a2 - 2a cos 0
Hence, we see that the magnitude of H(eia) does not change as the number of
zeros increases.
(b) For one zero at z = 0, we have

H(z)= z-a'

H(e") = e
e- a

We can calculate the phase of H(e ") by [Q - 4 (denominator)]. For two zeros
at 0, the phase of H(ein) is [29 - 4 (denominator)]. Hence, the phase changes
by a linear factor with the number of zeros.
(c) The region of the z plane where IH(z) I = 1 is indicated in Figure S22.5-3.

A z plane

+- Re z =

Figure S22.5-3
Signals and Systems
S22-6

S22.6

(a) (3)"[n]
n=O

= (3z)- 1 z
n=O i e z Z
Therefore, there is a zero at z = 0 and a pole at z = 1, and the ROC is

-<1 or IzI> ,
3,
as shown in Figure S22.6-1.

z plane

WZ
/
A, i - - - - ­

7
37

I Figure S22.6-1

(b) b[n +1] L bS[n + 1]z-" = z,

with the ROC comprising the entire z-plane, as shown in Figure S22.6-2.

Im z plane

Re

Figure S22.6-2

S22.7

(a) Using long division, we have

3
X(z) = 1 -!Z- + !z - z­ +

We recognize that
x[n] = (-i)"u[n]
 The z-Transform / Solutions
S22-7

(b) Proceeding as we did in part (a), we have

x[n] = (-I"u[n]

(c) X(z) = -a
-az

S_ a2)

(1
a 1-az

1 a2)z­
1 a2
a 1-a-IZ-1
Therefore,

a-[n] -
x[n) = -
aa I u[n ­

Solutions to
Optional Problems
S22.8
(a) 1{r-"x[n]} = 5{r-u[n]}

= r~"e -jan
n=O

= Z (re+yu-n
n=O

For the sum to converge, we must have

< 1
Thus, Ir| > 1.

(b) (i)
Im

r =1z plane

Re

Figure S22.8-1
Signals and Systems
S22-8

(ii)
Im

r= z plane

Re

Figure S22.8-2

(iii)

r= 3 z plane

Figure S22.8-3

(c)

Figure S22.8-4

S22.9
(a) The inverse transform of
1
1 -i -
 The z-Transform / Solutions
S22-9

is (i)"u[n]. But from the given relation, we have

2-"x[n] = (L)"u[n],

x[n] = (I)"u[n]
3 + 2z- 1 3 + 2z-'
(b) 2 + 3z-' + z-2
(2 + z-1)(1 + z-1)
1 1
2+z- 1 + z-'
2 1
1+ iz-' 1 + z- 1
= -1)u[n] + (-1)"u[n]

S22.10
Az - a)
(a) H(z) = ( - with A a constant
(1 - azTe)
Therefore,
H A(e -j' - a)
(e' ) 1 - ae j
I A 2(e ' a)(e" - a)
=
|H~es")2 a) = A2
(1 - ae-3)(1 - aej")

and thus,
|H(e'")| = Al
(b) (i) |v 112 = 1 + a2 - 2a cos 9
1 2
(ii) |v 2 12 = 1 + -2 - - cos 9
a a
1
= - (a 2 + 1 - 2a cos Q)
1
= -| vil2

S22.11

In all the parts of this problem, draw the vectors from the poles or zeros to the unit
circle. Then estimate the frequency response from the magnitudes of these vectors,
as was done in the lecture. The following rough association can be made:
(a) (i)
(b) (ii)
(c) (iv)
(d) (iii)
(e) (iv)
Signals and Systems
S22-10

S22.12
(a) x[n] = (I)"[u[n] - u[n - 101]. Therefore,
9
X(z) = E(i)"Z--"
n=o
1 0
-1(zY
= (2z)" = 1 (2z)-
n=O 1 (2z)-1
10 11

= |z| > 0,
zl
z9(z- (1)10I">0
- 2)

shown in Figure S22.12-1. The Fourier transform exists.

(b) x[n] = (.)In' = (I)"u[n] + (.) -u[-n - 1]. But
(l\n z z 1
u[n] , IzI> ,

and
(1)-n
u[-n - ]- z < 2
2 ~z - 2'
Summing the two z-transforms, we have
-2z
X(z) = 2 1 < |zI < 2
(z 2)D(z - 2) 2
(See Figure S22.12-2.) The Fourier transform exists.

(c) x[n] = 7()" cos + u[n]

Therefore,

X(z) = 76 cos + 4z

7 00 n j(2r/6)n+(,r/4)] j(2w/6)n+(-/4)]
Z
- l + e n
2n=0 3

[ej/4 ei(2/6) -1 -jr/

4
1 -j(2r/6)-1)]

7 e _ e
6__ jw/4

le(2r/6)Z- (2v/6) 1
7z e e -j4

,r
2z cosir
(2 / 6)

2 cos (2,r
z
Z ire-
,r)
(e j(21/6)
7

7z 4 3 6 4 1
z- I (2r/6>)(z - j(2/6) where |z| >

The pole-zero plot and ROC are shown in Figure S22.12-3. Clearly, the Fourier
transform exists.
 The z-Transform / Solutions
S22-11

9th order pole

pole -zero
cancel

Figure S22.12-1

z plane

Figure S22.12-2

Figure S22.12-3
Signals and Systems
S22-12

1 - z' 0 z10 -1
(d) X(z) = z- - - 1 = z ­
.=, 1 - Z' z"(z - 1

The ROC is all z except z = 0, shown in Figure S22.12-4. The Fourier transform
exists.

9th-order pole
.1.0 xPC 1 -1 1­

pole -zero
cancel

Figure S22.12-4

S22.13
(a) From

log(1 - w)= IwI < 1,

i=1
we find
(2z)'
log(1 - 2z) = 12z| < 1

z", |z| < ,

2-n
x[n] = s
0, n >- 0
(b) We solve this similarly to the way we solved part (a).
(Oz-')'f
log 1 12 - 1-1 Z11<1
1
z-< 1
n 2 2

01
n > 0,
x[n]
=t
0, n s 0
MIT OpenCourseWare
http://ocw.mit.edu

Resource: Signals and Systems

Professor 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.
23 Mapping Continuous-Time Filters
to Discrete-Time Filters
Solutions to
Recommended Problems
S23.1

(a) X,(z) = - 1 -2z 2Z <1,

so the ROC is IzI > i.

0
(b) X 2(z) = (-3)"z­

= (-3)-"z" 1 z | 3-'z| < 1,

n=O +3

so the ROC is IzI < 3.

We can also show this by using the property that

x[-n] < X(z-1)

Letting x[-n] = x 2[n], we have

x[-n] = (-3)"u[-n],
x[+n] = ()[n
1
X(z) = + iz'

Therefore,
1
X 2(z) = 1+ 1z3

and the ROC is IzI < 3.

(c) Using linearity we see that

X 3(z) = X 1(z) + X 2(z) 1 - iz- 1 + z

The ROC is the common ROC for XI(z) and X 2(z), which is - < z < 3, as shown
in Figure S23.1.

S23-1
Signals and Systems
S23-2

z plane

-3

Figure S23.1

(d) Using the time-shifting property

x[n - n 0] z "OX(z),
we have

X 4(z) = z- 5X 1(z) = -Z

Delaying the sequence does not affect the ROC of the corresponding z-trans­
form, so the ROC is Izi > i.
(e) Using the time-shifting property, we have

X 5(z) = z

and the ROC is Iz > .

(f) X 6(z) = =
n=0 3 1
and the ROC is I z-'| < 1, or Izi >i.
(g) Using the convolution property, we have

X7(z) =X1(z)X
6(z)= (1-1i) 1-z
and the ROC is Iz I > i, corresponding to ROC1 n ROC.

S23.2

(a) We have
y[n] - 3y[n - 1] + 2y[n - 21 = x[n]
Taking the z-transform of both sides, we obtain
Y(z)[1 - 3z-' + 2z -2 =X(z),

Y(z) _ 1 -2 _Z2 _z_2 z_2__

X(z) 1 - 3z- + 2z - 3z + 2 (z - 2)(z - 1)

 Mapping Continuous-Time Filters to Discrete-Time Filters / Solutions
S23-3

and the ROC is outside the outermost pole for the causal (and therefore right-
sided) system, as shown in Figure S23.2.

z plane

Figure S23.2

(b) Using partial fractions, we have

1 2 -1
H(z) = 1 = +
(1 - 2z- )(1 - z-1) 1 - 2z -1 1- z
By inspection we recognize that the corresponding causal h[n] is the sum of two
terms:
h[n] = (2)2"u[n] + (-1)1"u[n]
= 2"*'u[n] - u[n]
= (2"*' - 1)u[n].

The system is not stable because the ROC does not include the unit circle. We
can also conclude this from the fact that

E
n=O
12n+ 1 = o

(c) Since x[n] = 3"u[n],

1
X(z) = 3 1'
-
1 - 3z IzI > 3,

1 1
Y(z) = H(z)X(z) = _
(1 - 3z 1) (1 - 2z-1)(1 - z 1)

Using partial fractions, we have

-4 i
Y(z) = + + IzI > 3
1 - 3z-' 1 - 2z-' 1 - Z-1'
since the output is also causal. Therefore,
y[n] = (2)3"u[n] - (4)2"u[n] + in[n]
(d) There are two other possible impulse responses for the same
1
1
H(z) =
1 - 3z-' + 2z2
Signals and Systems
S23-4

corresponding to different ROCs. For the ROC Iz < 1 the system impulse
response is left-sided. Therefore, since
2 -1
H(z) = 1- 2z 1 z-'

then
h[n] = - (2)2"u[-n - 11 + (1)u[-n - 1]
= -2" 1u[-n - 1] + u[-n - 1]
For the ROC 1 < Iz I < 2, which yields a two-sided impulse response, we have
h[n] = -2"*lu[-n - 1] - u[n]
since the second term corresponding to -1/(1 - z-) has the ROC 1 < Iz|.
Neither system is stable since the ROCs do not include the unit circle.

S23.3
(a) Consider

X 1(z) = ( x[n - noiz-"

n= 0

Letting m = n - no, we have

X 1(z) = E x[m]z-m+ no)

m= -o
00

= z-"i' ( x [m]z-m
M = -_0

= z-oX(z)
It is clear that the ROC of Xi(z) is identical to that of X(z) since both require
that IU _0 x[n]z-" converge in the ROC.
(b) Property 10.5.3 corresponds to multiplication of x[n] by a real or complex expo­
nential. There are three cases listed in the text, which we consider separately
here.

(i) Xi(z) = E e x[n]Z-­

- : x[n] (ze -. 7 )- n

= X(ze --io),

with the same ROC as for X(z).

(ii) Now suppose that

X 2(z)= zox[n]z

x[n] ­

=X
 Mapping Continuous-Time Filters to Discrete-Time Filters / Solutions
S23-5

Letting z' = z/zo, we see that the ROC for X 2(z) are those values of z such
that z' is in the ROC of X(z'). If the ROC of X(z) is RO < Iz| < R 1, then
the ROC of X 2(z) isRozol < |zi < RjIzo|.
(iii) This proof is the same as that for part (ii), with a = zo.
(c) We want to show that
z dX(z)
nx[n] dz
Consider

X(z) = ( x[njz-"
n = -oo

Then
dX(z)
dz)= 700n­
-nx[niz-*­
dz n=_-o

= -z- n 1= -o ( nx[n]z-",

so
dX(z) = n
-Z dz nx[niz-',
dz n=_-o

which is what we wanted to show. The ROC is the same as for X(z) except for
possible trouble due to the presence of the z' term.

S23.4

(a)

IH(ein)|

r = 0.9
-r= 0.75
r = 0.5

0 iT 7 21r
4 2
Figure S23.4-1
Signals and Systems
S23-6

(b)

S23.5
(a)
Im z plane

()F X )( Re
3

Figure S23.5-1
 Mapping Continuous-Time Filters to Discrete-Time Filters / Solutions
S23-7

(b)

The ROC is IzI > 2. The system is not stable because the ROC does not include
the unit circle.
(c)

z plane

Figure S23.5-3

The ROC is 2 > Iz I > i, which for this case includes the unit circle. The corre­
sponding impulse response is two-sided because the ROC is annular. Therefore,
the system is not causal.
(d)
Im z plane

Re

Figure S23.5-4
Signals and Systems
S23-8

The remaining ROC does not include the unit circle and is not outside the out­
ermost pole. Therefore, the system is not stable and not causal.

S23.6
daytt) dytt)
(a) + 5 + 6y(t) = x(t) + 2 dxt)
dx(t)
dt2 dt dt
is the system differential equation. Taking Laplace transforms of both sides, we
have
Y(s)(s 2 + 5s + 6) = X(s)(1 + 2s),
so

Y(s) 1 + 2s
2
X(s) s + 5s + 6
1+2s 5 -3
(s + 3)(s + 2) s + 3 s + 2
Assuming the system is causal, we obtain by inspection
h,(t) = 5e - 3 u(t) - 3e -2'u(t)
(b) Using the fact that the continuous-time system function Ak/(s - s) maps to the
discrete-time system function Ak/(l - e*kz- 1) (see page 662 of the text), we
have
5 3
Hd(z) 1 e "z- - e -2TZI

(c) Suppose T = 0.01. Then

5 3
Hd(Z)
1 - e -0 0 3
z-'1- e ~o 02z- 1
0 03
Letting a = e - . , b = e- 002, we have
5 _ -
33
Hd(z) =
1 - az' 1- bz-

So by inspection, assuming causality,

hd[n] = 5a'u[n] - 3b u[n]
(d) From part (a), we have
hc(t) = 5e - 3 tu(t) - 3e 2
u(t)
Replacing t by nT = 0.01n, we have
hc(nT) = 5e 0~ 0 3 u(0.01n) - 3e 0 02
"u(0.02n)
Letting a = e -0.03 and b = e -0.02 yields
he(nT) = 5a'u[n] - 3b u[n],
which agrees with the result in part (c).
 Mapping Continuous-Time Filters to Discrete-Time Filters / Solutions
S23-9

Solutions to
Optional Problems
S23.7
(a) The differential equation is
dy(t) + 0.5y(t) x(t)
dt
Taking the Laplace transform yields
Y(s)[s + 0.5] = X(s),

H(s) = Y(S) = 1
X(s) s+ 0.5 '
H(w) = 0
jF + 0.5'
which is sketched in Figure S23.7-1.

20 logH(w)
20lgI H(0)

slope
0- -20 dB/decade

-20-. i W
.5 5 50 ...

Figure S23.7-1

(b) y[n + 1] - y[n] + 0.5y[n] = x[n]

T
Taking the z-transform of both sides yields
1
1 (z - 1)Y(z) + 0.5Y(z) = X(z),
T

Y(z) ( 0.5 + zT X(z)

Letting T = 2 yields
Y(z) 2
= Hd(z) = -
Z(z) z
Now since

|Hd(ej0 )I = |Hd(z)I I
Signals and Systems
S23-10

we have
|Hd(e j)| = 2, for all Q,
which is an all-pass filter and is sketched in Figure S23.7-2.

(c) HA(z) =
1 1
0.5- + - z
T T

(0.5T - 1) + z
The pole is located at zo = -(0.5T - 1) and, since we assume causality, we
require that the ROC be outside this pole. When the pole moves onto or outside
the unit circle, stability does not exist. The filter is unstable for
Izo l 1 or I-(0.5T - 1)1 1,
|0.5T - 11 1,
T 4
Therefore, for T > 4, the system is not stable.

S23.8

(a) X(z) = x[n]z-

X(z-1) = Z x[nZ"

Letting m = -n, we have

X(z 1) =
M= -0
x[-m]z- = ( x[mjz-"' = X(z)

f (z - ak)
(b) X(z) = A k
J (z - bk)

from the definition of a rational z-transform. Now

J1 (z-1 - ak)
X(z 1) = A k
1 (z- - bk)
k
Each pole (or zero) at zo in X(z) goes to a pole (or zero) z-1 inX(z-1). This implies
that zo = 1 or that X(z) must have another pole (or zero) at z--1.
 Mapping Continuous-Time Filters to Discrete-Time Filters / Solutions
S23-11

(c) (i) x[n] = b[n + 1] + b[n - 1],

_ _ z2 + _ (z + j)(z - j)
_ +
z z
The zeros are at z = j, 1/j, and the poles are at z = 0, z = oo.
(ii) x[n] = b[n + 1] - U[n] + 6[n - 1],
X(z) = z - 2 + z­
2 z + 1 (z - 1)(z - 2)

The zeros are at z = i, 2, and the poles are at 0, oo.

(d) (i) Y(z) = E y[n]z-",

Y*(z) = y[n]z- = L y[n](z*)" = Y(z*),

Y*(z*) = [ y[n]z-" = Y(z)

n= -00
(ii) Since Y(z) is rational,
fl(z - ak)
Y(z) = A k

f(z - bk)

Now if a term such as (z - ak) appears in Y(z), a term such as (z* - ak)
must also appear in Y(z). For example,
Y(z) = (z - ak)(z - ak),
*(z*) = [(z* - ak)(z - ak)*]
= (z - a*)(z - ak) = Yz)

So if a pole (or zero) appears at z = ak, a pole (or zero) must also appear
at z = a* because
(z - ak) = 0 = z = ak

(e) Both conditions discussed in parts (b) and (d) hold, i.e., a real, even sequence is
considered. A pole at z = z, implies a pole at 1/z, from part (b). The poles at
z = z, and z = 1/z, imply poles at z = z* and z = (1/zp)* from part (d). There­
fore, if z, = pej", poles exist at

pei"
1 =1
p
(pej")* = pe -',
( *1pei p

S23.9

(a) X 3 (z) = X3[n]z~

00 00

-
k=-
Y x1[k] [
( x2[n
n=--oo
- k]z-"
oO

= x1[kli2(z),
k= -oo
Signals and Systems
S23-12

where

X 2(z) = E X2[n - kz-­

= Z{x2[n - k]}
(b) Z{x 2 [n - k) = z-kX 2(z) from the time-shifting property of the z-transform, so

XA(z) = ( x 1 [k]z-kX2(z)
k=-oo

(c) XA(z) = X 2(z) T x 1[k]z k

k=-o
= X 1 (z)X 2(z)

S23.10
Consider x[n] to be composed of a causal and an anticausal part:
x[n] = x[n]u[-n - 1] + x[n]u[n]
Let
x 1 [n] = x[n]u[-n - 1],
X2[n] = x[n]u[n],
so that
x[n] = x 1 [n] + x 2[n]
and
X(z) = X 1(z) + X 2(z)
It is clear that every pole of X 2(z) is also a pole of X(z). The only way for this not
to be true is by pole cancellation from X1 (z). But pole cancellation cannot happen
because a pole ak that appears in X 2(z) yields a contribution (ak)"u[n], which cannot
be canceled by terms of x 1[n] that are of the form (bk )U[-n - 11.
From the linearity property of z-transforms, if
y[n] = y 1 [n] + y 2[n]
then
Y(z) = Y1(z) + Y 2(z),
with the ROC of Y(z) being at least the intersection of the ROC of Yi(z) and the ROC
of Y 2(z). The "at least" specification is required because of possible pole cancella­
tion. In our case, pole cancellation cannot occur, so the ROC of X(z) is exactly the
intersection of the ROC of X 1(z) and the ROC of X 2(z).
Now suppose X 2(z) has a pole outside the unit circle. Since x2[n] is causal, the
ROC of X 2(z) must be outside the unit circle, which implies that the ROC of X(z)
must be outside the unit circle. This is a contradiction, however, because x[n] is
assumed to be absolutely summable, which implies that X(z) has an ROC that
includes the unit circle.
Therefore, all poles of the z-transform of x[n]u[n] must be within the unit
circle.
 Mapping Continuous-Time Filters to Discrete-Time Filters / Solutions
S23-13

S23.11
(a) If h[n] = hc(nT), then

Sd[n] = E hc(kT)
k= -o

The proof follows.

Sd[n] = ( u[n - k]hd[k]

k= -o

= E hd[k],
k= -o

but hd[k] = he(kT), so

n

Sd[n] = E h,(kT)
k= -00

(b) If Sd[n] = se(nT), then hd[n] does not necessarily equal he(nT). For example,
hc(t) = e ~4'U(t),
sc(t) = f e -au(r)u(t -r) d­

= t - "'dr = - (1 - e -a'), t : 0
o 1a

Sd[n] = sc(nT) = - (1 - e -a"T), n >- 0

a
However,
n

Sd[n] = Z
k= -oo
hd[k],

so
Sd[n] - sd[n - 1] = hd[n]

and, in our case, for n 0,

1 1
hd[n] = - (1 - e -a"T) - e(1
e ~-"a"-)T)
a a
= 1 e -an(ea" - 1)
a
But, for n > 0,
he(nT) = e -a"Tu(nT)
1
#- e -anT(eae 1)
a

S23.12
(a) From the differential equation

( aks) Y(s) = ( bksk)X(S),

k-0 k=0
Signals and Systems
S23-14

we have
M
Y) bksk
Ys) = He(s) = k=O
X(s) N k
( aks"
k=0

Now consider

yldn] V"{x[n]} -x[n + 1] - x[n - 1]

2
Yi(z) = Z{y 1 [n]} = Z{Vo> {x[n]}) z -z X(z),

Y2[nJ = V(2>{X[n]= - y 1 [n + 1 ~ y1[n - 1]

2
Y 2 (z) = Z Z Yi(z) = (z 1)2X(z)
By induction,

Z{Vlk){X[n} =(z z1 X(Z)

Therefore,
N _(- k M b(k
a Y(z) =Yb ( X) )
k=0 k=

z - z­
b'
Hd(z) = Yz
X(z) N _(zk
(ak
k=0

=He,(s)

(b) He(s) = Hd(Z)

s =(z -z-/
from part (a). Consider s = jw, z = ej'. So

.' e j" - e -j"

= 2
and, thus, w = sin 0 is the mapping between discrete-time and continuous-time
frequencies. Since H(w) = w for IwI < 1, Hd(ej") is as indicated in Figure
S23.12.
 Mapping Continuous-Time Filters to Discrete-Time Filters / Solutions
S23-15

Hd(ejR)

7T T7T iT
2 -2

-1­

Figure S23.12

(c) From part (a) we see that

Hd(z) = Hd(z )

and that Hd(z) is a rational z-transform.

P
Afl (z - zo.)Mi

Hd(z) = Q'
] N,(z - zpi)Ni

Therefore, if a term such as (z - zo,) appears, (z-' - zo,) must also appear. If
Hd(z) has a pole within the unit circle, it must also have a pole outside the unit
circle. If the ROC includes the unit circle, it is therefore not outside the outer­
most pole (which lies outside the unit circle) and, therefore, Hd(z) does not cor­
respond to a causal filter.
Consider
1
He(s) =

corresponding to a stable, causal h,(t).

1
Hd(z) = He(s) Z - Z
=s=(z-z1)2 1
2 2
2z 2z
2
z +Z 1 [)(-1 + V5) (-1 - V5)
2 1[2

so poles of z are at 0.618, -1.618. Therefore, Hd(z) is not causal if it is assumed

stable because stability and causality require that all poles be inside the unit
circle.
Signals and Systems
S23-16

S23.13
(a) We are given that
A
He(s) = (S 2

From Table 9.2 of the text (page 604), we see that hc(t) = Ate'Ou(t).
To verify, consider

S O= f .e'O'utt)e -"dt,
d
1 d * sotu(t)e Stdt]
ds (s s) ds u

=SS) teso'ult)e~"dt

Therefore,
.Cr
tesotu(t) ( 2
(s -so

(b) hd[n] = hc(nT) = AnTeson Tu[n]

(c) Hd(z) = L hd[nlz~" = AT : nesonTz-n

From Table 10.2 of the text (page 655),
z az­
na~u[n]~-- -
(1 - az-1)2
This can be verified:
1 _

1 _ = a u[n]z-"
d 1) d *0
-z = C-z E a u[n]z-"
dz -- 1 z dz =
-az-2

= (-n)anu[n]z-"­
(1 - az )

= natu[n]z"
(1 -z) ,= _

In our case, a = e'oT, so

ATe soTz-
Hd(Z) = ( ez)2

(d) Hc(s) = = + +
(s+ 1)(s + 2)2 s + 1 s + 2 (s + 2)2
Using the first-order pole result for 1/(s + 1) and - 1/(s + 2) and the second-
order pole result for - 1/(s + 2)2, we have
1 1 Te 2Tz­
Hd(z) = 1 - e -Tz 1 - e 2 Tz- ­ e-2rz 1)2
1 e-T- (1
After some algebra, we obtain
z[z( -e -2T + e- - Te-2) + e -4T - e -3 T+ Te-3T]
Hd(z) = (z - e-T)(z - e -2T)2
 Mapping Continuous-Time Filters to Discrete-Time Filters / Solutions
S23-17

The corresponding pole-zero pattern is shown in Figure S23.13.

Im z plane

double pole

cl 1 C41 vRe
e-2T e-T

e-4T _ e-3 T (1 - T)
zero at z = ­
2
e-T e T (1+T)

Figure S23.13
MIT OpenCourseWare
http://ocw.mit.edu

Resource: Signals and Systems

Professor 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.
24 Butterworth Filters
Solutions to
Recommended Problems
S24.1
(a) For N = 5 and w = ( 2 7)1 kHz, IB(jw)1 2 is given by
1
IB(jw)1 2 = 10
1 + (2000 r)

(b) The denominator of B(s)B(-s) is set to zero. Thus

0 = 1 + (2000)1O, or s = (-1) 1 j20001r

Expressing -1 as e' andj as e j/ 2, we find that the poles of B(s)B(-s) are

2
)+<. 5isa,
s = 2000ei[(x/10 +<x/
as shown in Figure S24.1-1.

Im

poles of B(s)B(-s)

20007r
Re

x /
Figure S24.1-1

(c) For B(s) to be stable and causal, its poles must be in the left half-plane, as
shown in Figure S24.1-2.

Im

poles of B(s)

Re
-20007T

Figure S24.1-2

S24-1
Signals and Systems
S24-2

(d) Since the total number of poles must be as shown in part (b), the poles of
B(-s) must be given as in Figure S24.1-3.

Im

X
x poles of B(-s)

2000n
-Re

Figure S24.1-3

S24.2
(a) When there is no aliasing, the relation in the frequency domain between the
continuous-time filter and the discrete-time filter corresponding to impulse
invariance is

H(eJ.) = 1HH.
T \T
- Y |1 1
| 7r

Thus, there is an amplitude scaling of T and a frequency scaling given by

Q = oT, |1 1 !ir, Iw|I rT
The required transfer function can be found by reflecting TH(e'u) through the
preceding transformation, as shown in Figure S24.2-1.

2 = wT

I I
TH(e n) | |
Hai)(j)

T+0

Figure S24.2-1
 Butterworth Filters / Solutions
S24-3

Since the relation between 0 and w is linear, the shape of the frequency response
is preserved.
(b) For the bilinear transformation, there is no amplitude scaling of the frequency
response; however, there is the following frequency transformation:

9=2arctan (-2)
wT

As in part (a), we can find H(jw) by reflecting H(ej") through the preceding
frequency transformation, shown in Figure S24.2-2.

Because of the nonlinear relation between Q and w, H,(jo) does not exhibit a
linear slope as H(e'") does.
(c) We redraw the transformation of part (a) for the new H(e'0 ) in Figure S24.2-3.
As in part (a), the shape of the frequency response is preserved.
Signals and Systems
S24-4

2i
3
E2 wT

TH(ei2) I
Ha(jw)
T -­

I2 7

iT 2ir
3T 3T

Figure S24.2-3

We redraw the transformation of part (b) for the new H(ej") in Figure S24.2-4.
Unlike part (b), the general shape of H(eju) is preserved because of the piece­
wise-constant nature of H(ej").

2nT
3 =2 arctan w
2

7T
3

I |
H(e'*)
Ha(j) |

1 I-
I - I

2tan - 2 tan T
T 3 T 3

Figure S24.2-4
 Butterworth Filters / Solutions
S24-5

S24.3
(a) Using the bilinear transformation, we get
1 -

(b) Since H(s) has a pole at -a, we need a > 0 for H(s) to be stable and causal.
(c) Figure S24.3 contains a plot of (1 - a)/(1 + a), the pole location of H(z), ver­
sus a.

1- a
1+ a

-1-­

Figure S24.3

We see that for a > 0, (1 - a)/(1 + a) is between -- 1 and 1. Since the only pole
of H(z) occurs at z = (1 - a)/(1 + a), H(z) must be stable whenever H(s) is
stable, assuming that H(z) represents a causal h[n].

S24.4
(a) For T = 1 and the impulse invariance method, B(jw) must satisfy

1 IB(jw)|1 0.8 for 0 w ,

4,
0.2 IB(jw)|1 0 for 37 fCV

Therefore, if we ignore aliasing,

B (j =_ 1)2N
B (ij) 1 + j/4
4
_ 2N= (0. 2)
1j3r/4
1+ .'W
Signals and Systems
S24-6

(b) For T = 1 and the bilinear transformation, B(jw) must satisfy

1 1B(jo) I 0.8, 0 : o : 2 tan,

8)
0.2 1B(ji) 0, 2 tan - O
8
Therefore,

1 2= (0.8)2,
j2 tan (w/8)
1+
SJwc
1 2= (0.2)2
1+ j2 tan (3-/8)

S24.5
(a) The relation between 0 and w is given by Q = wT, where T = 1/15000. Thus,

1 - IH(e 0 )|
1 0.9 for 0 : - 2wr
5
0.1 - IH(e'")|1 0 for 3 -x s7
5
Note that while Hd(jw) was restricted to be between 0.1 and 0 for all o larger
than 2r( 4 500), we can specify H(eja) only up to Q = x. For values higher than
w, we rely on some anti-aliasing filter to do the attenuation for us.
(b) Assuming no aliasing,

H(e) = G j

Therefore,
27
3 1|G(jw)|1 2.7, 0 s o
15,
7w T
0.3 1
IG(jw) 1 0, s o < ­
5 3
(c) The relation between o and Qis given by Q = 2 arctan (w). Thus,
T
1 IG(jo)I - 0.9, 0 s s tan ,
5)
37r
0.1 - IG(jw)I 0, tan -5 o < oo
10
(d) If T changes, then the specifications for G(jw) will change for either the impulse
variance method or the bilinear transformation. However, they will change in
such a way that the resulting discrete-time filter H(e'") will not change. Thus,
He(jo) will also not change.
 Butterworth Filters / Solutions
S24-7

Solutions to
Optional Problems
S24.6
(a) We first assume that a B(s) exists such that the filter specifications are met
exactly. Since
2
1
IB(jw)1
1+

we require that
1
IB(j27r)1 2 = = (10-0.05)2 = 10-0,

1
IB(j3ir)1 2
= = 10-15

Substituting N = 5.88 and co, = 7.047, we see that the preceding equations are
satisfied.
(b) Since we know that N = 6, we use the first equation to solve for we:

10- = 1 1

1 +

Solving for co,, we find that w, = 7.032. The frequency response at ( = 0.31 is
given by
_1

IB(j3)1 2
- 12 = 0.02890,
1+ 7.032
20logo0 B(j3w)| = -15.4dB
(c) If we picked N = 5, there would be no value of we that would lead to a Butter-
worth filter that would meet the filter specifications.

S24.7
We require an Hd(z) such that
0 20 logio|Hd(eO)j -0.75, 0 s Q s 0.2613w,
-20 dB 20 logio
1 Hd(e)|), 0.4018w : Q ! w
We will for the moment assume that the specifications can be met exactly. Let 9, be
the frequency where
20 logI|Hd(eja,)| = -0.75, or |Hd(es0 ")1 2 = 10-0075

Similarly, we define Q, as the frequency where

20 log1oHd(eS')|
0 = -20, or |Hd(e "f)|2 = 10-2
Signals and Systems
S24-8

Using T = 1, we find the specifications for the continuous-time filter Ha(jo) as

IHa(jo,)| 2 = 10 -.075 IH(jW,)| 2 = 10 2,

where

o, = 2 tan = 2 tan =2 0.8703,

22
o, = 2 tan -= 2 tan (0418w= 1.4617
2 2
For the specification to be met exactly, we need N and co such that
2N
j.732N
/jO.8703 (jl.4617 2=10
1+ . = 1000" and 1 + . =102
Solving for N, we find that N = 6.04. Since N is so close to 6 we may relax the
specifications slightly and choose N = 6. Alternatively, we pick N = 7. Meeting the
passband specification exactly, we choose w, such that

1 + (O.= 03 )14 100.075, or we = 0.9805

jWc

The continuous-time filter H,(s) is then specified by

1
H(s)H(-s) =

1+ ( s
The poles are drawn in Figure S24.7.

We associate with Ha(s) the poles that are on the left half-plane, as follows:
s, = -0.9805, S2 = 0.9805e"/ 14, S3 = s*,
0
S4 = 0.9805e' "'/ S5 = S*, S6 = 0.9805e 2' 14
/" S 7 = s*
Ha(s) is given by

Ha(S) = (0.9805)7
7

f
i= 1
(s - s)
 Butterworth Filters / Solutions
S24-9

Hd(z) can be obtained by the substitution

Hd(Z) = H.(S)I|,= 2 1( _z-1)_(1+z )

S24.8
(a) Assuming no aliasing, Hd(e'u) is related to Hb(jw) by

Hdtej") = $o j , T = 2

Thus, the specifications for Hb(jw) are given by

2 IHb(jw)I 2a, 0 - w - 0.2r/2,
2b I
|Hb(jw)| 0, 0.37r/2 w
(b) Substituting
2 .2c
H3(jw) = H 33
for o = 0.27r/ 2 , we have

But

H 3a
=.2r

Thus
0.27r 23a=2
ft(3 2 33a=2a

Similarly,

(.3r)
j =2b

Thus, A,(s) satisfies the filter specifications for Hb(jw) exactly.

(c) H(ej") is given by

f(e") = - 1 [j (11 r)

But ft,(jo) = H, (jp,). Therefore,

ju 1 02 .j2 (Q 2x7k
ft(e ) =
2("
=
E- -
3
H0 3 2 2
1 00 [ ( 21rk) ej
k= 3 3

S24.9
(a) Using properties of the Laplace transform, we have
1
sY(s) = X(s), or H(s) ­
s
Signals and Systems
S24-10

(b) Here h is given by T, a is given by x[(n - 1)T], and b is given by x(nT). There­
fore, the area is given by
(a + b h = T[x((n - 1)T) + x(nT)] = An

(c) From the definition of 9[n], we find that

n-i
P[n - 11 = Z
k=-o
A,

Subtracting f[n - 1] from P[n], we find

n n-1

9[n) - [n - 11= Ak- A= A.

k= -ok= -- o

Therefore,
P[n] = y[n -1] + An.

(d) From the answer to part (a), we substitute for An, yielding

9[n] = P[n - 11 + T [x((n - 1)T) + x(nT)]

2

= P[n - 1] + {[n - 1] + &[n]}

(e) Using z-transforms, we find

f(z) = z -f(z) + T [z'i(z) + I(z)],

H(z) =k =zT 1+ z = H(s)
X(z) - 2 (1 zj s=2 t-21/1 +z-1>
MIT OpenCourseWare
http://ocw.mit.edu

Resource: Signals and Systems

Professor 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.
25 Feedback
Solutions to
Recommended Problems
S25.1

(a)

X(s) + ()o H(s) 0 Y(s)

YS)

Figure S25.1-1

We have
V(s) = X(s) - Y(s)K(s) (S25.1-1)
and
Y(s) = V(s)H(s) (S25.1-2)

From eq. (S25.1-2),

V(s) = H(s) (S25.1-3)

Substituting eq. (S25.1-3) into eq. (S25.1-1), we have

Y(S) = X(s) - Y(s)K(s),
H(s)
Y(s)[1 + H(s)K(s)] = H(s)X(s),
Y(s) H(s)
X(s) 1 + H(s)K(s)
Similarly,
Y(z) H(z)
X(z) 1 + H(z)K(z)
H(s) 1 H(z)
(b) Q(s) = Q(z) 1 + KHf(z)
1 + KH(s)'
For H(s) = 2/(s - 2) and H(z) = 2/(z - 2),
2 2
Q(s) = (s - 2) + 2K s - 2(1 - K)
2 2
Q(z) = (z - 2) + 2K z - 2(1 - K)

ForK = 0,
2 2
Q(S) = and Q(z) = z -2'
s- 2
as shown in Figures S25.1-2 and S25.1-3, respectively.

S25-1
Signals and Systems
S25-2

Im
s plane z plane

)( Re

2 / 2
IzI 1
Figure S25.1-2
Figure S25.1-3

For K = -1,
2 2
Q(s) = - and Q(z) = ,
s - 4 z -4
as shown in Figures S25.1-4 and S25.1-5, respectively.

Im
Im s plane z plane

X Re
4 Re
1 4
/ge 2
Figure S25.1-4 Figure S25.1-5

For K = 1,

Q(s) = 2S and Q(z) = 2

z
as shown in Figures S25.1-6 and S25.1-7, respectively.

Im
s plane

Re
Figure S25.1-6

2
(c) Q(s) =
s - 2(1 - K)
The pole is located at s = 2(1 -K), as shown in Figure S25.1-8.
 Feedback / Solutions
S25-3

Im s plane

K-*+00 2 Ko-
Re
K=O

Figure S25.1-8

Hence, the locus of the pole is the line Re{s} = 0. Similarly, for
2
Q(z) z - 2(1 - K)'

the locus of the pole is also the line Re{z} = 0, shown in Figure S25.1-9.

Im z plane

K++o 2 K--oo
Re
K=0

Figure S25.1-9

The root location decreases as K moves to infinity and increases as K moves to

negative infinity.
2
(d) Q(s) = - 2(1 - K)

The system is stable for 2(1 - K) < 0, or K > 1.

2
Q(z) = z - 2(1 - K)

The system is stable for -1 < 2(1 - K) < 1, or < K < i.

S25.2
We use Problem P25.1.

(a) (i) Y(s) H(s)

X(s) 1 + G(s)H(s)
(ii) E(s) = X(s) - R(s)
= X(s) - Y(s)G(s)
= X(s) - E(s)H(s)G(s),
E(s)[1 + H(s)G(s)] = X(s),
E(s) 1
X(s) 1 + H(s)G(s)
Signals and Systems
S25-4

(iii) E = H(s)
E(s)
1
R(s) G(s)

(b) W(z) = X(z) H (z)

1
1 + G(z)Hl(z)'
Y(z) = W(z) + X(z)Ho(z),

Y(z) = 1 + G(z)H 1 (z) + X(z)Ho(z)

Thus,
Y(z) = H 1 (z) + Ho(z)
X(z) 1 + G(z)H 1 (z)

(C)()W(S) = 1 + GsH(s) , as shown in Figure S25.2.

x(t) + H2(S) 1+G,(s)(s) --- y(t)

G2 (S):

Figure S25.2

H,(s)H2(s)
Y(s) + G1 (s)H1 (s)
1
X(s) 1 + G2(s)H 1(s)H 2(s)
1 + G1(s)H1 (s)
H1 (s)H 2(s)
1 + G1(s)H1 (s) + G2 (s)H 1 (s)H 2 (s)

S25.3

(a)
 Feedback / Solutions
S25-5

G(jw) H(jw)

100 - ~

---.. 5- 40-~10 J­

Figure S25.3-2

(b) From the frequency response in part (a), clearly system 1 tends to make the
response more constant and system 2 tends to resemble the inverse of G(jw).

S25.4
For the system in Figure S25.4-1, we denote the closed-loop system function by

H
V +GH

x(t) + H y(t)

Figure S25.4-1

1
(s + 1)(s + 3) 1
(a) V(s) =
1 (s + 1)(s + 3) + 1
1 + (s + 1)(s + 3)

s2 + 4s + 4 (s + 2)2

Therefore,
v(t) = te 2'u(t)

s + 3
(b) V(s)
(s + 3) + (s + 1)
1+ (s + 1)

2s + 4 2s + 2
In this case,
v(t) = le -2 t u(t)
(c) The system function G(s) = e-113 corresponds to a delay of 1, i.e., the feedback
system of Figure P25.4(a) becomes that shown in Figure S25.4-2.
Signals and Systems
S25-6

x(t) H(s) y(t)

Delay

Figure S25.4-2

We can now recursively obtain the impulse response by inspection. With

x(t) = 8(t),
y(t) =26t) i2116t -- 3)] + 1[46(t -- 3) ­

= O
- 1j ( t - )

2z­
(d) V(z) =

1+ 1 -(i' zz
z1 z
Z-1
(1-i ~)+ (!z~- Iz- 2)

z­1
1+ z-1 -z-2
-1

(1 -'-)(1 + i2z-I)
6 6
5 5
1
1-z- 1+ z-
Therefore,
v[n] = [)"u[n] -- (-)u[n]]
H(z) 2 - iz-1
(e) V(z) =
1 + H(z)G(z)
1 Z-) z-1
+2
1 +36 z-'
1 -iz­

(23 - iz-)(1 - iz-,)

(1 - iz-1) + (0 - iz-1)-1
2 - 2z- + nz-2
1-5z 1l~2
1+ 1z-1 - Iz-2

Thus,
v[n] = 3ib[n + 1] - 23[n] + 14[n - 1],
where f[n] is v[n] in part (d).
 Feedback / Solutions
S25-7

Solutions to
Optional Problems
S25.5
y(t) = K 2w(t) + KK 2v(t) (S25.5-1)
By taking the transform of eq. (S25.5-1), we have
Y(s) = K 2W(s) + KK 2Vs)

Also

V(s) = X(s) + s Y(s)

Therefore,

Y(s) = K 2W(s) + K 1 K 2 [X(s) + s Y(s),

s+ a

Y(s)( 1 - KK2S) = K2W(s) + KK 2X(s),

and
Y K2W(s) + K 1K 2X(s)
Y
=
- ) K 1K 2s
s + a
(s + a)[K 2W(s) + K1K 2X(s)]
(1 - K 1K2)s + a

S25.6
(a) The system function of the system given in Figure P25.6 must be determined
first. So we write down the difference equation
y[n] = x[n] + y[n - 1] + 4y[n - 2]

Taking the z-transform of the equation, we have

Y(z) 1
Y(z)(1 - z- - 4z- 2 ) = X(z), or H(z) =X(z) 1 - z-1 - 4z -2

The poles of this system are located at

z2 z - 4=0, or z 2
2 - 2
Since Iz I > 1 for at least one pole the system is unstable.
(b) With closed-loop feedback, the difference equation is
y[n] = x,[n] - Ky[n - 1] + y[n - 1] + 4y[n - 2]

Thus,
z2
H(z) = z2 + (K - 1)z - 4
Signals and Systems
S25-8

The poles are now located at

-(K - 1) ± \/(K - 1)2 + 16
Z= 2
Note that the roots are purely real because the term inside the square root is
always positive. For z = 1,
K 1 (K - 1)2 + 16
=+

2 2 - 2
K+ 1 \/(K - 1)2 + 16
Thus,
K 2 + 2K + 1 = K 2 - 2K + 17,
4K = 16, or K= 4
We can also calculate z 2:
Z2= -4
Similarly, zi = -1, z 2 = 4 for K = -2. Observe the root locus in Figure
S25.6-1.

Im Im
K> 0 z plane K <O z plane
K =-2
.00.9 K=4
Re Re
-1.56 2.56 -1.56 \ /12.56

Figure S25.6-1

Observe that if one of the poles is inside zI 1, the other is outside. Hence,
the system is unstable for all values of K.
(c) The difference equation can be written as
y[n] = x,[n] + y[n - 1] + (4 - K)y[n - 2]
Therefore,
z2
H(z) =
z- z + (K - 4)
In this case, the poles are located at
1 +/17 -4K
2 2
For a stable system, we want
izi < 1,
1 1/17 - 4K
|z| = ->+ 2
If we set 17 - 4K > 0, then

+ 17 - 4K <1 1
-<
 Feedback / Solutions
S25-9

/17 -4K 1
2 2'
17 - 4K < 1,
K> 4
Now suppose 17 - 4K < 0. Then

liIij\/V - KI77<1 or -K> ­

-K > -)
K< 5
Thus, for K in the range 4 < K < 5, we have a stable system. The root locus is
shown in Figure S25.6-2.

K>O K=5

-1.56 2.56

Figure S25.6-2

S25.7

(a) The de gain of the amplifier is IH(O)| = |G|.

(b) h(t) = Gae-"'u(t). Therefore, the time constant is 1/a.
G2 a2 1
(C) Hjc)2 22 a2 _ 2
a +w 2
Thus wc = ± a. Hence the bandwidth is a.
(d) The closed-loop transfer function is
Ga
ss~+ a Ga
V(s) V()
=
+ KGa (1 + KG)a + s
s +a
From part (a), the time constant is
1
(1 + KG)a
From part (c), the bandwidth is (1 + KG)a. From part (a), the dc gain is
| G |
1 + KG|
(e) We require (GK + 1)a = 2a. Hence, K = 1/G. So the bandwidth becomes 2a.
The time constant is 1/(2a), and IH(0) = I G/2 1,the de gain.
MIT OpenCourseWare
http://ocw.mit.edu

Resource: Signals and Systems

Professor 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.
MIT OpenCourseWare
http://ocw.mit.edu

Resource: Signals and Systems

Professor 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.
26 Feedback Example:
The Inverted Pendulum
Solutions to
Recommended Problems
S26.1
Ld 2 0(t)
(a) Ldz6(t) = g0(t) a(t) + Lx(t),
Ld2 (t)
dt2 - gA(t) = Lx(t)
Taking the Laplace transform of both sides yields
szLO(s) - gO(s) = LX(s),

O(s) = 2X(s)
s2 - g|L
0(s)
X(s) s- g/L (s + \g/L)(s - \ )

The pole at \g/L is in the right half-plane and therefore the system is unstable.
(b) We are given that a(t) = K0(t). See Figure S26.1-1.

x(t) o+2 1 - -+ 0(t)

Figure S26.1-1

0(s) H
X(s) 1 + GH'
so, with

and G=
H = s g/L

(s)/X(s) is given by
0(s) 1
X(s) s' - (g/L) + (K/L)

The poles of the system are at

s=+K-g L '

which implies that the system is unstable. Any K < g will cause the system
poles to be pure imaginary, thereby causing an oscillatory impulse response.

S26-1
Signals and Systems
S26-2

(c) Now the system is as indicated in Figure S26.1-2.

1
H(s) =1
2 g K1 K2
L L L
1
2K2
s+- s ±
K1 -- g
L L
The poles are at

-K2 + K2_ (K_ _ g)

2L ~ 2L L '
which can be adjusted to yield a stable system. A general second-order system
can be expressed as
Aw2
Hg(s) = s2 + 2Aws + w2'

so, for our case,

2 K 1 -g and 2,w = K2
Wn L L'
2
g = 9.8 m/s
L = 0.5 m
f = 1
O= 3 rad/s
2
K, = 14.3 m/s
K2 = 3 m/s

S26.2

(a) Here
2
H(s) =2 +2
2+ os + W2
G(s) = K
 Feedback Example: The Inverted Pendulum / Solutions
S26-3

The closed-loop transfer function He(s) is

H _2
He(s) = = "
1 + GH s 2 +2w.s+ w + nKw!
2
(A)n
s2 +2ws+ W2(1 +K
2

2
S +2 (n Wn
,, Z~
where w, = Wn(1 + K)"/2

± 2/(i) +
(s ( W,
here - = {­
s2 + F,+ )2 w
(A.n

Therefore,
= wn(l + K)" 2 ,

2
(1 + K) ,
2
A = =
An 1 +K'

for K = 1, = V2W, ands"= //2.

(b) Now we want to determine the poles of the closed-loop system

He(s) =
s2 + 2(,,s + (o2(1 + K)

The poles are at

-o ± /F~W - Wn(1 + K)

(c)
K = +oo Im
s plane

K=_xo=­
1 - ' Re

K = +oo

Figure S26.2

The poles start out at ± 0o, approach each other and touch at K = 2 - 1, and
then proceed to -fo, ±joo.
Signals and Systems
S26-4

S26.3

(a) = H 1 (s) = K 1K 2 (Os + r)K 1 K 2

X(s) Os + r + K 1K 2a
1 + K 1K 2a
#s + r
(b) Y(s) K2 _ (Os + r)K 2
= H 2 (s)
W(s) Os + r + K 1K 2a
1 + K1 K2a
Os + r
(c) For stability we require the pole to be in the left half-plane.

s= - r + KK 2a)

r + K1 K a
2 0
>0
If /3> 0, then r/a > -KK 2; if # < 0, then r/a < -K 1 K 2.

S26.4

K(s + 100)
H(s) = K
+ 1) s + 100 + Ks + K
1 + K(s
S + 100
K(s + 100)

(K + 1) s +100 + K)
K + 1]
(a) K = 0.01,
H(s) = 0.01(s + 100)
1.01(s + 99.0198)
The zero is at s = -100, and the pole is at s = -99.0198.
(b) K= 1,
S + 100
H(s) =
2 s+ 2
The zero is at s = -100; the pole is at s = -50.5.
(c) K =10,
H(s) = 10(s + 100)
11 s+ 110)
110
The zero is at s = -100; the pole is at s = -10.
(d) K = 100,
100(s + 100)
H(s) =
0200)
101 S
The zero is at s = -100; the pole is at s = -1.9802.
 Feedback Example: The Inverted Pendulum / Solutions
S26-5

S26.5

s-t+ 1 1
(a) H(s) = K +=1
+ K s+ 1+ K
1++1
The pole is at s = -1 - K, as shown in Figure S26.5-1.

Im Im
K>O s plane K<O s plane

- . Re Re
-1 -1

Figure S26.5-1

The pole moves from infinity to negative infinity as K changes from negative
infinity to infinity.
1
s-1i s+3
(b) H(s) =
1 (s + 3)(s - 1) + K
+1 (K
+ s
s + 3
2
s + 2s + K - 3
The poles are at s, = -1 ± \1 - (K - 3), as shown in Figure S26.5-2.

Im Im
K>O KK 0
s plane s plane
K =4

Re
-3 -1 +1 3

Figure S26.5-2

The poles start at ± oo when K = -oo, move toward -1, touch when K = 4,
and proceed to -1 ± joo as K approaches positive infinity.
Signals and Systems
S26-6

Solutions to
Optional Problems
S26.6
(a) The poles for the closed-loop system are determined by the denominator of the
closed-loop transfer function
Kz
1 + = 0
(z - 1)(Z - i')'

so
(z - -)(z - 1) + Kz = 0
Since we are told a pole occurs when z = -1, we want to solve the equation
for K:

K (z - 2)(z-) _ 15
z z=-1 8
(b) In a similar manner to that in part (a),

K (z - 2)(z -) -3
z 1z=1 8
(c) From the root locus diagram in Figure P26.6, we see that for K > 0 when K
exceeds a critical value of K = -', as determined in part (a), one root remains
outside the unit circle. Similarly, when K < -8, one root is outside the unit
circle. Therefore, to ensure stability, we need
- < K < '

S26.7
(a) The closed-loop transfer function is
Y(s) _ H(s) _ He(s)H,(s)
X(s) 1 + G(s)H(s) 1 + He(s)H,(s)
and, therefore, from the given He(s) and H,(s), we have
Ka
Y(s) S+ a Ka Ka
X(s) + Ka s + a + Ka s + (K + 1)a
s + a
The system is stable for denominator roots in the left half of the s plane; there­
fore -(K + 1)a < 0 implies that the system is stable.
Now since E(s)He(s)H,(s) = Y(s), we have
E(s) 1 s + a s + a
X(s) 1 + He(s)H,(s) s + a + Ka s + (K+ 1)a
The final value theorem, lim , e(t) = lim,_. sE(s), shows that

lim s(s= 0 for -(K + 1)a < 0

S-0 S+ (K + 1)a
 Feedback Example: The Inverted Pendulum / Solutions
S26-7

Note that if x(t) = u(t), then

E(s) = ( ) (K+ 1)a

and

lims s + # 0, for -(K + 1)a < 0

S-0 \sJ s + (K +1)a K+ 1
so limt-, e(t) # 0.

()Y(S_ He(s)H,(s)
X(s) 1 + He(s)H,(s)

K1+ -2 a
s )s + a

1 + K, + s2 a

SS + K2 Kia
(sK + K 2)a \ K 1
s(s + a) + (Kis + K 2)a S2 + sa(K + 1) + K 2a
The poles for this system occur at

- a(K + 1) (a(K+ 1) - K 2a

Note that if a(K + 1) > 0 and if K 2a > 0, we are assured that both poles are
in the left half-plane. Therefore, a(K + 1) > 0 and K 2a > 0 are the conditions
for stability. Now since
1
E(s) = X(s) 1 + He(s)H,(s)
1 s(s + a)
S s2 + a(K1 + 1)s + K 2a'
then
lim sE(s) = 0 implies that lim e(t) = 0,
s-0 t-CO
for a(KI + 1) > 0 and K 2a > 0, so we can track a step with this stable system.

S26.8

(a) = H(s)C(s)
X(s)

(s + 1)(s - 2)(s +
We can see from this expression that the overall transfer function for the sys­
tem is
Y(s) 1
X(s) (s + 1)(s + 3)'
a stable system. In effect, the system was made stable by canceling a pole of
H(s) with a zero of C(s). In practice, if this is not done exactly, i.e., if any com­
Signals and Systems
S26-8

ponent tolerances cause the zero to be slightly off from s = 2, the resultant
system will still be unstable.

(b) Y(s) C(s)H(s) K

X(s) 1 + C(s)H(s) (s + 1)(s - 2) + K
K
s2 - s + K - 2
The poles are at

2-v - (K - 2)

We see from this that at least one pole is in the right half-plane, i.e., there is
instability for all values of K.
1

(c)()Y(S) = 1+K(s + a) (s +
X() ____+_a)_(s+_1)(s_-_2 1 -
1)(s 2)
X(s) 1____
l+~sa(±+ )(- 2 )
K(s + a)
(s + 1)(s - 2) + K(s + a)
K(s + a) K(s + a)
s2 - s - 2 + Ks + Ka s 2 + (K - 1)s + (Ka - 2)
The poles are at

(K - 1) K_
- 2 -V -2- - (Ka - 2)

Now, if Ka - 2 > 0, the system is stable. K > 2/a because a > 0 is assumed.
This is true for 1 > a > 0 and 2 > a > 1. For a > 2, the system is stable for
K > 1.
Y(S) K(s + a)
(d) X(s) 2
s 2 + (K - 1)s + (Ka - 2)'
We want K - 1 = w,, 2K - 2 = W. So
(K - 1)2 = 2K - 2,
K = 3 or K= 1
IfK= 1, thenw, = 0, so K= 3 implies thatw, = 2.

S26.9

E(s) 1 S1
(a) -_____-where
X(s) 1 + H(s) s' + G(s)'

K f (s - #K
k=1
G(s) n-1

17 (s - aK)
k=1

For s = 0, G(s) constant = g.

(1/s)s' s___
E(s) = , and lim sE(s) = lim = 0
s + g s-0 s--o S + g

Thus, lim e(t) = 0.

 Feedback Example: The Inverted Pendulum / Solutions
S26-9

S-I
(b) E(s) = for 1 = 1, x(t) = u- 2(t)
s + G(s)
So
1 1
lim sE(s) = Constant
S-0 s + G(s) 1,=o g
5 1-k 2-k
5
(c) E(s) = s + G(s), sE(s) = s +
G(s)
For k > 2,
lim sE(s) = co, lim e(t) = oo
s-o

S 1-k
(d) (i) E(s) = ,G sE(s)
S' + G(s) s' + G(s)
If k :5 1, then
1-k+1
lim sE(s) = lim 0 0,= 0,
s-O s-o s' + G(s) 0+ g
so lim,_., e(t) = 0.
(ii) If k = 1 + 1 and since
1-k
E(s) = s' +
G(s)

then
1 1
lim sE(s) = lim = Constant
S-0 S-O S' + G(s) g
Thus, lim,me(t) = Constant.
(iii) If k > 1 + 1, then since
S1-k gl-k+1
E(s) = S' + G(s) sE(s)
s' + G(s)
lims- 0 sE(s) = oc implies lim_. e(t) = 00.

S26.10

E(z) 1
(a) X(z) 1 + H(z)
z
X(z) z- 1 z(z + i)
E(z) = 1 + H(z) 1 (z - 1)(z + i) + 1
1+(z - 1)(z + 1)
z 2 + .z - 2
1±2+
z2 2Z + -2 Z'2
- Z + 2

The poles are at i + - . These poles are inside the unit circle and therefore
yield stable inverse z-transforms, so e[n] = b[n] + (2 stable sequences). So
lim-,.e[n] = 0.
Signals and Systems
S26-10

(b) H(z) = A(z)

(z - 1)B(z)
since H(z) has a pole at z = 1. Now
E(z) 1 (z - 1)B(z)
X(z) 1 + H(z) (z - 1)B(z) + A(z)'

)(z - 1)B(z)
1I
-z(z - 1)B(z) for x[n] = u[n]
+ A(z)
zB(z)
(z - 1)B(z) + A(z)
Furthermore, we know that
Y(z) H(z) (z - 1)B(z)
X(z) 1 + H(z) (z - 1)B(z) + A(z)
There are no poles for Iz I > 1 because h[n] is stable. Therefore,
zB(z)
E(z) =
(z - 1)B(z) + A(z)
has no poles for IzI > 1, and lim,-e[n] = 0.

(c) H(z) =
- z 1'
E(z) 1 z - 1
X(z) 1 + H(z) z
z-1 fz-1\
E(z) = z1 X(z)
zz (z for x[n] = u[n]
= 1 =* e[n] = b[n],

so e[n] = 0, n -1
2z 1 + 1z- 2
(d) H(z)
(1+ -Iz ')(1 - z-1)'

E(z) 1 (1 + iz-1)(1 - z-1)

X(z) 1 + H(z) (1 + iz-')(1 - Z-1) + 4z-1 + ,z-2,

(1 + tz-1)
E(z) = 1
(1 + Iz- )(1 - z-1) + z + -2
= 1+ tz-'
Therefore,
e[n] = b[n] + '6[n - 1]
=0, n -2

E(z)
()X(z)
= 1H H(z) X(z 1
1 + H(z) ' E(z)
For x[n] = u[n], we have
1
X(z) =1 -1
 Feedback Example: The Inverted Pendulum / Solutions
S26-11

We would like
N-1
e[n] = a[n - k],
k=O

N-1

E(z) = akz-'
k=O

Therefore,
N-1
1 z1)
Z1- T akz- k)
(k=0
H(z) = N-­
(1-- 1) T akz -k)
(k=0

z-I + z- 2 - z- E(z) 1
(f) H(z) =
(1 + z-1)(1 - z)2 X(z) 1 + H(z)
Now x[n] = (n + 1)u[n] and

X(z) = (1 - z-1)2>

1
(1 + z-1)(1 - z-1)2 (1 - z-1)2
E(z) = 2 1 2
(1 + Z- 1)(1 - z-1) + z- + z- -z
1 + z-'

and
e[n] = b[n] + b[n - 1]
=0, n >- 2
MIT OpenCourseWare
http://ocw.mit.edu

Resource: Signals and Systems

Professor 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.