Solutions Manual for

Understanding Digital
Signal Processing
Third Edition

Richard G. Lyons
Antoine Trux

Upper Saddle River, NJ • Boston • Indianapolis • San Francisco

New York • Toronto • Montreal • London • Munich • Paris • Madrid
Capetown • Sydney • Tokyo • Singapore • Mexico City

Property of Pearson Education. Not permissible for redistribution.

The author and publisher have taken care in the preparation of this book, but make no
expressed or implied warranty of any kind and assume no responsibility for errors or
omissions. No liability is assumed for incidental or consequential damages in connection
with or arising out of the use of the information or programs contained herein.

Visit us on the Web: informit.com

Copyright © 2011 Pearson Education, Inc.

This work is protected by United States copyright laws and is provided solely for the use
of instructors in teaching their courses and assessing student learning. Dissemination or
sale of any part of this work (including the World Wide Web) will destroy the integrity of
the work and is not permitted. The work and materials from it should never be made
available to students except by instructors using the accompanying text in their classes.
All recipients of this work are expected to abide by these restrictions and to honor the
intended pedagogical purposes and the needs of their instructors who rely on these
materials.

ISBN-13: 978-0-13-218106-8
ISBN-10: 0-13-218106-1

Text printed in the United States at OPM in Laflin, Pennsylvania.

First printing, December 2010

Property of Pearson Education. Not permissible for redistribution.

 CHAPTER 1 PROBLEMS

Solution: 1.1
The solution is:

100
π ≈ 4 ⋅ ∑ ( −1) ⋅ 1 .
n
2n + 1
n =0

Solution: 1.2
There are many correct solutions to this problem. An example of a continuous
time-domain signal that has a finite number of amplitude values is a
squarewave, such as the x1(t) shown in Figure S1–2(a). Signal x1(t) has only two
possible amplitude values. Of course, any bi-level pulsed signal, like the x2(t)
shown in Figure S1–2(b) is also a correct solution to this problem. Figure S1–
2(c) shows a continuous signal having only three possible amplitude values.

Solution:
Three possible solutions:
x1(t)
1 ...
(a)
0
0 t

x2(t)
1 ...
(b)
0
0 t

x3(t)
2
(c) 1 ...
0
0 t

Figure S1–2

Solution: 1.3
The code: PI = 2*asin(1.0) correctly defines π
under the assumption that the arcsin's angle argument, "(1.0)",
is measured in radians.

Solution: 1.4
Property of Pearson Education. Not permissible for redistribution.
 From one of the Laws of Exponents, we can
write:
xp p–q
xq = x .
Solve the problem by setting q = p, giving us
xp p–p 0
xp = 1 = x = x .

Solution: 1.5
The cosine sequences are as follows:

(a) x1(n) = cos(2πfonts) = cos[2π(fs/2)nts] = cos[2π(fs/2)n(1/fs)] = cos(πn).

(b) x2(n) = cos(2πfonts) = cos[2π(fs/4)nts] = cos[2π(fs/4)n(1/fs)] = cos(πn/2).

(c) x3(n) = cos(2πfonts) = cos(2π0nts) = 1.

Solution: 1.6

Solution:
x1(n)
1 ...
1 3 5
(a) 0
0 2 4 6 n
–1

x2(n)
1 ...
2 6
(b) 0
0 1 3 4 5 n
–1

x3(n)
1 ...
(c) 0
0 1 2 3 4 5 6 n

Figure S1–6

Solution: 1.7

Property of Pearson Education. Not permissible for redistribution.

 Solution:
x1(n)
1 ...
1 3 5
(a) 0
0 2 4 6 n
–1

x2(n)
1 ...
3
(b) 0
0 1 2 4 5 6 n
–1

x3(n)
1 ...
(c) 0
0 1 2 3 4 5 6 n

Figure S1–7

(d) Sampled versions of sinewaves, whose frequencies are fs/2

and 0 Hz result in sequences where each time sample is zero.

Solution: 1.8
The desired xshift(n) = x(n+1) sequence is shown in Figure S1–8(b).

Solution:
x(n)
1 ...
2 6
(a) 0
0 1 3 4 5 7 n
–1

xshift(n) = x(n+1)
1 ...
1 5
(b) 0
0 2 3 4 6 7 n
–1

Figure S1–8

Solution: 1.9
In our text, we represent a sinusoidal sequence using the form

m(n) = sin(2πfonts)
Property of Pearson Education. Not permissible for redistribution.
 where fo is the sinusoid's frequency measured in Hz. Setting that expression
equal to the problem's m(n) expression, we can write:

m(n) = sin(2πfonts) = sin(0.8πn).

Setting the above angle arguments equal to each other, we write:

2πfonts = 0.8πn.

Recalling the definition that ts = 1/fs, solving the above expression for the
frequency fo, we have our solution of:
0.8πn 0.8
fo = = ⋅ f s = 0.4 ⋅ 2500 = 1000 Hz.
2πnts 2

Solution: 1.10
With N = 6 and n = 9, the computation needed to compute y(9) is

5
y(9) = ∑ x(9–p) = x(9) + x(8) + x(7) + x(6) + x(5) + x(4).
p=0

Solution: 1.11
(a) The block diagram implementing
n

y(n) =
∑
k=n–4
1
5 x(k)

is shown in Figure S1–11(a).

(b) The impulse response of a five-point moving averager is shown in Figure

S1–11(b).

Solution:
x(n) x(n–1) x(n–2) x(n–3) x(n–4)
Delay Delay Delay Delay

(a) 1/5 1/5 1/5 1/5 1/5

y(n)

y(n)
0.2 ...
(b)
0
0 1 2 3 4 5 6 7 8 9 n

Property of Pearson
FigureEducation.
S1–11 Not permissible for redistribution.
 (c) Implementing Eq. (P1–1) is preferred over implementing Eq. (P1–2)
because Eq. (P1–1) requires fewer multiplications (lower
computational workload) to compute each y(n) output sample.

Solution: 1.12
On the musical scale, the dimension of the x-axis is time, and the dimension of
the y-axis is frequency.

Solution: 1.13
Using the trigonometric identity:
cos(α+β) + cos(α–β) = 2cos(α)cos(β) (1.13–1)
and the problem's original
x(n) = cos(2πfonts + φ) + cos(2πfonts)
expression, we can write two simultaneous equations as
α + β = 2πfonts + φ (1.13–2)
and
α – β = 2πfonts. (1.13–3)
Solving Eqs. (1.13–2) and (1.13–3) for α and β yields
α = 2πfonts + φ/2, and β = φ/2.
Substituting α and β into Eq. (1.13–1) we write
cos(2πfonts + φ) + cos(2πfonts)

= 2cos(2πfonts + φ/2)cos(φ/2). (1.13–4)

Given Eq. (1.13–4), the solution to this problem is

x(n) = 2cos(2πfonts + φ/2)cos(φ/2).

Solution: 1.14
The x = α and y = sin(α) curves are shown in Figure S1–14. There we see that
over the range of roughly α = –0.1π to α = 0.1π the statement "For small α,
sin(α) = α" statement is valid.

Property of Pearson Education. Not permissible for redistribution.

 Solution:
2
π/2
x=α
1
y = sin(α)
0.5
0
–0.5
–1

–π/2

–0.5π –0.3π –0.1π 0 0.1π 0.3π 0.5π

α

Figure S1–14

Solution: 1.15
The solutions are:

(a) sin(2πfot + α) = cos(2πfot + α – π/2).

(b) cos(2πfot + α) = sin(2πfot + α + π/2).

Solution: 1.16
First we determine how many discrete samples are required to represent a single
cycle of the analog sinewave. That number of samples per analog signal cycle is
found using

fs in samples/sec
samples/cycle = analog sinewave frequency in cycles/sec

100x106 samples/sec
= 25x106 cycles/sec = 4 samples/cycle

as shown in Figure S1–16.

x(n)
1 ...
3 7
0
0 1 2 4 5 6 n
–1

Figure S1–16

The problem solution is the number of sampled-sinewave cycles that can be

stored in aProperty of Pearson
4x106-sample memory,Education.
or Not permissible for redistribution.
 4x106 samples
stored sinewave cycles = 4 samples/cycle = 106 cycles.

Solution: 1.17
The proportionality characteristic of a linear system, in the text's Eq. (1–14),
states that if input sequence x(n) yields output y(n),
results in
x(n) y(n),
then a scaled input sequence cx(n), where c is some constant scalar value, yields
a scaled output cy(n),
results in
cx(n) cy(n).
(a) For system ya(n) = x(n–1)/6, the answer is Yes.
For example, if we consider a new x'(n) = 2x(n) input, then the new ya'(n)
output sequence is
ya'(n) = x'(n–1)/6 = 2x(n–1)/6 = 2ya(n).

(b) For system yb(n) = 3 + x(n), the answer is No.

For example, if we consider a new x'(n) = 2x(n) input, then the new yb'(n)
output sequence is
yb'(n) = 3 + x'(n) = 3 + 2x(n) ≠ 2yb(n) = 6 + 2x(n).

(c) For system yc(n) = sin[x(n)], the answer is No.

For example, if we consider a new x'(n) = 2x(n) input, then the new yc'(n)
output sequence is
yc'(n) = sin[2x(n)] ≠ 2yc(n) = 2sin[x(n)].

Solution: 1.18
Decimation is not time-invariant.
An example of this, where yshift(m) ≠ y(m+1), is as follows:

Property of Pearson Education. Not permissible for redistribution.

 Solution:
x(n)
1 ...
2 6
(a) 0
0 1 3 4 5 7 n
–1

Decimated output sequence y(m) = x(2n)

1 ...
(b) 0 1 3 5 7
0 2 4 6 m
–1

Shifted time sequence xshift(n) = x(n+1)

1 ...
1 5
(c) 0
0 2 3 4 6 7 n
–1

Decimated time sequence yshift(m) = xshift(2n) = y(m+1)

1 ...
(d) 0
0 1 2 3 4 5 6 7 m

Figure S1–18

Solution: 1.19
We prove the two networks in Figure S1–19 exhibit the commutative property
of linear time-invariant systems as follows:
For the network in Figure S1–19(a) we write output y1(n) as

y1(n) = Ax(n) + By1(n–1).

Regarding the network in Figure S1–19(b) we write output y2(n) as

⎡ B ⎤
y2 (n) = A ⎢ x(n) + y2 (n − 1) ⎥ = Ax(n) + By2 (n − 1)
⎣ A ⎦

which is identical in form to the above y1(n) expression, which is what we set
out to prove.

x(n) y1(n) x(n) y2(n)/A y2(n)

Delay Delay
A A
y1(n–1) y2(n–1)/A

B B
Property
(a) of Pearson Education. Not
(b) permissible for redistribution.
 Figure S1–19

Solution: 1.20
The block diagram solutions to these problems are shown in Figure S1–20.

Solution:
4th-order comb filter
x(n) x(n–1) x(n–2) x(n–3) x(n–4)
Delay Delay Delay Delay

Notice the
(a)
minus sign
–
yC(n)

Integrator Leaky integrator yLI(n)

x(n) yI(n) x(n)

Delay Delay
A
yI(n–1) yLI(n–1)

(b) (c)
(1-A)

Differentiator (several equivalent versions)

x(n) x(n)
Delay Delay Delay Delay

0.5 –0.5 + –
yD(n)
yD(n)
(d) 0.5

x(n) + yD(n)
x(n)
Delay Delay
Delay –
0.5 + – 0.5
yD(n) Delay

Figure S1–20

Solution: 1.21
The impulse response solutions to these problems are shown in Figure S1–21.

Property of Pearson Education. Not permissible for redistribution.

 Solution:
y(n) 4th-order comb filter
1
...
(a) 4
0
0 1 2 3 5 6 7 8 9 n
–1

y(n) Integrator
1
(b) ...
0
0 1 2 3 4 5 6 7 8 9 n

y(n)
0.5 Leaky integrator

(c) 0.0312
0.25
0.0625
0.125 0.0156 ...
0
0 1 2 3 4 5 6 7 8 9 n

y(n)
Differentiator
0.5

...
(d) 0
0 1 2 3 4 5 6 7 8 9 n

–0.5

Figure S1–21

Solution: 1.22
The step response solutions to this problem are shown in Figure S1–22.

Property of Pearson Education. Not permissible for redistribution.

 Solution:
y(n) 4th-order comb filter
1
...
(a)
0
0 1 2 3 4 5 6 7 8 9 n

y(n) Integrator
9
7
(b) 5 ...
3
1
0 1 2 3 4 5 6 7 8 9 n

y(n) Leaky integrator

1.0
0.875
(c)
0.75
...
0.5
0 1 2 3 4 5 6 7 8 9 n

y(n) Differentiator
0.5
(d) ...
0
0 1 2 3 4 5 6 7 8 9 n

Figure S1–22

Solution: 1.23
(a) The original s(t) and the negative of the fundamental frequency
[(4A/π)sin(2πfot)] are shown in Figure S3–23(a). Adding those two
waveforms results in the interesting waveform in Figure S3–23(b) that is the
solution to this problem.

s(t) –4Asin(2πfot)/π

A
... ...
(a)
t=0 t (Time)

Property of Pearson Education. Not permissible for redistribution.

 Solution:
[s(t) – 4Asin(2πfot)/π]
A
... ...
(b)
t (Time)

–A

Figure S1–23

(b) The operating frequency range of an amplifier needed to exactly double the
ideal s(t) squarewave's peak-peak amplitude would be infinitely wide!

Solution: 1.24
Step 5 is the illegal step because it is an incomplete square root operation.
The square root of q2 is equal to ±q. So following Step 4, Step 5 should have
been:

±(4 – 9/2) = ±(5 – 9/2), or

±(0.5) = ±(–0.5).

Property of Pearson Education. Not permissible for redistribution.

 Chapter 2 Solutions

Solution: 2.1
(a) The photos would incorrectly indicate that the clock's minute
hand is rotating counterclockwise (anti-clockwise).

(b) With the idea of lowpass sampling in mind, because the minute hand rotates
at a frequency of one cycle/hour, to show the true clockwise minute hand
rotation:
Photos must be taken at rate of more than two photos/hour. That is, the
time between photos must be less than 1/2 hour.

Thinking a bit more deeply about this problem, it is true to say that if the
time T between photos is in the range K < T < (K + 0.5), where T is in
hours and K is a positive integer, then the sequence of photos would
indicate correct clockwise minute hand rotation. (The above Part (b)
solution is the case where K = 0.) For example, if T was one hour and
twenty minutes, clockwise minute hand rotation would be seen in the
sequence of photos.

Solution: 2.2
The important missing information is the fs sample rate.
Without knowing fs, we cannot use the x(n) samples to
characterize either the time-domain or frequency-
domain nature of the x(t) signal.

Solution: 2.3
(a) The ts period for a 2 GHz sample rate is:

1 1
ts = f = 2x109 = 0.5x10–9 seconds = 0.5 nanoseconds.
s

Note: 0.5 nanoseconds is a very short period of time—in that time light
travels only 5.9 inches (15 cm).

(b) With the time between samples being ts = 0.5x10–9 seconds, 256x106
samples represents a time interval of:

Max time interval ≈ (256x106 samples)(0.5x10–9 seconds/sample)

= 128x10–3 = 0.128 seconds.

Property of Pearson Education. Not permissible for redistribution.

Solution: 2.4
We can write the standard form of an x(n) discrete signal by replacing the
"t" in the continuous signal's x(t) equation with "nts". Doing so we have

x(n) = cos[2π(500)nts + π/7].

Knowing that ts = 1/fs, the problem solution is:

x(n) = cos[2π(500)n/fs + π/7] = cos[2π(500)n/4000 + π/7]

= cos[2πn/8 + π/7].

Solution: 2.5
The sample period ts must be less than half the sinewave's period of to = 1/fo.
to 1
Thus ts must be less than 2 = 2f , or
o

1
0 < ts < 2f .
o

Solution: 2.6
The answer is N = 2.
We must obtain no less than 2 discrete samples, in time, per analog
sinewave cycle. We verify this for an analog sinewave of frequency fo Hz
using the lowpass Nyquist criterion, fs ≥ 2fo. The periods of fs and 2fo must
satisfy

1 1 to
fs ≤ 2fo , or t s≤
2 , or 2ts ≤ to .

We interpret 2ts ≤ to as "No less than 2 sample-time intervals must be

contained within the time interval to = 1/fo.

Solution: 2.7
Yes, a continuous sinewave signal whose frequency is fs/2.
For example, a continuous x(t) signal defined by:

x(t) = sin[2π(fs/2)t].

With the continuous t replaced with nts, and recalling that ts = 1/fs, the
discrete samples are:

x(n) = sin[2π(fs/2)nts] = sin[2π(fs/2)n/fs] = sin(πn) = 0.

Property of Pearson Education. Not permissible for redistribution.

 The analog x(t) signal and the discrete x(n) sequence are shown in Figure
S2–7.

x(n)
x(t) ts
1 ...
0
0 1 2 3 4 5 6 n
–1

Figure S2–7

Solution: 2.8
Ignoring the days when the Stock Exchange is closed, the ts period
for standard stock market charts is one day, 24 hours.

Solution: 2.9
We can write x(t) in the standard form of:

x(t) = cos(4000πt) = cos(2πfot).

So the sinusoid's frequency is fo = 2000 Hz. We can write x(n) in the

standard form of:

x(n) = cos(nπ/2) = cos(2πfonts) = cos(2πfon/fs).

Equating the first and last angle arguments of the above cosine expressions
yields

nπ/2 = 2πfon/fs,

or
2πfon
fs = = 4fo = 4(2000) = 8000 Hz.
nπ/2

Sequence x(n) is shown in Figure S2–9.

x(n)
1
...
2 6
0
0 1 3 4 5 n
–1

Figure S2–9

Solution: 2.10

Property of Pearson Education. Not permissible for redistribution.

 First we find the highest-frequency spectral component in x(t) using the
trigonometric identity: cos(α)cos(β) = 0.5cos(α–β) + 0.5cos(α+β). Using
that identity, we express x(t) as

x(t) = 0.5cos(4000πt – 200πt) + 0.5cos(4000πt + 200πt)

= 0.5cos(3800πt) + 0.5cos(4200πt).
The highest-frequency spectral component in x(t) is 2100 Hz. So, the
minimum acceptable fs sample rate is twice that frequency, or

fs,min = 4200 Hz.

Solution: 2.11
The first step to solving this problem is finding frequency fo in terms of fs.
We can write x(n) in the standard form of:

x(n) = sin(nπ/4) = sin(2πfonts) = sin(2πfon/fs).

Equating the first and last angle arguments of the above sin expressions
yields
nπ/4 = 2πfon/fs,
or
fsnπ/4 fs 160
fo = = 8 = 8 = 20 Hz.
2πn
Next, knowing this one possible value for fo (20 Hz), we write the text's Eq.
(2–5) as
x(n) = sin(nπ/4) = sin(2πfonts) = sin(2π(fo + kfs)nts),
reminding us that "aliases" of fo are fo+kfs, where k is any integer. Using
k = 1 and k = 2, we solve the problem by stating that three possible positive
frequency values for fo that would result in sequence x(n) are:
fo = 20 Hz,
fo = 20 + 1·160 = 180 Hz, and
fo = 20 + 2·160 = 340 Hz.
or any of:
fo = 20 + k·160 Hz, for integer k.

Solution: 2.12
By drawing the Xd(f) spectrum in Figure S2–12 showing spectral
replications as dashed lines, the problem solution is the following table.

Property of Pearson Education. Not permissible for redistribution.

 Xd(f)
3 4 3 4 3 4
2 5 2 5
1 6

–fs –fs/2 0 fs/2 fs f

Figure S2–12

Aliased Spectral Segments:

Xd(f) spectral Associated Xa(f)

segment spectral segment
A 3
B 4
C 5
D 2
E 1
F 6

Solution: 2.13
Based on the expression x(t) = sin(2π700t), the frequency of the analog
sinusoidal tone is 700 Hz and its spectrum is shown in Figure S2–13(a).

Spectrum of analog x(t)

(a)
–700 0 700 Hz

Figure S2–13

The desired magnitude spectrum of x(n) is shown in Figure S2–13(b).

Solution:
|X(f)|, spectrum of x(n)

... ...
(b)
–2000 –1000 –fs/2 0 fs/2 1000 2000 Hz
(–2fs) (–fs) (fs) (2fs)
–700 700
–1300 –300 1700
(700–2fs) (700–fs) (700+fs)
–1700 300 1300
(–700–fs) (–700+fs) (–700+2fs)

Property of Pearson Education.

Figure S2–13 Not permissible for redistribution.
(Cont'd)
Solution: 2.14
(a) The spectrum of the analog s(t) signal is shown in Figure S2–14(a).The
desired spectrum of the x(n) sequence, X(f), is shown in Figure S2–14(b),
where the dashed curves are spectral replications.

|A(f)|

(a)
–70 0 70 Freq
(MHz)

Solution:
|X(f)|
56 56
MHz MHz

(b)

–70 –42 –14 0 14 42 70 Freq

–fs –fs/2 fs/2 fs (MHz)
(–56) (–28) (28) (56)

Figure S2–14

(b) The center frequency of the first positive-frequency

spectral replication in X(f) is 14 MHz.

(c) The 14 MHz from Part (b) is exactly one-fourth the fs

sample rate.
(This bandpass sampling scenario is quite useful for certain types of
follow-on processing of x(n).)

Solution: 2.15
As shown in Figure S2–15, the minimum fs sample rate for lowpass
sampling such that no spectral overlap occurs in the frequency range of 2 -
to- 9 kHz in the spectrum of the discrete x(n) samples is:

fs,min = 19 kHz.

Sampled |X(f)|
7 kHz 7 kHz

–1 0 1 2 9 10 17 19 Freq
(fs) (kHz)

Property of Pearson Education. Not permissible for redistribution.

 Figure S2–15

Solution: 2.16

This is "trick question". Bandpass sampling is not

possible for the opera singer's full-spectrum audio signal
shown in Figure S2–16.

The signal's bandwidth (928 Hz) is too wide relative to its center frequency
(711 Hz) to enable bandpass sampling.

|X(f)|

928 Hz

–2000 –1175 –247 0 247 1175 2000 Freq

–711 fc = 711 (Hz)

Figure S2–16

Solution: 2.17
(a) The spectrum of the u(t) output of the mixer will be the spectrum of signal
w(t), shown in Figure S2–17(a), translated up in frequency centered at 50
MHz + fLO and a copy of the spectrum of w(t) translated down in frequency
to be centered at 50 – fLO, as shown in Figure S2–17(b). To ensure that the
low-frequency image is centered at 15 MHz, then

15 MHz = 50 – fLO MHz, or

fLO = 50 – 15 = 35 MHz.

Property of Pearson Education. Not permissible for redistribution.

 bandwidth B = 10 MHz
|W(f)|

(a)

–50 0 50 MHz

|U(f)|

(b)
0
–85 –15 15 85 MHz
(–fc) (fc)

|X(f)|
bandwidth B = 5 MHz
(c)
0
–85 –15 15 85 MHz
(–fc) (fc)

Figure S2–17

(b) Analog bandpass filter# 2 has two purposes. First, it filters out (attenuates)
the high-frequency image spectrum that's centered at 85 MHz in U(f).
Second, it reduces the bandwidth of the desired image signal centered at 15
MHz to a bandwidth of 5 MHz, as shown in Figure S2–17(c).

(c) The correct table entries are as follows (don't forget, B = 5 MHz for x(t)):

Bandpass fs Sample Rate Ranges in MHz

m (2fc – B)/m (2fc + B)/(m+1) Positive-frequency
[(30 – 5)/m] [(30 + 5)/(m+1)] sampled spectrum is
inverted (Yes or No)
1 25.0 17.5 Yes
2 12.5 11.666 No
3 8.75 8.333 Nyquist is violated

(d) The text's Eq. (2–11) tells us that to force the sampled spectrum to be
centered at fs/4, then fs must satisfy:
4fc
fs = 2k – 1 ,

where k is an integer. Knowing that fs is 12 MHz, we can rewrite the above

equation as:
fs (2k–1) 12(2k–1)
fc = 4 = 4 = 6k – 3 Mhz.

As such, the fc center frequencies of the U(f) and X(f) spectra, and the
acceptable fLO local oscillator frequencies, can be list (in MHz) as:

k fc fLO
Property of Pearson Education. Not permissible for redistribution.
 1 3 50 – 3 = 47 MHz
2 9 50 – 9 = 41 MHz
3 15 50 – 15 = 35 MHz
4 21 50 – 21 = 29 MHz
... ... ...

So the solution to this problem, the maximum acceptable fLO local

oscillator frequency, is
max fLO = 47 MHz.
(In this case, the center frequency of Analog bandpass filter# 2 will be
fc = 3 MHz.)

Solution: 2.18
If we use the variable f–60 dB to represent the frequency at which the anti-
aliasing filter must have an attenuation value of –60 dB, as shown in Figure
S2–18(a), then

f–60 dB = fs –B Hz.

The solution to this problem is found by inspection as shown in Figure S2–

18(b).

|H(f)|
Analog filter
2B magnitude
response
(a) 60 dB

0 f–60 dB Freq

|X(m)|
B B

(b) 60 dB

–B 0 fs Freq

fs –2B
f–60 dB = B + (fs –2B) = fs –B

Figure S2–18

Solution: 2.19
The problem solution is the spectrum shown in Figure S2–19.

Property of Pearson Education. Not permissible for redistribution.

 Solution:
Spectrum of x(n) sequence

–1000 –fs/2 0 fs/2 1000 Hz

(–fs) (fs)
Lowpass signal
spectrum

Figure S2–19

Solution: 2.20
(a) For bandpass sampling, the absolute minimum sampling rate is fs = 2B, as
shown in Figure S2–20(a). So the minimum fc center frequency, in terms of
x(t)'s bandwidth B, that enables bandpass sampling of x(t) is

fc, minimum = 1.5B = 3B/2.

(b) Here we force the student to find out, on their own, that the AM broadcast
radio band in North America covers the frequency range of roughly 530
kHz to 1710 kHz (depending on who you ask). This AM broadcast-band
spectrum has a bandwidth of BAM = 1180 kHz and a center frequency of
fc,AM = 1120 kHz, as shown in Figure S2–20(b).
From the above Part (a), because fc,AM is less than 1.5BAM (1770
kHz), the full spectrum of the commercial AM broadcast band
cannot by bandpass sampled.

|X(f)|
B B B B

(a)

–fc 0 Freq
fc fs
1.5B

|XAM(f)| BAM = 1180

(b)

–fc,AM 0 fc,AM = 1120 kHz

Figure S2–20

Solution: 2.21
The correct table entries are as follows:

Acceptable Bandpass Sample Rate Ranges

Property of Pearson Education. Not permissible for redistribution.
 m (2fc –B)/m (2fc +B)/(m+1) Positive-frequency
sampled spectrum is
inverted (Yes or No)
1 45.0 27.5 Yes
2 22.5 18.33 No
3 15.0 13.75 Yes
4 11.25 11.0 No
5 9.0 9.17 Nyquist is violated

Solution: 2.22
The answer is no, the suggested algorithm does not compute the
absolute minimum fs bandpass sampling rate.

We show this by implementing the algorithm based on the text's Section 2.3
example of a B = 5 MHz-wide bandpass signal center at fc = 20 MHz. Doing
that, we have
4fc + 2B 80 + 10
Z=⎣ 4B ⎦ = ⎣ 20 ⎦ = ⎣4.5⎦ = 4.
Given Z = 4, the algorithm's fs,min,web is
4fc 80
fs,min,web = 2Z –1 = 8 –1 = 11.43 MHz.

From the text's Table 2-1, we see that the minimum fs sample rate (when
m = 3) is 11.25 MHz, which is less than the web site algorithm's fs,min,web.
This verifies that the suggested algorithm does not compute absolute
minimum fs bandpass sampling rates.

Figure S2–22(a) shows the spectrum of the analog bandpass signal. Figure
S2–22(b) and (c) show the spectra of the discrete bandpass signals sampled
at the two different sample rates.

Property of Pearson Education. Not permissible for redistribution.

 B = 5 MHz
X(f)

(a)
–25 –20 –15 –10 –5 0 5 10 15 20 25 MHz

fs = 11.43 MHz Xweb(f)

(b)
–25 –20 –15 –10 –5 0 5 10 15 20 25 MHz
11.43 11.43

fs = 11.25 MHz Xtext(f)

(c)
–25 –20 –15 –10 –5 0 5 10 15 20 25 MHz
11.25 11.25

Figure S2–22

Property of Pearson Education. Not permissible for redistribution.

 Chapter 3 Solutions
Solution: 3.1
Due to the DFT's output symmetry for real-valued time-
domain input samples, only 11 frequency-domain sample
(complex) values need be entered in the E-mail.
The 11 complex samples, X(0) through X(10), contain all non-redundant
spectral information. Sample X(10) corresponds to the fs/2 (Hz) spectral
sample and must be included in the E-mail.

Solution: 3.2
(a) The frequency-domain sample spacing of the DFT is:
fs
DFT sample spacing = ,
N
so
fs 1000
N= = = 22.2222.... samples
DFT sample spacing 45
which is not an integer. It is impossible to have a discrete
signal sequence having a non-integer number of samples.

(b) Tell your boss that 1000/45 is not an integer, so

either the fs sample rate, or the DFT sample
spacing, must be changed so that the ratio
fs
is an integer.
DFT sample spacing

Solution: 3.3
(a) Because the frequency-domain sample spacing of the DFT's X(m) is:

fs 44100
X (m) sample spacing = = = 1 Hz,
N N
the solution is N = 44100/1 = 44100 samples.

(b) The time duration of the x(n) sequence is the number of samples minus one,
N – 1, times the time period between samples (ts = 1/fs). We show that
notion in Figure S3–3, where a seven-sample w(n) sequence has a time
duration of six ts sample periods.

Property of Pearson Education. Not permissible for redistribution.

 (7–1)ts = 6ts

w(n)
0.5
3 4 5
0
1 2 6 n
–0.5
0
ts = 1/fs

Figure S3–3

So the solution to this part of the problem is

1 1
x(n) time duration = (N − 1) ⋅ = 44099 ⋅ = 0.999977 seconds .
fs 44100

Solution: 3.4
(a) The frequency spacing of the X(m) DFT samples is
fs 3000
X (m) sample spacing = = = 6 Hz.
N 500
(b) The highest-frequency spectral component that can be present in x(t) with no
aliasing errors in x(n) is half the fs sample rate, ... in this case that frequency
is
highest freq = fs/2 = 1500 Hz.

(c) The spacing between spectral replications is always equal to the fs sample
rate, ... in this case
the spectral replication spacing is 3000 Hz.

Solution: 3.5
(a) For this x1(n) = 9, 9, 9, 9, 9, 9, 9, 9 time-domain sequence, we use x1(n) and
go through the correlation steps used in the DFT Example 1 in Section 3.1.1.
In computing |X1(m)| in this way, the student should see that the |X1(m)|
samples are all zero when m is greater than zero. In computing |X1(0)|, the
student should use Eq. (3–13') to obtain:
7 7
|X1(0)| = ∑ x1(n) = ∑ 9 = 72.
n=0 n=0

the solution is:

|X1(0)| = 72, for freq. index m = 0 (DC term).
|X1(m)| = 0, for m = 1, 2, ..., 7.

(b) For theProperty

x2(n) = 1,of0,Pearson
0, 0, 0, 0,Education. Not permissible for redistribution.
0, 0 input sequence,
 the solution is:
|X2(m)| = 1, for all m.

(c) For the x3(n) = 0, 1, 0, 0, 0, 0, 0, 0 input sequence,

the solution is:

|X2(m)| = 1, for all m.

(d) The relationship of the |X2(m)| and |X3(m)| DFT samples is:

|X3(m)| = |X2(m)|.

Solution: 3.6
The positive-frequency spectral energy occurs at m = 3, so the cyclic
frequency of the x(n) sinusoid, with N = 8, is:
mfs 3(4000 Hz)
freq of x(n) in Hz, f = N = 8 = 1500 Hz.

The magnitude of X(3) is:

|X(3)| = 5.6572 + 5.6572 = 8.

Because x(n) is an integer number of cycles of a sinusoid, from the text's Eq.
(3–17) we find the sinusoid's peak amplitude be Ao = (2Mr)/N or:
2·|X(3)| 2·8
1500 Hz tone peak amplitude, Ao = N = 8 = 2.
The phase of X(3) is:
Imag. part of X(3) 5.657
X(3) phase, φ = tan–1[ Real part of X(3) ] = tan–1[ 5.657 ]

= tan–1(1) = π/4 radians (or 45o).

So the x(n) sinusoid is a cosine sequence with an initial phase of φ = π/4

radians. Given all this information, we write
x(n) = Aocos(2πfnts + φ) = Aocos(2πfn/fs + φ), or

x(n) = 2cos[2π(1500)n/4000 + π/4].

Simplifying the above x(n) expression, the problem solution is

x(n) = 2cos[3πn/4 + π/4].

Solution: 3.7 Property of Pearson Education. Not permissible for redistribution.

 (a) Given that an N-point DFT's input sequence x(n) is real-only and N is an
even number, using the following standard DFT equation,
N–1
X(m) = ∑ x(n)e–j2πnm/N ,
n=0
we can investigate what would be X(m) for various values of m other than
m = 0. When N is an even number, all the X(m>0) samples will be complex-
valued except when m = N/2. When m = N/2, from the above X(m)
expression we have
N–1 N–1
X(N/2) = ∑ x(n)e–j2πn(N/2)/N = ∑ x(n)e–jπn .
n=0 n=0
The factor e–jπn = (–1)n is a real-valued factor. (A sequence of alternating
plus and minus ones.) So the solution to our problem is:
DFT sample X(N/2) is always real-only
because
N–1
X(N/2) = ∑ x(n)(–1)n
n=0
is real-only when the x(n) input is real-only.

(b) The DFT equation is

N–1
X(m) = ∑ x(n)e–j2πnm/N .
n=0
Given that N is an odd number, we want to find a positive integer value for m
where e–j2πnm/N will be real-only. Stated in different words, we're looking for
integer values for m such that e–j2πnm/N will be equal to ±1, with odd N. The
problem solution is:

With odd N, there is no non-zero positive integer value

for m where 2m/N can be an integer, so the factor e–
j2πnm/N
will never be real-only, and the answer is No.

Solution: 3.8
(a) With m = N/2, the DFT equation becomes
N–1
X(N/2) = ∑x(n)[cos(2πn(N/2)/N) – jsin(2πn(N/2)/N)]
n=0

N–1 N–1
= ∑x(n)[cos(πn) – jsin(πn)] = ∑x(n)(–1)n.
n=0 n=0

Property of Pearson Education. Not permissible for redistribution.

 Next, examining the input sequence x(n) = sin[2π(fs/2)nts + θ], because
ts = 1/fs we can write
x(n) = sin[πn + θ].
Recalling the trigonometric identity: sin(α+β) = sin(α)cos(β) + cos(α)sin(β)
we express x(n) as
x(n) = sin(πn)cos(θ) + cos(πn)sin(θ)

= 0 + cos(πn)sin(θ) = (–1)nsin(θ).
Finally, applying our input x(n) expression to the above X(N/2) expression,
we have our solution of

N–1 N–1
X(N/2) = ∑(–1)nsin(θ)·(–1)n = ∑(1)2nsin(θ)
n=0 n=0

= N·sin(θ).

(b) When x(n) = sin[2π(fs/2)nts], θ = 0 and

X(N/2) = N·sin(0) = 0.

(c) When x(n) = cos[2π(fs/2)nts] = sin[2π(fs/2)nts + π/2], θ = π/2 and

X(N/2) = N·sin(π/2) = N.

Solution: 3.9
(a) The first challenge for the student is to correctly define the x(n) time-domain
expression for a complex sinusoid with magnitude Ao (i.e., x(n) = Aoej2πfnts),
having exactly three cycles over N samples. Substituting 1/fs for ts, and 3fs/N
for f, the x(n) expression becomes:

x(n) = Aoej2πfnts = Aoej2πfn/fs = Aoej2π3fsn/(Nfs) = Aoej6πn/N .

As such, the m = 3 bin output of the DFT of x(n) is:

N–1 N–1
X(3) = ∑ x(n)e–j6πn/N = ∑ Aoej6πn/Ne–j6πn/N
n=0 n=0
N–1 N–1
= Ao ∑ e0 = Ao ∑ 1 = AoN
n=0 n=0
so |X(3)| = AoN, which is what we set out to prove.

(b) The x(n) time-domain expression for a real-only sinusoid of peak amplitude
Ao with exactly three cycles over N samples is:
Property of Pearson Education. Not permissible for redistribution.
 x(n) = Aosin(2πfnts) = Aosin[2πfn/fs] = Aosin[2π3fsn/(Nfs)]

= Aosin(6πn/N).
Using Euler's identity, sin(α) = (ejα – e–jα)/2j, we write x(n) as:
Ao –jAo
x(n) = 2j (ej6πn/N –e–j6πn/N) = 2 (ej6πn/N –e–j6πn/N) .

So now the m = 3 bin output of the DFT of x(n) is:

N–1

∑
N–1
–jAo j6πn/N –j6πn/N –j6πn/N
X(3) = ∑ x(n)e–j6πn/N = 2 (e –e )e
n=0
n=0

–jAo N–1 j6πn/N –j6πn/N jAo N–1 –j6πn/N –j6πn/N

= 2 ∑ e e + 2 ∑e e .
n=0 n=0
From Part (a) we know the first summation in the above X(3) is equal to N,
so we write:

–jAoN jAo N–1 –j6πn/N –6πn/N

X(3) = 2 + 2 ∑ e e
n=0

–jAoN jAo N–1 –j12πn/N

= 2 + 2 ∑e .
n=0
Using the geometric series identity in Appendix B, the summation in the
above X(3) can be written in closed form as:
–jAoN jAo 1 –e–j12π –jAoN jAo 1–1
X(3) = 2 + ·
2 1 –e –j12π/N =
2 + ·
2 1 –e–j12π/N

–jAoN jAo 0 –jAoN

= 2 + 2 · 1 –e–j12π/N = 2 .

So the magnitude |X(3)| is:

|X(3)| = | –jA2 N | = A2N

o o

which is what we set out to prove.

Solution: 3.10
The first output sample, X(0), of an N-point DFT is equal to the sum of the
input time-domain samples. If X2(0) = X1(0) + 20, then the sum of x2(n) must
be 20 greater then the sum of x1(n). Therefore
Q = 41.

Property of Pearson Education. Not permissible for redistribution.

Solution: 3.11
From the definition of the DFT, X(m) is:

N–1 N–1 N–1

X(m) = ∑ x(n)e–j2πnm/N = ∑ ane–j2πnm/N = ∑ (ae–j2πm/N)n.
n=0 n=0 n=0
Because X(m) is a geometric series, from the identities in Appendix B we
can write:

(ae–j2πm/N)0 – (ae–j2πm/N)N 1 – aNe–j2πm

X(m) = = .
1 – ae–j2πm/N 1 – ae–j2πm/N

Because m is an integer, e–j2πm = cos(2πm) – jsin(2πm) = 1 – j0 = 1, we

arrive at our final solution of:
1 – aN
X(m) = .
1 – ae–j2πm/N

Solution: 3.12
Given a DFT's X(m) samples, we can compute the x(n) time samples using the
inverse DFT defined as:

1 N–1
x(n) = N ∑ X(m)ej2πnm/N , for n = 0, 1, 2, ..., N–1.
m=0

The first sample of x(n) is found by evaluating the above expression at n = 0, or:

1 N–1 1 N–1
x(0) = N ∑ X(m)ej2π(0)m/N
= N ∑ X(m)e0
m=0 m=0

1 N–1
= N ∑ X(m) .
m=0

Rearranging the above x(0) expression, we

prove the original statement to be true:

N–1
∑ X(m) = N·x(0).
m=0

Solution: 3.13
(a) The solution begins by conjugating both sides of the inverse DFT equation
as:
Property of Pearson Education. Not permissible for redistribution.
 1 ⎡ N–1 ⎤* 1 N–1
x*(n) = N ⎢ ∑X(m)ej2πnm/N ⎥ = N ∑ [X(m)ej2πnm/N]* .
⎢ ⎥
⎣m=0 ⎦ m=0

From Appendix A, we know that the conjugate of a product is equal to the

product of the conjugates. Using this fact, the above expression can be
written as:

1 N–1 1 N–1
x*(n) = N ∑X(m)*(ej2πnm/N)* = N ∑X(m)*e–j2πnm/N.
m=0 m=0

Next we recognize the right side of the above expression to be in the form of
a forward DFT.

Taking the conjugate of both sides of the above expression yields our
desired expression for x(n),
1⎡ N–1 ⎤
x(n) = N⎢ ∑X(m)*e–j2πnm/N ⎥*
⎢ ⎥
⎣m=0 ⎦
showing how to compute inverse DFTs using a forward DFT process.

(b) If the original frequency-domain X(m) sequence is conjugate

symmetric the inverse DFT's imaginary ximag(n) part will be all zeros.
In this situation the ximag(n) sequence need not be computed.

Solution: 3.14
Windowing the x(n) time-domain samples will reduce DFT leakage.
However, to minimize spectral leakage as much as possible we force the
analog signal's fo frequency to be one of the DFT's analysis frequencies,
such as in the scenario in the text's Figure 3–10(a). Thus,

the solution is to ensure that fo is given by

mfs
fo = N Hz

where m is an integer is in the range 1 < m < N/2–1 (to satisfy the
Nyquist sampling criterion.) The analog x(t) test signal becomes
x(t) = sin(2π[mfs/N]t).

Solution: 3.15
(a) In the X(m) DFT results the fundamental frequency spectral component is
located at m = 9. We find our Part (a) solution using
Property of Pearson Education. Not permissible for redistribution.
 mfs 9(22.255 kHz)
fundamental freq, at X(9), = N = 902 = 222.06 Hz.

(b) The frequency of the highest nonzero spectral component of the guitar's
"A3" note is located at m = 54. As was done above,
mfs 54(22.255 kHz)
highest freq, at X(54), = N = 902 = 1.332 kHz.

Solution: 3.16
(a) The original h1(n) comprises a 0 Hz (DC) component of amplitude 0.5 plus
a single-cycle sinusoid whose peak amplitude is 0.5. The DC component
accounts for the non-zero H1(0) sample in H1(m). The sinusoid component
of h1(n) is what produces the non-zero H1(1) sample in H1(m).

(b) The h2(n), in Figure S3–16(a), comprises a 0 Hz (DC) component and a

two-cycle sinusoid. The H2(m) sequence is shown in Figure S3–16(b). The
DC component of h2(n) component accounts for the non-zero H2(0) sample
in H2(m). The two-cycle component of h2(n) is what produces the non-zero
H2(2) sample in H2(m).

(c) Similarly, the h3(n), in Figure S3–16(c), comprises a 0 Hz (DC) component

and a three-cycle sinusoid. The |H3(m)| sequence is shown in Figure S3–
16(d). The DC component of h3(n) component accounts for the non-zero
H3(0) sample in H3(m). The three-cycle component of h3(n) is what
produces the non-zero H3(3) sample in H3(m).

(d) Considering the h1(n), h2(n), and h3(n) sequences, and their |H1(m)| ,
|H2(m)|, and |H3(m)| spectral magnitude samples, the statement that can be
made is:

"K repetitions of an h1(n) sequence results in an extended-

length time sequence whose spectral magnitudes have K–1
zero-valued samples inserted between each sample of the
original H1(m) sequence. In addition, the non-zero valued
samples of the extended-length spectral magnitudes are K times
the magnitude of the samples in the original H1(m) sequence.

Property of Pearson Education. Not permissible for redistribution.

 1
h (n)
0.8 2
0.6
(a) 0.4
0.2
0
0 5 10 15 n 20 25 31
20
|H2(m)| |H2(0)| = 16,
15
|H2(2)| = 8
(b) 10

5
0
0 1 2 3 m 4 5 15 16
1
h (n)
0.8 3
0.6
(c)
0.4
0.2
0
0 5 10 15 20 n 25 30 35 40 47
25
|H3(m)| |H3(0)| = 24,
20
|H3(3)| = 12
15
(d)
10
5
0
0 1 2 3 m 4 5 23 24

Figure S3–16

Solution: 3.17
We repeat the Problem's alternate Hanning window expression here as:
⎛ πn ⎞
whan,alt (n) = sin2⎜ N ⎟, for n = 0, 1, 2, . . ., N–1.
⎝ ⎠

1 –cos(2α)
Using the power relations trigonometry identity: sin2(α) = 2 ,

we can rewrite whan,alt (n) as

⎛2πn⎞
whan,alt (n) = 0.5 – 0.5cos⎜ N ⎟
⎝ ⎠
which is equal to the text's Eq. (3-29) Hanning window expression.

Solution: 3.18
An N-point DFT provides spectral samples whose cyclic frequency sample
spacingProperty
is fs/N. (fof
s isPearson
the sample rate in Hz of
Education. Notthepermissible
original x(n)for
sequence). A
redistribution.
 Q-point DFT provides spectral samples whose frequency sample spacing is
fs/Q. Because Q > N, the DFT spectrum of the zero-padded sequence has
finer spectral granularity (smaller spectral sample spacing).

Solution: 3.19
If the N-point DFT has a sample spacing of fs/N = 100 Hz, then the sample
spacing of a 5N-point DFT would be

Sample spacing = (fs/N)/5 = 100/5 = 20 Hz.

Solution: 3.20
The solution to this problem is found using the text's Eq. (3–33). The DFT
processing gain increase (in dB) achieved in using a one million-point DFT
compared to using 100-point DFT is:
⎛ 106 ⎞
SNRN=million = SNRN=100 + 20log10⎜ ⎟
⎝ 102 ⎠.

So the gain improvement in detecting a nonrandom spectral component is

⎛ 106 ⎞
Improved gain = 20log10⎜ ⎟
⎝ 102 ⎠ = 20log10(100) = 40 dB.

Solution: 3.21
(a) Because N = 16 for X2(m), the desired x2(n) is the 16-point time sequence
shown in Figure S3–21. The issues the student should consider in
determining x2(n) are:

• Because X2(m) is conjugate symmetric, x2(n) will be real-only samples.

• Because X2(m) exhibits no spectral leakage and only |X2(2)| and |X2(14)|
are non-zero, x2(n) will be exactly two sinusoidal cycles.
• Because the imaginary part of X2(m) are all zero-valued samples, x2(n) will
be a cosine sequence (as was x1(n)).
• Because N = 16 and |X2(2)| = 4, from the text's Eq. (3-17) we know that the
peak amplitude of the x2(n) cosine sequence will be Ao = (2Mr)/N =
(2·4)/16 = 0.5.

Solution:
0.5 x2(n)

0
x(t)
–0.5
0 2 4 6 n 8 10 12 14 15

Property of PearsonFigure
Education.
S3–21 Not permissible for redistribution.
 (b) The x2(n) sequence is a sampled version of the analog x(t) signal
with a sampling rate that is twice the sampling rate of x1(n).
Sequence x2(n) is an interpolated-by-two version of x1(n).

Solution: 3.22
The problem's rectangular expression for S, given as

S = P·[a + jb] – Q·[c + jd] – Q·[e + jg)],

can be written as

S = P·a(0) – Q·[c+ e] + j[P·b – Q·d – Q·g].

Now if the DFT's x(n) input sequence is real only, then X(0) = a + jb is real
only, and the DFT samples have conjugate symmetry. Thus we can state:

• b = 0,
• c = e, and
• d = –g.

So, the solution is that S can be written as

S = P·a – Q·[e+ e] + j[0 + Q·g – Q·g], or

S = P·a – 2Q·e = P·Real[X(0)] – 2Q·Real[X(1)].

where the notation "Real[X(0)]" means the real part of X(0).

Solution: 3.23
The DTFT of x(n) is a continuous function that cannot be plotted using a
digital computer. The N-point DFT of x(n) produces discrete frequency-
domain samples of the DTFT of x(n), which can be plotted.

Solution: 3.24
(a) The Ximp(ω) spectrum is computed using the DTFT equation as:
∞
Ximp(ω) = ∑ximp(n)e–jωn .
n=–∞
Because only ximp(0) is nonzero, we write:
0
Ximp(ω) = ∑ximp(n)e–jωn = ximp(0)e–jω0 = 1.
n=0
The problem solution is the constant-valued Ximp(ω) curve shown in Figure
S3–24.Property of Pearson Education. Not permissible for redistribution.
 Solution:

Ximp(ω)
0.5

0
0 π ω 2π
(fs/2) (fs)

Figure S3–24

(b) The answer to this question reflects on one of the most important principles
of signal analysis. That is, signals that are very narrow in the time domain
have spectra that are very wideband. (And vice versa. Bandlimited signals
that have their time durations expanded yield narrowed spectra.) So
lightening, which is narrow in time with high energy, has a very wideband
noise spectrum. Thus two AM radio receivers tuned to two very different
center frequencies both experience amplitude noise from lightening.

Solution: 3.25
The DTFT of x(n) is:
∞
X(ω) = ∑x(n)e–jωn .
n=–∞
Because only x(0) and x(1) are nonzero, we write:
1
X(ω) = ∑x(n)e–jωn = 1·e–jω(0) + 1·e–jω(1)
n=0

= 1 + e–jω = 1 + cos(ω) –jsin(ω).

The magnitude of X(ω) is:

|X(ω)| = [1 + cos(ω)]2 + [sin(ω)]2 = 1 + 2cos(ω) + cos2(ω) + sin2(ω) .

Because cos2(ω) + sin2(ω) = 1, we write |X(ω)| as:

|X(ω)| = 2 + 2cos(ω) .
To aid in drawing |X(ω)|, we evaluate |X(ω)| at five convenient frequency
points as:

at ω = ±π, |X(±π)| = 0; at ω = ±π/2, |X(±π/2)| = 2 ; and

atProperty of Pearson
ω = 0, |X(0)| = 2. Education. Not permissible for redistribution.
The problem solution is the |X(ω)| curve shown in Figure S3–25.

Solution:

2
1

0
–π –π/2 0 ω π/2 π

Figure S3–25

Property of Pearson Education. Not permissible for redistribution.

 Chapter 4 Solutions

Solution: 4.1
(a) There is no difference in the results of performing an N-point DFT
and an N-point FFT. The primary difference is that the FFT requires
far fewer multiplication and addition operations than does a DFT.

(b) The radix-2 FFT restriction is that the FFT size N must be integer
power of two.

Solution: 4.2
Based on the property that the FFT's X(m) frequency-domain sample
spacing is:
fs
X (m) sample spacing = = 1 Hz,
N
then the minimum number of time samples is:

fs 44100
N= = = 44100.
1 Hz 1

Because N = 44100 is not an integer power of two, to use the FFT

algorithm, we must make N be the next larger integer power of two greater
than 44100. So the solution to this problem is

N = 65536 samples.

(When N = 65536, the X(m) sample spacing will be 0.673 Hz.)

Solution: 4.3
(a) The next highest integer power of two greater than 3800 is 4096. Thus the
number of zero-valued samples that must be appended to the original time-
domain sequence to extend its length to 4096 is

4096–3800 = 296 zero-valued samples.

(b) The FFT sample spacing is the discrete signal's fs sample rate (in Hz)
divided by the length of the FFT. That is:
fs
FFT bin spacing = N .

The fs sample rate is the discrete signal sequence's length (in samples)
dividedProperty
by the signal's time duration,
of Pearson or Not permissible for redistribution.
Education.
 3800 samples
fs = two seconds = 1900 samples/second (1900 Hz).

Because the FFT's length is N = 4096, the solution is:

fs 1900
FFT bin spacing = N = 4096 = 0.464 Hz.

(c) In the case of lowpass sampling, the highest-frequency spectral component

permitted in the original analog signal such that no aliasing errors occur is
half the fs sampling rate (Nyquist criterion), or
highest-frequency = 1900/2 = 950 Hz.

Solution: 4.4
There are N2 complex multiplies necessary to compute an N-point DFT.
Thus a 32768-point DFT requires

MDFT = 327682 ≈ 1.074x109 complex multiplies.

There are (N/2)log2(N) butterfly operations in an N-point radix-2 FFT. Using

the optimized butterfly in the text's Figure 4–14(c), requiring one complex
multiply per butterfly, means that (N/2)log2(N) complex multiplies are
necessary to compute an N-point FFT. Thus a 32768-point FFT requires

N 32768
MFFT = 2 log2(N) = 2 log2(32768)

= (16384)(15) ≈ 2.46x105 complex multiplies.

The solution to this problem, the ratio of MDFT over MFFT, is:

MDFT 1.074x109
Ratio = M = 2.46x105 ≈ 4.37x103.
FFT

Solution: 4.5
(a) To satisfy the Nyquist criterion, fs must be greater than
2·50 = 100 Hz.

(b) If the average value of x(n) is 17, we use the property described in Section
3.4 stating the FFT's X(0) sample, which is always real, is the product of the
average value ("DC bias") of the x(n) sequence and the FFT size N. That is

X(0) = 17N = 17·2048 = 34,816.

Solution 4.6
Property of Pearson Education. Not permissible for redistribution.
 If the spectrum analyzer that can perform a 1024-point FFT in 50
microseconds, its maximum throughput rate would occur if a memory bank
finishes loading 1024 x(n) samples every 50 microseconds. In that situation
the ts time period between x(n) samples from the analog-to-digital converter
is
50x10–6 –8
ts = 1024 = 4.88x10 seconds/sample.
Because the fs sample rate is 1/ts, we know that
1 1
fs = t = 4.88 x 10–8 = 204.8x105 samples/second = 20.48 MHz.
s

In order to satisfy the Nyquist sampling criterion the analog x(t) signal can
have a maximum one-sided bandwidth of half the sampling rate. So our
maximum x(t) bandwidth solution is
fs 20.48 MHz
Bmax = 2 = 2 = 10.24 MHz.

Solution 4.7
When the measured frequency of a spectral component changes by
an amount equal to a change in sample rate, that spectral component
in x(t) has a center frequency greater than both fs/2 (20 kHz) and f's/2
(19 kHz), and aliasing is taking place.

We illustrate this situation in Figure S4–7(a) where the solid curves are the
spectrum of the analog x(t) signal and the dashed curves are spectral
replications in X1(m). The x(t) analog signal has a ±15 kHz spectral noise
component (greater than fs/2) whose negative-frequency spectral replication
resides at +5 kHz.
When the sample rate is reduced to f's = 19 kHz, the spectral
replication of x(t)'s negative 15 kHz spectral noise component now aliases to
4 kHz as presented in Figure S4–7(b).

Property of Pearson Education. Not permissible for redistribution.

 Spectral replication
aliased to 5 kHz
fs = 20 kHz
|X1(m)|

(a)
–20 –16 –12 –8 0 4 8 12 16 20 kHz
(fs)
–5 5 (fs/2)

Spectral replication
aliased to 4 kHz
|X2(m)| f's = 19 kHz

(b)
–20 –16 –12 –8 –4 0 4 8 12 16 20 kHz
–19 (f's/2) 19
(f's)

Figure S4–7

Solution: 4.8
(a) α = e–j2π(3)/16 = e–j6π/16 = e–j3π/8.

(b) α = 0.3827 – j9239.

Solution: 4.9
(a) All eight x(n) input samples affect the value of the X(2) output sample. That
is, there is a signal flow path from all x(n) inputs to the X(2) output.

(b) Likewise, all eight x(n) input samples affect the value of the X(5) output
sample.

Solution: 4.10
The signal flow solution using optimized decimation-in-time butterflies is
provided in Figure S4–10(a). Because W40 = 1, the 4-point FFT can also be
drawn as shown in Figure S4–10(b).

Property of Pearson Education. Not permissible for redistribution.

 Solution:
Correct
indices

x(0) X(0)

x(1) W40 –1 X(2)

(a)
x(2) W40 –1 X(1)

x(3) W40 –1 W41 –1 X(3)

x(0) X(0)

x(1) –1 X(2)
(b)
x(2) –1 X(1)

x(3) –1 W41 –1 X(3)

Figure S4–10

Solution: 4.11
By inspection, in a standard N-point decimation-in-time (DIT) FFT, the
number of unique twiddle factors in the qth stage is 2q. So

R = 2q.

The interval of the twiddle factors' exponents for the qth stage decrease by a
factor of two relative to the interval of the twiddle factors' exponents for the
(q–1)th stage. So P must increase as q increases. By inspection we see that

P = 2q,

allowing us to express the values of the 2q unique twiddle factors in the qth
stage of a general N-point DIT FFT as:
q
kth twiddle factor of qth stage: = WNkN / 2 , for k = 0,1,2,...,2q − 1.

Solution: 4.12
(a) To show the signal flow diagram of a standard 8-point decimation-in-time
FFT with bit-reversed outputs with the butterfly full twiddle factors shown
in rectangular notation, first we show the twiddle factors in complex
exponential notation as in Figure S4–12–I.
Property of Pearson Education. Not permissible for redistribution.
 stage 1 stage 2 stage 3
x(0) X(0)
–j2π0/8 e–j2π0/8
e–j2π0/8 e

x(1) e–j2π4/8 X(4)

e–j2π0/8
e–j2π0/8

x(2) e–j2π4/8 X(2)

e–j2π2/8
e–j2π0/8

x(3) e–j2π4/8 e–j2π6/8 X(6)

e–j2π0/8

x(4) X(1)
e–j2π1/8
e–j2π2/8

x(5) e–j2π5/8 X(5)

e–j2π2/8

x(6) e–j2π4/8 e–j2π6/8 X(3)

e–j2π3/8

x(7) e–j2π4/8 e–j2π6/8 e–j2π7/8 X(7)

Figure S4–12–I

Next we convert the complex exponential twiddle factors to rectangular

notation resulting in our desired FFT signal flow diagram shown in Figure
S4–12–II.

Property of Pearson Education. Not permissible for redistribution.

 Solution:
stage 1 stage 2 stage 3
x(0) X(0)
1 1
1
x(1) –1 X(4)
1
1
x(2) –1 X(2)
–j
1
x(3) –1 +j X(6)
1

x(4) –1 X(1)
–j 0.707–j0.707

x(5) –1 –0.707+j0.707 X(5)

–j

x(6) –1 +j X(3)
–0.707–j0.707

x(7) –1 +j 0.707+j0.707 X(7)

Figure S4–12–II

(b) What the student should notice about the twiddle factors in the FFT's 1st and
2nd stages in Figure S4–12–II is that those twiddle factors require no
multiplications. The computations within those two stages are only sign
changes or real/imaginary-part data assignments.

Solution: 4.13
We proceed through this grueling solution using the notation that
Wq = e–j2πq/8.
The samples at nodes A through F inside the FFT structure, in terms of the
x(n) input samples, are
A = x(0) + W0x(4) = x(0) + x(4),
B = x(2) + W0x(6) = x(2) + x(6),
C = x(1) + W0x(5) = x(1) + x(5),
D = x(3) + W0x(7) = x(3) + x(7),
E = A + W4B = x(0) + x(4) + W4x(2) + W4x(6),
F = C + W4D = x(1) + x(5) + W4x(3) + W4x(7).

Property of Pearson Education. Not permissible for redistribution.

 The FFT's X(2) output is
XFFT(2) = E + W2F
= x(0) + x(4) + W4x(2) + W4x(6) + W2x(1) + W2x(5) + W6x(3) + W6x(7).
XFFT(2) = E + W2F
= x(0) + x(4) + W4x(2) + W4x(6)
+ W2x(1) + W2x(5) + W6x(3) + W6x(7).

Next we determine the DFT's X(2) output using

7 7
XDFT(2) = ∑ x(n)e–j2π2n/8 = ∑ x(n)W2n
n=0 n=0
= x(0)W + x(1)W + x(2)W4 + x(3)W6
0 2

+ x(4)W8 + x(5)W10 + x(6)W12 + x(7)W14.

Because W8 = W0 = 1, W10 = W2, W12 = W4, and W14 = W6,

we can rewrite the DFT's XDFT(2) output as

XDFT(2) = x(0) + x(1)W2 + x(2)W4 + x(3)W6 + x(4) + x(5)W2 + x(6)W4 + x(7)W6

showing that XDFT(2) = XFFT(2) which is what we set out to prove. (Whew!)

Solution: 4.14
For a 16-point decimation-in-time FFT, the X(m) output index order is the
bit reversal of the x(n) input index order. With in-order inputs, the output
samples will be ordered as shown at the FFT output in Figure S4–14.

Solution:
Correct indices

x(0) X(0)
x(1) X(8)
x(2) X(4)
x(3) X(12)
x(4) X(2)
x(5) 16-point X(10)
x(6) FFT X(6)
x(7) X(14)
x(8) X(1)
x(9) X(9)
x(10) X(5)
x(11) X(13)
x(12) X(3)
x(13) X(11)
x(14) X(7)
x(15) X(15)

Property of Pearson Education.

Figure S4–14 Not permissible for redistribution.
Solution: 4.15
The answer is yes. If a standard butterfly's upward diagonal path contains a
twiddle factor, such as in Figure S4–15(a), then it is always a decimation-in-
time butterfly. Likewise, if a butterfly's downward diagonal path contains a
twiddle factor, such as in Figure S4–15(b), then it is always a decimation-in-
frequency butterfly.

Decimation in time Decimation in frequency

A C A C
e –j2πk/N

e –j2πk/N
–j2πm/N
B e D B e –j2πm/N D
(a) (b)

Figure S4–15

Solution: 4.16
There are many ways to draw a real-valued block diagram of a decimation-
in-time butterfly. One way is shown in Figure S4–16–I. All correct solutions
to this problem must perform the necessary computations shown on the right
side of Figure S4–16–I.

Solution:
Necessary computations
Decimation-in-time butterfly

AR CR = AR + BRcos(θ) + BIsin(θ)

AI CI = AI + BIcos(θ) –BRsin(θ)

DR = AR –BRcos(θ) –BIsin(θ)
–
DI = AI + BRsin(θ) –BIcos(θ)
–
BR
–

BI

cos(θ) –sin(θ)

Property of Pearson Education.

Figure S4–16–I Not permissible for redistribution.
 An alternate diagram of the butterfly is shown in Figure S4–16–II. The
necessary computations are shown in that figure.

Solution:
Necessary computations
Decimation-in-time butterfly
AR CR = AR + BRcos(θ) + BIsin(θ)

BR DR = AR –BRcos(θ) –BIsin(θ)
–
cos(θ)

sin(θ)

–sin(θ)

AI CI = AI + BIcos(θ) –BRsin(θ)

BI DI = AI + BRsin(θ) –BIcos(θ)
–
cos(θ)

Figure S4–16-II

Solution: 4.17
(a) We determine the maximum possible decimal value of the real part of the
complex output sample C as follows: The expression for the C output of the
FFT butterfly is
C = Areal + jAimag + e–j2πk/N(Breal + jBimag)
= Areal + jAimag + [cos(2πk/N) –jsin(2πk/N)](Breal + jBimag)
= Areal + cos(2πk/N)Breal + sin(2πk/N)Bimag
+ j[Aimag + cos(2πk/N)Bimag –sin(2πk/N)Breal] .

So the real part of the C output sample, Creal , comprises three terms
Creal = Areal + cos(2πk/N)Breal + sin(2πk/N)Bimag .
Here's the solution to this problem:
If Areal, Breal, and Bimag are each equal to 127 when they enter the butterfly,
and if the twiddle factor angle 2πk/N happens to be π/4 = 45o, then output
Creal can have a decimal value as large as
Creal = 127 + cos(π/4)127 + sin(π/4)127 = 306.61
Property of Pearson Education. Not permissible for redistribution.
 which is more than twice the maximum possible decimal value (127) for the
real or imaginary parts of input samples A and B.

(b) Because the maximum possible decimal value for the real part of output
sample C is greater than 255 (which would require 9 binary bits), we require
10 binary bits to store the real part of output sample C.

The real part of C is equally likely to be more negative than a decimal –255
after a single FFT butterfly if A and B are large negative numbers.) We
remind the student that without accounting for, or mitigating, this potential
binary-word growth in fixed-point FFT systems, overflow errors could
render FFT results useless.

Solution: 4.18
The number of butterfly operations in a single 16384-point FFT is:

N 16384
BFFT = 2 log2(N) = 2 log2(16384)

= (8192)(14) = 114688 butterflies/FFT.

The number of butterfly operations performed per second is:

Bsec = BFFT·(FFTs/second) = 114688 ·1.744x105

≈ 2x1010 butterflies/second.

The optimized decimation-in-frequency FFT butterfly structure, shown in

Figure S4–18 requires four real multiplications per butterfly operation. So
the number of real multiplications per second, performed by the spectrum
analyzer is:

Multiplications/second = Bsec·4 real multiplies/butterfly = 4·2x1010

≈ 8x1010 (80 billion) real multiplies/second.

A C

Four real multiplies

–1 A–B
B D
–j2πk/N
e

Figure S4–18

Property of Pearson Education. Not permissible for redistribution.

Property of Pearson Education. Not permissible for redistribution.
 Chapter 5 Solutions

Solution: 5.1
(a) The h(k) impulse response of the system is:

h(k)
1 ...
0
0 1 2 3 4 5 k

Figure S5–1

(b) The block diagram of the system is:

x(n) y(n)
x(n) y(n)

or
1

Figure S5–1 (Cont'd)

(c) The impulse response of such a trivial system is

h(k) = 1, 0, 0, 0, 0, ...

The student should recall that we can estimate the frequency response of the
system by taking the discrete Fourier transform (DFT) of h(k), the solution
of which has a magnitude of

|H(m)| = 1

for all m where m = 0, 1, 2, ..., N–1.

For this h(k) impulse response, there are two methods the student might use to
determine the frequency response, i.e., the DFT magnitude samples.

Solution Method# 1: The student could use, say, a standard N = 4-point

DFT as:
N–1 3
H(m) = ∑ h(k)e –j2πmk/N
= ∑ h(k)e–j2πmk/4.
k=0 k=0

= h(0)e–j2πm0/4 + h(1)e–j2πm1/4 + h(2)e–j2πm2/4 + h(3)e–j2πm3/4.

Because h(1) = h(1) = h(1) = 0,

Property of Pearson Education. Not permissible for redistribution.
 |H(m)| = |h(0)e–j2πm0/4| = h(0) = 1.

Solution Method# 2: Recalling the text's Eq. (3–43) for the DFT of a
general rectangular time sequence, the magnitude of that expression is:
sin(πmK/N)
|H(m)| = .
sin(πm/N)

With K = 1, the above |H(m)| becomes:

sin(πm/N)
|H(m)| = = 1, for all m ≠ 0.
sin(πm/N)

To find |H(m=0)|, we apply L'Hopital's rule to the first |H(m)| expression

above and set m = 0. Doing so we find: |H(0)| = 1. Thus |H(m)| = 1, for all
m.

Solution: 5.2
(a) Based on the expression x(t) = cos(2π800t), the frequency of the analog
sinusoidal tone is 800 Hz and its spectrum is shown in Figure S5–2(a).

The spectral magnitude of x(n) is shown in Figure S5–2(b). With a

sample rate fs = 1000 Hz, aliasing has occurred in the spectrum of x(n).

(b) The spectral magnitude of y(n) is shown in Figure S5–2(c), where we show
replications of the filter's frequency magnitude response.

(c) Because x(n)'s ±200 Hz sinusoid is within the filter's unity-gain passband,

the y(n) output sinusoid has a peak amplitude of

one (no attenuation).

Property of Pearson Education. Not permissible for redistribution.

 Spectral magnitude of x(t)

(a)
–0.8 0 0.8 kHz

Spectral replications
Spectrum of x(n)

... ... Solution to

(b) Part (a)
–2 –1 –f /2 0 fs/2 1 2 kHz
s
(–2fs) (–fs) (fs) (2fs)
–1.8 –0.2 0.2 0.8 1.8
–1.2 1.2

Filter magnitude response

Spectrum of y(n)

... ... Solution to

(c) Part (b)
–2 –1 –fs/2 0 fs/2 1 2 kHz
(–2fs) (–fs) (fs) (2fs)
–0.2 0.2

Figure S5–2

Solution: 5.3
The time delay through the linear-phase FIR filter is equal to the filter's
group delay. From the text's description of group delay (measured in
seconds), D/(2fs) where D is the number of filter delay elements, we can
write
Max number of delay elements Dmax
6x10–3 > 2fs = 2f ,
s

or
Dmax < 6x10–3 · 2fs = 6x10–3 · 96x103 = 576.
So the number of filter delay elements must be less than 576 which will be
satisfied so long as:

the maximum number of filter taps (coefficients) is 576.

Solution: 5.4
The problem solution is the network shown in Figure S5–4, where the 'D'
symbol indicates a single delay element.
If input x(n) is a unit impulse sequence, the sequence at node L will be the
hLow(n) impulse response, and the sequence at node H will be the required
hHigh(n) impulse response.
Property of Pearson Education. Not permissible for redistribution.
 Solution:
Woofer
L D/A
hLow(k)
converter
x(n)
– H D/A
D D D D D D D
+ converter
Tweeter
D = Delay

Figure S5–4

Solution: 5.5
The answer is simple. First we design the desired lowpass filter
to obtain filter coefficients h(k). Then we merely multiply the
h(k) coefficients by 2, yielding the FIR filter a gain of 2.

Solution: 5.6
The answer to this problem is "Nothing."

The lowpass filter's h(k) coefficients need not be changed in any way. With
the actual x(n) sample rate being 40 kHz, the X(f) spectral plot and the
filter's H(f) magnitude response plot remain unchanged with the exception
that their frequency axis values (Hz) are increased by a factor of two.

Solution: 5.7
(a) The characteristic of the filter's frequency response causing the filter's
output sequence to go to all zeros is:

the filter has a magnitude null (infinite attenuation)

at the frequency of the input sinusoid.

Hopefully the student realizes that fact, and validates that notion by noticing
the filter input sequence has 8 samples/cycle making its frequency fs/8. An
8-point moving average FIR filter has a frequency magnitude response null
at fs/8.

(b) Those initial nonzero-valued filter output samples are

called the "transient response" of the filter.

Solution: 5.8
Abrupt changes in the amplitude of a signal are associated
with high frequency spectral content.
Property of Pearson Education. Not permissible for redistribution.
 Likewise, a time-domain signal represented by only small amplitude
changes, as time advances, is a low frequency signal.

Solution: 5.9
(a) Because FIR digital filter performs time-domain convolution, we apply the
text's Eq. (5–29). With a filter impulse response h(k) being of length P = 3
and the filter's input sequence being of length Q = 3,

the filter's nonzero output sample sequence length is

P + Q –1 = 3 + 3 –1 = 5 samples.

(b) The nonzero filter output sequence samples are [0.5, 1.5, 2.0, 1.5, 0.5], so

the maximum filter output sample value is 2.0.

Solution: 5.10
To compute the filter's y(n) output sequence, we convolve the x(n) input
sequence with the h(k) impulse response. Using the text's Eq. (5–6) with a
filter impulse response length of M = 3, our convolution is described by
2
y(n) = ∑ h(k)x(n–k).
k=0

We flip the time order of the filter's x(n) input sequence, pass the flipped
sequence over the h(k) sequence, and perform the sums of products as
shown in Figure S5–10(a).

The desired y(n) output sequence is given in Figure S5–10(b)

where y(n) = [0, 0, 0.375, 1.25, 2.125, 1.75, 0.75].

Alignment to compute first output, y(0).

h(k) y(n)
1.0 2.25
0.75 2.0
0.5 1.75
0.25 1.5
0 1.25
x(n) flipped 0 1 2 k
1.0 1.0
Slide x(n) one
0.75 sample to the right
0.75
0.5 for each new y(n) 0.5
computation.
0.25 0.25
0 0
4 3 2 1 0 n 0 1 2 3 4 5 6 n
(a) (b)

Figure S5–10
Property of Pearson Education. Not permissible for redistribution.
Solution: 5.11
The important topic we learned in Chapter 5 that verifies the spectral
replications of discrete sequences is the Convolution Theorem. The
spectrum of a discrete version of the product y(t) = s(t)x(t) is the
convolution of the individual X(f) and S(f) spectra.

This principle is illustrated in Figure S5–11.

X(f)
S(f)

–2fs –fs 0 fs 2fs Freq 0 Freq

Convolution

Y(f)
... ...

–2fs –fs 0 fs 2fs Freq

Figure S5–11

Solution: 5.12
(a) Recalling the convolution theorem, which states that multiplication in the
time domain is equivalent to convolution in the frequency domain, the Y(m)
spectra is the convolution of X(m) with itself. The convolution of a
rectangular function with itself is a triangular function as shown in Figure
S5–12(b). (The "*" symbol means convolution.)

X(m)
Given:
(a)
–0.4fs 0 0.4fs Freq

Solution:
Y(m) Spectral
Y(m) = X(m) * X(m)
replication

(b)
–fs 0 fs Freq
–0.8fs 0.8fs

Property of Pearson Education. Not permissible for redistribution.

Figure S5–12
 (b) Yes, y(n) will experience aliasing errors because
Y(m) has overlapped spectral replications.

(c) The maximum one-sided bandwidth of x(n) to avoid aliasing [overlapped

spectral replications in Y(m)] is fs/4 as shown in Figure S5–12(c).

X(m)
Given:

–fs/4 0 fs/4 Freq

(c)

Y(m) = X(m) *X(m) Y(m) No aliasing

Spectral
replication

–fs –fs/2 0 fs/2 fs Freq

Figure S5–12 (Cont'd)

Solution: 5.13
(a) Based on the interpretation in Figure S5–13(a), the difference equation for
linear interpolation is:
x(n) – x(n–1) x(n) x(n–1)
y(n) = x(n–1) + 2 = 2 + 2 .

y(n)
x(1)

x(0) x(1) – x(0)

(a) y(1) = x(0) + [x(1) –x(0)]/2

x(0)
0
1 n

Figure S5–13

(b) The various (equivalent) block diagrams for a linear interpolation system
are shown in Figure S5–13(b).

Property of Pearson Education. Not permissible for redistribution.

 Solutions:
x(n) x(n)
Delay Delay

0.5 0.5

y(n)

(b) y(n)
0.5

x(n) x(n) y(n)

Delay

0.5 Delay
0.5

y(n)

0.5

Figure S5–13 (Cont'd)

Solution: 5.14
The DC gain of filter h1 is equal to the sum of filters' coefficients. That is:
6
Filter h1 DC gain = H1(0) = ∑h1(k) = 99.6.
k=0
Likewise, the DC gain of filter h2 is equal to:
6
Filter h2 DC gain = H2(0) = ∑h2(k) = 103.6.
k=0
So H2(0) and H1(0) differ by a value of four. Each of filter h1's two 25.5-
valued coefficients were changed to Q. Thus 2Q = 2·25.5 + 4, giving us our
solution of:

Q = 25.5 + 2 = 27.5.

Solution: 5.15
The filter's difference equation is:

y(n) = h(0)·x(n) + h(1)·x(n–1) + h(2)·x(n–2) + h(3)·x(n–3) + h(4)·x(n–4).

Due to phase linearity, h(0) = h(4) and h(1) = h(3). So we may rewrite the
difference equations as:

y(n) = h(0)·x(n) + h(1)·x(n–1) + h(2)·x(n–2) + h(1)·x(n–3) + h(0)·x(n–4)

= h(0)·[x(n) + x(n–4)] + h(1)·[x(n–1) + x(n–3)] + h(2)·x(n–2)

Property of Pearson Education. Not permissible for redistribution.

 which requires only three multiplies per y(n) output sample. The problem's
solution is the FIR filter block diagram (structure) shown in Figure S5–15.

Solution:
x(n–4) x(n–3)
z–1 z–1

x(n) x(n–1)
z–1 z –1

x(n) + x(n–4) x(n–1) + x(n–3) x(n–2)

h(0) = h(4) h(1) = h(3) h(2)

y(n)

Figure S5–15

Solution: 5.16
To derive an equation for the H1(ω) and H2(ω) frequency responses, we
write the discrete-time Fourier transform (DTFT) of the filters' coefficients
as:
∞ 2
H1(ω) = ∑h1(k)e-jkω = ∑h1(k)e-jkω = e–j0ω + 2e–jω + e–j2ω
k=–∞ k=0

= 1 + 2e–jω + e–j2ω.
and
∞ 1
H2(ω) = ∑h2(k)e -jkω
= ∑h2(k)e-jkω = e–j0ω + e–jω = 1 + e–jω.
k=–∞ k=0

Next we set the H1(ω) and H2(ω) expressions equal to each other to
determine frequency ωo as:
H1(ωo) = H2(ωo) → 1 + 2e–jωo + e–j2ωo = 1 + e–jωo.
With algebraic simplification the above equality can be written as:
e–jωo = –e–j2ωo.
or
1 = –e–jωo.
or
e–jωo = –1.
Because e–j±π = –1, the desired ωo frequency is:
ωo = ±π radians/sample
Property
which is theofsolution
Pearson Education.
to our problem. Not permissible for redistribution.
 In terms of cyclic frequency, the ωo = ±π frequency is equivalent to ±fs/2 Hz
(half the sample rate). The student could have explicitly solved the 1 = –e–jωo
equation for the desired ωo frequency using natural logarithms as:
ln(1) = ln(–1)·(–jωo)
0 = ±jπ –jωo
ωo = ±π.
This problem deserves additional discussion. The frequency magnitude
responses of the two FIR filters is shown in Figure S5–16(a). Their phase
responses are provided in Figure S5–16(b). At a frequency of ω = 2.095
radians, the two filters have identical frequency magnitude responses.
However, at that frequency their phase responses differ, so their complex
frequency responses are not equal at ω = 2.095 radians.

4
|H1(ω)|
(a) 2
|H2(ω)|
0
–π –π/2 0 π/2 π
ω
ω = 2.095

180 H1(ω) phase

Degrees

90
(b) 0 H2(ω) phase
–90
–180
–π –π/2 0 π/2 π
ω
ω = 2.095
|H2(ω)|
0
–20
dB

(c) |H1(ω)|
–40
–60
–π –π/2 0 π/2 π
ω

Figure S5–16

Solution: 5.17
(a) The filter's center coefficient is:
h1 = 1.6617.

The derivation of h1 is as follows: We write the discrete-time Fourier

transform (DTFT) of the filter's impulse response, which is equal to the
filter's h(k) = [1, h1, 1] coefficients, as:
Property of Pearson Education. Not permissible for redistribution.
 ∞ 2
H(ω) = ∑h(k)e-jkω = ∑h(k)e-jkω
k=–∞ k=0

= e–j0ω + h1e–jω + e–j2ω = 1 + h1e–jω + e–j2ω

where the normalized frequency variable ω is in the range –π ≤ ω ≤ π. At
the filter's magnitude null frequency ωn the frequency response is zero, so
we can write:
H(ω) = 1 + h1e–jωn + e–j2ωn = 0.
Solving the above equation for h1 yields our desired expression for h1 in
terms of ωn. Doing that, we have
–e–j2ωn –1
h1 = = –e–jωn –ejωn.
e–jωn
Remembering Euler's equation of cos(α) = (ejα + e–jα)/2, we can simplify the
h1 expression as:
h1 = –2cos(ωn)
where the normalized null frequency ωn is in the range 0 ≤ ωn ≤ π.

Because the normalized frequency of ω = 2π radians corresponds to a radian

frequency of 2πfs radians/second (where fs is the signal data sample rate in
Hz), the h1 coefficient can also be defined using
2π
h1 = –2cos[(2πfn) ] = –2cos(2πfn/fs)
2πfs
where cyclic null frequency fn is in the range 0 ≤ fn ≤ fs/2 Hz. So, given a
null frequency of 3.35x106 Hz, the desired h1 filter coefficient is
h1 = –2cos[2π(3.35x106/8.25x106)] = –2cos(2.5514) = 1.6617.

(b) The DC gain of the filter is the sum of its coefficients, or:

DC gain = 1 + 1.6617 + 1 = 3.6617.

Using h1 = 1.6617 yields a filter magnitude response shown by the solid

curve in Figure S5–17.

0
|H(ω)| Noise
–20
dB

–40

–5 –4 –3 –2 –1 0 1 2 3 4 5
Freq (MHz)
3.35 MHz

Property of PearsonFigure
Education.
S5–17 Not permissible for redistribution.
 (c) The 3-tap FIR filter has linear phase because its
coefficients are symmetrical, h(k) = [1, h1, 1].

Solution: 5.18
As described in the text, to achieve a linear phase response the
h(k) coefficients of an N-tap tapped-delay line FIR filter must
be either symmetrical,
h(k) = h(N–k–1)
or anti-symmetrical,
h(k) = –h(N–k–1)
where 0≤k≤(N–1)/2 for odd N, and 0≤k≤(N/2)–1 for even N.

Solution: 5.19
A tapped-delay line FIR filter having N = 512 taps requires N – 1 = 511
addition operations to compute each filter output sample.

Solution: 5.20
From Section 5.10 we learned that the group delay, measured in samples, of
a tapped-delay line filter network having symmetrical (or antisymmetrical)
coefficients is found using
D
group delay = G = 2 samples,

where D is the number of unit-delay elements in the filter's delay line. With
the number of FIR filter delay elements, D, being one less than its number
of taps, the group delay of the filter is
255 – 1
group delay = 2 = 127 samples.

Solution: 5.21
Because the input signal's frequency is within the lowpass filter's passband,
the filter's description of "linear-phase" is the key here. That means we have
symmetrical coefficients and the time delay, equal to the filter's group delay,
through the 70-tap filter is independent of frequency. With the number of
FIR filter delay elements, D, being one less than its number of taps, from
Section 5.10 the group delay (measured in seconds) of the filter is
D 70 – 1 69
group delay = G = 2f = 2f = 4x106 = 17.25 microseconds.
s s

Property of Pearson Education. Not permissible for redistribution.

Solution: 5.22
(a) From Section 5.10 we learned that the group delay, measured in samples, of
a tapped-delay line filter network having symmetrical (or antisymmetrical)
coefficients is found using
D
group delay = G = 2 samples,

where D is the number of unit-delay elements in the filter's delay line. With
the number of FIR filter delay elements, D, being one less than its number
of taps, the group delay of the first filter is

17 – 1
group delay G1 = 2 = 8 samples.

(b) The group delay of the 17-tap H2(f) half-band FIR filter depends on how it
is implemented. Recall, the first and last coefficients of a 17-tap half-band
filter are zero-valued. If the filter is implemented with a tapped-delay line
structure with 17 multipliers, then its group delay will be

17 – 1
group delay G2,17 taps = 2 = 8 samples.

However, that filter can be implemented with only 15 taps by discarding the
first and last zero-valued coefficients with no change in its frequency
response. In that case the group delay of the 15-tap H2(f) half-band FIR filter
will be

15 – 1
group delay G2,15 taps = 2 = 7 samples.

Solution: 5.23
(a) The reverberator's h(n) impulse response is shown in Figure S5–23(a).

Solution:
1
h(n)

(a) 0

–1
0 5 10 15 n 20 25 30
8

Figure S5–23

(b) We find the digital reverberator's |H(ω)| by performing the discrete-time

Fourier transform (DTFT) of the reverberator's impulse response as:
∞
H(ω) = ∑h(n)e–jωn = 8 [1,0,0,0,0,0,0,0,1]e–jωn
Property of Pearson
n=0 Education. Not permissible for redistribution.
∑

n=–∞
 = e–j0ω + e–j8ω = cos(0)–jsin(0) + cos(8ω)–jsin(8ω) = 1 + cos(8ω) –jsin(8ω).
So |H(ω)| is

|H(ω)| = |1 + cos(8ω) –jsin(8ω)|.

The expression for |H(ω)| can be simplified as:

|H(ω)| = [1 + cos(8ω)]2 + sin2(8ω)

= 1 + 2cos(8ω) + cos2(8ω) + sin2(8ω) .

Using the identity cos2(α) + sin2(α) = 1, we can write
4 + 4cos(8ω) 1 + cos(8ω)
|H(ω)| = 2 + 2cos(8ω) = 2 =2 2 .

Recalling the half-angle trigonometric identity

{[1 + cos(α)]/2}1/2 = ±cos(α/2), and that our |H(ω)| must be positive, we
have an equivalent solution of

|H(ω)| = |2cos(4ω)|.

(c) The plot of |H(ω)| = |2cos(4ω)| is given in Figure S5–23(b). A simple delay
line implementation of a digital reverberator does not have a flat frequency
response, and should not be used in practice.

Solution:
2 |H(ω)|

1
(b)

0
–π –π/2 0 π/4 π/2 π
(–fs/2) ω (fs/2)
(–fs/4) (fs/8) (fs/4)
π/8
(fs/16)

Figure S5–23 (Cont'd)

Solution: 5.24
The number of unit-delay elements in the upper path of the parallel-path
filter must be equal to the group delay of the tapped-delay line subfilter in
the bottom path. Because the bottom path subfilter has symmetrical
coefficients and four unit-delay elements, as shown in Figure S5–24, its
group delay is
D 4
G = 2 = 2 = 2 samples.

As such, for time synchronization, we must use two unit-delay

Property
elements of Pearson
in the upper Education.
path as shown NotS5–24.
in Figure permissible for redistribution.
 Time-synchronization
delay line

Delay Delay
y(n)
x(n) x(n–4) –
Delay Delay Delay Delay

w(n)
Figure S5–24

Solution: 5.25
The student should recall that the discrete (sampled) frequency response of
an FIR filter is the DFT of the filter's impulse response (the DFT of the
filter's h(k) coefficients). For an N-tap FIR filter (k = 0, 1, 2, ..., N–1) the N-
sample discrete frequency response is:
N–1
H(m) = ∑ h(k)e–j2πmk/N , for m = 0, 1, 2, ..., N–1.
k=0

A lowpass filter's gain at DC (zero Hz), H(0), is the above

H(m) expression evaluated at m = 0, or:

N–1 N–1 N–1

H(0) = ∑ h(k)e–j2π(0)k/N = ∑ h(k)e0 = ∑ h(k)
k=0 k=0 k=0

which is what we set out to prove.

Solution: 5.26
The magnitude response solution is:

(N–1)/2
|H(ω)| = |h(0) + 2 ∑ h(k)cos(ωk)| .
k=1

Below is a derivation of the above |H(ω)|.

Using the discrete time Fourier transform (DTFT), an odd N-length
symmetrical FIR filter, having an h(k) impulse response, has a frequency
response defined as:
∞ (N–1)/2
H(ω) = ∑h(k)e –jωk
= ∑h(k)e–jωk
k = –∞ k = –(N–1)/2
where ω is a continuous frequency variable. Due to the symmetrical range
of the k index, and the h(k) sequence symmetry, we can write
Property of Pearson Education. Not permissible for redistribution.
 (N–1)/2
H(ω) = h(0)e–j0 + ∑[h(k)e–jω(–k) + h(–k)e–jωk]
k=1

(N–1)/2
= h(0) + ∑h(k)[ejωk + e–jωk]
k=1
Recalling Euler's 2cos(α) = ejα + e–jα identity, we write H(ω) as
(N–1)/2
H(ω) = h(0) + 2 ∑ h(k)cos(ωk).
k=1

Our solution is the magnitude of H(ω) as

(N–1)/2
|H(ω)| = |h(0) + 2 ∑ h(k)cos(ωk)| .
k=1

Note: The above H(ω) applies to odd-N FIR filters whose symmetrical
coefficients (impulse response) are defined by a symmetrical k index. (With
h(0) being the center coefficient.) Such filters are called "noncausal". For
the more standard, causal, filter indexing where h(0) is the first coefficient,
as shown by example in Figure S5–26, the correct frequency response
expression is
(N–1)/2
Hcausal(ω) = [h(0) + 2 ∑ h(k)cos(ωk)]e–jω(N–1)/2
k=1

with its linear phase shift factor. However, the magnitude of H(ω) and
Hcausal(ω) are equal.

Causal h(k)
0.2

0.1

–0.1
0 1 2 3 4 5 6 k

Figure S5–26

Solution: 5.27
The solution is simple. Just rearrange
Atten
Nfir ≈ 22(f – f )
stop pass

to solve for the (fstop – fpass) transition region width, yielding

Atten 55
fstop – fpass ≈ 22N = (22)(191) = 0.013.
fir

That is, for a lowpass FIR filter, having 55 dB of stopband attenuation,

Property
using this device of
thePearson Education. Not permissible for redistribution.
 most narrow transition region width ≈ 0.013fs.

Solution: 5.28
(a) The time-domain difference equation for the filter can be found by way of
inspection of the filter's block diagram, or by redrawing the block diagram
to that shown in Figure S5–28.

x(n) x(n–1) x(n–2) x(n–3) x(n–4)

Delay Delay Delay Delay

b 256 –2b b

y(n)

Figure S5–28

The difference equation is:

y(n) = bx(n) + (256 – 2b)·x(n–2) + bx(n–4).

(b) The filter's five coefficients are:

h(0) = b,
h(1) = 0,
h(2) = 256 – 2b,
h(3) = 0,
h(4) = b.

Because the filter's coefficients are symmetrical, the

filter will have a linear-phase frequency response.

(c) From Figure S5–28, the number of delay elements in the filter's tapped-
delay line is D = 4. So the group delay, measured in samples, is:

D 4
group delay = G = 2 = 2 = 2 samples.

Solution: 5.29
Computing the product of 24 and 13 by way of convolution is shown in
Figure S5–29.

Property of Pearson Education. Not permissible for redistribution.

Solution:
Multiplicand = 24
Multiplier = 13
Multiplier digits 0 2 10 12 = 2x100 + 10x10
+ 12x1 = 312
flipped in order
0 0 2 4
3 1
t Sum the
products
0
24 x 13 = 312

0 0 2 4
3 1
t Sum the
products
2

0 0 2 4
3 1
t Sum the
products
10

0 0 2 4
3 t Sum the
products
12

Figure S5–29

Property of Pearson Education. Not permissible for redistribution.

 Chapter 6 Solutions
Solution: 6.1
The shaded semi-circular band on the z-plane:
• covers a finite "stable" region of the z-plane,
• one boundary of the shaded circular band is on the unit circle (the boundary
between stability and instability),
• the shaded semi-circular band extends over the frequency range of –fs/4 to
+fs/4 Hz. (fs is the filter sample rate.)

A rough sketch of the Laplace s-plane showing a shaded area that corresponds
to the z-plane shaded band's characteristics is shown in Figure S6–1.

Solution:
jω
(Imag.)
s-plane
(s = σ + jω) fs/4

σ
(Real)

–fs/4

Figure S6–1

The s-plane shaded band is a finite-width vertical stripe that:

• is on the left side (stable) of the s-plane,
• borders on the σ = 0 axis (the boundary between stability and instability),
• extends over a frequency range of fs/2 Hz, centered at 0 Hz, on the s-plane.

Solution: 6.2
The filter transfer functions are:

1
(a) H(z) = 1 + z–2 .

1 + 3z–1 + 2z–2
(b) H(z) = 1 + z–3 .

1 + z–1 + z–3 + z–4

(c) H(z) = 1 + z–2 .

Property of Pearson Education. Not permissible for redistribution.

Solution: 6.3
The order of a digital filter is defined as the largest exponent in that filter's
transfer function. The filter orders are:

(a) Order is 2.

(b) Order is 3.

(c) Order is 4.

Solution: 6.4
The filter frequency response equations are those filters' z-domain transfer
functions with variable z replaced with ejω, or:

1
(a) Hpolar(ω) = .
1 + e–j2ω

1 1
(a) HRect(ω) = = .
1 + [cos(2ω) –jsin(2ω)] 1 + cos(2ω) –jsin(2ω)

1 + 3e–jω + 2e–j2ω
(b) Hpolar(ω) = .
1 + e–j3ω
1 + 3[cos(ω) –jsin(ω)] + 2[cos(2ω) –jsin(2ω)]
HRect(ω) =
1 + [cos(3ω) –jsin(3ω)]

1 + 3cos(ω) + 2cos(2ω) –j[sin(ω) + sin(2ω)]

(b) HRect(ω) = .
1 + cos(3ω) –jsin(3ω)

1 + e–jω + e–j3ω + e–j4ω

(c) Hpolar(ω) = .
1 + e–j2ω
1 + [cos(ω) –jsin(ω)] + [cos(3ω) –jsin(3ω)] + [cos(4ω) –jsin(4ω)]
HRect(ω) =
1 + [cos(2ω) –jsin(2ω)]

1 + cos(ω) + cos(3ω) + cos(4ω) –j[sin(ω) + sin(3ω) + sin(4ω)]

(c) HRect(ω) = .
1 + cos(2ω) –jsin(2ω)

Solution: 6.5
(a) One or more poles outside the unit circle means the filter is unstable.

Property of Pearson Education. Not permissible for redistribution.

 (b) A zero on the unit circle means that at some frequency the filter has
infinite attenuation (a "magnitude null").

Solution: 6.6
(a) Given H(z) as
z2 + 0.3z + 1
H(z) = z2 + 0.3z + 0.8 ,

we multiply H(z) by z–2/z–2 providing the desired form with z having

negative-only exponents.
1 + 0.3z–1 + z–2
H(z) = 1 + 0.3z–1 + 0.8z–2 .

(b) The filter is stable, because the filter's poles are within the unit circle. We
verify this as follows: Given the original form of H(z) as
z2 + 0.3z + 1
H(z) = z2 + 0.3z + 0.8 ,

we find the roots of the denominator, which are the locations of the filter's
poles, by first setting the denominator of H(z) to zero as
z2 + 0.3z + 0.8 = 0.
Using the quadratic factorization formula, we factor the above 2nd-order
polynomial to yield
z2 + 0.3z + 0.8 = (z + p1)(z + p2)

= (z + 0.15 +j0.8818)(z + 0.15 – j0.8818).

So the first pole location is z = –p1 = –0.15 – j0.8818, and the second pole
location is z = –p2 = –0.15 + j0.8818.

The two conjugate poles are each located inside the z-

plane's unit circle at a radius of:

(–0.15)2 + (±0.8818)2 = 0.8944,

therefore the filter is stable.

(c) The Direct Form I structure for H(z) are shown in Figure S6–6(a).

(d) The Direct Form II structure for H(z) are shown in Figure S6–6(b).

Property of Pearson Education. Not permissible for redistribution.

 Solution:
Direct Form I Simplified Direct Form I
x(n) y(n) x(n) y(n)

z–1 1 z–1 z–1 z–1

(a) OR

0.3 –0.3 0.3 –0.3

z–1 z–1 z–1 z–1

1 –0.8 –0.8

Solution:
Direct Form II Simplified Direct Form II
x(n) y(n) x(n) y(n)

z–1 1 z–1
(b) OR

–0.3 z–1 0.3 –0.3 z–1 0.3

–0.8 1 –0.8

Figure S6–6

Solution: 6.7
(a) The difference equation describing the filter is

y(n) = h(0)x(n) + h(1)x(n–1) + h(2)x(n–2).

(b) The z-transform of the y(n) difference equation from Part (a) is

Y(z) = h(0)X(z) + h(1)X(z)z–1 + h(2)X(z)z–2.

(c) The z-domain transfer function of the filter is

Y(z)
H(z) = X(z) = h(0) + h(1)z–1 + h(2)z–2.

(d) The order of the filter is 2 due to the z–2 term in H(z).

Solution: 6.8
(a) The answer is "No."

Property of Pearson Education. Not permissible for redistribution.

 If we know the time-domain difference equation of an IIR filter, we can
determine the filter's frequency response by converting that difference
equation to a z-domain transfer function, replace z with ejω, and evaluate the
new frequency-domain expression over the range of –π ≤ ω ≤ π
(corresponding to a cyclic frequency range of –fs/2 ≤ f ≤ fs/2 Hz).

If we know the impulse response of the filter, we can determine the

frequency response of that filter by taking the discrete Fourier transform
(DFT) of that impulse response sequence. In practice, the impulse response
sequence is padded with zero-valued samples to improve the resolution
(granularity) of the frequency response DFT results.

(b) If we know the H(z) z-domain transfer function equation of a digital filter, to
determine the filter's frequency response we replace z with ejω in H(z),
creating H(ω), and evaluate the H(ω) frequency-domain expression over the
range of –π ≤ ω ≤ π.

Solution: 6.9
The filter block diagrams can be drawn by inspection of Hcas(z), rewritten as:
⎛ 1 − 4 z −1 + 2 z −2 ⎞
H cas ( z ) =
Y ( z) ⎛
=⎜
1 ⎞
( −1
⎟ ⋅ 1 − 4z + 2z
X ( z ) ⎝ 1 − 0.3 z −1 ⎠
−2
)
=⎜ −1 ⎟
.
⎝ 1 − 0.3z ⎠
Or if necessary, the block diagrams can be drawn based on the filter's time-
domain difference equation found by taking the inverse z-transform of Hcas(z).
Doing that, the filter's difference equation is:
y(n) = x(n) – 4x(n–1) + 2x(n–2) + 0.3y(n–1).
Based on either Hcas(z) or y(n), the problem's solutions are the Direct Form I and
Direct Form II block diagrams given in Figure S6–09.

Solution:
Direct Form I Direct Form II
x(n) y(n) x(n) y(n)

z–1 z–1 z–1

–4 0.3 0.3 –4
z–1 z–1

2 2

Figure S6–09

Solution: 6.10
Property of Pearson Education. Not permissible for redistribution.
 The h1(k) impulse response of Filter H1 is given in Figure S6–10(a)
showing this filter to be an IIR filter.
The H1 filter's z-plane pole/zero plot is provided in Figure S6–10(b)
showing a pole at z = 1 on the unit circle.

The h2(k) impulse response of Filter H2 is given in Figure S6–10(c)

showing this filter to be an FIR filter.
That H2 filter's z-plane pole/zero plot is provided in Figure S6–10(d) where
the pole at z = 1 is canceled by a zero at z = 1, yielding a filter having only
zeros on the z-plane.

Filter H1
h1(k) 1

Imaginary part
2
pole
1.5
0
1
0.5
–1
0 –1 0 1
0 5 k 10 15 Real part
(a) (b)

Filter H2
h2(k) 1
1 Imaginary part
pole
0
0.5 zero
–1
0 –1 0 1
0 5 k 10 15 Real part
(c) (d)

Figure S6–10

Solution: 6.11
Because the networks contain cascaded subnetworks, we start our solution
by examining the subnetworks. Subnetwork-A is shown in Figure S6–11(a).

Subnetwork-A
Network-A
x(n) w(n) y(n)
(a)
z–1 z–1 z–1

Subnetwork-B
Network-B
x(n) w(n) y(n)
(b)
z–1 z–1 z–1

Property of Pearson Education. Not permissible for redistribution.

 Figure S6–11

Subnetwork-A's time-domain difference equation is

w(n) = x(n) + w(n–1).
Its z-domain transfer function is
W(z) 1
HSub-A(z) = X(z) = 1 – z–1 .

Substituting ejω for z in Hsub-A(z), Subnetwork-A's frequency response

expression is:
1
HSub-A(ω) = .
1 – e–jω
Thus the frequency response of the full cascaded Network-A is:
1
HNet-A(ω) = [HSub-A(ω)]3 = .
(1 – e–jω)3
Next, using the same analysis steps, and Figure S6–11(b), the frequency
response of Subnetwork-B is:
e–jω
HSub-B(ω) = .
1 – e–jω
The frequency response of the full cascaded Network-B is:
(e–jω)3
HNet-B(ω) = [HSub-B(ω)]3 = = (e–jω)3 ·HNet-A(ω).
(1 – e–jω)3
Because the magnitude of (e–jω)3 = e–j3ω is always unity, we can relate the
two network frequency magnitude responses as

|HNet-B(ω)| = 1·|HNet-A(ω)|.

So the solution to this problem is "Yes", the frequency magnitude

responses of Network-A and Network_B are identical.

Solution: 6.12
We prove that the z-plane pole locations for the two filters are identical by
examining their H(z) transfer functions. For Filter# 1 we write
Y(z) = X(z) + 2Acos(α)Y(z)z–1 – A2Y(z)z–2.
Rearranging the above, we have
Y(z) 1
H1(z) = X(z) = .
1 – 2Acos(α)z–1 + A2z–2
Next, analyzing Filter# 2 we write
U(z) = X(z) – Asin(α)Y(z)z–1 + Acos(α)W(z)z–1
Rearranging the above,
Property we have
of Pearson Education. Not permissible for redistribution.
 X(z) – Asin(α)Y(z)z–1
U(z) = . (6.12–1)
1 – Acos(α)z–1
For Filter# 2 we can also write
Y(z) = Asin(α)U(z)z–1 + Acos(α)Y(z)z–1.
Rearranging the above, we have
Y(z)[1 – Acos(α)z–1]
U(z) = . (6.12–2)
Asin(α)z–1
Setting Eqs. (6.12–1) and (6.12–2) equal to each other, we write

Y(z)[1 – Acos(α)z–1]·[1 – Acos(α)z–1]1 = [X(z) – Asin(α)Y(z)z–1]·[Asin(α)z–1].

Rearranging the above to solve for H2(z) = Y(z)/X(z), and recalling the trig
identity: cos2(α) + sin2(α) = 1, we write
Y(z) Asin(α)z–1
H2(z) = X(z) = .
1 – 2Acos(α)z–1 + A2z–2
Because H1(z) and H2(z) have identical denominator terms their z-plane
pole locations will be identical, which is what we set out to prove.

Solution: 6.13
(a) We can derive the DC bias removal filter's H(z), from Figure S6–13–I(a), by
writing
y(n) = x(n) – x(n–1) + Ay(n–1).
Next we convert that expression to the z-domain and write
Y(z) = X(z) – X(z)z–1 + A(z)Y(z)z–1.
Solving the above expression for Y(z)/X(z) we have our solution of
Y(z) 1 – z–1
H(z) = X(z) = 1 – Az–1 .

Alternatively, the student could find the above H(z) using the text's Eq. (6–25).

Imag.
z-plane
x(n) y(n)
zero at
z=1
z–1 z–1
Real
Unit pole at
circle z=A
–1 A
(a) (b)

Figure S6–13–I

Property
(b) We find of z-plane
the filter's Pearson Education.
zero Not permissible
by setting H(z)'s for redistribution.
numerator equal to zero as:
 1 – z–1 = 0.
Solving the above expression for z we see that the filter indeed has a
z-plane zero at z = 1 (at the cyclic frequency of zero Hz), as shown
in S6–13–I(b).

(c) Two equivalent versions of the Direct Form II implementation of the filter
are provided in S6–13–II. This DC bias removal filter is both stable and
often used in practice.

Solution:
Direct Form II Alternate Direct Form II
x(n) y(n) x(n) y(n)
–
z–1 z–1

A –1 A

Figure S6–13–II

Solution: 6.14
(a) This part of the problem is simple because we gave the student the factored
form of the notch filter's transfer function as:
Y(z) 1 –2cos(ωc)z–1 + z–2 (1 –ejωcz–1)(1 –e–jωcz–1)
H(z) = X(z) = = ,
1 –2Rcos(ωc)z–1 + R2z–2 (1 –Rejωcz–1)(1 –Re–jωcz–1)
and R = 0.9. Finding the poles and zeros locations on the z-plane merely
means determining the values of z that make the H(z) numerator and
denominator factors equal to zero. Thus we write
(1 –ejωcz–1) = 0, and (1 –e–jωcz–1) = 0,
specifying two zeros at, z = z0 and z = z1, where
z0 = e+jωc, and z1 = e–jωc. [Both zeros are on the unit circle.]
The location of the filter's poles are found by writing
(1 –Rejωcz–1) = 0, and (1 –Re–jωcz–1) = 0,
specifying two poles, z = p0 and z = p1, where
p0 = 0.9e+jωc, and p1 = 0.9e–jωc.
The pole/zero diagram for this generic notch filter is provided in Figure S6–
14(a).

Property of Pearson Education. Not permissible for redistribution.

 Solution:
Imag.
z-plane Radii of poles
are 0.9.

ωc
(a) –ωc
Real

–1 1

Imag.
z-plane ωc = 0.419
radians
(120 Hz)

ωc
(b) –ωc Real

–1 1

Mag. Response (filter notch at 120 Hz)

0
(c)
–10
dB

–20
–30
–800 –600 –400 –200 0 200 400 600 800
Freq (Hz) fs/2
120

Figure S6–14

(b) For a signal sample rate of fs = 1800 Hz, to position the notch filter's center
frequency at 120 Hz we set ωc to

120
ωc = 2π . 1800 = 0.419 radians

which corresponds to a cyclic frequency of 120 Hz as shown in the

frequency magnitude response curve in Figure S6–14(c). The pole/zero
locations for a notch filter centered at 120 Hz when fs = 1800 Hz, are
provided in Figure S6–14(b).

Solution: 6.15
Property of Pearson Education. Not permissible for redistribution.
 To find the locations of the filter's two z-plane zeros, first we set the transfer
function polynomial equal to zero as:
1 + Bz–1 + z–2 = 0.
Next we multiply both sides of the above expression by z2 converting it to
z2 + Bz + 1 = 0
so we can use the following quadratic factorization formula to factor the
polynomial
⎛ b b2 –4ac ⎞ ⎛ b b2 –4ac ⎞
az2 + bz + c = ⎜z + 2a + 2 ⎟ · ⎜z +
2a –
⎟
4a2 ⎠ .
⎝ 4a ⎠ ⎝
Because a = 1, b = B, and c = 1 for our filter polynomial, we can write
⎛ B B2 –4 ⎞ ⎛ B B2 –4 ⎞
z2 + Bz + 1 = ⎜z + 2 + ⎟ · ⎜z +
2 –
⎟
⎝ 4 ⎠ ⎝ 4 ⎠

= (z – z0)(z – z1)
where z0 and z1 are the locations of the filter's two zeros on the z-plane. The
sum of z0 and z1 is
⎛ B B2 –4 ⎞ ⎛ B B2 –4 ⎞
z0 + z 1 = ⎜– 2 – ⎟ ⎜
4 ⎠ + ⎝– 2 +
⎟
⎝ 4 ⎠
or
B B
z0 + z1 = – 2 – 2 = –B

which is what we set out to prove.

Solution: 6.16
(a) The z-domain transfer function of the filter is, by inspection,:
Y(z) 1 – 6z–1 + 8z–2
H(z) = X(z) = 1 – 2.5z–1 + z–2 .

(b) In preparation for polynomial factoring, to find the filter's poles and zeros,
we multiply H(z) by z2/z2 to the exponents of z positive. Doing so yields
z2 – 6z + 8 (z – 2)(z – 4)
H(z) = z–2 – 2.5z + z–2 = (z – 0.5)(z – 2) .

From the above factored form of H(z) we see that the filter has zeros at z = 1
and z = 4. In addition, the filter has poles at z = 0.5 and z = 2. So the
problem solution is the z-plane pole/zero diagram shown in Figure S6–16(a).

Property of Pearson Education. Not permissible for redistribution.

 Solutions:
1 z-plane
Pole/zero

Imaginary part
0.5 cancelation
0
(a)
–0.5
–1
–1 0 1 2 3 4
Real part

Figure S6–16

(c) Canceling the common (z – 2) factors in the numerator and denominator of

the above H(z) gives us the simplified transfer function of
(z – 4) 1 – 4z–1
H(z) = (z – 0.5) = 1 – 0.5z–1 .

The block diagram of the simpler equivalent filter (having fewer multipliers)
is shown in Figure S6–16(b).

Solutions:
X(z) Y(z)

z–1
(b)

0.5 –4

Figure S6–16 (Cont'd)

Solution: 6.17
The answer is "No." It is not possible to have a real-coefficient filter
whose frequency-domain phase response is of the form:
θ(ω) = C.
The above θ(ω) implies that at zero Hz (ω = 0) we have a non-zero phase
response of C radians, and this situation has no meaning. Stated in different
words, a phase of θ(0) = C implies that with a constant (zero Hz) input signal
applied to the filter, the constant (zero Hz) filter output will be shifted in phase
by C radians. However the notion of phase is not defined (has no meaning) for a
constant-amplitude, DC, signal. All digital filters (having real-valued
coefficients) have a phase response of zero radians at a frequency of zero Hz.

Solution: 6.18
The time-domain difference equations are.
Property of Pearson Education.
w(n) Not permissible for redistribution.
= x(n) + 0.25w(n–1)
 and
y(n) = 3[2x(n) + w(n)] + 3w(n–1).
Converting those expressions to the z-domain we write
X(z)
W(z) = X(z) + 0.25W(z)z–1 = 1 – 0.25z–1 (6.18–1)

and
Y(z) = 3[2X(z) + W(z)] + 3W(z)z–1

= 6X(z) + 3W(z)(1 + z–1)

X(z)
= 6X(z) + 3 1 – 0.25z–1 (1 + z–1). (6.18–2)

From Eq. (6.18–2), we write our desired H(z) as

Y(z) 1 + z–1
H(z) = X(z) = 6 + 3 1 – 0.25z–1

6(1 – 0.25z–1) 3(1 + z–1)

= 1 – 0.25z–1 + 1 – 0.25z–1

or
9 + 1.5z–1
H(z) = 1 – 0.25z–1 .

Solution: 6.19
(a) The network's time-domain difference equation is:
y(n) = x(n) + y(n–1) –Qy(n–1) = x(n) + (1–Q)y(n–1).
To simplify the notation we temporarily replace the (1–Q) factor with the
symbol β, and replace x(n) with the constant D, to write:
y(n) = D + βy(n–1).
At various values of time index n the network's outputs are:
n = 0, y(0) = D,
n = 1, y(1) = D + βD,
n = 2, y(2) = D + β(D + βD) = D + βD + β2D,
n = 3, y(3) = D + β(D + βD + β2D) = D + βD + β2D + β3D,
...
n = N, y(N) = D + βD + β2D + β3D + ... + βND
N
= D + D ∑ βn .
n=1
At time n = 100 we can write:
Property of Pearson Education. Not permissible for redistribution.
 100
y(100) = D + D ∑ βn .
n=1
The above y(100) contains a geometric series which can, after reviewing
Appendix B, be written as:

β – β101
y(100) = D + D .
1–β

Because Q is a positive number less than one, β is also a positive number

less than one. Thus we assume β101 is zero allowing us to rewrite y(100) as:
β 1–Q
y(100) = D + D = D + D 1 – (1–Q)
1–β
Or

1–Q D D
y(100) = D + D Q = D + Q –D = Q

which is what we set out to prove.

The time-domain behavior of the recursive network, for an input of D = 2,

is shown in Figure S6–19(a).

Input x(n) = D = 2
10 Imag.
Q = 0.2 z-plane Single pole at
8 zpole = 1–Q
y(n)
6
Q = 0.4
4 Q = 0.6 Real

2 Q = 0.8

0 -1 1
0 5 10 15 20 25
n
(a) (b)

Figure S6–19

(b) The network's time-domain difference equation is:

y(n) = x(n) + (1–Q)y(n–1),
so the z-domain expression describing the network is:
Y(z) = X(z) + (1–Q)Y(z)z–1.
As such, the network's transfer function is:

Property of Pearson Education. Not permissible for redistribution.

 Y(z) 1 1
H(z) = X(z) = 1– (1–Q)z–1 = 1– (1–Q)/z .

At zero Hz, where z = 1, the network's DC (zero Hz) gain is:

1 1
H(z)|z=1 = 1– (1–Q) = Q

which is what we set out to prove.

Alternatively, we could replace z with ejω in H(z) yielding a frequency

response of:
1
H(ω) = .
1– (1–Q)e–jω
At zero Hz, ω = 0, the network's DC (zero Hz) gain is:
1 1 1
H(ω)|ω=0 = 1– (1–Q)e–j0 = 1– (1–Q) = Q

which agrees with the above H(z)|z=1.

(c) To verify stability, the location of the network's poles is found by setting its
H(z) transfer function's denominator equal to zero, and solving for z. Doing
so we have
1– (1–Q)z–1 = 0.

Solving the above expression for z yields a single pole at:

zpole = 1–Q

so the filter is stable when |zpole| = |1–Q| < 1. With Q being a real
number, the |1–Q| < 1 condition means that 0 < Q < 2 guarantees
stability. Thus any Q in the range of 0 < Q ≤ 1 yields a stable system as
shown in Figure S6–19(b).

Solution: 6.20
(a) The pole/zero plot is found, of course, by determining the poles and zeros of
the system's transfer function. We find the desired H(z) transfer function
from the time-domain difference equation of
y(n) = x(n) –y(n–1), or

y(n) + y(n–1) = x(n).

Performing the z-transform of the difference equation, we have
Y(z) + Y(z)z–1 = X(z),
yielding the H(z) transfer function of
Property of Pearson Education. Not permissible for redistribution.
 Y(z) 1 z
H(z) = X(z) = 1 + z–1 = z + 1

with a zero at z = 0, and a pole at z = –1. The desired pole/zero plot is shown
in Figure S6–20-I.

Solution:
Imag.

z-plane
pole zero

Real

–1 1

Figure S6–20-I

(b) The pole's location of z = –1 corresponds to a cyclic frequency of fs/2 Hz.

(c) The time-domain impulse response of the system is shown in Figure S6–20-II.

Solution:
h(n)
1

–1
0 1 3 5 7 9 11 13 15 17 19 21 Time
n

Figure S6–20-II

(d) Because the system's impulse response is oscillatory having two

samples/cycle, its frequency is fs/2 Hz. This corresponds to the system's pole
location on the z-plane at a cyclic frequency of fs/2 Hz.

Solution: 6.21
The solutions to this problem are found using the following steps:

1. Determine the Hk(z) transfer function for each difference equation.

2. Find the locations of the poles for each Hk(z).
3. Examine the z-plane angle of the poles to determine the resonant
frequencies of each filter.
Property of Pearson Education. Not permissible for redistribution.
 For example, analyzing the fourth difference equation, we write:
y4(n) = x(n) – 0.81y4(n–2).
The H4(z) transfer function is
1
H4(z) = 1 + 0z–1 + 0.81z–2 .

The poles of H4(z) are located at z = 0 + j0.9, and z = 0 – j0.9 as shown in Figure
S6–21. The angles of those poles are ±π/2 radians which correspond to cyclic
frequencies of ±fs/4. Thus the y4(n) filter is associated with the |HA(m)|
frequency magnitude response, and we indicate that by writing:
y4(n) → |HA(m)|.

Imag.
y1(n) filter Pole at
z = 0 +j0.9

Real

–1 1

Figure S6–21

The complete solution to this problem is:

Y1(z) has only one pole at 0 radians so, y1(n) → |HD(f)|.

Y2(z) has only one pole at –π radians so, y2(n) → |HF(f)|.
Y3(z) has poles at 0 and π radians so, y3(n) → |HE(f)|.
Y4(z) has poles at ±π/2 radians so, y4(n) → |HA(f)|.
Y5(z) has poles at ±3π/4 radians so, y5(n) → |HC(f)|.
Y6(z) has poles at ±π/4 radians so, y6(n) → |HB(f)|.

Solution: 6.22
Finding the DC gain of the IIR filter starts with the filter's transfer function:

b(0) + b(1)z–1 + b(2)z–2 b(0) + b(1)/z + b(2)/z2

HIIR(z) = –1
1 –a(1)z –a(2)z –2 = 1 –a(1)/z –a(2)/z2 .

Next we set z = 1 in HIIR(z) because this corresponds to a frequency of zero

radians/sample, giving us our desired DC gain solution of:

b(0) + b(1) + b(2)

DC gain = HIIR(z)|z=1 = 1 –a(1) –a(2) .
Property of Pearson Education. Not permissible for redistribution.
 Of course, a student could have substituted ejω for z in HIIR(z), and then set
ω = 0 to produce the above DC gain solution.

Solution: 6.23
(a) The allpass filter's Hap(ω) frequency response is the filters' z-domain transfer
function with variable z replaced with ejω. That is
–K + e–jω
Hap(ω) = .
1 – Ke–jω
Next we describe the frequency magnitude squared, |H(ω)|2, as
–K + e–jω [–K + cos(ω)]2 + [–sin(ω)]2
|Hap(ω)|2 = | |2 =
1 – Ke–jω [1 –Kcos(ω)]2 + [Ksin(ω)]2 ]

K2 –2Kcos(ω) + cos2(ω) + sin2(ω) K2 –2Kcos(ω) + 1

= 2 2 2 2 =
1 –2Kcos(ω) + K cos (ω) + K sin (ω) 1 –2Kcos(ω) + K2[cos2(ω) + sin2(ω)]
or

2 K2 –2Kcos(ω) + 1
|Hap(ω)| = = 1, for all ω.
1 –2Kcos(ω) + K2

(b) The Direct Form I and Direct Form II structures for Hap(z) are shown in
Figure S6–23.

Solution:
Direct Form I Direct Form II
x(n) y(n) x(n) y(n)

z–1 –K z–1 z–1 –K

K K

Figure S6–23

(c) Explanation 1: A transfer function zero on any filter's z-plane

unit circle would cause that filter to have a zero-valued
magnitude response (infinite attenuation) at some frequency.
Because the allpass filter has a unity magnitude response at
all frequencies, it cannot have a zero on the unit circle.

Explanation 2: The allpass filter has a transfer function zero at

z = 1/K, and a pole at z = K. To place a zero on the unit circle
K must be equal to one. However, when K = 1 the filter's
pole will reside on top of the zero at z = 1 and the zero would
be canceled (annihilated) by the pole.

Property of Pearson Education. Not permissible for redistribution.

Solution: 6.24
We can solve the problem by looking at the filter's z-plane pole/zero
locations. The filter's single pole location is found by setting the Hg(z)
transfer function's denominator to zero, and solving for z. Doing that we
have
1 + 0.1127z–1 = 0
yielding a pole location at z = p0 = –0.1127. Setting Hg(z)'s numerator to
zero,
0.9152 + 0.1889z–1 = 0,
and solving for z gives us a zero location at z = z0 = –0.206. Figure S6–24 is
the filter's z-plane pole/zero diagram. Examining that diagram, we see that
the filter's zero is closer to z = –1 (fs/2 Hz) than it is to z = 1 (0 Hz).

So the filter will have a smaller frequency magnitude response at fs/2 Hz

than at 0 Hz. Because of this, the Hg(z) filter is a lowpass filter.

Alternatively, the student could have replaced z in Hg(z) with ejω to obtain
the Hg(ω) frequency response expression for the filter. Evaluating that
expression at ω = 0 (0 Hz) yields an Hg(ω = 0) of 0.99. Evaluating Hg(ω) at
ω = π (fs/2 Hz) gives an Hg(ω = π) of 0.82. Again showing that the filter is a
lowpass filter.

1 z-plane
Imaginary Part

–1
–1 0 1
Real Part

Figure S6–24

Solution: 6.25
The derivation of the z-domain transfer function of the general 2nd-order
recursive network, in Figure S6–25, proceeds as follows:

Property of Pearson Education. Not permissible for redistribution.

 General 2nd-order recursive network
x(n) w(n) y(n)

z–1 b(0) a(0) z–1

z–1 b(1) a(1) z–1

b(2) a(2)

Figure S6–25

We start be writing the network's time-domain difference equations as:

w(n) = b(0)x(n) + b(1)x(n–1) + b(2)x(n–2).

y(n) = a(0)[w(n) + a(1)y(n–1) +a(2)y(n–2)]

= a(0)[b(0)x(n) + b(1)x(n–1) + b(2)x(n–2) + a(1)y(n–1) + a(2)y(n–2)].

Collecting like terms, we have:

y(n) –a(0)[a(1)y(n–1) +a(2)y(n–2)] = a(0)[b(0)x(n) + b(1)x(n–1) + b(2)x(n–2)].

Converting (transforming) the above to a z-domain expression, we have:

Y(z) –a(0)[a(1)Y(z)z–1 +a(2)Y(z)z–2] = a(0)[b(0)X(z) + b(1)X(z)z–1 + b(2)X(z)z–2].

or

Y(z)[1 –a(0)a(1)z–1 –a(0)a(2)z–2] = X(z)[a(0)b(0) + a(0)b(1)z–1 + a(0)b(2)z–2].

So one solution for the transfer function of the general recursive 2nd-order
network is:

Y(z) a(0)b(0) + a(0)b(1)z–1 + a(0)b(2)z–2

HGen(z) = X(z) = 1 –a(0)a(1)z–1 –a(0)a(2)z–2 .

Dividing through by a(0) yields a somewhat simpler-form solution of:

b(0) + b(1)z–1 + b(2)z–2

HGen(z) = 1/a(0) –a(1)z–1 –a(2)z–2 .

Solution: 6.26
(a) The filter's time-domain difference equation is
y(n) = 2x(n) – e–0.88y(n–1).
Next we convert that expression to the z-domain and write
Property of PearsonY(z)Education. Not
= 2X(z) – e–0.88 permissible
Y(z)z–1
. for redistribution.
 Solving the above expression for Y(z)/X(z) we have the z-domain transfer
function of
Y(z) 2
H(z) = X(z) = 1 + e–0.88z–1 .

Replacing H(z)'s variable z with ejω, we write our desired H(ω) frequency
response as:
2
H(ω) = –0.88 –jω .
1+e e
(b) Setting the denominator of H(z) to zero, we find the location of the single z-
plane pole as:
1 + e–0.88zpole–1 = 0.
Solving for zpole, the z-plane pole value is

zpole = –e–0.88 = –0.4148.

(c) Evaluating H(ω) at ω = 0, and ω = π, we have

2 2
M0 = H(ω)|ω=0 = 1 + e–0.88e–j0 = 1 + e–0.88 = 1.4136.

2 2
Mπ = H(ω)|ω=π = –0.88 –jπ = –0.88
1+e e 1+e [cos(π) –jsin(π)]

1 1
Mπ = 1 + e–0.88[–1 – j0] = 1 – e–0.88 = 3.4175.

Solution: 6.27
(a) The difference equation for the averaging network, shown at the left side of
Figure S6–27(a), is:
x(n) n–1
y(n) = n + n y(n–1).
The right side of Figure S6–27(a) is provided to remind us that
multiplication by 1/n can be interpreted as a division operation.
Implementing the y(n) difference equation requires 2 divisions, 1 multiply,
and 1 addition per output sample. We can do better. (Division operations in
binary arithmetic are computation intensive and are to be avoided whenever
possible.)

Property of Pearson Education. Not permissible for redistribution.

 Original network
x(n) y(n) x(n) _. y(n)
.n

z–1 _. z–1
1/n = .n
(a) 1/n _..
n

(n–1)/n (n–1)

Figure S6–27

The original network's difference equation can be rewritten as

1
y(n) = n [x(n) + (n–1)y(n–1)]

yielding the block diagram shown in Figure S6–27(b). This Solution 1

network requires only 1 division, 1 multiply and 1 addition per output
sample.

Solution:
Solution 1
x(n) y(n) x(n) _.. n y(n)
– –
(b) 1/n z–1 z–1

y(n–1) y(n–1)
(n–1) (n–1)

Figure S6–27 (Cont'd)

We can do better still. Realizing that

n–1 1
n = 1 – n ,
we can rewrite the original real-time averaging network's difference
equation as:
x(n) y(n–1) x(n) –y(n–1)
y(n) = n + y(n–1) – n = n + y(n–1).

The implementation of this final Solution 2 difference equation is shown in

Figure S6–27(c). In this second solution scenario, the network requires
merely 1 division, and 2 additions per output sample.

Solution:
Solution 2
x(n) y(n) x(n) _.. n y(n)
(c) – –
1/n z–1 z–1
y(n–1) y(n–1)

Property of Pearson Education.

Figure Not permissible for redistribution.
S6–27 (Cont'd)
 (b) The original averaging network, shown in Figure S6–27(a), and the two
alternative networks provided in the solution of Part (a) of this problem,
have the following transfer function:
1/n
H(z) = 1 –(1 – 1/n)z–1 .

Setting the H(z) denominator equal to zero to find the network's single pole
location gives us
1
zpole = 1 – n .

1
zpole = 1 – n [for all equivalent networks].

When n = 1 the networks' poles are located a z = 0 which is inside the unit
circle guaranteeing stability. As n increases, the magnitude of the pole
locations are always less than one revealing that the networks in Figure S6–
27 are always stable.

Solution: 6.28
(a) Given the two poles and single zero on the z-plane, defined by
p0 = 0.25 + j0.25, p1 = 0.25 – j0.25, and z0 = –1,
the associated H(z) transfer function is
G(z – z0) z +1
H(z) = (z – p )(z – p ) = G· (z –0.25 –j0.25)(z –0.25 + j0.25) .
0 1

Multiplying the denominator factors, we have:

z +1
H(z) = G · z2 –0.5z + 0.125 .

Next, multiplying the numerator and denominator terms by z–2, we have our
solution of:

z–1 + z–2
H(z) = G · 1 –0.5z–1 + 0.125z–2 .

Scalar G is an arbitrary constant.

(b) The Direct Form I block diagram of the H(z) filter is given in Figure S6–
28(a). An alternate, but equivalent, block diagram depiction is given in
Figure S6–28(b).

Notice that the block diagram in Figure S6–28(a) is an implementation of:

0 + z–1 + z–2
H(z) = G · 1 –0.5z–1 + 0.125z–2 .

Property of Pearson Education. Not permissible for redistribution.

 (c) There are three equivalent filter implementations that eliminate one of the
multipliers in the H(z) Direct Form I block diagram in Figures S6–28(a) and
(b).
Those implementation diagrams are shown in Figures S6–28(c), (d), and (e).

Direct Form I Alternate Structure

x(n) y(n) x(n) y(n)
z–1

z–1 z–1 G z–1

z–1

G 0.5 z–1 G 0.5

z–1 z–1

G 0.125 0.125
(a) (b)

Simplified Structure-I Simplified Structure-II

x(n) y(n) x(n) y(n)
z–1 z–1

G z–1 G z–1
z–1 z–1

0.5 0.5 z–1

z–1

0.125 0.125
(c) (d)

Simplified Structure-III
x(n) y(n)
z–1

z–1 G
z–1

0.5 z–1

(e) 0.125

Figure S6–28

Solution: 6.29
To find the roots of polynomial P, we set P equal to zero, as
Roots of P → z2 + bz + c = 0,
and solve for z. To find the roots of polynomial Q, we set Q equal to zero, as
Roots of Q → Gz2 + Gbz + Gc = 0,
and solve for z. If we multiply both sides of the above 'Roots of Q'
expression by 1/G we have
Roots of Q → z2 + bz + c = 0
Property of Pearson Education. Not permissible for redistribution.
 which is equal to the 'Roots of P' expression. Thus the roots of polynomial
Q are equal to the roots of polynomial P, which is what we set out to prove.

Solution: 6.30
(a) Given the pole/zero characteristics of z-plane A, the associated |HA(f)|
magnitude response is shown in Figure S6–30–I(a).

Solution:
Approximate |HA(f)|
∞
Note
(a)

0
–fs/2 –3fs/8 –fs/4 –fs/8 0 fs/8 fs/4 3fs/8 fs/2
Frequency

Figure S6–30–I

Because the poles are directly on the unit circle, the magnitude response is
infinity at f = ±fs/8 Hz.

(b) Given the pole/zero characteristics of z-plane B, the associated |HB(f)|

magnitude response is shown in Figure S6–30–I(b).

Solution:
Approximate |HB(f)|
max |Hb(f)|
(< ∞)

(b)

|HB(–fs/2)| = 0 |HB(fs/2)| = 0
0
–fs/2 –3fs/8 –fs/4 –fs/8 0 fs/8 fs/4 3fs/8 fs/2
Frequency

Figure S6–30–I (Cont'd)

Because the poles are inside the unit circle, the magnitude response is less
than infinity at f = ±fs/8 Hz. Because there is a zero at z = –1, the magnitude
response is zero at f = ±fs/2 Hz.

(c) Given the pole/zero characteristics of z-plane C, in Figure S6–30–II(a), we

know that the dual poles near z = 1 are lying exactly on top of the dual zeros
so those pole/zero pairs cancel each other. The result is the pole/zero plot
shown in Figure S6–30–II(b).

Property of Pearson Education. Not permissible for redistribution.

 The solution to this problem is the approximate |HC(f)| magnitude response
shown in Figure S6–30–II(c).

z-plane C Approximate z-plane C

1 1

Imaginary Part

Imaginary Part
0 0

–1 –1
–1 0 1 –1 0 1
Real Part Real Part
(a) (b)

Solution:
Approximate |HC(f)| Passband (main lobe)
max |HC(f)|
Sidelobes Sidelobes

(c)

0
–fs/2 –3fs/8 –fs/4 –fs/8 0 fs/8 fs/4 3fs/8 fs/2
Frequency

Figure S6–30–II

(d) This problem is a bit "tricky". All of the zeros in z-plane D, shown in Figure
S6–30–III(a), affect the |HD(f)| magnitude response but to generate a rough
sketch of |HD(f)| we can ignore all the zeros not lying on the unit circle. As
such, we can use the approximate pole/zero plot shown in Figure S6–30–
III(b) providing the problem solution's |HD(f)| magnitude response shown in
Figure S6–30–III(c).

z-plane D Approximate z-plane D

1 1
Imaginary Part

Imaginary Part

0 0

–1 –1
–1 0 1 –1 0 1
Real Part Real Part
(a) (b)

Property of Pearson Education. Not permissible for redistribution.

 Solution:
Approximate |HD(f)|
max |HD(f)|

(c)

0
–fs/2 –3fs/8 –fs/4 –fs/8 0 fs/8 fs/4 3fs/8 fs/2
Frequency

Figure S6–30–III

Solution: 6.31
The solution begins by examining the filter's pole/zero plot. We find the
poles and zeros by factoring the filter's transfer function as
1 – z–2 (1 – z–1)(1 + z–1)
H(z) = 1 – z–1 = 1 – z–1 .

The two factors in the numerator of H(z), when each is set equal to zero,
produce the two z-plane zeros at z = ±1 as shown in Figure S6–31(a). The
single H(z) denominator factor, when set equal to zero, produces the pole at
z = 1. The pole-zero pair at z = 1 cancel each other so the filter's equivalent
pole/zero plot is that shown in Figure S6–31(b). A filter having the pole/zero
plot of Figure S6–31(b), having no poles (no feedback), with a zero at z = –1
will have the following transfer function:

HEquiv(z) = 1 + z–1.

So the problem solution is the simplified filter, implementing

HEquiv(z), shown in Figure S6–31(c).

H(z) z-plane Equivalent HEquiv(z) z-plane

1 1
Imaginary part
Imaginary part

0 0

–1 –1
–1 0 1 –1 0 1
Real part Real part
(a) (b)
Solution:
Simplified HEquiv(z)
x(n) y(n)

(c)
z–1

Property of Pearson Education. Not permissible for redistribution.

Figure S6–31
 (Of course the perceptive student may have canceled the common factors in
the numerator and denominator of H(z) and quickly produced HEquiv(z) and
the simplified filter structure.)

Solution: 6.32
First, we need expressions for the locations of the two poles in terms of
coefficients ap1 and ap2. The transfer function of the IIR notch filter is
Y(z) 1 + b(1)z–1 + z–2
H(z) = X(z) = 1 –a(1)z–1 –a(2)z–2 .

Factoring the denominator of H(z), we can write

1 + b(1)z–1 + z–2
H(z) = (1 + a z–1)(1 + a z–1) .
p1 p2

Setting those H(z) denominator factors equal to zero allows us to specify the
filter's pole locations in terms of coefficients ap1 and ap2. Doing so, we write
(1 + ap1z–1) = 0, and (1 + ap2z–1) = 0 ,
which specifies two poles, p0 and p1, located at
p0 = –ap1, and p1 = –ap2 .
Values –ap1 and –ap2 are complex numbers. OK, from the original problem
specification, we know that
p0 = Rejωn, and p1 = Re–jωn ,
defining coefficients ap1 and ap2 as
ap1 = –Rejωn and ap2 = –Re–jωn .
We can now write H(z) as
1 + b(1)z–1 + z–2
H(z) =
(1 –Rejωnz–1)(1 –Re–jωnz–1)

1 + b(1)z–1 + z–2 1 + b(1)z–1 + z–2

= = .
1 –Re–jωnz–1 –Rejωnz–1 + R2z–2 1 –R(e–jωn + ejωn)z–1 + R2z–2

Recalling Euler's identity: (ejθ + e–jθ) = 2cos(θ), we write

1 + b(1)z–1 + z–2
H(z) = .
1 –2Rcos(ωn)z–1 + R2z–2

H(z) is now in the standard form allowing us to identify the IIR notch filter's
feedback coefficients. So the solutions to this problem are:

a(1) = 2Rcos(ωn),
a(2) = –R2 .
The IIR notch filter's structure is shown in Figure S6–32.
Property of Pearson Education. Not permissible for redistribution.
 IIR biquad, Direct Form II
x(n) y(n)

z–1

2Rcos(ωn) –2cos(ωn)
z–1

–R2

Figure S6–32

Solution: 6.33
(a) We determine the filter's stability by determining the locations of the filter's
H(z) transfer function zeros on the z-plane. H(z) is found from the filter's
difference equation of
y(n) = x(n) + 0.5y(n–1) – 0.81y(n–2).
Taking the z-transform of the above expression, rearranging terms, and
solving for H(z) = Y(z)/X(z), we have
Y(z) 1
H(z) = X(z) = 1 – 0.5z–1 + 0.81z–2 .

Multiplying H(z) by z2/z2, yielding the desired form with z having positive-
only exponents, we have
z2
H(z) = z2 – 0.5z1 + 0.81 .

Using the quadratic factorization formula, we factor the denominator's 2nd-

order polynomial to yield
z2 – 0.5z + 0.81 = (z – 0.25 – j0.8646)(z – 0.25 + j0.8646).

The first factor is zero when z = p0 = 0.25 + j0.865, and the second pole
location is z = p1 = 0.25 – j0.8646. The poles are each located inside the z-
plane's unit circle at a radius of 0.9 as shown in Figure S6–33, therefore

The filter is unconditionally stable.

Property of Pearson Education. Not permissible for redistribution.

 Imag.
z-plane p0

Radii of poles
are 0.9.
ωr
–ωr Real

1
–1
p1

Figure S6–33

(b) We determine the range of negative values of A for which the filter will be
stable by examining H(z)'s denominator polynomial in terms of the A
coefficient. That expression is
z2 – 0.5z – A.
Using the quadratic factorization formula, we factor the above 2nd-order
polynomial showing the two filter poles located on the z-plane at
1 1
p0 = 0.25 + 16 + A and p1 = 0.25 – 16 + A .
We can write:

A p0 p1 Radii of Poles
0 0.5 0 0.5, 0
–1/32 0.4268 0.0732 0.4268, 0.0732
–1/16 0.25 0.25 0.25
–1/8 0.25 + j0.25 0.25 –j0.25 0.3536
–1/4 0.25 + j0.433 0.25 –j0.433 0.5
–1/2 0.25 + j0.6614 0.25 –j0.6614 0.7071
–3/4 0.25 + j0.8292 0.25 –j0.8292 0.866
–1 0.25 + j0.9682 0.25 –j0.9682 1.0

When A is in the range –1/16 ≤ A ≤ 0, the filter is stable because the two
poles lie on the z-plane's real axis at radii that do not exceed 0.5. When
A < –1/16 the poles move off the real axis and become conjugate poles with
equal radii.

The p0 pole in the upper portion of the z-plane is located at

1 1 1
p0 = 0.25 + 16 + A = 0.25 + –j2( 16 + A) = 0.25 + j –A – 16 .

When A ≤ –1/16 the magnitude of p0 is

1
|p0|A ≤ –1/16 = 0.252 – A – 16 = |A| .
Property of Pearson Education. Not permissible for redistribution.
 So the filter is stable so long as the square root of |A| is less than one, and we
can write:
Guaranteed stable when: –1 < A ≤ 0.

(c) From the above analysis, the magnitude of p0 is unity when |A| is unity, and
we write:
Conditionally stable when: A = –1.

(d) The positive resonant frequency of the filter, in terms of fs, is found from the
angle (shown in Figure S6–33) of the p0 pole when A = –0.81. From the
above p0 = 0.25 + j0.8646 we find angle ωr using
⎛ 0.8646 ⎞
ωres = tan–1⎜ 0.25 ⎟ = 1.289 radians.
⎝ ⎠
Because an angle of 2π radians corresponds to fs Hz, the filter's cyclic
resonant frequency is found using

ωresfs 1.289fs
fres = = 6.283 = 0.205fs Hz.
2π

(Plowing through this filter analysis makes us appreciate the value of

commercially-available digital filter analysis software!)

Solution: 6.34
Given one of the filter's zeros is z0 = 0.5657 + j0.5657 = 0.8ejπ/4, and the
filter has real-valued coefficients, there must be a complex conjugate zero
whose value is z1 = 0.5657 – j0.5657 = 0.8e–jπ/4. Because the filter is linear-
phase, the z0 and z1 zeros must be accompanied by two zeros having the
same angles with magnitudes equal to 1/0.8 = 1.25. Those two zeros are
shown as z2 and z3 in Figure S6–34.

The solutions to this problem are the following four z-domain zeros:
z0 = 0.5657 + j0.5657 = 0.8ejπ/4 (Given)
z1 = 0.5657 – j0.5657 = 0.8e–jπ/4
z2 = 0.8839 + j0.8839 = 1.25ejπ/4
z3 = 0.8839 – j0.8839 = 1.25e–jπ/4.

Property of Pearson Education. Not permissible for redistribution.

 1
z0 = 0.8ejπ/4

Imaginary part
z2 = 1.25ejπ/4
0 α
–α
z3 = 1.25e–jπ/4
z1 = 0.8e–jπ/4
–1
–1 0 1 2
Real part

Figure S6–34

Solution: 6.35
(a) The two-stage bandpass IIR filter will detect the B4 musical note (approximately 493
Hz), when the sample rate is 8000 samples/second. We determine this by finding the
pole locations of the two 2nd-order IIR filters by factoring the denominators of their
transfer functions. From the text's Eq. (6–25), the transfer functions are
0.1032 –0.1837z–1 + 0.1032z–2 0.1032z2 –0.1837z + 0.1032
H1(z) = 1 –1.8275z–1 + 0.9834z–2 = z2–1.8275z + 0.9834
and
0.3034 –0.5768z–1 + 0.3034z–2 0.3034z2 –0.5768z + 0.3034
H2(z) = –1
1 –1.8462z + 0.9843z –2 = z2–1.8462z + 0.9843 .

Using the quadratic factorization formula from the text's Eq. (6–15), we
have
First filter denominator polynomial: z2–1.8275z + 0.9834

= (z –0.9138 –j0.3853)(z –0.9138 + j0.3853).

Thus the first filter has two poles at z = 0.9138 + j0.3853, and z = 0.9138 –
j0.3853.

Likewise for the second filter, we have

Second filter denominator polynomial: z2–1.8462z + 0.9843

= (z –0.9231 –j0.3636)(z –0.9231 + j0.3636).

Similarly, the second filter has two poles at z = 0.923 + j0.3636, and
z = 0.9231 –j0.3636. The very similar pole/zero plots for the two filters are
shown in Figure S6–35.

Property of Pearson Education. Not permissible for redistribution.

 Imag. Imag.
First Pole at Second Pole at
filter z = 0.9138 +j0.3853 filter z = 0.9231 +j0.3636

α1 α2
Real Real

-1 1 -1 1

Figure S6–35

The angles of the filters' poles determine the resonant frequency of each
filter. The angle of the first filter's positive-frequency pole is
α1 = 0.3991 radians.
Multiplying α1 by 8000/2π yields a resonant frequency of 508.1 Hz. The
angle of the second filter's positive-frequency pole is
α2 = 0.3752 radians.
Multiplying α2 by 8000/2π yields a resonant frequency of 477.7 Hz. The
average of these two resonant frequencies is our desired fc resonant
frequency of the cascaded filter. So the solution is:

508.1 + 477.7
fc = 2 = 493 Hz. [Musical note B4]

(b) Yes, the two 2nd-order IIR filters are stable. The first filter's pole
magnitudes are 0.9917, and the second filter's pole magnitudes are 0.9921.
Because these pole magnitudes are less than unity (inside the unit circle),
both filters are stable.

Solution: 6.36
There are two correct solutions to this problem. The transfer function of the IIR
filter is:
B + Bz −1
H ( z) = ,
1 − Az −1
which can be re-written as:
1 + z −1
H ( z) = ⋅B.
1 − Az −1
The above H(z) gives us one solution shown in Figure S6–36(a) requiring only
two multiplies per filter output sample. Because they are linear, the factor B and
the ratio of polynomials factor in H(z) can be swapped in order to give us the
second correct solution, shown in Figure S6–36(b), requiring only two
multiplies Property of Pearson
per filter output sample.Education. Not permissible for redistribution.
 Solution:
x(n) y(n) x(n) y(n)

z–1 B B z–1

A A
(a) (b)

Figure S6–36

Solution: 6.37
The transposed structure of a traditional 3-coefficient tapped-delay line FIR
filter is given Figure S6–37.

Solution:
x(n)

h(2) h(1) h(0)

y(n)
z–1 z–1

Figure S6–37

Solution: 6.38
The transposed structure of Network I is given Figure S6–38(a). An equivalent
depiction is provided in Figure S6–38(b). The transposed structure of Network
II is given Figure S6–38(c).

Solution:
Transposed Transposed Network I
Network I (alternate depiction)
y(n)
x(n)
z–1 A C B
z–1
x(n)
y(n)
C B A
(a) (b)

x(n) Transposed Network II y(n)

Note the
A minus sign!
–
z–1
(c)

Figure S6–38
Property of Pearson Education. Not permissible for redistribution.
Solution: 6.39
(a) Starting at time index n = 0, we compute the values of the internal filter
nodes as shown in Table S6–1:

Table S6–1 Internal filter values when q = 0.05.

Time x(n) B(n) C(n) A(n) y(n)
index n
0 1.0 0 0 1.0 1.0
1 0 1.0 –0.5 –0.5 –0.5
2 0 –0.5 0.25 0.25 0.25
3 0 0.25 –0.125 –0.125 –0.15
4 0 –0.15 0.075 0.075 0.1
5 0 0.1 –0.05 –0.05 –0.05
6 0 –0.05 0.025 0.025 0.05
7 0 0.05 –0.025 –0.025 –0.05
8 0 –0.05 0.025 0.025 0.05
9 0 0.05 –0.025 –0.025 –0.05
10 0 –0.05 0.025 0.025 0.05

Given the above y(n) impulse response output, the problem solution is
shown in Figure S6–39.

Solution:
10
Quantized y(n) with rounding to nearest q = 0.05,
when x(n) = 1, 0, 0, 0, 0, ...
5
y(6) = 0.05
0
y(7) = –0.05
–5
0 5 n 10 15

Figure S6–39

(b) Comparing the problem's given impulse responses and the above Figure S6–
39, we can state:
The peak-to-peak amplitude of the limit cycles is equal to
twice the value of the rounding precision factor q.

Solution: 6.40
The impulse response of the hCas(k) cascaded combination filter is the
convolution of the h1(k) and h2(k) impulse responses of the two filters.

Solution: 6.41
Property of Pearson Education. Not permissible for redistribution.
 (a) The first part of the problem is solved as follows: We define the maximum
and minimum passband gain with the (1 + R) and (1 – R) values,
respectively, as shown in Figure S6–41.

|H(ω)|
1+R
1 P = 2 dB
1–R

Linear
scale
0 fpass Freq

Figure S6–41

The linear passband peak-peak ripple is (1 + R) – (1 – R). That peak-peak

ripple expressed in dB is
20log10(1 + R) – 20log10(1 – R) = 2 dB.
Now we're on our way. Next, recalling the property: log(x) –
log(y) = log(x/y), we write
⎛1+R⎞
20log10⎜ 1 – R ⎟ = 2,
⎝ ⎠
followed by
⎛1+R⎞ 2
log10⎜ 1 – R ⎟ = 20 = 0.1.
⎝ ⎠
Next, using the law of logarithms, we write
1+R 0.1
1 – R = 10 .
Solving for R, associated with our P = 2 dB value, we have

100.1 – 1 1.259 – 1 0.259

R|P=2 = 1 + 100.1 = 1 + 1.259 = 2.259 = 0.1147

(b) The general equation that defines the linear R deviation parameter as a
function of the logarithmic peak-peak passband ripple parameter P is taken
from the above derivation, and is

10P/20 – 1
R(P) = 1 + 10P/20 .

Solution: 6.42
The overall frequency response of cascaded and parallel filters was
discussed in Section 6.8.1 of this chapter. The solution to this problem is:
Property of Pearson Education. Not permissible for redistribution.
 Hcombination(ω) = H1(ω)H2(ω)[H3(ω) + H4(ω)]
as indicated in Figure S6–42 below.

H1(ω)H2(ω)[H3(ω) + H4(ω)]

H3(ω)
x(n) y(n)
H1(ω) H2(ω) +

H4(ω)

H1(ω)H2(ω) H3(ω) + H4(ω)

Figure S6–42

Solution: 6.43
(a) The feedback system is shown in Figure S6–43(a).

x(n) y(n)
+ A(z)
–

B(z)

(a)

Original system Simplified

x(n) y(n) representation
+ A(z) B(z)
– – x(n) y(n)
+ A(z)B(z)
–
C(z)

C(z) + D(z)
D(z)

(b) (c)

Figure S6–43

By inspection of Figure S6–43(a) we write

Y(z) = [X(z) – Y(z)B(z)]A(z)
or
Y(z)[1 + A(z)B(z)] = X(z)A(z).
Solving the above expression for Y(z)/X(z) we obtain our desired H(z) of:
Y(z) A(z)
H(z) = X(z) = 1 + A(z)B(z) .

Property of Pearson Education. Not permissible for redistribution.

 (b) We redraw the original system, shown in Figure S6–43(b), using the
principles of cascaded and parallel subsystems, to arrive at the simplified
network structure shown in of Figure S6–43(c). Notice how the simplified
system in Figure S6–43(c) has the same form as the system in Figure S6–
43(a). The transfer function of the simplified system can be determined,
then, by inspection to be:
Y(z) A(z)B(z)
H(z) = X(z) = 1 + A(z)B(z)[C(z) + D(z)] .

By substituting ejω for z in the A(z) and B(z) polynomials the desired
expression for the H(ω) frequency response of the system is:
Y(ω) A(ω)B(ω)
H(ω) = = .
X(ω) 1 + A(ω)B(ω)[C(ω) + D(ω)]

Solution: 6.44
(a) As discussed in the text, the impulse response of a parallel combination of
subfilters is the sum of the individual subfilters' impulse responses. For this
problem we write
hPar(k) = h(k) + [–hHigh(k)].
Solving for h(k), we have
h(k) = hPar(k) + hHigh(k).
Given the following sample values for hPar(k) and hHigh(k),

hPar(k) = –hHigh(0), –hHigh(1), ..., –hHigh(5), –hHigh(6) + 1,

–hHigh(7), ..., –hHigh(11), –hHigh(12)

hHigh(k) = hHigh(0), hHigh(1), ..., hHigh(5), hHigh(6), hHigh(7), ..., hHigh(11), hHigh(12)

we write our desired h(k) = hPar(k) + hHigh(k) as

h(k) = 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0.

(b) The parallel lowpass filter network, showing the h(k) subfilter being a delay
line whose length is six samples, is shown in Figure S6–44.

Solution:
z–6
x(n) y(n)

–hHigh(k)

Figure S6–44

Property of Pearson Education. Not permissible for redistribution.

Solution: 6.45
The original filter has a Direct Form II-like structure (feedback operations
precede the feedforward operations), so the two acceptable solutions to this
problem are the scaled filters shown in Figure S6–45.

Solution:
x(n) y(n)
–
G2 z–1 z–M G1
–
z–1 z–M
(a) –
A/G2 G1B
z–1 z–M

1/G2 G1

x(n) y(n)
–
G2 z–1 z–M G1
–
(b) z–1 z–M
–
A G1B
z–1 z–M

G1

Figure S6–45

Solution: 6.46
(a) The impulse invariance Method 2 integrator design proceeds as follows:

Step 1: The Laplace s-domain transfer function of the analog integrator is

given as:
1
H ( s) = .
s
Step 2: For the digital integrator the fs sample rate is fs = 0.5 Hz, so the
sample period is ts = 2.

Step 3: The analog integrator's H(s) transfer function is already in the form
of individual single pole filters,
1
H (s) = where p1 = 0,
s + p1
so Step 3 can be skipped.

Property of Pearson Education. Not permissible for redistribution.

 Step 4: Substituting 1–e–p1tsz–1 for s + p1 in H(s), when p1 = 0, yields the
desired impulse invariance z-domain transfer function of

1 1
H ii ( z ) = = .
1 − e0 z −1 1 − z −1

(b) The bilinear transform integrator design proceeds as follows:

Step 1: As before, H(s) = 1/s.

Step 2: As before, the sample period is ts = 2.

Step 3: Substituting
2 ⎛ 1 − z −1 ⎞
⋅
ts ⎜⎝ 1 + z −1 ⎟⎠
for s in H(s), when ts = 2, yields the desired bilinear transform z-domain
transfer function of

1 1 + z −1
H blt ( z ) = = .
1 − z −1 1 − z −1
1 + z −1

(c) The H(s) analog integrator has an s-plane pole at s = 0 corresponding to a

frequency of zero Hz. Setting their denominators equal to zero, we see
that the Hii(z) and Hblt(z) digital integrators each have a z-plane pole at
z = 1 corresponding to a frequency of zero Hz.

(d) We recommend using the bilinear transform-designed integrator because

its frequency magnitude response at fs/2 Hz, unlike the impulse
invariance-designed integrator, is less than the specified value of 0.01.
We prove this by evaluating Hii(z) and Hblt(z) at z = –1 (fs/2 Hz) as:

1 1 1
H ii ( z ) = = = , unacceptable gain at fs/2,
z =−1 1− z −1
1 − (−1) 2

1 + z −1 1 −1
H blt ( z ) = =
−1 1 − (−1)
= 0 , acceptable gain at fs/2.
z =−1 1− z

The frequency magnitude responses of the two digital integrators are

illustrated in Figure S6–46.

Property of Pearson Education. Not permissible for redistribution.

 4
|Hii(f)|
3

Mag (linear)
|Hblt(f)|
2

0
–fs/2 0 Freq fs/2

Figure S6–46

Solution: 6.47
(a) Given the analog filter's transfer function
1
H (s) = ,
1 + RCs
using the impulse invariance Method 2 process we skip to Method 2 Step 3
of the design process. Because the filter is 1st-order having one path (no
parallel paths) and no partial fraction expansion is necessary, from Method 2
Step 3 we write
A1 1/ RC
H ( s) = = ,
s + p 1 s + 1/ RC

where A1 = p1 = 1/RC. From Method 2 Step 4, with ts = 1, we write our

desired transfer function as
A1 1/ RC
H ii ( z ) = = .
1− e
−p1
z −1 1 − e−1 / RC z −1
Two equivalent Direct Form II block diagrams of the Hii(z) digital filter are
shown in Figure S6–47(a). Note: the feedback coefficients in the figure are
the negative of the z–1 coefficient in the denominator of Hii(z).

Solution:
Hii filter
x(n) y(n) x(n) y(n)

z–1 1 1 z–1
(a) RC RC

e–1/RC e–1/RC

Figure S6–47

(b) Using the Bilinear transform, with ts = 1, we substitute

2 ⎛ 1 – z–1 ⎞
1 ⎜⎝ 1 + z–1 ⎟⎠
Property of Pearson Education. Not permissible for redistribution.
 for s in H(s) yielding the z-domain transfer function
1 1
H bt ( z ) = = ,
⎛ 1 − z −1 ⎞ ⎛ 1 − z −1 ⎞
1 + RC ⎜ 2 ⋅ ⎟ 1+ β ⎜ ⎟
⎜ 1 + z −1 ⎟ ⎜ 1 + z −1 ⎟
⎝ ⎠ ⎝ ⎠

where β = 2RC. Simplifying H(z), we write the desired transfer function as

1 + z −1 1 + z −1 1 1 + z −1
H bt ( z ) = = = ⋅
1 + z −1 + β(1 − z −1 ) 1 + β + (1 − β) z −1 1 + β 1 + ⎛ 1 − β ⎞ z −1
⎜ ⎟
⎝ 1+ β ⎠
or

1 1 + z −1
H bt ( z ) = ⋅ .
1 + 2 RC 1 + ⎛ 1 − 2 RC ⎞ z −1
⎜ ⎟
⎝ 1 + 2 RC ⎠

Three equivalent Direct Form II block diagrams of the Hbt(z) digital filter
are shown in Figures S6–47(b) and S6–47(c).

Solution:
Hbt filter
x(n) y(n) x(n) y(n)

z–1 1 z–1 1
1 + 2RC 1 + 2RC
(b)

2RC – 1 1 2RC – 1
1 + 2RC 1 + 2RC 1 + 2RC

Hbt filter
x(n) y(n)

1 z–1
(c) 1 + 2RC

2RC – 1
1 + 2RC

Figure S6–47 (Cont'd)

(c) The |Hbt(z)| response has large attenuation at high frequencies because
it has a z-plane zero at z = –1 (infinite attenuation at fs/2 Hz).

Solution: 6.48
Given the analog filter's Laplace-domain transfer function of
s
H ( s) = ,
s +
Property of Pearsonωo Education. Not permissible for redistribution.
 using the bilinear transform we substitute
2 ⎛ 1 – z–1 ⎞
ts ⎜⎝ 1 + z–1 ⎟⎠
for s in H(s) yielding the z-domain transfer function
2 1 − z −1
⋅
ts 1 + z −1
H ( z) = .
2 1 − z −1
⋅ + ωo
ts 1+ z −1
Multiplying H(z)'s numerator and denominator by ts·(1 + z–1), we have

H ( z) =
( )
2 1 − z −1
.
2 (1 − z −1 ) + ts (1 + z −1 ) ω o

Dividing the numerator and denominator terms of H(z) by 2 and combining

like terms, we have
1 − z −1
H ( z) = .
(1 + 0.5ts ωo ) − (1 − 0.5ts ωo ) z −1
To arrive at the desired form for H(z), we divide the numerator and
denominator terms of H(z) by (1 – 0.5tsωo), yielding

H ( z) =
1
1 + 0.5ts ωo (
⋅ 1 − z −1
.
)
1 − 0.5ts ωo −1
1− z
1 + 0.5ts ωo

With ts = 1/fs = 1/100 Hz = 0.01, and ωo = 62.832 radians/second, we have

the desired highpass filter transfer function of

0.761 − 0.761z −1
H ( z) = .
1 − 0.522z −1

Solution: 6.49
Given the analog filter's transfer function
5
H (s) = ,
s( s − 0.8)
using the bilinear transform we substitute
2 ⎛ 1 – z–1 ⎞ –1
⎛1–z ⎞
⎜ –1 ⎟ = a ⎜
ts ⎝ 1 + z ⎠ –1 ⎟, where a = 2/ts,
⎝1+z ⎠
for s in H(s) yielding the z-domain transfer function

Property of Pearson Education. Not permissible for redistribution.

 5
H(z) = 1 – –1 –1 .
⎛ z ⎞⎛ 1–z ⎞
a⎜ 1 + z–1 ⎟ ⎜a 1 + z–1 – 0.8⎟
⎝ ⎠⎝ ⎠

5
= 1 – –1
1 – –1 –1 .
⎛ z ⎞⎛ z ⎞ ⎛ 0.8a(1 – z ) ⎞
a2⎜ 1 + z–1 ⎟ ⎜ 1 + z–1 ⎟ – ⎜ 1 + z–1 ⎟
⎝ ⎠⎝ ⎠ ⎝ ⎠

Multiplying the numerator and denominator of H(z) by (1 + z–1)2, we have

5(1 + z–1)2
H(z) = a2(1 – z–1)2 – 0.8a(1 + z–1)(1 – z–1)

5(1 + 2z–1 + z–2) 5 + 10z–1 + 5z–2

= a2(1 – 2z–1 + z–2) – 0.8a(1 – z–2) = a2 –2a2z–1 + a2z–2 – 0.8a + 0.8az–2

5 + 10z–1 + 5z–2
= (a2 – 0.8a) –2a2z–1 + (a2 + 0.8a)z–2 .

Substituting 2/ts = 2/0.001 = 2x103 for a, we have a primary solution of:

5 + 10z–1 + 5z–2
H(z) = 3.9984x106 –8x106z–1 + 4.0016x106z–2 .

To model filter transfer functions using commercial signal processing

software we can divide the numerator and denominator of H(z) by
3.9984x106 to make the denominator's first coefficient equal to 1. Doing so
yields an equivalent secondary solution of:
0.12505x10–5 + 0.2501x10–5z–1 + 0.12505x10–5z–2
H(z) = 1 –2.0008z–1 + 1.0008z–2 .

Solution: 6.50
(a) Given that the analog lowpass filter's cutoff frequency is fa = 3.8 kHz, the
digital filter's fd cutoff frequency is found by mapping fa to fd using

fs tan–1(πfa/fs ) (11000) tan–1(3800π/11000 )

fd = =
π π
or
fd = 2.893 kHz.
(b) To achieve a digital lowpass filter cutoff frequency of fd = 3.8 kHz, the new
analog lowpass filter's cutoff frequency is found by mapping fd to fa using

fs tan(πfd/fs ) (11000) tan(3800π/11000 )

fa = =
π π
or
fa = 6.636Education.
Property of Pearson kHz. Not permissible for redistribution.
 Chapter 7 Solutions
Solution: 7.1
First-difference Differentiator: from the first-difference differentiator's time-
domain difference equation of yFd(n) = x(n) – x(n–1), the differentiator's z-
domain transfer function is

YFd(z)
HFd(z) = X(z) = 1– z–1.

To determine the differentiator's frequency response, we substitute ejω for z in

HFd(z), yielding

HFd(ω) = 1 – e–jω = 1 –cos(ω) + jsin(ω).

The differentiator's frequency magnitude response is the square root of the sum
of HFd(ω)'s real and imaginary parts squared, or

|HFd(ω)| = [1 – cos(ω)]2 + sin2(ω) = 1 – 2cos(ω) + cos2(ω) + sin2(ω) .

Using the trigonometric identity cos2(α) + sin2(α) = 1, we can write

4 – 4cos(ω) 1 – cos(ω)
|HFd(ω)| = 2 – 2cos(ω) = 2 =2 2 .

Using the half-angle trigonometric identity sin(α / 2) = ± [1 − cos(α)]/ 2 , we

can write the solution to this part of the problem as

|HFd(ω)| = ±2sin(ω/2) = 2|sin(ω/2)|.

Central-difference Differentiator: from the central-difference differentiator's

time-domain difference equation of yCd(n) = [0.5x(n) – 0.5x(n–2)], the
differentiator's z-domain transfer function is

YCd(z)
HCd(z) = X(z) = 0.5 – 0.5z–2.

To determine the differentiator's frequency response, we substitute ejω for z in

HCd(z), yielding

HCd(ω) = 0.5 – 0.5e–j2ω = 0.5[1 – cos(2ω) + jsin(2ω)].

The differentiator's frequency magnitude response is the square root of the sum
of HCd(ω)'s real and imaginary parts squared, or
2 2
|HCd(ω)| = [0.5
Property – 0.5cos(2ω)]
of Pearson + 0.25sin
Education. Not (2ω)
permissible for redistribution.
 = 0.25 – 0.5cos(2ω) + 0.25cos2(2ω) + 0.25sin2(2ω) .

Using the trigonometric identity cos2(α) + sin2(α) = 1, we can write

1 – cos(2ω)
|HCd(ω)| = 0.5 – 0.5cos(2ω) = 2 .

Using the half-angle trigonometric identity sin(α/2) = ±([1 –cos(α)]/2)1/2, we

can write the solution to this part of the problem as

|HCd(ω)| = ±sin(ω) = |sin(ω)|.

Solution: 7.2
Two block diagrams of a central-difference differentiator having only one
multiplier are shown in Figure S7–2.

Solution:
Central-difference differentiators
x(n) x(n)
z–1 z–1 z–1 z–1
– –
0.5
ycd(n) ycd(n)

0.5

Figure S7–2

Solution: 7.3
Rocky would be incorrect because the web page
statement is not true!
Solution Method# 1: The impulse response of two cascaded first-difference
differentiators is a first-difference differentiator's impulse response
[hFd(n) = [1,–1]) convolved with itself, or

hFd,cascaded(n) = [1,–2,1].

Because the impulse response of a central-difference differentiator

(hCd(n) = [0.5,–0.5]) is not equal to hFd,cascaded(n),
the original Internet statement is not true. The lesson we learn
here is: Do not believe everything you read on the Internet!

Solution Method# 2: Here's a frequency-domain solution: the first-difference

differentiator's time-domain difference equation is yFd(n) = x(n) – x(n–1), and its
z-domain transfer function is
Property of Pearson Education. Not permissible for redistribution.
 YFd(z)
HFd(z) = X(z) = 1– z–1.

To find that differentiator's frequency response, we substitute ejω for z in HFd(z),

giving us
HFd(ω) = 1 – e–jω.
Now, from the text's Section 6.8 we know that two first-difference
differentiators in cascade (in series) will have a frequency response equal to the
product of their individual frequency responses, or
HFd,cascaded(ω) = HFd(ω).HFd(ω) = [1 – e–jω ].[1 – e–jω]

= 1 – 2e–jω + e–j2ω.
Next, obtained in a similar manner, the central-difference differentiator's
frequency response is
HCd(ω) = 0.5 – 0.5e–j2ω.
Because HFd,cascaded(ω) ≠ HCd(ω),
again, the original Internet statement is not true.

Solution: 7.4
We can estimate the acceleration of the motor shaft by first computing the
derivative of the Apos(n) shaft position signal to obtain the Avel(n) motor shaft
velocity signal as shown in Figure S7–4–I(a). Next, we compute the derivative
of Avel(n) to obtain the desired Aacc(n) acceleration signal. Thus our Acceleration
Measurement Network is the cascade of two digital differentiators as shown in
Figure S7–4(a).
Due to the high-frequency noise in the Apos(n) signal, we eliminate first-
difference differentiators as possible solutions. Due to the restriction of performing
no more than one multiplication per Apos(n) input sample both the text's Lanczos
and wideband differentiators are eliminated, leaving only central-difference
differentiators as possible solutions as shown in Figure S7–4(b).

Apos(n) Avel(n) Aacc(n)

Differentiator Differentiator
(a)

Apos(n)
z–1 z–1 z–1 z–1

(b) + – + –
Aacc(n)
Avel(n)
0.5 0.5

Figure S7–4
Property of Pearson Education. Not permissible for redistribution.
 The two multipliers in Figure S7–4(b) can be combined into one multiplier giving
us the two equivalent correct solutions shown in Figures S7–4(c) and (d).

Solutions:
Apos(n)
z–1 z–1 z–1 z–1

(c) + – + –
Aacc(n)
Avel(n)
0.25

Apos(n)
z–1 z–1 z–1 z–1

(d) + – + –
Aacc(n)
Avel(n)
0.25

Figure S7–4 (Cont'd)

Solution: 7.5
The student should use the text's Eq. (7–14),
(–1)k
h(k)ωc=π = k ,
to solve this wideband differentiator design problem. Doing so, with –3≤k≤3 and k ≠ 0,
using a table would produce

k hωc=π(k)
–3 (–1)–3
–3 = 0.333
–2 (–1)–2
–2 = –0.5
–1 (–1)–1
–1 = 1
0 0 (set to 0 by
design)
1 (–1)1
1 = –1
2 (–1)2
2 = 0.5
3 (–1)3
3 = –0.333

Because N = 7 is odd we set the center coefficient, h((N–1)/2) = h(3), to zero.

Property
Thus our solution is:of
thePearson Education.
7-tap wideband Not permissible
differentiator's for redistribution.
coefficients are:
 h(k) = 0.333, –0.5, 1, 0, –1, 0.5, –0.333.

Note: If the student uses the text's more general Eq. (7–13),
ωccos(ωck) sin(ωck)
hgen(k) = – ,
πk πk2
that equation reduces to
πcos(πk) sin(πk) cos(πk)
hgen(k) = – 2 =
πk πk k
yielding the desired coefficients of
hgen(k) = 0.333, –0.5, 1, 0, –1, 0.5, and –0.333.
(Again, because N = 7 is odd we must set the center coefficient, hgen(3), to
zero.)

Solution: 7.6
From Section 5.10 we learned that the group delay, measured in samples, of a
tapped-delay line filter network having antisymmetrical coefficients is found
using

D
Gdiff = 2 samples,

where D is the number of unit-delay elements in the filter's delay line. The block
diagram (structure) of the ydiff(n) differentiator is shown in Figure S7–6(a).
Redrawing that figure to show the individual unit-delay elements (z–1) gives us
Figure S7–6(b) showing the differentiator to have D = 6 unit-delay elements.
Thus the solution to this problem is

6
group delay Gdiff = 2 = 3 samples, which
is an integer number of samples.

x(n) x(n–2) x(n–4) x(n–6)

z–2 z–2 z–2
–1/16 1/16
(a)
–
ydiff(n)

x(n)
z–1 z–1 z–1 z–1 z–1 z–1
–1/16 1/16
(b) –
ydiff(n)

Property of Pearson Education. Not permissible for redistribution.

Figure S7–6
Solution: 7.7
We determine the validity of the problem statement by finding the expression
for the real part of the HRe(ω) frequency response of the rectangular rule
integrator, given by:
1
H Re (ω) = .
1 − e − jω
Multiplying the numerator and denominator of HRe(ω) by ejω, we can write:
e jω
H Re (ω) = jω .
e −1
Using Euler's identity, ejω = cos(ω) + jsin(ω), we write HRe(ω) in the rectangular
format of:
cos(ω) + j sin(ω)
H Re (ω) = .
cos(ω) − 1 + j sin(ω)
Next from Eq. (A–20), in Appendix A, we write the expression for the ratio of
two complex numbers as:
cos(ω)[cos(ω) − 1] + sin(ω) sin(ω) + j {[cos(ω) − 1]sin(ω) − cos(ω) sin(ω)}
H Re (ω) = .
[cos(ω) − 1]2 + sin 2 (ω)
The real part of HRe(ω) is:
cos(ω)[cos(ω) − 1] + sin(ω) sin(ω)
Real[ H Re (ω)] = .
[cos(ω) − 1]2 + sin 2 (ω)

cos 2 (ω) − cos(ω) + sin 2 (ω)

= .
cos 2 (ω) − 2 cos(ω) + 1 + sin 2 (ω)
Using the trig identity cos2(ω) + sin2(ω) = 1, we write:
1 − cos(ω) 1
Real[ H Re (ω)] = = .
2 [1 − cos(ω)] 2

So the real part of HRe(ω) is equal to 0.5 for all ω, which validates the
problem's statement.

Solution: 7.8
(a) Given the x(n) input, shown in Figure S7–8(a), the integrator's y(n) output
sequence is that shown in Figure S7–8(b).

x(n)
1 ...
(a) 3 5 7
0
0 1 2 4 6 8 n
Property
–1 of Pearson Education. Not permissible for redistribution.
 Solution:
y(n)
(b) 1
...
0
0 1 2 3 4 5 6 7 8 n

Figure S7–8

(b) Comparing the peak-peak amplitudes of x(n) input and y(n) output
sequences we see that, for an input sinusoid whose frequency is fs/2 Hz,
the integrator has an amplitude loss by a factor of 0.5.

(c) The text's equation for the frequency response of the integrator is:

1
H Re (ω) = .
1 − e − jω

The cyclic frequency of fs/2 Hz corresponds to a normalized frequency of

ω = π in the expression for HRe(ω). So the integrator's gain factor at fs/2 Hz
is:
1 1 1
H Re (π) = − jπ
= = = 0.5
1− e 1 − (−1) 2
as illustrated in Figure S7–8(c).

|HRe(ω)|

(c)
0.5
0
0 π 2π ω
(fs/2) (fs)

Figure S7–8 (Cont'd)

Solution: 7.9
Using a trapezoidal rule integrator to estimate the area under the x(t) curve will
produce integrator output samples equal to the areas of the shaded trapezoids
shown in Figure S7–9. Those integrator output samples (areas of the shaded
trapezoids) are:

x(1) + x(0) x(2) + x(1) x(3) + x(2)

A(1) = 2 , A(2) = 2 , and A(3) = 2 .

From Figure S7–9 we see that

Property of Pearson Education. Not permissible for redistribution.
 the sum of shaded areas A(1) + A(2) + A(3) is larger than the true area
under the x(t) curve during the time interval of 0 to 3ts seconds.

x(n)
Continuous x(t)

A(1) A(3)
A(2)
0
0 1 2 3 n
(ts) (2ts) (3ts)

Figure S7–9

Solution: 7.10
(a) From Eq. (6–105) in the text's Section 6.11, to perform a bilinear
transformation we substitute:
2 1 – z–1
s = t 1 + z–1
s

for s in the problem's Hint(s) to yield the discrete integrator's z-domain

transfer function of
ts 1 + z–1
HBilin(z) = 2 1 –z–1 .

Substituting e–jω for z in the above HBilin(z) yields the desired bilinear
transform-designed integrator's frequency response of
ts 1 + e–jω
HBilin(ω) = 2 .
1 –e–jω

(b) Recognizing that the above HBilin(ω) can be written as

1 + e–jω 0.5 + 0.5e–jω
HBilin(ω) = 0.5ts = ts .
1 –e–jω 1 –e–jω
For simplicity, we assume ts = 1 and see that

the above bilinear transform-designed HBilin(ω) integrator is

equal to the text's trapezoidal rule integrator

whose frequency response is

0.5 + 0.5e–jω
HTr(ω) = .
1 –e–jω

Solution: 7.11

Property of Pearson Education. Not permissible for redistribution.

 (a) The rectangular rule integrator is the most computationally-simple
integrator because it requires no multiply operations, and only a single
addition, to compute each integrator output sample.

(b) To determine the z-plane pole/zero characteristics of the four discrete

integrators, we write their z-domain transfer functions as
YRe(z) 1
HRe(z) = X(z) = 1 –z–1

YTr(z) 0.5 + 0.5z–1

HTr(z) = X(z) = 1 –z–1

YSi(z) 1 1 + 4z–1 + z–2

HSi(z) = X(z) = 3 1 –z–2

YTi(z) 0.3584 + 1.2832z–1 + 0.3584z–2

HTi(z) = X(z) = 1 –z–2 .

Solving for the roots of the four transfer functions' denominators, we find
that the four integrators have z-domain pole locations of:

Rectangular rule: a single pole at z = 1.

Trapezoidal rule: a single pole at z = 1.
Simpson's rule: poles at z = 1 and z = –1.
Tick's rule: poles at z = 1 and z = –1.

Thus, the answer to this problem is that

the Simpson's rule and Tick's rule integrators have a
single pole (infinite gain) at z = –1.

(c) To determine at what frequency an integrator has a zero-valued magnitude

response we must find the location of any transfer function zeros located on
its z-plane unit circle. We find the integrators' z-plane zeros by solving for
the roots of the four transfer functions' numerators. Doing so we find that
only the trapezoidal rule integrator has a z-plane zero on the unit circle,
located at z = –1. The cyclic frequency associated with z = –1 is fs/2 Hz, so
the solution to this problem is the trapezoidal rule integrator.

Solution: 7.12
(a) The block diagram of the FIR matched filter is shown in Figure S7–12.

Property of Pearson Education. Not permissible for redistribution.

 Solution:
s(n) s(n–1) s(n–2) s(n–3)
z–1 z–1 z–1
1 –1 4 2

y(n)

Figure S7–12

(b) The y(n) output sequence of the filter is:

y(n) = 2, 2, 3, 22, 3, 2, 2

Solution: 7.13
Given the problem's xs(n) signal-of-interest, and assuming the filter's
coefficients are a time-reversed version of xs(n), the maximum filter output
sample is the sum of the xs(n) samples squared. The algebra expression is:
Maximum y(n) = N–1xs(k)2 .
∑

k=0

Solution: 7.14
Using x1(n) as our signal-of-interest in a matched filter, the filter's output
sequence is:
y1(n) = [9, 30, 43, 30, 9].
The matched filter's y1(n) output SNR is:

Max y1(n) sample 43

SNR1 = Max x1(n) sample = 5 = 1.31, or

SNR1,dB = 20*log10(1.31) = 2.35 dB.

Using x2(n) as our signal-of-interest in a matched filter, the filter's output
sequence is:
y2(n) = [9, 24, 46, 64, 75, 64, 46, 24, 9].
The matched filter's y1(n) output SNR is:

Max y2(n) sample 75

SNR2 = Max x2(n) sample = 5 = 1.73, or

SNR2,dB = 20*log10(1.73) = 4.76 dB.

The problem solution is SNR2,dB – SNR1,dB, or

SNR improvement = 2.41 dB.

Property of Pearson Education. Not permissible for redistribution.

Solution: 7.15
The larger M, the narrower must be the transition region width of the
image-reject subfilter, thus
The larger M becomes, the larger must be the
number of taps in the image-reject subfilter.

Solution: 7.16
(a) The passband width of the lowpass prototype FIR filter is M = 5 times the
lowpass IFIR filter's desired passband width (4 kHz), or
Prototype filter passband width = 5·4 kHz = 20 kHz
as shown in Figure S7–16(a).

(b) Because the shaping subfilter's passband images are centered at integer
multiples of fs/M = fs/5 Hz,
the shaping subfilter will have two passband
images residing between zero and fs/2 Hz
as shown in Figure S7–16(b).

|Hp(f)|

(a)
0 20 fs/2 fs
kHz
|Hsh(f)| M=5

(b)
0 fs/5 2fs/5 3fs/5 4fs/5 fs
fs/2

Figure S7–16

Solution: 7.17
(a) The desired IFIR filter's transition bandwidth is 61.2 kHz minus 60 kHz, or
1.2 kHz. Thus the normalized transition region bandwidth, and the
normalized passband width, are:
ftrans = 1.2x103/3x106 = 0.0004, and fpass = 60x103/3x106 = 0.02.
Using those normalized ftrans and fpass frequencies in the text's Figure 7–
18(a), as shown in Figure S7–17 , yields an
optimum shaping subfilter expansion factor of M = 16.

Property of Pearson Education. Not permissible for redistribution.

 (b) The maximum passband peak-to-peak ripple requirement for the prototype
filter is half the passband ripple required of the IFIR filter. That is,
maximum passband peak-to-peak ripple 0.2/2 = 0.1 dB.

(c) Using those normalized ftrans and fpass frequencies in the text's Figure 7–
18(b), the computational reduction we expect from the IFIR filter is
roughly
92%.

Optimum expansion factor (M)

25
0.01
20 fpass = 0.02
M ≈ 16

10 0.04
0.06
5 0.08

0.0001 0.001 0.01 0.1

ftrans = 0.0004 Transition region bandwidth, ftrans

Figure S7–17

Solution: 7.18
(a) The time-domain difference equation of the complex multiplier output is

w(n) = ejω·v(n) = [cos(r) + jsin(ω)]·[vR(n) + jvI(n)]

= cos(ω)vR(n) – sin(ω)vI(n) + j[cos(ω)vI(n) + sin(ω)vR(n)].

(b) The complex multiplier block diagram, showing real-only variables, is

shown in Figure S7–18.

Solution:
+
vR(n) wR(n)
–

vI(n) w(n) =
wR(n) + jwI(n)
v(n) =
vR(n) + jvI(n) wI(n)

cos(ω) sin(ω)

e jω = cos(ω) + jsin(ω)

Figure S7–18
Property of Pearson Education. Not permissible for redistribution.
 (c) A single complex multiply requires four real multiplies and two real
additions.

Solution: 7.19
(a) From the first comb filter we see that N = 8 and r = 1. The pole/zero
plots of the cascaded comb and the resonators are given in Figures S7–19–
I(a), (b) and (c). With their pole–zero cancellations, the desired pole/zero
plot solution is shown in Figure S7–19–I(d).

Cascaded-comb zeros 1st resonator stage (k = 0) poles

1 1

Imaginary part
Imaginary part

Two
0.5 zeros 0.5
2 2 0 2
0

–0.5 Two –0.5 Two

zeros poles
–1 –1
–1 0 1 –1 0 1
Real part Real part
(a) (b)

2nd and 3rd resonator stages Solution:

(k = 1 & 2) poles Total FSF poles-zeros
1 1
Imaginary part

Imaginary part

0.5 k=2 0.5

poles
0 0 2

–0.5 k=1 –0.5

poles
–1 –1
–1 0 1 –1 0 1
Real part Real part
(c) (d)

Figure S7–19–I

(b) The filter is a lowpass filter because it has z-plane zeros

located in the vicinity of z = –1 (ω = 0 radians/sample, or
fs/2 Hz) on the unit circle as shown in Figure S7–19–I(d).

(c) This part of the problem is a bit tricky. For the three stages of the FSF to
have equal gains, the gain factors following the second and third resonators
must be twice the value of the gain factor following the first (k = 0, or zero
Hz) resonator. However all three gain factors, 1/16, are equal so the second
and third stages have half the gain as the first stage. Thus the FSF's
frequency magnitude response looks roughly as that shown in Figure S7–
19–II.

Property of Pearson Education. Not permissible for redistribution.

 Solution:
FSF magnitude response
1
0.8
0.6
0.4
0.2
0
–0.5 –0.4 –0.3 –0.2 –0.1 0 0.1 0.2 0.3 0.4 0.5
–(fs/2) ω (fs/2)
(–3fs/8) (3fs/8)

Figure S7–19–II

Solution: 7.20
The cascaded (series combination) comb filters are shown below in
Figure S7–20–I(a).

Unit sample impulse [1, 0, 0, 0, 0, . . .]

x(n) w(n) y(n)

(a)
z–8 z–2
r=1
–(18) = –1 –(12) = –1

Solution:
Cascaded-comb impulse response
1
h(n)
(b) 0

–1
0 2 4 6 8 10 12 14 16
n

Figure S7–20–I

(a) The cascaded-comb combination filter's h(n) impulse response is shown in

Figure S7–20–I(b).

(b) Yes, the cascaded-comb combination filter is linear phase because the
nonzero samples of the h(n) impulse response are symmetrical as shown in
Figure S7–20–I(b). (The cascade of two linear-phase filters will always
have linear phase.)

(c) With damping factor r = 0.9, the cascaded combination of comb filters in
Figure S7–20–II(a) will not have linear phase because the nonzero samples
Property of Pearson Education. Not permissible for redistribution.
 of their h(n) impulse response are not symmetrical. This is shown in Figure
S7–20–II(b).

Unit sample impulse [1, 0, 0, 0, 0, . . .]

x(n) w(n) y(n)

(a)
z–8 z–2
r = 0.9
–(0.98) = –0.43 –(0.92) = –0.81

Combined impulse response

1
h(n) 0.35
(b) 0
–0.81 –0.43
–1
0 2 4 6 8 10 12 14 16
n

Figure S7–20–II

Solution: 7.21
The purpose of this problem is to test the student's understanding of the
text's Figure 7–50. If we expand a portion of Figure 7–50, shown below as
Figure S7–21, we see that

a Type-IV FSF having a single non-unity transition band coefficient can

achieve stopband attenuations roughly in the range of 42 -to- 54 dB.

0
attenuation (dB)
Min stopband

256 No transition coefficients

(R ≈ 1.6 dB)
–20
64 32 16
One transition coefficient
(R ≈ 0.7 dB)
This range –40 128
32
is roughly
–42 to –54 256 64
dB. 16
–60
0 0.05fs 0.1fs 0.15fs
Transition bandwidth

Figure S7–21

Solution: 7.22
Property of Pearson Education. Not permissible for redistribution.
 Yes, it is possible to design an FSF that achieves 85 dB of stopband
attenuation.

From the text's Figure 7–50, shown here as Figure S7–22, we see that
Type-IV FSFs with three transition band coefficients can easily achieve 85
dB of stopband attenuation.

0
256 No transition coefficients R = passband
(R ≈ 1.6 dB)
Min stopband attenuation (dB)

peak-peak ripple
–20
64 32 16
One transition coefficient
(R ≈ 0.7 dB)
–40 128

256 64 32
–60 16 Two transition coefficients
128
(R ≈ 0.35 dB)
32
Three transition
–80 256 64 16 coefficients
–85 dB (R ≈ 0.16 dB)
–100 256 128 64
32
16
–120
0 0.05fs 0.1fs 0.15fs 0.2fs 0.25fs
Transition bandwidth

Figure S7–22

Solution: 7.23
(a) From Table H–4 (for odd N), in Appendix H, we use the values N = 23 and
BW = 3 to find the transition coefficient values of
H(3) = T1 = 0.64635467,

H(4) = T2 = 0.16260027, and

H(5) = T3 = 0.0077623.

(b) From the same row of Table H–4 as was used above, we can expect
an FSF stopband attenuation of roughly 95 dB.

(c) The structure (block diagram) of the six-section Type-IV FSF is shown in
Figure S7–23 below.

Property of Pearson Education. Not permissible for redistribution.

 Solution:
Type-IV resonators

k=0
0.5
k=1
x(n) – y(n)
–
k=2
z–23 z–2 0.64635467 –

k=3
–r23 –r2 0.16260027
k=4
0.0077623
k=5

Figure S7–23

Instructor: The key here is to see that the student filled in all the details such as
the comb filters' coefficients, the 0.5 gain factor for the k = 0 (DC) section,
and the correct plus and minus signs for the final summation.

Solution: 7.24
There are two ways to solve this problem; a DFT analysis method, and a z-
domain transfer function analysis method. Both solution methods are provided
below.

(a) DFT Analysis Method: Given this problem, we expect the student to begin
their solution by drawing a block diagram like that shown below in either
Figure S7–24–I(a) or Figure S7–24–I(b).

Property of Pearson Education. Not permissible for redistribution.

 FSF H(k)/N
gain factors

1
H0(z) =
1 –e j2π0/16 z–1
H(0)/16
1
H1(z) =
1 –e j2π1/16 z–1
x(n) H(1)/16 X2(n)
1 –z–16 1
H2(z) =
1 –e j2π2/16 z–1
(a)
.. .. H(2)/16
. . X2(n)X2(n)*
1
H15(z) = |X2(n)|2
j2π15/16 –1
1 –e z
H(15)/16

z–1
H(0)/16

x(n) e j2π0/16
X2(n)

z–16 z–1 H(1)/16

(b)
e j2π1/16 X2(n)X2(n)*
–1
|X2(n)|2
z–1
H(2)/16

e j2π2/16
. .
. .
. .

z–1 H(15)/16

e j2π15/16

Figure S7–24–I

The FSF's H(k) gain terms at the outputs of the resonators are the DFT of
the FIR filter's coefficients in the problem's Figure P7–24–II(a). So a key
part of the solution is determining the values for the H(k)/16 gain factors in
the above Figure S7–24–I. Based on the text's Figure 7–20 discussion that
the FSF H(k) gain terms are the DFT of the equivalent nonrecursive FIR
filter's coefficients, we find the desired H(k) gain terms by performing the
DFT of the problem's Figure P7–24–II(a) filter coefficients when m = 2.

To avoid confusion, we represent the nonrecursive filter's coefficients as

h(i), where 0 ≤ i ≤ N–1. The N-point DFT of the nonrecursive filter's
h(i) = e–j2πm(N–1–i)/N coefficients for DFT bin m = 2 is:

Property of Pearson Education. Not permissible for redistribution.

 N–1 N–1 N–1
H(k) = ∑h(i)e–j2πki/N = ∑e–j2π2(N–1–i)/Ne–j2πki/N = ∑e–j2π[2(N–1–i)+ki]/N
i=0 i=0 i=0

N–1 N–1
= ∑e –j2π[2N–2–i(2–k)]/N
= ∑e–j2π2N/Nej2π2/Nej2πi(k–2)/N .
i=0 i=0
Because e–j2π2N/N = 1, we can write the FSF's H(k) gain terms as
N–1
H(k) = ej2π(2)/N ∑e–j2πi(k–2)/N .
i=0
The summation in the above H(k) is a geometric series that can be
converted to a closed form expression (See Appendix B) as
N–1
1 – e–j2π(k–2)
H(k) = e j2π(2)/N
∑e –j2π(k–2)i/N
=e j2π(2)/N
1 – e–j2π(k–2)/N
i=0

j2π(2)/N e–jπ(k–2)[ejπ(k–2) – e–jπ(k–2)]

=e
e–jπ(k–2)/N[ejπ(k–2)/N – e–jπ(k–2)/N ]

sin[π(k–2)]
= e–jπ(k–2)(N–1)/N ej2π(2)/N .
sin[π(k–2)/N]
Substituting various values of k in H(k), to obtain the FSF's gain terms,
reveals that the sin[π(k–2)] terms are zero for all k except k = 2. So we are
left with a single-resonator FSF whose H(2) gain term is

sin[π(k–2)]
H(2) = ej2π(2)/N .
sin[π(k–2)/N]

When k = 2, the phase of H(2) is ej2π(2)/N = ej4π/16 but its magnitude is 0/0
(indeterminate). Using L'Hopital's Rule on
sin[π(k–2)]
|H(2)| = .
sin[π(k–2)/N]
yields
d{sin[π(k–2)]}/dk cos[π(k–2)] d[π(k–2)]/dk
|H(2)| = = ·
d{sin[π(k–2)/N]}/dk cos[π(k–2)/N] d[π(k–2)/N]/dk

cos[π(k–2)] πk
= · .
cos[π(k–2)/N] πk/N
When k = 2, the above |H(2)| becomes

cos(0) 2π
|H(2)| = cos(0) · = N = 16.
2π/N

So our FSF's single resonator gain term, for N = 16, is

j2π(2)/N jπ/4
H(2) = 16e
Property of Pearson = 16e
Education. Not permissible for redistribution.
yielding a final gain factor of H(2)/16 = ejπ/4. Thus the FSF block diagram
is greatly simplified to that in Figure S7–24–II, shown in both the z-domain
form and the actual FSF structure, which is the solution to this part of the
problem.

Part (a) Solutions:

x(n) Hres(z) = X2(n) |X2(n)|2

1–z–16 1 X2(n)X2(n)*
(I) 1 –e j2π2/16 z–1
16e j2π2/16/16
= e jπ/4
OR

x(n) X2(n) |X2(n)|2

X2(n)X2(n)*
–
(II) z–16 z–1 e jπ/4

e jπ/4

Figure S7–24–II

Transfer Function Analysis Method: First we compute the transfer

function of the filter in the problem's Figure P7–24–II(a), and then find the
FSF in Figure P7–24–II(b) by analogy. For m = 2, letting y(n) = Xm=2(n) we
can write the filter's difference equation as

y(n) = e–j2π(2)15/16x(n) + e–j2π(2)14/16x(n–1) + ... + e–j2π(2)0/16x(n–15)

and the filter's z-domain expression is

Y(z) = e–j2π(2)15/16X(z)z–0 + e–j2π(2)14/16X(z)z–1 + ... + e–j2π(2)0/16X(z)z–15

= 15e–j2π(2)(15–k)/16X(z)z–k
∑ .
k=0

Thus the FSF's transfer function is

Y(z)
HFSF(z) = X(z) = 15e–j2π(2)(15–k)/16z–k = 15e–j2π(2)15/16ej2π(2)k/16z–k
∑ ∑

k=0 k=0

= e–j2π(2)15/16 15ej2π(2)k/16z–k
∑ .
k=0

From the abcdbe = (acde)b law of exponents, we can write

HFSF(z) = e–j2π(2)15/16 15(ej2π(2)/16z–1)k
∑ .
k=0

Using the material in Appendix B, we can write the above geometric series
in closed form as
1 – (ej2π(2)/16z–1)16 1 – ej2π(2)z–16
HFSF(z) = e–j2π(2)15/16 j2π(2)/16 –1 = e–j2π(2)15/16
1–e z 1 – ejπ/4z–1

Property of Pearson Education. Not permissible for redistribution.

 1 – z–16 1
=e –j2π(2)15/16
= (1 – z–16) e–j2π(2)15/16
1 – ejπ/4z–1 1 – ejπ/4z–1
Realizing that e–j2π(2)15/16 = ej2π(2)/16, we write our desired transfer function as
1
HFSF(z) = (1 – z–16) jπ/4 –1 e
j2π(2)/16
1–e z

1
= (1 – z–16) ejπ/4
1 – ejπ/4z–1
whose implementation is that shown in S7–24–II(b).
Note: The Figure S7–24–II(b) solution to this problem is called a "sliding
DFT", and is well known in the field of DSP. A detailed discussion of the
sliding DFT is provided in Section 13.18.

(b) There are two ways to solve this part of the problem. First, we could use our
knowledge of DFT bin center frequencies. If the fs sample rate of the x(n)
input is 200 kHz, the center frequency of the m = 2 DFT bin is

mfs (2)(200 kHz)

f(m=2 center) = N = 16 = 25 kHz.

The second way to find the center frequency of the m = 2 DFT bin is to find
the resonant frequency of the single-section FSF resonator in Figure S7–
24–II(b) by determining the angle of the resonator's z-plane pole location.
From the text's Eq. (7–40), the resonator's Hres(z) transfer function is
1
Hres(z) = .
1 – ejπ/4z–1
Setting Hres(z)'s denominator equal to zero yields
1 – ejπ/4zpole–1 = 0
or
zpole = ejπ/4
defining the resonator's z-plane pole location to be on the unit circle at an
angle of π/4 radians. Because a resonant frequency of 2π radians
corresponds to fs Hz, the filter's cyclic resonant frequency (and the desired
single-bin DFT center frequency) is found using
π . fs 200 kHz
fr = f(m=2 center) = 4 = 8 = 25 kHz.
2π

Property of Pearson Education. Not permissible for redistribution.

Property of Pearson Education. Not permissible for redistribution.
 Chapter 8 Solutions
Solution: 8.1
a) The completed table is as follows:

Real-only Integer Rational Complex

x+6=0 Yes Yes Yes (–6/1) No
2x –7 = 0 Yes No Yes No
x2 –2 = 0 Yes No No No
x2 + 2 = 0 No No No Yes

b) The rectangular and polar representations are:

1. x = –6 + j0 = 6ejπ = 6e–jπ
2. x = 7/2 + j0 = 7ej0/2 = 7/2
3. x = ± 2
x1 = 2 + j0 = 2 ·ej0 = 2
x2 = – 2 + j0 = 2 ·ejπ = 2 ·e–jπ
4. x = 0 ± jsqrt(2)
x1 = 0 + j 2 = 2 ·ejπ/2
x2 = 0 – j 2 = 2 ·e–jπ/2 = 2 ·ej3π/2

Solution: 8.2
(a) Rectangular notation proof: Given a complex number C = a +jb,
multiplication by j is:
jC = ja – b = –b + ja.
Numbers C = a +jb and jC = –b + ja are shown in Figure S8–2.

Imag

jC a
θ3
θ2
b C

θ1
–b a Real

Figure S8–2

that anglesofθPearson
Notice Property 1 and θ2 are
Education.
equal as: Not permissible for redistribution.
 ⎛b⎞
θ1 = tan–1⎜ a ⎟ = θ2.
⎝ ⎠
The amount of rotation is angle θ3 minus angle θ1 or:
⎛π ⎞ π π
θ3 – θ1 = ⎜ 2 + θ2⎟ – θ1 = 2 + θ1 – θ1 = 2 = 90o.
⎝ ⎠

Polar notation proof: Given a complex number C = Mejθ, and using the text's
Eq. (8–12), j = ejπ/2, multiplication by j is:
jC = Mejθejπ/2 = Mej(θ + π/2).
Thus the angle of jC is θ + π/2 radians, or θ + 90o, which is what we set out
to prove. Again, using polar notation yields a simpler solution.

(b) Given a complex number C = Mejθ:

jC = jMejθ = Mejθejπ/2 = Mej(θ + π/2).
Thus we may write
|jC| = M = |C|.

Solution: 8.3
(a) With a rectangular form of C as C = Mcos(φ) + jMsin(φ):
CC* = [Mcos(φ) + jMsin(φ)][Mcos(φ) – jMsin(φ)]

= M2cos2(φ) – jM2cos(φ)sin(φ) + jM2sin(φ)cos(φ) + M2sin2(φ)

= M2[cos2(φ) + sin2(φ)] + jM2[sin(φ)cos(φ) – cos(φ)sin(φ)]

= M2[cos2(φ) + sin2(φ)] + jM2[sin(φ)cos(φ) – sin(φ)cos(φ)]

= M2[1] + jM2[0] = M2.

(b) With a polar form of C as C = Mejφ:

CC* = (Mejφ)(Me–jφ) = M2ej[φ−φ] = M2ej[0] = M2.

(c) The polar of C = Mejφ was much easier to use.

Solution: 8.4
If the sum of a complex number C plus its reciprocal (C + 1/C) is
real only, then the magnitude of C must be unity. That is, |C| = 1.

Justification of this solution is as follows: If C = R + jI, then from Appendix

A (Section A.3.5,ofInverse
Property Pearsonof aEducation.
Complex Number) 1/C is
Not permissible for redistribution.
 1 R – jI
C = R2 + I2 .
So we may write C + 1/C as:
1 R – jI R I
C + C = R + jI + R2 + I2 = R + jI + R2 + I2 –j R2 + I2

R I
= (R + R2 + I2 ) + j(I – R2 + I2 ).

If the imaginary part of C + 1/C is zero, then we write:

I
I – R2 + I2 = 0,

leading to
I
I = R2 + I2 .

For the above expression to be true, R2 + I2 = |C|2 = 1. If the magnitude

squared of C is one, then magnitude of C is also one, which is what we set
out to show.

Solution: 8.5
This is a trick question. It is not valid to use Qefficient to compute the
desired Q value! In general, the real part of the ratio of two complex
numbers is not equal to the ratio of their individual real parts.
We show this as by writing the original ratio of complex numbers:
Ca Ra + jIa
Cb = Rb + jIb .
Using the principles given in Appendix A's Eq. (A–20) , we write:
Ca (RaRb + IaIb) + j(RbIa – RaIb)
Cb = Rb2 + Ib2 .

The correct value for our desired Q is the real part of the above ratio, or:
⎡ Ca ⎤ RaRb + IaIb
Q = real part of ⎢ C ⎥ = R 2 + I 2 .
⎣ b⎦ b b

The expression for Qefficient given in the problem is:

real part of [Ca] Ra
Qefficient = real part of [C ] = R .
b b

Therefore, in general, Qefficient ≠ Q, so the proposal of computing Qefficient to

obtain the desired Q is not correct.

Solution: 8.6
Property of Pearson Education. Not permissible for redistribution.
 The solution proceeds as follows:
cos(α + β) = Re{ej(α + β)} = Re{ejαejβ}
= Re{[cos(α) + jsin(α)]·[cos(β) + jsin(β)]}
= Re{cos(α)cos(β) + jcos(α)sin(β) + jsin(α)cos(β) – sin(α)sin(β)}
= cos(α)cos(β) – sin(α)sin(β).

Solution: 8.7
Given the notation for the spectrum of a continuous cosine wave in the text's Figure
8–10(a), shown here as Figure S8–7(a), we describe the problem's given x(t) signal,
in the time domain, as shown in S8–7(b). Thus the problem solution is:

x(t) = cos(2π100t ± π) = –cos(2π100t).

Imag
Real
–fo
cos(2πfot) =
e j2πfot + e–j2πfot
0
2 2
(a) fo
Freq

Imag
–100 Real

0 100
(b)
Freq (Hz)

e j(2π100t +– π) e –j(2π100t +– π)
2
+
2 – π)
= cos(2π100t +

Figure S8–7

Solution: 8.8
To add the two complex exponentials, we convert them to rectangular form for
addition, and convert the sum back to polar form, as follows:

(
Ae j (ωt +α) + Be j (ωt +β) = e jωt Ae jα + Be jβ )
= e jωt [ A cos(α) + B cos(β) + j ( A sin(α) + B sin(β) )]

⎡ j ⋅ tan −1 ⎡⎢
A sin(α) + B sin(β) ⎤
⎣ A cos(α) + B cos(β) ⎥⎦
⎤
⋅ ⎢ [ A cos(α) + B cos(β) ] + [ A sin(α) + B sin(β) ] ⋅ e
jωt 2 2
=e ⎥
⎢⎣ ⎥⎦
Property of Pearson Education. Not permissible for redistribution.
 {
j ωt + tan −1 ⎡⎢
A sin(α) + B sin(β) ⎤
}
⎣ A cos(α) + B cos(β) ⎥⎦
[ A cos(α) + B cos(β) ] + [ A sin(α) + B sin(β) ] ⋅e
2 2
= .

So the problem solutions are:

M= [ A cos(α) + B cos(β)]2 + [ A sin(α) + B sin(β)]2

and

⎡ A sin(α) + B sin(β) ⎤
θ = ωt + tan −1 ⎢ ⎥.
⎣ A cos(α) + B cos(β) ⎦

Solution: 8.9
The proof proceeds as follows:

sin(α)
tan(α) = = sin(α) · [1/cos(α)]
cos(α)
ejα – e–jα 2
= · jα
2j e + e–jα
ejα – e–jα
= .
j(ejα + e–jα)

Solution: 8.10
Given two complex numbers, C1 = a + jb and C2 = c + jd, either rectangular
or polar notation may be used to prove |C1|·|C2| = |C1C2|.

Rectangular notation proof: The product of the magnitudes is

ProdOfMag = |C1|·|C2| = |(a + jb)|·|(c + jd)|

= (a2 + b2)1/2(c2 + d2)1/2 = [(a2 + b2)(c2 + d2)]1/2 .

Next, the magnitude of the product is:
MagOfProd = |C1C2| = |(a + jb)(c + jd)| = |(ac –bd) + j(ad + bc)|

= [(ac –bd)2 + (ad + bc)2]1/2 .

Squaring the terms of inside the brackets gives:
MagOfProd = [(ac)2 –2abcd + (bd)2 + (ad)2 + 2abcd + (bc)2]1/2

= [(ac)2 + (bd)2 + (ad)2 + (bc)2]1/2

= [a2(c2 + d2) + b2(c2 + d2)]1/2

= [(a2 + bEducation.
Property of Pearson
2
) (c2 + d2)]1/2Not permissible(above).
= ProdOfMag for redistribution.
 which is what we set out to prove. (Whew!)

Polar notation proof: Using C1 = M1ejθ and C2 = M2ejφ, the product of the
magnitudes is:

ProdOfMag = |C1|·|C2| = |M1ejθ|·|M2ejφ| = M1M2.

Next, the magnitude of the product is:

MagOf Prod = |C1C2| = |M1ejθM2ejφ| = |M1M2ej(θ+φ)|

= M1M2 = ProdOfMag (above).

Solution: 8.11 Using polar notation:

(a) If C = Mejθ, then C* = Me–jθ. So:

|C/C*| = |Mejθ/Me–jθ| = |ej2θ| = 1.

Hopefully the student is learning that polar notation is often easier to use
than rectangular notation for solving complex-valued algebra problems.

(b) Using C1 = M1ejθ and C2 = M2ejφ, we write:

(C1·C2)* = C1*·C2*

(M1ejθM2ejφ)* = M1e–jθM2e–jφ

(M1M2ej(θ+φ))* = M1M2e–j(θ+φ)

M1M2e–j(θ+φ) = M1M2e–j(θ+φ)
which is what we set out to prove.

Solution: 8.12
Because the signals' magnitudes are |c1(t)| = 1, and |c2(t)| = 0.75, the equation for
the sum of the two signals, having exponents measured in radians, is:

c1(t) + c2(t) = ej[2π(5)t + π/4] + 0.75e–j[2π(7)t + π/3].

Solution: 8.13
Converting to polar notation makes the proof quite simple. That is,

[cos(φ) +jsin(φ)]N = (ejφ)N = ejNφ = cos(Nφ) + jsin(Nφ).

Here we learn that polar notation is often easier to work with than
rectangular notation in the algebra of quadrature processing.

Property of Pearson Education. Not permissible for redistribution.

Solution: 8.14
Given q(n) = 0.9nej2πn/8 :

(a) The magnitude of q(n) decreases (exponentially) as time index n increases.

(b) The phase of q(n) increases linearly as time index n increases.

(c) The real part of q(n) begins with a unity-valued sample (when n = 0) and is
a cosine wave sequence decreasing in amplitude as time index n increases.

(d) The imaginary part of q(n) begins with a zero-valued sample (when n = 0)
and is a sinewave sequence decreasing in amplitude as time index n
increases.

(e) The polar plot of q(n) is shown below in Figure S8–14(a).

A more descriptive 3-dimensional plot of q(n) is provided in Figure S8–14(b).

Solution:
1
q(n) q(2)
0.8
q(1)
0.6
q(3)
0.4
Imaginary

0.2

(a) 0
q(4) q(0)
-0.2
q(5) q(7)
-0.4
-0.6 q(6)

-0.8
-1
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
Real

Property of Pearson Education. Not permissible for redistribution.

 q(n)

2
1.5

Imag Part
1
0.5
0
(b)
-0.5
-1

-2
0 2
2 1
Time 4 0 Real Part
6
8 -1

Figure S8–14

Solution: 8.15
Using Euler's equation, the cosine sequence x(n) may be written in complex
exponential form as
ej[2π(–1000)nts + π/4] e–j[2π(–1000)nts + π/4]
x(n) = 2 + 2 .

Using the Law of Exponents, we express x(n) as

e–j2π1000nts ejπ/4 ej2π(1000)nts e–jπ/4
x(n) = 2 + 2
where the first term above is x(n)'s negative frequency component having a
phase shift of π/4 radians, or
+45 degrees.

Solution: 8.16
The text's Figure 8–11 gives us the answer, but we can also use algebra to arrive at
the solution. If the highest frequency spectral component of |XC(ω)| is ωo
radians/sample, shown by the bold arrow in Figure S8–16(a), in the discrete time
domain we can express that spectral component as:

Mejωon/fs = M[cos(ωon/fs) + jsin(ωon/fs)].

Property of Pearson Education. Not permissible for redistribution.

 |XC(ω)| |XR(ω)|
M P = M/2

0 π/2 π ω –π/2 0 π/2 π ω

ωo –ωo ωo
(a) (b)

Figure S8–16

From Euler's equation, the real part of that ωo spectral component is

M·ejωon/fs M·e–jωon/fs
M[cos(ωon/fs) = 2 + 2 .

So the magnitudes of the ±ωo spectral components in |XR(ω)| are

P = M/2
as shown in Figure S8–16(b).

Solution: 8.17
The complex constant ejπ/2 is equal to the j operator,
ejπ/2 = j.
So we may write y(n) as
y(n) = j·x(n) = j·[xi(n) + jxq(n)] = –xq(n) + jxi(n)
leading to the simple problem's block diagram solution shown in Figure S8–
17.

Solution:
xi(n) yi(n)

–1
xq(n) yq(n)

Figure S8–17

Solution: 8.18
Using the function-product trigonometric identity:
cos(α – β) cos(α + β)
sin(α)sin(β) = 2 – 2
we write:
cos(2πft – 2πft – θ) – cos(2πft + 2πft + θ)
sin(2πft)sin(2πft + θ) = 2 .
Property
Because cos(–θ) of Pearson
= cos(θ), we Education. Not permissible
write our solution as: for redistribution.
 cos(θ) cos(2π(2f)t+θ)
sin(2πft)sin(2πft + θ) = 2 – 2 .

Alternatively, using Euler's sin(α) = (ejα – e–jα)/j2, we may write:

ej2πft – e–j2πft ej(2πft+θ) – e–j(2πft+θ)
sin(2πft)sin(2πft + θ) = j2 · j2

ej[2π(2f)t+θ] – ej(–θ) –ej(θ) + e–j[2π(2f)t+θ]

= –4

ejθ + e–jθ ej[2π(2f)t+θ] + e–j[2π(2f)t+θ]

= 4 – 4 .

Using Euler's 2cos(α) = (ejα + e–jα), we write our solution:

cos(θ) cos(2π(2f)t+θ)
= 2 – 2
as shown in Figure S8–18 for θ = π/4.

1
0
sin(2πft)
–1
1 The average of
θ = π/4 sin(2πft)sin(2πft + π/4)
0 equals
sin(2πft + θ) cos(π/4)/2 = 0.354.
–1
1
θ = π/4
0
sin(2πft)sin(2πft + θ)
–1
0 0.2 0.4 0.6 0.8 1
t

Figure S8–18

Solution: 8.19
To determine the constants A, B, and C, we start with
x(n) = 5ej(2πn/20 + π/4) + 3ej(2πn/20 + π/6)
and factor out the ej2πn/20 term yielding
x(n) = ej2πn/20(5ejπ/4 + 3ejπ/6).
Next we convert the complex exponentials inside the parenthesis to
rectangular form as
x(n) = ej2πn/20[5cos(π/4) + j5sin(π/4) + 3cos(π/6) + j3sin(π/6)].
The terms inside the brackets are constants which we evaluate as
x(n) = ej2πn/20[3.54 + j3.54 + 2.598 + j1.5] = ej2πn/20[6.13 + j5.04]
Property of Pearson Education. Not permissible for redistribution.
 = ej2πn/20[7.94ej0.687] = 7.94ej(2πn/20 + 0.687) .
So our desired A, B, and C constants are:

A = 7.94, B = 2π/20 = π/10, C = 0.687.

Solution: 8.20
(a) The real part of x(t) is 1 – cos(2π1t), as shown in Figure S8–20(a).

(b) The imaginary part of x(t) is –sin(2π1t), as shown in Figure S8–20(b).

(c) The magnitude of x(t) is

|x(t)| = [1 – cos(2π1t)]2 + [–sin(2π1t)]2

= 1 – 2cos(2π1t) + cos2(2π1t) + sin2(2π1t) .

Recalling trig identity: cos2(α) + sin2(α) = 1, we write

|x(t)| = 1 – 2cos(2π1t) + 1 = 2 – 2cos(2π1t)

4 – 4cos(2π1t) 1 – cos(2π1t)
= 2 =2 2 .

Recalling the half-angle trigonometric identity

{[1 – cos(α)]/2}1/2 = ±sin(α/2),
and that our |x(t)| must be positive, we have our desired solution of

|x(t)| = |2sin(πt)|.

which is shown in Figure S8–20(c).

Property of Pearson Education. Not permissible for redistribution.

 Solution:
2 Real part of x(t)

(a) 1
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
1
Imaginary part of x(t)
(b) 0
–1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

2
|x(t)|
(c) 1
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time (seconds)

Figure S8–20

Solution: 8.21
(a) The θ2 arctangent algorithm has the lowest average error magnitude.

(b) The computational workload of the three arctangent algorithms is given in

the table below.

Algorithm Multiplies Additions Divisions

θ1 5 2 0
θ2 5 2 1
θ3 4 1 0

(c) The required number of processor clock cycles for the three algorithms are:

Algorithm clock cycles

θ1 5+2=7
θ2 5 + 2 + 30 = 37
θ3 4+1=5

Given the computational workload of 30 processor clock cycles for a single

division, the θ2 algorithm is disqualified for its computational intensity.

Because of its lower computational overhead and its lower average error,
the θ3 arctangent algorithm should be chosen over the θ1 algorithm.

Property of Pearson Education. Not permissible for redistribution.

 (d) Given that the true angle θ is always in the range of –22.5o ≤ θ ≤ 22.5o, the
θ1 algorithm is optimum because of its computational efficiency and greatly
reduced average error over the true θ angular range.

Solution: 8.22
Given that the algorithm's error is
X
E(X) = tan-1(X) – 1 + AX2 radians,

we can find the value Xmax error where the E(X) function is at its maximum by
taking the derivative of E(X) with respect to X, setting that derivative equal
to zero, and solving for Xmax error. Knowing Xmax error, we can compute the
desired angle θ associated with Xmax error. We do this as follows, Given the
above E(X), its derivative with respect to X is
d(1 + AX2)
(1 + AX2) –
dE(X) 1 dX 1 1 + AX2 – 2AX2
dX = 1 + X2 – 2 2
(1 + AX )
= 1 + X2 – 1 + 2AX2 + A2X4 .

Setting the derivative equal to zero leads to

1 1 + AX2 – 2AX2 1 –AX2
1 + X2 = 1 + 2AX2 + A2X4 = 1 + 2AX2 + A2X4
where variable X now represents the ratio Q/I having the maximum
arctangent approximation error. Solving for X we re-write the above
expression as
1 + 2AX2 + A2X4 = (1 + X2)(1 –AX2)
or
2AX2 + A2X4 = (1 –A) X2 –AX4
or
2A –(1 –A) = –(A2 + A) X2
or
–(3A –1)
X2 = A2 + A .

With A = 0.28125,
–(0.84375 –1)
X2 = 0.36035 = 0.4336.

So X is
X = ± 0.4336 = ±0.6585
giving the two values for X at which the error is maximized. Our solution is
the arctangent of X = ±0.6585, or
Property of Pearson Education. Not permissible for redistribution.
 θmax error = tan–1(±0.6585) = ±0.5823 radians = ±33.36o.

Those angles are shown in Figure S8–22.

0.4

Error (degrees)
0.2

–0.2
–0.4
–40 –20 0 20 40
True angle θ (degrees)
–33.36o 33.36o

Figure S8–22

Solution: 8.23
In our text, we have represented a complex exponential sequence using the form
m(n) = ej2πfonts
where fo is the complex exponential's frequency measured in Hz. Setting that
expression equal to the problem's m(n) expression, we may write:
m(n) = ej2πfonts = ej0.8πn
Setting the above angle arguments equal to each other, we write:
2πfonts = 0.8πn.
Recalling the definition that ts = 1/fs, solving the above expression for the
frequency fo, we have our solution of:

0.8πn 0.8
fo = = ⋅ f s = 0.4 ⋅ 5000 = 2000 Hz.
2πnts 2

Solution: 8.24
(a)

xi(t) = cos[2π(fc+10)t]cos(2πfct).

Using the trig identity: cos(α)cos(β) = cos(α–β)/2 + cos(α+β)/2, xi(t)

becomes:

xi(t) = cos[2π(fc+10)t –2πfct]/2 + cos[2π(fc+10)t + 2πfct]/2.

= cos(2π10t)/2 + cos[2π(2fc+10)t]/2.

The solution is a 10 Hz cosine wave and a (2fc+10) Hz cosine wave.

Property of Pearson Education. Not permissible for redistribution.
 (b)
xq(t) = –cos[2π(fc+10)t]sin(2πfct).

Using the trig identity: cos(α)sin(β) = sin(α+β)/2 –sin(α–β)/2, xq(t) becomes:

xq(t) = –sin[2π(fc+10)t + 2πfct]/2 + sin[2π(fc+10)t –2πfct]/2.

= sin(2π10t)/2 –sin[2π(2fc+10)t]/2.

The solution is a 10 Hz sinewave and a negative (2fc+10) Hz sinewave.

(c) The minus sign on the second term for xq(t) represents a 180o (π radians)
phase shift between the low and high frequency components in xq(t).

This property is indicated in the text's Figure 8–19(b).

(d) After lowpass filtering (removing the high frequency term in xi(t)),

the filter output is the 10 Hz cosine wave of

i(t) = cos(2π10t)/2.

(e) After lowpass filtering (removing the high frequency term in xq(t)),

the filter output is the 10 Hz sinewave of

q(t) = sin(2π10t)/2.

Solution: 8.25
The first challenge for the student is to correctly define the xc(n) time-
domain expression for a discrete complex sinusoid, whose frequency is –fc
Hz, used to translate (mix) the x(n) = cos(2πfonts) signal down in frequency
by fc Hz. That discrete complex sinusoid expression is:
xc(n) = e–j2πfcnts.
Using Euler's identity, cos(α) = (ejα + e–jα)/2, we write x(n) as:
1
x(n) = 2 (ej2πfonts + e–j2πfonts).

The problem solution is the xp(n) product of x(n) and xc(n) expressed as:

1
xp(n) = 2 (ej2πfonts + e–j2πfonts)e–j2πfcnts
1
= 2 (ej2π(fo–fc)nts + e–j2π(fo+fc)nts)
Property of Pearson Education. Not permissible for redistribution.
 containing spectral components at frequencies 2π(fo–fc) and –2π(fo+fc)
radians/second.

(b) With the spectrum of the original x(n) being that shown in Figure S8–25(a),
the solution is the two spectral components of xp(n) as shown in Figure S8–
25(b).

Original X(m)
1/2 1/2

–fo 0 fo Freq
(Hz)

Solution:
Xp(m)

1/2 1/2

–fo 0 fo Freq
–fo–fc fo–fc (Hz)
[–(fo+fc)]

Figure S8–25

(Complex signals need not have spectral magnitude symmetry around the
zero Hz point as do real signals.)

(c) There is no spectral amplitude loss, or gain, when multiplying a

discrete x(n) time-domain signal by e–j2πfct.

Solution: 8.26
Given the expression

mimp(t) = Acos(ωot) – jA(1+ε)sin(ωot + α)

we convert mimp(t) to polar form using Euler's expressions for cos and sin,
yielding:
ejωot + e–jωot je–j(ωot+α) – jej(ωot+α)
mimp(t) = A· 2 – jA(1+ε)· 2

ejωot + e–jωot je–jωote–jα – jejωotejα

= A· 2 – jA(1+ε) · 2 .

Collecting terms to determine the complex-amplitude terms of the ωo and

–ωo frequency components gives us:
–jα
⎡ A A(1+ε)e ⎤ jωot ⎡ A A(1+ε)e ⎤ –jωot
jα
mimp(t) = ⎢ 2 – ⎥e + ⎢ 2 + ⎥e
⎣ 2 ⎦ ⎣ 2 ⎦
Property of Pearson Education. Not permissible for redistribution.
 ⎪⎧ A A(1+ε) ⎪⎫ jωot
=⎨2 – 2 · [cos(α) + jsin(α)] ⎬e
⎩⎪ ⎭⎪

⎡A A(1+ε) ⎤ –jωot
+ ⎢2 +
⎣ 2 ·[cos(α) – jsin(α)]⎥⎦e

⎧⎪ ⎡ A A(1+ε) ⎤ ⎡ A(1+ε) ⎤ ⎫⎪⎬ jωot

=⎨⎢2 – · cos(α) ⎥ – j ⎢ · sin(α)⎥ e
⎩⎪ ⎣ 2 ⎦ ⎣ 2 ⎦ ⎭⎪

⎪⎧ ⎡ A A(1+ε) ⎤ ⎡ A(1+ε) ⎤ ⎪⎫⎬ –jωot

+⎨⎢2 + · cos(α) ⎥ – j ⎢ · sin(α) ⎥ e .
⎩⎪ ⎣ 2 ⎦ ⎣ 2 ⎦ ⎭⎪

Because α<<1, cos(α) ≈ 1 and sin(α) ≈ α, and we write:

⎡ –Aε A(1+ε)α ⎤ jωot ⎡ ε A(1+ε)α ⎤ –jωot

mimp(t) ≈ ⎢ 2 – j ⎥ e + ⎢A(1 + 2 ) – j ⎥e .
⎣ 2 ⎦ ⎣ 2 ⎦

Next, because ε<<1, (1+ε/2) ≈ (1+ε) ≈ 1 and we have:

⎡ –Aε Aα ⎤ ⎡ Aα ⎤
mimp(t) ≈ ⎢ 2 – j 2 ⎥ ejωot + ⎢A – j 2 ⎥ e–jωot
⎣ ⎦ ⎣ ⎦

giving us the complex-amplitude terms of the ωo and –ωo frequency

components. Because we want the magnitude of the unwanted ωo component of
mimp(t) to be 0.1% of the magnitude of the desired –ωo component, we write:

A2ε2 A2α2 –3 A2α2

4 + 4 = 10 A2 + 4 .

Factoring A from within the square roots and dividing both sides by A we have:

ε2 α2 –3 α2
4 + 4 = 10 1+ 4 .

Squaring both sides of the above equation, we write:

ε2 α2 –6 ⎡ α2 ⎤
+ = 10 ⎢ 1 + ⎥
4 4 ⎣ 4 ⎦.

Finally, because α<<1, (1+α2/4) ≈ 1 and we write the solution to this problem
as:
ε2 + α2 ≈ 4x10–6
telling us that the sum of errors ε2 and α2 must not be greater
than 4x10–6.

Solution: 8.27
(a) The original incorrect filter structure, repeated here in Figure S8–27(a), has
incorrect subtractions, and incorrect signs for the "B" coefficients.
Property of Pearson Education. Not permissible for redistribution.
 Incorrect signs

yi(n)
xi(n) –
z–1
–B
(a)

+B
xq(n)
z–1 yq(n)
–

Figure S8–27

To prove this, we determine the real-valued difference equation for the quadrature
bandpass filter as:
yi(n) + jyq(n) = [xi(n) + jxq(n)] + [A + jB][yi(n–1) + jyq(n–1)]

= [xi(n) + jxq(n)] + Ayi(n–1) – Byq(n–1) + j[Ayq(n–1) + Byi(n–1)]

= xi(n) + Ayi(n–1) – Byq(n–1) + j[xq(n) + Ayq(n–1) + Byi(n–1)].

So to compute yreal(n) + jyimag(n), we implement the following expressions.
yreal(n) = xi(n) + Ayi(n–1) – Byq(n–1)

yimag(n) = xq(n) + Ayq(n–1) + Byi(n–1).

There are several equivalent correct implementations of the quadrature
bandpass filter. Three correct solutions are provided in Figures S8–27(b),
S8–27(c), and S8–27(d).

Property of Pearson Education. Not permissible for redistribution.

 Solutions:
A A –A

yi(n) yi(n) yi(n)

xi(n) xi(n) xi(n) –
z–1 z–1 z–1
– –
B –B –B

–B B B
xq(n) xq(n) – xq(n) –
z–1 yq(n) z–1 yq(n) z–1 yq(n)
–

A A –A
(b) (c) (d)

Figure S8–27 (Cont'd)

(b) In complex notation, the filter's difference equation is:

y(n) = x(n) + ej2πfr/fsy(n–1).

(c) The complex filter's H(z) transfer function is:

Y(z) 1
H(z) = X(z) = .
1 –ej2πfr/fsz–1

(d) The filter is only conditionally stable. Setting the denominator of H(z) equal
to zero and solving for z gives:
1 –ej2πfr/fsz–1 = 0, or

z = ej2πfr/fs.
This results in a single filter pole at z = ej2πfr/fs directly on the z-plane's unit
circle, and that's what we call "conditionally stable".

(e) Setting z = ejω = ej2πf in H(z), yields:

1
H(f) = H(z)|z=ej2πf = j2πfr/fs –j2πf , or
1 –e e

1
H(f) = –j2π(f – fr/fs) ← Polar form
1 –e

1
= . ← Rectangular form
1 –cos[2π(f – fr/fs)] +jsin[2π(f – fr/fs)]

Property of Pearson Education. Not permissible for redistribution.

Solution: 8.28
Finding the solutions to these problems require that we start with the text's
complex frequency translation multiplier, shown in Figure S8–28(a), and to
modify that multiplier appropriately.

(a) Assuming we want to frequency translate a real-only time sequence, the

block diagram of the discrete complex frequency translation network whose
input is a real-only valued discrete sequence, and whose output is a
complex-valued discrete sequence is shown in Figure S8–28(b).

+
i(n) i'(n)
–

q(n)

(a) q'(n)

cos(2πfcnts) sin(2πfcnts) up-conversion

–sin(2πfcnts) down-conversion

Solution:
i(n) i'(n)
xc'(n) = i'(n) + jq'(n)
(b) q'(n)
cos(2πfcnts)
sin(2πfcnts) up-conversion
–sin(2πfcnts) down-conversion

Figure S8–28

(b) The block diagram of a discrete complex frequency translation network

whose input is a complex valued discrete sequence and whose output
sequence is real-valued is shown in Figure S8–28(c).

Solution:
+
i(n) i'(n)
–
q(n)
(c)
cos(2πfcnts) sin(2πfcnts) up-conversion
–sin(2πfcnts) down-conversion

Figure S8–28 (Cont'd)

Solution: 8.29
First we need to know the difference equation for our complex resonator,
which is:
Property of Pearson Education. Not permissible for redistribution.
 y(n) = x(n) + ejωry(n–1)

= x(n) + [cos(ωr) + jsin(ωr)] y(n–1).

Writing y(n) and y(n–1) in rectangular form we have
yreal(n) + jyimag(n) = x(n) + [cos(ωr) + jsin(ωr)][yreal(n–1) + jyimag(n–1)]

= x(n) + cos(ωr)yreal(n–1) – sin(ωr)yimag(n–1)

+ j[cos(ωr)yimag(n–1) + sin(ωr)yreal(n–1)].
So we build the complex oscillator by implementing
yreal(n) = x(n) + cos(ωr)yreal(n–1) – sin(ωr)yimag(n–1)
and
yimag(n) = cos(ωr)yimag(n–1) + sin(ωr)yreal(n–1).

The block diagram for the real-valued implementation of the complex

resonator (oscillator) is shown below in Figure S8–29.

Solution:
Complex digital resonator (oscillator)
x(n) yreal(n)

z–1
yimag(n)

z–1
+
–

sin(ωr) cos(ωr)

Figure S8–29

Solution: 8.30
(a) Because the real-coefficient lowpass filter's transfer function is
Hreal(z) = 1 + z–1,
that filter has a z-plane zero at z = –1. We want a filter having a z-plane zero
at z = e–jπ/2, so the transfer function of our desired complex filter is:
HProperty of Pearson
cmplx(z) = 1 – e
–jπ/2 –1 Education. Not permissible for redistribution.
z .
 Because –e–jπ/2 = ejπ/2, an alternate (and equivalent) Hcmplx(z) transfer
function is:

Alternate Hcmplx(z) = 1 + ejπ/2z–1 .

(b) Block diagrams of the two equivalent Hcmplx(z) filters are shown in Figures
S8–30(a) and (b).

Solutions:
Hcmplx(z) Alternate Hcmplx(z)
x(n) y(n) x(n) y(n)
–
z–1 z–1

e–jπ/2 ejπ/2
(a) (b)

Figure S8–30

Solution: 8.31
The operation of the communications system is shown in Figure S8–31.

I(t)cos(ωct)
I(t)

cos(ωct)
–sin(ωct) +
I(t)cos(ωct)
–Q(t)sin(ωct) – Q(t)sin(ωct)
Q(t)

Antenna

I(t)
LPF
I(t) [cos(0t) + cos(2ωct)]/2 2
–Q(t) [sin(2ωct) - sin(0t)]/2
cos(ωct)

Q(t)
LPF
–I(t)[sin(2ωct) – sin(0t)]/2 2
–sin(ωct) + Q(t) [cos(0t) – cos(2ωct)]/2

Figure S8–31

In the demodulator the lowpass filters (LPF) eliminate the signal components
whose frequencies are 2ωc. Recalling that cos(0) = 1, and sin(0) = 0,
the output of the top LPF is:

Property of+ Pearson

I(t)cos(0t)/2 Q(t)sin(0t)/2 = I(t)/2 +Not
Education. Q(t)·0/2 = I(t)/2.for redistribution.
permissible
The output of the bottom LPF is:

I(t)sin(0t)/2 + Q(t)cos(0t)/2 = I(t)·0/2 + Q(t)/2 = Q(t)/2

Property of Pearson Education. Not permissible for redistribution.

Property of Pearson Education. Not permissible for redistribution.
 Chapter 9 Solutions
Solution: 9.1
As shown in Figure S9–1, the expressions for xa(t) and xb(t) are:

xa(t) = cos(ωt) – [–cos(ωt)] = 2cos(ωt)

xb(t) = cos(ωt) – cos(ωt) = 0.

xr(t) = cos(ωt)
+
xa(t) = 2cos(ωt)
sin(ωt) –cos(ωt) –
HT HT
+
xb(t) = 0
cos(ωt) +

Figure S9–1

Solution: 9.2
(a) Given the 3-dimensional Xr(f) spectrum of xr(t) = Asin(2πfot) shown in
Figure S9–2(a), the three-dimensional spectrum of the Hilbert transform of
xr(t), Xi(f), is shown in Figure S9–2(b).

Solution:
Xr(f) Xi(f)
Imag Imag
A/2
Real Hilbert –fo Real
–fo transform
–A/2
0 fo 0 fo
–A/2
–A/2 Freq Freq
(a) (b)

Figure S9–2

(b) Representing the Hilbert transform of xr(t) as xi(t), from Figure S9–2(b) we
see that xi(t) = –Acos(2πfot). Thus the equation for the analytic signal, xa(t),
associated with xr(t) is

xa(t) = xr(t) + jxi(t) = Asin(2πfot) + j[–Acos(2πfot)]

= Asin(2πfot) – jAcos(2πfot).

Property of Pearson Education. Not permissible for redistribution.

 (c) Using Euler's equations, we can represent the xa(t) analytic signal associated
with xr(t) as
xa(t) = Asin(2πfot) – jAcos(2πfot)
–j2πfot
jAej2πfot j2πfot
jAe–j2πfot
=A[ jAe2 – 2 ] – j[ jAe2 + 2 ]
or
xa(t) = –jAej2πfot.

(d) The xa(t) analytic signal is a positive-frequency exponential

because its angle becomes more positive as time advances.

(e) The xa(t) analytic signal, on a complex plane, at time t = 0 is shown as the
dot in Figure S9–2(c).

Imag
jA

(c) –A A
xa(0) Real

t=0
–jA

Figure S9–2 (Cont'd)

Solution: 9.3
As shown by the table in Figure S9–3(a):
(a) y(t) = x(t).
(b) v(t) = –x(t).
(c) w(t) = HT–1[x(t)].
(d) The system that uses a single Hilbert transform operation to compute
inverse Hilbert transforms is shown in Figure S9–3(b).

Property of Pearson Education. Not permissible for redistribution.

 HT2[x(t)]= HT[HT[(x(t)]]

Solution:
x(t) = x(t) =
HT3[x(t)]
cos(ωot) sin(ωot)
u(t), HT[(x(t)] sin(ωot) –cos(ωot) Inverse HT2[x(t)]
Hilbert
v(t), HT2[(x(t)] –cos(ωot) –sin(ωot) transform x(t)
HT
w(t), HT3[(x(t)] –sin(ωot) cos(ωot)
Inverse
y(t), HT4[(x(t)] cos(ωot) sin(ωot) –1 Hilbert
transform
(a) (b) of x(t).

Figure S9–3

Solution: 9.4
(a) We should expect their frequency magnitude responses to be zero (a
magnitude null) at zero Hz because the DC gain (gain at zero Hz) is
equal to the sum of the systems' impulse responses. For both Hilbert
transformers the sum of the impulse response samples is equal to zero.
(b) The z-domain transfer function for a 6-tap Hilbert transform filter is:

H(z) = h(0) + h(1)z–1 + h(2)z–2 + h(3)z–3 + h(4)z–4 + h(5)z–5 .

We determine the filter's frequency response at zero Hz (DC) by setting z = 1 in

H(z). Thus the filter's response at zero Hz is:
DC response = H(z)z=1 = h(0) + h(1) + h(2) + h(3) + h(4) + h(5) .
Because the coefficients have odd symmetry [h(0) = –h(5), h(1) = –h(4),
h(2) = –h(3)] the sum of the h(k) coefficients is zero as:

DC response = H(z)z=1 = h(0) + h(1) + h(2) – h(2) – h(1) – h(0) = 0.

which is what we set out to find.

Solution: 9.5
There are two correct solutions to this problem. The equivalent structures of an
11-tap FIR Hilbert transformer are shown in Figure S9–5.

Property of Pearson Education. Not permissible for redistribution.

 Solutions:
x'r(n)

xr(n)
z–1 z–1 z–1 z–1 z–1 z–1 z–1 z–1 z–1 z–1
(a)
h(0) h(2) h(4) h(6) h(8) h(10)

+
xi(n)

Notice the use of "two-unit" x'r(n)

delay (z–2) symbols
xr(n)
z–2 z–2 z–1 z–1 z–2 z–2
(b)
h(0) h(2) h(4) h(6) h(8) h(10)

+
xi(n)

Figure S9–5

Solution: 9.6
(a) There are several correct solutions. They are:
(I) H(z) = h(0) + h(1)z–1 + h(2)z–2 + h(3)z–3 + h(4)z–4 + h(5)z–5 + h(6)z–6
+ h(7)z–7 + h(8)z–8 + h(9)z–9 + h(10)z–10,
10
(II) H(z) = ∑ h(k)z–k.
k=0

Or, eliminating the zero-valued coefficients,:

(III) H(z) = h(0) + h(2)z–2 + h(4)z–4 + h(6)z–6 + h(8)z–8 + h(10)z–10,
5
(IV) H(z) = ∑ h(2k)z–2k.
k=0

(b) The Hilbert transformer is a 10th-order system.

(c) The Hilbert transformer requires 10 delay elements.
(d) The Hilbert transformer requires 6 multipliers.
Property of Pearson Education. Not permissible for redistribution.
Solution: 9.7
In the text, our expression for the impulse response of a discrete Hilbert
transformer is:
fs [1 – cos(πn)]
h(n) = .
πn
A common expression in the literature of DSP for the impulse response of a
discrete Hilbert transformer, under the assumption that fs is equal to unity, is:
2sin2(πn/2)
h'(n) = .
πn
We can equate h(n) and h'(n), based on the assumption that the fs sampling rate
is normalized to unity, by first substituting one for fs in h(n), yielding
[1 – cos(πn)]
hfs=1(n) = .
πn
Next, recalling the trigonometric power relations identity:
sin2(α) = [1 – cos(2α)]/2, we can substitute 2sin2(πn/2) for [1 – cos(πn)] in
hfs=1(n) to give the desired

2sin2(πn/2)
hfs=1(n) = = h'(n)
πn
which is what we set out to prove.

Solution: 9.8
The operations performed on X(m) to create an Xnew(m) sequence are shown
below:
N
Xnew(m) = –j·X(m), for positive frequencies (1≤m≤ 2 –1),

N
Xnew(m) = j·X(m), for negative frequencies ( 2 +1≤m≤N–1).

Solution: 9.9
The equation for the hhilb(k) coefficients when the half-band filter h(k)
coefficients' index k is defined using our standard notation of k = 0, 1, 2, ..., N–1
is found by replacing the original sin function's n index with
N–1
n=k– 2 .
This substitution shifts the sin function's indexing to correspond with our
traditional filter coefficient k indexing. The correct hhilb(k) expression is
Property of Pearson Education. Not permissible for redistribution.
 π N–1
hhilb(k) = 2sin[ 2 (k – 2 )]h(k).

Solution: 9.10
Because h(0) = –h(6) and h(2) = –h(4), two equivalent alternatives of the
original 7-tap FIR Hilbert transformer, shown in Figure S9–10(a), implemented
to reduce the number multipliers by a factor of two are shown in Figure S9–
10(b) and Figure S9–10(c). The difference between the two structures are:

1) the coefficients used, and

2) how the subtraction is performed in the first two adders.

x'r(n)
xr(n) xr(n–2) xr(n–4)
z–1 z–1 z–1 z–1 z–1 z–1 xr(n–6)

h(0) h(2) h(4) h(6)

(a)

+
xi(n)

Solutions:
Using the h(0) and h(2) coefficients
x(n–6) x(n–4)
z–1 z–1
x(n) x(n–2)
z–1 z–1 z–1 z–1

(b) – –
x'r(n)
h(0) h(2)

xi(n)

Using the h(4) and h(6) coefficients

x(n–6) x(n–4)
z–1 z–1
x(n) x(n–2)
z–1 z–1 z–1 z–1
– –
(c) x'r(n)
h(6) h(4)

xi(n)

Figure S9–10

Property of Pearson Education. Not permissible for redistribution.

Solution: 9.11
(a) The 6-tap FIR Hilbert transform filter used to generate both x'r(n) and xi(n)
is shown in Figure S9–11. This implementation is necessary to keep the
x'r(n) and xi(n) samples time-aligned.

Solution:
This process is difficult to
implement in an accurate,
computationally-efficient, way Half-sample x'r(n)
delay

xr(n)
z–1 z–1 z–1 z–1 z–1

h(0) h(1) h(2) h(3) h(4) h(5)

+
xi(n)

Figure S9–11

(b) The problem with an odd-order (even number of taps) Hilbert transformer
is that a non-integer (2.5 samples) delay is needed to obtain a time-aligned
x'r(n) sequence. Such non-integer, fractional, delay systems can be built, but
they require many arithmetic computations per output sample. From a
computational standpoint, even-order (odd number of taps) Hilbert
transformers are preferred because the x'r(n) output sequence is always
available at no computational cost.

Solution: 9.12
A complex-coefficient FIR filter's frequency magnitude response is periodic
because the filter's impulse response comprises discrete samples. All digital
filters have periodic frequency responses because the Fourier transforms of
discrete impulse response sequences are always periodic.

Solution: 9.13
(a) To determine an expression for the frequency response of a (K–1)th-order
FIR Hilbert transformer, first we write the transformer's time-domain
difference equation as
K–1
xi(n) = ∑ h(k) x (n–k).
r
k=0

Property of Pearson Education. Not permissible for redistribution.

 Just as we did with IIR filters in Chapter 6, we can determine the expression
for the frequency response of our Hilbert transformer by next taking the z-
transform of xi(n). Doing that, the z-domain expression for the transformer is
Κ–1 Κ–1
Xi(z) = ∑ h(k) X (z)z r
–k
= Xr(z) ∑ h(k)z –k
.
k=0 k=0

Next we can describe the transfer function of the K-tap Hilbert transformer
defined as H(z) = Xi(z)/Xr(z). Rearranging the above expression yields
Κ–1
Xi(z)
H(z) = X (z) =
r ∑ h(k)z –k
.
k=0

Finally, substituting ejω for z in the H(z) transfer function gives us our
desired expression for the (K–1)th-order Hilbert transformer's H(ω)
frequency response, in radians per sample, as
Κ–1
H(ω) = H(z)|z = ejω = ∑ h(k)e –jkω
.
k=0

If the student substituted ej2πf for z in the H(z) transfer function, the desired
expression for the (K–1)th-order Hilbert transformer's H(f) frequency
response, in cyclic frequency Hz, would be
Κ–1
H(f) = HHT(z)|z = ej2πf = ∑ h(k)e –jk2πf
.
k=0

(b) The range of continuous variables ω and f are:

ω in H(ω) would be defined over the range of –π to π radians/sample.

f in H(f) would be defined over the range of –fs/2 to fs/2, where fs is the
sample rate in Hz.
Because the h(k) coefficients are real-valued, the Hilbert transformer's
frequency magnitude responses |H(ω)| and |H(f)| will be symmetrical on
either side of the zero-frequency point. So in plotting frequency magnitude
responses the range of the radian and cyclic frequency variables could be 0
to π and 0 to fs/2 respectively.

(c) The equation for a K-point DFT of a (K–1)th-order Hilbert transformer is

Κ–1
HDFT(m) = ∑ h(k)e –j2πkm/K
, for 0≤m≤K–1.
k=0

(d) For a 43rd-order Hilbert transformer, the above DFT equation will yield 44
frequency-domain
Property ofsamples.
PearsonTo increase theNot
Education. number of frequency-domain
permissible for redistribution.
 samples, we would pad (append to) the h(k) impulse response sequence with
zero-values samples, and perform a larger-sized DFT of the lengthened
time-domain unit impulse response sequence.

Solution: 9.14
With K being an odd number, we can create a table whose entries are:

K Number of
multipliers
3 2
5* 2
7 4
9* 4
11 6
13* 6
15 8
17* 8
and so on.
* values where K – 1 is an integer multiple of 4.

When K – 1 is an integer multiple of 4, similar to half-band FIR filters, the

first and last h(k) coefficients are zero-valued and two multipliers can be
eliminated as shown for K = 5 in Figure S9–14.

K=5
xr(n)
z–1 z–1 z–1 z–1 xr(n)
z–1 z–1 z–1
0 h(1) 0 h(3) 0
h(1) h(3)

+
+ xi(n)
h(0) = h(2) = h(4) = 0 xi(n)

Figure S9–14

The problem's solution is:

K–2 K+2
Nmults = 2 + 2 ⎣ 4 ⎦ = 2⎣ 4 ⎦. ( ⎣X⎦ means the "integer part of X".)

Solution: 9.15
Writing an expression for Xc(ω) in terms of Xr(ω) and H(ω), we have
Xc(ω) = XProperty
r(ω) + jXiof
(ω)Pearson
= Xr(ω) +Education.
jXr(ω)H(ω) Not permissible for redistribution.
 = Xr(ω)[1 + jH(ω)].

To satisfy our constraints for Xc(ω) we can write

⎧⎪ 1, for − π < ω ≤ 0
1 + jH (ω) = ⎨
⎪⎩ 0, for 0 < ω < π

which leads to our solution of:

⎧⎪ 0, for − π < ω ≤ 0
H (ω) = ⎨
⎪⎩ j, for 0 < ω < π.

Solution: 9.16
The problem's solution, found by determining the spectra at two intermediate
nodes, A(f) and B(f), is shown in Figure S9–16(d). The network eliminates all
negative-frequency spectral components in x(t).

A(f)
xr(t) yr(t) Imag
0.5 Real
xi(t) yi(t)
–5
HT 0 10
a(t) –1 b(t)
Hz
HT
1
(a) (b)

Solution:
B(f)
Imag Y(f) = X(f) + B(f)
Imag
Real Real
–5 1
0.5
0 2
10 0
10
Hz
Hz
(c) (d)

Figure S9–16

Solution: 9.17
Over the time interval 100 ≤ t < 150 milliseconds, for example, signal xa(t) is
xa(t)100≤t<150 = 3cos(ωt) + jHT[3cos(ωt)] = 3cos(ωt) + j3sin(ωt)

where the "HT[]" notation means Hilbert transform. The magnitude of xa(t)100≤t<150 is
2 2
|xa(t)|Property
100≤t<150 = of [3cos(ωt)] + [3sin(ωt)]Not
Pearson Education. 9[cos2(ωt) for
= permissible + sin 2
(ωt)] .
redistribution.
Using the trig identity: cos2(α) + sin2(α) = 1, we write

|xa(t)|100≤t<150 = 9 = 3.

Using that same analysis over the full signal time duration, the problem solution
is the |xa(t)| waveform shown in Figure S9–17. Signal |xa(t)| is called the
"envelope" of the original xr(t) signal.

Solution:
4
|xa(t)|
3
2
1
0.5
0
0 50 100 150 200 250
Time (milliseconds)

Figure S9–17

Property of Pearson Education. Not permissible for redistribution.

Property of Pearson Education. Not permissible for redistribution.
 Chapter 10 Solutions
Solution: 10.1
(a) The x(n) sequence should be lowpass filtered before discarding
every fourth sample.

(b) Figure S10–1 shows the frequency magnitude response of an ideal (zero
transition region width) decimation-by-four lowpass filter.

Solution:
Filter mag. response
1
... ...

–2π –π –π/2 0 π/2 π 2π Rad./sample

(–fs) (–fs/2) –π/4 π/4 (fs/2) (fs) (Hz)
(–fs/8) (fs/8)

Figure S10–1

(c) The filter's DC gain should be 1, as shown in Figure S10–1, because

there is no time-domain amplitude gain or loss inherent in the
downsample-by-four process.

Solution: 10.2
There is no frequency difference between the x(n) and y(m) sequences. As
shown in Figure S10–2, the set of 36 samples in the original x(n) and y(m)
plots extend over two different intervals of time. Both sinusoids repeat every
18ts1 = 9ts2 = 18 milliseconds, so the x(n) and y(m) sinusoidal sequences
have identical frequencies.

Property of Pearson Education. Not permissible for redistribution.

 1
x(n) fs1 = 1000 Hz
0.5
(a) 0
–0.5
–1
0 5 10 15 20 25 30 35
Time (milliseconds)
ts1 = 1 millisecond
1
y(m) fs2 = 500 Hz
0.5
(b) 0
–0.5
–1
0 10 20 30 40 50 60 70
Time (milliseconds)
ts2 = 2ts1 = 2 milliseconds

Figure S10–2

Solution: 10.3
(a) Given that w(n)'s frequency is 128 Hz, the FFT size is N = 2048, and
fs = 1024 Hz, the positive-frequency value of index m for the FFT sample
having the largest magnitude in |W(m)| is
(128)N (128)2048
mmax = fs = 1024 = 256.

(b) Figure S10–3(a) shows the first 20 samples of the original w(n). Figure
S10–3(b) shows the first 20 samples of x(n). Now x(n)'s frequency is still
128 Hz, the FFT size is N = 1024, and fs = 512 Hz, the positive-frequency
value of index m for the FFT sample having the largest magnitude in |X(m)|
is
(128)N (128)1024
mmax,dec=2 = fs = 512 = 256.

(c) This is almost a trick question. Figure S10–3(c) shows the first 20 samples
of y(n). They are all zero-valued samples! Because y(n) are all zeros, the
Y(m) sequence is also all zeros. Therefore:

the question of what is the frequency index of the

largest magnitude |Y(m)| sample has no meaning.

Property of Pearson Education. Not permissible for redistribution.

 1

0
(a)
w(n)
–1
0 5 n 10 15

1
x(n)
(b) 0

–1
0 5 n 10 15

1
y(n)
(c) 0

–1
0 5 n 10 15

Figure S10–3

Solution: 10.4
(a) The sample rate of y(n) is fs2 = fs1/M = 1000/4 = 250 Hz.

(b) Because 100 Hz is within the unity-gain filter's passband, the peak
amplitude of the 100 Hz sinusoid in the w(n) sequence is P.

(c) Because decimation that does not violate the Nyquist criterion causes no
amplitude loss, the peak amplitude of the 100 Hz sinusoid in the y(m)
sequence is P.

(d) This is a little tricky for beginners in DSP. From Chapter 3, hopefully the
student will remember that DFT magnitudes are proportional to DFT size.
Because the x(n) 100 Hz sinusoid's 4N-point DFT magnitude was K,
the y(m) 100 Hz sinusoid's N-point DFT magnitude will be K/4.
(Magnitude loss by a factor of M = 4.)
The lessons to be learned from this problem, so far, are that decimation of a
finite-length sequence does not cause a change in time-domain amplitudes,
but decimation of a finite-length sequence does cause a loss in frequency-
domain magnitudes.

(e) The equation defining the downsampled y(m) sequence, in terms of w(n), is
y(m) = w(4n).

Solution: 10.5
Property of Pearson Education. Not permissible for redistribution.
 The spectral magnitudes of the complex signals out of the filter pairs, and the problem's
solution of the |Xc(f)| spectrum, are provided in Figure S10–5.

xr(n) uI(n) wI(n)

hcos(k) hHP(k) 3 xI(m)

wQ(n) xc(m)
(a) uQ(n)
hsin(k) hHP(k) 3 xQ(m)

hBP(k) xc(m) = xI(m) + jxQ(m)

|Xr(f)|

(b)
–12 –8 –4 0 4 8 12 kHz
(–fs/2) (fs/2)
|U(f)|

(c)
–12 –8 –4 0 4 8 12 kHz
(–fs/2) (fs/2)
|W(f)|
(d)
–12 –8 –4 0 4 8 12 kHz
(–fs/2) (fs/2)

Solution:
|Xc(f)|

(e)
kHz
–16 –12 –8 –4 0 4 8 12 16
(–2fs) (–fs) (fs) (2fs)

Figure S10–5

Solution: 10.6
(a) Using the text's Eq. (10–3), and Atten = 50, the number of taps in the LPF0
lowpass filter is
Atten 50
N LPF0 ≈ = ≈ 455 taps.
22( fstop − f pass ) 22(2.3/120 − 1.7 /120)

(b) With M = 30, B' = 1700 Hz, and the fstop frequency is 2300 Hz we compute
F = (fstop–B')/fstop = F = 600/2300 = 0.261.
From the text's Eq. (10–2), we estimate the optimum M1 decimation factor
as
1 − MF /(2 − F ) 1 − 30 ⋅ 0.261/(2 − 0.261)
M1,opt ≈ 2M = 60 = 11.05.
Property2 −ofFPearson
( M + 1) Education.
2 − 0.261(30 + 1)
Not permissible for redistribution.
 The integer submultiple of 30 closest to 11.05 is 10, so we set
M1 = 10, and M2 = M/M1 = 30/10 = 3.

(c) Because M1 = 10 the output sample rate of the M1 downsampler is 12000

Hz, so the LPF1 filter, operating at a 120 kHz sample rate, must have the
frequency response shown in Figure S10–6(a). Its fstop frequency is 12000–
1700 = 10300 Hz. Using the text's Eq. (10–3), the number of taps in the
LPF1 filter is
Atten 50
N LPF1 ≈ = ≈ 32 taps.
22( f stop − f pass ) 22(10.3/120 − 1.7 /120)
Because M2 = 3 the output sample rate of the M2 downsampler is 4000 Hz,
so that LPF2 filter, operating at a 12000 Hz sample rate, must have the
frequency response shown in Figure S10–6(b). Its fstop frequency is 4000–
1700 = 2300 Hz. Using the text's Eq. (10–3), the number of taps in the LPF2
filter is
Atten 50
N LPF2 ≈ = ≈ 45 taps.
22( fstop − f pass ) 22(2.3/12 − 1.7 /12)

(d) The number of filter taps using the single-filter decimation system is 455
taps. The number of filter taps using the two-stage decimation system is
32+45 = 77 taps.
The reduction in the number of filter taps using the two-
stage decimation system is 455–77 = 378 taps.

LPF1 response

(a)
0 1.7 6 10.3 12 Freq
(fs,old /10) (kHz)

LPF2 response
(b) B'
0 1 1.7 2.3 3 4 Freq
(fs,new ) (kHz)

Figure S10–6

Solution: 10.7
We need 23 xold(n) input samples to fill the LPF1 filter with data. After that we
need M1 xold(n) input samples to LPF1 for each delay element in the LPF2 filter.
So the number of xold(n) input samples needed to fill both filters with input data
is:
Number of input samples = 23 + M1·(10–1) = 23 + 63 = 86 samples.

Property of Pearson Education. Not permissible for redistribution.

Solution: 10.8
(a) The x(n) sequence should be upsampled (insertion of two zero-valued
samples between each x(n) sample) before lowpass filtering.

(b) Figure S10–8 shows the frequency magnitude response of an ideal (zero
transition region width) interpolation-by-three lowpass filter.

Solution:
Filter mag. response
3
... ...

–2π –π π 2π Rad./sample
(–fs) (–fs/2) –π/3 π/3 (fs/2) (fs) (Hz)
(–fs/6) (fs/6)

Figure S10–8

(c) The filter's DC gain should be 3, as shown in Figure S10–8, to

compensate for the time-domain amplitude loss of three inherent in
the upsample-by-three process.

Solution: 10.9
(a) The equation defining the upsampled y(n) sequence, in terms of x(n), is
⎧⎪ x(n / 3), for n = 0,3,6,9,...
y ( p) = ⎨
⎪⎩ 0, otherwise.

(b) The 24-point Y(m) spectrum of the upsampled-by-3 y(n) sequence is shown
in Figure S10–9.

Solution:
|Y(m)|
4

0
0 7 9 15 17 23 m

Figure S10–9

The above Y(m) spectrum is obtained by merely repeating the original X(m)
spectrum three times over 24 frequency-domain samples.

Solution: 10.10
Property of Pearson Education. Not permissible for redistribution.
 Upsampling (zero sample insertion) by a factor of two results in a spectrum that
is compressed by a factor of two. That is, given the original |Xold(f)| spectrum in
Figure S10–10(a), the spectral magnitude of xnew(m) will be the |Xnew(f)| shown
in Figure S10–10(b). The desired expression for the frequency points of spectral
symmetry of |Xnew(f)| is:

kfs
Points of spectral symmetry = ± 4

where k is an integer.

Points of spectral symmetry

... ...

(a) |Xold(f)|
... ...

–fs –fs/2 0 fs/2 fs Freq

Points of spectral symmetry

... ...
(b)
|Xnew(f)|
... ...

–3fs/4 –fs/2 –fs/4 0 fs/4 fs/2 3fs/4 Freq

Figure S10–10

Solution: 10.11
The spectral magnitudes of the complex sequences at nodes A, B, C, and the
real part of the y(m) output sequence are shown in Figure S10–11.

|X(f)|

(a)
–4 –2 –fs/2 0 fs/2 2 4 Freq
(MHz)

Property of Pearson Education. Not permissible for redistribution.

 Solutions:
A

(b)
–4 –2 –fs/2 0 fs/2 2 4 Freq
(MHz)
B

(c)
–4 –fs/2 –1 0 1 fs/2 4 Freq
(MHz)
C

(d)
–4 –fs/2 –1 0 1 fs/2 4 Freq
(MHz)

Output of 2nd
multiplier
(e)
–4 –fs/2 –1 0 1 fs/2 4 Freq
(MHz)

Solution:
|Y(f)|
(f)
–4 –fs/2 –1 0 1 fs/2 4 Freq
(MHz)

Figure S10–11

Solution: 10.12
(a) The solution to this problem is interpolation combined with appropriate
bandpass filtering as shown in Figure S10–12(a).

Solution:
4
x(t) u(n) v(m) Bandpass w(m) y(m)
A/D 4
(a) filter, h(k)

fs2 = 32 kHz 9 kHz sinusoid

fs1 = 8 kHz at fs2 = 32 kHz

Figure S10–12

(b) Justification of this design is provided in the spectral plots in Figure S10–
12(b) through (e). Interpolation by four exhibits a gain loss of four as shown
by the K/4 value in Figure S10–12(c), so if the bandpass filter has a gain of
one, we need to multiply the w(m) sequence by four.(We can eliminate the
multiplier by designing a filter whose passband gain is four.)
Property of Pearson Education. Not permissible for redistribution.
 Solutions:
|U(f)|
... K ...
(b)
–3 –2 –1 0 1 2 3 kHz
–4 4
(–fs1/2) (fs1/2)
|V(f)|
K/4
... ...
(c)
–12 –8 –4 0 4 8 12 kHz
–16 16
(–fs2/2) (fs2/2)
|W(f)|
... K/4 ...
(d)
–12 –8 –4 0 4 8 12 kHz
–16 16
(–fs2/2) (fs2/2)
|Y(f)|
... K ...
(e)
–12 –8 –4 0 4 8 12 kHz
–16 16
(–fs2/2) (fs2/2)

Figure S10–12 (Cont'd)

Solution: 10.13
(a) Based on the interpretation in Figure S10–13(a), the difference equation for
linear interpolation is
x(n) – x(n–1)
y(n) = x(n–1) + 2 = 0.5x(n) + 0.5x(n–1).

Given this y(n) difference equation, we can write linear interpolation's z-

domain expression for Y(z) as:
Y(z) = 0.5X(z) + 0.5X(z)z–1.
Thus the H(z) transfer function of linear interpolation is expressed as:
Y(z)
H(z) = X(z) = 0.5 + 0.5z–1.

y(n) x(1)

x(0) x(1) – x(0)

(a)
y(1) = x(0) + [x(1) –x(0)]/2

x(0)
0
1 n
Property of Pearson Education. Not permissible for redistribution.
 Figure S10–13

(b) The linear interpolator's frequency response is its H(z) transfer function
with variable z replaced with ejω, or:
H(ω) = 0.5 + 0.5e–jω = 0.5 + 0.5cos(ω) –j0.5sin(ω).
The frequency magnitude response of a linear interpolation filter is
|H(ω)| = |0.5 + 0.5cos(ω) –j0.5sin(ω)|
= [0.5 + 0.5cos(ω)]2 + [0.5sin(ω)]2
= 0.25 + 0.5cos(ω) + 0.25cos2(ω) + 0.25sin2(ω)
= 0.25 + 0.5cos(ω) + 0.25 = 0.5 + 0.5cos(ω) .
A sketch of this |H(ω)| over the frequency range of ω = ±π radians/sample
(±fs/2 Hz) is given as the solid curve in Figure S10–13(b).

Solution:
1
|H(ω)|
0.8
0.6
(b)
0.4
0.2
0
–π/2 –π/4 0 π/4 π/2
(–fs/2) (–fs/4) Frequency (fs/4) (fs/2)

Figure S10–13 (Cont'd)

(c) Linear interpolation;

Advantages:
• Simple to compute. It's merely a two-point moving averager.
• The multiplies by 0.5 can be performed by binary data right shifts
requiring no actual multiplication operations.

Disadvantage:
• Linear interpolation's frequency magnitude response is highly inferior
to a high-performance interpolation by two lowpass filter, shown as
the dashed lines in Figure S10–13(b).
• Least accurate of all methods of interpolation.

Solution: 10.14
Sample rate conversion by a rational factor establishes the following
relationships:
L ⋅ f s ,in L ⋅ f s ,CD
f s ,out = = f s ,DAT =
.
PropertyMof Pearson Education.
M Not permissible for redistribution.
 or
L ⋅ f s ,CD 160 ⋅ 44100
M = = = 147.
f s ,DAT 48000

Solution: 10.15
The solutions to this problem are:

(1) The time duration of interpolated sequences will equal the

time duration of the original sequence.

(2) The time duration of decimated sequences will be less than,

or equal to, the time duration of the original sequence.

Verification of the solutions are as follows: Concerning interpolation, its very

nature means the time duration of the interpolated sequence will equal the time
duration of the original sequence. On the other hand, the time duration of a
decimated sequence will be less than or equal to the time duration of the
original sequence depending on whether the number of samples in the original
sequence was odd or even. Figure S10–15 illustrates these principles.

2
Length = 20
0
xo(n)
Decimation of an –2
even-length
0 5 n 10 15
sequence. Time ts = 1 msec. Time = 19ts = 19 msec.
durations are not
equal. 2 Decimation of an
Length = 10 odd-length
0 sequence. Time
xD(mD) durations are
–2 equal.
0 2 4 mD 6 8
tsD = 2 msec. Time = 9tsD = 18 msec.

2
Length = 39
0
xI(mI)
–2
0 5 10 15 20 mI 25 30 35
tsI = 0.5 msec. Time = 38tsI = 19 msec.

Figure S10–15

Solution: 10.16
The table should be filled in as follows:

Sample Rate Conversion Gain

Property of Pearson Education. Not permissible for redistribution.
 Time Domain Frequency Domain
Decimation Gain = 1 Gain = 1/M
by M
Because DFT magnitudes are
proportional to time sequence's length,
reducing a time sequence's length by a
factor of M results in the DFT
magnitudes of the shorter sequence to
also be reduced by a factor of M.

Interpolation Gain = 1/L Gain = 1

by L Although there is a time-domain gain
The stuffed zeros due to
upsampling causes the (amplitude) loss of L by upsampling and
output of a unity gain filtering, that loss is canceled in the
lowpass filter to be reduced frequency domain by the L-fold gain of
in amplitude by L. the DFT of a sequence that is L times
longer than the original (un-interpolated)
time sequence.

Solution: 10.17
The desired spectral plots are as follows (dashed curves are spectral replications
at multiples of the sample rate):

Solutions:
|Q(f)| Spectral replications

K/4

(a)
-1600 -1200 -800 -400 0 400 800 1200 1600 Freq
(fs) (Hz)
-100 100

Spectral images Spectral

Notice the solid and (Must be solid curves!) replication
dashed curves! |W(f)|
K/4

(b)
-1600 -1200 -800 -400 0 400 800 1200 1600 Freq
(fs) (Hz)
-100 100

Notice the Filter

passband |H(f)|
4 magnitude
gain = 4 response
(c)
-1600 -800 0 800 1600 Freq
(fs) (Hz)

Figure S10–17

Property of Pearson Education. Not permissible for redistribution.

 (d) The answer is no: Upsampling (zeros insertion) by 4 does not reverse the effect
of downsampling by 4 because |W(f)| is not equal to |X(f)|. Time sequences x(n)
and w(p) have the same sample rate but not the same spectra.

Solution: 10.18
(a) The frequency responses of the L(f) lowpass and H(f) highpass filters are
shown in Figure S10–18(a). The system's first scrambling network is shown
in Figure S10–18(b), with signal nodes A through F appropriately marked.
The spectral magnitude of the x(n) input to the first scrambling network is
shown in Figure S10–18(c). The spectral magnitudes at the various signal
nodes are shown in Figures S10–18(d) through Figures S10–18(f).

The solution to this Part (a) of the problem, the spectrum of the first
scrambling network's output, is shown in Figure S10–18(g). (Compare that
spectrum with the original X(f) spectrum in Figure S10–18(c).)

|L(f)| |H(f)|
(a)
–8 –4 0 4 8 kHz –8 –4 0 4 8 kHz
(–fs) (fs) (–fs) (fs)

First scrambling network

x(n) A B C
L(f) 2 2 H(f)
y(n)
+
(b) D E F
H(f) 2 2 L(f)

|X(f)|

(c)
–8 –4 0 kHz
4 8
(–fs) (fs)
Node A Node D
(d)
–8 –4 0 4 8 kHz –8 –4 0 4 8 kHz
(–fs) (fs)
Node B Node E
(e)
–8 –4 0 4 8 kHz –8 –4 0 4 8 kHz
(–fs) (fs)
Node C Node F

(f)
–8 –4 0 4 8 kHz –8 –4 0 4 8 kHz
(–fs) (fs)

Property of Pearson Education. Not permissible for redistribution.

 Solution:
|Y(f)|

(g)
–8 –4 0 4 8 kHz
(–fs) (fs)

Figure S10–18

(b) Because the scrambling networks swap the low- and high-frequency
spectral components of their inputs,

the solution to this Part (b) of the problem is the

original X(f) spectrum shown in Figure S10–18(c).

Solution: 10.19
(a) Polyphase filters are useful for interpolation because no unnecessary
multiplications are performed on zero-valued time samples.

(b) The solution is shown in Figure S10–19.

Solution:
x(n) at
fs rate
H0(z) 4
z–1
H1(z) 4

z–1
H2(z) 4
y(m) at
z–1 4fs rate
H3(z) 4

Figure S10–19

Solution: 10.20
This problem can be a bit tricky for a DSP novice to contemplate. The solution
is to maintain the time relationships between the four time-domain sequences.
That is, make sure the sequences are delayed from each other by one sample-
time period, with the oldest (earlier in time) sequence being applied to the "A"
input port. Using standard "z–1" delay elements, the solution is shown in Figure
S10–20(a). (If desired, the system could be re-drawn as that in Figure S10–
20(a).)

Property of Pearson Education. Not permissible for redistribution.

 Solutions:
x(n-3) x(n)
A D
z–1
z–1
x(n-2) Some x(n-1) Some
B C useful
useful y(n) y(n)
z–1 processing z–1 processing
x(n-1) x(n-2)
C B
z–1
z–1
x(n) x(n-3)
D A

(a) (b)

Figure S10–20

Solution: 10.21
(a) In the first system 12 multiplications must be performed for each x(n) input sample. With
30 input samples arriving per second, the total multiplication rate is:

12 multiplies 30 input samples

System# 1 multiplication rate = input sample · second

= 360 multiplications/second.

(b) In the second system 12 multiplications must be performed for every three x(n) input
samples. With 30 input samples arriving per second, the total multiplication rate is:

12 multiplies 30 input samples

System# 2 multiplication rate = 3 input samples · second
= 120 multiplications/second.

Solution: 10.22
The decimation by four filter using the switches is shown in Figure S10–22(a),
and a decimation by four polyphase filter having 12 multipliers is shown in
Figure S10–22(b). Let's define ts as the time between xold(n) samples.

In real-time implementations, from a time-domain computational standpoint, the

fundamental difference between the Figure S10–22(b) filter and the polyphase
decimation filter is:
• The Figure S10–22(b) filter must be capable of performing 12
multiplies and 11 additions within a period of ts seconds.
• The polyphase filter need perform only 3 multiplies and 2
additions within a period of ts seconds.

Property of Pearson Education. Not permissible for redistribution.

 xold(n) x(n–1) x(n–2) x(n–11)
z–1 z–1 z–1 ... z–1

h(0) h(1) h(2) h(11)

(a)

xnew(m)

z–1 z–1
h(3) h(7) h(11)

z–1 z–1

xold(n) h(0) h(4) h(8)

+
(b)
z–1 z–1
xnew(m)
h(1) h(5) h(9) +

z–1 z–1
h(2) h(6) h(10)

Figure S10–22

Solution: 10.23
(a) Polyphase filters are useful for decimation because they only
compute output samples that will be retained. No unnecessary
computations are performed.

(b) The solution is shown in Figure S10–23.

Property of Pearson Education. Not permissible for redistribution.

 Solution:
x(n) at fs
rate
4 H0(z)
z–1
4 H1(z) y(m) at fs/4
rate
z–1 +
4 H2(z)
–1
z
4 H3(z)

Figure S10–23

Solution: 10.24
(a) Resampling by the rational factor 5/4 mandates that we perform interpolation by L = 5
followed by decimation by M = 4. With output y(m)'s index is m = 7, the commutating
switch's port position value (index) k is computed as

k = <mM>L = <7·4>5 = <28>5 = 3.

where the subscripted 5 means modulo-5. The index n of the most recent
x(n) input sample applied to the subfilters, when m = 7, is computed as

n = ⎣mM/L⎦ = ⎣7·4/5⎦ = ⎣28/5⎦ = 5.

So the solution is k = 3, and n = 5.

(b) Interpolation by L has an inherent gain loss by a factor of L. For the resampler to have a DC
(zero Hz) gain of unity the original prototype lowpass FIR filter must have a
DC gain of L = 5.

Solution: 10.25
Given the spectrum of the filter output a(n) sequence shown in Figure S10–
25(a), the spectra of the b(n), c(m), and y(p) sequences are those shown in
Figures S10–25(b), (c), and (d).

|A(f)|

... ...
(a)
–1600 –800 –400 0 400 800 1600 Freq
(fs)

Property of Pearson Education. Not permissible for redistribution.

 Solutions:
|B(f)|

... ...
(b)
–800 –400 0 400 800 Freq
(fs)
|C(f)|

... ...
(c)
–800 –400 –200 0 200 400 800 Freq
(fs)
|Y(f)|

... ...
(d)
–400 –200 0 200 400 Freq
(fs)

Figure S10–25

Solution: 10.26
(a) The answer is no, the shapes of a comb filter's passband curves are not a
function of f2 which would make them parabolic. We prove this statement
by finding the equation for the frequency response of a comb filter. That
frequency response is the Hcomb(z) transfer function evaluated on the unit
circle. We start by substituting ejω for z in Hcomb(z), because z = ejω defines
the unit circle, giving
Hcomb(ejω) = Hcomb(z)|z=ejω = (1 - e-j8ω).
Factoring out the half-angled exponential e-jω8/2, we have
Hcomb(ejω) = e-jω8/2 (ejω8/2 - e-jω8/2) = e-j4ω (ej4ω - e-j4ω).
Using Euler's identity 2jsin(α) = ejα - e-jα, we arrive at
Hcomb(ejω) = e-j4ω [2jsin(4ω)] = je-j4ω [2sin(4ω)]
or
Hcomb(f) = je-j4ω [2sin(8πf)].
Ignoring the phase shift term (complex exponential) above, the frequency-
domain magnitude response of a comb filter is

|Hcomb(ejω)| = |2sin(4ω)|

or
|Hcomb(f)| = |2sin(8πf)|.

So we see that the |Hcomb(f)| magnitude response shape is sinusoidal, not

parabolic.
Property of Pearson Education. Not permissible for redistribution.
 (b) From the above |Hcomb(f)|, the peak value of the frequency magnitude curve
is P = 2.

Solution: 10.27
The impulse response of the decimation CIC filter, shown in Figure S10–27(a),
can be obtained from the y(n) column in the table in Figure S10–27(b). At time
index n = 5 sequence w(n–5) becomes all ones and the comb's output becomes
all zeros. The y(n) impulse response is plotted in Figure S10–27(c).

Solution:
n x(n) w(n) w(n–5) y(n)
Decimation CIC filter –1 0 0 0 0
Integrator Comb 0 1 1 0 1
1 0 1 0 1
x(n) y(n) 2 0 1 0 1
w(n)
3 0 1 0 1
–
4 0 1 0 1
z–1 z–5 5 0 1 1 0
6 0 1 1 0
w(n-1) w(n-5)
7 0 1 1 0
(a) (b)

Decimation CIC filter impulse response

1
(c)
0.5
0 ...
0 5 10
n

Figure S10–27

Solution: 10.28
(a) The solution for this part of the problem are the u(n) and y(n) sequences for
the D = 4 interpolation and decimation CIC filters shown in Figure S10–28.

x(n) input step signal

1 ...
0.5
0
0 5 n 10

Property of Pearson Education. Not permissible for redistribution.

 Solutions:
u(n) in interpolation CIC filter y(n) in interpolation CIC filter
6
1 ...
4
0.5
2
0 ...
0 5 10 0
n 0 5 n 10

u(n) in decimation
.
CIC filter .. y(n) in decimation CIC filter
10
6
8 ...
4
6
2
4
0
2 0 5 n 10
0
0 5 n 10

Figure S10–28

The u(n) sequence in the decimation filter is unbounded, but this does not
mean that decimation CIC filters are not used in practice. (We never apply a
step sequence to CIC decimators in practice. In practice, we apply signals
that have both positive and negative sample values to CIC decimation
filters.)

(b) The number of binary bits needed to accommodate the u(n) and y(n) samples,
for each CIC filter, up to time index n = 500 are shown in Table S10–2. At
n = 500, u(500) = 501 requiring a 9-bit memory location (or hardware register)
for the decimation CIC filter.

Memory, or hardware register, bit-width requirements

Sequence Interpolation Decimation
CIC filter CIC filter
u(n) 1 bit 9 bits
y(n) 3 bits 3 bits

(c) The CIC decimation filter requires a larger integrator (accumulation)

register than does the CIC interpolation filter even though the two
CIC filters have identical time-domain impulse responses and
identical frequency-domain frequency responses.

Solution: 10.29
(a) This problem is solved by evaluating the 5th-order CIC filter's frequency
magnitude response equation at the frequency fo indicated by Atten. That
equation is
Property of Pearson Education. Not permissible for redistribution.
 M
⎪ sin(πfD) ⎪
|Hcic,Mth-order(f)| = ⎪ ⎪
⎪ sin(πf) ⎪
with M = 5 and D = 6. Because the first CIC filter magnitude null is at
fs,in/D = 0.1667 as shown in Figure S10–29, our fo frequency of interest
(normalized to fs,in) is
fo = 1/D – B/2 = 0.1667 – 0.04/2 = 0.1467
where bandwidth B is also normalized to fs,in.

0
|HCIC(f)|

B Atten
dB

–0.1fs,in 0 0.1fs,in 0.2fs,in 0.3fs,in 0.4fs,in fs,in/2

Frequency (before decimation)

fs,in/D – B/2 = 0.1467fs,in fs,in/D = 0.1667fs,in

Figure S10–29

Evaluation |Hcic,Mth-order(f)| at fo = 0.1467 yields

5
⎪ sin(0.1467π/6) ⎪
|Hcic,Mth-order(0.1467)| = ⎪ ⎪ = (0.828)5 = 0.389.
⎪ sin(0.1467π) ⎪
The peak magnitude response of the filter is DM = 65, so the magnitude
response at fo = 0.1467 in decibels is

Atten = 20 log10( 0.389/65) = –86 dB,

or 86 dB below the CIC filter's passband peak, which is what we set out to
find.

(b) The Gain loss is found in a similar manner by evaluating the 5th-order CIC
filter's frequency magnitude response equation at the normalized frequency
f = B/2 = 0.02. Doing so we have
5
⎪ sin(0.02π/6) ⎪
|Hcic,Mth-order(0.02)| = ⎪ ⎪ = (5.863)5 = 6.926x103.
⎪ sin(0.02π) ⎪
The magnitude response at f = 0.02 in decibels is
Gain Loss = 20 log10( 6.926x103/65) = –1 dB,
or 1 dB below the passband peak, which is what we set out to find.

Solution: 10.30
The number of unit-delay elements in the upper path of the two-path filter
must beProperty
equal to of
thePearson Education.
group delay of the CICNot permissible
filter. formay
The student redistribution.
recall
that a D = 9 CIC filter is equivalent to a D = 9-tap tapped-delay line moving
average filter (having symmetrical coefficients) as shown in Figure S10–30.
Because that moving average filter (and the CIC filter) has eight delay
elements, its group delay is:

Number of delay elements 8

Group delay = 2 = 2 = 4 samples.

Thus the problem solution is that

the upper path of the parallel filter must contain four unit-delay elements.

x(n)
z–1 z–1 z–1 z–1 z–1 z–1 z–1 z–1

y(n)

Figure S10–30

The more mathematical solution is as follows:

The number of unit-delay elements in the upper path of the two-path filter
must be equal to the group delay of the CIC filter. The group delay of a
filter, as described in Appendix F, is defined as the negative of the
derivative of the filter's phase response with respect to frequency. The
frequency response of the single-stage CIC filter is
sin(πfD)
Hcic(f) = e–jπf(D–1)
sin(πf)
where the frequency variable f is in the range of –0.5 to 0.5. The phase
response of Hcic(f) is

Hø(f) = –πf(D–1) = –πf(9–1) = –8πf.

Given that Hø(f), as shown in Appendix F, the group delay of this filter is:

–1 d(Hø(f)) –1 d(–8πf)
G(f) = · d(f) = · d(f) = 4 samples.
2π 2π

So, for a D = 9 CIC filter in the bottom path, the upper path of the parallel
filter must have four unit-delay elements.

Property of Pearson Education. Not permissible for redistribution.

 Chapter 11 Solutions
Solution: 11.1
(a) With N = 4, the average of x(n) is

1 N 1 4 1
xave = N ∑ x(n) = 4 ∑ x(n) = 4 [1 + 2 + 3 + 4] = 2.5.
n=1 n=1

(b) The variance of x(n) is

1 N
xvar = σ = N ∑ [x(n) – xave]2 , or
2

n=1

1
xvar = 4 [(–1.5)2 + (–0.5)2 + (0.5)2 + (1.5)2] = 1.25.

(c) The standard deviation of x(n) is

xstd = xvar = 1.25 = 1.118.

Solution: 11.2
(a) By its inherent nature, xave is the single value such that the sum of
diff(n) = x(n) – xave is zero. That is,
6 6
∑ diff(n) = ∑ [x(n) – xave] = 0.
n=1 n=1

So for our problem,

–3 + (–6) + (–1) + (–8) + 2 + diff(6) = 0.
Or,
diff(6) = 16.

(b) Given diff(6) = x(6) – xave,

x(6) = diff(6) + xave = 16 + 4 = 20.

Solution: 11.3
The answer is yes, it is valid to state that zave = xave + yave.
We verify this as follows: Given that
1 N 1 N
zave = N ∑ z(n) = N ∑ [x(n) + y(n)]
Property
n=1 of Pearson
n=1 Education. Not permissible for redistribution.
 [x(1) + y(1)] + [x(2) + y(2)] + [x(3) + y(3)] + ... + [x(N) + y(N)]
= N ,

we can write
x(1) + x(2) + x(3) + ... + x(N) + y(1) + y(2) + y(3) + ... + y(N)
zave = N

x(1) + x(2) + x(3) + ... + x(N) y(1) + y(2) + y(3) + ... + y(N)
= N + N

1 N 1 N
= N ∑ x(n) + N ∑ y(n) = xave + yave.
n=1 n=1
which is what we set out to prove.

Note: The statement that zave = xave + yave tells us that the average of a signal-
plus-additive-noise is equal to the average of the noise-free signal plus the
average of the noise signal. If the noise signal's average is zero (noise
samples that are equally likely to be plus or minus), we can measure the
average of a signal by computing the average of the signal-plus-additive-
noise. This is what signal averaging is all about!

Solution: 11.4
The average phase angle (in degrees) of π/4 radians, –3π/4 radians, and –π/4
radians is found by first finding the average of three complex numbers
having the appropriate arguments, as:
ejπ/4 + e–j3π/4 + e–jπ/4
Averagecomplex = ( 3 )

= ( (0.7071 + j0.7071) + (–0.7071 3– j0.7071) + (0.7071 – j0.7071) )

= ( 0.7071 –3j0.7071 ) = (0.2357 – j0.2357) = e –jπ/4

.

The argument of e–jπ/4 is –π/4 radians, so the average phase angle in radians
is:
Average phaseradians = tan–1(e–jπ/4) = –π/4 radians.

Our average angle solution, measured in degrees, is:

Average phasedegrees = (Average phaseradians) · ( 180 degrees

π radians
), or

Average phasedegrees = (–π/4 radians) · ( 180 degrees

π radians
) = –45 degrees.

Property of Pearson Education. Not permissible for redistribution.

Solution: 11.5
FFT magnitude variance reduction is expressed by the text's Eq. (11–19),
repeated here as:
σ2 k FFTs 1
Variance reduction = 2 = k .
σ single FFT
Because our desired variance reduction is a factor of 1/20, we write:
1 1
Variance reduction = k = 20 , and

k = 20 FFTs.

Solution: 11.6
Using the geometric series-to-closed form expression description in Appendix
B, we can write Hma(z) as

1 N–1 1 z–1(0) – z–1(N)

Hma(z) = N ∑ z–n = N 1 – z–1
n=0

1 1 – z–N
= N 1 – z–1 = Hrma(z)

which is what we set out to show.

Solution: 11.7
To determine the frequency magnitude responses of a recursive running sum
filter when N = 4, 8, and 16, we take the DFT of the filter's impulse response for
the various N values. Those impulse responses are rectangular sequences of all
ones, with lengths of 4, 8, and 16. The DFT of those impulse responses are
sin(x)/x-like functions whose mainlobe amplitudes are equal to N, with
magnitude nulls at multiples of fs/N, as shown in Figure S11–7.

Solutions:
Recursive running sum
frequency magnitude response
16
N = 16
12
N=8
8
N=4
4

0
0 fs/8 fs/4 fs/2

Property of PearsonFigure S11–7 Not permissible for redistribution.

Education.
Solution: 11.8
The phase responses of both nonrecursive and recursive N-point
moving averagers are linear because their (identical) time-
domain impulse responses are symmetrical.

Solution: 11.9
The solution to this problem is the bold curve shown in Figure S11–9(b). Over
the positive frequency range, the location of the N = 3 averager's z-plane zero is
found using the text's Eq. (11–26) for k = 1. That zero location is
2π 2π
θzeros = k N = 3 radians

as shown in Figure S11–9(a). From the text's Eq. (11–26'), the cyclic frequency
associated with an angle of 2π/3 radians is fs/3 Hz as shown by magnitude null
at fs/3 Hz in the bold curve in Figure S11–9(b). The magnitude at fs/2 (half the
sample rate) is found using the text's Eq. (11–25) with f = 0.5 as

1 sin(πfN) 1 sin(3π/2) 1 –1 1
|Hma(f)| = N · | | = 3 ·| | = 3 ·| 1 |= 3 .
sin(πf) sin(π/2)

Solution:
θ = 2π/3 radians |Hma(f)|
= fs/3 Hz 1

1 0.8
Imaginary part

N=3 N=4 N=2

θ 0.6
N=3
0 0.4 1
0.2 3
–1
–1 0 1 0
Real part 0 fs/8 fs/4 fs/3 Freq fs/2
(a) (b)

Figure S11–9

Solution: 11.10
(a) Cascading an N = 4 moving averager and an N = 2 moving averager results
in a cascaded network whose frequency magnitude response is the product
of the individual moving averagers' frequency magnitude responses as
shown in by the bold curve in Figure S11–10(a).

Property of Pearson Education. Not permissible for redistribution.

 Solution:
Moving averager frq. mag. responses
1
0.8
N=2
0.6
(a) N=4
0.4
0.2 |Hma, cascade(f)|
0
0 fs/8 fs/4 3fs/8 fs/2
Freq

Figure S11–10

(b) The phase of the cascaded (two-stage) filter is linear because the product of
two linear-phase frequency responses is itself linear phase.

One way to show this is as follows: Assume two linear-phase cascaded

filters have frequency responses, each comprising a magnitude and a phase
factor, of
|M1(f)|ejKf and |M2(f)|ejLf
where f is the normalized frequency variable (in the range of –0.5 to 0.5
corresponding to a cyclic frequency range of –fs/2 to fs/2 Hz) and K, and L
are constants. Then the cascaded filter will have a frequency response of
[|M1(f)|ejKf ].[|M2(f)|ejLf ] = |M1(f)|.|M2(f)|ej(Kf + Lf).
The phase response of the cascaded filter is
θ(f) = Kf + Lf = (K + L)f radians
which is a linear function of frequency, and that's what we set out to show.

Another way to show that the phase of the cascaded (two-stage) filter is
linear is to remember that the combined hcasc(k) impulse response of two
cascaded filters is the convolution of their individual impulse responses. As
shown in Figure S11–10(b), convolving the two moving averagers' impulse
responses yields a cascaded hcasc(k) impulse response that is symmetrical,
meaning that the cascaded (two-stage) filter exhibits linear phase.

h1(k)
0.5
0.25
0 hcasc(k)
0 1 2 3 k 0.5
(b) Convolve 0.25
h2(k) 0
0.5 0 1 2 3 4 k
0.25
0
0 1 k

Figure S11–10 (Cont'd)

Property of Pearson Education. Not permissible for redistribution.
Solution: 11.11
(a) We proceed as follows: A noise variance reduction of 13 dB corresponds to
a linear noise reduction factor as
13 = 10log10(Linear reduction factor).
So
13
log10(Linear reduction factor) = 10 ,
and
Linear reduction factor = 1013/10 = 19.95 ≈ 20.
Based on the text's Eq. (11–9'), we can achieve a factor of
σ2ave 1 1
2 = N = 20
σ in
noise variance reduction when N = 20.

The number of delay elements needed in an N = 20 nonrecursive

moving average filter is N–1 = 19 delays.

(b) The number of delay elements needed in an N = 20 recursive moving

average filter is N = 20 delays.

(c) There are three ways to determine an exponential averager's weighting

factor α to achieve a noise variance reduction of 13 dB.

[1] We could examine Figure 11–14 and see that an

α = 0.1
value would achieve the desired noise reduction.
[2] Or we could assign R = 20 in the text's Eq. (11–30) as
2 2
α = R + 1 = 20 + 1 = 0.0952

[3] Or, assign N = 20 in the text's Eq. (11–31) as

2 2
α = N + 1 = 20 + 1 = 0.0952.

(d) The implementation requirements table is as follows:

Table S11–1 Implementation requirements for 13 dB noise reduction

Computations per Nonrecursive Recursive Exponential
output sample and moving moving averager
memory requirements averager averager
Multiplies 1 1 2
Additions 19 2 1
Property of Pearson Education. Not permissible for redistribution.
 Data memory locations 19 20 1

Solution: 11.12
When α = 0 the filter's feed forward coefficient is zero and the filter output
samples will be an all-zero sequence. (What filter could be more stable than
this?) Thus we would never actually use a value of α = 0.

Solution: 11.13
The alternate (feedback only) exponential averager's transfer function is:

Y(z) 1
H(z) = W(z) = .
1– (1–α)z–1

Replacing z with ejω yields the feedback loop's frequency response of:

1
H(ω) = .
1– (1–α)e–jω

At zero Hz, ω = 0, the feedback loop's frequency response (its DC gain) is:

1 1 1
H(ω)|ω=0 = –j0 = = .
1– (1–α)e 1– (1–α) α

which is what we set out to prove.

Solution: 11.14
Given the averager's frequency response of
α
Hexp(ω) =
1 – (1–α) cos(ω) + j(1–α).sin(ω)
.
the desired frequency magnitude response is found by finding the magnitude
expression for the complex denominator of Hexp(ω) as:
⎪ α ⎪
|Hexp(ω)| = ⎪ . . ⎪
⎪ 1 – (1–α) cos(ω) + j(1–α) sin(ω) ⎪
α
=
[1 – (1–α).cos(ω)]2 + [(1–α).sin(ω)]2
Squaring the terms in brackets, our solution is:
α
|Hexp(ω)| = .
1 – 2.(1–α).cos(ω) + (1–α)2

Solution: 11.15
Property of Pearson Education. Not permissible for redistribution.
 When a unit impulse sequence (1,0,0,0,0,...,) input is applied to the standard
integrator, the unity-valued sample's contribution to the y(n) output sequence
lasts forever as shown in Figure S11–15(a). When a unit impulse is applied to
an exponential averager, the unity-valued sample makes an immediate
contribution to the averager's output sequence, however that sample's
contribution to the output diminishes with time (it leaks away) as shown in
Figure S11–15(b).

Exp. averager impulse

Rectangular rule integrator response, α = 0.4
0.4
impulse response
0.4
0.3
0.3
0.2 0.2

0.1 0.1
0
0 1 2 3 4 5 6 7 8 9 0
n 0 1 2 3 4 5n6 7 8 9
(a) (b)
Figure S11–15

Solution: 11.16
(a) The exponential averager's response, to a unity-valued input sample applied
at time n = 0, is
h(0) = α,
h(1) = α(1–α),
h(2) = α(1–α)(1–α) = α(1–α)2,
h(3) = α(1–α)(1–α)(1–α) = α(1–α)3,
h(4) = α(1–α)(1–α)(1–α)(1–α) = α(1–α)4,
etc.
The student should see a pattern occurring here, and write:
h(0) = α,
h(n) = α(1–α)n, for n > 0.
The above h(0) and h(n) results are an acceptable solution to this problem.
However, a perceptive student may realize that h(0) = α = α(1–α)0, and
write the h(n) impulse response in the more concise form of:
h(n) = α(1–α)n, for n ≥ 0.
which is what we seek.

Note: The hyperperceptive student may notice that there is a problem with
the h(n) formula when α = 1, because then h(0) = 00. Now, 00 is sometimes
Property
regarded of Pearson
as indeterminate. AsEducation.
Antoine TruxNot permissible
points for redistribution.
out for example, the
 Standard C library pow(x,y) (which computes xy) generates an error if
x = y = 0. However, most mathematicians now consider 00 to be equal to 1,
and this agrees well with the first output sample (h(0) = 1) of our impulse
response when α = 1.

(b) Here we hope the student remembers that a filter's gain at zero Hz (DC
gain) is the sum of the filter's impulse response samples. Using the h(n)
expression from Part (a), we can write the exponential averager's gain at 0
Hz (DC gain) as:

∞ ∞
Hexp(ω)|ω=0 = Hexp(0) = ∑ h(n) = ∑ α(1–α)n.
n=0 n=0

That Hexp(0) is a geometric series, and can be converted to a closed form

expression using the technique in Appendix B. Doing so we can write our
solution as
∞
(1–α)0 – (1–α)∞ 1–0
DC gain = Hexp(0) = ∑ α(1–α) = α n
=α = 1.
1 –(1–α) α
n=0

We could have, just as well, evaluated the text's Eq. (11–34) Hexp(ω)
frequency response, repeated here as
α
Hexp(ω) = ,
1 – (1–α)e–jω
at ω = 0 . Doing so we have
α α
DC gain = Hexp(0) = –j0 = = 1.
1 – (1–α)e α
which is what we set out to prove.
Note: We ignore the indeterminate case when α = 0 because the exponential
averager has all zero-valued output samples in that situation.

(c) The averager's gain at zero Hz is independent of the α weighting factor.

Property of Pearson Education. Not permissible for redistribution.

Property of Pearson Education. Not permissible for redistribution.
 Chapter 12 Solutions
Solution: 12.1
(a) Unsigned binary 1100 0111 = 19910.
(b) Sign-magnitude 1100 0111 = –7110.
(c) Two's complement 1100 0111 = –5710.
(d) Offset binary 1100 0111 = 7110.

Solution: 12.2
(a) $A231 = 4152110.
(b) 0x71F = 182310.

Solution: 12.3
Hex numbers $07 and $E2 converted to binary format are:
$07 = 0000 0111 and $E2 = 1110 0010.
To perform the desired subtraction we add the two's complement of 1110 0010
($E2) to 0000 0111 ($07). Doing so yields:

0000 0111 ← ($07)

+0001 1110 ← ($1E), two's compliment of 1110 0010 ($E2)
--------------
0010 0101 ← ($25)
The binary two's complement number 0010 0101 ($25) converted to decimal
gives our problem solution of 3710.

Alternatively, we have: $07 = 710, $E2 = –3010. Performing the subtraction in

decimal yields our solution of:

710 ← ($07)
–(–3010) ← ($E2)
--------------
3710

Solution: 12.4
(a) Converting $45 to binary format results in: 0100 01012.
Sign-extending to sixteen bits yields: 0000 0000 0100 01012.
Converting back to hex format gives the solution: $0045.
Property of Pearson Education. Not permissible for redistribution.
 (b) Converting $B3 to binary format results in: 1011 00112. (negative number!)
Sign-extending to sixteen bits yields: 1111 1111 1011 00112.
Converting back to hex format gives the solution: $FFB3.

Solution: 12.5
Using the symbol "◊" to represent a binary point:

(a) In a binary 7.1 two's complement fractional format, the decimal

representation of the addition is as follows:

0000 111◊1 → 7.510

+0000 110◊1 → 6.510
---------------
0001 110◊0 → 14.010 Correct decimal results.

(b) In a binary 6.2 two's complement fractional format, the decimal

representation of the addition is as follows:

0000 11◊11 → 3.7510

+0000 11◊01 → 3.2510
---------------
0001 11◊00 → 7.010 Correct decimal results.

(c) In a binary 4.4 two's complement fractional format, the decimal

representation of the addition is as follows:

0000 ◊1111 → 0.937510

+0000 ◊1101 → 0.812510
---------------
0001 ◊1100 → 1.7510 Correct decimal results.

Solution: 12.6
With 16-bit data words (b = 16), in the two's complement integer number format
the most positive representable decimal number is:
most positive = 2b–1–1 = 215 –1 = 32,76710
and the most negative representable decimal number is:
most negative = –2b–1 = –215 = –32,76810.

Solution: 12.7
The solution is given in the following tables:

Binary Decimal Binary Decimal

0000 0.00 1000 2.00
Property of Pearson Education. Not permissible for redistribution.
 0001 0.25 1001 2.25
0010 0.50 1010 2.50
0011 0.75 1011 2.75
0100 1.00 1100 3.00
0101 1.25 1101 3.25
0110 1.50 1110 3.50
0111 1.75 1111 3.75

Solution: 12.8
Multiplying decimal 0.165 by decimal 32768 gives 540.672. Rounding that
product to the nearest integer yields 541. Converting decimal 541 to a binary
integer proceeds as follows:

Decimal 541 to binary 1.15-format conversion

Operation Binary word
541 ÷ 2 = 270 + remainder of 0.5 → 1
270 ÷ 2 = 135 + remainder of 0 → 01
135 ÷ 2 = 67 + remainder of 0.5 → 101
67 ÷ 2 = 33 + remainder of 0.5 → 1101
33 ÷ 2 = 16 + remainder of 0.5 → 1 1101
16 ÷ 2 = 8 + remainder of 0 → 01 1101
8 ÷ 2 = 4 + remainder of 0 → 001 1101
4 ÷ 2 = 2 + remainder of 0 → 0001 1101
2 ÷ 2 =1 + remainder of 0 → 0 0001 1101
1 ÷ 2 = 0 + remainder of 0.5 → 10 0001 1101
Append six MSB zeros 0000 0010 0001 1101

Placing the binary point to the right of the MSB of the 16-bit binary number
gives us our binary and hexadecimal solutions of
Tax rate = 0◊000 0010 0001 1101 = 021D16 ($021D).

Solution: 12.9
The number system allowing decimal 1/3 to be exactly represented with a
finite number of digits must have a base that is a multiple of three.

For example; using fixed-point numbering systems having bases of three, six,
and nine, decimal 1/3 can be represented by
1/310 = 0.13 = 0.26 = 0.39.

Solution: 12.10
This is a trick question! In a base 6 numbering system, the
digit "7" in 42736 has no meaning (is not defined).
Property of Pearson Education. Not permissible for redistribution.
Solution: 12.11
(a) To a sixteen decimal digit precision, the maximum positive decimal value is
2(# of integer bits – 1) – 2–(# of fraction bits) = 2(1 – 1) – 2–31 = 1 – 2–31

= 1 – 4.656612873077393 x 10–10 = 0.9999999995343387....

(b) The maximum negative decimal value is

–2(# of integer bits – 1) = –2(1 – 1) = –20 = –1.

Solution: 12.12
The frequency resolution of the AD9958 chip, in Hz, is the signal
generator's maximum frequency (250 MHz) minus the generator's minimum
frequency (0 Hz) divided by two raised to the power of the number of
frequency-control bits. That is:

250 MHz – 0 MHz

Frequency resolution = 231 = 0.1164 Hz.

Solution: 12.13
The combined data output rate of the digital portion, measured in bytes (8-bit
binary words) per second, of a stereo CD player is:

16 bits 44100 samples/sec.

Data Ratestereo CD = sample · channel · 2 channels(stereo)

= 1.4112x106 bits/sec. = 176.4x103 bytes/sec. = 176.4 kbytes/sec.

Solution: 12.14
As stated in the text's Eq. (12–22) with regard to digital networks, such as
that shown in Figure S12–14, the hardware register containing the y(n)
sequence must have a word width that accommodates a value equal to the
network's DC (zero Hz) gain G times the input signal, i.e., G·x(n). That
number of y(n) bits is
y(n) bits = number of bits in x(n) + ⎡log2(G)⎤

where ⎡log2(G)⎤ means: if log2(G) is not an integer, round it up to the next

larger integer. So we must compute G for the filter in Figure S12–14.

Property of Pearson Education. Not permissible for redistribution.

 x(n) y(n)

z–1

0.85 z–1

–0.12

Figure S12–14

The network's transfer function is:

Y(z) 1 1
H(z) = X(z) = 1 –0.85z–1 + 0.12z–2 = 1 –0.85/z + 0.12/z2 .

Next we set z = 1 giving us our desired DC (zero Hz) gain solution of:
1
G = H(z)|z=1 = 1 –0.85 + 0.12 = 3.704.

Using the G = 3.704 value, the solution to this problem is:

y(n) bits = number of bits in x(n) + ⎡log2(G)⎤

= 8 + ⎡1.89⎤ = 8 + 2 = 10 bits.

Solution: 12.15
The decimal values of all possible filter coefficient values using a four-bit
unsigned binary words in a 2.2 (two dot two) "integer plus fraction" format are
given in the following tables:

Binary Decimal Binary Decimal

0000 0.00 1000 2.00
0001 0.25 1001 2.25
0010 0.50 1010 2.50
0011 0.75 1011 2.75
0100 1.00 1100 3.00
0101 1.25 1101 3.25
0110 1.50 1110 3.50
0111 1.75 1111 3.75

The quantized value for desired coefficient B = 2.5 is BQ = 10102 = 2.510, in

which case there is no BQ-coefficient quantization error.
However, the quantized value for the desired coefficient A = 1/B =
1/2.5 = 0.4 is AQ = 00102 = 0.510, in which case there is coefficient
quantization error. Because AQ ≠ 1/BQ, the quantized allpass filter does
not achieve the desired constant frequency magnitude response.

Property of Pearson Education. Not permissible for redistribution.

Solution: 12.16
The maximum amplification occurs with an arithmetic left-shift by 7 bits.
Performing a left-shift by 7 bits is equivalent to multiplying a number by 2
raised to the 7th power, or:
linear gain = 27 = 128.
Taking 20 times the log of this linear gain yields our solution of :
dB gain =20*log10(128) = 42.1 dB.

Solution: 12.17
As shown in Figure S12–17, the input to the summation operation is
x(n)/22 + x/24 + x/26 + x/28 + x/210 + x/212 + x/214

= x(n)·(1/22 + 1/24 + 1/26 + 1/28 + 1/210 + 1/212 + 1/214)

≈ x(n)·(0.3333) ≈ x(n)/3.

The algorithm approximates division by three, thus integer K = 3.

x(n) 2
2
2
x(n)/22 2
x(n)/24 2
x(n)/26 2
x(n)/28
x(n)/210 2
x(n)/212
x(n)/214

x(n)/3

Figure S12–17

Solution: 12.18
The solution to this problem is A = 3, and B = 1 as shown in Figure S12–18.

9 . x(n)
x(n)
Left shift
by 3 bits y(n) = 54.x(n)
Left shift
by 3 bits –
Left shift
by 1 bit
Multiplication
by 9, (23 + 1) Multiplication
by 6, (23 - 21)

Property of PearsonFigure
Education.
S12–18Not permissible for redistribution.
Solution: 12.19
The data-word bit manipulations necessary to eliminate the two multipliers in
Figure S12–19(a) are negation and binary bit shifting. The desired multiplier-
free implementation is shown in Figure S12–19(b).
• The two's complement negation comprises inverting the bits of the x(n–2)
sample and adding one. The bits of the negated result are arithmetically shifted
left by one bit before being added to x(n) and x(n–4).
• The magnitude bits of the sum, w(n), are then shifted right by two bits.

An "Arithmetic left shift" in Figure S12–19(b) means that a zero bit is inserted
as the new least significant bit (LSB). After that shift, the overflow and MSB
bits must be examined to maintain proper data word polarity. An "Arithmetic
right shift" means that the appropriate bit, to maintain proper polarity, is
inserted as the new most significant bit.

x(n) x(n–2) x(n–4)

z–2 z–2

Two's complement
negation
x(n) x(n–2) x(n–4)
z–2 z–2
Arithmetic left
–2 shift by 1 bit

y(n)
w(n) w(n)

0.25 Arithmetic right y(n)

shift by 2 bits

(a) (b)

Figure S12–19

Solution: 12.20
With a sample rate of fs = 2x109 samples/second, the time period between
x(n) samples is:
1 1
ts = f = 2x109 = 0.5x10–9 seconds.
s

The duration, T, of N time samples is T = N·ts. So N is computed as:

T 5x10–6
N = t = 0.5x10–9 = 104 samples.
s

Thus we need 10000 computer memory locations.

Property of Pearson Education. Not permissible for redistribution.

Solution: 12.21
 Because x(n) is an integer number of cycles of a sinusoid, from the Chapter
3's Eq. (3–17) we find the discrete x(n) sinusoid's peak amplitude as:
2·|Xave(3)| 2·65536
x(n) peak amplitude, Ao = N = 1024 = 128.
Having a 10-bit binary (sign-magnitude format) output, the maximum
positive A/D converter output word has a decimal value of:
max positive A/D output = 210–1–1 = 511.
Because the A/D converter's maximum positive voltage is 5 volts, the peak
voltage of the continuous x(t) sinewave is:
peak amplitude of x(n)·5 volts
peak voltage of x(t) = max positive A/D output

128·5
= 511 ≈ 1.25 volts.

Solution: 12.22
(a) The A/D converters' quantization-level voltage, q, is the converter's full
peak-peak voltage range divided by two raised to a power equal to the
number of converter bits. Thus the quantization-level voltage for a 12-bit
A/D converter is:

10 volts 10 volts
q= 2 12 = 4096 = 0.0024 volts = 2.4 millivolts.

(b) The A/D converters' maximum positive and maximum negative quantization
error (noise) voltages are plus and minus half the converter's quantization
voltage:
±q ±0.0024 volts
max plus and minus quantization error = 2 = 2

= ±0.0012 volts = ±1.2 millivolts.

(c) With a 7-volt peak-peak sinusoidal voltage at the converter's analog input,
we use the text's Eq. (12–13) to compute the converter's SNRA/D output
signal to quantization noise value, in dB. That SNRA/D value is:
SNRA/D = 6.02(12) + 4.77 + 20log10(LF) = 77.01 + 20log10(LF).
The loading factor, LF, is
rms voltage of the input (7/2)(0.707)
LF = peak converter input voltage = 5 = 0.495.

Plugging the LF = 0.495 value into the SNRA/D expression, we compute our
desired SNR result as:

SNRA/D = 77.01 + 20log10(0.495) = 77.01 –6.11 = 70.9 dB.

Property of Pearson Education. Not permissible for redistribution.
 An alternate solution to this problem is to compute the loading factor
(0.495) in dB which is
LFdB = 20log10(0.495) = –6.11
and apply that value to the 12-bit curve in the text's Figure 12–4 as shown in
Figure S12–22,
giving us an alternate solution for SNRA/D of roughly 71 dB.

SNRA/D-max
90
80
71 70
60 12-bit
dB

50 10-bit
40
8-bit
30
6-bit
20

–21 –18 –15 –12 –9 –6 –3 Loading

factor (dB)
–6.11

Figure S12–22

Solution: 12.23
Given that a 12-bit A/D converter's signal to quantization noise level is 67
dB, when driven to its full-scale input voltage range, we compute the
desired effective bit value beff using the text's Eq. (12–16) as:

SNR –1.76 67 –1.76

beff = 6.02 = 6.02 = 10.84 effective bits.

Solution: 12.24
(a) The quantization error as a function of the continuous x(t) input for the
truncating and rounding A/D converters are shown as the bold diagonal lines
in Figure S12–24.

Property of Pearson Education. Not permissible for redistribution.

 Solutions:
x(n), quantized by x(n), quantized by
truncation rounding
3q 3q
2q 2q
q –1.5q q
–3q –2q –q 2q 3q 4q 3.5q
x(t) –3.5q x(t)
–q –q 1.5q
–2q –2q
–3q –3q

Figure S12–24

(b) The table entries of quantization error properties, in terms of the A/D
converters' quantization-level voltage q, are provided in the following table.

A/D Type Most negative Most positive Max peak-peak

quantization quantization quantization error
error error
Truncating –q 0 q
Rounding –q/2 q/2 q

Solution: 12.25
To determine the A/D converter's number of bits, b, to accommodate the
accuracy and operating range of the thermocouple, we first determine the
dynamic range of the measurement system's voltage v(t). That is done as
follows:
Temp. range
Voltage v(t) dynamic range = Temp. resolution

2(212o) –32o 424o –32o 392o

= 2 o = 2 o = 2o = 196.
Next, the linear dynamic range of a b-bit A/D converter is
dynamic rangelinear = 2b – 1,
so the minimum number of bits to accommodate a linear dynamic
range of 196 + 1 is:
196 + 1 ≤ 2b.
Thus:
bminimum = log2(197) = 7.6 bits,

leading us to choose a b = 8-bit A/D converter.

Property of Pearson Education. Not permissible for redistribution.
Solution: 12.26
Assuming that an analog sinewave, whose peak value is 1, is used as an
analog input for A/D converter output histogram testing, the solution to this
problem is the rough sketch of the histogram of converter output samples
shown in Figure S12–26(a).

Solution:
# of occurrences

(a)

0
–1 –0.5 0 0.5 1
Sample value

Figure S12–26

That bathtub-like curve in Figure S12–26(a) shows that there are more
occurrences of A/D output samples whose values are near ±1 than
occurrences of sample values close to zero amplitude.
We can see that this is true from the sinewave A/D converter output in
Figure S12–26(b), where we see that there are more samples in the
amplitude Range-1 and Range-3 (near ±1) than there are in the amplitude
Range-2 (near zero). Histogram testing of A/D converters is described in
more detail in Chapter 13.
The precise histogram of an ideal A/D converter's output samples,
when the converter's input is an analog sinewave whose peak value is one, is
shown in Figure S12–26(c).

1
Range-1
# of occurrences

0.5

0 Range-2

–0.5
Range-3
–1
–1 –0.6 –0.2 0 0.2 0.6 1
0 Time Sample value
(b) (c)

Figure S12–26 (Cont'd)

Solution: 12.27
To determine the value of A of a uniform pdf whose variance is equal to
two, we use Appendix D's Figure D–4, repeated below in Figure S12–27,
and Eq. (D–12) from Appendix D. That equation, which applies to Figure
Property
S12–27, is of Pearson Education. Not permissible for redistribution.
 (b + a)2
2
uniform pdf variance = σ = 12 .

p(f)
1
b+a

–a 0 b f

Figure S12–27

For this problem, the limits of the Figure S12–27 uniform pdf are
a = A, and b = A. So we can write the above σ2 expression as:
(A + A)2 4A2 A2
σ2 = 2 = 12 = 12 = 3 .
Solving the above for A we have:
A2 2
3 = 2, or A = 6,
so
A= 6 ≈ 2.4495.

Solution: 12.28
Given a single data sample value in binary floating point format, we can
perform a multiply-by-4 by:
(i) Shifting the sample's fraction bits to the left by two bits, or
(ii) Incrementing the sample's exponent bits two times.

Solution: 12.29
To solve this problem we use the text's expression
valueIEEE = (–1)s . 1◊ f . 2e–127 ,
where symbol ◊ denotes the binary point and f and e are decimal numbers.
To examine the floating-point number's individual bits, we convert the given
hex number to binary format as:
$C2ED0000 = 1100 0010 1110 1101 0000 0000 0000 0000

= 1|100 0010 1|110 1101 0000 0000 0000 0000

exponent fraction

Our exponent value e is

Property of Pearson
e = 1000 0101 = Education.
13310 . Not permissible for redistribution.
The fraction bits are
110 1101 0000 0000 0000 0000
so our fraction's decimal value f is
f = 2–1 + 2–2 + 2–4 + 2–5 + 2–7 = 0.851562510.
The floating-point number's sign bit s = 1 making the number negative, so
after including the hidden one (to the left of the decimal point) the solution
to this problem is:
valueIEEE = (–1)s . 1◊ f . 2e–127 = –1.8515625 . 2133–127

= –1.8515625 . 64 = –118.510.

Property of Pearson Education. Not permissible for redistribution.

Property of Pearson Education. Not permissible for redistribution.