FUNDAMENTALS OF DIGITAL AUDIO New Edition

THE COMPUTER MUSIC AND DIGITAL AUDIO SERIES John Strawn, Founding Editor James Zychowicz, Series Editor Digital Audio Signal Processing Edited by John Strawn Composers and the Computer Edited by Curtis Roads Digital Audio Engineering Edited by John Strawn Computer Applications in Music: A Bibliography Deta S. Davis The Compact Disc Handbook Ken C. Pohlman Computers and Musical Style David Cope MIDI: A Comprehensive Introduction Joseph Rothstein William Eldridge, Volume Editor Synthesizer Performance and Real-Time Techniques Jeff Pressing Chris Meyer, Volume Editor Music Processing Edited by Goffredo Haus Computer Applications in Music: A Bibliography, Supplement I Deta S. Davis Garrett Bowles, Volume Editor General MIDI Stanley Jungleib Experiments in Musical Intelligence David Cope Knowledge-Based Programming for Music Research John W. Schaffer and Deron McGee Fundamentals of Digital Audio Alan P . Kefauver The Digital Audio Music List: A Critical Guide to Listening Howard W. Ferstler The Algorithmic Composer David Cope The Audio Recording Handbook Alan P . Kefauver Cooking with Csound Part I: Woodwind and Brass Recipes Andrew Horner and Lydia Ayers Hyperimprovisation: ComputerInteractive Sound Improvisation Roger T. Dean Introduction to Audio Peter Utz New Digital Musical Instruments: Control and Interaction Beyond the Keyboard Eduardo R. Miranda and Marcelo M. Wanderley, with a Foreword by Ross Kirk Fundamentals of Digital Audio New Edition Alan P . Kefauver and David Patschke

Volume 22

THE COMPUTER MUSIC AND DIGITAL AUDIO SERIES

FUNDAMENTALS OF DIGITAL AUDIO New Edition

Alan P . Kefauver and David Patschke

A-R Editions, Inc.

Middleton, Wisconsin

Library of Congress Cataloging-in-Publication Data

Kefauver, Alan P. Fundamentals of digital audio / By Alan P. Kefauver and David Patschke. -- New ed. p. cm. -- (Computer music and digital audio series) ISBN 978-0-89579-611-0 1. Sound--Recording and reproducing--Digital techniques. I. Patschke, David. II. Title. TK7881.4.K4323 2007 621.389'3--dc22 2007012264

A-R Editions, Inc., Middleton, Wisconsin 53562 2007 All rights reserved. Printed in the United States of America 10 9 8 7 6 5 4 3 2 1 1. MusicData
processing. 2. MusicComputer programs.

Contents

List of Figures Preface to the New Edition Chapter One The Basics
Sound and Vibration The Decibel 9 16 19 The Analog Signal Synchronization 1

ix xiii 1

Chapter Two

The Digital PCM Encoding Process

Sampling 23 27 32 Quantization

23

Analog-to-Digital Conversion

Chapter Three The Digital Decoding Process

Data Recovery 45 Error Detection and Correction Demultiplexing Sample and Hold Output Amplier 50 50 53 53 56 Digital-to-Analog Converter Reconstruction Filter 47

45

Chapter Four

Other Encoding/Decoding Systems

Higher-Bit-Level Digital-to-Analog Converters Oversampling Digital-to-Analog Converters Oversampling Analog-to-Digital Converters One-Bit Analog-to-Digital Converters Direct Stream Digital (DSD) 61 62 59 57 57 59

57

High Denition Compact Digital (HDCD)

Chapter Five

Data Compression Formats

Lossless Compression Lossy Compression 65 64

63

Chapter Six

Tape-Based Storage and Retrieval Systems

Rotary Head Tape Systems Record Modulation Systems 76 81 86 84 Digital Audiotape (DAT) Systems Multitrack Rotary Head Systems

75

8mm, 4mm, and Digital Linear Tape (DLT) Storage Systems 91 Fixed-Head Tape-Based Systems 92

Chapter Seven Disk-Based Storage Systems

Optical Disk Storage Systems 101 Magnetic Disk Storage Systems (Hard Disk Drives) Solid State (Flash) Memory 125

101
120

Chapter Eight

Digital Audio Editing

Tape-Based Editing Systems Disk-Based Editing Systems 127 130 133

127

Personal Computers and DAWs

Chapter Nine

The Digital Editing and Mastering Session

Tracks, Input/Output Channels, and Tracks The Editing Session 152 161 The Multitrack Hard-Disk Recording Session 149

147

Chapter Ten

Signal Interconnection and Transmission

Electrical Interconnection Optical Interconnection Digital Audio Broadcast 167 170 174

167

Glossary Further Reading/Bibliography Index

177 185 187

List of Figures

Chapter 1 Figure 1.1 Figure 1.2 Figure 1.3 Figure 1.4 Figure 1.5 Figure 1.6 Figure 1.7 Figure 1.8 Chapter 2 Figure 2.1 Figure 2.2 Figure 2.3 Figure 2.4 Figure 2.5 Figure 2.6 Figure 2.7 Figure 2.8 Figure 2.9 Figure 2.10 Figure 2.11 Figure 2.12 Figure 2.13 Chapter 3 Figure 3.1 Figure 3.2 Figure 3.3

A sound source radiating into free space. A musical scale. The musical overtone series. The envelope of an audio signal. A professional sound level meter (courtesy B and K Corporation). Typical sound pressure levels in decibels. The inverse square law. The Robinson-Dadson equal loudness contours.

Waveform sampling and the Nyquist frequency. Waveform sampling at a faster rate. Waveform sampling and aliasing. Voltage assignments to a wave amplitude. Comparison of quantization numbering systems. Offset binary and twos complement methods. A waveform and different types of digital encoding. Block diagram of a PCM analog-to-digital converter. Filter Schematics. Various lter slopes for anti-aliasing. The sample-and-hold process. A multiplexer block schematic. Interleaving.

Block diagram of a digital-to-analog converter. Reshaping the bit stream. Three error correction possibilities.
ix

FUNDAMENTALS OF DIGITAL AUDIO

Figure 3.4 Figure 3.5 Figure 3.6 Figure 3.7 Chapter 4 Figure 4.1 Figure 4.2 Chapter 5 Figure 5.1 Figure 5.2 Figure 5.3 Figure 5.4 Figure 5.5 Figure 5.6 Figure 5.7 Chapter 6 Figure 6.1 Figure 6.2

An 8-bit weighted resistor network converter. A dual slopeintegrating converter. The effects of hold time on the digital-to-analog process. Reconstruction of the audio signal.

The effect of oversampling digital-to-analog converters. PCM and PWM (1-bit) conversion.

Compression/decompression in the digital audio chain. A block diagram of lossless encoding and decoding. A block diagram of an MPEG-1 Layer I (MP1) encoder. A block diagram of an MPEG-1 Layer III (MP3) encoder. A block diagram of an AAC encoder. A block diagram of an AC-3 encoder. ATRAC codec overview.

Figure 6.3 Figure 6.4 Figure 6.5 Figure 6.6 Figure 6.7 Figure 6.8 Figure 6.9 Figure 6.10 Figure 6.11

a. Perspective view of tape wrap around a video head drum; b. top view of tape wrap around a video head drum showing details. a. Track layout on an analog 3/4 helical scan videotape recorder; b. PCM-1630 processor and associated DMR-4000 U-matic recorder (courtesy Sony Corp.). a. Tape wrap on a DAT recorders head drum; b. track layout on a DAT recorder. Channel coding for record modualtion. Frequency allocation for a. an 8mm and b. a Hi8 video recorder. A Hi8mm-based eight-channel multitrack recorder (courtesy Tascam) showing a. the main unit and b. the controller. An S-VHS-based eight-channel multitrack recorder (courtesy Alesis). Cross-fading between the read and write heads on a digital tape recorder. Track layout on a DASH 48- and 24-track digital tape recorder. Cross-fading between the two write heads on a DASH digital tape recorder. a. Transport layout on a Sony PCM-3324A digital tape recorder; b. a 48track multitrack digital tape recorder (photo courtesy Sony Corp.).

LIST OF FIGURES

xi

Figure 6.12 Chapter 7 Figure 7.1 Figure 7.2

Cross-fading between the read and write heads on a ProDigi digital multitrack recorder. Pit spacing, length and width for the compact disc. Note how pit length denes the repeated zeroes. Light reected directly back maintains the 0 series while the transition scatters the beam, denoting a change. Dust particles on the substrate are out of focus. Compact disc specications. The compact disc pressing process. Compact Disc Interactive (CD-i) audio formats. a. The front panel of a CD-R machine; b. the rear of the same machine showing the RS-232 interface as well as the ES/EBU, S/PDIF, and analog inputs and outputs (courtesy Tascam). Super CD. The recording system for the MiniDisc. The playback system for the MiniDisc. A professional magneto-optical recorder (courtesy Sony Corp.). MiniDisc specications. A computer hard disk showing sectors and tracks. A computer hard disk stack showing the concept of cylinders. Various ash-memory products, from left to right: Secure Digital, CompactFlash, and Memory Stick (courtesy SanDisk). The assemble edit process. A professional digital editor for tape-based digital recorders (photo courtesy Sony Corp.). a. A 1-Terabyte RAID Level-0 conguration diagram; b. a 500-Gb RAID Level-1 conguration diagram. a. The front panel of an 8-channel A-to-D (and D-to-A) converter; b. the back panel of the same unit. Notice the FireWire connections (photos courtesy Tascam). A compressor software plug-in for a computer-based digital audio workstation. A software instrument plug-in (specically, a Hammond B3 emulator/ virtual instrument) for use with a MIDI controller on a computer-based digital audio workstation.

Figure 7.3 Figure 7.4 Figure 7.5 Figure 7.6

Figure 7.7 Figure 7.8 Figure 7.9 Figure 7.10 Figure 7.11 Figure 7.12 Figure 7.13 Figure 7.14 Chapter 8 Figure 8.1 Figure 8.2 Figure 8.3 Figure 8.4

Figure 8.5 Figure 8.6

xii

FUNDAMENTALS OF DIGITAL AUDIO

Figure 8.7 Figure 8.8 Figure 8.9

Figure 8.10 Figure 8.11 Chapter 9 Figure 9.1 Figure 9.2 Figure 9.3 Figure 9.4 Figure 9.5 Figure 9.6

Diagram of a computer-based digital audio workstation. A DAW processor card for a PCI slot in a personal computer (photo courtesy Avid/Digidesign). A multi-channel A-to-D (and D-to-A) converter that connects directly to a processor card located in a computer-based DAW (photo courtesy Avid/Digidesign). An example of a control surface to manually adjust DAW software (courtesy Tascam). An example of a stand-alone DAW unit (photo courtesy Tascam).

Figure 9.7

Figure 9.8 Figure 9.9

Figure 9.10 Figure 9.11 Figure 9.12 Figure 9.13

Editing screen from Digidesign Pro Tools software. Editing screen from Apple Logic Pro software. Editing screen from Steinberg Nuendo software. Mixing window from Digidesign Pro Tools software. Mixing window from Apple Logic Pro software. a. A portion of 2-track (stereo) material selected in Digidesign Pro Tools software; b. that same portion being duplicated and dragged into a new stereo track. An edit window in Digidesign Pro Tools software showing a. nominal resolution of a waveform, b. zooming in on a portion of the waveform, and c. maximum magnication of the waveform for precise selection. The cross-fade and -3dB down point of a digital edit. The crossfade editing process in Digidesign Pro Tools software: a. selecting the portion of material to crossfade, and determining the precise options for the fade; b. The resulting waveform after the fade. A magnetic tape splice with a linear slope. The crossfade editor in Apple Logic Pro software. a. An example of an input matrix for Digidesign Pro Tools software; b. An output matrix example for the same. A multitrack project opened in Digidesign Pro Tools software. Note that although there are 18 tracks (some of them stereo, even), in this example the hardware supports only up to 8 analog outputs and inputs. An equalization software plug-in for a software-based DAW. A reverb software plug-in for a software-based DAW.

Figure 9.14 Figure 9.15 Chapter 10 Figure 10.1

Frequencies and their uses.

Preface to the New Edition

Back when I was in schooland it wasnt so long agobeing in a recording studio was an immense (and immersive) experience. Humongous mixers lled whole rooms, with just enough space left for a couple oor-to-ceiling racks full of graphic equalizers, reverb units, noise reduction, gates, and all manner of sound modication devices. Another wall was lled with large multitrack tape recorders. Im remembering the good old days as I sit on my deck, typing this on my laptop computer that has built into it a humongous mixer; equalizers, reverb units, gates and all manner of sound modication devices; and the capability to record or export my nished audio programs in any number of professional formats. Whether because of the demand for quality and portability or because of its leaner data requirements, audio was at the forefront of the digital entertainment revolution. Audio and music today exist almost exclusively in digital form for three reasons: reduced manufacturing costs, Moores Law, and the Internet. The Internet you know about: a sprawling, worldwide network of computers through which increasing numbers of people get their information and entertainment and do increasing amounts of their business. Moores Law goes something like this: Computers double in speed for the same cost every 18 months. In addition, the price-per-byte of memory continues to fall similarly from year to year. I remember spending $200 for 256 kilobytes of memory for my rst computer (an 8 MHz model). Today, that same $200 would buy about two gigabytesalmost an 8,000x increase. What all this means is that even off-theshelf computers today are more than adequate to record, process, edit, and listen to multiple channels of digital audio. Once available only to the most elite recording institutions, high quality digital audio recording is now accessible to almost everyone. Consequently, unlike years ago, when high-quality equipment and restricted distribution dictated what would be recorded and how it would be heard, today anyone can be an artist, engineer, and label. Some people do well with this accessibility and others do not. Learning the fundamentals wont give you insight for making subjective production decisions, but it will give you a solid basis for making them. Dont get too comfortable with your

xiii

xiv

FUNDAMENTALS OF DIGITAL AUDIO

new high tech equipment and skills though, because the one fact about this industry is that everything will continue to change, like it or not, and faster than you would prefer. That makes understanding the basic concepts and principles underlying your equipment all the more important. And that is what this book hopes to accomplish.

ONE
The Basics

SOUND AND VIBRATION

Sound has been discussed many times and in many ways in almost every book written about recording and the associated arts. Yet, without a thorough knowledge of the nature of sound, understanding how analog sound is handled and converted to and from the digital environment is impossible. Sound, by its nature, is illusive, and dening it in concise terms is not easy.

The Basic Characteristics of Sound

The major components of sound are frequency, amplitude, wavelength, and phase. Other components are velocity, envelope, and harmonic structure. Because we discuss sound as a wave later in this book, now may be a good time to visualize a sound wave in the air. Figure 1.1 shows a sound wave traveling through free space. At the source of the sound (point a), a sound wave radiates omnidirectionally away from itself. The sound energy from the sound source is transferred to the carrying medium (in this case air) as air compressions (point b) and rarefactions (point c). The air itself does not move, but the movement of individual air particles is transferred from one particle to the next in much the same way that waves move on the surface of the sea. It is probably easiest to visualize the creation and movement of soundwaves by imagining a large drum begin struck: the head (or skin) of the drum moves in and out after it is struck, causing subsequent air compressions and rarefactions emanating away from the drum.

Velocity
The speed of energy transfer equals the velocity of sound in the described medium. The velocity of sound in air at sea level at 70 degrees Fahrenheit is 1,130 feet per second (expressed metrically as 344 meters per second). The velocity of sound depends on the medium through which the sound travels. For example, sound travels
1

FUNDAMENTALS OF DIGITAL AUDIO

a. Sound source

b. Compression

c. Rarefaction

FIGURE 1.1 A sound source radiating into free space.

through steel at a velocity of about 16,500 feet per second. Even in air, the velocity of sound depends on the density of the medium. For example, as air temperature increases, air density drops, causing sound to travel faster. In fact, the velocity of sound in air rises about 1 foot per second for every degree the temperature rises. The formula for the velocity of sound in air is: V 49 459.4 F

THE BASICS

Note in Figure 1.1 that as sound moves farther from its source, its waves become less spherical and more planar (longitudinal).

Wavelength
The distance between successive compressions or rarefactions (i.e., the sound pressure level over and under the reference pressure) is the wavelength of the sound that is produced. The length of the sound wave is itself not that useful for our purposes. Wavelength Velocity Frequency

However, the frequency of the sound is important. Because we know the velocity of sound, it is easy to determine the frequency when the wavelength is known. The simple formula can be changed to read: Frequency Velocity Wavelength

Thus, if the distance from compression to compression is 2.58 feet, the frequency is 440 cycles per second, or 440 hertz (Hz). The period of the wave, or the time it takes for one complete wave to pass a given point, can be dened by the formula Period 1 Frequency

Therefore, the period of this wave is 1/440 of a second. As the period becomes shorter, the frequency becomes higher. This sound, with a frequency of 440Hz and a period of 1/440 of a second, is referred to in musical terms as A, that is, the A above middle C on the piano. Although sounds exist both higher and lower and range varies from person to person, well assume generally that the range of human hearing is approximately 20 Hz to 20,000Hz (20kHz).

Musical Notation and Frequency

It is beyond the scope of this book to teach the ability to read musical notation. However, this skill is essential to becoming a competent recording engineer, and the student is strongly advised to pursue the study of music if preparing for a career in the recording arts. As mentioned above, the note A has a frequency of 440Hz (this is the note occupying the second space of the treble-clef staff). The A that is located on the top line

FUNDAMENTALS OF DIGITAL AUDIO

of the bass clef staff is an octave below 440Hz and has a frequency of 220Hz. An octave relationship is a doubling or halving of frequency. Figure 1.2 shows a musical scale with the corresponding frequencies.

Harmonic Content
Very few waves are a pure tone (i.e., a tone with no additional harmonic content). In fact, only a sine wave is a true pure tone. When an object (e.g., a bell) is struck, several tones at different frequencies are produced. The fundamental resonance tone of the bell is heard rst, followed by other frequencies at varying amplitudes. The next tone is a doubling of the fundamental frequency (an octave) above that, followed by another doubling that is heard as a musical interval of a fth. Each of these harmonics changes amplitude slightly over time, adding to the individual characteristics of the sound. For example, a bell with a fundamental frequency of 64Hz produces harmonics of 128Hz and 192Hz, which is a G, a fth above the second C. Many other harmonics at varying amplitudes are produced, depending on the metallic composition of the bell. These harmonics are arranged in a relationship called the overtone series, and the combination of these harmonies, or overtones, gives sound its specic timbre, or tone coloration. The difference in harmonic content makes an oboe sound different from a clarinet. Although an oboe and a clarinet produce the same note with the same fundamental frequency, the number of overtones and the amplitude of each differs. The overtone series is most often notated in the musical terminology of octaves, fths, fourths, and so on, but actually corresponds to the addition of the fundamental frequency. Therefore, an overtone series based on the note C is, in hertz, 65, 130, 195, 260, 325, 390, 455, 520, 585, 650, 715, and so on. In musical terms this is a series of the fundamental followed by an octave, a perfect fth, a perfect fourth, a major third, a minor third, another minor third, three major seconds, and a minor

C 130 Hz

A 220 Hz

B c 260 Hz d e f g a 440 Hz b c1 520 Hz

FIGURE 1.2 A musical scale.

THE BASICS

second (see Figure 1.3). The frequency that is twice the frequency of the fundamental is called the second harmonic even though it is the rst overtone. Confusion often exists concerning this difference in terminology. The third harmonic is the rst non-octave relationship above the fundamental (it is the fth), and as such, any distortion in this harmonic tone is often detected before distortion is heard in the tone a fth below (the second harmonic). Many analog audio products list the percentage of third harmonic distortion found (because it is the most audible) as part of their specications.

The Sound Envelope

All sounds have a start and an end. The drop in amplitude, called a decay, often occurs after the start of the sound and is followed by a sustain and a nal drop in amplitude before the event ends completely. The four main parts of the sound envelope are (1) the attack, (2) the initial decay, (3) the sustain (or internal dynamic), and (4) the nal decay (often referred to as the release). These are shown in Figure 1.4. The attack is the time that it takes for the sound generator to respond to whatever has set it into vibration. Different materials, with their different masses, have different attack times. How an object is set into vibration also affects its attack time. A softly blown ute has a longer attack time than a sharply struck snare drum. In general, struck instruments have an attack time that is much faster (in the 1- to 20-millisecond range) than wind instruments (in the 60- to 200-millisecond range). String attack times vary, depending on whether the instrument is bowed or plucked. The attack time of an instrument can be represented by an equivalent frequency based on the following formula: T 1 Frequency

4th harmonic 3rd harmonic 2nd harmonic Fundamental

FIGURE 1.3 The musical overtone series.

FUNDAMENTALS OF DIGITAL AUDIO

Initial attack

Sustain Initial decay

Final decay

FIGURE 1.4 The envelope of an audio signal.

Rearranging this formula gives: Frequency 1 T This means that i an instrument has an attack time of 1 millisecond, the equivalent frequency is 1 kilohertz (kHz). This fact is important to remember when trying to emphasize the attack of an instrument. The prudent engineer remembers this when applying equalization, or tonal change, to a signal. The initial decay, which occurs immediately after the attack on most instruments, is caused by the cessation of the force that set the tube, string, or medium into vibration. It is the change in amplitude between the peak of the attack and the eventual leveling-off of the sound, known as the sustain. The length of the sustain

THE BASICS

varies, depending on whether the note is held by the player for a specic period of time (e.g., when a trumpet player holds a whole note) or on how long the medium continues to vibrate (medium resonance) before beginning the nal decay, which occurs when the sound is no longer produced by the player or by the resonance of the vibrating medium. As the trumpet player releases a held note, the air column in the instrument ceases to vibrate, and the amplitude decays exponentially until it is no longer audible. Final decays vary from as short as 250 milliseconds to as long as 100 seconds, depending on the vibrating medium. However, not all frequencies or harmonics decay at the same rate. Most often, the high-frequency components of the sound decay faster than the low-frequency ones. This causes a change in the sounds timbre and helps dene the overall sound of the instrument (or other sound-producing device).

Masking
Many references can be found in the literature about the equal loudness contours (discussed later in this chapter) developed by Fletcher and Munson and later updated by Robinson and Dadson. These curves relate our perception of loudness at varying frequencies and amplitudes and apply principally to single tones. How many times have you sat in a concert hall listening to a recital in which one instrumentalist drowned out another? Probably more than once. When you consider the fact that a piano, even at half-stick (i.e., with the lid only partially raised), can produce a level of around 6 watts, whereas a violinist, at best, produces 6 milliwatts, it is easy to understand why the violinist cannot be heard all the time. Simply stated, loud sounds mask soft ones. The masking sound needs to be only about 6 decibels higher than the sound we want to hear to mask the desired sound. It makes perfect sense that a loud instrument will cover a softer one if they are playing the same note or tone, but what if one instrument is playing a C and another an A? Studies by Zwicker and Feldtkeller have shown that even a narrow band of noise can mask a tone that is not included in the spectrum of the noise itself. For example, a low-frequency sine wave can easily mask, or apparently reduce the level of, a higher sinusoidal note that is being sounded at the same time. This masking occurs within the basilar structure of the ear itself. The louder tone causes a loss in sensitivity in neighboring sections of the basilar membrane. The greatest amount of masking occurs above the frequency of the masking signal rather than below it, and the greater the amplitude of the masking signal, the wider the frequency range masked. However, when a note is produced by an instrument that has a complicated, dense sound spectrum (i.e., a note with a rich harmonic structure), that sound will usually mask sounds that are less complicated (less dense). In fact, many newer buildings that use modular, or carrel, types of ofce space are equipped to send constant low-level random wide-band noise through loudspeakers in their ceilings.

FUNDAMENTALS OF DIGITAL AUDIO

This masking signal keeps the conversations in one ofce carrel from intruding into adjacent ofce space. The level of the masking signal is usually around 45 decibels, and this (plus the inverse square law, which is discussed later) effectively provides speech privacy among adjacent spaces. This effect is used also by some noise reduction systems when, as the signal rises above a set threshold, processing action is reduced or eliminated. In addition, the masking effect is a critical part of the data-reduction systems used in some of the digital audio storage systems discussed later in this book.

Localization
A person with one ear can perceive pitch, amplitude, envelope, and harmonic content but cannot determine the direction from which the sound originates. The ability to localize a sound source depends on using both ears, often referred to as binaural hearing. Several factors are involved in binaural hearing, depending on the frequency of the sound being perceived. The ears are separated by a distance of about 6 1/2 or 7 inches so that sound waves diffract around the head to reach both ears. When the wavelength of sound is long, the diffraction effect is minimal and the comparative amplitude at each ear about the same. However, when the wavelength is short, the diffraction effect is greater and the sound attenuated at the farther ear. Because the sound has to travel farther relative to wavelength, there is a perceptual time difference factor as well. You may have noticed that it is easier to locate high-frequency sounds than low-frequency ones. In fact, low-frequency signals often appear to be omnidirectional because of this effect. We can say that high frequencies (above 1kHz) are localized principally by amplitude and time differences between the two ears. How, then, do we localize low frequencies? With all sound there is a measurable time-of-arrival difference between the two ears when the sound is to one side or the other. As the sound moves to a point where it is equidistant from both ears, these time differences are minimized. With longer wavelengths the time-of-arrival differences are less noticeable because the ratio of the time difference to the period of the wave is large. However, this creates phase differences between the two ears, allowing the brain to compute the relative direction of the sound. Therefore, we can say that low frequencies are located by intensity and phase differences.

The Haas Effect and the Inverse Square Law

Although much has been said and written about the Haas Effect, also known as the precedence effect, and the inverse square law, we can say that the sound we hear rst denes for us the apparent source of the sound. Consider the following example. Two trumpet players stand in front of you, one slightly to the right at a distance of 5 feet and another slightly to the left at a distance of 10 feet. If they play

THE BASICS

the same note at the same amplitude, you will localize the sound to the right because it is louder. This amplitude difference is due to the inverse square law, which states that for every doubling of distance there is a 6-decibel loss in amplitude in free space (i.e., where there are no nearby reecting surfaces). Now suppose that both players sustain their notes. If player A (on the left) increases his amplitude by 6 decibels, the sound levels will balance, but your ear-brain combination will insist that player B (on the right) is closer to you than player A. Although the levels have been equalized, you perceive the nearer player to be closer. You would think that as long as the levels are identical at both ears, the players would appear to be equidistant from you. According to Haas, Our hearing mechanism integrates the sound intensities over short time intervals similar, as it were, to a ballistic measuring instrument. This means that the ear-brain combination integrates the very short time differences between the two ears, causing the sound with the shortest timing differences to appear louder and therefore closer. Haas used two sound sources and, while delaying one, asked a subject to vary the loudness of the other source until it matched the sound level of the delayed sound. He found that where the delay was greater than 5 but less than 30 milliseconds, the amplitude of the delayed source had to be 10 decibels louder than the signal from the nondelayed source for the two sounds to be perceived as equal. Beyond a delay of 30 milliseconds a discrete echo was perceived, and prior to 5 milliseconds the level needed to be increased incrementally as the delay lengthened.

THE DECIBEL
Earlier in this chapter we discussed the basic characteristics of sound, but one that was conspicuously absent was amplitude. The unit of measure that is normally used to dene the amplitude of sound levels is the decibel (dB). A Bel is a large, rather cumbersome unit of measure, so for convenience it is divided into 10 equal parts and prefaced with deci to signify the one-tenth relationship. Therefore, a decibel is one tenth of a Bel. A sound level meter measures sound in the environment. A large commercial aircraft taking off can easily exceed 130dB, whereas a quiet spot in the summer woods can be a tranquil 30dB. Most good sound level meters are capable of measuring in the range of 0dB to 140dB. You might think that the 0dB level is the total absence of sound, but it is not. Actually, 0dB is the lowest sound pressure level that an average listener with normal hearing can perceive. This is called the threshold of hearing. The 0dB reference level corresponds to a sound pressure level of 0.00002 dynes per square centimeter (dynes/cm2 ), which, referenced to power or intensity in watts, equals 0.000000000001 watts per square meter (W/m2 ). Also referred to in this book is the threshold of feeling, which is typically measured as an intensity of 1W/m2. A typical professional sound level meter is shown in Figure 1.5.

10

FUNDAMENTALS OF DIGITAL AUDIO

FIGURE 1.5 A professional sound level meter (courtesy B and K Corporation).

As you can see, the decibel must have a reference, which, when we discuss sound levels in an acoustic environment, is the threshold of hearing. In fact, the decibel is dened as 10 times the logarithmic relationship between two powers. The formula for deriving the decibel is: dB 10log PowerA Power B

where PowerA is the measured power and PowerB the reference power. We can use this formula to dene the amplitude range of human hearing by substituting the threshold of hearing (0.000000000001W/m2 ) for the reference power and using the threshold of feeling (1W/m2) as the measured power. The formula, with the proper values inserted, looks like this:

THE BASICS

11

10log

1W/m2 120dB 1 1012 W/m2

Therefore, the average dynamic range of the human ear is 120dB. Figure 1.6 shows typical sound pressure levels, related to the threshold of hearing, found in our everyday environment. Note also the level called the threshold of pain. It is interesting to note that other values can be obtained using the power formula. For example, if we use a value of 2W in the measured power spot and a value of 1W in the reference power spot, we nd that the result is 3dB. That is, any 2-to-1 power relationship can be dened simply as an increase in level of 3dB. Whether the increase in power is from 100W to 200W or from 2,000W to 4,000W, the increase in level is still 3dB. The inverse square law was mentioned earlier in this chapter. You might surmise that doubling distance would cause a sound level loss of one half, or -3dB. However, you must remember that sound radiates omnidirectionally from the source of the sound. Recall from your high school physics class that the area of a sphere is determined by the formula a 4r2. It follows that when a source radiates to a point that is double the distance from the rst, it radiates into four times the area instead of twice the area. This causes a 6dB loss of level instead of a 3dB loss. The formula for the inverse square law is: 2 2 2 Level drop 10log r 20log r 6dB r1 r1

where r1 equals 2 feet and r2 equals 4 feet. Figure 1.7 shows this phenomenon. Note that at a distance of 2 feet, the sound pressure level is 100dB. When the listener moves to a distance of 4 feet (i.e., twice the original distance), the level drops to 94dB.

Equal Loudness Contours

Ambient, or background, noise is all around us. Even in a quiet concert hall, the movement of air in the room can be 30dB or more above the threshold of hearing at one or more frequencies. However, the ear is not equally sensitive to sound pressure at all frequencies. The ear is most sensitive to the frequency range between 3,000Hz and 5,000Hz. A tone that is heard just at the threshold of hearing (1 x 10-12W/m2) at a frequency of 4,000Hz corresponds to a sound pressure level of 0dB. For the tone to be perceived by the same listener at the same loudness level when the frequency is lowered to 400Hz, the amplitude of the tone must be raised about 12dB. If we lower the frequency to 40Hz, the level must be raised nearly 45dB to be perceived at the same volume. Figure 1.8 shows the equal loudness contours developed by Fletcher and Munson and updated later by Robinson and Dadson.

12

FUNDAMENTALS OF DIGITAL AUDIO

Level in dB

180 170 160 150 140

Rocket engines

Jet engine (close up)

Threshold of pain Jackhammer Threshold of feeling Full symphony orchestra @FFF Subway train Heavy truck traffic

Danger Levels

130 120 110 100 90 80 70 60 50 40 30 20 10 0

Average full symphony orchestra Power lawn mower Average factory Chamber orchestra

Average conversation Average office Subdued conversation

Quiet concert hall Quiet recording studio Dripping water

Threshold of hearing

FIGURE 1.6 Typical sound pressure levels in decibels.

THE BASICS

13

r1 = 2 feet r2 = 4 feet r1

Sound source

r2

FIGURE 1.7 The inverse square law.

The contours are labeled phons, which range from 10 to 120; at 1,000Hz the phon level is the same as the sound pressure level in decibels. Therefore, the phon is a measure of equal loudness. Figure 1.8 has a curve labeled MAF, which stands for minimum audible eld. This curve, equal to 0 phons, denes the threshold of hearing. Note that at low sound pressure levels, low frequencies must be raised substantially in level to be perceived at the same loudness as 1kHz. Frequencies above 5kHz (5,000Hz) need to be raised as well, although not as much. Looking at Figure 1.8, you can see that the discrepancies between lows, highs, and mid tones are reduced as the level of loudness increases. The 90-phon curve shows a variation of 40dB, whereas the 20-phon curve shows one of nearly 70dB. The sound level meter in Figure 1.5 has several weighting networks so that at different sound pressure levels the meter can better approximate the response of the human ear. The A, B, and C weighting networks correspond to the 40-, 70-, and 100phon curves of the equal loudness contours.

14

FUNDAMENTALS OF DIGITAL AUDIO

120

120 110

Loudness Level (phons)

100

100 90

Sound Pressure Level (dB re 20 N/m

80

80 70

60

60 50 40 30

40

20 10 0 MAF

20 10

20

100 Frequency (Hz)

1,000

5,000

10,000

FIGURE 1.8 The Robinson-Dadson equal loudness contours.

Logarithms
As you may have noticed, the formula used to dene the decibel applied logarithms, abbreviated log. Anyone involved in the audio engineering process needs to understand logarithms. In brief, a logarithm of a number is that power to which 10 must be raised to equal that numbernot multiplied, but raised. The shorthand

THE BASICS

15

notation for this is 10x, where x, the exponent, indicates how many times the number is to be multiplied by 10. Therefore, 103 is 10 raised to the third power, or simply 10 to the third: 101 = 10 102 = 100 103 = 1,000 109 = 1,000,000,000 Numbers whose value is less than 1 can be represented with negative exponents, such as 0.1 = 10 -1 0.01 = 10 -2 0.001 = 10 -3 0.000001 = 10 -6 Because we are talking about very large and very small numbers, prex names can be added to terms such as hertz (frequency) and ampere (a measure of current ow) to indicate these exponents. Therefore, for large numbers, 1,000 cycles per second = 103 hertz (or 1 kilohertz [1kHz]) 106 hertz = 1 megahertz or (1MHz) 109 hertz = 1 gigahertz (1GHz) 1012 hertz = 1 terahertz (1THz) For small numbers, ampere (A) = 10-3 A (or 1 milliamp [1mA]) 10-6A = 1 microamp (1A) 10-9A = 1 nanoamp (1nA) 10-12A = 1 picoamp (1pA) Now you can see that the threshold of hearing, dened earlier as 0.000000000000W/ m2, appears in the formula as 1 x 10-12W/m2. It is also helpful to know that when powers of 10 are multiplied or divided, you simply add or subtract the exponents. Therefore, 106 x 103 = 106+3 = 109, and 1012 109 = 1012-9 = 103. Logarithms of numbers also exist between, for example, 1 and 10 and 10 and 100, but we will leave those problems to the mathematicians. Today, the logarithm of any number can easily be found by either looking them up in a table or pushing the log button on a calculator.

16

FUNDAMENTALS OF DIGITAL AUDIO

THE ANALOG SIGNAL

In order to record, reproduce, or transmit sound, it rst needs to be transduced into an electrical signal. The beginning of this process requires a microphone. A microphone has a thin diaphragm that is suspended in or attached (in some fashion, depending on the type of microphone) to a magnetic eld. The diaphragm moves back and forth in reaction to the sound waves that pass through it, and that movement within the magnetic eld creates a small electrical signal, which is an electrical representation of the compressions and rarefactions of the sound wave. The signal is transmitted from the microphone along its cable to be amplied. Microphones generate only a tiny amount of signal (measured in volts), which needs ample amplication before it can be used for any recording or reproduction. It should be noted that entire volumes (and, indeed, careers) exist that detail microphones and capturing sound, which have been generalized for our purposes here.

Reference Levels and Metering

When we discuss reference levels, we are talking about the values of the current and voltage that pass through a cable between two pieces of equipment. Ohms law establishes some fundamental relationships that we should be aware of. Power, expressed in watts, is equal to either the square of the voltage divided by the resistance or the square of the current multiplied by the resistance in the circuit, that is, P = E2/R and P = I2R, respectively. In professional audio circuits we work with voltage levels (or their corresponding digital values) instead of power levels, and because the resistance in the circuit is constant, we do not need to be concerned with R at this time. To calculate the difference between two powers, we use the power formula: dB 10log PowerA Power B

Because we really want to know the decibel level referenced to volts, the formula should read:
2 dB 10log E A E2 B

To remove the squares from the voltages in the formula we can rewrite the expression as: dB 20log EA EB

THE BASICS

17

Note now that any 2-to-1 voltage relationship will yield a 6dB change in level instead of the 3dB change of the straight power formula. There is always a standard reference level in audio circuits. This reference level, an electrical zero reference, is equivalent to the voltage found across a common resistance in the circuit. Therefore, we can compare levels by noting how many decibels the signal is above or below the reference level.

The dBm
The standard professional audio reference level is +4dBm. The dBm is a reference level in decibels relative to 1 milliwatt (mW). Zero dBm is actually the voltage drop across a 600-ohm resistor in which 1mW of power is dissipated. Using Ohms law (P = E2/R) we nd that the voltage is 0.775 volts RMS. (This value is merely a convenient reference point and has no intrinsic signicance.) The meters that were used on the original audio circuits when this standard was enacted were vacuum tube volt meters (VTVMs). As the demand for more meters grew (as stereo moved to multitrack), a less expensive meter was needed. An accurate meter, which needs a low impedance, would load down the circuit it was measuring and thereby cause false readings. To compensate for the loading effect of the meter, a 3.6K resistor was inserted in the meter path. Now the meter no longer affected the circuit it was measuring, although it read 4dBm lower. When the meter was raised to 0VU (volume units), the actual line level was +4dBm. The notation dBu is often found in the specications in manuals that come with digital equipment. As was just mentioned, we know that the dBm is referenced to an impedance of 600 ohms. (This was derived from the early telephone company standards. Many of our audio standards originated with Ma Bell.) However, most circuits today are bridging circuits instead of the older style matching circuits, and the dBu is used as the unit of measure. Without going deeper into the subject of matching and bridging circuits, we can say that the dBu is equal to the dBm in most cases.

The Volume Unit

If the meters on the recording devices that we use to store digital audio displayed signal level in dBm, they would be very large and difcult to read. Considering the variety of system operating levels found on recording devices today, a comprehensive meter would have to range from -40dBm or -50dBm to +16dBm. This would be a range of around 76dB. To make this easier, we use the volume unit (VU) to measure signal level through the device. In most professional applications, this meter is calibrated so that an input level of +4dBm equals 0VU. However, if the device operates at -10dBm, as many consumer-type ones do, then -10dBm equals 0VU.

18

FUNDAMENTALS OF DIGITAL AUDIO

Consider an experiment where a calibration tone of 1kHz at 0VU is played back from an analog consumer device operating at a level of -10dBm and from a professional analog device operating at +4dBm. If those signals were both sent to an analog-to-digital converter, the digitized signal level would be the same from both machines. As this signal is played back from the storage medium, the digital-toanalog converter outputs the calibration tone and produces a level of -10dBm or +4dBm at the devices output, depending on the device. The same signal played back on either a consumer or professional device will produce the reference level output at the devices specied line level. If we were to compare the output of the two devices, the +4dBm machine would play back 14dB louder, but this difference is due to the output gain of the playback section amplier. The classic VU meter is calibrated to conform to the impulse response of the human ear. The ear has an impulse response, or reaction time, that is dened as a response time and a decay time of about 300 milliseconds. This corresponds to the reaction time of the ears stapedius muscle, which connects the three small bones of the middle ear to the eardrum. Therefore, any peak information shorter than 300 milliseconds will not be fully recognized by the meter. The VU meter is designed to present the average content of the signal that passes through it.

The Peak Meter

On average, certain types of music have shorter attack and decay times than others, and, as you will see in Chapter 2 during the discussions of digital recording, once full level is reached, there is no margin for error. A peak meter is a device that is designed to respond to all signals (no matter how short) that pass through it and is much more suitable for digital recording. The ballistics, or response time, of a standard DIN peak meter are designed to fulll these requirements. A peak meter will reach full scale in 10 milliseconds. However, if the peak meter reaching the input signal were allowed to fall back at the same rate, the eye would be unable to follow its rapid movement. The metered value is held in a capacitor for a specied amount of time and allowed to fall back using a logarithmic response. Because the peak meter is not a volume indicator, it is correct to read it in decibels instead of volume units. Analog recording uses several standards for what is called the zero level for a peak meter. The International Electrotechnical Commission (IEC) has dened these meters as the BBC, where Mark 4 equals 0dBm; the Type 1, where 0dB equals +6dBm; and the Type 2, where 0dB equals 0dBm. Traditionally, a volume indicator is calibrated so that there is some headroom above 0VU. Analog magnetic tape saturates slowly as levels rise, and typically 9dB of headroom above zero is allowed before reaching critical saturation levels. This scaling also accommodates peak information, as fast peaks tend to be about 9dB or 10dB higher than the average level. The BBC meters, as well as the Type 1 and Type 2 IEC meters, also allow for headroom above their zero reference. On the other hand, meters designed for digital audio are calibrated differently. Zero on a

THE BASICS

19

digital meter means full quantization level; that is, all bits at full value (there is no headroom). Therefore, it is important to note that 0VU on the console (whether it is +4dBm, -10dBm, or another level) does not equal 0dB on the digital meter. In early cases, -12dB was chosen as the calibration point for digital metering, but with the advent of systems with higher quantization levels, -18dB is more often used. To differentiate between peak meters and digital meters, the term dBfs is used (fs stands for full scale). This implies that the meter is on some kind of digital machine where 0dBfs equals full quantization level (all bits are used). Manufacturers have used a variety of standards, and some allow several decibels of level above zero as a hidden headroom protection factor. Perhaps in the future a digital metering standard will be adapted that everyone can adhere to. In the meantime the prudent engineer will read the manual for the piece of equipment in use and be aware of its metering characteristics. Most professional equipment in production today uses the standard of 0VU = -18dBfs.

SYNCHRONIZATION
Although not a basic characteristic or function of sound, time code is an important part of digital systems. Without time code, position locating and synchronization in the tape-based digital domain would be extremely difcult. Time code, as we know it today, was developed by the video industry to help with editing. In 1956, when videotape made its debut, the industry realized that the lm process of cut-andsplice would not work in video. The images that were visible on lm were not so on videotape. Certain techniques (e.g., magnetic ink that allowed you to see the recorded magnetic pulses) were developed, but these did not prove satisfactory. Another technique was to edit at the frame pulse or control track pulse located at the bottom edge of the videotape. This pulse tells the head how fast to switch in a rotating-head system. In the 1960s, electronic machine-to-machine editing was introduced, providing precise machine control and excellent frame-to-frame match-up. However, the splice point was still found by trial and error.

Time Code
A system was needed that would uniquely number each frame so that it could be precisely located electronically. Several manufacturers introduced electronic codes to fulll this task, but the codes were not compatible with one another. In 1969 the Society of Motion Picture and Television Engineers (SMPTE) developed a standard code that became recognized for its accuracy. That standard was also adopted by the European Broadcasting Union (EBU), making the code an international standard. The SMPTE/EBU code is the basis for all of todays professional video- and audiotape editing and synchronization systems.

20

FUNDAMENTALS OF DIGITAL AUDIO

The SMPTE time code is an addressable, reproducible, permanent code that stores location data in hours, minutes, seconds, and frames. The data consist of a binary pulse code that is recorded on the video- or audiotape along the corresponding video and audio signals. The advantages of this are (1) precise time reference, (2) interchange abilities among editing systems, and (3) synchronization between machines. On analog-based tape machines, the code can be stored in two different ways: longitudinal time code (LTC), which is an audio signal that is stored on a separate audio track of the video or audio machine; and vertical interval time code (VITC), which is small bursts of video integrated into the main video signal and stored in the vertical blanking interval (i.e., between the elds and frames of the video picture).

LTC
Longitudinal time code is an electronic signal that switches from one voltage to another, forming a string of pulses, which can be heard as an audible warbling sound if amplied. Each 1-second-long piece of code is divided into 2,400 equal parts when used in the NTSC standard of 30 frames per second or into 2,000 equal parts when used with the PAL/SECAM system of 25 frames per second. Notice how each system generates a code word that is 80 bits long: PAL/SECAM: 2,000 bits per second divided by 25 frames per second equals 80 bits per frame; NTSC: 2,400 bits per second divided by 30 frames per second equals 80 bits per frame. Most of these bits have specic values that are counted only if the time code signal changes from one voltage to another in the middle of a bit period, forming a 12-bit pulse, which represents a digital 1, whereas a full-bit pulse represents a digital 0. Following is one frames 80-bit code: Bits 03, 8, 9 1619, 2426 3235, 4042 4851, 56, 57 6479 10 11 27 Function Frame count Second count Minute count Hour count Sync word Drop frame Color frame Field mark

The remaining eight groups of 4 bits are called user bits, which can be used to store data such as the date and reel number. Three bits are unused. A typical time code number might be 18:23:45:28. This code number indicates a position 18 hours, 23 minutes, 45 seconds, and 28 frames into the event. This could be on the fortieth reel of tape. Time code does not need to start over on each reel.

THE BASICS

21

Bit 10, the drop-frame bit, tells the time code reader whether the code was recorded in drop-frame or non-drop-frame format. Black-and-white television has a carrier frequency of 3.6MHz, whereas color uses 3.58MHz. This translates to 30 frames per second as opposed to 29.97 frames per second. To compensate for this difference, a dened number of frames are dropped from the time code every hour. The offset is 108 frames (3.6 seconds). Two frames are dropped every minute of every hour except in the tenth, twentieth, thirtieth, fortieth, and ftieth minute. Frame dropping occurs at the changeover point from minute to minute. Bit 11 is the color frame bit, which tells the system whether the color frame identication has been applied intentionally. Color frames are often locked as AB pairs to prevent color shift in the picture. As mentioned earlier, user bits can accommodate data for reel numbers, recording date, or any other information that can be encoded into eight groups of 4 bits.

VITC
Vertical interval time code is similar in format to LTC. It has a few more bits, and each of the 9 data-carrying bits is preceded by 2 sync bits. At the end of each frame there are eight cyclic redundancy check (CRC) codes, which are similar to the codes used in digital recording systems. This generates a total of 90 bits per frame. The main difference between LTC and VITC is how they are recorded on tape, LTC being recorded on one of the videotapes longitudinal audio tracks or on a spare track of the audiotape recorder. Some specialized two-track recorders have a dedicated time code track between the two standard audio tracks. Playback and record levels should be between -10dB and +3dB (-3dB is recommended). This allows 12dB of headroom on high-output audiotape operating at a reference level of 370 nanowebers per meter (nw/m), where +4dBm equals 370nw/m of magnetic uxivity. Time code appears similar to a 400Hz square wave with many odd harmonics. Time code is difcult to read at low speeds and during fast wind or rewind. Vertical interval time code was developed for use with 1-inch-tape-width helical scan SMPTE type-C video recorders, which were capable of slow-motion and freezeframe techniques. During these functions, LTC is impossible to read. However, VITC is readable (as long as the video is visible on the screen) because the indexing information for each eld/frame is recorded in the video signal during the vertical blanking interval. When viewed on a monitor that permits viewing of the full video signal, VITC can be seen as a series of small white pulsating squares at the top of the video eld. Normally, VITC is recorded on two nonadjacent vertical blanking interval lines in both elds of each frame. Recording the code four times in each frame provides a redundancy error that lowers the probability of reading errors due to dropouts. Because VITC is recorded as an integrated part of the analog videotape recorders video track, it can be read by the rotating video heads of a helical scan

22

FUNDAMENTALS OF DIGITAL AUDIO

recorder at all times, even in freeze-frame or fast-wind modes. Video technology is explained more fully in Chapter 6. In todays digital video acquisition and editing environment, time code is still a valuable reference. Frequently, audio and video are now being handled digitally on the same computer system, so the need for mechanical synchronization of separate tape machines is lost; however, the importance of internal synchronization is still present. The need to identify each eld of video is addressed by embedding the time code reference during acquisition into the data for each frame, whether it is recorded onto tape or on a data storage device. The time code information is part of the data stream that gets transferred concurrently with the audio and video, intertwining all three elements.

Word Clock
SMPTE time code can only help synchronize devices up to its nite resolution 1/30 th of a secondthough on some equipment it can be used up to 1/100 of a frame. When connecting digital devices that divide time into slices of 1/96,000 th of a second (or smaller), a higher resolution reference is needed to ensure that all the data are being sent and received, and that they are correctly interpreted into the destination device. Word clock differs from time code in that it doesnt stamp each sample point with another tag of data, but rather it is a constant signal that sets the reference sampling rate of the source device in order to avoid data errors and maximize performance in the digital domain. The word clock signal is integrated in the S/PDIF and AES/EBU digital audio signals (discussed in Chapter 10), which allows the destination device to correctly process the digital signal. Some facilities also use a separate, or master word clock device, to dictate the word clock to all studio devices from a single source. If no master word clock is present, then the source device would provide the reference clock to its destination. This ends our discussion of the basic characteristics of sound. An understanding of these concepts will help you with the information yet to come. We have touched only briey on some very important areas, so you should consult the excellent general texts on sound and recording cited in the list of suggested readings at the end of this book.