Visit https://ebookfinal.

com to download the full version and

explore more ebooks

Notes on Digital Signal Processing Practical

Recipes for Design Analysis and Implementation
1st Edition C. Britton Rorabaugh

_____ Click the link below to download _____

https://ebookfinal.com/download/notes-on-digital-
signal-processing-practical-recipes-for-design-
analysis-and-implementation-1st-edition-c-britton-
rorabaugh/

Explore and download more ebooks at ebookfinal.com

Here are some suggested products you might be interested in.
Click the link to download

Digital Signal Processing System Analysis and Design 2nd

Edition Paulo S. R. Diniz

https://ebookfinal.com/download/digital-signal-processing-system-
analysis-and-design-2nd-edition-paulo-s-r-diniz/

Real Time Digital Signal Processing Implementation and

Application 2nd Edition Sen M. Kuo

https://ebookfinal.com/download/real-time-digital-signal-processing-
implementation-and-application-2nd-edition-sen-m-kuo/

Digital Signal Processing International Edition Proakis

https://ebookfinal.com/download/digital-signal-processing-
international-edition-proakis/

Digital Signal Processing 3rd Edition S. Salivahanan

https://ebookfinal.com/download/digital-signal-processing-3rd-edition-
s-salivahanan/
Digital Signal Processing and Applications Second Edition
Dag Stranneby

https://ebookfinal.com/download/digital-signal-processing-and-
applications-second-edition-dag-stranneby/

Digital Signal Processing DSP and Applications 1st Edition

Dag Stranneby

https://ebookfinal.com/download/digital-signal-processing-dsp-and-
applications-1st-edition-dag-stranneby/

Analysis and Design of Quadrature Oscillators Analog

Circuits and Signal Processing 1st Edition Luis B.
Oliveira
https://ebookfinal.com/download/analysis-and-design-of-quadrature-
oscillators-analog-circuits-and-signal-processing-1st-edition-luis-b-
oliveira/

Digital Signal Processing Laboratory Second Edition B.

Preetham Kumar

https://ebookfinal.com/download/digital-signal-processing-laboratory-
second-edition-b-preetham-kumar/

Understanding Digital Signal Processing 2nd Edition

Prentice Hall Ptr

https://ebookfinal.com/download/understanding-digital-signal-
processing-2nd-edition-prentice-hall-ptr/
Notes on Digital Signal Processing Practical Recipes for
Design Analysis and Implementation 1st Edition C.
Britton Rorabaugh Digital Instant Download
Author(s): C. Britton Rorabaugh
ISBN(s): 9780131583344, 0131583344
Edition: 1
File Details: PDF, 28.35 MB
Year: 2010
Language: english
Notes on Digital Signal Processing
This page intentionally left blank
Notes on Digital Signal
Processing
Practical Recipes for Design, Analysis,
and Implementation

C. Britton Rorabaugh

6QQFS 4BEEMF 3JWFS /+ t #PTUPO t *OEJBOBQPMJT t 4BO 'SBODJTDP

/FX :PSL t 5PSPOUP t .POUSFBM t -POEPO t .VOJDI t 1BSJT t .BESJE
$BQF 5PXO t 4ZEOFZ t 5PLZP t 4JOHBQPSF t .FYJDP $JUZ
Many of the designations used by manufacturers and sellers to distinguish their products are
claimed as trademarks. Where those designations appear in this book, and the publisher was
aware of a trademark claim, the designations have been printed with initial capital letters or in all
capitals.

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.

The publisher offers excellent discounts on this book when ordered in quantity for bulk purchases
or special sales, which may include electronic versions and/or custom covers and content
particular to your business, training goals, marketing focus, and branding interests. For more
information, please contact:

U.S. Corporate and Government Sales

(800) 382-3419
[email protected]

For sales outside the United States please contact:

International Sales
[email protected]

Visit us on the Web: informit.com/ph

Library of Congress Cataloging-in-Publication Data

Rorabaugh, C. Britton.
Notes on digital signal processing : practical recipes for design,
analysis, and implementation / C. Britton Rorabaugh.
p. cm.
Includes bibliographical references and index.
ISBN 0-13-158334-4 (hardcover : alk. paper)
1. Signal processing—Digital techniques. I. Title.
TK5102.9.R673 2011
621.382'2--dc22
2010040954
Copyright © 2011 Pearson Education, Inc.

All rights reserved. Printed in the United States of America. This publication is protected by
copyright, and permission must be obtained from the publisher prior to any prohibited reproduction,
storage in a retrieval system, or transmission in any form or by any means, electronic, mechanical,
photocopying, recording, or likewise. For information regarding permissions, write to:

Pearson Education, Inc

Rights and Contracts Department
501 Boylston Street, Suite 900
Boston, MA 02116
Fax (617) 671 3447

ISBN-13: 978-0-13-158334-4
ISBN-10: 0-13-158334-4

Text printed in the United States on recycled paper at Edwards Brothers in Ann Arbor, Michigan
First printing, November 2010
To Joyce
This page intentionally left blank
 Contents

Preface .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . xi
About the Author .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . xiii

Part I DSP Fundamentals

Note 1 Navigating the DSP Landscape .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 1-1
Note 2 Overview of Sampling Techniques .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 2-1
Note 3 Ideal Sampling . ........................................................................................................... 3-1
Note 4 Practical Application of Ideal Sampling .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 4-1
Note 5 Delta Functions and the Sampling Theorem . .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 5-1
Note 6 Natural Sampling .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 6-1
Note 7 Instantaneous Sampling .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 7-1
Note 8 Reconstructing Physical Signals . .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 8-1

Part II Fourier Analysis

Note 9 Overview of Fourier Analysis . .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 9-1
Note 10 Fourier Series . .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. ..10-1
Note 11 Fourier Transform.. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. ..11-1
Note 12 Discrete-Time Fourier Transform .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. ..12-1
Note 13 Discrete Fourier Transform .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. ..13-1
Note 14 Analyzing Signal Truncation .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. ..14-1
Note 15 Exploring DFT Leakage .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. ..15-1
Note 16 Exploring DFT Resolution . .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. ..16-1

vii
viii Contents

Part III Fast Fourier Transform Techniques

Note 17 FFT: Decimation-in-Time Algorithms . .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. 17-1
Note 18 FFT: Decimation-in-Frequency Algorithms.. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. 18-1
Note 19 FFT: Prime Factor Algorithm .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. 19-1
Note 20 Fast Convolution Using the FFT . .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. 20-1

Part IV Window Techniques

Note 21 Using Window Functions: Some Fundamental Concepts . .. .. .. .. .. .. .. .. .. 21-1
Note 22 Assessing Window Functions: Sinusoidal Analysis Techniques .. .. .. .. .. .. 22-1
Note 23 Window Characteristics . .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. 23-1
Note 24 Window Choices. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. 24-1
Note 25 Kaiser Windows . .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. 25-1

Part V Classical Spectrum Analysis

Note 26 Unmodified Periodogram . .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. 26-1
Note 27 Exploring Periodogram Performance:
Sinusoids in Additive White Gaussian Noise .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. 27-1
Note 28 Exploring Periodogram Performance:
Modulated Communications Signals. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. 28-1
Note 29 Modified Periodogram . .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. 29-1
Note 30 Bartlett’s Periodogram . .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. 30-1
Note 31 Welch’s Periodogram . .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. 31-1

Part VI FIR Filter Design

Note 32 Designing FIR Filters: Background and Options.. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. 32-1
Note 33 Linear-Phase FIR Filters .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. ..33-1
Note 34 Periodicities in Linear-Phase FIR Responses .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. 34-1
Note 35 Designing FIR Filters: Basic Window Method . .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. 35-1
Note 36 Designing FIR Filters: Kaiser Window Method .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. 36-1
Note 37 Designing FIR Filters: Parks-McClellan Algorithm.. .. .. .. .. .. .. .. .. .. .. .. .. .. .. 37-1
Contents ix

Part V Analog Prototype Filters

Note 38 Laplace Transform. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 38-1
Note 39 Characterizing Analog Filters. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 39-1
Note 40 Butterworth Filters .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 40-1
Note 41 Chebyshev Filters.. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 41-1
Note 42 Elliptic Filters.. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 42-1
Note 43 Bessel Filters .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 43-1

Part VI z-Transform Analysis

Note 44 The z Transform .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 44-1
Note 45 Computing the Inverse z Transform Using the Partial Fraction Expansion .. .. .. .. .. .. . 45-1
Note 46 Inverse z Transform via Partial Fraction Expansion
Case 1: All Poles Distinct with M < N in System Function .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 46-1
Note 47 Inverse z Transform via Partial Fraction Expansion
Case 2: All Poles Distinct with M ≥ N in System Function (Explicit Approach) .. .. .. .. . 47-1
Note 48 Inverse z Transform via Partial Fraction Expansion
Case 3: All Poles Distinct with M ≥ N in System Function (Implicit Approach) .. .. .. .. . 48-1

Part VII IIR Filter Design

Note 49 Designing IIR Filters: Background and Options .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 49-1
Note 50 Designing IIR Filters: Impulse Invariance Method . .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 50-1
Note 51 Designing IIR Filters: Bilinear Transformation .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 51-1

Part VIII Multirate Signal Processing

Note 52 Decimation: The Fundamentals.. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 52-1
Note 53 Multistage Decimators . ................................................................................................................. 53-1
Note 54 Polyphase Decimators . .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 54-1
Note 55 Interpolation Fundamentals . .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 55-1
Note 56 Multistage Interpolation .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 56-1
Note 57 Polyphase Interpolators.. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 57-1
x Contents

Part IX Bandpass and Quadrature Techniques

Note 58 Sampling Bandpass Signals. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 58-1
Note 59 Bandpass Sampling: Wedge Diagrams . .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 59-1
Note 60 Complex and Analytic Signals.. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 60-1
Note 61 Generating Analytic Signals with FIR Hilbert Transformers . .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 61-1
Note 62 Generating Analytic Signals with Frequency-Shifted FIR Lowpass Filters . .. .. .. .. .. .. . 62-1
Note 63 IIR Phase-Splitting Networks for Generating Analytic Signals .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 63-1
Note 64 Generating Analytic Signals with Complex Equiripple FIR Filters.. .. .. .. .. .. .. .. .. .. .. .. . 64-1
Note 65 Generating I and Q Channels Digitally: Rader’s Approach .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 65-1
Note 66 Generating I and Q Channels Digitally: Generalization of Rader’s Approach . .. .. .. .. . 66-1

Part X Statistical Signal Processing

Note 67 Parametric Modeling of Discrete-Time Signals .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 67-1
Note 68 Autoregressive Signal Models .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 68-1
Note 69 Fitting AR Models to Stochastic Signals: Yule-Walker Method.. .. .. .. .. .. .. .. .. .. .. .. .. .. . 69-1
Note 70 Fitting All-Pole Models to Deterministic Signals: Autocorrelation Method . .. .. .. .. .. . 70-1
Note 71 Fitting All-Pole Models to Deterministic Signals: Covariance Method.. .. .. .. .. .. .. .. .. . 71-1
Note 72 Autoregressive Processes and Linear Prediction Analysis . .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . 72-1
Note 73 Estimating Coefficients for Autoregressive Models: Burg Algorithm . .. .. .. .. .. .. .. .. .. . 73-1

Index . .......................................................................................................................................................................... I-1

 Preface

S tandard advice for writing a preface tells the author to begin by answering the
question, “Why did you write this book?” The published answers almost always
include an explanation of how something is still missing in the already vast body of
existing literature, and how the book in question represents a valiant attempt to fill the
void at least partially.
This book is no exception. There still is a dearth of good collections of step-by-
step procedures, or recipes, for design and implementation of anything beyond just
the most elementary DSP procedures. This book is an attempt to fill this void—at least
partially. However, the tagline for this book is most definitely not meant to be, “Get
a result without really gaining much understanding along the way.” Here, the focus is
clearly on the recipes, but supporting explanations and mathematical material are also
provided. This supporting material is set off in such a way so that it is easily bypassed
if the reader so desires.
This book provides an opportunity to delve deeper into the nuances of certain in-
teresting topics within DSP. A good alternative title might be Exploring the Nooks and
Crannies of Digital Signal Processing. As with all books, every reader will not resonate
with every topic, but I’m confident that each reader will share an interest in a large sub-
set of the topics presented.
Note 1, Navigating the DSP Landscape, provides diagrams that map the relationships
among all the book’s various topics. One diagram is dedicated to processing techniques
that operate on real-valued digital signals to modify in some way the properties of those
signals while leaving their fundamental real-valued and digital natures intact. A sec-
ond diagram is dedicated to processing techniques that are concerned primarily with
conversion between real-valued digital signals and other entities such as analog signals,
complex-valued signals, and estimated spectra.
Many of the Notes include examples that demonstrate an actual application of the
technique being presented. Most sections use Matlab tools for routine tasks such as
designing the digital filters that are used in the reference designs. When appropriate, the

xi
xii Preface

use of these tools is discussed in the text. The results provided in Note 66, Generating I and Q
Channels Digitally: Generalization of Rader’s Approach, were generated by a modified version of
the PracSim simulation package that is described in Simulating Wireless Communication Sys-
tems (Prentice Hall, 2004). However, the filter coefficients used in the simulation were generated
using Matlab. The examples make heavy use of Matlab as a convenience. However, it is not my
intent to make this a Matlab “workbook” with projects and exercises that require the reader to
use Matlab, because I want the book to remain useful and attractive to readers who do not have
access to Matlab. The m-files for the Matlab programs discussed in the book, as well as for
programs used to generate some of the illustrations, can be found at the website www.informit.
com/ph.
This book is not the best choice for a first book from which to learn DSP if you’re start-
ing from scratch. For this task, I recommend Understanding Digital Signal Processing by Rich-
ard Lyons (Prentice Hall, 2004). This book is, however, a good N+1st book for anyone—from
novice to expert—with an interest in DSP. Its contents comprise an assortment of interesting
tidbits, unique insights, alternative viewpoints, and rarely published techniques. The following
are some examples.
t The set of five techniques for generating analytic signals presented in Notes 60 through 64 do
not appear together in any single text.
t The visualization techniques used in Note 22 probably are not discussed anywhere else, be-
cause I came up with them while writing this book. These techniques follow directly from
first principles, but I’ve never seen them explicitly presented elsewhere.
t Natural sampling, as discussed in Note 6, usually can be found only in older texts that cover
traditional (that is, analog) communication theory.
My overarching goal was to write an easy-to-read book loaded with easy-to-access informa-
tion and easy-to-use recipes. I hope I have succeeded.
 About the Author

C. BRITTON RORABAUGH has a B.S. and M.S. in electrical engineering from

Drexel University and currently holds the position of Chief Scientist for a company
that develops and manufactures specialized military communications equipment. He
is an associate editor for IEEE Signal Processing Magazine, and the author of several
publications on topics such as wireless communications, DSP, digital filters, and error
coding, with experience in communication system design and analysis, geolocation,
radar processing, real-time software, numerical methods, computer graphics, C++, C,
Matlab, and assembly languages for various microprocessors and DSP devices.

xiii
This page intentionally left blank
Note 1

Navigating the DSP Landscape

D igital signal processing (DSP) is based on the

notions that an analog signal can be digi-
tized and that mathematical operations can effec-
Hundreds of techniques (and variations thereof)
are used in DSP, and it can be difficult to see the big
picture—how all these various techniques relate to
tively take the place of (or even surpass) electronic each other and to a particular application at hand.
manipulations performed on the original analog Rather than a comprehensive treatment of all the
signal. In the earliest days of DSP, its applications possible topics within DSP, this book is an attempt
were limited to sonar and seismology because these to document in-depth explorations of some of the
fields utilized low-bandwidth signals that could be “nooks and crannies” in DSP that often are glossed
sampled at adequate rates using the available tech- over in traditional texts. Figures 1.1 and 1.2 show
nology. As digital processing circuits and analog- diagramatically the realtionships among the vari-
to-digital converters have become faster and faster, ous processing techniques explored in this book.
the number of applications for DSP has exploded. The topic areas are arbitrarily split into two groups

Window
Functions

N21 - N25

FIR Filters
optimal designs
N32-N34, N37 FIR Filters
designs based
on windows
N35, N36

Real-
Valued
Digital
Signals z Transform
IIR Filters used in IIR
Decimation derived from design &
to decrease analog designs analysis
sample rate
N49 - N51 N44 - N48
N52 - N54
Interpolation
to increase
sample rate
N55 - N57 Analog Filter
Designs

N38 - N43

Figure 1.1 Processing techniques that modify the properties of real-valued digital signals. The
numbers “Nnn” indicate the Notes in which each technique is discussed. Solid paths indicate “run-
time” data connections. Dashed paths indicate “design-time” connections.

1-1
1-2 Notes on Digital Signal Processing

for organizational purposes. Figure 1.1 shows those most of these techniques are easily extended to cor-
techniques that are concerned primarily with op- responding complex cases. Figure 1.2 shows those
erating on real-valued digital signals to modify in techniques that are concerned primarily with con-
some way the properties of those signals while leav- version between real-valued digital signals and
ing their fundamental real-valued and digital na- other entities such as analog signals, complex-
tures intact. Given that complex-valued signals are valued signals, and spectrum estimates.
really just quadrature pairs of real-valued signals,

Natural Sampling
model for analog
signal commutation
N6

Sampling
Analog models for Parametric Modeling of
Signals A/D conversion Discrete-Time Signals
N2 - N5, N58 N67 - N73

Instantaneous Window
Real-
Sampling Functions
Valued
model for Digital N21 - N25
D/A conversion Signals
N7, N8

Generating
Complex & Improved
DFT Sample Periodogram
Analytic Signals Estimate
Spectrum Techniques
N13 - N19 of Signal’s
N60 - N66
N26 - N31 Spectrum

Complex-
Parametric Modeling of
Valued
Discrete-Time Signals
Digital
Signals N67 - N73

Figure 1.2 Processing techniques that convert real-valued digital signals to or from other things such as analog signals,
complex-valued digital signals, or spectral estimates. The numbers “Nnn” indicate the Notes in which each technique is
discussed. Solid paths indicate “run-time” data connections. Dashed paths indicate “design-time” connections.
Note 2

Overview of Sampling Techniques

This note discusses the difference between implicit and explicit sam- 2.2 Explicit Sampling Techniques
pling. It introduces three different mathematical models of explicit
sampling techniques—ideal sampling, natural sampling, and in- Unlike implict sampling, in which samples are trig-
stantaneous sampling. gered by some aspect of signal behavior, in explicit
sampling, signal values are measured at specified

D igitization of an analog signal is the one pro-

cess above all others that makes DSP such
a useful technology. If it were limited to work-
times without regard to the signal’s behavior. Con-
sider the continuous-time sinusoidal signal and its
two-sided magnitude spectrum depicted in Figure
ing with only those signals that orginate in digital 2.1. There are three explicit sampling techniques—
form, DSP would be just an academic curiosity. natural sampling, instantaneous sampling, and ideal
Digitization actually comprises two distinct opera- sampling—that can be used to sample such a signal.
tions: sampling and quantization, which are usually The results produced by these techniques, and the
analyzed separately. corresponding impacts on the signal’s spectrum are
compared in Key Concept 2.1.
2.1 Implicit Sampling Techniques
Ideal Sampling
In implicit sampling, a sample measurement is trig-
As depicted in Key Concept 2.1, zero-width sam-
gered by the signal attaining some specified value
ples take on instantaneous values of the analog
or crossing some specified threshold. Recording
signal. Neglecting quantization and timing errors,
the times at which zero-crossings occur in a bipolar
the sequence of values produced by an analog-to-
signal is an example of implict sampling.
digital converter can be modeled as the output of

x(t) X(f )

t f

Figure 2.1 Continuous-time sinusoid and its two-sided magnitude spectrum

2-1
2-2 Notes on Digital Signal Processing

an ideal sampling process. Ideal sampling

Key Concept 2.1
is discussed further in Note 3.
Explicit Sampling Techniques
Natural Sampling Ideal Sampling
Nonzero-width samples have time-vary- Models the output of an A/D converter

ing amplitudes that follow the contours of

the analog signal, as shown in Key Con-
cept 2.1. Commutator systems for time- t
division multiplexing of telegraph signals, f
first proposed in 1848, used an approxima-
Spectral images all scaled equally
tion to natural sampling. The sample pulses Zero width samples take on
were created by gating a signal with rotat- instantaneous values of
the analog signal.
ing mechanical contacts. This multiplex-
ing technique was subsequently applied to Instantaneous Sampling
Models the output of a sample and hold amplifier,
telephone signals in 1891. It was the ap- zero order data hold, or D/A converter

plication of natural sampling to telephony

that first led to consideration of just how
rapidly a continuous-time signal needed to t
be sampled in order to preserve fidelity and
ensure the ability to reconstruct exactly the f

original, unsampled signal. Natural sam- Spectral images, including baseband,

are distorted by (sin x)/x envelope.
Non zero width samples, each with
pling is explored further in Note 6. amplitude held constant over the
width of the pulse

Instantaneous Sampling Natural Sampling

Models the output of an analog demultiplexer
In instantaneous sampling, nonzero-width
samples each have a constant amplitude
that corresponds to the instantaneous
value of the analog signal at the beginning t

of the sample. The sample values are held f

constant long enough to create flat-topped Spectral images, with each image
scaled by a factor that is constant
sample pulses. The output of a digital-to- Non zero width samples with
over the image but varies image
to image
time varying amplitudes that
analog converter (DAC) can be modeled follow the contours of the
analog signal
as the output of an instantaneous sampling
process, often as the limiting case in which
the sample width equals the sampling in-
terval. As discussed in Note 7, the results
of the instantaneous sampling model play
a key role in the specification of the analog
filter used to smooth the DAC output.
Reference
1. H. D. Lüke, “The Origins of the Sampling
Theorem,” IEEE Communications, April 1999,
pp. 106–108.
Note 3

Ideal Sampling

T he concept of ideal sampling can be viewed as

the foundation upon which the rest of DSP is
built. The simplest way to view ideal sampling is
ideal continuous-to-discrete (C/D) converter, as
shown in Figure 3.2(b).
Without some augmentation, the ideal sampler
as a mapping process, such as the one depicted in and C/D converter concepts are both just notational
Figure 3.1, that “grabs” uniformly spaced instanta- conveniences to indicate that “sampling happens
neous values of the continuous-time function, x(t), here.” They don’t provide any mathematical basis for
and uses these values to construct the discrete-time the spectral images that are known to arise due to
sequence, x[n]: sampling. It is possible to successfully practice DSP
at the journeyman level without this mathematical
x[n] ← x(nT ) (3.1)
basis, and many introductory texts simply present
where n is the is the integer-valued sample index the facts concerning spectral images and aliasing
and T is the real-valued sampling interval. In or- without delving into the mathematics needed to
der to depict ideal sampling in a block diagram, derive these facts. However, the mathematical back-
some authors adopt the concept of an ideal sampler, ground is useful for exploring advanced topics like
which is usually depicted as a time-driven switch, sampling jitter and nonuniform sampling.
as in Figure 3.2(a). Oppenheim and Schafer [1] in- The formal mathematical model for ideal sam-
troduce a more elegant depiction in the form of an pling is presented in Note 5. The important result

x(t ) x[n]
x(2T ) x[2]

13T 13
t n
T 1 2
2T

x(13T ) x[13]

Figure 3.1 Ideal sampling viewed as a simple mapping from a continuous-time

function to a discrete-time sequence

a b

x(t ) x[n] xa(t )

C/D
x[n] xa(nT )

p(t ) T

Figure 3.2 Block diagram representation of an ideal sampling

process: (a) ideal sampler, (b) ideal continuous-to-discrete
converter
3-1
3-2 Notes on Digital Signal Processing

produced by this model shows that sampling in the thereby causing Eq. (3.2) to become
time domain can be expected to create periodic
∞
⎛ m⎞
images of the original signal’s spectrum in the fre-
quency domain. Physical measurements confirm
Xs ( f ) = ∑
m = −∞
X⎜ f − ⎟
⎝ T⎠
that creation of images predicted by the mathemat-
ical model does in fact occur in the real world. The creation of the images is the important idea—
Consider an arbitrary continuous-time signal, the presence or absence of the 1/T scaling factor is
x(t), having a spectrum that is bandlimited to the usually lost in the overall scaling strategy of most
frequency range, ±fH, as shown in Figure 3.3. The real-world sampling implementations.
spectrum for the ideally sampled version of the sig-
nal will consist of copies or images of the spectrum 3.1 Aliasing
for x(t) periodically repeated along the frequency Figure 3.4 shows the case in which fs > 2fH . The
axis with a center-to-center spacing that is equal to original baseband spectrum and the images are all
the sampling rate, as shown in Figure 3.4. The im- distinct, and theoretically the original signal could
age corresponding to the original signal’s spectrum be recovered exactly via ideal lowpass filtering to
is often called the baseband. completely remove all of the images and leave the
The sample signal’s spectrum can be expressed undistorted baseband spectrum. The sampling the-
in terms of the original signal’s spectrum, X( f ), as orem, presented in the next section, provides a for-
mal statement of this property, as well as a formula
1 ∞ ⎛ m⎞
Xs ( f ) = ∑ X⎜ f − ⎟
T m = −∞ ⎝ T⎠
for reconstructing the original signal from the ideal
samples.
(3.2)
∞ In the case where fs < 2fH , the spectral images
= fs ∑ X ( f − mf s )
m = −∞
overlap, as shown in Figure 3.5. This overlap cre-
ates the condition known as aliasing. Components
A few texts (such as [2]) redefine the sampling rela- in the original signal at frequencies greater than
tion of Eq. (3.1) to be fs /2 will appear at frequencies below fs /2 after the
x[n] ← Tx(nT )

X S (f )
Baseband
X (f ) Images Images

A
A T

f
2fs fs fH 0 fH fs 2fs
f
fs fH fs
fH fH 2
fs fH fs fH

Figure 3.3 Bandlimited spectrum of an Figure 3.4 Frequency relationships for sampling-induced
arbitrary analog signal images when fs > 2fH
Ideal Sampling 3-3

X S (f )
sampling is performed. As depicted in Figure 3.6,
the result is as though the portion of the spectrum
above fs /2 folds over the line at f = fs /2 and adds
into the spectrum immediately below fs /2. Based f
2fs fH 0 fH fs 2fs
on this viewpoint, fs /2 is sometimes called the fold- fs
fs fH
ing frequency. fs fH fs fH
Note 4 shows how the concept of ideal sampling is Figure 3.5 Frequency relationships for sampling-induced
applied to real-world signals in practical applications. images when fs < 2fH

Aliased content adds to content

Similar aliasing originally present in frequency
occurs at the range (fS − fH ) ≤ f ≤ (fS 2), raising
negative edge the apparent spectrum level
of the baseband. to here.

f = − fs 2 f = fs 2

Image Baseband Image

f
0 fH
fs
fs − fH

Overlap from the lower skirt of

the image looks like the upper
skirt of the baseband spectrum
was folded back at fS 2.

Figure 3.6 Aliasing viewed as spectral folding at f = fs/2

References
1. A. V. Oppenheim and R. W. Schafer, Discrete-
Time Signal Processing, Prentice Hall, 1989.
2. R. A. Roberts and C.T. Mullis, Digital Signal
Processing, Addison-Wesley, 1987.
Note 4

Practical Application of Ideal Sampling

This note shows how theoretical results from Note 3 are used
Key Concept 4.1
to develop a practical sampling strategy that minimizes the
effects of aliasing. Aliasing

* n the discussions of aliasing in Note 3, The sampling process creates periodic images of the origi-
nal signal’s spectrum in the frequency domain. Overlap
the frequency fH is portrayed as an abso-
between these images creates a condition called aliasing,
lute upper frequency. Prior to sampling, the in which components in the original signal at frequencies
signal of interest has zero spectral content greater than half the sampling rate will appear, or alias, at
at frequencies greater than fH . Under these frequencies below half the sampling rate in the spectrum of
conditions, sampling at rates greater than the sampled signal.
2fH would result in no aliasing. However,
Spectrum of
this ideal situation is impossible to achieve analog signal
in practical applications, where there will al-
ways be some aliasing. In many cases, even
to achieve extremely low levels of aliasing,
Spectrum of
fH would have to be set so high that, in most fH fH sampled signal
cases, sampling at a rate of 2fH would be pro- Sa
m Images
hibitively difficult and expensive. In practi- pl Baseband Images
in
g
cal situations, rather than trying to avoid all
aliasing, the design goal is to minimize the
effects of aliasing while recognizing that they f
can not be completely eliminated. Overlap
Baseband and
creates
Practical sampling is performed at a rate aliasing. first image cross
at half the
Center-to-center spacing
greater than 2fH, where the signal of inter- of images equals the sampling frequency.
sampling frequency.
est is known to have negligible (rather than
zero) spectral content above some upper In practical systems, a lowpass anti-aliasing filter is typically
frequency, fH . The definition of “negligible” used prior to the sampling operation in order to attenuate
varies based on the particular application. components at frequencies greater than half the sampling
In some cases, fH might be set so restrictively rate and thereby minimize the effects of aliasing.
that less than 0.01 percent of the signal’s en-
ergy is at frequencies greater than fH . In less
demanding applications in which the analog-
to-digital converter (ADC) cost may need to
be kept low, fH may be set lower to allow up
to 5 percent of the signal’s energy to be at fre-
quencies greater than fH .
In most applications, the signal is passed
through an anti-aliasing filter prior to being
digitized. The purpose of that filter is to ensure

4-1
Practical Application of Ideal Sampling 4-2

that aliasing is held to tolerably low levels. A t The signal of interest may contain spuri-
value for fH can be chosen based on the prop- ous self-interference caused by nonlinear
erties of the signal of interest and the require- processing in a mixer or a saturated am-
ments of the application, but in real-world plifier prior to sampling. Because nonlin-
situations, it is generally not possible to guar- ear processing creates components at new
antee that the chosen value for fH bounds the frequencies, some of this self-interference
frequency extent of the actual signal that is may occur at frequencies above fH .
presented to the ADC. There are several rea-
The selection of the sampling rate, fs, and
sons why this is so:
the design of the anti-aliasing filter are co-
t The signal of interest may be contaminat- ordinated, observing the guidelines called
ed with wideband additive noise. out in Design Strategy 4.1, so that the filter
produces minimal attenuation or distortion
t The signal of interest may be contami-
for frequencies below fH , but provides severe
nated with an unanticipated interfering
attenuation for all frequencies above fs/2.
signal having a bandwidth that extends
beyond fH .
References
1. A.V. Oppenheim and R.W. Schafer, Discrete-
Time Signal Processing, Prentice Hall, 1989.
2. R.A. Roberts and C.T. Mullis, Digital Signal
Processing, Addison-Wesley, 1987.
Note 5

Delta Functions and the

Sampling Theorem

W IFO JEFBM TBNQMJOH JT WJFXFE BT B EJ-

SFDU NBQQJOH QSPDFTT BT JU JT JO /PUF
 UIF DSFBUJPO PG spectral images is simply
Math Box 5.1

Sampling Theorem
stated as a fact and accepted without mathe- If the spectrum, X( f ), of a function, x(t ), van-
matical justification. To generate mathemati- ishes beyond an upper frequency of fH Hz,
then x(t ) can be completely determined by its
cal support for the existence of these images,
values at uniform intervals of less than 1/(2 fH).
a more complicated mathematical model of
the sampling process must be adopted. In ad- Reconstruction Formula
dition to correctly predicting the appearance
If a function is sampled at a rate, fs = T -1, that
of spectral images, such a model can also satisfies the sampling theorem, the original
be used to derive the discrete-time Fourier function, x(t ), can be reconstructed from the
transform (DTFT) and the discrete Fourier samples as
transform (DFT) from the “usual” contin-
⎡ ⎛t ⎞⎤
uous-time Fourier transform (CTFT). (See ∞
sin ⎢π ⎜ − n ⎟ ⎥
⎝T ⎠⎦
Math Boxes 12.1 and 13.2.) x(t ) = ∑ x(nT ) ⎣
⎛t ⎞
If the result of ideal sampling is considered n =− ∞
π ⎜ −n⎟
in the continuous-time domain, each sample ⎝ T ⎠ (MB 5.1)
∞
⎛t ⎞
exists for a single instant on the continuous- = ∑ x(nT ) sincπ ⎜ − n ⎟
time axis, and the value of the sampled wave- n =− ∞ ⎝T ⎠
form is zero between sample instants. This
differs from the approach corresponding to
Eq. (3.1) in Note 3 in which the result of ideal
sampling is viewed in the discrete-time do-
main where the result is defined only at the
sample instants—between samples the result
is not zero, it is simply not defined at all.
An ideal sampling process that produces
5.1 Dirac Delta Function
a continuous-time result can be constructed
using a generalized function: The impulse is As far as mathematicians are concerned,
sometimes called the delta function (due to the Dirac delta function is not a function in
its usual notation) or the Dirac delta func- the usual sense, and its use in engineering is
tion, in honor of the English physicist Paul sometimes controversial. Nevertheless, the
Dirac (1902–1984), who made extensive use delta function provides a convenient way to
of impulse functions in his work on quantum relate the spectra of ideal sampling, natural
mechanics.

5-1
Delta Functions and the Sampling Theorem 5-2

sampling, and instantaneous sampling that Math Box 5.2

are discussed in Notes 3, 6, and 7.
A number of nonrigorous approaches for Properties of the
defining the impulse function can be found Dirac Delta Function
throughout the literature. A unit impulse can ∞

be described loosely as having zero width ∫ δ (t )dt = 1

−∞
(MB 5.2)
and an infinite amplitude at the origin such
that the total area under the impulse is equal ∞

to unity. ∫ δ (t − t ) f (t )dt = f (t )
−∞
0 0 (MB 5.3)
The impulse function is usually denoted
as δ(t) and is depicted as a vertical arrow at 1
the origin. The rigorous definition of δ(t), δ (at ) = δ (t ) (MB 5.4)
a
introduced in 1950 by Laurent Schwartz
[1], rejects the notion that the impulse is
δ (t − t 0 ) f (t ) = δ (t − t 0 ) f (t 0 ) (MB 5.5)
an ordinary function and instead defines
it as a distribution. The rigorous definition
FT
notwithstanding, most engineers are more δ (t ) ←→ 1 (MB 5.6)
comfortable defining the impulse in an op-
erational sense. Specifically, δ(t) is taken as
that function which exhibits the so-called
sifting property:
∞

∫ δ (t ) f (t )dt = f (0)
−∞
(5.1)

It has been shown [2] that the impulse func-

tion exhibits the properties listed in Math
Box 5.2.

5.2 Comb ΔT (t)—to indicate an impulse sequence

In DSP, the delta function is most often en- having a period of T seconds:
∞
countered in the form of an infinite periodic
sequence of impulses:
δT (t ) ∑ δ (t − kT )
k = −∞
(5.3)

∑ δ (t − kT )
k = −∞
(5.2)
Other texts refer to this function as the shah
function and denote it using the Cyrillic letter
shah, which somewhat resembles the graph
The DSP literature is about evenly split be-
of the function
tween two different names and notations for
the function represented by Eq. (5.2). Many ∞

texts refer to this function as a Dirac comb шT (t ) ∑ δ (t − kT )

k = −∞
(5.4)

and use the notation δT (t)—or sometimes

5-3 Notes on Digital Signal Processing

5.3 Sampling Model Math Box 5.3

A sequence of samples can be modeled as Properties of the Dirac Comb
the result of multiplying the original signal
and a comb of Dirac impulses: ∞

T (t )x ( ) ∑ kt (t kT )
(MB 5.7)
x s (t ) = x a (t )δ T (t ) k = −∞

where ∞ ∞
δT (t ) ∑ δ (t − nT )
n=−∞
δ T (t ) ⊗ x(t ) = ∑ x(t − kT )
k = −∞
(MB 5.8)

Figure 5.1 shows a C/D converter that has

1 ∞
been modified to include a representation of T (t ) = ∑ exp( 2
T k = −∞
t /T)
(MB 5.9)
this multiplication.
As indicated by Eqs. (MB 5.10) and
(MB 5.11), the Fourier transform of a Dirac 1 ∞ ⎛ k⎞
∑
FT
comb with a spacing of T in time is a second δ T (t ) δ⎜f − ⎟ (MB 5.10)
T k = −∞ ⎝ T⎠
Dirac comb with a spacing in frequency of
T -1 Hz, or 2π/T radians per second:
2π ∞
⎛ 2π ⎞ (MB 5.11)
∑ δ ⎜⎝ ω − k T
FT
F δT (t )
T( ) = fs ( f ) ⎟
s
(5.5) T k = −∞ ⎠
= s ωs (ω )
where ∞
δ fS ( f ) ∑ δ ( f − mf )
m = −∞
s

∞
δ ωS (ω ) ∑ δ (ω − mω )
m = −∞
s

Multiplication in time corresponds to

convolution in frequency, so the spectrum of
the sampled signal can be obtained by con- C/D converter
volving the original signal’s spectrum with
the right-hand sides of Eq. (5.5): Convert from
xa (t ) x s (t ) a weighted
X s ( f ) = X ( f ) ⎡⎣ f s δ fS ( f )⎤⎦ impulse train
in continuous x[n]
1
X s (ω ) =
2π
( X(ω) ⋅[ωs δωS (ω )]) time to a
discrete-time
Dirac comb sequence.
Finally, exploiting the shifting property of ∞

the delta function yields įT (t ) ∑ į(t − nT )

n = −∞
∞
Xs ( f ) fs ∑ X( f − mf )
m = −∞
s

ωs ∞ T
X s (ω ) ∑ δ (ω − mωs )
2 πm = −∞ Figure 5.1 Continuous-to-discrete converter with two
separate processing steps
Delta Functions and the Sampling Theorem 5-4

References
1. L. Schwartz, Théorie des distributions, Herman &
Cie, Paris, 1950.
2. R. N. Bracewell, The Fourier Transform and Its
Applications, 3rd ed., McGraw-Hill, 2000.
3. H. D. Lüke, “The Origins of the Sampling
Theorem,” IEEE Communications, April 1999,
pp. 106–108.
4. H. Nyquist, “Certain topics in telegraph
transmission theory,” Trans. AIEE, vol. 47, Apr.
1928, pp. 617–644, republished in Proc. IEEE,
vol. 90, no. 2, Feb. 2002, pp. 280–305.
5. C. E. Shannon, “Communications in the
presence of noise,” Proc. IRE, vol. 37, 1949, pp.
10–21.
6. V. A. Kotelnikov, “O propusknoj sposobnosti
‘efira’ i provoloki v elektrosvjazi,” (“On the
transmission capacity of ‘ether’ and wire in
electro-communications”), First All-Union
Conference on Questions of Communication,
Jan. 14, 1933.
7. H. Raabe, “Untersuchungen an der
wechselzeitigen Mehrfachubertragung
(Multiplexubertragung),” Elektrische
Nachrichtentechnik, vol. 16, 1939, pp. 213–228.
8. W. R. Bennett, “Time division multiplex
systems,” Bell Syst. Tech. J., vol. 20, 1941, pp.
199–221.
Note 6

Natural Sampling

I n natural sampling, an analog signal is gated in

such a way that the resulting signal consists of
pulses with time-varying amplitudes that follow
At a Glance
t Natural sampling is a mathematical model that
can be used to analyze the impacts of analog mul-
the contours of the original waveform as shown in tiplexing and signal commutation and decommu-
Figure 6.1. In this example, the original signal is a tation.
sinusoid with a period of Tx, and the sampled sig-
t In natural sampling, the samples have nonzero
nal has a sampling interval of T, and a sample width width and the amplitude of the sample varies
of τ. Natural sampling is mathematically equivalent across the width of the sample to match the am-
to multiplying the original signal with a train of plitude of the analog signal.
unit-amplitude rectangular sampling pulses. There-
fore, the spectrum of a naturally sampled signal can
be determined by convolving the original signal’s
spectrum with the spectrum of the train of sam-
pling pulses.
Figure 6.2 summarizes the relationships be- t
tween the original signal (labeled T4), the naturally
sampled signal (labeled T5), and the corresponding
spectra (labeled F4 and F5, respectively). The train
of sampling pulses (labeled T3) can be generated by t In the magnitude spectrum of a naturally sam-
convolving a single pulse of width τ (labeled T1) with pled signal, each image is scaled by a factor that
a Dirac comb having impulses spaced at intervals of is constant over the image but that varies from
T (labeled T2). The resulting pulse train has a magni- image to image according to the magnitude of a
tude spectrum like the one shown in Figure 6.3. sinc envelope.

Figure 6.1 In natural samping, the amplitudes of the sample

pulses follow the varying amplitudes of the original function.

6-1
Natural Sampling 6-2

Time Frequency
T1 Domain Domain F1
Rectangular pulse
­ IJ
CTFT Spectrum of pulse
°1, t 
p(t ) ® 2 P(f ) IJ sinc Ⱥ (f IJ )
°¯0, otherwise
T2 F2

Dirac comb Spectrum of Dirac comb

f f
1 1
įT (t ) ¦ į(t nT ) CTFT ƩT (t ) T ¦ į f mT
n f m f

Multiply
Convolve

F3
T3
CTFT Spectrum of pulse train
Rectangular pulse train P(f ) ƩT (f )

F4
T4
CTFT Spectrum of original signal
Original signal X (f )
x(t )

Frequency Convolve
Multiply Time
Domain Domain

T5 F5

Naturally sampled signal Spectrum of naturally sampled signal

CTFT
f § sin Ⱥ mfs IJ ·
IJ
X s (f )
T
¦ ¨¨ Ⱥ mf IJ ¸¸ X (f mfs )
m f© s ¹

Figure 6.2 Natural sampling is mathematically equivalent to multiplying the original spectrum with a train of rectangular
sampling pulses. Therefore, the spectrum of a naturally sampled signal can be determined by convolving the original
signal’s spectrum with the spectrum of the train of sampling pulses. (The multiply operation that creates spectrum F3
is not mathematically rigorous, but it is consistent with typical engineering use of Dirac delta functions as discussed in
Note 5. “CTFT” indicates the continuous-time Fourier transform.)

τ
T

1 1
τ τ
1
fs T
1
fs T
0
Figure 6.3 Magnitude spectrum of rectangular pulse train (corresponds to block F3
in Figure 6.2)
6-3 Notes on Digital Signal Processing

The magnitude spectrum for a naturally sam- corresponding to the center of the image. In other
pled sinusoid is shown in Figure 6.4. The spacing words, the image centered at f = nfs is scaled by
of the images is equal to the reciprocal of the sam- τT -1|sincπ( nfsτ)|. This factor changes from image to
pling interval, and each image is amplitude scaled image, but remains constant across the width of each
by the value of τT -1|sincπ( fτ)| at the frequency image.

f =0
IJ
T
IJ
sinc Ⱥ (f IJ )
T

IJ 1 fx Tx 1 IJ 1
fs T 1
fs T 1

Figure 6.4 Magnitude spectrum of naturally sampled sinusoid (corresponds to block F5

in Figure 6.2)

Reference
1. W. R. Bennett, “Time division multiplex
systems,” Bell Syst. Tech. J., vol. 20, 1941,
pp. 199–221.
Note 7

Instantaneous Sampling

I n instantaneous sampling, the value of the ana-

log signal is captured at the sampling instant and
held constant for the duration of the sample pulse,
At a Glance
t Instantaneous sampling (also known as zero-
order-data-hold (ZODH) sampling) is a math-
as shown in Figure 7.1. Instantaneous sampling is
ematical model that can be used to analyze the
sometimes called “flat-topped” sampling or zero- impact on signal quality of digital-to-analog con-
order-data-hold sampling. version, as shown in Note 8.
Instantaneous sampling is mathematically
t In instantaneous sampling, the samples have
equivalent to convolving a single rectangular sam- nonzero width and the amplitude of the sample
pling pulse with an ideally sampled version of the remains constant across the width but varies
original signal. Therefore, the spectrum of an in- from sample to sample to match the amplitude of
stantaneously sampled signal can be determined by the analog signal at each sample’s starting instant.
multiplying the ideally sampled signal’s spectrum
with the spectrum of a single sample pulse.
Figure 7.2 summarizes the relationships between
the original signal (labeled T2), the instantaneously
sampled signal (labeled T5) and the corresponding t
spectra (labeled F2 and F5 respectively). A single
pulse, p(t ), (labeled T4 in the figure) defined by
⎧⎪ τ
p(t ) = ⎨ 1, t< 2 t In the magnitude spectrum of an instantaneous-
⎪⎩ 0, otherwise ly sampled signal, all spectral components are
scaled by the magnitude of a sinc envelope. Be-
IBT UIF 'PVSJFS USBOTGPSN P(f MBCMFE ' HJWFO CZ cause the scaling varies across the width of each
spectral image, the signal is distorted.
P( f ) τ sincπ ( f τ )

Figure 7.1 In instantaneous sampling, the amplitude

of each sample remains constant over its width.

7-1
7-2 Notes on Digital Signal Processing

The magnitude spectrum of an ideally sampled X s (f )

sinusoid is shown in Figure 7.3. Multiplying this
spectrum (F3 in Figure 7.2) by P( f ) yields the re-
sult shown in Figure 7.4. The scaling factor does
not remain constant across each image, as it does in
the spectrum for natural sampling. If the spectrum
f
of the original spectrum is represented as X( f ),
then the spectrum of the corresponding instanta- fx Tx 1
1
fs T
neously sampled signal is given by 1
fs T
τ § sin(π f τ ) · f 1 0
Xs( f ) ¨ ¸ ¦ X f  nT
T © π fτ ¹n f Figure 7.3 Magnitude spectrum of ideally sampled sinusoid
(corresponds to block F3 in Figure 7.2).

Time Frequency
T1 Domain Domain F1

Dirac comb Spectrum of Dirac comb

CTFT f
f 1 1
įT (t ) ¦ į(t nT ) ƩT (t ) T ¦ į f mT
m f
n f
Original signal
T2 F2
x(t )
Spectrum of
CTFT
original signal

Multiply X (f )

T3
Convolve
Ideally sampled signal CTFT

F3

Spectrum of ideally sampled signal

T4 Rectangular pulse f
1 1
­ IJ
X s (f ) T ¦ X (f mT )
°1, t  m f
p(t ) ® 2
°
¯0, otherwise CTFT F4

Spectrum of pulse
P(f ) IJ sinc Ⱥ (f IJ )

Convolve

Multiply
Time Frequency
Domain Domain

T5 F5

CTFT Spectrum of instantaneously

Instantaneously sampled signal
sampled signal
IJ § sin Ⱥ fs IJ · f 1
X s (f ) ¨¨ ¸¸ ¦ X (f  mT )
T © Ⱥ fs IJ ¹m f

Figure 7.2 Instantaneous sampling is mathematically equivalent to convolving a single rectangular sam-
pling pulse with an ideally sampled version of the original signal. Therefore, the spectrum of an instanta-
neously sampled signal can be determined by multiplying the ideally sampled signal’s spectrum with the
spectrum of a single sample pulse. (“CTFT” indicates the continuous-time Fourier transform.)
Instantaneous Sampling 7-3

IJ
T

Each component is gain scaled

by the sinc envelope.

f
1
IJ 1 IJ
f=0 1
fs T
1
TP

Figure 7.4 Magnitude spectrum of instantaneously sampled sinusoid (corresponds to block F5 in Figure 7.2)

Reference
1. W. R. Bennett, “Time division multiplex
systems,” Bell Syst. Tech. J., vol. 20, 1941,
pp. 199–221.
Note 8

Reconstructing Physical Signals

The Basic Idea

An analog signal is typically reconstructed from a discrete-time signal using a

digital-to-analog converter (DAC) followed by a reconstruction filter.

Digital Analog
samples Digital-to- output
Reconstruction
analog
filter
converter

The spectrum of the DAC output signal is distorted by a sinc envelope, with the nulls
of the sinc envelope falling in the center of each image other than the baseband.

Baseband
Images Images

Spectrum of
digital signal

f
−T −1 0 T −1

Re
con
str
uc tio
n

Higher frequencies
in baseband image Spectrum of
are attenuated by DAC output
sinc envelope. signal

Images are
sinc envelope
severely distorted
by sinc envelope.

f
−2T −1 −T −1 0 T −1 2T −1

The reconstruction filter must remove the images and correct for the sinc distor-
tion in the baseband. The close spacing between the baseband and the adjacent
images can make the design of this filter relatively difficult.

8-1
T he mathematical signal reconstruction tech-
niques presented in Note 5 do have practical
uses, but these techniques are not really suited for
interval. The DAC output is a special case in that the
DAC typically holds each output value for an entire
sampling interval, thereby generating a “stair-step”
converting a sequence of digital signal values back signal, such as the one shown in Figure 8.2, in which
into a continuously time-varying voltage that can the sample width equals the sampling interval.
be used to drive a speaker or a pair of headphones. The spectrum of the DAC output contains im-
Reconstruction of a physical analog signal is usually ages and is multiplied by a (sin x) /x envelope, as dis-
accomplished using a digital-to-analog converter cussed in Note 7. However, the illustrations in Note
(DAC). The input to the DAC is a sequence of digi- 7 depict the case in which the sampling interval, T, is
tal words, and the output is a time-varying voltage several times larger than the sample width, τ. When
that is proportional to the sequence of values rep- the sample width is equal or nearly equal to the sam-
resented by the input words. Each output voltage is pling interval, the distortion effects caused by the
held constant until the input value changes. (sin x) /x envelope are much more severe.
The output of the DAC can be viewed as a special Assume that the ideally sampled signal inside
case of the instantaneously sampled signal described the processing computer has a simple trapezoi-
in Note 7. In Note 7, each voltage pulse is depicted dal baseband spectrum, as depicted in Figure 8.3.
as being significantly narrower than the sampling The corresponding DAC output has a spectrum,

digital stepped
smooth
signal analog
analog T IJ
values digital -to- waveform
analog signal
analog
reconstruction
converter
filter
(DAC )
Figure 8.2 Output of DAC modeled as the limiting case of
Figure 8.1 Block diagram of the signal reconstruction process instantaneous sampling

b b b a b b b

f
−1 −1 −1 −1
−2T −T 0 T 2T
Figure 8.3 Idealized trapezoidal spectrum for a sampled signal, showing (a) the baseband
spectrum of the original signal, and (b) spectral images created by the sampling process

8-2
8-3 Notes on Digital Signal Processing

as shown in Figure 8.4, with the main lobe of the output signal in a way that separates the spectral im-
(sin x) /x envelope having a null-to-null width of ages, as shown in Figure 8.5. In many practical sys-
just twice the sampling rate. All of the images are tems, the sample rate is already significantly higher
severely distorted by the side lobes. Because it oc- than twice the signal bandwidth, which causes the
cupies such a large portion of the main lobe, the images to be spread farther apart and simplifies the
baseband component of the spectrum also experi- filter design task. In other applications, such as au-
ences distortion from the (sin x) /x envelope. dio CD players, it is necessary to take explicit steps
The reconstruction filter that follows the DAC to make the reconstruction problem easier to man-
needs to have both a stopband response that se- age by increasing the Nyquist bandwidth (which is
verely attenuates the spectral images and a pass- equal to one-half the sample rate) without increas-
band response that is designed to correct the ing the utilized bandwidth, as shown in Figure 8.5.
(sin x) /x distortion present on the baseband spec- Audio CD players use a sample rate of 44.1 kHz
trum. When the sample rate equals exactly twice to support a utilized bandwidth of about 20 kHz,
the highest frequency component in the signal’s so there would be a gap of about 4 kHz between
original spectrum, the signal is said to be critically images in the DAC output. Design of an acceptable
sampled. When the signal is critically sampled, as reconstruction filter is not impossible, and there
in the case depicted in Figure 8.4, the images are were many early CD players built that used direct
close together, thus making it almost impossible to reconstruction of the 44.1 kHz sample stream.
design a filter that can both remove the images and However, CD players are consumer products, and
compensate for (sin x) /x distortion. Most filter de- there is constant pressure to make them smaller,
signs that can remove the images under these con- lighter, cheaper, and better-sounding. Increasing
ditions are likely to introduce phase distortion into the CD sample rate would make it easier to build
the baseband signal. cheaper reconstruction filters with good perfor-
Design and implementation of the reconstruc- mance, but an increased sample rate for the re-
tion filter can be made easier by modifying the DAC corded signal would require more samples for each

f
1 1 1
2T T 0 T 1 2T

Figure 8.4 Spectrum at output of DAC (solid trace). Shown for comparison are the
undistorted images of the trapezoidal spectrum (dashed trace) and sinc envelope
(dotted trace).
Reconstructing Physical Signals 8-4

second of audio, thus resulting in reduced playing coincides with a more central, flatter portion of the
time for a disc having a given total bit capacity. main lobe, thus lessening the severity of the distor-
Most newer CD players advertise 4⫻, 8⫻, or tion that the reconstruction filter must correct.
even 16⫻ oversampling, but the increased sample If oversampling is carried throughout the digi-
rate is not used for the recorded signal. Instead, the tal processing and if the utilized bandwidth has
digital signal is interpolated to create new sample not already been limited by this processing, then
values in between the sample values that are actu- it may be prudent to perform digital filtering to
ally read from the disc. This type of oversampling limit the utilized bandwidth just prior to sending
does not increase the utilized bandwidth; that the signal to the DAC. Because of the increased
is, it does not increase the bandwidth of the re- Nyquist bandwidth (due to the higher sampling
corded signal or the reconstructed signal. What it rate), earlier processing may have inadvertantly in-
does do is increase the Nyquist bandwidth, which troduced components outside of the intended sig-
moves the spectral images farther apart so that it nal bandwidth and thereby increased the utilized
becomes relatively easy to design a reconstruction bandwidth, reducing the effectiveness of the over-
filter that will reject all of the non-baseband im- sampling strategy. On the other hand, if the over-
ages while simultaneously compensating for the sampling is introduced by an interpolation process
(sin x) /x distortion in the baseband spectral com- just prior to sending the signal to the DAC, there
ponent. As shown in Figure 8.5, this oversampling is no opportunity to inadvertently increase the uti-
will also have the effect of widening the main lobe lized bandwidth, and additional filtering would not
of the sinc function so that the utilized bandwidth be necessary.

Utilized bandwidth

Nyquist bandwidth

f
T 1 fS 0 fS T 1
2 2
Figure 8.5 Spectrum at output of DAC with 2× oversampling
Note 9

Overview of Fourier Analysis

F ourier analysis is based on the notion

that signals can be represented and
analyzed as weighted sums of sinusoidal
Major Categories of Fourier Analysis

Fourier Series
components over a range of different fre- t Signal: periodic function of continuous
quencies. The collection of amplitudes time
and phases for the sinusoids needed to
t Spectrum: nonperiodic function of
completely represent the signal is usually discrete frequency
called the spectrum of the signal. Depend-
ing upon the nature of the signal being t Detailed in Note 10
analyzed, the spectrum can span either
Fourier Transform
a continuum of frequencies or a count-
t Signal: nonperiodic function of continu-
able (but possibly infinite) set of discrete ous time
frequencies.
t Spectrum: nonperiodic function of con-
tinuous frequency
9.1 Fourier Series
t Detailed in Note 11
Periodic continuous-time signals have fi-
nite power (but infinite energy), and can Discrete-Time Fourier Transform
be analyzed using the Fourier series (FS) t Signal: nonperiodic function of discrete
defined by time

X[n] = T1 ∫ x(t )exp ( − jnω0t ) dt t Spectrum: represented as a periodic func-

T tion of continuous frequency
t Detailed in Note 12
for n = 0, ±1, ±2, ±3, . . . , ±∞
where ω0 is the fundamental radian fre- Discrete Fourier Transform
quency of x(t), T is the period of x(t), and t Signal: periodic function of discrete time
the single limit of T on the integral indi- t Spectrum: periodic function of discrete
cates that the integration is performed over frequency
one period of x(t). t Detailed in Notes 13 through 16
The spectrum produced by the Fourier
series is a function of discrete frequency— Fast Fourier Transform
or in other words, the spectrum has non- t Collection of techniques for efficient
zero values only at discrete frequencies implementation of the discrete Fourier
that correspond to integer multiples (in- transform
cluding zero) of the periodic time signal’s t Detailed in Notes 17 through 19
fundamental frequency. The Fourier series
is discussed in more detail in Note 10.

9-1
Overview of Fourier Analysis 9-2

9.2 Fourier Transform 9.4 Discrete Fourier Transform

Mathematically defined continuous time Periodic discrete-time signals have finite
signals having finite energy can be analyzed power and are analyzed using the discrete
using the Fourier transform (FT, or some- Fourier transform (DFT), defined by
times CTFT to emphasize the continuous- N −1

time nature of the input) defined by X[m] = ∑ x[n]exp(− j 2π mFnT )

n=0
∞

X (ω ) = ∫ x(t )exp(− jω t )dt The spectrum produced by the DFT is dis-

∞
crete in frequency and periodic, with a
where ω represents continuous radian fre- period equal to the reciprocal of the time-
quency. There are also some “special” signals domain sampling interval. Many practical
that have a Fourier transform even though applications do not involve periodic discrete-
the signals do not have finite energy. The time signals—in such applications, a finite-
spectrum produced by the Fourier trans- duration segment of the sampled signal is
form is a function of continuous frequency. assumed to represent exactly one period of
Details of the Fourier transform are dis- a periodic (and infinite duration) signal and
cussed in Note 11. analyzed using the DFT. If the segment of the
discrete-time signal is obtained by sampling
9.3 Discrete-Time Fourier a properly bandlimited analog signal, then
Transform one period of the DFT’s periodic spectrum
Discrete-time signals that have finite en- can be interpreted as a frequency-sampled
ergy can be analyzed using the discrete-time estimate of the original analog signal’s spec-
Fourier transform (DTFT), defined by trum. The DFT and its properties are dis-
∞
cussed in Notes 13 through 16.
X (ω ) = ∑ x[n]exp(− jω nT )
n=−∞ 9.5 Fast Fourier Transforms
The spectrum produced by the DTFT is a Direct computation of an N-sample DFT in-
function of continuous frequency and is peri- volves a number of arithmetic operations that
odic, with a period equal to the reciprocal of is proportional to N2 . There are a number of
the time-domain sampling interval. If the dis- different algorithms that exploit periodicities
crete-time signal is viewed as having been cre- in the sine and cosine functions to compute
ated via sampling of a properly bandlimited the DFT using significantly fewer arithmetic
continuous-time signal, then one period of operations. These algorithms are collectively
the DTFT’s periodic spectrum can be inter- referred to as the fast Fourier transform (FFT).
preted as the spectrum of the original analog The most commonly used FFT algorithms
signal. The DTFT is discussed in Note 12. require a number of arithmetic operations
that is proportional to N log2 N. Fast Fourier
transform algorithms are presented in Notes
17 through 19. Using FFT algorithms to per-
form fast convolution is discussed in Note 20.
Exploring the Variety of Random
Documents with Different Content
And Alexander shaped the conquered world,
But Aristotle shaped the conqueror’s mind.
He had shaped that mind to ends not all its own.
His was the well-thumbed Odyssey that reposed
Under the conqueror’s pillow; his the love,
Fragrant with memories of the hills and sea,
That had rebuilt Stagira; his the voice
In the night-watches; his the harnessed thoughts
That, like immortal sentries, mounted guard
In the dark gates of that world-quelling mind.
His was the whisper, the dark vanishing hint,
The clue to the riddle of slowly emerging life
That, imaged in Egyptian granite, rose
Before the silent conqueror when he stared
At that strange shape, half human and half brute,
The Sphinx, who knew the secret of the world
And smiled at him, and all his victories,
Under the desert stars, while the deep night
Silently deepened round him.
Far away,
In Athens, towered the bearer of the fire.
His was the secret harmony of law
That, while the squadrons wheeled in ordered ranks,
Each finding its full life only in the whole,
Flashed light upon the cosmos; his the quest
That taught the conqueror how to honour truth
And led him, while he watered his proud steeds
In all the streams from Danube to the Nile,
To send another army through the wilds,
Ten thousand huntsmen, ranging hills and woods
At Aristotle’s hest, for birds and beasts;
So that the master-intellect might lay hold
Upon the ladder of life that mounts through Time,
From plants to beasts, and up, through man, to God.
So all the might of Macedon had been turned
To serve the truth, and to complete his work
At Athens, for the conquering age to come;
When Athens, like the very City of Truth,
Might shine upon all nations, and might wear,
On her clear brows, his glory as her own.

Then came a flying rumour through the night.

Earth’s overlord, the autocrator, his friend,
Alexander the Great had fallen in Babylon.
A little cup of poison, subtle drops
Of Lethe—in a cup of delicate gold,—
And the world’s victor slept, an iron sleep;
The conqueror, stricken in his conquered city,
Cold, in the purple of Babylon, lay dead:
And the slow tread of his armies as they passed,
Soldier by soldier, through that chamber of death,
To look their last upon his marble face,
Pulsed like a muffled drum across the world.
Had Aristotle’s cunning mixed the draught
That murdered tyranny? Let that whispered lie
Estrange the heart of Macedon.
There, in Athens,
It was enough, now that his friend lay dead,
To know that, as the body is rent away
From the immortal soul, his greatness now
Had lost its earthly stay. His mighty mind
Walked like a ghost in Athens. It was enough
To hint that he had taught his king too well;
Served him too well; and played the spy for him;
While, for main charge, since he had greatly loved
The mother who had borne him, since he had poured
His love out on her tomb, it would suffice
To snarl that rites like these were meant for gods
And that this man who had seen behind the world
The Mover of all things, the eternal God,
The supreme Good, by these fond rites of love,
Too simple and too great, too clear, too deep,
Had robbed the little sophists of their dues
And so blasphemed against their gods of clay.
 ...

Hurrying footsteps neared. He turned and saw

His young adopted son and Tyrtamus.
“Nicanor! Theophrastus!—nay, lift up
Your heads. You cannot bring me bitterer news
Than I foresaw. I must be brought to judgment.
But on what grounds?”—
“Dear father of us all——”
The youth, Nicanor, answered, “When the crowd
Grins in the very face of those who ask,
Or think, or dream that truth should be their guide;
Nay, grins at truth itself, as at a fool
Tricked in his grandsire’s rags, a rustic oaf,
A blundering country simpleton who gapes
At the great city’s reeling dance of lies,
How can the grounds be wanting?”
“The true grounds,”
His ‘Theophrastus’ muttered, “we know too well.
Eurymedon, and the rest, those gnat-like clans,
The sophists’ buzzing swarms, desire a change.
They hold with Heraclitus—all things change.”
His irony stung the youth. His grey eyes gleamed.
His voice grew harsh with anger. “Ay, all things change!
So justice and injustice, right and wrong,
Evil and good, must wear each other’s cloaks;
And, in that chaos, when all excellence
And honour are plucked down, and the clear truth
Trampled into the dirt, themselves may rise.
Athens is dying.”
“They speak truly enough
Of all that they can know,” the Master said.
“Change is the rhythm that draws this world along.
They see the change. Its law they cannot see.
But man who is mortal in this body of earth
Has also a part, by virtue of his reason,
In an enduring realm. Their prophet knew
And heard what sophists have no souls to hear,—
The Harmony that includes the pulse of change;
The divine Reason, past the flux of things;
The eternal Logos, ordering the whole world.”
And, as he spoke, I heard, through his own words,
Tones that were now a part of his own mind,
The murmur of that old legend which he read
So long ago, in boyhood, by the sea.
Time never fails. Not Tanais or the Nile
Can flow for ever. All things pass away
But One, One only; for the eternal Mind
Enfolds all changes, and can never change.
Tyrtamus touched his arm. “Time presses now.
Come with us. All is ready. On the coast,
In a lonely creek, the quiet keel is rocking.
Three trusty sailors wait us, and at dawn
We, too, shall find new life in a new world
With all that could endure. The voyager knows
The blindness of the cities. Each believes
Its narrow wall the boundary of the world;
And when he puts to sea, their buzzing cries
Fade out behind him like a wrangle of bees.”—

“If I remain, what then?”—

The hill-top shone
In the last rays. Athens was growing dark.
Tyrtamus answered him. “A colder cup
Of hemlock, and the fate of Socrates.”
The Master looked at Athens. Far away
He traced the glimmering aisle of olive-trees
Where, for so long, with many a youthful friend
He had walked, and taught, and striven himself to learn.
Southward, below the Acropolis, he could see
The shadowy precincts of the Asclepiads,
Guarding their sacred spring, the natural fount,
Loved for his father’s memory.
Close beside,
The Dionysiac theatre, like a moon
Hewn from the marble of Hymettus, gleamed,
A silvery crescent, dying into a cloud.
There, though the shade of Sophocles had fled,
Long since, he heard even now in his deep soul
The stately chorus on a ghostly stage
Chanting the praise of thought that builds the city,
Hoists the strong sail to cross the hoary sea,
Ploughs the unwearied earth, yokes the wild steed
And the untamed mountain-bull; thought that contrives
Devices that can cure all ills but death:

Of all strong things none is more strong than man;

Man that has learned to shield himself from cold
And the sharp rain; and turns his marvellous arts
Awhile to evil; and yet again, to good;
Man that is made all-glorious with his city
When he obeys the inviolable laws
Of earth and heaven; but when, in subtle pride,
He makes a friend of wrong, is driven astray
And broken apart, like dust before the wind.

All now, except the heights, had died away

Into the dark. Only the Parthenon raised
A brow like drifted snow against the west.
He watched it, melting into the flood of night
With all those memories.
Then he turned and said,
“If in a moment’s thoughtless greed I grasped
The prize that Athens offers me to-night,
She is not so rich but this might make her poor.
Death wears a gentle smile when we grow old;
And I could welcome it. But she shall not stain
Her hands a second time. Let Athens know
That Aristotle left her, not to save
His last few lingering days of life on earth
But to save Athens.
 I have truly loved her,
Next to the sea-washed town where I was born,
Best of all cities built by men on earth.
But there’s another Athens, pure and white,
Where Plato walks, a City invisible,
Whereof this Athens is only a dim shadow;
And I shall not be exiled from that City.”

The hilltop darkened. The blind mist rolled down;

The voices died. I saw and heard no more.
 III—MOVING EASTWARD
I
Farabi and Avicenna
Grey mists enfolded Europe; and I heard
Sounds of bewildered warfare in the gloom.

Yet, like a misty star, one lampad moved

Eastward, beyond the mountains where of old
Prometheus, in whose hand the fire first shone,
Was chained in agony. His undying ghost
Beheld the fire returning on its course
Unquenched, and smiled from his dark crag in peace,
Implacable peace, at heaven.
Eastward, the fire
Followed the road Pythagoras trod, to meet
The great new morning.
The grey mists dissolved.
And was it I—or Shadow-of-a-Leaf—that saw
And heard, and lived through all he showed me then?

I saw a desert blazing in the sun,

Tufts of tall palm; and then—that City of dreams.
As though an age went past me in an hour
I saw the silken Khalifs and their court
Flowing like orient clouds along the streets
Of Bagdad. In great Mahmoun’s train I saw
Nazzam, who from the Stagirite caught his fire.
Long had he pondered on the Eternal Power
Who, in the dark palm of His timeless hand
Rolls the whole cosmos like one gleaming pearl.
Had he not made, in one pure timeless thought,
All things at once, the last things with the first,
The first life with the last; so that mankind,
Through all its generations, co-exists
For His eternal eyes? Yet, from our own
Who in the time-sphere move, the Maker hides
The full revolving glory, and unfolds
The glimmering miracles of its loveliness
Each at its destined moment, one by one,
In an æonian pageant that returns
For ever to the night whence it began.
Thus Nazzam bowed before the inscrutable Power,
Yet found Him in his own time-conquering soul.

I saw the hundred scribes of El Mansour

Making their radiant versions from the Greek.
I saw Farabi, moving through the throng
Like a gaunt chieftain. His world-ranging eyes
Beheld the Cause of causes.
In his mind,
Lucid and deep, the reasoning of the Greeks
Flooded the world with new celestial light,
Golden interpretations that made clear
To mighty shades the thing they strove to say.

He carried on their fire, with five-score books

In Arabic, where the thoughts of Athens, fledged
With orient colours, towered to the pure realm
Of Plato; but, returning earthward still,
Would wheel around his Aristotle’s mind
Like doves around the cote where they were born.
Then the dark mists that round the vision flowed
Like incense-clouds, dividing scene from scene,
Rolled back from a wide prospect, and I saw,
As one that mounts upon an eagle’s wing,
A savage range of mountains, peaked with snow,
To northward.
They glowed faintly, for the day
Was ending, and the shadows of the rocks
Were stretched out to the very feet of night.
Yet, far away, to southward, I could see
The swollen Oxus, like a vanishing snake
That slid away in slippery streaks and gleams
Through his grey reed-beds to the setting sun.
Earthward we moved; and, in the tawny plain,
Before me, like a lanthorn of dark fire
Bokhara shone, a city of shadowy towers
Crimsoned with sunset. In its turreted walls
I saw eleven gates, and all were closed
Against the onrushing night.
Then, at my side,
My soul’s companion whispered, “You shall see
The Gates of Knowledge opening here anew.
Here Avicenna dwelt in his first youth.”

At once, as on the very wings of night,

We entered. In the rustling musky gloom
Of those hot streets, thousands of falcon eyes
Were round us; but our shadows passed unseen
Into the glimmering palace of the Prince
Whom Avicenna, when all others failed,
Restored to life, and claimed for all reward
Freedom to use the Sultan’s library,
The pride of El Mansour; a wasted joy
To the new Sultan. Radiances were there
Imprisoned like the innumerable slaves
Of one too wealthy even to know their names;
Beautiful Grecian captives, bought with gold
From tawny traffickers in the Ionian sea.
A shadow, with a shadow at my side,
I saw him reading there, intent and still,
Under a silver lamp; his dusky brow
Wreathed with white silk, a goblet close at hand
Brimmed with a subtle wine that could uncloud
The closing eyes of Sleep.
 Along each wall
Great carven chests of fragrant cedar-wood
Released the imprisoned magic,—radiant scrolls,
Inscribed with wisdom’s earliest wonder-cry;
Dark lore; the secrets of the Asclepiads;
History wild as legend; legends true
As history, all being shadows of one light;
Philosophies of earth and heaven; and rhymes
That murmured still of their celestial springs.
He thrust his book aside, as in despair.
Our shadows followed him through the swarming streets
Into the glimmering mosque. I saw him bowed
Prostrate in prayer for light, light on a page
Of subtle-minded Greek which many a day
Had baffled him, when he sought therein the mind
Of his forerunner.
I saw him as he rose;
And, as by chance, at the outer gates he met
A wandering vendor of old tattered books
Who, for three dirhems, offered him a prize.
He bought it, out of gentle heart, and found
A wonder on every page,—Farabi’s work,
Flooding his Greek with light.
He could not see
What intricate law had swept it into his hand;
But, having more than knowledge, he returned
Through the dark gates of prayer; and, pouring out
His alms upon the poor, lifted his heart
In silent thanks to God.

II
Avicenna’s Dream
But all these books—for him—were living thoughts,
Clues to the darker Book of Nature’s law;
For, when he climbed, a goat-foot boy, in Spring
Up through the savage Hissar range, he saw
A hundred gorges thundering at his feet
With snow-fed cataracts; torrents whose fierce flight
Uprooted forests, tore great boulders down,
Ground the huge rocks together; and every year
Channelled raw gullies and swept old scars away;
So that the wildered eagle beating up
To seek his last year’s eyry, found that all
Was new and strange; and even the tuft of pines
That used to guide him to his last year’s nest
Had vanished from the crags he knew no more.

There, pondering on the changes of the world,

Young Avicenna, with a kinglier eye,
Saw in the lapse of ages the great hills
Melting away like waves; and, from the sea,
New lands arising; and the whole dark earth
Dissolving, and reshaping all its realms
Around him, like a dream.
Thus of his hills
And of their high snows flowing through his thoughts
Was born the tale that afterwards was told
By golden-tongued Kazwini, and wafted thence
Through many lands, from Tartary to Pameer.
For, cross-legged, in the shadow of a palm,
The hawk-eyed teller of tales, in years unborn
Holding his wild clan spell-bound, would intone
The deep melodious legend, flowing thus,
As all the world flows, through the eternal mind.

I came one day upon an ancient City.

I saw the long white crescent of its wall
Stained with thin peach-blood, blistered by the sun.

I saw beyond it, clustering in the sky,

Ethereal throngs of ivory minarets,
Tall slender towers, each crowned with one bright pearl.
It was no desert phantom; for it grew
And sharpened as I neared it, till I saw,
Under the slim carved windows in the towers,
The clean-cut shadows, forked and black and small
Like clinging swallows.
In the midst up-swam
The Sultan’s palace with its faint blue domes,
The moons of morning.
Wreaths of frankincense
Floated around me as I entered in.
A thousand thousand warrior faces thronged
The glimmering streets. Blood-rubies burned like stars
In shadowy silks and turbans of all hues.

The markets glowed with costly merchandise.

I saw proud stallions, pacing to and fro
Before the rulers of a hundred kings.
I saw, unrolled beneath the slender feet
Of slave-girls, white as April’s breathing snow,
Soft prayer-rugs of a subtler drift of bloom
Than flows with sunset over the blue and grey
And opal of the drifting desert sand.

Princes and thieves, philosophers and fools

Jostled together, among hot scents of musk.
Dark eyes were flashing. Blood throbbed darker yet.
Lean dusky fingers groped for hilts of jade.
Then, with a roll of drums, through Eastern gates,
Out of the dawn, and softer than its clouds,
Tall camels, long tumultuous caravans,
Like stately ships came slowly stepping in,
Loaded with shining plunder from Cathay.
I turned and asked my neighbour in the throng
Who built that city, and how long ago.
He stared at me in wonder. “It is old,
Older than any memory,” he replied.
“Nor can our fathers’ oldest legend tell
Who built so great a city.”
I went my way.
And in a thousand ages I returned,
And found not even a stone of that great City,
Not even a shadow of all that lust and pride.
But only an old peasant gathering herbs
Where once it stood, upon the naked plain.

“What wars destroyed it, and how long ago?”

I asked him. Slowly lifting his grey head,
He stared at me in wonder.
“This bleak land
Was always thus. Our bread was always black
And our wine harsh. It is a bitter wind
That scourges us. But where these nettles grew
Nettles have always grown. Nothing has changed
In mortal memory here.”
“Was there not, once,
A mighty City?” I said, “with shining streets,
Here, on this ground?” I spoke with bated breath.
He shook his head and smiled, the pitying smile
That wise men use to poets and to fools.—
“Our fathers never told us of that City.
Doubtless it was a dream.”
I went my way.
And in a thousand ages I returned;
And, where the plain was, I beheld the sea.
The sea-gulls mewed and pounced upon their prey.
The brown-legged fishermen crouched upon the shore,
Mending their tarry nets.
I asked how long
That country had been drowned beneath the waves.
They mocked at me. “His wits are drowned in wine.
Tides ebb and flow, and fishes leap ashore;
But all our harvest, since the first wind blew,
Swam in deep waters. Are not wrecks washed up
With coins that none can use, because they bear
The blind old images of forgotten kings?
The waves have shaped these cliffs, dug out these caves,
Rounded each agate on this battered beach.
How long? Ask earth, ask heaven. Nothing has changed.
The sea was always here.”—
I went my way.

And in a thousand ages I returned.

The sea had vanished. Where the ships had sailed
Warm vineyards basked, among the enfolding hills.
I saw, below me, on the winding road,
Two milk-white oxen, under a wooden yoke,
Drawing a waggon, loaded black with grapes.
Beside them walked a slim brown-ankled girl.
I stood beneath a shadowy wayside oak
To watch them. They drew near.
It was no dream.
Blood of the grape upon the wrinkled throats
And smoking flanks of the oxen told me this.
I saw the branching veins and satin skin
Twitch at the flickering touch of a fly. I saw
The knobs of brass that sheathed their curling horns,
The moist black muzzles.
Like many whose coats are white,
Their big dark eyes had mists of blue.
Their breath
Was meadows newly mown.
By all the gods
That ever wrung man’s heart out in the grave
I did not dream this life into the world.—
Blood of the grape upon the girl’s brown arms
And lean, young, bird-like fingers told me this.
Her smooth feet powdered by the warm grey dust;
The grape-stalk that she held in her white teeth;
Her mouth a redder rose than Omar knew;
Her eyes, dark pools where stars could shine by day;
These were no dream. And yet,—
 “How long ago,”
I asked her, “did the bitter sea withdraw
Its foam from all your happy sun-burnt hills?”
She looked at me in fear. Then, with a smile,
She answered, “Nothing here has ever changed.
My father’s father, in his childhood, played
Among these vines. That oak-tree where you stand
Had lived a century, then. The parent oak
From which its acorn dropped had long been dead.
But hills are hills. I never saw the sea.
Nothing has ever changed.”
I went my way.
Last, in a thousand ages I returned,
And found, once more, a City, thronged and tall,
More rich, more marvellous even than the first;
A City of pride and lust and gold and grime,
A City of clustering domes and stately towers,
And temples where the great new gods might dwell.
But, turning to a citizen in the gates,
I asked who built it and how long ago.
He stared at me as wise men stare at fools;
Then, pitying the afflicted, he replied
Gently, as to a child:
“The City is old,
Older than all our histories. Its birth
Is lost among the impenetrable mists
That shroud the most remote antiquity.
None knows, nor can our oldest legends tell
Who built so great a City.”
I went my way.
 IV—THE TORCH IN ITALY
Leonardo Da Vinci
I
HILLS AND THE SEA
The mists rolled back. I saw the City of Flowers
Far down, upon the plain; and, on the slope
Beside us—we were shadows and unseen,—
Giulio, the painter, sketching rocks and trees.
We watched him working, till a pine-cone crackled
On the dark ridge beyond us, and we saw,
Descending from the summits like a god,
A deep-eyed stranger with a rose-red cloak
Fluttering against the blue of the distant hills.

He stood awhile, above a raw ravine,

Studying the furrows that the rains had made
Last winter. Then he searched among the rocks
As though for buried gold.
As he drew near
Giulio looked up and spoke, and he replied.
Their voices rose upon the mountain air
Like a deep river answering a brook,
While each pursued his work in his own way.

Giulio.

What are you seeking? Something you have lost?

The Stranger.

Something I hope to find.

 Giulio.

You dropped it here?

Was it of value? Not your purse, I hope.

The Stranger.

More precious than my purse.

Giulio.

Your lady’s ring?

A jewel, perhaps?

The Stranger.

A jewel of a sort;
But it may take a thousand years to trace it
Back to its rightful owner.

Giulio (laughing).

O, you are bitten

By the prevailing fashion. Since the plough
Upturned those broken statues, all the world
Is relic-hunting; but, my friend, you’ll find
No Aphrodite here.

The Stranger (picking up a fossil).

And yet I think

It was the sea, from which she rose alive,
That shaped these rocks and left these twisted shells
Locked up, like stone in stone. They must have lived
Once, in the sea.

Giulio.

Ah, now I understand.

You’re a philosopher,—one of those who tread
The dusty road to Nowhere, which they call
Science.

The Stranger.

All roads to truth are one to me.

Giulio.

Sir, you deceive yourself. Your road can lead

Only to error. The Adriatic lies
How many miles away? We stand up here
On these unchanging hills; and yet, to fit
Your theory, you would roll the seas above
The peaks of Monte Rosa.

The Stranger.

But these shells?

How did they come here?

Giulio.

Obviously enough,
The sea being where it is, it was the Flood
That left them here.

The Stranger.

Then Noah must have dropped them

Out of his Ark. They never crept so far;
And Noah must have dumped his ballast, too,
Among our hills; for all those rippled rocks
Up yonder were composed of blue sea-clay.
I have found sea-weed in them, turned to stone,
The claws of crabs, the skeletons of fish.
Think you that, if your Adriatic lay
Where it now lies, its little sidling crabs
Could scuttle through the Deluge to the hills?
Your Deluge must have risen above the tops
Of all the mountains. If it rose so high,
Then it embraced the globe, and made our earth
One smooth blue round of water. When it sank
What chasm received those monstrous cataracts?
Or was the sun so hot it sucked them up
And turned them into a mist?
Is not that tale
A racial memory, lingering in our blood,
Of realms that now lie buried in the sea,
Or isles that heaved up shining from the deep
In old volcanic throes?

Giulio.

I must confess
I always feel a pang, sir, when I see
A man of talent wasting his fine powers
On this blind road.

The Stranger.

Show me a better way.

Giulio.

The way of Art, sir.

The Stranger.

Yes. That is a road

I have wished that I might travel. But are you sure
Our paths are not eventually the same?
Why have you climbed up here? To paint the truth,
As you perceive it, in those rocks and trees.
Suppose that, with your skill of hand, you saw
The truth more clearly, saw the lines of growth,
The bones and structure of the world you paint,
And the great rhythm of law that runs through all,
Might you not paint them better even than now?
Might you not even approach the final cause
Of all our art and science,—the pure truth
Which also is pure beauty?

Giulio.

Genius leaps
Like lightning to that mark, sir, and can waive
These pains and labours.

The Stranger.

O, I have no doubt
That you are right. I speak with diffidence,
And as a mere spectator; one who likes
To know, and seizes on this happy chance
Of learning what an artist really thinks.

Giulio.

We artists, sir, are not concerned with laws,

Except to break them. Genius is a law
Unto itself.

The Stranger.

And that is why you’ve made

Your wood-smoke blue against that shining cloud?
Against the darker background of the hill
It is blue in nature also; but it turns
To grey against the sky.

Giulio.
Welcome to our website – the ideal destination for book lovers and
knowledge seekers. With a mission to inspire endlessly, we offer a
vast collection of books, ranging from classic literary works to
specialized publications, self-development books, and children's
literature. Each book is a new journey of discovery, expanding
knowledge and enriching the soul of the reade

Our website is not just a platform for buying books, but a bridge
connecting readers to the timeless values of culture and wisdom. With
an elegant, user-friendly interface and an intelligent search system,
we are committed to providing a quick and convenient shopping
experience. Additionally, our special promotions and home delivery
services ensure that you save time and fully enjoy the joy of reading.

Let us accompany you on the journey of exploring knowledge and

personal growth!

ebookfinal.com