Libro de Pds Traducido

Pgina 1
captulo 1
Qu es el procesamiento de seales?
Una mirada en el reproductor de CD
1
Pgina 2
S procesamiento ignal es un campo inmenso y diverso. Quizs hay
50,000 ingenieros que consideran el procesamiento de seal como su
especialidad y
cientos de miles ms cuyo trabajo implica el procesamiento de seales. 1 Se
es tambin un campo que no exista hace 50 aos 2 y uno que permanece Mysteri,
o casi desconocido para la mayora de las personas, aunque muchas de sus tareas,
como
conversin analgica, digital, error, codificacin de correccin, sntesis de voz y
compresin de imagen, se han vuelto familiares para los legos de la comunidad
actual,
nicaciones, y computadora, mundo dominado.
Antes de intentar una definicin de procesamiento de seal en general,
permtanos
eche un vistazo a un dispositivo cotidiano, el compacto, reproductor de disco,
cuyo rendimiento,
depende de varios tipos de procesamiento de seal.
Para comprender mejor el reproductor de CD, considere primero la acstica
fongrafo, es decir, uno sin amplificacin electrnica. En el estudio a
el registro se realiza de la siguiente manera: las ondas de presin en el aire
mueven un diafragma, el
cuyo centro est conectado a un lpiz, y el lpiz corta un surco
las ondulaciones registran el sonido. En la reproduccin, el movimiento de la
ranura debajo del
lCraig Marven y Gillian Ewers, un enfoque sencillo para Procesamiento Digital de Seales (Nueva
York:
John Wiley & Sons, 1996), p. 3.
2 El trmino "procesamiento de seal" no exista hace 50 aos, ni tampoco haba
mucho de lo que
hoy se llama procesamiento de seal digital; hubo, sin embargo, una gran
cantidad de ingenieros con ...
cemed con seales analgicas (como por telfono o radio), y muchas de sus
actividades, tales como
el diseo de filtros, hoy se considerara como procesamiento de seales.
Pgina 3
Captulo 1 Qu es Procesamiento de Seales? Una mirada en el reproductor de CD
3
la aguja causa vibraciones en el tono ... arma; las vibraciones viajan a un dia ...
phragm en la base de un habla hom, que produce las ondas de presin
en el aire. Grabacin elctrica, introducida en la dcada de 1920, hizo esta
operacin
ms complicado al convertir la seal acstica en seal elctrica
tanto al hacer la grabacin como al reproducirla. (Ver Figura 1.) El
la ventaja vital era que las seales elctricas, a diferencia de las seales acsticas,
podan
ser fcilmente amplificado
La operacin bsica de un reproductor de CD es casi tan simple. El elctrico
la seal se convierte de una forma de onda que vara continuamente a una
secuencia de
dgitos binarios, Os y Is. Estos se graban en un camino espiral alrededor del
disco. (Ver Figura 2.) Hay hoyos o "hoyuelos" en la pista, y la cada
y los bordes ascendentes de un hoyuelo corresponden a los bordes descendentes y
tren de pulsos. En la reproduccin, un rayo lser rastrea la trayectoria de la
espiral. Dnde estn
la altura cambia, el rayo no se refleja de forma pronunciada, sino que se
atena. Una luz
el sensor detecta las reflexiones, produciendo nuevamente la secuencia de Os e
Is,
EUCTIUCAL
GRABACIN SOtJND
DDKCtlTTER
Figura 1. El dibujo muestra cmo elctrica trabajos de grabacin de sonido. Una
micro-
El telfono convierte las ondas de sonido en una seal elctrica que, despus de
grabar en un
cinta magntica, impulsa electroimanes para cortar una pista en el disco
maestro. Cuando el
se reproduce el registro, el movimiento de la aguja genera una pequea corriente,
que se amplifica
y enviado a los altavoces, que convierten la seal de nuevo en ondas de
sonido. (Reproduccin
fabricado con el permiso del artista.)
Pgina 4
4
/, "I-, r
<:: ~~
Figura 2. El dibujo muestra cmo las obras digitales de grabacin de sonido. El
elctrico
la seal proveniente del micrfono se digitaliza, es decir, se convierte de forma
continua
forma de onda a una secuencia de dgitos binarios por muestreo y
cuantificacin. La secuencia
se graba en un disco compacto como una serie de pozos poco espaciados. En el
reproductor de CD a
La captura lser rastrea la pista en espiral, generando la secuencia original de
dgitos binarios.
El cdigo digital se convierte a una seal analgica que, despus de la
amplificacin,
ves los altavoces. (Reproducido con el permiso del artista)
que luego se convierte nuevamente a una seal de sonido analgica. 3 (la
secuencia de
Ceros y unos es larga: el sonido digital se compone de 44.100 muestras por
segundo ..
y cada muestra est representada por 16 bits, por lo que se puede calcular que
74 minutos de msica que un CD puede contener requerir unos 3,1 mil millones de bits. Actu ..
aliado, la correccin de errores y otros requisitos de codificacin significan que
tres veces
este nmero es obligatorio). 4
Los reproductores de CD proporcionan un mejor sonido que los reproductores de
discos convencionales: en ..
ancho de banda arrugado (un lmite inferior de 20 hercios en lugar de 30 hercios),
ms plano
3Pred Guterl, "disco compacto" (IEEE Spectrum, vol 25 (1988), no 11, pp 102-108...); Senri
Miyaoka, "audio digital es compacto y robusto" (IEEE Spectrum, vol. 21 (1984), no. 7,
pp.
35-39); y John G. Truxal, La edad de los mensajes electrnicos (Cambridge, MA: MIT Press,
1990), p. 245.
Op 4Guterl. cit. Puesto que hay 31 millones de segundos en un ao, si una persona
fuera a leer fuera
la secuencia de ceros y unos, sin parada, una por segundo, tardara 300 aos, ms
que
incluso un adolescente querra gastar con un CD.
Pgina 5
5
frecuencia ... curva de respuesta (ms o menos 0.5 decibeles en lugar de 3 deci ...
Bels), mayor rango dinmico (90 dB en lugar de 70 dB), mejor seal ... a ...
relacin de ruido (90 dB en lugar de 60 dB), distorsin armnica mucho ms
pequea
(0,01 por ciento en lugar de 1 a 2 por ciento), y mucho mejor separacin sean ...
entre los dos canales estreo (90 dB en lugar de 30 dB). Otra ventaja ...
tages son de menor tamao, ms tiempo de juego ... tiempo y mayor durabilidad
(hay
no haber contacto mecnico en la reproduccin para causar el desgaste de los
discos). En anuncio...
Dition, uno tiene acceso inmediato a diferentes pistas, y el juego programado
es posible. Finalmente, los CD son casi inmunes al polvo, los araazos y las
aletas ...
gerprints que alteran el sonido de LPs.5 No es, entonces, sorprendente que en
menos de una dcada despus de su introduccin a principios de la dcada de
1980, reproductores de CD
condujo a los tocadiscos del mercado. 6
Si bien la ventaja vital de la grabacin elctrica era que permita
amplificacin, la ventaja vital de la grabacin de CD es que permiti
muchos tipos de procesamiento de seal, que confieren las ventajas recin
nombradas.
El tubo de electrones hizo posible la amplificacin; microelectrnica hecha
complejo, real ... procesamiento de seal de tiempo posible.
Una de las tareas de procesamiento de seales ms importantes es la con ...
versin de la seal de forma analgica a digital, que se realiza en tal
forma en que la forma de onda original puede recrearse con un alto grado de accu
...
picante. La codificacin digital en s misma confiere una ventaja importante: leve
im ...
perfecciones en el disco o bajos niveles de ruido elctrico en el circuito digital ...
Cuits: lo suficientemente pequeos para no cambiar un 0 en un 1 o viceversa.
ninguna degradacin de la seal en absoluto.
Errores inevitablemente ocurren, pero la seal ... capacidad de procesamiento del
CD
el jugador por lo general puede corregirlos. Primero, hay redundancia en el
digital
codificacin de modo que si se lee mal un nmero pequeo de bits, la secuencia
original es
restaurado automticamente En segundo lugar, los dgitos binarios sucesivos no
se graban en ...
Juntos, pero fsicamente separados. Esto significa que si un defecto en el disco af
...
5 Peter J. Bloom, "Audio digital de alta calidad en la industria del
entretenimiento: una visin general de
logros y desafos"(IEEE Revista ASSP, vol. 2 (1985), no. 4, pp. 2-25), y
Op Miyaoka. cit.
6Steven Lubar, Infoculture: El libro de la era de la informacin Smithsonian Invenciones (Boston:
Houghton Mifflin, 1993), p. 195. A fines de la dcada de 1990 hay algo as como
un culto de LP ...
que cree en la superioridad de los discos de vinilo en CD. La revista Time inform
[14 de julio
1997, p. 20] que el nmero de LP vendidos haba aumentado de 625,000 en 1994
a 1,100,000 en
1996, y que a fines de junio de 1997 ya se haban vendido 2.200.000 ese ao. Y
tornamesas, como parte de los centros de entretenimiento en el hogar, continan
vendindose; Atlantic Monthly re ..
portado [diciembre de 1997, p. 106] que haba ms tornamesas de alto ... para
elegir en
1997 que nunca antes.
Pgina 6
6
ENTREVISTADOR:
Una vez que la gente comenz a hablar sobre "Procesamiento digital de seales"
eso defini retroactivamente el "procesamiento de seales"? Fue el
"procesamiento de seales"
campo coherente antes de mediados de los aos 60?
JAMES KAISER:
Nunca lo pensamos realmente como procesamiento de seales. En el
electrnica, tal vez fue la generacin de seal o la configuracin de la seal o ese
tipo de
cosa. Nunca me recuerdo diciendo: "Lo haremos por procesamiento de
seal". Pero
luego, una vez que te digitalizaste, ahora tambin te das cuenta, "Caramba,
precios de bolsa
son seales digitales. Son seales muestreadas. Y una seal de muestra de un
instrumento musical, una seal de msica. Todos estos tienen algo en comn. "Es
no solo esta pequea cosa de comunicaciones en la que estabas trabajando, o este
radar
cosa. Es muy amplio Las operaciones que ests haciendo son, esencialmente, las
mismas
tipo de cosas en cada caso. Y el desarrollo del procesador de la computadora
el poder es lo que provoc esa realizacin. Fue la herramienta que te permiti,
con
el software, a hacer las cosas por las seales no importa de dnde venan .... 1
ENTREVISTADOR:
Al mirar su libro de 1965, qued muy impresionado por cmo
Tanto usted mismo us computadoras en esos primeros aos, tanto para hacer
clculos
ciones y para simular fenmenos. Debe haber sido muy temprano en esos ap
plicaciones.
JAMES FLANAGAN:
Descubrimos que puede obtener soluciones tiles para casi
cualquier proceso de seal que puedas convertir en ecuaciones de diferencia. Eso
es
forma en que hicimos la mayor parte de la sntesis del habla, con la simulacin
del tracto vocal por
ecuaciones funcionales, luego resolverlas simultneamente. No en tiempo real,
obviamente.
Este temprano trabajo de procesamiento del habla requiri datos muestreados, y
el
posicin de las seales de datos muestreadas. Pero tambin necesitaba todo tipo
de filtrado y
anlisis espectral.
Creo que toda el rea del procesamiento de seal digital, particularmente el filtro
digital
El diseo final fue impulsado por la comunidad de procesamiento de voz. Hice
una marca
aqu. Roger Golden y yo hicimos algo llamado el vocoder de fase en 1966.
Esto requiri la simulacin de filtros elctricos. Tuvimos un impulso infinito
filtros de esponja que se aproximan a las caractersticas de Bessel. No habamos
pensado
acerca de los filtros de respuesta de impulso finita mucho despus, se
desarrollaron
poco despus, pero los usamos con buenos resultados. Todo el asunto de tener
que
hacer filtracin de seales, anlisis espectral y operaciones algortmicas en sam-
datos pled, de reconocer lo que sucede cuando cuadras una seal o tomas un
raz cuadrada, o viendo lo que sucede con el ancho de banda, todo esto llev a la
desarrollo del procesamiento digital de seales en ese momento. Pudo haber
habido un
paralelo en el procesamiento de imgenes que no conozco, pero el procesamiento
de voz
fue una actividad de investigacin que galvaniz las tcnicas de seal digital.
2
1James Kaiser, entrevista de historia oral, 11 de febrero de 1997, p. 36.

2 James Flanagan entrevista de historia oral 8 abril de 1997, pp. 16-17.
Pgina 7
7
afecta a una gran cantidad de bits, estos se extraern de diferentes muestreos
reas de la seal original.
7
Por lo tanto, una partcula de polvo, un rasguo, una huella digital o
incluso un pequeo orificio en el disco no afectar el sonido. Finalmente, ms
grande
errores, que el programa no puede corregir completamente, son detectados y
minimizados
a travs del filtrado digital. 8
El reproductor de CD exhibe un procesamiento de seal sofisticado, y muchos de
los
las tareas no podran haber sido realizadas 25 o incluso 10 aos antes. En esto
monografa revisaremos la evolucin, en los ltimos 50 aos, de la seal pro ...
proceso, describiendo el desarrollo de tcnicas nuevas y mejoradas y
sealando muchas de sus aplicaciones. Revisaremos tambin la evolucin de
la organizacin profesional hoy conocida como IEEE Signal Processing So ...
la sociedad civil, que desempe un papel vital en el desarrollo tecnolgico y en
la
surgimiento de una profesin reconocida de procesamiento de seales. 9 (Vase la
figura 3.)
Una seal puede definirse como una cantidad de tiempo ... variante que transmite
informacin. 1o En el mundo de la seal ... procesamiento, esa cantidad casi
siempre
es, o se convierte en, una cantidad elctrica tal como corriente o voltaje. 11
Y, cada vez ms en las ltimas dcadas, las seales continuamente variables son
con ...
vertido a secuencias de bits, lo que significa que las seales analgicas se
convierten a
seales digitales. Un tpico sistema de procesamiento de seales ... se muestra en
la Figura 4,
donde la fila superior muestra la seal ... tareas de procesamiento y la fila inferior
muestra cmo cambia la seal. 12
Una seal analgica, como podra provenir del cartucho de recogida de un
fongrafo, pasa a travs de un bajo ... filtro de paso, tambin llamado anti ...
aliasing fil ...
ter, que elimina todas las componentes de frecuencia de la seal por encima de
un cer ...
tain frecuencia. El bajo ... filtro de paso se llama anti ... filtro de aliasing, porque
si
no estaba presente la muy ... alta ... los componentes de frecuencia podran
producir, en
el procesamiento posterior de la seal, menor ... frecuencia "alias". El siguiente
Op 7Truxal. cit., p. 250.
Op 8Miyaoka. cit.
9 Como se seala en el prefacio, una historia mucho ms completa de la Sociedad de Procesamiento de Seales es
contenida en la monografa de Frederik Nebeker, La Sociedad de Procesamiento de Seal IEEE:
Cincuenta
Aos de Servicio, 1948-1998 (New Brunswick, Nueva Jersey: IEEE Centro de Historia, 1998).
lOJames A. Cadzow, Fundamentos de Procesamiento Digital de Seales y Anlisis de Datos (Nueva York:
Macmillan, 1987), p. 1.
Esto tambin es cierto en el mundo biolgico de los organismos con sistema
nervioso central: sentido
los rganos convierten seales visuales, auditivas, tctiles, olfativas y de otro tipo
en seales elctricas para con ...
valos al cerebro.
12Este descripcin de un ... sistema de procesamiento de seal tpico y el dibujo
en la Figura 4 vienen
del Captulo 3 (pp. 31-72) de Marven y Ewers, op. cit.
Pgina 8
8
Figura 3. La mayor de las conferencias anuales patrocinados por el IEEE Seal
Processing Society es la Conferencia Internacional sobre Acstica, Discurso y
Sig-
procesamiento final (ICASSP). En el ICASSP de 1997 en Munich, los asistentes
recibieron un CO-
ROM que contiene todo el proceso de la conferencia de 4 das. (Reproducido por
permiso de IEEE.)
paso, que produce la forma de onda de la escalera, se llama muestra ... y ...
la seal de entrada se muestrea en puntos particulares en el tiempo, y el valor
muestreado es
mantenido constante hasta el prximo muestreo. Entonces las alturas se
cuantifican, eso
es decir, un convertidor analgico a digital asigna uno de un conjunto discreto de
valores a
cada una de las alturas. Aqu los valores muestreados se ordenan en clases de
acuerdo.
ing a la amplitud. Cada altura puede representarse como un nmero binario.
Si el nmero contiene, por ejemplo, 6 bits, esto permitira cada altura
a ser codificada como una de 2 6 o 64 niveles diferentes. Toda la seal tiene as
convertido en una secuencia de nmeros binarios. Luego viene un poco de digital
..
Pgina 9
9
LPF
010101
010110
011100
101101
010101
010110
011100
101101
LPF
FIGURA 4. Las dos filas muestran una secuencia tpica de procesamiento de
seales, la parte superior
fila que muestra las tareas de procesamiento de seal y la fila inferior cambia la
seal.
Falta en esta secuencia la tarea particular de procesamiento de seales para la
cual
todo el proceso se lleva a cabo. (Redibujado despus de la Figura 3.55 de Craig
Marven y
Gillian Ewers, un enfoque sencillo para procesamiento de seales
digitales (Nueva York: John
Wiley & Sons, 1996)).
seal ... tarea de procesamiento. Esto podra mejorar algunas caractersticas del
sig ...
nal; o extraer la informacin deseada de una seal ruidosa; o analizando una
forma de onda en frecuencias componentes; o encriptacin de informacin; o
com ...
presionando la informacin para la transmisin y de ... compresin despus de
trans ...
misin. Luego, un proceso inverso recrea una forma de onda analgica: digital ...
a ...
la conversin analgica produce una forma de onda de escalera, que pasa a travs
de un
low ... pass filter para eliminar los componentes de alta frecuencia, suavizando
forma de onda
Cada paso de esta secuencia ha sido sometido a un gran escrutinio y
mejorado por un diseo ingenioso. Por ejemplo, la necesidad de cuantificacin
no ser uniforme (con bandas del mismo tamao ...), pero se puede adaptar al tipo
de
seal que se procesa (con, por ejemplo, bandas estrechamente espaciadas donde
varia ...
la seal es ms significativa). El esquema de cuantificacin puede incluso
cambiar de momento a momento para lidiar mejor con la seal cambiante;
esta tcnica se llama cuantificacin adaptativa.
La conversin de una seal analgica a una secuencia de bits no necesita
implican una prdida significativa de informacin. Si una seal contiene solo
frecuencia ...
cias por debajo de cierto lmite, dicen fmax, a continuacin, la seal original puede
ser reconstruido
exactamente de los valores muestreados, siempre que la frecuencia de muestreo
sea al menos dos veces
f mx.! 3 No es necesariamente cierta prdida de informacin cuando el val muestra
...
13Esto suele llamarse el ndice de Nyquist, en honor al ingeniero de Bell Labs
Harry Nyquist, quien
en la dcada de 1920 estudi la velocidad mxima de sealizacin que podra
utilizarse en un telegrama chan ..
Nel de ancho de banda dado [Sidney Millman, ed., Una historia de Ingeniera y Ciencia de la
Campana
Sistema: Ciencias de la Comunicacin. (1925-1980) (AT & T Bell Telephone Laboratories, 1984),
pp. 9-10]. Ver Harry Nyquist, "Ciertos temas teora de la transmisin de
telgrafo" (Transac ...
ciones de la ALEE, vol. 47 (1928), pgs. 617-644).
Pgina 10
10
se digitalizan, pero esta prdida puede hacerse tan pequea como se desee al
aumentar
la cantidad de bits dados a cada muestra
Hay dos grandes ventajas de tener informacin en forma digital.
En primer lugar, como ya hemos mencionado, las seales digitales con
frecuencia pueden ser recon;
estructurado exactamente a pesar del ruido y el error en la transmisin,
almacenamiento o re;
trieval. En segundo lugar, las seales digitales pueden ser manipuladas por
circuitos electrnicos o
computadoras de una variedad bastante ilimitada de formas. El resto de esto
monografa dejar muy en claro la verdadera importancia mundial de estos
dos ventajas.
El procesamiento de seal no es la transmisin de seales, como a travs de tele;
cables telefnicos o por ondas de radio, pero los cambios realizados en las
seales para im;
probar la transmisin o el uso de las seales. Entre los procesos estudiados y
ideado por seal; los ingenieros de proceso estn filtrando, codificando,
estimando, de;
informando, analizando, reconociendo, sintetizando, registrando y
reproduciendo.
Aunque el procesamiento de la seal se refiere tanto a las tcnicas analgicas
como a las digitales, la
el campo est cada vez ms dominado por las tcnicas digitales. De hecho, como
veremos
en el Captulo 4, la aparicin de las tcnicas digitales en los aos 60 y 70
jug un papel importante en la creacin de una comunidad de ingenieros
preocupados por
procesamiento de la seal. 14 Otra distincin es entre lo real; proceso de seal de
tiempo;
procesamiento de seal de tiempo real y no real; el primero est restringido a
proce;
condiciones que pueden llevarse a cabo tan rpido como llega la seal. La
disciplina de
procesamiento de seales abarca el desarrollo de teoras, la creacin de
algoritmos, la implementacin de algoritmos en hardware y la aplicacin;
de software y hardware para comunicaciones, instrumentacin y
otros tipos de tareas; es decir, abarca teora, tcnicas, difcil;
ware y aplicaciones.
A fines del siglo XX, el procesamiento de la seal es una tecnologa vital;
en muchas reas: comunicaciones, procesamiento de informacin, consumidor
electrnica, sistemas de control, radar y sonar, diagnstico mdico, seismol;
oga, y la instrumentacin cientfica en general. Adems de la amplia
gama de reas de aplicacin, existe una amplia gama de tareas de procesamiento
de seales.
Ejemplos de estas tareas son eliminar el eco de las lneas telefnicas, scram;
bling celular; conversaciones telefnicas, control de la suspensin de un
automvil;
mvil para que responda a las condiciones del camino, lo que permite imgenes
de satlite
sistemas para resolver pequeos objetos en el suelo y crear rganos internos
sobresalir en escaneos CAT. Que tantos asuntos relacionados con el
procesamiento de
14Las seales digitales no son nuevas: los incendios de seales de los testigos,
las comunicaciones de las banderas del buque ... del buque, Morse
cdigo y Braille, pero su uso en comunicaciones, computacin, instrumentacin
y control.
Los sistemas trol han aumentado enormemente en las ltimas dcadas.
Pgina 11
11
seales han surgido en las ltimas dcadas y han asumido tal economa
la importancia es el resultado del hecho -y lo confirma an ms- de que el
nuestro es un
Edad de informacin.
Hoy hacemos uso de una gran variedad de medios electrnicos de comunicacin.
catin y entretenimiento. Las computadoras estn involucradas en la conduccin
de la mayora
las empresas y las computadoras, en forma de microprocesadores, se encuentran
todas
a nuestro alrededor en oficinas, en casas y en autos. En tiempos anteriores, la
mayora de las personas
trabaj en la agricultura o la industria. Hoy una gran parte del trabajo
fuerza son los trabajadores de la informacin, como profesores, investigadores,
diseadores,
banqueros, abogados, agentes de seguros, anunciantes, emisoras y diarios.
ists. 15 En el siguiente captulo, de una manera muy parcial, asignar una
determinada
fecha al comienzo de la era de la informacin.
15 Segn una estimacin, el 55 por ciento de las personas empleadas en los Estados Unidos en 1990
eran trabajadores de la informacin [Karen Wright, "El camino a la aldea
global" (Scientific Ameri ..
puede, vol. 262 (1990), no. 3, pp. 83-94)]. La importancia del procesamiento de la informacin en los indus ..
intentarlo en general est indicado por el hecho de que en los Estados Unidos, a
mediados de la dcada de 1990, unos 40 por ..
El centavo del presupuesto de capital de la industria era para computadoras y
equipos de telecomunicaciones [W.
Wayt Gibbs, "Toma de ordenadores para tareas" (Scientific American, vol. 277 (1997),
no. 1, pp.
82-89)]. Si uno mira especficamente a la informtica, electrnica de consumo y
telecomunicaciones.
industrias de cationes, la imagen sigue siendo impresionante: en 1996 en los
EE.UU. estas industrias em ..
atrajo a casi 4,3 millones de personas y cre el 6,2 por ciento de la produccin
nacional de bienes y
servicios, que era una parte ms grande de la economa que la industria del
automvil, la interpretacin ..
industria cin, o la industria alimentaria [IEEE Spectrum, vol. 35 (1998), no. 1,
p. 16T4].
Pagina 12
Capitulo 2
Certificado de nacimiento
de la era de la informacin:
El annus mirabilis de 1948
Lo que suceda en 1948
pelculas que la gente estaba mirando:
Hamlet con Laurence Olivier
El tesoro de la Sierra de Walter Huston
Madre
TV muestra personas estaban viendo:
"Hopalong Cassidy", el primero de la TV
Westems, comenz en 1948
"The Howdy Doody Show"
"Conoce a la prensa"
gente de la msica estaban escuchando:
canciones de Bing Crosby y Frank Sinatra
libros que la gente estaba leyendo:
Comportamiento sexual de Alfred Kinsey de la
Hombre humano
Norman Mailer Los desnudos y los muertos
Walden Dos de BF Skinner
13
Pgina 13
E n la historia de la ciencia el ao 1666 se conoce como annus de Isaac Newton
mirabilis (ao increble). En ese ao que escribi dos documentos sobre m ...
chanics, "On Violent Motion" y "The lawes of Motion", y un tratado
sobre las nuevas matemticas que haba inventado, a las que llam el mtodo de
fluxiones y que hoy conocemos como clculo. Annus 1 de Albert Einstein
mirabilis era de 1905, cuando public tres artculos clsicos que presentaron,
respectivamente, la hiptesis de la naturaleza cuntica de la luz, la estadstica
la fsica del movimiento browniano y la teora especial de la relatividad. 2
El ao 1948 puede considerarse como el annus mirabilis en la emergencia
de la disciplina del procesamiento de seales. Porque en ese ao Claude Shannon
pub ...
la poca ... haciendo "una teora matemtica de la comunicacin";
Bernard Oliver, John Pierce y Claude Shannon publicaron el ar ..
mtodo para el uso de la modulacin por pulso de cdigo; mtodos digitales
modernos de
la estimacin del espectro se introdujo; error ... los cdigos de correccin fueron
introducidos ...
ducado; la ingeniera de audio alcanz una nueva prominencia; y la seal IEEE
Procesamiento de la sociedad, aunque con un nombre diferente, se
estableci. Adems de..
lRichard Westfall, en su biografa de Newton [Nunca en reposo (Cambridge:
Cambridge Uni,
versidad Press, 1980)], sostiene que uno debe hablar en lugar de tres Mirabiles
anni (1664, 1665,
y 1666) y ese mito ha exagerado cunto logr Newton en este perodo
(aunque "Por cualquier otra norma que un mito newtoniano, la realizacin del anni
Mirabiles fue sorprendente ") [pp. 140-175, cita en la p. 174].
2 Abraham Pais, 'sutil es el Seor ...' -La ciencia y la vida de Albert Einstein (Oxford: Ox'
Ford University Press, 1982), p. 522.
Pgina 14
Captulo 2 Certificado de nacimiento de la era de la informacin: El Annus
Mirabilis 1948
15
En ese ao, se realiz una demostracin del primer programa informtico ...
y el anuncio de la invencin del transistor, que anuncia dos
nuevas tecnologas que luego estimularan en gran medida el procesamiento de la
seal. Ser...
Antes de discutir estos y otros logros con ms detalle, establezcamos el
escena.
Cuando comenz el ao 1948, la Segunda Guerra Mundial era solo dos aos y
medio
aos en el pasado. Hubo disturbios polticos en todo el mundo: el conflicto sera
...
entre los estados recin creados de India y Pakistn sobre cmo el subcon ...
tinente debe ser particionado (y en enero Mahatma Gandhi fue asesinado ...
engendrado por un hind que resenta el papel de Gandhi en la particin); guerra
civil
en China entre los nacionalistas encabezados por Chiang Kai ... shek y el
Comunistas encabezados por Mao Tse ... tung; creacin de dos estados, dividido
por
el 38 paralelo, en Corea; conflicto entre judos y rabes despus de la
La evacuacin britnica de Palestina y la proclamacin del estado de Israel; y
polarizacin de los Estados europeos en un bloque comunista y un bloque
occidental
(con las tensiones aument en abril cuando los soviticos negaron la occidental
Los aliados tienen acceso a Berln Oeste, lo que lleva al suministro de esa ciudad
por aire
por 18 meses). Pero haba razones para ser optimista sobre el futuro. En
los Estados Unidos la economa era fuerte, y la aprobacin del Congreso en
Abril de lo que se conoce como el Plan Marshall prometi acelerar el Euro ...
pea la recuperacin econmica. La gente en todas partes esperaba que la
tecnologa de la guerra ...
Los avances tecnolgicos ahora se dirigen a elevar el nivel de vida.
Haba aviones a reaccin y la promesa de la energa atmica, pero para muchos
la gente la ms impresionante de las nuevas tecnologas fue la electrnica, que
sali de la Segunda Guerra Mundial un campo mucho ms grande de lo que
ingres. Antes de la
guerra, la radio era con mucho el rea ms grande de la electrnica, aunque los
tubos de electrones
se utilizaron tambin durante mucho tiempo ... telefona a distancia y telefona de
operador (usando una
lnea para muchas conversaciones simultneamente), para pelculas de sonido, en
fono ...
grficos y pblicos ... sistemas de direcciones, y en muchos
instrumentos. Durante el
guerra surgieron nuevas aplicaciones de la electrnica, como radar, sonar y
informtica y reas de aplicacin anteriores, como las comunicaciones,
sistemas electrnicos de navegacin, instrumentacin y control expandidos
enormemente. Y, justo despus de la guerra, la electrnica industrial surgi como
campo principal, y la televisin y la radio FM pareca listo para suc comercial ...
impuesto.
El florecimiento de la electrnica afect a las dos grandes soci ...
eties en los Estados Unidos que representaban a los que ahora llamaramos electri
...
cal ingenieros: el Instituto Americano de Ingenieros Elctricos (AlEE) y
el Instituto de Ingenieros de Radio (IRE). El ms grande de los dos, el AlEE,
haba sido fundada en 1884, y la mayora de sus miembros trabajaban en
electricidad
potencia, aplicaciones industriales, comunicaciones por cable o instrumentacin.
Pgina 15
diecisis
Mirabilis 1948
El IRE, fundado en 1912, difera del AlEE no solo en trminos tcnicos.
pero tambin en eso, desde su inicio, pretenda ser una organizacin
transnacional ...
nizacin (de ah que nunca hubo un "Instituto Americano de Radio Engi ...
neers "}. 3
El crecimiento de la electrnica hizo que cada vez fuera ms parte
de los campos tradicionales de la ingeniera elctrica y, por lo tanto, en el mbito
de
el AlEE. De hecho, en 1963, cuando el AlEE y el IRE se fusionaron para formar
el
Instituto de Ingenieros Elctricos y Electrnicos, aproximadamente la mitad de
los miembros
del AlEE estaban preocupados con la electrnica. 4
Para la IRE, el crecimiento de la electrnica significaba no solo un aumento de
tamao
pero tambin un campo tcnico mucho ms amplio. Porque a menudo eran
ingenieros de radio
quien fue pionero en nuevas reas de la electrnica, era natural que la IRE repre
...
envi las nuevas reas. Sin embargo, la mayor diversidad dentro del Instituto
como
todo sugiri a sus lderes que haya grupos establecidos de miembros
en especialidades especiales, y en 1948 el Sistema de Grupo Profesional IRE
fue establecido. Aunque anteriormente haba habido comits tcnicos de IRE
para diferentes especialidades, esto era nuevo ya que cada grupo era una
membresa
organizacin (abierta a miembros de IRE) que podra publicar sus propias
transacciones
y organizar sus propias conferencias. El sistema del Grupo IRE, adoptado por
IEEE en el momento de la fusin en 1963, evolucion en el sistema actual de
Sociedades Tcnicas IEEE.
5
El primero de estos grupos tcnicos en ser configurado fue el Profesional
Group on Audio, cuya solicitud de formacin fue aprobada por la IRE Ex ...
Comit EJECUTIVA el 2 de junio de 1948. 6 Oliver Angevine, Leo Beranek, y
Benjamin Bauer estuvo entre los que trabajaron para establecer la nueva organi ...
zacin. El Grupo organiz sesiones en conferencias IRE y pronto comenz
publicando sus propias transacciones, as como un boletn de noticias.
Los intereses del Grupo incluyen fongrafo y grabacin magntica,
amplificadores, altavoces, tcnicas de medicin de audio, electroacstica,
y comunicaciones de voz. Estos involucraron lo que se puede llamar anlogo
procesamiento de seal y se bas en tecnologas desarrolladas durante la Guerra
Mundial
II. Por ejemplo, los investigadores en el Electro ... Acstico Laboratorio y el Psy
...
cho ... El Laboratorio Acstico de la Universidad de Harvard avanz mucho en el
anlisis ...
3 Juan Ryder y Donald G. Fink, Ingenieros y electrones: Un siglo de P elctrico; ogress
(Nueva York: IEEE Press, 1984), pp 209-215..
4Ivan S. Coggeshall, "la dotacin de IEEE en la electrnica de alee" (Ingeniera Elctrica,
vol. 82 (1963), pp. 2-20).
SRyder y Fink op. cit., pp. 216-217.
6Institute of Radio Engineers, Comit Ejecutivo, minutos 2 de junio de 1948.
Pgina 16
Mirabilis 1948
17
sis y procesamiento de seales en el rango de audio. Filtros analgicos para
determinar ...
se disearon los espectros del habla, la msica o el ruido. El desempeo
de amplificadores y de equipos de comunicacin y grabacin fue im ...
demostrado, como en el "recorte" y la remodelacin de las seales de voz para
aumentar la
seal ... a ... relaciones de ruido en las comunicaciones en entornos de alto ruido
... tales
como avin militar. Otras seales ... tcnicas de procesamiento salieron de la
extensin ...
un tiempo de guerra en el sonido submarino, como filtros y formas espectrales ...
circuitos para enfatizar los ruidos de los submarinos. 7
As que, aunque el procesamiento de seales no exista como una disciplina
en 1948, el
IRE Professional Group on Audio se ocup de un trabajo que, sobre la
las prximas dos dcadas contribuyeron al surgimiento del nuevo campo. Ah
fueron, sin embargo, otras lneas de trabajo que contribuyeron, y al mirar atrs
1948 debemos tener en cuenta estas otras lneas de trabajo y podemos considerarlos como
parte del patrimonio del procesamiento de seales.
El ao 1948 se destaca en una historia de la tecnologa, ya que vio la
publicacin de "Una teora matemtica de la comunicacin ... de Claude Shannon
...
cin ".8 Robert Lucky ha escrito:" No conozco ningn trabajo de genio mayor en
los anales del pensamiento tecnolgico ". Irving Reed ha comentado:" Por
este documento histrico y sus varios documentos posteriores sobre la
informacin ...
ory ha alterado ms profundamente todos los aspectos de la teora de la
comunicacin y
practice."lO
Shannon, a researcher at Bell Telephone Laboratories, analyzed com...
munication as the transmission of a message from a source through a chan...
nel to a receiver. (See Figure 1.) He quantified this process, measuring the
information source in bits per second, and found limits to the channel ca...
pacity. Perhaps the most celebrated result contained in the original paper is
the proof that the capacity of a band...limited channel, in bits per second,
~h!V}6 --+~ f2ir~
pe p06+ e cl.?
7Leo Beranek personal communication 2 January 1998.
!fLl')
8Claude Shannon, "A mathematical theory of communication" (Bell System TechnicalJour-
l9I
nal, vol. 27 (1948), pp. 379-423, 623-656). Several months later Shannon published an..
other paper, which elaborated and extended the first paper and which adopted an
engineer.. @~ ,
ing rather than a strictly mathematical viewpoint: "Communication in the
presence of
~
noise" (Proceedings of the IRE, vol. 37 (1949), pp. 10-21). This paper is reprinted in Proceed...
ings of the IEEE, vol. 86 (1998) along with an introductory paper written by Aaron D.
Wyner and Shlomo Shamai (Shitz): "Introduction to 'Communication in the
presence of
ruido' por CE Shannon" (Actas de la IEEE, vol. 86 (1998), pp. 442-446).
9Robert W. Lucky, Sueos de Silicio: Informacin, el hombre y la mquina (Nueva York: St. Martin
Press, 1989), p. 37.
lOQuoted en Neil lA Sloane y Aaron D. Wyner, eds,. Claude Elwood Shannon: Collected
Papeles (Nueva York: IEEE Press, 1993), p. xiii.
Pgina 17
18
Mirabilis 1948
Fuente de informacin
Transmisor
Receptor
Destino
Mensaje
I --- Re-ce-IV-ed ..... DM-ES-Sa-ge .... D
Seal
ruido
Fuente
Figura 1. Esta es la figura 1 desde "A teora matemtica de la comunicacin", show-
La concepcin de ING Shannon de un sistema de comunicacin en
general. (Redibujado despus Fig-
Ure 1 de Claude Shannon, "Una teora matemtica de la comunicacin" (Bell
System
Technical Journal, vol. 27 (1948), pp. 379-423, 623-656)).
es igual a la anchura de banda del logaritmo (base 2) del signal..to..noise
relacin ms 1. 11 Shannon, por supuesto, no estaba solo en la creacin de una
disciplina de
teora de la informacin: que estaba muy influido por un papel de 1928 por Ralph
VL
Hartley, comenzando donde Hartley fue apagado, y de Norbert Wiener, que em ..
phasized la naturaleza estadstica de la comunicacin. 12 Un nmero de otros pio ..
nieros de la nueva disciplina, incluyendo Dennis Gabor y Philip M. Wood ..
Ward, hizo contribuciones vitales. 13
Otra publicacin de referencia de 1948 fue "La filosofa de la PCM",
escrito por Bernard M. Oliver, John R. Pierce, y Claude Shannon. 14 Pulso ..
cdigo de modulacin (PCM) es un medio de transmisin de informacin en la
forma
on..or..off de pulsos. En telefona y radio, las seales eran tradicionalmente trans
..
Mitted travs de la modulacin de amplitud (AM), aunque por 1,948 frecuencia
modulacin (FM) haba entrado en uso generalizado tambin. con tradicional
AM o FM, la seal es analgica (variacin continua), mientras que con PCM la
la seal es digital (representado en unidades discretas). REM se haba inventado
Paul M. Rainey en 1926 e independientemente por Alec H. Reeves en 1937, pero
llSidney Millman, ed, A. Historia de la Ingeniera y Ciencia en el Bell System: Comunicacin
Sciences (1925-1980) (AT & T Bell Telephone Laboratories, 1984), p. xiv.
12Wiener de extrapolacin, interpolacin y suavizado de Estacionaria de series de tiempo (1949), re-
nerales para a continuacin, se emiti originalmente como un informe clasificado
de 1942.
13Lucky op. cit., p. 38, y William S. Burdic, Acoustic Signal Analysis Submarino (Engle
madera Cliffs, Nueva Jersey: Prentice-Hall, 1984), p. 12.
14Bernard M. Oliver, John R. Pierce, y Claude Shannon, "La filosofa de la
PCM" (Pro-
ceedings del Instituto de Ingenieros de Radio, vol. 36 (1948), pp. 1324-1332).
Pgina 18
Mirabilis 1948
19
poco notado. 15 Oliver, Pierce, y Shannon fueron tan impresionado por su advan,
tajas, explican con ms detalle en el captulo 4, que "espera PCM para barrer
el campo de las comunicaciones". Esto no sucedi, aunque su arti,
CLE era extremadamente. influyente. Durante varias dcadas, recibiendo
refuerzos de
el desarrollo de tecnologas integradas, circuitos, la proliferacin de com,
ordenadores, y el uso de fibra ptica para comunicaciones, PCM, efectivamente,
ser,
llegado generalizada. diecisis
Una tendencia dominante en muchas reas de la tecnologa en la segunda mitad
del
el siglo 20 ha sido la sustitucin de los mtodos analgicos por digitales
mtodos. Adems del artculo de Oliver, Pierce, Shannon, haba otros signos
de esta tendencia en 1948. Debido a la mayora de los sistemas de radar y sonar
vas
trabajadas, tanto radiante y recibir pulsos de radiacin electromagntica,
muchos ingenieros durante la Segunda Guerra Mundial tratan datos discretos. El
trabajo sobre
control de incendios y otros sistemas de radar, por tanto, generan una gran
cantidad de literatura sobre
muestreada, sistemas de control de datos. 17 Un problema enfrentado por los
ingenieros de radar y
ms tarde por los ingenieros en muchos otros campos, estimaba una
especificacin continua,
Trum, tales como el espectro de potencia de una seal de radar, sobre la base de
la muestra
datos. En 1948 Maurice Bartlett en Inglaterra, y al ao siguiente John
Tukey en los Estados Unidos, comenz a desarrollar los mtodos digitales de
especificacin,
estimacin trom que han permanecido en uso desde entonces. 18
Una de las ventajas de las seales digitales se puso de manifiesto en 1948 por
Richard
El invento de W. Hamming de error, cdigos correctores. Hamming propuesta es,
codificacin de un mensaje en bloques de dgitos binarios, con cada cer bloque
de satisfacer,
ecuaciones algebraicas Tain, de modo que si ocurre un error en la transmisin (a
0
convertido en un viceversa 1 o) el destinatario no slo poda detectar esto,
15Millman op. cit., pp. 403-404.
16John R. Pierce y A. Michael Noll, Seales: La Ciencia de Telecomunicaciones (Nueva York:
Scientific American Library, 1990), pp. 78-79; the quotation is from p. 79.
17Kaiser reports Dames F. Kaiser, "Digital filters" (Chapter 7 of System Analysis by Digital
Computer, Franklin F. Kuo and James F. Kaiser, eds. (New York: John Wiley, 1966), pp.
218-285)] that this work yielded a well..developed theory of z..transforms. Ben Gold [The In..
stitute, June 1997, p. 13] has said that "the major 'spark' that generated my interest [in digital
signal processing] was a chapter in 'Theory of Servomechanisms' (a volume in
the Radiation
Laboratory Series) on sampled..data systems."
18Millman op. cit., p. 75. Two standard methods of measuring the average power
spectral den..
sity are the frequency smoothing method (FSM) and the time averaging method
(TAM).
FSM was introduced by Einstein in 1914, rediscovered by Norbert Wiener
in 1930, and redis..
covered by Percy John Daniell in 1946. TAM was proposed by Bartlett in 1948 and
rediscov..
ered by PD Welch in 1967. [William A. Gardner, "History and equivalence of two
methods
of spectral analysis" (IEEE Signal Processing Magazine, vol. 13 (1996), no. 4, pp. 20-23).]
Pgina 19
20
Chapter 2 Birth Certificate of the Information Age: The Annus Mirabilis 1948
To get to [Bell Labs in] NYC I would go a bit early to the Murray Hill, NJ
location where I worked and get a ride on the company mail delivery car.
Well, riding through north Jersey in the early morning is not a great sight, so
I was ... reviewing the rectangular codes. Suddenly, and I can give no reason
for it, I realized if I took a triangle and put the parity checks along the
diagonal ... then I would have a more favorable redundancy.... A few miles of
thought on the matter [and] I realized a cube of information bits [would be bet-
ter] .... It was soon obvious (say five miles) a 2 x 2 x 2 x ... x 2 cube, with n+1
parity checks would be the best ....
- Richard W. Hamming 1
1Richard W. Hamming, The Art of Doing Science and Engineering: Learning to
Learn
(Amsterdam: Gordon and Breach, 1997), pp. 139-140. Harold S. Black first
conceived
the negative feedback amplifier [Harold S. Black, "Inventing the negative
feedback am-
plifier" (Spectrum, vol. 14, no. 12 (December 1977), pp. 54-60)] while taking the
ferry
from New Jersey to New York City, and Bishnu Atal's ideas for linear predictive
coding
(LPC) came first while he was traveling by train from Bell Labs in Murray Hill,
New Jer-
sey to the Polytechnic Institute of Brooklyn [Atal personal communication 2
February
1998]. On the basis of these three examples, one might theorize that travel from
New
Jersey to New York stimulates great thoughts. More prosaically, one might point
out
that the three inventors were Bell Labs researchers and that Bell Labs facilities
were lo-
cated on both sides of the New Jersey-New York boundary.
but also correct the error. (See Figure 2). Hamming showed how to achieve
this efficiently and, working with Bernard D. Holbrook, even how to design
equipment to do the checking and correcting automatically, which, in the
pre...computer era, was not an easy task. I9
In June 1948 Bell Telephone Laboratories announced the invention
of the transistor, a solid...state amplifying device. (See Figure 3.) Even as a
discrete component, its small size and low power requirements would in the
1950s and 1960s expand the realm of signal processing, and this effect was
magnified manyfold when it later became possible to incorporate huge
numbers of transistors on a single chip of silicon. Later chapters, especially
Chapter 7, will make these effects clear.
In 1948 there was a milestone in computer history also: on 21 June
1948 a small prototype computer, built at Manchester University, became
the first operational stored...program electronic computer. 21 Full...size stored...
19Millman op. cit., p. 53.
Pgina 20
21
segundo
UN
do
UN
Three overlapping circles with the
regions a through g defined
UN
Insertion of the digits of the signal
UN
segundo
do
Determination of transmitted signal
Received signal including.error in e
FIGURE 2. These figures illustrate Hamming's error-correcting code. Para cada
block of four binary digits d, e, f, g, send seven digits a, b, c, d, e, f, g, where a, b,
and c are chosen so that the number of 1s in each of the three overlapping circles
A,
B, C is even. Then if an error occurs in anyone of the digits a through g, it can be
detected and corrected. Suppose, for example, that the original message is 1 0 1
1.
Then, a would be 1 and band c would be 0, so you would send 1 0 0 1 0 1 1. Sup-
pose the fifth digit is received incorrectly. The receiver can see this immediately,
since the two lower circles have an odd number of 1s (hence the common
element in
those two circles must be the erroneous).20 (Redrawn after figures on pp. 18 and
19
of John G. Truxal, The Age of Electronic Messages (Cambridge, MA: MIT Press,
1990).)
20) ohn G. Truxal, The Age of Electronic Messages (Cambridge, MA: MIT Press, 1990), pp.
16-19.
Pgina 21
22
FIGURE 3. William Shockley (left), Walter Brattain (standing), and John
Bardeen
received the Nobel Prize in 1956 for their invention of the transistor. (Bell Labs
photo
reproduced by permission.)
program computers were under construction in many places-the EDSAC
by Maurice Wilkes at Cambridge University, the BINAC by Presper Eckert
and John Mauchly at the University of Pennsylvania, John von Neumann's
computer at the Institute for Advanced Study in Princeton, and others-
but, as with the transistor, several years would pass before computers at..
tained commercial success. Computers, too, would over the next decades
have enormous influence on the development of signal processing.
21Martin Campbell...Kelly and William Aspray, Computer: A History of the Information Ma..
chine (New York: Basic Books, 1996), p. 100. The ENIAC, the first large general...purpose
electronic digital computer, was completed in 1946, but it was programmed with
switches
and patch...cords. Later it was modified to operate as a stored..program computer.
[Michael R.
Williams, A History of Computing Technology (Englewood Cliffs, NJ: Prentice...Hall, 1985),
pp.275-287.]
Pgina 22
23
ENDERS ROBINSON:
I came back to MIT in the fall of 1950, and I was in the
Mathematics Department as a graduate research assistant. I was working
under Norbert Wiener to find applications for his time-series analysis. As you
know, Norbert Wiener was an eminent mathematician at MIT, and he worked
in generalized harmonic analysis back in 1930. He always felt that people did-
n't realize that was probably his most important contribution. But during World
War II, he worked on prediction theories for anti-aircraft fire control. He devel-
oped a theory-elassified at the time-but published by MIT Press in 1949.
We used the book when I was an undergraduate student there, so I was famil-
iar with it. By 1950, we were ready to apply it. Meanwhile, Wiener had written
another book called Cybernetics which was first published about 1948. That
book was an instant success; he was the one that introduced the word cyber-
netics.
So Wiener was a celebrity by 1950, and my research assistantship de-
pended on finding applications of his work. Another MIT mathematics profes-
sor, George Wadsworth, my graduate advisor, was working in weather predic-
tion which he originally started during World War II; Wiener's classified book
had come out, but it was too difficult mathematically to be applied as such. Asi
que
Wadsworth asked Norman Levinson, who was another eminent mathemati-
cian at MIT, to take Wiener's book and to simplify it into numerical algorithms
that he, Wadsworth, could use. Levinson published these papers in the Jour-
nal of Mathematics and Physics in 1947. They were added as appendices to
Wiener's 1949 book, so Levinson's algorithms with Wiener's theory became
the way to do this type of thing. We could then apply it to geophysics because
seismic records are essentially noisy records. 1
1Enders Robinson oral-history interview 6 March 1997, pp. 2-3.
The potential impact of computers on society was made known to a
wide audience by a book published in 1948, Norbert Wiener's Cybernetics,
which compared biological systems with electrical communications.. and
control..systems. 22 It contained some discussion of signal processing, but a
libro de Wiener public el siguiente ao, extrapolacin, interpolacin y
Suavizado de Estacionaria de series temporales, demostrado de mucho ms valor para la
seal
procesamiento, mientras que, como es lgico, de recibir la notificacin mucho
menos popular. 23
Todava no hemos mencionado una de las mayores historias del ao: en
Septiembre Peter Goldmark, jefe de CBS Laboratories, present un documento
de
22Norbert Wiener, Ciberntica o el control y comunicacin en animales y la
Mquina
(Cambridge, MA: MIT Press, 1948).
23Norbert Wiener, extrapolacin, interpolacin y suavizado de Estacionaria serie de tiempo con

~
Aplicaciones de Ingeniera (Cambridge, MA: MIT Press, 1949).
~
Pgina 23
24
Mirabilis 1948
el Instituto de Ingenieros de Radio en "The Columbia Largo - Reproduccin Mi--
Sistema de grabacin crogroove" .24 En realidad, no fue esta charla, pero una
prensa
conferencias y lanzamiento comercial de la nueva larga - duracin (LP) Registro
tres meses antes de que era la gran story.25 (Ver Figura 4.)
Larga - registros de juego se han desarrollado en la dcada de 1920 por Western
Elctrica con el fin de proporcionar un sonido de pelculas, y en la dcada de
1930 los estudios de
las industrias de radio y msica usado 33 mquinas rpm de transcripcin. Oro--
1/3 ..
marca y Columbia hicieron importantes innovaciones tecnolgicas, incluyendo

una
nueva tcnica de corte y un nuevo pico de peso ligero - brazo hacia arriba, y se
enfrent a la
desafo formidable negocio de persuadir a las compaas discogrficas y cus--
tomers para aceptar el nuevo format.26The mucho ms tiempo de juego y el
notablemente mejor sonido les ayud succeed.27
RCA respondi a la larga Columbia - que juega el expediente mediante la
introduccin de
el 45 - rpm microsurco rcord en 1949. A pesar de 45s a cabo tan slo cuatro
minutos
de la msica en un lado, que eran mucho ms pequeos (por lo que podra caber
en un estante),
se podra cambiar rpidamente, y podran venderse a un precio mucho ms
bajo. Un expediente
formato que haban sido la norma durante veinte aos por lo tanto se enfrent a
dos rivales
en "la batalla de los plazos de envo" .28 La fabricacin de 78s disminuy, pero
ambos
las nuevas velocidades tuvieron xito, convirtindose en el LP ms popular y el
45 main--
TaiNing grandes ventas, sobre todo al "mercado de la juventud", por
decades.29The
24Peter Goldmark, "El sistema de grabacin de Microgroove Long..Playing
Columbia" (Continuar ..
Ings del IRE, vol. 37 (1949), pp. 923-927).
25William E. Butterworth, Hi..Fi: A partir del fongrafo de Edison a cuadrafnico Sound (Nueva
York: Four Winds Press, 1977), pp 156-166..
26en el desarrollo de su brazo pick..up ligera, Columbia se bas en la obra de
Frederick Hunt en una
el brazo de peso ligero a finales de 1930. Una relacin de trabajo de Hunt est
contenida en el Beranek
entrevista oral..history [parte que comienza con el Laboratorio Cruft].
27Steven Lubar, Infoculture: El libro de la era de la informacin Smithsonian Invenciones (Boston:
Houghton Mifflin, 1993), p. 184, y Millard, p. 205.
28Cuando Emile Berliner introdujo el gramfono en la dcada de 1890, los
registros se volvieron a 70
revoluciones por minuto. En la dcada de 1920 el uso de motores sncronos para
impulsar las placas giratorias
causado un cambio a 78 rpm: el motor, impulsado por corriente alterna 60..Hertz,
se volvi a
3600 rpm, se fij a un 46: 1 de reduccin de engranajes (dando 78,26
rpm). [Butterworth op. cit., p.
30.]
29Andre Millard, Amrica de registro: Una Historia del sonido grabado (Cambridge, Reino
Unido: Cam ..
Puente University Press, 1995), pp. 206-208. Todava en 1954, sin embargo,
todava vendi ms que el 78s
otros formatos: en ese ao la industria discogrfica de Estados Unidos envan 121
millones de discos de 78 $ US 24 milsimas de pulgada ..
len, por valor de 76 millones de 45s $ 21 millones, y 24 millones de discos por
valor de $ 22 millones [de Editores
Electrnica, una era de innovacin: El mundo de la electrnica 1930-2000 (Nueva York: McGraw ..
Hill, 1981), p. 98].
Pgina 24
Mirabilis 1948
25
Figura 4. En esta fotografa Peter Goldmark demuestra con gravedad
compresin de datos
sin: la pequea pila de discos en el centro contiene la misma msica que los dos
grandes
pilas de registros 7a-rpm. (CBS foto reproducida con permiso.)
nuevos formatos de registro estimulado el inters en la msica y volvieron a
despertar el deseo
tener tan completa una experiencia de msica en el pas como en presentaciones
en vivo.
La dcada de 1940 fue testigo de finales de otro nuevo medio de
grabacin. inventado en
1898 por el ingeniero dans Valdemar Poulsen, grabacin magntica hizo
no lograr mucho xito comercial hasta que la empresa alemana AEG
(Allgemeine Elektricitats..Gesellschaft) sustituy al alambre de acero o cinta de
acero
que haba sido utilizado como el medio de grabacin con cinta de plstico
recubierto en su
Pgina 25
26
Magnetophone, first marketed in 1935.30This type of magnetic recording
came to the United States after the war, and in 1948 the major recording
studios in the United States adopted magnetic recording. A vital advantage
was that it permitted easy editing.31Similarly, and at about the same time,
the film industry began using magnetic sound..recording for the original
takes and for editing,32 and, as part of the "hi..fi movement", tape recorders
began to be sold in significant numbers to individuals.33
In 1934 radio and phonograph salesmen had begun using the term
'high fidelity'.34 Though not defined precisely, it meant greater faithfulness
in the sound coming from the radio or phonograph to the original sound.35
In the 1930s, however, it was little more than an advertising slogan, as
radio and phonograph engineers worked mainly to lower the price of equip..
ment rather than to improve its performance.36There was also the belief
among manufacturers that most people preferred the mellow sound and low
fidelity of existing sets to the sharp sound of higher fidelity.37
Many music enthusiasts, however, did not accept the quality of sound
available from the radios and phonographs they could purchase and began
experimenting with putting together their own sound systems. Some of
30AEG engineers made other vital improvements, notably a shifting of the audio
signal to a
much higher frequency, where the magnetic material displays a more nearly
linear response.
This technique, called electrical bias, had been patented in the United States in
the 1920s,
but it was not then put into practice, nor was it known to the German engineers.
[Millard
op. cit., pp. 195-197.]
310liver Read and Walter L. Welch, From Tin Foil to Stereo: Evolution of the Phonograph (In,
dianapolis, IN: Howard W. Sams, 1959), p. 350.
32Max C. Batsel and Glenn L. Dimmick, "Film recording and
reproduction" (Proceedings of
the IRE, vol. 50 (1962), pp. 745-751).
33Ampex, which came to dominate the US market in the 1950s, began selling
tape
recorders in 1948 [US Consumer Electronics Industry Today (Arlington, VA: Consumer
Electronics Manufacturers Association, 1997), p. 83].
34Both the Oxford English Dictionary and the Merriam,Webster Collegiate
Dictionary give
1934 as the date for the earliest use of 'high fidelity'.
35Radio engineers had given clear quantitative meanings to terms such as
'sensitivity', 'selec,
tivity', and 'fidelity', but not to 'high fidelity'. See Frederik Nebeker, "The
development in
the 1920s of test,and,measurement techniques for the design of radio
receivers" (Wescon/92
Conference Record, IEEE and ERA, Anaheim CA; 17-19 November 1992, pp. 802-807).
36Frederik Nebeker, "Harold Alden Wheeler: a lifetime of applied
electronics" (Proceedings
of the IEEE, vol. 80 (1992), pp. 1223-1236).
37Roland Gelatt, The Fabulous Phonograph: From Tin Foil to High Fidelity (Philadelphia: JP
Lippincott, 1954), p. 297.
Pgina 26
27
them were amateur radio operators, and some of them had worked with
communications, radar, or other electronics for the military in the war.
They found that by judiciously combining amplifiers, pickups, turntables,
receivers, speakers, and other components purchased from radio supply
houses, whose supplies were augmented by surplus military equipment, they
could achieve a much better sound. It was in 1947 and 1948 that this hi,fi
movement began. 38 What had been a hobby of a relatively small number of
people at the end of World War II soon produced a multimillion,dollar
component manufacturing business, with annual sales in the us reaching
$140 million in 1954. 39 In the next chapter we see how audio engineering
both responded to and stimulated the hi,fi movement.
38Gelatt op. cit., pp. 297-298, and Read and Welch op. cit., pp. 343-352.
39Gelatt op. cit., p. 298, Millard op. cit., pp. 208-211, and Read and Welch op.
cit., pp.
343-352.
Pgina 27
Chapter 3
Halcyon Days for Audio
Engineering:
The 1950s
WHAT WAS HAPPENING IN THE 1950s
movies people were watching:
The African Queen with Humphrey Bogart
and Katherine Hepburn
Ben..Hur with Charleton Heston
The Bridge on the River K wai
Rebel Without a Cause with James Dean
TV shows people were watching:
"I Love Lucy"
"Dragnet"
"Leave It to Beaver"
Westerns: "Gunsmoke", "Have Gun, Will
Travel", "Wagon Train", and many more
music people were listening to:
Leonard Bernstein's "West Side Story"
songs of Elvis Presley
books people were reading:
Norman Vincent Peale's The Power of
Positive Thinking
Jack Kerouac's On the Road
JD Salinger's The Catcher in the Rye
29
Pgina 28
P opular enthusiasm for higher quality sound made the 1950s an excit...
ing time to be an audio engineer. Those concerned with disk record...
ing made improvements associated with the new LPs and 45s, such as
better recording, mastering, and processing techniques and a new lower...
force pickup. Though accurate measurement of the features and perfor...
mance of phonograph systems began in the 1920s, it was not until the mid
1950s that the industry adopted standardized tests of recording characteris...
tics, making it easier for engineers to communicate and for listeners to
judge the relative merits of different phonograph sets or components.!
(Standardization of tests of radio...receiver performance had been achieved
by the IRE in about 1930.)2
Experiments in the use of two or more microphones and two or more
speakers to recreate "auditory perspective", the spatial distribution of sound,
go back to a demonstration of 2...channel sound transmission at the Paris
Electrical Exposition in 1884. 3 The 1930s saw the development of several
systems for stereophonic phonograph recording, notably by Alan Blumlein
lWilliam S. Bachman, Benjamin B. Bauer, and Peter C. Goldmark, "Disk
Recording and Re..
production" (Proceedings of the IRE, vol. 50 (1962), pp. 738-744).
2Frederik Nebeker, "The development in the 1920s of test..and..measurement
techniques for
the design of radio receivers" (Wescon/92 Conference Record, IEEE and ERA, Anaheim CA;
17-19 November 1992, pp. 802-807).
3Morton D. Fagen, ed., A History of Engineering and Science in the Bell System: The Early Years
(1875-1925) (New York: Bell Telephone Laboratories, 1975), p. 68.
Pgina 29
Chapter 3 Halcyon Days for Audio Engineering: The 1950s
31
FIGURE 1. Three decades before car CD-players, the Goldmark car phonograph
was available in all 1956 cars manufactured by the Chrysler Corporation.
(Chrysler
Corporation image reproduced by permission.)
in England and Arthur Keller in the US, but stereophonic recording did
not then become a commercial product. 4 When stereo records were finally
offered to the public in 1957, they became popular immediately.5
Two other areas of work by phonograph engineers in the 1950s were
dictating machines and car phonographs. Phonograph recording for office
dictation was a successful, and continually improved, technology from the
late 19th century-Edison saw it as one of the most important uses of his
invention-through the 1950s. Peter Goldmark and CBS Laboratories de...
veloped an automobile phonograph system in 1955 that was available in all
1956 Chrysler Corporation cars (see Figure 1).6 There was, however, little
4Stereo did come to movie theaters earlier: first in Paris in 1932 (for a sound
version of Abel
Gance's 1927 classic "Napoleon Bonaparte") and later to great popular acclaim
with Walt
Disney's "Fantasia" in 1941 [Ed Lyon, "Stereophonic sound-part 2" (Radio Age, vol. 20
(1995), no. 7, pp. 1-7)].
sMark Kahrs, "Professional and consumer gear: hardware & software" (IEEE Signal
Processing
Magazine, vol. 14 (1997), no. 5, pp. 51-57).
6Bachman, Bauer, and Goldmark op. cit., and William E. Butterworth, Hi...Fi: From
Edison's
Phonograph to Quadraphonic Sound (New York: Four Winds Press, 1977), p. 166.
Pgina 30
32
Chapter 3 Halcyon Days for Audio Engineering: The 19505
LEO BERANEK:
We found that the old methods of calibrating microphones that
the Bell Labs were using were in need of change and new methods were de-
veloped. We were running the only real scientific airborne laboratory, in the
world, I'd say it turned out. The rest of the world had shut these things down,
were all doing underwater sound. That included Japan and Germany as well
as England and France-France was probably taken over by the Germans by
then.
This then led to another point. Remember, we were dealing now with
things that really belong in the radio engineer's field. We were dealing with mi-
crophones and earphones and communication.
... My paper became the basis for the American National Standards that
were written later on. My other major publication as a result of this war work
[still available from the Acoustical Society of America] was my book Acoustic
Measurements. 1
ALFRED FEITWEIS:
We designed filters, but now in a somewhat higher fre-
quency range, with the modern method. That's the task Belevitch gave me,
and I had to fight with these numerical intricacies. It was very difficult, of
course. It took me two months, from morning to night, calculating on an
electromechanical desk-calculator to design the filter. Then I got assigned
some young girl who had only attended grade school, and she then did the
work for me. I had to divide up all the work into small steps which she could
understand. She did not know what a zero of a polynomial equation would be,
or anything like this. So I did something like what you do for programming a
computer at that early stage. 2
JAMES FLANAGAN:
[The MIT Acoustics Lab in the 1950s] was multidisciplinary
with people from physics, mathematics, engineering, psychology and archi-
tecture. All these folks worked together in a very effective way. The primary
thrust of the laboratory was physical acoustics and communications
acoustics. 3
1 Leo Beranek oral-history interview 22 November 1996, pp. 39-40.
2Alfred Fettweis oral-history interview 24 April 1997, p. 29.
3 James Flanagan oral-history interview 8 April 1997, p. 7.
Pgina 31
33
public interest, which Goldmark attributed to apathetic marketing by Co...
lumbia and Chrysler. 7 For this application, as with office dictation, a differ...
ent technology-magnetic recording-came to the fore in the 1950s.
Magnetic recording, too, was part of the hi...fi movement. Ampex
began to sell tape recorders in about 1948-these were, of course, reel...to...
reel recorders-and went on to dominate the us market. In 1954, when
the difficulties of mass...producing them had been solved, pre...recorded tapes
began to be sold, though they did not approach LPs in sales. 8 The 1950s was
a period of intense work on understanding the physical processes underlying
magnetic recording. 9
Microphone design was another important area of audio engineering.
Employing a wide range of transducers-earbon, condenser, piezoelectric,
moving conductor, moving armature, and many others-engineers im...
proved the performance of microphones and designed them for a great
many specialized applications. In the 1950s two types that received much
attention were directional microphones and wireless microphones. 1o
Continuing work on loudspeakers included efforts to achieve
smoother frequency...response characteristics, more uniform directional pat...
terns, lower nonlinear distortion, and superior response to transients. 11 In
the 1950s and 1960s the design of loudspeaker enclosures was put on a firm
theoretical foundation; Leo Beranek's Acoustics, published in 1954, was a
landmark. 12 For public...address systems, there were new techniques to
achieve desired directional patterns from individual loudspeakers and from
loudspeaker arrays. 13
There were other contributors to the hi...fi movement. One factor was
the great interest in classical music, a musical taste often associated with
7Peter Goldmark with Lee Edson, Maverick Inventor: My Turbulent Years at CBS (New York:
Saturday Review Press, 1973), pp. 148-155.
8Roland Gelatt, The Fabulous Phonograph: From Tin Foil to High Fidelity (Philadelphia: JP
Lippincott, 1954), p. 298.
9Marvin Camras, "Current problems in magnetic recording" (Proceedings of the IRE, vol.
50
(1962), pp. 751-761).
lOBenjamin B. Bauer, "A century of microphones" (Proceedings of the IRE, vol. 50
(1962), pp.
719-729).
llHarry G. Olson, "Loudspeakers" (Proceedings of the IRE, vol. 50 (1962), pp.
730-737). C>
12Leo Beranek, Acoustics (New York: McGraw...Hill, 1954), and Kahrs op. cit.
W
13Winston E. Kock, "Speech communication systems" (Proceedings of the IRE, vol. 50
(1962),
pp.769-776).
Pgina 32
34
concern for high fidelity.14 A second factor, unnoticed by most consumers,
was the availability of accurate test equipment. This equipment allowed en..
gineers to set standards of performance for the various parts of electroa..
coustic systems, so that components could be combined to achieve pre..
dictable performance.15 Another factor was a vigorous group of
entrepreneurs-Amar Bose, Avery Fisher, Sidney Harman, Jim Lansing,
Saul Marantz, Hermon Scott, and many others-whose technologically su..
productos Perior estimularon demand.16
Entre los nuevos productos fue de radio de FM, que haba comenzado justo antes
la guerra. Se dio no slo la msica ms alta fidelidad, sino tambin la recepcin
que era
casi libre de esttica. YO? La eventual aceptacin de FM debi en gran parte a
trabajo terico en el diseo de altavoces. Edward Howard Armstrong, el
desarrollador principal de radio FM, haba estado promoviendo durante ms de
una
dcada con xito slo moderado, cuando en 1950 se consult con
Leo Beranek y Jerome Wiesner en el MIT. Beranek y Wiesner analizaron
el problema como uno de costo y el tamao de los equipos: adems de un
receptor de FM,
el oyente necesita un amplificador y un altavoz de gran armario para conseguir
hi..fi
calidad. Beranek ide un circuito analgico que modela todas las partes de un
fuerte ..
orador, incluyendo el circuito elctrico, el transductor, el cono de radiacin,
y el aire dentro de la cual pasa el sonido. Edgar Villchur de la AR en voz alta ..
compaa de altavoces utiliza la teora Beranek para disear un pequeo hi..fi voz
alta ..
altavoz que podra fabricarse de manera econmica, y contribuy a
ofFM el xito en el 1950s.18
La dcada de 1950 se retratan a menudo como un perodo tranquilo en la historia
de Estados Unidos.
La economa estaba creciendo de manera constante, la mayora de los votantes
estadounidenses eran bastante contenido
con Dwight Eisenhower como presidente (1953-1961), y no haba poltica
la estabilidad, al menos en comparacin con los problemas de 1940-en la
mayora de la
mundo. Las personas estaban gastando ms tiempo en casa. Era una edad
do..it..yourself
14En los EE.UU. en 1954, se gastaron $ 70 millones de dlares en grabaciones
de msica clsica, mientras que 20
aos antes, quiz $ 750.000 fue [Gelatt op. cit., p. 302].
150liver Lea y Walter L. Welch, De la hoja de lata a estreo: Evolucin del fongrafo (En ..
dianapolis, IN: Howard W. Sams, 1959), p. 350. Un hito en la ciencia de la
acstica mea ..
surement era de Leo Beranek Mediciones Acsticas (Nueva York: John Wiley & Sons,
1949).
16U.S. Industria Electrnica de Consumo Hoy (Arlington, VA: Electrnica de Consumo Manu ..
fabri- Asociacin, 1997), p. 84.
17En la dcada de 1950 la Asociacin Nacional de Radiodifusores de radio y
televisin pro ..
promovida el lema "En radio, no hay alta fidelidad excepto en la transmisin
FM" [Newsweek,
21 diciembre de 1953, p. 69].
18Leo Beranek comunicacin personal 2 enero de 1998.
Pgina 33
Captulo 3 Halcyon Days de Ingeniera de Audio: La dcada de 1950
35
y la dcada mayora de los estadounidenses descubri la barbacoa del patio
trasero. televi,
sin haba convertido rpidamente en una parte importante de la vida
estadounidense: en 1957 el 85 por ciento
de los hogares tena un aparato de televisin, que se observ un promedio de
cinco horas
un da.19 En Europa, tambin, la economa creca (liderada por Alemania
Occidental
"Milagro econmico"), y Europa Occidental dio un paso hacia la unidad en
1957, cuando seis naciones firmaron el Tratado de Roma dirigida a un comn
market.20
La tranquilidad fue, sin embargo, con frecuencia perturbada. Haba en contra,
tinuing conflictos entre las potencias occidentales y los comunistas, coun
intentos, sobre todo la guerra de Corea, que alcanz una tregua en 1953 y
la represin sovitica de la revuelta hngara 1956. Estadounidenses recibieron
una
shock psicolgico cuando los soviticos colocaron un satlite en rbita el 4 de
Oc,
tubre de 1957, una fecha que puede ser tomado como el comienzo de la era
espacial.
La tranquilidad fue perturbado tambin por la msica de Elvis-Presley roca
estall en
estrellato en 1956, y el radio de transistores.
En 1954 la regencia de la empresa, utilizando componentes fabricados por Texas
In'
mentos, fabricado la primera radio transistor. A pesar de bolsillo, tamao, su
el rendimiento era pobre y su precio fue alto ($ 49.95) .21 En 1957 Sony
(Que hasta ese ao llevaba el nombre de Tokio Telecomunicaciones Engi,
niera Company) introdujo un transistor de radio (el TR, 63) que establece un
nuevo
estndar y atrajo a decenas de imitadores. Pronto, ayudado por la prolifera,
cin de todos, estaciones de radio de rock, radios de transistores se convirti en
una parte esencial de
la cultura juvenil. radios de transistores de Sony tambin ayudaron a establecer
un mercado para
electrnicos personales, dispositivos destinados a un solo user.22
Ms o menos al mismo tiempo que la introduccin de la radio de transistores
vino la celebracin del primer cable telefnico transatlntico. (Ver figura
2.) Aunque los cables de telgrafo a travs de la fecha del Atlntico de 1858, el
NE,
cessity para la telefona de la utilizacin de repetidores (es decir, amplificadores)
y la difi,
cultades en el diseo de los que iba a funcionar bajo el agua durante un largo
perodo de
19Erik Bamouw, Tubo de Plenty: La evolucin de la televisin americana (segunda edicin revisada)
(Nueva York: Oxford University Press, 1990), p. 198.
lomo de 1950 la economa alemana estaba empezando a recuperarse de la
devastacin de la
guerra; por 1960 que representaron one..fifth del comercio mundial de productos
manufacturados y tena
superado la economa britnica [David Reynolds, "Europa dividida y se
reuni, 1945-1995"
(TCW en Blanning, ed,. El Oxford Historia Ilustrada de la Europa moderna (Oxford: Oxford
University Press, 1996), pp. 279-304).]
21Michael Brian Schiffer, La Radio porttil en American Life (Tucson, AZ: Universidad de
Arizona Press, 1991), pp. 176-178.
22Schiffer op. cit., pp. 206-211.
Pgina 34
36
FIGURE 2. The transatlantic telephone cable, laid by ship as shown, was put into
operation in 1956. (Bell Labs photo reproduced by permission.)
time meant that until 1956 the only way to transmit a telephone signal
from the United States to Europe was by radio. 23
The 51 deep...sea repeaters in the transatlantic cable used electron
tubes, since at the time appropriate transistors of established reliability were
no disponible. For sheathing and insulation the cable made use of new syn...
thetic materials, such as polyethylene. Two cables were actually laid, each
carrying signals in a single direction, as this allowed a simpler design for the
repeaters. Together the cables carried 64 bidirectional voice channels. 24
So valuable were the undersea telephone channels that engineers de...
vised a system, called time assignment speech interpolation (TASI), to take
advantage of the pauses in speech during a telephone conversation. A per...
23AT&T began operating a transatlantic radiotelephone service in 1927, but static
and
ionospheric disturbances degraded voice..quality and sometimes interrupted
service [Eugene
F. O'Neill, ed., A History of Engineering and Science in the Bell System: Transmission Technol..
ogy (1925-1975) (New York: AT&T Bell Telephone Laboratories, 1985), p. 337].
240'Neill op. cit., pp. 341-345. A common misconception, which even appears in print
(eg
Frank N. Magill, ed., Great Events from History II (Pasadena CA: Salem Press, 1991), pp.
1502-1507), is that transistors made possible the transatlantic telephone cable. The elec..
tron..tube repeaters of the 1956 cable all exceeded the 20.. year expected life; there
had been
no electronic failures when the system was taken out of service in 1979 [O'Neill op.
cit., p.
345]. The fifth transatlantic telephone cable, which was completed in 1970, was the first to
use transistors rather than tubes [Britannica Yearbook of Science and the Future 1971 (Chicago:
Encyclopaedia Britannica, 1970), p. 165].
Pgina 35
37
JAMES FLANAGAN:
My thesis was on an automatic formant tracker which in-
volved building a real-time spectrum analyzer. This was essentially a custom
designed filter bank to do the spectrum analysis, an electromechanical scan-
ner that would convert the spectral envelope into signals that could be ana-
lyzed, and then an electronic logic set that would attempt to identify the reso-
nant peaks in that spectrum. Part of the thesis was then to evaluate how well
that formant tracker did and how well one could synthesize speech from that
informacin.
INTERVIEWER:
Was this actually built?
FLANAGAN:
Yes, over about a 3-year period. It ended up being all vacuum
tubes. This was before integrated circuits obviously, even before transistors. Eso
must have been about four, or maybe five, 6-foot relay racks of electronic
equipment, with lots of heat generated. 1
INTERVIEWER:
You said you started to drift into speech. Can you tell me about
how that happened?
BEN GOLD:
Yes, I was interested in problems of this sort. I was 16 when I
went to see the New York World's Fair. At the fair they had an exhibit of the
Voder. It was fascinating. Here was a machine that kind of spoke. Not very
well, but it spoke. I was more interested in some of the other things. Ah
was an exhibit where if you won a lottery you could make a long distance call
to anybody in the country through the Bell System, and everyone could listen
in. I got a real kick out of that, but that didn't lead to anything later. The Voder,
on the other hand, I always remembered. 2
1 James Flanagan oral-history interview 8 April 1997, pp. 4-5.
2Ben Gold oral-history interview 15 March 1997, p. 4.
son using the telephone talks, on average, less than 40 percent of the time.
TASI, using speech detectors and fast electronic switches, could transmit
parts of other telephone conversations during the pauses. A signaling and
switching system at the receiving end ensured that a listener is always con...
nected to the correct line. Put into service in 1960, TASI doubled the ca...
pacity of the cable and may be the first example of a commercial time...divi...
sion switch. 25
25Kock op. cit., and O'Neill op. cit., pp. 348-349.
Pgina 36
38
The telephone cable was an important step toward a worldwide net..
work for instantaneous voice communication. Ironically, after the long wait
for undersea telephony, it was soon challenged. In 1962 AT&T placed the
first commercial communications satellite, Telstar, into orbit. With ground
stations in the US, France, and Britain, the satellite system could carry
600 telephone calls or one television channe1. 26 There soon followed the
Communications Satellite Corporation (COMSAT), the International
Telecommunications Satellite Consortium (INTELSAT), and several se..
ries of communications satellites.
Efforts to increase the capacity of a communication channel have a
long history: the first duplex telegraphs, able to send two signals simultane..
ously on the same wire, were invented in the 1850s, and Alexander Gra..
ham Bell's first interest in electrical communication was his scheme for
multiplex telegraphy, sending signals at different frequencies on the same
wire. 27 Data compression, too, goes back to the early decades of telegraphy,
with code books that allowed businesses to compress whole sentences to
single words. 28
For the telephone companies the most important way of increasing
transmission capacity was the introduction of carrier telephony, or fre..
quency..division multiplexing, which allowed many telephone signals to be
sent over the same line, each within a particular bandwidth. 29 The tech..
nique made heavy use of electron tubes, as oscillators, modulators, demodu..
lators, and amplifiers. Also vital were wave filters, that is, circuits that
passed only the frequencies within a specified range. 30 Since these methods
260'Neill op. cit., pp. 377-394. In Telstar there were some 6000 transistors and only one
vacuum tube (a traveling...wave tube used as an amplifier). The first conversation
over Tel...
star was less memorable than Samuel Morse's "What hath Ood wrought?" or
Alexander Ora...
ham Bell's "Mr. Watson, come here. I want you": AT&T Chairman Frederick
Kappel in An...
dover, Maine asked Vice President Lyndon Johnson in Washington "How do you
hear me?",
to which Johnson replied "You're coming through nicely" [Communications (a volume
of the
series Understanding Computers) (Alexandria, VA: Time...Life
Books, 1986), pp. 79-80].
27W.A. Atherton, From Compass to Computer: A History of Electrical and Electronics Engineer...
ing (San Francisco: San Francisco Press, 1984), pp. 98-100.
28Jim Reeds, "Data compression-for telegraph" (Antenna, vol. 9
(1997), no. 2, p. 10). Caas
gives two examples from Harvey's Mining Code (1889), one of many trade...specific code
books: 'convicteth' meant 'Old shaft will need retimbering' and 'archons' meant
'Send two
good Cornish miners'.
29A voice signal occupies a bandwidth of about 4000 Hz, and this signal may be
shifted to
another position in the frequency spectrum for transmission and shifted back to
the audio
range before it reaches the receiver.
30Morton D. Fagen, ed., A History of Engineering and Science in the Bell System: The Early
Years (1875-1925) (New York: Bell Telephone Laboratories, 1975), pp. 279-280.
Pgina 37
39
BEN GOLD:
... One day in late 1959 or early 1960, I found myself at Bell Labs
talking to a gentleman named John Kelly, who is now gone [he died in 1965],
and he was describing something called the Pitch Detector to me, and he
talked about vocoders. I had heard of vocoders, but I hadn't been very aware
of what it was all about. And Kelly inspired me to look very carefully at the
problem of finding the fundamental frequency of human speech.
I was at Lincoln at that time, I had already built the Morse Code Transla-
tor, and I came back and started working on this pitch problem. That was my
entry into signal processing. I did it just out of curiosity. In those days at Lin-
coln Lab you could almost pick and choose what you wanted to work on, up to
a certain point. Working on pitch detection was considered okay, but my boss
wanted to turn it into something "useful".
INTERVIEWER:
Who was that?
GOLD:
A fellow named Paul Rosen. What he said was, "Well, if you have
such a good pitch detector, shouldn't we build a vocoder that includes that
pitch detector?" So that put me onto vocoders. Now, a vocoder, among other
things, has many filters in it, and once you get into filters you are into signal
tratamiento. It was frustrating at the beginning because although the computer
was capable of doing a good program for pitch detection, it really wasn't capa-
ble of simulating an entire vocoder. It was just too complicated. Nobody knew
how to do filters on computers. So we were in a way poised when we realized
that, "Hey, maybe there's a way to do this." We got very, very excited, and we
started doing a lot of work very quickly over a period of maybe two or three
years, and that's where most of the work came from. 1

were standard in radio engineering, carrier telephony illustrates a general
historical trend toward convergence of techniques of wire and wireless com...
munications. 31
Karl Willy Wagner in Germany and George A. Campbell in the
United States helped establish a theory of wave filters. 32 The subject re...
ceived great impetus from the work of Wilhelm Cauer, Sidney Darlington,
31Ivan S. Coggeshall, "The compatible technologies of wire and radio" (Proceedings
of the
IRE, vol. 50 (1962), pp. 892-896).
32Pioneer publications were Karl Willy Wagner's "Spulen... und
Kondensatorleitungen"
(Archiv fur Elektrotechnik, vol. 3 (1915), pp. 315-322) and George A. Campbell's "Physical
theory of the electrical wave filters" (Bell System Technical]oumal, vol. 1 (1922), pp. 1-32).
Others who made important early contributions to filter theory were RM Foster
and o.
Brune.
Pgina 38
40
ENDERS ROBINSON:
So Professor Hurley obtained eight seismic records, like
this one, so that I actually had the data in the fall of 1950.... The first step was
for me to hand-digitize these eight records by putting aT-square down with a
scale and reading off the traces point-by-point, putting them in numerical form.
INTERVIEWER:
You probably had a pretty modest sample rate!
ROBINSON:
Yes ... it took a couple of days to do a record. And then you'd
have to check to make sure. 1
ENDERS ROBINSON:
To get back to 1951, we had a method that worked, which
was called deconvolution. The next question was how to compute it. It took all
summer to deconvolve a few of these traces with Virginia Woodward [working
with a Marchant desk calculator]. At that time MIT had a computer they called
Whirlwind.... Like the ENIAC, it took whole rooms, a whole building, the Barta
Building at MIT.... So in the spring of 1952 I went to Whirlwind with Howard
Briscoe and put deconvolution on Whirlwind. 2
1 Enders Robinson oral-history interview 6 March 1997, p. 6.
2Enders Robinson oral-history interview 6 March 1997, p. 10.
y otros. Cauer's 1926 dissertation on The Realization of Impedances of
Specified Frequency Dependence was a landmark contribution to the system...
atic understanding of electrical filters. Cauer stated clearly the objective of
being able to design a filter having prescribed characteristics: "... it is less
important for the electrical engineer to solve given differential equations
than to search for systems of differential equations (circuits) whose solu...
tions have a desired property."33
AT&T introduced carrier systems as early as 1918 and 1920, but the
first system to be widely adopted was introduced in 1924. On a single pair of
wires, three 2...way voice channels were carried at frequencies above the
voice frequencies (in the 5 to 30 kHz range), thus quadrupling capacity.34
Continued improvements to carrier systems were made possible by the neg...
ative...feedback amplifier and by new electron tubes for use with higher fre...
quencies (to 100 MHz and above).35 Coaxial cable, with much greater
33Emil Cauer and Wolfgang Mathis, "Wilhelm Cauer (1900-1945)" (Archiv fur Elektronik
und Obertragungstechnik, vol. 49 (1995), pp. 243-251); the quotation is from p. 244.
340'Neill op. cit., pp. 5-6.
350'Neill op. cit., p. 2.
Pgina 39
41
ENDERS ROBINSON:
The water layeritself is reverberating like a drumhead and
that hid the signals coming from the depths. Deconvolution removed those
reverberations.
By the late 1950s there were lots of areas they wanted to explore off-
shore, like the Gulf of Mexico and the Persian Gulf, and in Venezuela they
had Lake Maracaibo. They realized that the only way they could get rid of
water reverberations was by deconvolution. The only way they could decon-
volve was by signal processing, because analog cannot do it. So they were
sort of forced into digital that way.
... The analog methods they used would be electrical band-pass filter-
ing, high-pass filtering, and low-pass filtering. They could also adjust the
traces in time: move one trace with respect to the other. In other words, they
might think, "Well, the signal's coming in at 40 hertz, so we will band-pass it
close to 40 hertz to find the signal." That would be analog processing because
they did it by an electric circuit.
INTERVIEWER:
And that was supposed to strip out the reverberations?
ROBINSON:
They could use band-pass filtering if the reverberations were at a
different frequency than the deep reflections. That was the idea, but it didn't
work because all these signals overlapped in frequency. The companies de-
cided they had to get rid of those reverberations, which they could do through
deconvolution. That meant going digital, so they started spending the money.
That was the advantage of the oil industry, they had money. The other thing is
that they had a history of spending money on computers through their refining
operaciones. They had close contacts with IBM and other computer compa-
nies.1

bandwidth, and, in the 1950s, microwave relay systems permitted carrier
systems of higher capacity, such as the L,3 system bearing 1800 voice chan,
nels in one circuit. 36
A different approach to increasing transmission capacity was that
taken by Homer Dudley at Bell Labs in the 1930s. He reasoned that speech
is formed by modulating, with slowly changing vocal resonances, the sound
produced by vocal sources, either the vocal cords or the turbulent airflow at
constrictions in the vocal tract. The source could be characterized as either
aperiodic (unvoiced sounds) or periodic (voiced sounds), and if it is peri,
odic its frequency could be measured. The modulations of the speech spec,
36Kock op. cit.
Pgina 40
42
FIGURE 3. Speech-processing technology had never before been as glamorous
as it was at the 1939 New York World's Fair, where the Voder, a speech
synthesizer,
was demonstrated. The operator worked at a keyboard, with a wrist bar to control
the voicing parameter and a pedal for pitch control. (Bell Labs photo reproduced
by
permission.)
trum could be measured by the relative energy in contiguous filter bands.
This information might then be transmitted and the speech reconstituted
at the receiver. The analysis..synthesis system, called the "vocoder", could
achieve speech transmission with a 300..Hz bandwidth, while a traditional
telephone channel required a bandwidth of 3000 Hz. 3
7 The so..called
"Voder" used hardware conceptually similar to the vocoder synthesizer, but
the Voder was a human..controlled synthesizer. (See Figure 3.)
37Sidney Millman, ed., A History of Engineering and Science in the Bell System:
Communication
Sciences (1925-1980) (AT&T Bell Telephone Laboratories, 1984), pp. 99-102; y
Ben
Gold personal communication 30 January 1998.
Pgina 41
43
INTERVIEWER:
Can you re-cap the transition from when people were skeptical
to when the companies started investing in it?
ENDERS ROBINSON:
It came when Texas Instruments Company, which was
Geophysical Services Incorporated, said, "give us your exploration records
from the Gulf of Mexico and we'll remove the reverberation by deconvolution."
Then they showed the results, and everybody flip-flopped.
INTERVIEWER:
And that happened in the late 1950s?
ROBINSON:
Early 1960s. They were working on it in the late 1950s. The geo-
physical people involved at Texas Instruments were previously my research
assistants at MIT. Mark Smith was one who was really working on this; he be-
came vice-president at Texas Instruments. Cecil Green was so ecstatic he
gave MIT a building-the Green Building-which was his first real act of phil-
anthropy, and from there he gave professorships and more buildings to sev-
eral universities.
INTERVIEWER:
You're saying that Texas Instruments led the way, then other
exploration companies developed that ability too?
ROBINSON:
It is fair to say that of all the geophysical companies, Texas Instru-
ments had the ability and motivation to carry out this program. 1
MANFRED SCHROEDER:
... In early June 1954, I got a letter from Bell Labs of-
fering me a job at $640 per month. I could have gone to Siemens for 500
marks a month, but here was a better company offering me five times as
much, so of course I accepted. I would have gone for nothing. 2
MANFRED SCHROEDER:
I thought I should do something of fundamental interest
to Bell Laboratories, and I elected on my very first day at Bell Labs to go into
speech. I felt this should be of interest to the telephone system, and it was
something new.
So I started speech research. After two years or so I discovered I couldn't
really make my speech synthesizers sound better if I didn't know more about
the human ear, so I got into hearing research. Then quite a few years later we
were starting to think about speakerphones and conference telephone
arrangements. Then room acoustics came in again. So the main parts of my
career at Bell Labs dealt with speech, hearing, and room acoustics, in that
temporal order. 3
2Manfred Schroeder oral-history interview 2 August 1994, p. 40.
Pgina 42
44
100,000,000
1,000,000
10,000
100
0
~
a::
>-C)
a::
w
.01
z
w
.0001
.000001
.00000001
-
~
--.......
"
V ~HRESHOLD OF FEELING " "
.....
yo
\.
.....
.;
"""
" ;'
.....
\. I..-
.....
yo
;';'~
.....
:-..1
'"
i' "-
"
/
"
'"
J
'~ESHOLD OF AUDIBILITY /
yo
yo
I ./
I--,-~
10,000
1000
Cii
w
z e. 100
w
Cl
::::>
10
~
::J
a..
~
w
a::
::::>
es
0.1
es
w
a::
a..
.01
es
~
Ii
.001
.0001
8
diecisis
32
64
128
256
512
1024 2048
4096 8192 16384 32768
FREQUENCY
FIGURE 4. The range of human audition as depicted by Harvey Fletcher and Ray-
mond L. Wegel in 1927. The vertical axis is pressure amplitude; the horizontal
axis is
frecuencia. (Bell Labs image reproduced by permission.)
Dudley's work provides one example of the research on human speech
and hearing conducted, over more than 70 years, at Bell Telephone Labora,
tories in order better to design telephone equipment that faithfully, or at
least intelligibly, conveys the human voice. 38 An earlier example is the
graphical depiction of the range of human hearing published in 1927 by
Harvey Fletcher and Raymond L. Wegel; it has become one of the land,
marks of 20th,century science. (See Figure 4.) In conducting such research,
engineers developed the new electronic tools of audio engineering, such as
oscillators, amplifiers, attenuators, and frequency analyzers.
In the 1940s Ralph K. Potter and other researchers at Bell Labs began
producing sound spectrograms, which are portrayals of the frequency con,
tent of speech as it varies with time. 39 The sound spectrograph, the rna,
chine that produces the spectrograms, is an early example of signal,process,
ing hardware. (See Figure 5.) It became standard in phonetics laboratories
around the world. 40 Figure 6 shows a spectrogram with some of the charac,
teristics of the speech marked. This display format for speech continues to
38Fagen op. cit., pp. 926-958.
39Fagen op. cit., p. 957.
Pgina 43
45
--ELECTRICALLY
SENSITIVE PAPER
STYLUS RECORDS FILTER
OUTPUT ON ELECTRICALLY
SENSITIVE PAPER AS FILTER
TUNING IS CHANGED
......MECHANICAL LINK
/
r-:--------+-oI
... SPECTROGRAM
/ RECORDING
yo
yo
MAGNETIC,
RECORDING \
\
\
RECORDER AND...,
REPRODUCER UN IT
..........
LOOP OF MAGNETIC TAPE ON
WHICH SOUNDS ARE RECORDED
FOR REPEATED PLAYBACK
FIGURE 5. This diagram shows how the spectrograph works. (Audio
Engineering
Society image reproduced by permission.)
be used today.41 And the technique itself-time..frequency signal represen..
tation-eame to be widely used in signal processing.42
One approach to understanding human speech and hearing was by
simulation, and in 1950 Bell Labs researchers constructed electrical analogs
of the vocal tract and of the inner ear.43In a few fields, such as meteorology,
numerical simulation had been tried, but the technique was little known
and would in most cases, since computers were not available, have been im..
practical.44
Understanding of human speech was employed to devise more effec..
tive means of transmission. For example, the late 1950s saw pioneering
work, both at Bell Labs and at MIT's Lincoln Laboratory, on automatic
recognition of phonemes (speech sounds). This had many potential appli..
cations, one being even greater bandwidth compression than the vocoder,
41Robert W. Lucky, Silicon Dreams: Information, Man, and Machine (New York: St. Martin's
Press, 1989), p. 226. A landmark book published in 1947, Visible Speech by Ralph K.
Potter,
George A. Kopp, and Harriet C. Green, made this technique widely known
[Lucky op. cit.,
p.226].
42Franz Hlawatsch and G. Faye Boudreaux..Bartels, "Linear and quadratic
time...frequency sig...
nal representations" (IEEE Signal Processing Magazine, vol. 9 (1992), no. 2, pp. 21-67).
44Frederik Nebeker, Calculating the Weather: Meteorology in the 20th Century (New York:
Academic Press, 1995), pp. 107-110. One of the first extensive uses of numerical
simulation
in engineering was in the Manhattan Project to build the atomic bomb, but this
work was of
course not publicized.
Pgina 44
46
RISING
INFLECTION
UNVOICED
SOUND
VOICED
RESONANCE
SOUND
BARS
FALLING
INFLECTION
3500 ~
JOOOffi
2500 ~
2000 ;:
1500 ~
1000 ~
500
@
o
ff:
FIGURE 6. In the spectrogram the horizontal axis is time, the vertical axis is fre-
quency, and the intensity of the sound is shown by the darkness. (Bell Labs photo
since all that would need to be transmitted is the indication of what
phonemes make up an utterance. 45
In the improvement of communications, as well as the audio technolo...
gies discussed earlier, electroacoustics played an important part. Electro...
acoustics is the technology of converting acoustic energy into electric energy
and vice versa. A technology comprises a set of techniques and the scientific
understanding of them. Though there were earlier examples of electro...
acoustic transducers (devices that convert energy from one form into an...
other), such as electric bells and telegraph sounders, it was Alexander Gra...
ham Bell's invention of the telephone in 1876 that made electroacoustics a
major field of endeavor. 46 The development of electron tubes in the first
decades of the 20th century enlarged the field, as it led to new applications of
microphones, loudspeakers, and other types of electroacoustic transducers in
radio, public...address systems, the electric phonograph, sound movies, mag...
netic recording, and underwater sound...ranging (later known as sonar). aqu
we see the electron tube as an enabling technology that, like the integrated
circuit later, allowed a great expansion of the realm of signal processing. 47
45Kock op. cit., and Eva C. Freeman, ed., MIT Lincoln Laboratory : Technology in the National
Interest (Lexington, MA: MIT Lincoln Laboratory, 1995), p. 231.
46An excellent history of electroacoustics, which takes the story up to the 1930s,
is in Fred..
erick V. Hunt's Electroacoustics: The Analysis of Transduction, and Its Historical Background
(Cambridge, MA: Harvard University Press, 1954).
47Gerald Tyne's Saga of the Vacuum Tube (Indianapolis, IN: Howard W. Sams, 1977) de..
scribes the development and manufacturing of electron tubes up to 1930.
Pgina 45
47
MANFRED SCHROEDER:
One early highlight was the following: The main problem
with synthesizing not just intelligible speech, which we could do, but natural
sounding speech was the so-called pitch problem, extracting the fundamental
frequency from telephone quality-pardon the expression-speech signal. En
about 1955 John Pierce asked me to use Homer Dudley's vocoder principle, not
to compress the bandwidths of telephone speech so you could send five times
as many conversations over a transatlantic cable, but to take a high-fidelity
speech signal off a bandwidth of 10 kilohertz or more, compress that down to 3
kilohertz, and send it over regular telephone channels. In other words, send
high quality speech over regular telephone channels. That was what John
Pierce asked of me. His idea was another application of Dudley's vocoder.
I said, "John, we want to send something that's even better than tele-
phone speech, and we have this pitch problem. How can we ever get some-
thing better if we can't solve the pitch problem?" Well, it was quite clear it
couldn't
be solved. So my idea was to take a so-called baseband speech signal up to 2
kilohertz, then synthesize the rest from 2 kilohertz to 10 kilohertz by Dudley's
vocoder method, and circumvent the pitch problem by generating the higher
harmonics from applying a very severe nonlinear distortion to the baseband.
So you transmit the baseband from 300 to 2000 hertz. At the other end
you would take the baseband, distort it nonlinearly to generate frequencies up
to 10 kilohertz, and then vocoder-fashion give them the right amplitude. Esta
thing was christened "voice-excited vocoder" by Ed David. (David later be-
came Science Advisor to the President with an office in the White House.)
Everybody was flabbergasted. Pierce said, "Manfred, that's the first
recorder that sounds like a human!" Yes, because it had the natural intonation.
What makes human speech sound so natural is the intonation pattern andnot
the buzzy kind of synthetic speech that you usually hear. That was a real break-
through, and so shortly after that I was put in charge of all acoustics research. 1
1 Manfred Schroeder oral-history interview 2 August 1994, pp. 55-56.
A science of electroacoustics requires the ability to measure the per..
formance of transducers, amplifiers, and filters, both steady..state and tran..
sient response. A basic tool, the sound..level meter, was improved in the
1950s through the use of transistors and rugged condenser..microphones.
Engineers specified protocols for measuring the characteristics of micro..
phones and loudspeakers, such as directivity vs. frequency, linearity vs.
level, and power..available efficiency vs. frequency.48 A valuable application
4
8 Leo L. Beranek, "Electroacoustic measuring instruments and techniques" (Proceedings of the
IRE, vol. 50 (1962), pp. 762-768).
Pgina 46
48
SIGNAL
SOUND APPEARS TO
COME FROM UNDELAYED
SOURCE
FIGURE 7. A modern version of the Buridan's-ass problem is solved by signal
delay: if the same sound comes from two sources, the hearer perceives the sound
as coming from the source with least delay. (IEEE image reproduced by
permission.)
of electroacoustic transduction was the electronic artificial larynx devel..
oped in 1959 at Bell Labs and marketed by Western Electric; when pressed
against the throat it supplied sound similar to that of the vocal cords. 49
Electroacoustics is closely allied to the study of human speech and
hearing, since such study frequently leads to improved design of electro..
acoustic devices. For example, in 1951 Helmut Haas discovered that if the
same sound comes from two sources, and if the signal from the one is delayed
5 to 35 milliseconds, then the hearer perceives the sound as coming from the
undelayed source even if it is many decibels weaker. (See Figure 7.) Shortly
after its discovery, this so..called Haas effect was used in designing the pub..
lic..address system for St. Paul's Cathedral in London. (See Figure 8.)
Also related to electroacoustics are underwater acoustics and the de..
sign of sonar systems. In the first world war, hydrophones, or underwater
microphones, were much used as a means of detecting submarines.5oTo de..
termine the direction of a sound, arrays of hydrophones were used, and they
were steered either mechanically (repositioning the array) or-an early ex..
ample of beamforming-electrically (using electrical compensators to intro..
duce delays in the different hydrophone channels).51 Some hydrophones
4
9 Millman op. cit., p. 127.
50Willem D. Hackmann, Seek & Strike: Sonar, Anti--Submarine Warfare and the Royal Navy
1914-54 (London: Her Majesty's Stationery Office, 1984), pp. 45-71.
51William S. Burdic, Underwater Acoustic Signal Analysis (Englewood Cliffs, NJ: Prentice..
Hall, 1984), p. 8.
Pgina 47
49
FIGURE 8. A public address system that incorporates delays in the lines to loud-
speakers can give the audience the impression that the sound is coming directly
from the human speaker. (As so often, the worshippers tend to sit at the back of
the
church: "If there were enough back pews, they could fill the church.") (IEEE
image
were equipped with electrical filters to improve the signal--to--noise ratio.52
The early systems were analog, but in 1960 VC Anderson showed how to
steer a hydrophone array using digital shift registers to introduce the desired
delays.53
Radar was another area where digital signal processing was used in this
perodo. One US radar system, known as SAGE (Semi--Automatic Ground
Environment), developed in the 1950s, did much to stimulate the develop--
ment of both computer and communications technologies.54When com--
pleted in 1963, SAGE consisted of 23 interconnected direction centers,
52Burdic op. cit., p. 9.
53Burdic op. cit., p. 13.
54The MIT computer project known as Whirlwind, headed by Jay Forrester,
became part of
the SAGE project. Among the innovations of Whirlwind was magnetic..core
memory. por
SAGE, IBM transformed the prototype Whirlwind into the IBM AN/FSQ..7
computer. An..
other innovation for SAGE was the cathode..ray..tube computer terminal. [Martin
Camp..
bell..Kelly and William Aspray, Computer: A History of the Information Machine (New York:
Basic Books, 1996), pp. 165-169.]
Pgina 48
50
HANS SCHUESSLER:
Las estructuras que se utilizan para simular la transferencia de fun-
ciones en una computadora analgica son precisamente el mismo que se utiliza
para el SIG-digitales
procesamiento nal. La diferencia es que en el caso de una computadora analgica
que
tener slo un integrador como el elemento bsico. En DSP que tiene el retraso
ele-
ment. Adems necesita multiplicadores. En un caso anlogo es slo un poten-
tiometer; en el otro caso, se trata de un multiplicador, un multiplicador digital. la
suma
se hace simplemente con el circuito integrado, el integrador se puede hacer
fcilmente:
la estructura es la misma. Todas las estructuras que tenemos, sabemos por ejem-
ple las estructuras en cascada, cosas por el estilo, se pueden hacer y se han hecho,
y lo hice, de vuelta en los ltimos aos 50 con medios analgicos.
... [Sola] la computadora analgica slo para jugar con los potencimetros, shift
ing los polos y ceros alrededor y mirar las respuestas de impulso; y nosotros
encontr por fin, como hiptesis, que el mnimo se lograr si la frecuencia
respuesta en el comportamiento de la banda de parada es de Chebyshev, y la
respuesta al impulso
en el dominio del tiempo se extingue en un Chebyshev manera tambin. Algo
como esto.
ENTREVISTADOR:
determinado empricamente?
SCHUESSLER:
Empricamente. No pudimos demostrarlo, pero encontramos muy buenos
resultados
slo por jugar con el ordenador analgico. Una vez ms esta hiptesis -no ms-
era un punto de partida para el diseo de filtros de este tipo en el ordenador
digital.
ENTREVISTADOR:
Por qu se mueve a una computadora digital?
SCHUESSLER:
Bueno, digamos que encontrar cientos de ejemplos slo por jugar
todo es una posibilidad, pero no es satisfactorio. En este caso tenemos el
regla general, y el desarrollo del programa para hacer que bajo ciertas
restricciones
es, con mucho, mejor. Y ms tarde se nos ha dicho que estos circuitos que pu-
cado los resultados han sido muy utilizado en la prctica, que es de alguna
manera satisfactoria
fying. Por cierto, un informe acerca de que ha sido publicado en el IEEE Trans
acciones sobre Teora de Circuitos, que es '65. 1
1Hans Wilhelm Schuessler entrevista-historia oral 21 de abril de 1997 pp. 5-7.
cada uno alimentado por los datos de radar de algunos cientos de estaciones. En
total hay
fueron 1,5 millones de millas de lneas de comunicacin, y los datos fueron
transferidos
en forma digital. Para este fin, los ingenieros del MIT y Bell Labs desarrollaron
los primeros mdems high..speed (modulator..demodulators para el envo digital
seales a travs de los telephone..lines analgicas), que eran capaces de
transmisin ..
ting 1600 bits por segundo. 55 En relacin con el trabajo sobre SAGE fue la intro
..
duccin en 1958 por AT & T del sistema Datfono, el primer comercial
mdems especficamente para la transmisin de datos de la computadora a travs
de lneas telefnicas. 56
procesamiento de radar y de seal digital permiti al presidente Eisenhower a
reclamar un xito para los Estados Unidos en la exploracin espacial en la
secuela
Op 55Communications. cit., p. 18.
Pgina 49
51
del lanzamiento del Sputnik. Esta fue la medicin precisa de la distancia de
la luna por los ingenieros de Lincoln Lab. Utilizaron continua ... radar de onda
con
seudo ... secuencias de pulsos aleatorios y, a continuacin cruz ... correlacionado
la sig reflejada ...
nales con la secuencia transmitida. Las seales se cuantifican a ocho bits
y grabado en la cinta. El pico de correlacin cruzada ... indica el tiempo de viaje
de las ondas de radar a la luna y back.57
Tanto sonar y radar involucran el procesamiento de seales dbiles en presencia
de un ruido considerable, un reto que se presenta en muchas otras reas,
tales como las comunicaciones de imgenes y el espacio biomdicas. Durante la
guerra mundial
II una serie de personas contribuy a una teora matemtica de las seales y
ruido, sobre todo Norbert Wiener y Steven O. Rice. En 1950, James Lawson
y George Uhlenbeck public los influyentes seales de umbral, el cual
discute la relacin entre el filtro receptor y la seal de salida ...
... a ruido y se entreg a un procedimiento para la aproximacin de la filter.58
ptima
El anlisis de los datos ssmicos, como el anlisis de los datos de radar, estimul
el desarrollo ...
rrollo tanto de la tecnologa informtica y la seal ... tcnicas de procesamiento.
Texas Instruments, originalmente un fabricante de instrumentos ssmicos
utilizados en
la exploracin de petrleo, de 59 aos se inici en 1956 para disear una
computadora digital para la i ...
cessing datos ssmicos. 60 Seal ... tcnica de procesamiento fue avanzado por,
entre
otros, Enders Robinson. A principios de la dcada de 1950 mostr cmo derivar
la de ...
engendrados seales de reflexin a partir de datos ssmicos, llevando a cabo uno
de descontaminacin ... dimensiones ...
volution. 61 La digitalizacin y los clculos se llevaron a cabo a mano, sobre una
57Bernard Widrow personal communication 2 February 1998. Widrow wrote the
following also:
"During the summer of 1959, Widrow worked at Lincoln Lab. He contacted Bob
Price about
the radar data. He told them about his theory of quantization noise. He suggested
that they
could have used 2 bits per sample instead of 8, and gotten the same correlation
result. They reo'
quantized to 2o'bits per sample and reo'ran the correlation and found the moon in
the same
lugar. They reo'ran the correlation with lo'bit samples, and still got the moon in
the same place." CJ
58Burdic op. cit., p. 12. Threshold Signals was Volume 24 of the celebrated series of books
that@
summarized and systematized the work done during the war at MIT's Radiation
Laboratory.
It was based on the work of many investigators during the war.
590eophysical Service Incorporated (OSI) was founded in 1930 to offer
reflection seismoo'
graph exploration services to the oil industry, and by 1939 it had crews working
in nine
countries around the world. During the war, OSI assembled radar equipment and
manufaco'
tured airborne magnetometers for the government. Texas Instruments was formed
in 1950
from the manufacturing operations of OS I. [The Leading Edge, October 1996, p. 1151.]
60Harvey O. Cragon, "The early days of the TMS320 family" (Texas Instruments Technical
Journal, vol. 13 (1996), no. 2, pp. 17-26).
61Jerry M. Mendel, "Seismic modeling problems in reflection seismology"
(IEEE ASSP Mag...
azine, vol. 3 (1986), no. 2, pp. 4-17). Robinson's Ph.D. thesis Predictive Decomposition of
Time Series with Applications to Seismic Exploration (completed at MIT in 1954) was reprinted
in full in Geophysics (vol. 32 (1967), pp. 418-484).
Pgina 50
52
desk calculator; the deconvolution of 32 traces, each one 600 to 800 readings,
took the entire summer of 1951.62In the spring of 1952 Robinson and Howard
Briscoe programmed the MIT Whirlwind digital computer to do the numeri...
cal filtering at high speed.63A group at the Raytheon company contracted
with MIT to do programming and computation tasks relating to the analysis of
seismograms, and in March 1954 Raytheon offered to the industry at large
what must have been the first commercial digital...signal...processing service.64
In the 1950s the economy was growing, and many companies increased
their R&D budgets. Higher education was growing even faster, and, added to
this, government funding of research increased markedly, particularly after
the launch of Sputnik in 1957. Hence there was much more research in sci...
ence and engineering, and growth of the professional societies was one con...
secuencia. While membership in the AlEE increased modestly during the
decade, reaching 65,000, IRE membership almost tripled, reaching 90,000.65
The IRE Professional Group on Audio concerned itself with micro...
phones, loudspeakers, disk recording and reproduction, stereo sound repro...
duction, film recording and reproduction, magnetic recording, electroacoustic
measurement, and speech communication systems (telephone, radio broad...
casting, public...address systems, and bandwidth...conserving systems).66 Some
62Robert Dean Clark, "Enders Robinson" (Geophysics: The Leading Edge of Exploration, Februo'
ary 1985, pp. 16-20).
63Gaining access to the Whirlwind computer was not easy for the MIT
Geophysical Analysis
Group. In response to Robinson's request for five hours per week, a meeting was
called at the
MIT President's office which included Julius Stratton (MIT Vice President), Jay
Forrester
(director of the Whirlwind project), head of the Geology and Geophysics
Department,
Robinson, and others. The outcome was that the MIT Geophysical Analysis
Group was
granted one hour per week. [Robinson personal communication 8 January 1998.]
64Robinson personal communication 8 January 1998, and Richard F. Clippinger,
Bernard
Dimsdale, and Joseph H. Levin, "Utilization of electronic digital computers in
analysis of
seismograms" (Waltham, MA: Raytheon Manufacturing Company, 29 March
1954). See
also three reports from the MIT Geophysical Analysis Group (GAG): Enders A.
Robinson,
"Linear operator study of a seismic profile of the Texas Company" (MIT GAG
Report No.4,
Cambridge, MA; 1953; 181 pp.); Enders A. Robinson, Stephen M. Simpson, and
Mark K.
Smith, "On the theory and practice of linear operators in seismic analysis" (MIT
GAG Reo'
port No.5, Cambridge, MA; 1953; 95 pp.); and Enders A. Robinson, Stephen M.
Simpson,
and Mark K. Smith, "Further research on linear operators in seismic analysis"
(MIT GAG
Report No.6, Cambridge, MA; 1954; 203 pp.).
65John Ryder and Donald G. Fink, Engineers & Electrons: A Century of Electrical Progress
(New York: IEEE Press, 1984), p. 216.
66This listing of technical areas comes from the subjects of the eight papers
contained in the
50th anniversary issue of Proceedings of the IRE (May 1962) in "Section 3: Audio",
which was
prepared with the assistance of the Professional Group on Audio.
Pgina 51
53
of the areas of research mentioned in this chapter were concerns of other IRE
Groups; filter design, for example, was an important subject for the IRE Pro,
fessional Group on Circuit Theory (which evolved into the IEEE Circuits
and Systems Society). The AlEE, as well, represented engineers doing some
of the work mentioned, and leaders of electrical engineering often published
in both AlEE and IRE journals. For example, Claude Shannon's "A symbolic
analysis of relay and switching circuits" (which a historian of computing has
praised as "a landmark in that it helped to change digital circuit design from
an art to a science")67 appeared in the AlEE Transactions, while his argument
for PCM mentioned in Chapter 1 appeared in the IRE Proceedings. 68
At the end of the 1950s there occurred a technological advance that
was to have an enormous impact on all of electronics and especially on the
field of signal processing. Indeed, it may be argued that it was crucial to the
emergence of a recognized field of signal processing. In February 1959 Jack
Kilby of Texas Instruments filed a patent forthe integrated circuit-a set of
electronic components and their interconnections on a single slice of sili,
con or other semiconductor. Some six months later Robert Noyce and Jean
Hoerni at Fairchild Semiconductor demonstrated the so'called planar
process by which the components could be economically connected.69Up
to this time the transistor was not revolutionary: as a discrete component it
was comparable in cost and performance with electron tubes.70It was
through the integrated circuit that the transistor revolutionized technology.
67Herman Goldstine quoted in Neil jA Sloane and Aaron D. Wyner, eds., Claude
Elwood
Shannon: Collected Papers (New York: IEEE Press, 1993), p. xii.
68Claude Shannon, "A symbolic analysis of relay and switching
circuits" (Transactions of the
AlEE, vol. 57 (1938), pp. 713-723), and Bernard M. Oliver, John R. Pierce, and Claude
Shannon, "The philosophy of PCM" (Proceedings of the Institute of Radio Engineers, vol. 36
(1948), pp. 1324-1332).
69Ernest Braun and Stuart Macdonald, Revolution in Miniature: The History and Impact of
Semiconductor Electronics, second edition (Cambridge: Cambridge University Press, 1982), p.
88.
70Even in the 1950s the transistor had important advantages-smaller size, greater
robust..
ness, lower power-that made it much better than the electron tube for certain
applications,
such as hearing aids and missile..guidance systems.
Pgina 52
Chapter 4
Going Digital: The 1960s

WHAT WAS HApPENING IN THE 1960s
The Graduate with Dustin Hoffman
James Bond movies
Midnight Cowboy
TV sh()IWs people were watching:
"Twilight Zone"
"Perry Mason"
"The Beverly Hillbillies"
"Mission Impossible"
the Beatles and the Rolling Stones
Motown
Simon and Garfunkel's "Mrs. Robinson"
Rachel Carson's Silent Spring
Joseph Heller's Catch..22
Harper Lee's To Kill a Mockingbird
55
Pgina 53
Q uite in contrast to the 1950s, the 1960s in the United States were
turbulent: the civil rights movement, which had begun a decade
earlier, had grown in strength; the anti...war movement escalated
even ore rapidly than US military involvement in Vietnam; otro
movement too, the one for women's liberation, aroused great energies and
passions; and there were the assassinations of John Kennedy, Robert
Kennedy, and Martin Luther King. In Europe as well as the United States,
youth were questioning the established order and demanding change or,
in the case of the Hippies, "dropping out". Despite a decade of economic
growth based in part on technical advances and of individual technologi...
cal triumphs, such as the IBM 360 computer, the first heart transplants,
and the Moon landing, there arose a widespread anti...technology senti...
ment.
Telephones were much in the news in the 1960s: the use of satellites
to relay telephone signals (and a song entitled "Telstar" headed the popular
music charts for three weeks in 1962)1 (see Figure 1); the introduction of,
and the storm of opposition to, All Number Calling (ending the use of let...
ters, as in PE(nnsylvania) 6...5000);2 the "hot line", with a red telephone at
each end, between Washington and Moscow, which was put into service in
1 Communications (a volume of the series Understanding Computers) (Alexandria, VA:
Time..Life Books, 1986), p. 89.
2 Juan Brooks, Telfono: Los primeros cien aos (Nueva York: Harper y Row, 1975), pp.
270-273.
Pgina 54
Captulo 4 a la fase digital: La dcada de 1960
57
FIGURA 1. El satlite de comunicaciones Telstar fue lanzado en 1962. (Bell
Labs
foto reproducido con autorizacin.)
1963; y la introduccin del mismo ao de los telfonos Touch..Tone (que
usado un par de tonos en lugar de una secuencia de impulsos para indicar un
dgito de un
nmero de telfono). Un pequeo culto, los "phreakers", se fascin
con la tecnologa telefnica, y en la prxima dcada muchos de ellos se movi
a la computacin como un hobby.3 Y en 1970, tanto en el extranjero y
direct..dialing
Picture..Phone servicio comenz. 4
.
Como hemos visto, Bell Labs ingenieros de la dcada de 1920 en la que haba
sido ex ..
ploring maneras de aumentar la capacidad del sistema telefnico, ya finales de
1955 direccin decidi desarrollar una modulacin por impulsos
codificados (PEM) sys ..
3En 1971 Steven Jobs se asoci con Steve Wozniak para disear y vender "cajas
azules" que
producido los tonos de controllong..distance de conmutacin telefnica (de modo
que se podra colocar una
llame sin ser facturado). Ms tarde, Jobs y Wozniak fundaron Apple Computer y
lanzaron
la era del ordenador personal. Dames w. Cortada, Diccionario histrico de Procesamiento de Datos:
Biografas (Nueva York: Greenwood Press, 1987), pp. 144-146,287-288].
4Britannica Anuario de la Ciencia y el Futuro 1971 (Chicago: Enciclopedia Britnica,
1970), pp. 166-167. En 1968 la pelcula de Stanley Kubrick "2001: Odisea del
espacio" hizo audi ..
cias conscientes ambos de telfonos de imagen y el habla computer..synthesized.
Pgina 55
58
JAMES COOLEY:
Una parte importante de mi educacin fue que la gente siempre
pens que estaba en el procesamiento de seales digitales y saba mucho sobre l,
pero no lo estaba.
As pues, en el principio, cuando los programas no estaban disponibles, las
personas se llaman ASK
cin de programas, y tengo que hablar con ellos. He aprendido de ellos mientras
pensaban que estaban aprendiendo de m. Fue una buena manera de empezar en
procesamiento de seales digitales. 1
JAMES FLANAGAN:
Una vez que tuvimos ordenadores [en la dcada de 1960] que podran hacer rea-
sonably simulaciones rpidas y obtener las seales de entrada y salida, podramos
disear cosas
como PCM diferencial adaptativa, que ahora se despliega en el telfono sis-
TEM. Esta tcnica ahorra ancho de banda y duplica la capacidad de PCM
canales. Tambin comenzamos reconocimiento automtico del habla, que haba
sido una
temprano inters en los Laboratorios Bell ....
ENTREVISTADOR:
Parece que su investigacin fue muy influenciado por lo
ordenadores que tenas.
Flanagan:
S, en efecto. Fue depende de lo que los ordenadores eran capaces
que hacer. Usted no fue en un gran proyecto que no tena ninguna posibilidad de
calcular una re-
sultado en su vida. Me hizo llegar cerca de eso, sin embargo, con un colega
llamado
Kenzo Ishizaka, que es profesor en Japn ahora. l trabaj con nosotros para una
nmero de aos. Tratamos de hacer una imitacin de voz sistema es todava una
corriente
problemas en la investigacin, que los modelos informticos del sistema
articulatorio, y
sintetiza una seal que intenta hacer coincidir la entrada de voz naturales
arbitraria. Parece
en la diferencia espectral entre la entrada natural y la versin sinttica.
Entonces se trata de conducir el error a cero mediante el ajuste de los parmetros
de la articu-
modelo miento de reglamentacin. Es un algoritmo de descenso de
gradiente. Nos encontramos con nuestra primera versin de este
en un Cray I, y el sintetizador corri algo as como varios cientos de veces reales
hora. Que estaba en un Cray I. As que ahora lo tenemos hasta cinco o seis veces
reales
tiempo en un C-90. 2
1James Cooley-entrevista de historia oral 11 marzo de 1997, p. 2.
2James Flanagan entrevista-historia oral 8 de abril de 1997 pp. 21, 23-24.
golondrina de mar. Se obtuvieron, el sistema de soporte T..1 so..called, que se
puso en
servicio en 1962 como comunicaciones digitales primera common..carrier del
mundo
sistema. The speech waveform was sampled 8000 times per second to repre..
sent the usual 4000..Hz channel. The amplitude of each sample was de..
scribed by seven binary digits, using nonlinear quantizing to cover better
the wide dynamic range of speech, and an eighth bit was used for signaling.
Pgina 56
Chapter 4 Going Digital: The 1960s
59
Voice was thus encoded at 64,000 bits per second. The system was designed
to allow several conversions from analog to digital to analog, since digital
lines would have to work with analog lines. 5
The T,l system spread rapidly, helped by new integrated,circuit de,
signs that offered smaller size, lower power,consumption, better perfor,
mance, and lower costs, and T,1 set a standard for voice coding that
dominated telecommunications for decades. 6 By the end of the 1960s
other developed countries had implemented their own PCM systems, 7
and in 1983 more than half of the Bell System exchange trunks were
digita1. 8
Digital transmission speeded the adoption of digital electronic switch,
ing, which was introduced in France in 1970,with Platon, a time,division
switching system, and in the Bell System in 1976. 9 Indeed, there emerged a
synergy of digital communication, digital switching, and computer,aided
information processing in the telephone system. Here the unification of
technologies reversed a historical trend of divergence between transmission
and switching. 10
A 1968 ruling of the Federal Communications Commission had far,
reaching effects. The Bell System had always forbidden direct attachment
of non,Bell equipment to the telephone network, arguing that it might
damage the network, but in 1968 the FCC ruled that people could buy and
use the Carterphone, a device to link citizen,band radios and telephones.
An important precedent for the Carterphone ruling was the case brought
before the FCC in about 1960 by the Hushaphone Company. The Husha,
phone was a mechanical device, invented by Leo Beranek, that attached to
a telephone handset so that speech could not be overheard by other people
in the room. After a lengthy hearing, the FCC ordered the Bell System to
SEugene F. O'Neill, ed., A History of Engineering and Science in the BeU System: Transmission Tech...
nology (1925-1975) (New York: AT&T Bell Telephone Laboratories, 1985), pp. 538-542.
6Robert W. Lucky, Sueos de Silicio: Informacin, el hombre y la mquina (Nueva York: St. Martin
Press, 1989), p. 235.
7Robert J. Chapuis y Amos E. Joel, Jr., Electrnica, Informtica y Conmutacin telefnica
(Amsterdam: North Holland, 1990), p. 298.
80'Neill op. cit., pp. 562-563.
9Chapuis y Joel op. cit., p. 223.
lOPatrice Flichy, Dinmica del mdem de comunicaciones: La Formacin e Impacto de Nueva Com ...
municacin Tecnologas (Londres: Sage Publications, 1995), p. 128.
Pgina 57
60
BEN GOLD:
Despus de un tiempo hemos sido capaces de probar nuestro propio codificador
de voz con nuestro pro-
detector de tono gramo. Era un proceso lento: analizar dos segundos de discurso
tomaron la
equipo de unos dos minutos, 60 a 1 en tiempo real. As que aqu es lo que
tenamos que hacer.
Tuvimos que tomar el discurso hacia el equipo, ejecute el discurso a travs de
el ordenador, ejecute el programa detector de paso, grabar, hacer nuestra 2 pistas
la grabacin, y llevarlo arriba a donde estaba el codificador de voz. Fue bastante
lento.
As que seguimos diciendo: "No sera bueno si pudiramos programar el
codificador de voz en
el equipo?" As que regres a los Laboratorios Bell, y visit Jim Kaiser. Hay
puede haber sido otras personas all, pero l es el nico que recuerdo. l dijo que
l
saba cmo construir ciertos filtros digitales. Eso era justo lo que
necesitbamos. Nosotros
dijo: "Dios mo, esto es fantstico." De hecho, podramos construir codificadores
de voz digitales. 1
BEN GOLD:
La FFT fue muy amable de una bomba. Eso fue lo que cre
procesamiento de seales digitales como un verdadero campo, porque tiene
tantos apli- posible
cationes. Todava siento que el perodo que comienza con algunas de las cosas de
Kaiser, algunos de
nuestras cosas, y trabajar su camino hasta la divulgacin de la FFT, del '63 al '67,
Fue cuando todo explot y se convirti en un campo real. 2
1Ben Oro entrevista de una historia oral transcurridos 15 marzo de 1997, p. 7.
2Ben Oro entrevista de una historia oral transcurridos 15 marzo de 1997, p. 3.
permitir el uso de Hushaphones. 11 Estas resoluciones abren la puerta a la de ..
signo, fabricacin y comercializacin de una amplia gama de telephone..related
equipos, tales como mdems, contestadores automticos, to y comunicacin de
oficina ..
sistemas cin. 12
La velocidad a la que pueden operar los mdems estaba limitada por la distor ..
cin de seales digitales durante la transmisin (ya que la ms corta es la
duracin de
cada bit, ms fcilmente un 0 podra ser confundido con un 1 versa o vice). A
reducir la distorsin y por lo tanto aumentar la velocidad de los mdems, Robert
Lucky, una
llBeranek escribe [comunicacin personal 2 de enero de 1998]: "El Hushaphone
Co. tena
tres de nosotros en la sala de audiencia en Washington: su Presidente, el cual se
llamaba Tuttle,
joven abogado, y yo como testigo experto. El sistema de Bell haba trado una
tribuna real
en la sala del tribunal para mantener todas sus abogados. Se contrat a la mejor
abogado en Nueva York para
cruzar ... examinar Tuttle y yo. Tres jueces de la corte de circuito de Estados
Unidos fueron asignados a conocer el caso
y tomar una decisin. El caso se prolong durante varias semanas. Hushaphone
gan el caso, y
[Bell System] se le dijo que no poda evitar que este accesorio extranjera en
particular de ser
usado."
Pgina 58
61
joven ingeniero de Bell Labs, ha diseado un ecualizador adaptativo, qu anuncio
..
reajustarse automticamente para adaptarse a la call..path especial
establecida. Como un
resultado, los mdems en la dcada de 1970 podra operar a 4800 bits por
segundo
(Y, en lneas dedicadas, a 9600 bits por segundo) .13 Un avance similar,
tambin hizo en la dcada de 1960 por los ingenieros de Bell Labs, era un eco
adaptativo puede ..
celler.14
En el sistema T..1 descrito anteriormente, el habla se transmite a 64
kbps (kilobits por segundo). Una invencin realizado en 1967 por Bishnu Atal
y Manfred Schroeder, llamado codificacin predictiva adaptativa o APC, por ..
mitted speech transmission of fair quality at just 4.8 kbps.15 Unlike the
T..1 system, which (in digital form) transmitted the actual waveform,
APC is, like the vocoder, a parameter..transmitting system: the voice sig..
nal is analyzed at the transmitter, parameters derived from this analysis
are sent to the receiver, and there the voice signal is synthesized. los
trade..off is a great deal of calculation at both transmitter and receiver for
a low bit rate.16
Carrying further the idea of APC, Atal invented linear predictive
coding or LPC.17 While APC is a waveform synthesis technique used as a
speech coder, LPC is a general method of speech analysis, which can be
used for speech compression, speech recognition, speech synthesis, and
other purposes. LPC, in a variety of forms, came to be widely used.18In
Japan at about the same time and independently, Fumitada Itakura and
Shuzo Saito invented maximum likelihood analysis, and Itakura developed
13Communications op. cit., pp. 18-20; and Robert W. Lucky, "Techniques for adaptive equal..
ization of digital communication systems" (Bell System Technical]ournal, vol. 45 (1966), pp.
255-286).
14Sidney Millman, ed., A History of Engineering and Science in the Bell System: Communication
Sciences (1925-1980) (AT&T Bell Telephone Laboratories, 1984), pp. 114-115. In 1980
the adaptive echo canceler became available on a single chip and became widely
used in the
telephone system.
15Bishnu S. Atal and Manfred Schroeder, "Adaptive predictive coding of speech
signals"
(Bell System Technical Journal, vol. 49 (1970), pp. 1973-1986). Atal and Schroeder demon..
strated APC at an IEEE meeting in Boston in 1967; the auditors heard the original
speech,
the transmitted signal, which was noise..like, and the reconstituted speech.
16Lucky op. cit., pp. 249-255.
17Bishnu S. Atal and Suzanne L. Hanauer, "Speech analysis and synthesis by
linear predic..
tion of the speech wave" (Journal of the Acoustical Society of America, vol. 50 (1971), pp.
637-655).
Pgina 59
62
BEN GOLD:
The other thing that happened was Oppenheim got very inter-
ested in what Charlie [Rader] and I were doing. And just around that time the
FFT hit. And it was actually an interesting story about how it hit. I was teach-
ing this course, and it was mainly a course on digital filtering the Z transform,
different kinds of filters. There was a lot of stuff along the lines of applications
to vocoders. I had a TA, a teaching assistant, named Tom Crystal, who was
still a graduate student. Well, a lot of MIT students spend time at Bell Labs.
One day, during the 1966-67 academic year, when the course was nearly fin-
ished, he brought a little document to me after class. It was by Cooley and
Tukey. At that time it hadn't been published as a paper, as a journal article,
but simply as an internal memo.
I can tell you my reaction. After the first few paragraphs the hair stood
up on my head. I said, "This is unbelievable, and I know that this is very, very
important." The rest of the paper, I couldn't understand at all. It was all mathe-
matics, and I was just not good at that. It was quadruple sums, really compli-
cated stuff. It was just algebra, but it was very hairy. So I asked Charlie, who
is better at that than me, and Tom Stockham, to "translate", because I knew it
was important. They came up with some wonderful, wonderful ways of looking
at it, which I think actually sparked the whole field.
At that point, given Oppenheim's work, given the FFT, and given the
stuff on digital filters, we said, "There's a book here," and Charlie and I sat
abajo. I had written a fair amount of stuff already for my class, and we just sat
down, we said ''we're going to write a book, we're going to get Oppenheim to
write a chapter, going to get Stockham to write another chapter." Charlie
wrote the chapter on the FFT, and that was our book. 1
BEN GOLD:
So if something came along that the managers [at Lincoln Lab]
felt was good but not sponsorable, they'd use in-house money. The FOP, the
Fast Digital Processor, was built with in-house money. It cost a lot of money,
and the directors got very antsy about it towards the end. So they said, "We've
got to use this for something useful. Let's use it for radar." So we all became
radar people. For a few years we worked on radar, and Ed Muehe and Bob
Purdy were people that I worked with, but it was really the same people, like
Charlie and me, who were pushing radar DSP. 2
1 Ben Gold oral-history interview 15 March 1997, p. 12.
Pgina 60
63
the PARCOR (partial correlation) method, which is essentially identical
with LPC. 19
It is remarkable that, at the same time that speech researchers in..
vented APC and LPC, John P. Burg, who was analyzing geophysical signals,
invented a procedure, called maximum entropy spectral analysis, that, it
turns out, is mathematically equivalent. 20 Furthermore, these results were
closely related to methods in statistics for fitting autoregressive models to
datos. 21
A great deal of calculation was required not only for the implementa..
tion of LPC and other coding schemes, but also for the research leading to
such schemes. It was not just a coincidence that this happened in the
1960s, the decade in which computer technology became widely available.
It became usual for universities and large businesses to have computers, and
much more powerful ones, such as the IBM 360 introduced in 1964, be..
came available. Perhaps even more important for advances in engineering
were the so..called minicomputers, such as Digital Equipment Corporation's
PDP..8, which first appeared in the 1960s. 22 These machines had the speed
of mainframes (though a shorter word..length), but were smaller and much
less expensive. This permitted an epoch..making change: for the first time
ever, many researchers could have their own computers. 23
19Fumitada Itakura and Shuzo Saito, "A statistical method for estimation of
speech spectral
density and formant frequencies" (Electronics and Communications in Japan, vol. 53...A
(1970),
pp.36-43). The first public presentation of the PARCOR invention was on 21
July 1969 at
a meeting of the Technical Committee on Speech of the Acoustical Society of
Japan
[Itakura personal communication 17 February 1998]. In 1986 the IEEE Morris N.
Liebmann
Award was given to Atal and Itakura for the invention of LPC.
20Manfred R. Schroeder, "Linear prediction, entropy and signal analysis"
(IEEE ASSP Maga...
zine, vol. 1 (1984), no. 4, pp. 3-13). Maximum entropy spectral analysis came to be applied
in a great many areas, such as to determine the periodicity of the polarity
reversals of the
earth's magnetic field (1972), to produce radio...brightness maps (1977), and to
improve the
angular resolution of microwave antenna arrays (1980) [Thomas Kailath,
ed., Modem Signal
Processing (Washington, DC: Hemisphere Publishing, 1985), pp. 153-154].
21Thomas Kailath personal communication 20 January 1998. Kailath pointed out
that in fit...
ting the autoregressive models to data certain equations arose that were solved
using the
Levinson algorithms for prediction theory mentioned by Enders Robinson in the
oral...history
excerpt in Chapter 2.
22The term 'mainframe' dates from 1964, 'minicomputer' from 1968. The DEC
PDP...1, in...
troduced in 1960, is usually considered, retrospectively, the first minicomputer.
The ex...
tremely popular PDP...8 was introduced in 1965.
23Martin Campbell...Kelly and William Aspray, Computer: A History of the Information Ma...
chine (New York: Basic Books, 1996), pp. 222-226.
Pgina 61
64
FIGURE 2. This is not a pizza oven but the Lincoln Lab TX-2 computer, which
Ben
Gold and Charles Rader used to perform numerical simulations of the
performance
of analog filters. The photograph shows Don Ellis removing one bit-plane from
the
64K-word ferrite-core memory. (Lincoln Lab photo reproduced by permission.)
At the forefront in the use of computers in research was MIT's Lin.,
coIn Laboratory, itself the site of much computer development.24In 1962 or
1963 Ben Gold and Charles Rader, using Lincoln Lab's TX.,2 computer,
began to simulate the performance of wave (band.,pass) filters numeri.,
cally.25 (See Figure 2.) Before then, a wave filter, such as might be used in a
vocoder, was built from wires and components in order to be tested. At this
stage, Gold and Rader-thought of the c~mputer only as a tool to speed up
the design process for a device that would be implemented in analog hard.,
ware. 26
24Two landmark computers of the 1950s, the Whirlwind and the AN/FSQ..7,
were devel..
oped in part at Lincoln Lab. Two Lincoln Lab researchers, Kenneth Olsen and
Harlan An..
derson, left in 1957 to found Digital Equipment Corporation, which changed the
computer
industry by introducing minicomputers. [Eva C. Freeman, ed., MIT Lincoln Laboratory:
Tech..
nology in the National Interest (Lexington, MA: MIT Lincoln Laboratory, 1995), pp.
221-222.]
25Charles Rader personal communication 2 January 1998.
26Charles Rader oral..history interview 27 February 1997, pp. 3-4.
Pgina 62
sesenta y cinco
BEN GOLD:
One of the reasons that we were doing radar work was that the
funding for speech work had dried up, and that was one of the reasons direc-
tors ordered us to do radar work, "because we can give you money for that."
All of a sudden LPC came along, just another bombshell.
INTERVIEWER:
Tell me when.
GOLD:
I'd say very late sixties, early seventies, probably going into up to the
mid-seventies. In any case, we jumped into that pretty quickly. We had a fel-
low named Ed Hofstetter who was actually an old microwave person. He wasn't
that old, but he came from microwave. He got interested in speech, and got
interested in programming, something he had never done, and he got very
good at it. He was also a good mathematician. When LPC came along, he
was one of the first to really pick up on it and understand it. He said, "I could
write a real time program using the FOP." At that time nobody could do real-
time computation on a general purpose computer to do LPC.
INTERVIEWER:
The FOP?
GOLD:
The Fast Digital Processor. It was a fast computer. He actually pro-
grammed the first real-time LPC on the FOP. So that got us back into the
speech area, and we actually did quite a bit of work on LPC.
INTERVIEWER:
Why did that get you back into the speech area?
GOLD:
Well, because LPC caused the funding switch to open again, and we
got money, and were able to work on stuff. 1
THOMAS HUANG:
[In about 1960] image processing was so primitive then that
we [the Image Processing Group at MIT, headed by Bill Schreiber] had to be
concerned with equipment as well. We had to build our own scanner for digi-
tizing images and reproducing them. We built one of the first image scanners
in our lab using a CRT. I was using a Lincoln Lab prototype computer called
the TX-O, probably the first transistorized computer. I had to program in as-
sembly language. After digitizing each image we had to store the data on
paper tape with punched holes. We fed that into the computer and then had to
write the result on paper tape again. I remember an image of just 240 x 240
pixels took three rolls of tape. 2
2Thomas S. Huang oral-history interview 20 March 1997, p. 3.
Pgina 63
66
The 1960s was a time when numerical simulation was coming into use
in many areas of science and engineering.27Not only Gold and Rader, but
others in the speech research community made early use of computers. por
example, in 1964 A. Michael Noll used computer simulation to show that
John Tukey's concept of the cepstrum (the Fourier transform of the loga..
rithm of the amplitude spectrum) was well suited to solving the problem of
pitch..extraction in a vocoder,28 and James L. Flanagan's classic Speech
Analysis: Synthesis and Perception, published in 1965, describes many com..
puter simulations.29This was part of a general trend toward calculational
approaches to science and engineering problems, especially numerical ex..
perimentation, which is the study of phenomena or systems by numerical
simulation rather than by observation or manipulation of physical systems.
A circuit or device may be described by its transfer function, the
mathematical relation of input to output. With analog circuits, composed
of resistors, capacitors, and so on, only certain transfer functions are practi..
cal, while a digital device incorporating a computer could realize almost any
conceivable transfer function.3oYet before the 1960s few engineers even
considered digital circuits for signal processing because computers were
hardly available, they were extremely expensive, and they were not fast
enough to do real..time signal processing.3 !
As computers became widely used for simulation in the 1960s, there
arose the idea, in a number of places, that the signal processing itself might
be done by computer. As early as 1961 a computer was built specifically for
digital signal processing. This was the TI187 transistorized computer built
by Texas Instruments for analyzing seismic data, though not in real time.32
At Lincoln Lab, Gold and Rader used a computer as a vocoder, but the
computer needed 10 seconds to process 1 second of speech.33Also in the
Academic Press, 1995), pp. 177-183.
29James F. Flanagan, Speech Analysis: Synthesis and Perception (Berlin: Springer..Verlag,
1965).
30Thomas Kailath recalls that Robert Price at Lincoln Lab used to make this
point in the
early 1960s [personal communication 20 January 1998]. As mentioned below,
Price did digi..
tal filtering (in the modem sense) around 1960 in his analysis of radar signals
from Venus.
31Kailath op. cit., pp. xi, 370.
32Harvey G. Cragon, "The early days of the TMS320 family" (Texas Instruments Technical
Journal, vol. 13 (1996), no. 2, pp. 17-26). Even earlier, in 1955, J. Fred Bucy and others at
Texas Instruments designed a hybrid analog..digital computer for seismic data
reduction
[Cragon op. cit.].
33Freeman op. cit., p. 227.
Pgina 64
67
FUMITADA ITAKURA:
Suppose that we sample that linear predictive parameter
every twenty milliseconds, fifty times a second, and suppose that we have ten
parameters, and each parameter is quantized with ten bits. So for each frame
we need 100 bits. If we multiply by fifty, that is five kilobits per second just for
the linear predictive parameters. That is too much for vocoding. So we have to
find a better method of quantizing linear predictive parameters. I tried to re-
duce the number of quantizing parameters of LPC parameters, but it was not
Perfecto. I had to think a little more in detail. I went to the new concept of partial
autocorrelation, PARCOR. By using those parameters we could quantize pa-
rameters to any number of bits. Of course, there might be some degradation
of speech resulting in instability.
INTERVIEWER:
Was this your own conception, this partial autocorrelation?
ITAKURA:
It is a well-known statistical concept.
INTERVIEWER:
That was the first time it was applied to the speech problem?
ITAKURA:
Right.
INTERVIEWER:
This was what you applied for a patent for in May 1969, and
then in July you presented the PARCOR vocoder at the special group meeting
of the Acoustical Society of Japan.
ITAKURA:
Yes, that was a very special day.
INTERVIEWER:
The same day as the moon landing.
ITAKURA:
S. The interesting thing about that is that in Japan we have regu-
lar meetings, so every eight months speech scientists and engineers get to-
gether to talk and encourage the technology and science of speech. It is infor-
mal, with one person designated to talk. Usually twenty or even forty people
get together, but that was the day of the more interesting moon landing. Muy
few attended: the chairman of the committee, myself, Saito, and a few other
researchers very close to the work, but they kept coming in and going out. Asi
que
it was essentially only three people. 1
1 Fumitada Itakura oral-history interview 22 April 1997, pp. 16-17.
early 1960s, James Kaiser and Roger M. Golden at Bell Labs began work to
transfer "the extensive continuous filter art of the electrical engineer" from
the analog to the digital domain. 34
34James F. Kaiser, "Digital filters" (Chapter 7 of System Analysis by Digital
Computer, Franklin
F. Kuo and James F. Kaiser, eds. (New York: John Wiley, 1966), pp. 218-285). Qu
puede
have been the first hardware implementation of digital filters was at Bell Labs
in 1966 and
1967; this work is described in Leland B. Jackson, James F. Kaiser, and Henry S. McDonald,
"An approach to the implementation of digital filters" (IEEE Transactions on Audio and
Elee...
troacousties, vol. 16 (1968), pp. 413-421).
Pgina 65
68
Though 'filter' originally meant a device to select certain frequencies
or ranges of frequencies from a signal, 'digital filter' soon acquired a very
general definition: "a discrete time system which operates on an input se...
quence to produce an output sequence according to some computational al...
gorithm".35 There emerged a field of digital...filter theory, which drew on the
extensive earlier work done in classic circuit theory, numerical analysis, and
more recently sampled...data systems.36In the United States, James Kaiser,
Enders Robinson, Sven Treitel, Ken Steiglitz, Ben Gold, Charles Rader,
Alan Oppenheim, Lawrence Rabiner, and Thomas Stockham did impor...
tant work in the 1960s.37In Europe, Hans Wilhelm Schuessler, Anthony
Constantinides, Vito Cappellini, and Fausto Pellandini were among the
early contributors to digital...filter theory.
The rapid advances in computing technology in the 1960s promised
to make this theory of practical importance. Mass production of integrated
circuits began in 1962, when Fairchild and Texas Instruments sold a few
thousand logic chips, and the capabilities of ICs grew rapidly.38 The avail...
ability of software to make it easier to program a computer was very impor...
tant. At Bell Labs, Kaiser and Golden used BLODI (from 'block diagram')
that could translate a block...diagram description of a system into a program
that simulated the system; at Lincoln Lab, Rader wrote a block...diagram
compiler of his own.39
Also in the mid 1960s, the discovery of a particular algorithm, the fast
Fourier transform or FFT, incited a great deal more activity. The Fourier
35Bede Liu, ed., Digital Filters and the Fast Fourier Transform (Stroudsburg, PA: Dowden,
Hutchinson & Ross, 1975), p. 1.
36Some important works on sampled..data systems were two textbooks of
Eliahu I. Jury,
Sampled..Data Control Systems (New York: John Wiley and Sons, 1958) and Theory and Ap..
plication of the Z..Transform Method (New York: John Wiley and Sons, 1964), and the article
of Rudolf E. Kalman and John E. Bertram "A unified approach to the theory of
sampling sys..
tems" (Journal of the Franklin Institute, vol. 267 (1959), pp. 405-436).
37Kailath op. cit., p. xi. A particularly influential publication was Kaiser's chapter
on digital
filters in the book System Analysis by Digital Computer (New York: John Wiley, 1966), edited
by Franklin F. Kuo and James F. Kaiser [Hans Wilhelm Schuessler oral..history
interview 21
April 1997, p. 34].
38Stan Augarten, State of the Art: A Photographic History of the Integrated Circuit (New Haven,
CT: Ticknor & Fields, 1983), p. 10.
39Rader's compiler, called PATSI (Programming Aid To System Investigation),
was used by
a few other people at Lincoln Lab; Bert Sutherland built a graphical interface for
PATSI, so
that a user could draw his block diagram and the computer would simulate the
system
[Charles Rader personal communication 2 January 1998].
Pgina 66
69
transform, named after the French mathematician Jean Baptiste Joseph,
Baron de Fourier (1768-1830), is a mathematical procedure to convert a
time...domain signal into a frequency...domain signal. (The human ear auto...
matically performs a Fourier...like transform on impinging sound waves, send...
ing signals to the brain that give the intensity of the sound at different fre...
quencies.)40 It has been extensively used by communication engineers in
studying noise, by physicists in solving partial differential equations, and by
statisticians in studying statistical distributions. It can simplify the mathe~at ...
ics, and it also often makes the phenomenon of interest easier to understand.41
The straightforward calculation of a discrete Fourier transform of N
points requires 4N2multiplications. In 1965 James Cooley and John Tukey
showed how to do the calculation with just 2N log2N multiplications.42
(See Figures 3 and 4.) Thus, even for a thousand...point transformation, the
FFT reduces the calculation required by a factor of 200, and for larger sam...
pIe sizes the reduction factor is much greater. Rabiner writes "... the algo...
rithm remained a mathematical curiosity to most electrical engineers until
an engineering interpretation was given to the procedure by Charlie Rader
and Tom Stockham.... [They] constructed a Mason flow graph interpreta...
tion of the FFT from which a wide variety of the properties of the FFT, in...
cluding bit reversal, in...place computation, twiddle factors, N log2N opera...
tion count, decomposition ideas etc., became clear."43
4 This statement is only roughly correct: the standard Fourier transform
0
corresponds fairly
closely to a bank of equal bandwidth filters, while the human ear corresponds to
a filter bank
where filter bandwidths increase with increased center frequencies [Ben Gold
personal com..
munication 30 January 1998]. James Kaiser [personal
communication 11 February 1998] ar..
gues that the ear is more a transient detector than a frequency analyzer.
41W. Morven Gentleman and Gordon Sande, "Fast Fourier transforms-for fun
and profit"
(AFIPS Proceedings of the 1966 Fall Joint Computer Conference, vol. 29
(1966), pp. 563-578).
42James W. Cooley and John W. Tukey, "An algorithm for the machine
calculation of com..
plex Fourier series" (Mathematics of Computation, vol. 19 (1965), pp. 297-301). Cooley has
authored or co..authored several articles on the re..discovery and acceptance of
the FFT:
James W. Cooley, Peter AW Lewis, and Peter D. Welch, "Historical notes on the
fast
Fourier transform" (IEEE Transactions on Audio and Electroacoustics, vol. 15 (1967), no. 2,
pp. 76-84); James W. Cooley, "The re..discovery of the fast Fourier transform
algorithm"
(Mikrochimica Acta, vol. 3 (1987), pp. 33-45); and James W. Cooley, "How the FFT gained
acceptance" (IEEE Signal Processing Magazine, vol. 9 (1992), no. 1, pp. 10-13). The deep
history of the FFT is given in Michael T. Heideman, Don H. Johnson, and C.
Sidney Burrus,
"Gauss and the history of the fast Fourier transform" (IEEE ASSP Magazine, vol. 1 (1984),
no. 4, pp. 14-21).
43Lawrence R. Rabiner, "The Acoustics, Speech, and Signal Processing Society-
A Histori..
cal Perspective" (IEEE ASSP Magazine, January 1984, pp. 4-10), quotation from pp. 4-5.
Pgina 67
70
O~-----+-_---<:>------
........ --u
(m-1)st
stage
wr
norte
mth
stage
O~-----+-_---<:>------ ........ --u
-1
FIGURE 3. In a typical step of the FFT, a pair of values are added or subtracted
after being multiplied by predetermined sine or cosine values. This operation
may be
depicted in the so-called "butterfly structure" shown. (Redrawn after Figure 8 of
Gene Frantz and Panos Papamichalis, "Introduction to DSP solutions" (Texas In-
struments Technical Journal, vol. 13 (1996), no. 2, pp. 5-16).)
The IEEE Group on Audio and Electroacoustics (G..AE), formerly the
IEEE Group on Audio, had a major role in the widespread adoption of the
FFT. It co..sponsored a special workshop on FFT and spectral analysis in the
spring of 1966, and in 1967 papers from this workshop were part of a special
issue of the G..AE Transactions, which included a classic tutorial paper on the
FFT written by members of the G..AE Subcommittee on Measurement Con..
cepts. In 1968 it organized the first of the so..called Arden House Workshops;
a hundred researchers exchanged ideas about the FFT, and many of the pa..
pers were published in a special issue of the G..AE Transactions in 1969.44
Researchers developed other algorithms for digital signal processing,
such as the Viterbi algorithm in 1967 (used especially in speech recogni..
tion),45 the chirp z..transform algorithm in 1968 (which widened the range
of application of the FFT),46 use of the maximum likelihood principle also
in 1968 (for sensor..array signal processing),47 and adaptive delta modula..
tion in 1970 (for speech encoding).48
44Rabiner op. cit. The two special issues were the following: Special Issue on Fast
Fourier
Transforms and Applications to Digital Filtering and Spectral Analysis,
IEEE Transactions
on Audio and Electroacoustics, vol. 15, no. 2 (June 1967), and Special Issue on Fast Fourier
Transforms, IEEE Transactions Audio and Electroacoustics, vol. 17, no. 2 (June 1969).
4 5 Harvey F. Silverman and David P. Morgan, "The application of dynamic
programming to
connected speech recognition" (IEEE ASSP Magazine, vol. 7 (1990), no. 3, pp. 6-25).
46Lawrence R. Rabiner, Ronald W. Schafer, and Charles Rader, "The chirp
z...transform and
its applications" (Bell System Technical]ournal, vol. 48 (1969), pp. 1249-1292).
4 7 Hamid Krim and Mats Viberg, "Two decades of array signal processing
research" (IEEE
Signal Processing Magazine, vol. 13 (1996), no. 4, pp. 67-94).
48Millman op. cit., p. 115, and Jerry D. Gibson, "Adaptive prediction for speech
encoding"
(IEEE ASSP Magazine, vol. 1 (1984), no. 3, pp. 12-26).
Pgina 68
Chapter 4 Going Digital: The 19605
71
x[7]
x[O] ...... --o---........--(:)--........
t"\-_~-O-----e-~').-- x[O]
----<>---~o
x[1] o--~-o-----I---(,)--"",,--o---A........--(::>--........ ----<:>--t----t-.o x[4]
x[3]
x[6]
-1
-1
WO
norte
x[1]
x[4]
w1
norte
x[5]
x[5]
W 2
norte
x[6]
x[3]
x[7] ......--o------e--<"l---........
n-_~-n-----e----("l--- ----C:>---_~O
-1
-1
-1
FIGURE 4. This Art Deco design is actually the structure of a 2-point, radix-2, deci-
mation-in-frequency fast Fourier transform. (Redrawn after Figure 7 of Gene
Frantz
and Panos Papamichalis, "Introduction to DSP solutions" (Texas Instruments Tech-
nical Journal, vol. 13 (1996), no. 2, pp. 5-16).)
Pgina 69
JAMES KAISER:
That got me very interested in signal processing. Now, at the time
I arrived at Bell Laboratories [ca. 1960], a change in the means of doing research
in the speech area, in the coding area, was under way. Instead of the old way,
which was to test an idea about a new way to do things by designing the electron-
ics that embodied that idea and then running tests on that physical embodiment,
we were starting to simulate the system-a compression system, an encoding
system, whatever-on the general purpose digital computer. Then we would just
run a test of the new idea, taking speech and running that through the simulated
system and listening to the result. It was much faster and more versatile.
So I got much more interested in how you took continuous systems and
got the discrete models necessary for the simulation. With my control back-
ground, I knew continuous systems and filter design quite well, and I tried to
carry over some of the same ideas to the discrete world. A lot of it carries over
as far as the recursive filters are concerned. These design techniques carry
over directly via the different transform techniques, the Z transform, the
bilinear Z transform, the matched Z transform, and so forth. But one feature of
the digital systems is that it's very easy to build finite impulse response digital
filters, whereas these are very difficult to build as continuous filters. 1o
JAMES KAISER:
The thing was, there were old timers-especially in industry-
that had been designing filters, continuous filters, for years, and all of the sud-
den these fellows were told, "Now look, I want you to build digital filters." Su
reaction was, "I don't know what a digital filter is." They were completely lost
and some of these fellows didn't want to learn the new stuff. They said
''There's too much new to learn! Don't bother me with the new stuff." So one of
my goals was to preserve all the knowledge that those fellows already had
and say to them, "Look, all you've got to do is run this little program with that
knowledge you already have and it will design a digital filter for you. You know
how to do it!" That's basically what the bilinear transformation does. It carries
the continuous filter designs over to the discrete design. 11
BEDE Llu: Yes, radar processing, and you also found sonar people inter-
ested. That was the group for serious applications [in the 1960s]. It was a seri-
ous application because people actually built it to put in systems for that pur-
pose. Speech, I think at that time was mostly in the research stage, but they
did have an application goal in that, too.
... I think signal processing is becoming so popular for two reasons. Es
application-driven, particularly as seen now, and it is very much tied with tech-
nology. So you can identify applications; you have needs to actually build sys-
tems to do certain tasks.12
lJames Kaiser oral-history interview 11 February 1997, pp. 6-7.

2James Kaiser oral-history interview 11 February 1997, p. 24.
3Bede Liu oral-history interview 10 April 1997, pp. 20-21.
72
Pgina 70
73
Besides the new digital communications, digital control systems and
digital instruments began to be more common. The first commercial digital
instruments, introduced in the early 1950s, measured frequency, pulse rate,
and time interva1. 49 There soon followed digital voltmeters, digital pulse"
delay generators, and digital frequency synthesizers.5oThe advantages of
digital systems were many: accurate realization, no adjustments, no aging,
temperature.,independence, reliability, programmability, flexibility, mono.,
lithic integration (often the entire circuit on a single chip), and little or no
signal degradation in transmission and copying.51
But engineers needed to learn about the new possibilities, and here
the workshops, conferences, and publications of the IEEE Audio and Elec.,
troacoustics Group played an important role. A new field of digital signal
processing was emerging, and it received its first textbook formulation in
1969 with Gold and Rader's Digital Processing of Signals (with chapters con.,
tributed by Stockham and Oppenheim) (see Figure 5).52 According to Don
Johnson, "Their book, more so than the algorithmic developments of the
time, pushed digital signal processing into the limelight."53 The second
textbook in the new field, and the first in Europe, appeared in 1973: Hans
Wilhelm Schuessler's Digitale Systeme zur Signalverarbeitung. 54
Though speech and other l.,dimensional signals were the concern of
most of the pioneers of the new field, image processing came to the fore in a
number of ways in the 1960s. It was then that Marshall McLuhan first
called attention to the cultural importance of the mass media ("the medium
is the message") and argued that society was moving from a "print culture"
to a "visual culture". He introduced the concept of the "global village", say.,
ing, "Today, after more than a century of electric technology, we have ex.,
49Perhaps the first commercial digital instrument was an events'per,unit,time
meter offered
by Berkeley instruments in 1950; Hewlett'Packard offered a combined frequency,
and pe,
riod,measuring instrument in 1951 [Bernard M. Oliver, "Digital display of
measurements in
instrumentation" (Proceedings of the IRE, vol. 50 (1962), pp. 1170-1172)].
500liver op. cit.
51Piet J. Berkhout and Ludwig DJ Eggermont, "Digital audio systems" (IEEE ASSP
Maga,
zine, vol. 2 (1985), no. 4, pp. 45-67). There were often, of course, disadvantages such as in,
creased bandwidth for transmission and increased power consumption.
/h"l
52Bemard Gold and Charles M. Rader, with chapters by Alan V. Oppenheim and
Thomas ( (b
G. Stockham, Jr., Digital Processing of Signals (New York: McGraw,Hill, 1969).
53Don H. Johnson, "Rewarding the pioneers" (IEEE Signal Processing Magazine, vol. 14
(1997), no. 2, pp. 20-22), quotation from p. 22.
54Hans Wilhelm Schuessler, Digitale Systeme zur Signalverarbeitung (Berlin: Springer,Verlag,
1973).
Pgina 71
74
DIGITAL PROCESSING
OF SIGNALS
BERNARD GOLD
y
CHARLES M. RADER
Li1&lXJln Laboratnry
Ma88achU8ett3 In8titute of Technology
with chapter8 by
ALAN V. OPPENHEIM
Ruearch Laboratory of Electronic8
MaoachmeU8 Imtitute of Technology
y
THOMAS G. STOCKHAM, JR.
Computer Science Department
U of Utah
nUJer8ity
McGRAWHILL
BOOK COMPANY
Ne1D York
St.Louia
San FrancUco
Londres
Sydney
Toronto
Mjico
Panam
FIGURE 5. This is the title page of Gold and Rader's Digital Processing of
Signals
(1969), which was the first textbook on digital signal processing. (Reproduced by
permission of McGraw-HilI.)
tended our central nervous system itself in a global embrace, abolishing
both space and time as far as our planet is concerned."55 For example, in
1963 both Europe and Japan received, via satellite, live television coverage
ofJohn F. Kennedy's funera1.56
In the 1960s the public was eager for photographs from space: as early
as 1959 the Soviet Luna 1 had transmitted back a photograph of the previ...
ously unseen far side of the moon; in 1966 the uS Surveyor 1 soft...landed
55Marshall McLuhan, Understanding Media: The Extensions of Man (Cambridge, MA: MIT
Press, 1994 (first published 1964)), quotation from p. 3.
56Walter A. McDougall, ... the Heavens and the Earth: A Political History of the Space Age
(New York: Basic Books, 1985), p. 396.
Pgina 72
75
BEDE Llu:
With floating point arithmetics-without getting into detail-the er-
rors are introduced through a slightly different mechanism. Well, the same
mechanism, but in a different way, which makes the analysis very hard. Irwin
Sandberg presented a very good paper at the Allerton Conference which
pointed out the problem and ways to handle it. The approach he took was a
deterministic approach. I guess you can call it classical error-analysis, or the
numerical analysis approach.
I found it very interesting. I came back and talked to Toyohisa Kaneko [a
graduate student] and said the proper way to look at the problem is through
statistical probability. I said, "This is a very interesting paper, but the underly-
ing problem is one of many, many small errors being introduced, and we
should be able to analyze the problem using this probabilistic model." We
talked about this for a little while, and a few days later Kaneko said, "Yes,
everything carries through fine." So we wrote it up and sent the abstract to
Allerton again for the '68 conference. One week before the conference he
came in and he said, "No, it didn't work." Sure enough, it did not work; nosotros
overlooked something. We quickly tried to do a lot of things, and finally I was
able to find a way to analyze the problem, which actually made the problem
much more interesting. It got the result out. And that's my first work on digital
filters. 1
HANS GEORG MUSMANN:
During that time I worked on my thesis [in the early
1960s], I also built up a small computer with transistors-I had to connect the
transistors of course by hand! I wanted to learn how computers work and op-
erate. So, together with a colleague, I built up a small computer which could
be operated by voice.
INTERVIEWER:
How was that done? Did you have any voice recognition tech-
niques?
MUSMANN:
Well, to a certain extent. I developed a very simple technique in
which the voice is split up into frequency bands. Then I sampled the output of
the frequency bands and used these patterns to distinguish between ten dig-
its. Some commands, like "add in," or "subtract," or "multiply," and so on, were
for operating the computer ....
INTERVIEWER:
Why did you choose to use voice input for that device?
MUSMANN:
I thought it would be very nice to have a computer you can oper-
ate with voice. That is what you need still today!2
12Hans Georg Musmann oral-history interview 30 August 1994, pp. 4-5.
Pgina 73
76
on the moon and sent back high..quality photographs of the Ocean of
Storms; and later in 1966 the uS Lunar Orbiter 1 attracted even greater
attention with photographs of the far side of the moon and of the earth
from beyond the moon. In 1969 the public marveled at live television of
the first steps onto the moon and at detailed pictures telemetered from
Mars.57These and other ventures into space posed great challenges for
image coding, transmission, and reconstruction. Among the early advances
in image coding was the block transform method introduced in 1969 by
Grant Anderson and Thomas S. Huang,58 and not long afterwards came the
use of the discrete cosine transform for image coding.59
In the United States, color television finally caught on, and beginning
in 1964 television sets had to have all..channel tuners, that is, for UHF as
well as VHF, which gave a great boost to UHF broadcasters.6oThe 1960s
saw the beginning, in Japan, of research on high..definition, wide..screen
television.61Other developments related to image processing were the
adoption in 1968 by the International Telecommunications Union of a
standard for facsimile apparatus and transmission.62At Bell Labs, Leon Har..
mon studied how much visual information is required for recognition of
faces; his quantized portrait of Abraham Lincoln (with only 756 bits of in..
formation) became a popular image (Figure 6).63
57Better encoding and a higher bit rate (16,000 bits per second rather than 8 bits per
second)
made the pictures sent by Mariner 6 and 7 in 1969 much more detailed than the
pictures
sent by Mariner 4 in 1965 [Britannica Yearbook of Science 1971 op. cit., p. 165].
58Grant Anderson and Thomas S. Huang, "Picture bandwidth compression by
piecewise
Fourier transformation" (IEEE Transactions on Communications, vol. 19 (1971), pp.
133-140); an earlier version of this paper appeared in the proceedings of the Purdue Univer,
sity Centennial Symposium on Systems and Information Sciences, held 28-
30 April 1969.
An important early paper on block quantization in general is JJY Huang and
Peter M.
Schultheiss, "Block quantization of correlated Gaussian random variables" (IEEE
Transac--
tions on Communications Systems, vol. 11 (1963), pp. 289-296).
5~asir Ahmed, T. Natarajan, and KR Rao, "Discrete cosine transform" (IEEE Transactions
on Computers, vol. 23 (1974), pp. 90-93).
60 The US Consumer Electronics Industry in Review: 94 Edition (Washington, DC: Electronic
Industries Association, 1994), p. 19.
61Simon Haykin, An Introduction to Analog and Digital Communications (New York: John
Wiley, 1989), p. 320.
62Henry Petroski, Invention by Design: How Engineers Get from Thought to Thing (Cambridge,
MA: Harvard University Press, 1996), p. 113.
63Lucky op. cit., pp. 345-346. Another Bell Labs researcher, Manfred Schroeder, won
first
prize at the 1969 Las Vegas Computer Art Competition with a computer,generated
image of
a human eye composed of a 65,000, letter text (illustrating, one might say, the
proverb that
one picture is worth a thousand words).
Pgina 74
77
FIGURE 6. Leon Harmon's quantized portrait of Lincoln became a popular
image.
(Bell Labs photo reproduced by permission.)
In science, imaging technology was advancing rapidly. Computerized
tomography or CT scanning (reconstructing a 2, or 3,dimensional image of
an object from the data from projections through the object) moved from
conception to a practical device in 1971.64One of the pioneers of CT scan,
ning, Ronald Bracewell, helped develop another new imaging technology
in the late 1960s: very,long,baseline interferometry for high,resolution as,
tronomy and geodesy.65
Radar imaging was improved by the application of signal,processing
concepts; a milestone in this field was Fred E. Nathanson's 1969 book Radar
64Bettyann Holtzmann Kevles, Naked to the Bone: Medical Imaging in the Twentieth Century
(New Brunswick, NJ: Rutgers University Press, 1997), pp. 145-172. Kevles identifies
three
pioneers of the technique: Ronald Bracewell, who wished to construct images of
the sun and
the moon from radio telescope data, and William Ohlendorf and Alan Cormack,
both of
whom wanted to use x..ray data to form images of parts of the body.
65Kenneth L. Kellerman and A. Richard Thompson, "The very..long baseline
array" (Scien..
tific American, vol. 258 (1988), no. 1, pp. 54-61). The two imaging technologies that
Bracewell pioneered-CT..scanning and very..long..baseline interferometry-are
entirely dif..
ferent, though he came to both of them from the practice of radio astronomy;
very..long..
baseline interferometry has had even wider application than has CT..scanning.
Pgina 75
78
HANS GEORG MUSMANN:
That was about 1966, 1967 [that I became interested
in digital image signals]. Since I was looking especially for representation of
visual information, I started with facsimile.
INTERVIEWER:
Why were you particularly interested in visual information?
MUSMANN:
I thought, "We have a present type of communication, the tele-
phone; visual information should be the next step in communications." But it
took a long time!1
ALAN OPPENHEIM:
When I finished the thesis I decided I was sick of it, and so I
thought what I would do is change fields. I would do something different. Asi
que
what I decided to do was get involved with Ken Stevens and his group on
speech stuff. He was very well known in the speech area, and I thought I was
going to be taking off in a totally different direction. But within a couple of
weeks I realized that speech was the perfect application for deconvolution
using homomorphic filtering, because in speech processing deconvolution is a
really key aspect. You think of speech as essentially, in an engineering sense,
modeled as the convolution of the air flow through the vocal chords and the
impulse response in the vocal cavity.
... And then all speech compression systems and lots of speech recog-
nition systems are oriented toward doing this deconvolution and then process-
ing things separately and then going on from there. So speech really became
a significant focus for me for applying this homomorphic signal processing
and it's become one of the important tools for that. 2
ALAN OPPENHEIM:
... I also remember that around the time I graduated I was
talking to Tom Stockham, and I said, "It's kind of depressing that nobody is
picking up on this stuff" and Tom said, "It's not depressing. Actually what's
great is that we have this all to ourselves for a while until other people really
discover it.,,3
SIDNEY MILLMAN:
[The FFT] revolutionized much of engineering practice,
haciendo disponibles muchas tcnicas que anteriormente habran necesitado
prohibitiva
cantidades tivos de computacin ....4
1 Hans Georg Musmann entrevista de historia oral 30 de agosto de 1994, p. 8.

2Alan Oppenheim entrevista de historia oral 28 de febrero de 1997, pp. 9-10.
3Alan Oppenheim entrevista de historia oral 28 de febrero de 1997, p. diecisis.
4Sidney Millman, ed,. Una historia de Ingeniera y Ciencia en el Bell System:
Com-
Ciencias municacin (1925-1980) (AT & T Bell Telephone Laboratories, 1984),
p. 76.
Pgina 76
Captulo 4 a la fase digital: El 19605
79
ALAN OPPENHEIM:
.. si iba para identificar una especie de punto de vista [de la Op
penheim y libros de texto Schafer], dira lo siguiente: Una forma tradicional
que mucha gente considera el procesamiento de seal digital era como una
aproximacin
a analgico de procesamiento de seales. En el procesamiento de seal
analgica; la matemtica implica
derivadas e integrales; realmente no se puede hacer eso en el equipo, por lo que
tener que aproximar. Cmo se puede aproximar la integral? Cmo AP-
prximo al derivado? El punto de vista que tom en el curso era de empezar
desde el principio, reconociendo que usted est hablando de procesamiento
discreto
seales de tiempo, y que su procedencia es un tema aparte, pero el Matemticas-
matem- de que no es una aproximacin a nada. Una vez que tenga las cosas
que son de tiempo discreto, entonces hay cosas que haces con ellos. Existen
Los ejemplos que he usado en el camino correcto desde el principio que
claramente
mostr que si se toma la otra vista le cuesta mucho, que hara tonto
cosas mediante la adopcin de sistemas analgicos y tratando de aproximarse a
ellas de forma digital y
usando eso como su procesamiento digital de seales. Yo dira que era un fuerte
componente del mismo. 1
LAWRENCE RABINER:
Ellos preguntaron, "Cmo puedo hacer que sea ms rpido, con mayor
precisin,
ms correctamente? Cmo puedo aprender realmente algo por lo que hago
mejor?" Ellos
todos fueron impulsados por aplicaciones especficas.
... Las aplicaciones nos llev a mirar el anlisis del espectro, se
nos llev a mirar las tcnicas de filtrado, que nos llev a mirar inter
polacin. Pero una vez que ests all, es un mundo completamente nuevo. Uno se
pregunta,
"Debo dejar en la pieza que necesito para resolver ese problema y luego
volver?"
porque entonces es una distraccin, "O es ms importante para ayudar a
establecer que
campo antes de volver?" Siempre tom la segunda tachuela. 2
1Alan Oppenheim entrevista de historia oral 28 de febrero de 1997, pp. 25-26.
2Lawrence Rabiner entrevista-historia oral 13 noviembre de 1996, pp. 10, 30.
Principios de diseo: Procesamiento de Seales y el medio ambiente. 66 Los filtros
digitales jugado
una parte en el anlisis de las seales de radar reflejadas desde Venus a principios
de 1960. 67
Para procesar los datos de un radar Doppler de impulsos, Herbert Groginsky y
George
Obras diseado y construido un procesador de seal FFT cableada en 1969. 68
PP (f;).
66S imon Haykin, "proceso de seales Radar" (IEEE Revista ASSP, vol 2 (1985), no 2,..
2-18).
67Robert precio en el Lincoln Lab utiliza un filtrado digital en el anlisis de las
seales de radar [Thomas
Kailath comunicacin personal 20 de enero de 1998].
68Herbert L. Groginsky y George A. Works "una transformada rpida de Fourier
tubera" (IEEE
Transacciones en Informtica, vol. 19 (1970), pp. 1015-1019).
Pgina 77
80
ENTREVISTADOR:
Djame ser seguro de entender. Usted est diciendo ahora que esto digitales
experimentacin se consider como el procesamiento de seales. Me estaba
preguntando si
la gente nunca hablaron de procesamiento de seales en el dominio analgico?
LAWRENCE RABINER:
El procesamiento de seales se convirti en un concepto digital.
ENTREVISTADOR:
As que la gente no habla de ello hasta que se ha dado cuenta de forma digital?
Rabiner:
S. De hecho, el procesamiento de la seal plazo es siempre sinnimo de
DSP- "procesamiento de seales digitales." Nunca se ha odo el trmino
analgico ASP- "
procesamiento de la seal ". Se desarroll a partir de la simulacin plazo. misma
simulacin es una dig-
ital modo de realizacin de un proceso anlogo, que es todo el concepto. Despus
de un tiempo
Se ha dado cuenta que podra haber una simulacin digital de un proceso
digital. Qu son
realmente hacer entonces? Slo ests teniendo una seal y que slo ests
procesarlo.
ENTREVISTADOR:
As que antes de eso, lo que ahora llamaramos tecno- procesamiento de seales
nicas estaban incrustadas en las tecnologas especficas, las aplicaciones
especficas en
que se realizaron, y nunca se juntaron como general o universal
campo de estudio.
Rabiner:
Esa es mi percepcin de ella, en su mayor parte. Hice un curso en el MIT
en el diseo de filtros. Lo llamaron Filter Design. Ellos no lo llaman filtro
analgico De-
firmar. Fue RLEs y redes pasivas. Nadie pens en llamar a ese analgico
procesamiento de la seal. 1
CHARLES RADER:
Ben [Gold] had been working with the TX-2 computer, which
was a very interesting machine in the history of computing. If you look at the
history of computing and computers as a kind of a map or a tree, where every
computer was the predecessor to one or several others, and so on, right up
the main trunk of that tree you'll find the TX-2 computer, and then it branches
from there. But, it was for its day, which was 1961-and of course it was sev-
eral years old then-quite an impressive machine. It had a significantly large
memory, about 64,000 words. They were 36 bit words. Actually 38, but some
of those bits were parity bits. It had built-in hardware index registers, and a
thin film memory, and a whole lot of other nice features. One of its most un-
usual characteristics was that it was accessible to the user. You could attach
equipment to it, and you accessed it directly rather than submitting your
punched card deck to a cabal of operators.2
1 Lawrence Rabiner oral-history interview 13 November 1996, pp. 8-9.

2Charles Rader oral-history interview 27 February 1997, p. 3.
Pgina 78
81
CHARLES RADER:
But the major thing is that if you wanted to try something out
that involved changing a filter, changing the bandwidths, changing center fre-
quency, etc., you had to build another one. That took a few weeks at best. Asi
que
speech research, in effect, was being hampered by the need to actually build
the hardware to try it out. And one of the thoughts that we had was if we could
simulate the vocoder, instead of building it, it would speed the pace of re-
search, because we thought, "Gee, programming things doesn't take any
time." Nowadays people understawj that that isn't true. 1
CHARLES RADER:
The butterfly diagram was a huge breakthrough. It enabled
us to understand fast Fourier transform algorithms. It enabled us to explain it
to other people. And, I suppose, with a bit of a messianic streak, we did. Nosotros
organized a talk in which we both explained what we understood about the al-
gorithm. 2
CHARLES RADER:
We began to have the idea that we should publish. We wrote
a paper called "Digital Filter Design Techniques in Frequency Domain," which
was published in IEEE Proceedings. We also organized, along with AI Oppen-
heim and Tom Stockham, a two-week summer course at MIT on this new field
of digital signal processing. That was perhaps a turning point in my career, be-
cause we prepared extensive notes for the course. We had a thick stack of
notes, and we realized this could be the basis of a book. So we got permission
from the Laboratory, and we wrote the book. We thought it was just going to be
a simple matter of transferring the notes. It actually took about two years to go
from the notes to a book. The course was '67. The book came out in '69.
It was almost the first book to cover any of the material, with one very
significant exception. There was a book by a group of people from Bell Labs
called Systems Analysis by Digital Computer. The major authors were Jim
Kaiser and Frank Kuo. But it had a chapter on digital filters that Jim Kaiser
had authored. That chapter, in my opinion, deserves to be called the first book
that introduced any of this material. 3
CHARLES RADER:
In 1969, I thought digital signal processing theory had gone
as far as it was going to go for a while, and I changed fields completely. Talk
about a bad decision. I changed into a group that was doing communications
satellite work, and I became an assistant group leader of what was then called
Group 69. And I worked on a couple of communication satellites that were
launched in 1975, and are still working.4
1 Charles Rader oral-history interview 27 February 1997, p. 4.

3Charles Rader oral-history interview 27 February 1997, pp. 11-12.
Pgina 79
82
INTERVIEWER:
How did the digital revolution in signal processing effect the oil
exploration business?
ENDERS ROBINSON:
The thing that converted electrical engineering to digital
was the Fast Fourier Transform. Tukey had been looking for that for many
years-it wasn't just something he did instantly, he had been working on that
problema. Somehow he had this vision that we needed that Fast Fourier Trans-
form, and we did because convolution takes a lot of computer time in the time
domain, but if you use the Fast Fourier Transform, it's much faster. A lot of op-
erations were possible with the Fast Fourier Transform. Tukey knew that,
worked on it, and published it with Cooley in 1965. I think as soon as he did
that the electrical engineers went digital. That was a big transition in electrical
Ingenieria. The oil company people had switched a few years before be-
cause they could see reflections in marine records by deconvolution. 1
MANFRED SCHROEDER:
In 1964 or '65 we were beginning to buy smaller com-
puters for online research. Before that we used mainframe computers doing
batch processing, machines from IBM, General Electric, and so forth. Entonces
we began using small machines from DEC and Honeywell for online speech
and hearing research. Much of our batch processing was shifted to the
smaller machines that we owned.
Incidentally, we had a hard time getting these minicomputers. Bill Baker
said, "Look Manny, you know we spent all that money on the big machines,
and now you want all these smaller machines." I said, "But this is a completely
different thing." We demonstrated for the management the online use of these
small computers for speech research or whatever. I still remember Bill Baker
saying "Manny, we are grateful for your having insisted." That is, he saw that
you needed big computers and small computers. This was in 1964. We were
probably pioneers in the use of small computers. 2
1 Enders Robinson oral-history interview 6 March 1997, pp. 41-42.
The field of signal processing was, even in the 1960s, wider still.
There were studies of sound propagation in the ocean and of hydrophone
arrays.69 Digital filters were used with missile data. 70 A wide range of appli;
cations, often involving statistical, nonstationary, and multichannel data,
served to broaden the range of theory and techniques brought into the field
7oJ.FA Ormsby, "Design of numerical filters with applications to missile data processing"
(Journal of the ACM, vol. 8 (1961), pp. 440-466).
Pgina 80
83
MANFRED SCHROEDER:
That day in 1967 I was pacing up and down my office
[in the presence of Bishnu Atal], saying "We have to do something about the
vocoder speech quality." We had solved the pitch problem with the cepstrum
method, but it still didn't sound right like a real human voice. The trouble in a
vocoder was that we went on the basis of a fixed and inflexible
paradigm. Nosotros
needed to code speech so as to leave room for error. From this conversation
with Bishnu evolved the idea of predictive coding.
The idea is the following. As speech is being encoded, continuously
predict, on the basis of preceding samples, the subsequent speech; comparar
the prediction with the actual speech; and transmit the prediction error, which
is called prediction residual. In decoding, use of the same prediction algorithm
and knowledge of the prediction residual allow accurate reconstruction of the
actual speech.
Well, that idea, which we called adaptive predictive coding, turned out to
be absolutely marvelous. The idea was already in existence for television pic-
ture coding, but those coders were fixed. With speech we needed predictors
that changed with every speech sound, and to make that clear we called this
adaptive predictive coding, APe. We wrote a paper for the Bell System Tech-
nical Journal and presented the technique at Wescon, a major IEEE confer-
ence, in 1967. Some Japanese also discovered it, but happily somewhat later. 1
HANS SCHUESSLER:
The essential point was that the first people who used dig-
ital signal processing had been those who had to transmit speech.
INTERVIEWER:
The speech-processing people?
SCHUESSLER:
S. I think that was the reason that the people at Bell Labs-
this group of Flanagan's-became interested in signal processing as such,
since they saw a chance to improve the speech-processing business, to
transmit speech with higher efficiency. Really a very important point was that
this group-these people under Flanagan, Larry Rabiner especially-had the
time and support to push the whole area. They did a lot of ground-breaking
work, no doubt about that. 2
2Hans Wilhelm Schuessler oral-history interview 21 April 1997, p. 17.
of signal processing. For example, Wiener filtering theory was extended to
solve problems in missile tracking and guidance and in signal detection
with radar, sonar, and multipath communications. 71 Another example is
71See, por ejemplo, Robert Price, "la deteccin ptima de seales aleatorias en el
ruido, con una p ..
plicatura a scatter..multipath comunicacin I" (IRE Transacciones sobre Teora de la Informacin,
vol. 6 (1956), pp. 125-135), y Robert Price y Paul E. Green, Jr., "Una
comunicacin
tcnica para los canales de trayectorias mltiples" (Actas de la IRE, vol. 46 (1958), pp.
555-570).
Pgina 81
84
HANS SCHUESSLER:
Pero de vuelta en Alemania en el 64 hemos publicado un documento sobre la
clculo de la respuesta de impulso de las redes continuas utilizando el z-
transformacin. Y eso es slo el tipo de trabajo que se ha hecho ms o
menos al mismo tiempo en los Laboratorios Bell por Jim Kaiser. En ese
momento tenan una, con mucho,
programa ms amplio para simular sistemas continuos en un ordenador digital,
calculat-
ing la respuesta al impulso y todo eso. Pero estbamos siguiendo, al menos, la
misma
idea. Y lo que hicimos fue justo lo ves aqu, comenzando con una cascada de
con-
nection de las diferencias, que transforma este sistema continuo de segundo
orden
ms o menos en digital. Y entonces estamos en lo que llamamos ahora el
dominio IIR.
Y todas estas cosas se han hecho utilizando el z-transformacin y terminamos
con funciones de transferencia en el dominio z. 1
ENTREVISTADOR:
Podemos hablar de cmo se involucr con filtrado adaptativo?
BERNARD WIDROW:
... Estaba pensando en tratar de construir una mquina que sera
cerebral similar, pero slo iba a tomar mucho tiempo. Yo tena que hacer algo
que
tendra algn valor prctico, pero todava quera trabajar en ese campo. Asique
Tengo una idea, que se remonta a los sistemas de datos de muestra, de lo que hoy
llamaramos
un filtro digital. Un filtro digital tiene parmetros, tiene coeficientes. Tengo la
idea
para desarrollar un mecanismo que variar los coeficientes del filtro de modo que
el filtro podra mejorarse a s misma y convertirse en un filtro mejor. En un
sentido un filtro puede
aprender a mejorarse a s misma, actuando como un filtro, y me di cuenta de una
manera de hacer esto. yo
no tena un cerebro, pero que tena un filtro. No he tenido una serie de conceptos
de aprendizaje del cerebro
y aprender cosas sofisticadas; Tena un filtro de aprender cosas muy
simples. Pero
que estaba aprendiendo. Era algo que podramos hacer incluso en aquellos das, y
era
algo prctico y algo til y algo bueno para Engineer-
ING, y que era mi negocio.
... La aplicacin ms grande que existe en la actualidad es el mdem en su com-
putadora. Casi todos los mdems en el mundo utilizan varios filtros adaptativos,
y la mayora de
ellos se construyen como filtros FIR, y filtran la seal procedente de la tele-
lnea telefnica.
... Estbamos usando filtros adaptativos para otro propsito. Estbamos haciendo
antenas adaptables. Hemos publicado el primer artculo sobre antenas adaptativas
en 1967
en las Actas de la IEEE. Creo que fue diciembre de 1967. Ese documento
result ser un clsico de la cita, sin embargo, estuvo muy cerca de ser rechazado.
Tres revisores examinaron de la misma. Un crtico dijo que est bien para
publicar; otro
crtico tena un montn de problemas con l, pero dijo con un poco de fijacin
que podra ser
publicable-que era una especie de en la valla; y el tercer revisor realmente
cerr de golpe. El dijo: "Esto es slo una mala idea. Esto no va a funcionar. ,, 2
1Hans Wilhelm Schuessler entrevista de historia oral 21 de abril de 1997, p. 10.
2Bernard Widrow entrevista-historia oral 14 marzo de 1997, pp. 30, 47,51.
Pgina 82
85
BERNARD WIDROW:
Yo estaba ocupado con filtros adaptativos [en la dcada de 1950]. Asique
continuado en el Lincoln laboratorios hace el trabajo sobre filtrado adaptativo, y
con la com-
ordenadores que tenamos que podran simular un filtro adaptativo. Yo saba que
los ordenadores
iban a ser mejor, que iban a llegar ms rpido, pero no estaba think
ing acerca ms barato. No haba manera en ese momento uno podra anticipar
lo que sucedera, que iba a tener un ordenador personal, o que
usted puede tener un chip de procesamiento de seales que puede tener un chip
que cuesta unos pocos
dlares y se puede hacer la programacin digital adaptativo con l. A quin se le
imaginar una cosa as? Pero eso no era realmente lo que nos conduca. Era un
inters intelectual. Y en mi mente, era el principio. 1
JAMES COOLEY:
Esto [la actividad de las personas de procesamiento del habla y de la seal
en el MIT] fue la primera evidencia realmente impresionante para m de la
importancia de la
FFT. 2
GENE FRANTZ y Panos PAPAMICHALlS:
La FFT ha sido probablemente el ms
herramienta ampliamente utilizada en el procesamiento de la seal y su
desarrollo a mediados de los aos 60 del re-
aliado inici la revolucin DSP. 3
JACK DELLER:
Trate de imaginar la profesin de procesamiento de seal sin la
FFT. No es tan fcil es? ... En un poco ms de N log N tiempo, la FFT ha
ayudado
para crear un grupo de especialistas que, sin ella, podra ser, por ejemplo,
elctricamente
Los ingenieros de cal que en realidad saba algo de electricidad. Para ayudar a
desovar
an entire field of engineering: Now that's hyperl 4
1 Bernard Widrow oral-history interview 14 March 1997, pp. 39-40.
2James W. Cooley, "How the FFT gained acceptance" (IEEE Signal Processing
Maga-
zine, vol. 9 (1992), no. 1, pp. 10-13), p. 11.
3Gene Frantz and Panos Papamichalis, "Introduction to DSP solutions" (Texas
Instru-
ments Technical Journal, vol. 13 (1996), no. 2, pp. 5-16), p. 9.
41EEE Signal Processing Magazine, vol. 9 (1992), no. 1, p. 6.
the extension of Kalman filter theory for handling large or growing amounts
of data in control systems and signal...detection systems; in these and other
areas time...variant filters were needed. 72 Closed...form solutions often had to
nSee, for example, Rudolf E. Kalman, "A new approach to linear prediction and filtering
problems" (Journal of Basic Engineering, vol. 82 (1960), pp. 34-45).
Pgina 83
86
give way to recursive numerical (computer..based) algorithms and time..
variant implementations.73
Though the hi..fi movement reached its peak in the 1950s, the 1960s
were also exciting years for audiophiles. There was the beginning of stereo
FM broadcasting in the United States in 1961;74 the standards for stereo
FM, established by the National Stereophonic Radio Communications
Commission, were later adopted by other countries.75In 1963 Philips intro..
duced the audio cassette. In 1964 an MIT researcher, Amar G. Bose,
founded the Bose Corporation to exploit some of his patents in acoustics
and electronics, and the next year the young engineer, Ray Dolby, started
another company whose first product was a "SIN (Signal..to..Noise)
Stretcher".76 The year before Robert Moog began marketing a music syn..
thesizer, which was an analog device, and in 1968 Walter Carlos's
"Switched On Bach", which used a synthesizer, sold more copies than any
classical recording before it.77Electret microphones were commercialized by
Sony Corporation in 1968.78And in 1970 quadraphonic records began to
be sold in the United States.
The 1960s saw major changes in the professional organization for sig..
nal processing. First of all, on 1 January 1963 the AlEE and the IRE merged
to form the Institute of Electrical and Electronics Engineers, and so came
into being the IEEE Professional Group on Audio. Second, the Group on
Audio expanded its interests, notably in the direction of electroacoustics,
speech processing, and the newly recognized field of digital signal process..
En g. There was, thus, a name change in 1965 to Professional Group on
Audio and Electroacoustics, and in the late 1960s the creation of technical
committees, within the Group, on speech communications and digital sig..
nal processing. There were also technical committees on electroacoustics
73See, for example, Thomas Kailath, "The innovations approach to detection and
estimation
theory" (Proceedings of the IEEE, voL 58 (1970), pp. 680-695).
74Mark Kahrs, "Professional and consumer gear: hardware & software"
(IEEE Signal Process~
ing Magazine, voL 14 (1997), no. 5, pp. 51-57).
76RLE Currents, voL 8 (1996), no. 1, pp. 8-9, and IEEE Spectrum, voL 25 (1988), no. 11,
p.
57.
77Steven Lubar, Infoculture: The Smithsonian Book of Information Age Inventions (Boston:
Houghton Mifflin, 1993), p. 191.
78Gerhard M. Sessler, "What's new in electroacoustic transducers" (IEEE ASSP
Magazine,
voL 1 (1984),no. 3,pp. 3-11).
Pgina 84
87
and underwater sound. 79 In this decade engineers who dealt exclusively
with audio tended to associate with the Audio Engineering Society rather
than the IEEE Group. As before, and since, some aspects of signal process,
ing were concerns of other organizations. For example, the IRE Professional
Group on Instrumentation, which evolved into the IEEE Instrumentation
and Measurement Society, dealt with "the electrical generation, processing,
recording, reproduction and display of data in digital form".8o Other IRE
Professional Groups concerned with signal processing were those for Auto,
matic Control, Circuit Theory, and Information Theory.
For the future of the Group on Audio and Electroacoustics, and for
the field of signal processing generally, a most significant event was the em,
brace by that professional group of the new possibilities opened up, in com,
munications, control, instrumentation, and other areas, by the fast Fourier
transform. The person within the Group who did most to bring this about
was IBM researcher William Lang, who helped to organize workshops and
conferences, to arrange for publications, to attract the speech processing
community, and to change the name of the Group.81 As we will see in later
chapters, the Group, which became the IEEE Signal Processing Society, has
continued its work in expanding the realm of signal processing.
79Rabiner op. cit.
8001iver op. cit., p. 1170.
81Rabiner op. cit.
Pgina 85
Chapter 5
DSP Comes of Age:
The 1970s
Woody Allen's Annie Hall
The Godfather with Marlon Branda
Saturday Night Fever
"All in the Family"
"M*A*S*H"
"Saturday Night Live"
Bruce Springsteen
disco music
punk rock
Solzhenitsyn's Gulag Archipelago
Richard Adams's Watership Down
89
Pgina 86
S ome historians have seen in the 1970s the beginning of the Third In..
dustrial Revolution. The First Industrial Revolution, characterized by
steam power and factory production, began in the late 18th century.
The Second Industrial Revolution, characterized by electric power, the in..
temal combustion engine, and telegraph and telephone communications,
began in the late 19th century. The defining technology of the Third In..
dustrial Revolution is the computer, particularly the microprocessor. 1 Be..
sides in computer technology itself, there were revolutionary changes in
communications, instrumentation, and control systems, and all of these
areas became much more important economically.
In the exploitation of the technologies of the First and Second Indus..
trial Revolutions, the Soviet bloc had been able to compete with the West
{though users of Soviet cars and telephones might have their doubts}. Con
the new computer and communications technologies, however, the Com..
munist states fell further and further behind. 2 Indeed, failure to participate
fully in the Third Industrial Revolution was a contributing cause of Com..
munist defeat in the Cold War. 3
I).AS Grenville, A History of the World in the Twentieth Century (Cambridge, MA: Harvard
University Press, 1994), p. 927; and Walter B. Wriston, "Bits, bytes, and
diplomacy" (Foreign
Affairs, vol. 76 (1997), no. 5, pp. 172-182).
2Grenville op. cit., p. 927.
3The historian Richard Bessel has written, "Indeed, one reason for the
destabilization of
eastern European socialist regimes during the 1980s was that their populations
had come to
expect the fruits of a consumer society which the state...socialist regimes were
unable to de...
liver." [Richard Bessel, "European society in the Twentieth Century." In TCW
Blanning,
ed., The Oxford Illustrated History of Modern Europe (Oxford: Oxford University Press, 1996),
pp. 231-254; quotation on p. 241.]
Pgina 87
Chapter 5 DSP Comes ofAge: The 1970s
91
Many people would remember the 1970s for detente (relaxation of ten..
sions), highlighted by trips by President Nixon to China and the Soviet
Union in 1972 and the Strategic Arms Limitation Treaty the same year.
Even more memorable were the 1972 Watergate break..in (leading to
Nixon's resignation in 1974); the 1973 oil embargo by the Arab oil..produc..
ing states (causing an energy crisis and contributing to the worldwide infla..
tion of the mid 1970s); the end of the Vietnam war in 1975; the overthrow
of the Shah of Iran in 1979; and the Soviet invasion of Afghanistan the
mismo ao. Prominent were environmental concerns (the Club of Rome
published The Limits of Growth in 1972) and the women's movement (Ms
magazine first appeared in 1972).
Engineers may remember new aircraft (such as the Boeing 747, the
Airbus, and the Concorde), the first pocket electronic..calculators,4 the
blackout of New York City in 1977, and the beginning of bar codes and
electronic scanners in supermarkets. In 1975 came the first personal com..
puter, the Altair 8800, sold in kit form, and an assembled computer, the
Apple II, was offered in 1977. In 1975 Sony and JVC began marketing
video cassette recorders (JVC in VHS format, Sony in Betamax), and at
about the same time Citizens Band radio burst into popularity.5 The 1970s
was the first decade of video games (Pong appeared in 1972), of word pro..
cessing (on typewriters with the capability of storing text), and of ATMs
(automatic teller machines). It was also a time when many people thought
the world was changing too fast, as suggested by the popularity of Alvin
Toffler's Future Shock (1970), which discussed the "bombardment of the
senses" and "information overload".6
In the 1970s consumers began to be aware of digital signal processing. En
Japan, the recordings of Nippon Columbia began to be digitally mastered in
1972. 7 The same year in Britain, the BBC began using PCM for high..quality
41t was the HP.. 35, the first with exponentiation and trig..function keys, that sold the engi..
neering community on pocket calculators; Hewlett..Packard sold more
than 300,000 of these
in the three years following its marketing launch in 1972 [IEEE Spectrum, vol. 25 (1988),
no. 11, p. 128]. When calculators with such capabilities became inexpensive, the
slide rule
became history.
5Citizens Band radio started in the United States in 1958, but first came to wide
public at..
tention with the gasoline shortage and truckers' strike of 1974. Suddenly millions of
Ameri..
cans wanted to have CB radio in their homes, cars, or boats.
In 1976 approximately 11 mil..
lion units were sold. [The us Consumer Electronics Industry in Review: 94 Edition
(Washington, DC: Electronic Industries Association, 1994), p. 20.]
6Alvin Toffler, Future'Shock (New York: Random House, 1970).
7Peter J. Bloom, "High..quality digital audio in the entertainment industry: an
overview of
achievements and challenges" (IEEE ASSP Magazine, vol. 2 (1985), no. 4, pp. 2-25).
Pgina 88
92
FIGURE 1. The Speak & Spell toy was introduced by Texas Instruments in 1978.
(Texas Instruments photo reproduced by permission.)
sound distribution for radio and television;8 and in its studios it began using an
8,track digital audio recorder with error correction. 9 In 1975 Tom Stockham
showed how DSP could improve historical recordings of Enrico Caruso,10 and
digitally restored recordings began to appear the following year. ll In 1978
Texas Instruments introduced a toy called Speak & Spell. 12 (See Figure 1.) It
taught a child to spell by pronouncing a word and indicating whether an at,
tempted spelling was correct. Two things made the toy practical: an efficient
8Bloom op. cit.
9Guy M. McNally, "Digital audio in broadcasting" (IEEE ASSP Magazine, vol. 2 (1985), no.
4, pp. 26-44). Professional audio equipment went digital before the consumer
products: in
1975 real..time digital reverberation systems became available, and in 1977 several profes..
sional digital audio recorders were being sold [Bloom op. ciLl.
lOThomas G. Stockham, Thomas M. Cannon, and Robert B. Ingebretsen, "Blind
deconvolu..
tion through digital signal processing" (Proceedings of the IEEE, vol. 63 (1975), pp.
678-692).
llBloom op. cit.
12'Speak y encanto' es una marca comercial de Texas Instruments Incorporated.
Pgina 89
Captulo 5 DSP viene ofAge: La dcada de 1970
93
MAURICE BELLANGER:
La idea de filtrado de mltiples velocidades vino de nuestro trabajo en el delta
modulacin. Dado que nuestro banco de canales PCM tena un filtro digital por
canal, nos
tenido que reducir al mnimo el nmero de multiplicaciones. El multiplicador fue
el ms
rea consume operacin. Una segunda razn es que, con el fin de realizar fil-
tering en ese contexto, tuvimos que partir de una alta frecuencia de muestreo. La
ltima mues-
pling tasa es la tasa regular de PCM, 8 kHz, y tuvimos que empezar de al menos
32 kilohertz. As que tratamos de optimizar la conversin de 32 kilohercios a 8
kilo
hertz el uso de filtros digitales, y ah es donde el concepto de mltiples
velocidades entr.
Nos dimos cuenta de que dos filtros de media banda reducen drsticamente el
nmero de multi-
alicates, y pensaron que el concepto podra aplicarse en diferentes
campos. Nosotros
dio algunas presentaciones del concepto dentro de Philips, y se utiliza en
otras reas, no slo para la reduccin de frecuencia de muestreo, sino tambin
para la Tasa de muestreo in-
pliegue, que fue llamada decimacin e interpolacin.
ENTREVISTADOR:
Por qu tiene la tasa de muestreo inicial ms alto?
BELLANGER:
Debido a lo que queramos hacer era un filtro de banda telefnica
desde 0 Hz a 3400 Hz. Pero, como es sabido, un filtro digital es un sis-
muestreado
TEM, por lo que tiene una periodicidad de frecuencia. Y despus de la banda
base, que es lo
que realmente quiere mantener con filtros digitales, tenemos bandas de imgenes
alrededor de todo el
mltiplos de la frecuencia de muestreo, que tiene que ser canceladas por fil-
analgica
tros en todo caso, ya que la salida tiene que ser analgica. Pero el mayor de estos
im-
edades, ms fcil es para reprimirlos. Nuestro objetivo era ser capaz de
deshacerse de
esas bandas de imagen con slo resistencias y condensadores no-bobina, sin
filtro-severa
En g. Haciendo que requiere al menos 32 kilohertz de sobre-muestreo. Sesenta y
cuatro sera
Probablemente han sido mejor, pero era demasiado caro, as que elegimos el 32. 1
ENTREVISTADOR:
El desarrollo tiene hardware influy en la direccin
de los algoritmos FFT?
JAMES COOLEY:
Seguro. Se dio un incentivo para el diseo de estos simpticos
algoritmos de planificacin para romper la FFT en bloques y la programacin de
la
bloques de clculo. documentos enteros se han escrito slo en la programacin
de datos a travs de almacenamiento jerrquico. 2
1Maurice Bellanger entrevista-22 de la historia oral de abril de 1997 p. 5.
2 James Cooley-entrevista de historia oral 11 de marzo de 1997 p. 10.
algoritmo para la sntesis de voz y los circuitos integrados de bajo costo para
llevar a cabo
el algoritmo. Contena un chip speech..synthesis, un chip de memoria read..only,
generadores de tono, generadores de ruido y filtros elctricos variables. discurso
fue
Recreated, no sintetizado de novo; las 165 palabras que el juguete poda
pronunciar
haba sido hablado por un ser humano y luego codificada por LPC (predictiva
lineal
Pgina 90
94
Habla
J \ J \ -I
Sampler
TRANSMISOR
Muestra
diferencias
F~ '-------'
Las predicciones
Los bits transmitidos
10010110
Cosa anloga
Convertidor
, .. J -, -, L ,.
D/A
Los bits transmitidos
10010110
RECEPTOR
diferencias
Las muestras del habla
.JJ. .. RL-
Habla
FIGURA 2. La idea de ADPCM (diferencial adaptativa Pulse Code Modulation)
es
colocar predictores idnticos en el transmisor y el receptor, de modo que todo lo
que necesita ser enviada, en
Para recrear el discurso original, son las diferencias entre el discurso predicho
las muestras y las muestras reales. (Reproducida despus figura en la pg. 247 de
Robert W. Lucky,
Sueos de Silicio: Informacin, el hombre y la mquina (Nueva York: St. Martin
Press, 1989)).
codificacin) para el almacenamiento eficiente. 13 (Speech sntesis de novo se
realiz por
Kurzweil mquina de lectura, un dispositivo mucho ms sofisticado construido
en 1976 por lo
que las personas ciegas pudieran leer el texto ordinario.)
Entre los muchos avances en el procesamiento del habla durante esta dcada
debe mencionarse diferencial modulacin de cdigo de pulso adaptativo
(ADPCM), sugerido por James Flanagan en 1973. 14 La idea era explotar
the fact that a small sample of speech can be predicted fairly accurately
from preceding samples: place identical predictors at the transmitter and
the receiver, and send only the difference between the actual sample and
the prediction. (See Figure 2.) In 1984 ADPCM was adopted as a new stan..
dard for encoding speech; it could, in 32 kbps, achieve the same quality of
13Robert W. Lucky, Silicon Dreams: Information, Man, and Machine (New York: St. Martin's
Press, 1989), pp. 202, 256.
14Sidney Millman, ed., A History of Engineering and Science in the Bell System: Communication
Sciences (1925-1980) (AT&T Bell Telephone Laboratories, 1984), pp. 115-117.
Pgina 91
95
ALFRED FETTWEIS:
How can we do something in the digital domain, which is
the equivalent of these resonant transfer filters? I tried first to do it on the
basis of voltages and currents, and then realized that this is not
feasible. Solamente
with mediocre properties was it feasible. So I had to do something different. yo
therefore tried to directly convert the equations of resonant transfer filters,
thus to transpose them into equations that would be realizable by digital algo-
rithms involving multiplications, additions, and delays, that is, the kind of oper-
ations commonly available in digital signal processing. And that's how I dis-
covered these wave digital filters. I first obtained them indeed by means of the
resonant transfer. I never published how I did that. It was very much more
complicated than the approach I did publish. I did not even mention in the pa-
pers-the first papers certainly not-that I had obtained the results from reso-
nant transfer. But, in fact, I did. So the resonant transfer work was fundamen-
tal for getting to this signal processing method. 1
ALFRED FETTWEIS:
.. Digital signal processing, as opposed to analog signal
processing, requires sequential ordering. You must carry out operations in se-
quence. You must be able to first carry out one operation, and the next one,
y as. This is very different in an analog circuit, where the Kirchhoff equa-
tions are all continuously satisfied, so to speak. There is no sequential order-
ing involved. In digital signal processing, however, you have an algorithm, by
definition computable, so you must be able to assign a sequential order in
which the operations have to be carried out. I realized this only later, I must
confess, but that is really basic for these wave digital ideas. You have on the
one hand passivity-that is, you can carry over the passivity of analog circuits.
On the other hand, you have sequential ordering due to the fact that we base
the analogy, not on voltages and currents, but on waves. And therefore you
get sequential operations and satisfy the computability requirement. Estas
are really the two important aspects: on the one hand computability, and on
the other passivity and losslessness, which then give you the excellent sensi-
tivity and robustness properties.
... I was quite confident that I could get access to such a Lyapunov
function and therefore could also solve the stability and, more generally, ro-
bustness problem. I had to give the filters a name. They're digital filters, but
they were based on the wave concept. That's why I called them wave digital
filters. 2
Pgina 92
96
Chapter 5 DSP Comes of Age: The 1970s
speech as the earlier standard could in 64 kbps.15 In 1976 Ronald
Crochiere, Susan A. Webber, and Flanagan showed that one could achieve
moderate savings in digital rate through subband coding (SBC), which di,
vided the signal spectrum into bands and adaptively quantized each inde,
pendently.16 In 1978 Larry Rabiner and Ron Schafer published the highly
influential Digital Processing of Speech Signals, which showed how DSP per,
vaded the field of speech processing.17
The use of satellites for telephony increased the need for reducing
echo. The older technique, echo suppression, was replaced by echo cancel,
lation, a technique that, instead of attenuating the echo, synthesizes it and
subtracts it from the total signal. (See Figure 3.) Proposed by M. Mohan
Sondhi in 1967, the process became commercially feasible with large,scale
integration technologies of the late 1970s. The first single,chip echo,can,
celer was designed by Donald L. Duttweiler in 1979.18
Speech processing researchers continued to innovate in digital,filter
design, but the field was enlarged also by the work of researchers concerned
with other types of signals. In 1969 Alfred Fettweis introduced a new type
of digital filter, called the wave digital filter.19A seminal advance was the
algorithm created by James McClellan and Thomas Parks for the design of
equiripple finite,impulse response (FIR) filters; Larry Rabiner further devel,
oped the Parks,McClellan algorithm and applied it to a number of different
types of filter. 20 Sidney Burrus and Thomas Parks showed how filters could
15Lucky op. cit., p. 245.
17Lawrence R. Rabiner and Ronald W. Schafer, Digital Processing of Speech Signals (Engle..
wood Cliffs, NJ: Prentice..Hall, 1978).
18Charles WK Gritton and David W. Lin, "Echo cancellation algorithms"
(IEEE ASSP
Magazine, vol. 1 (1984), no. 2, pp. 30-38), y Shoji Makino, "Acoustic cancelacin de eco"
(IEEE Revista Signal Processing, vol. 14 (1997), no. 5, pp. 39-41).
19El papel primero full..length en el filtro digital de onda fue Alfred Fettweis,
"Los filtros digitales re ..
lated a redes de filtros clsicos" (Archiv Elektrische piel Obertragung, vol. 25 (1971), pp.
79-89), que fue reimpreso en el Comit de Procesamiento Digital de Seales, eds., Papeles seleccionados en
Procesamiento Digital de Seales II (Nueva York: IEEE Press, 1976), pp. 510-524. Un tutorial sobre
papel
el sujeto es Alfred Fettweis, "filtros digitales Wave: teora y prctica" (Actas de la
IEEE, vol. 74 (1986), pp. 270-327).
2'Thomas Parques y W. James H. McClellan, "Chebyshev aproximacin para no
recursiva digi ..
filtros Tal con fase lineal"(IEEE Transacciones sobre Circuit Theory, vol. 19 (1972), pp. 189-194);
James H. McClellan, Thomas W. Parks, y Lawrence R. Rabiner, "Un programa
de ordenador para
el diseo de filtros digitales FIR ptimo de fase lineal"(IEEE Transacciones en audio y
Electroa ..
coustics, vol. 21 (1973), pp. 506-526); y Millman op. cit., p. 112. Un filtro digital cuya salida
slo depende de los valores de entrada anteriores se llama un filtro de respuesta
de impulso finito; en una infinita
filtro de respuesta de impulso, la salida depende de los valores de salida
anteriores, as como valores de entrada.
Pgina 93
97
telfono 1
trayecto de eco
+
ECO
"Mary Jane tena
COCIDO un cordero"
1 --------- + - 1 Cancelador de eco
"Mara tena un cordero"

telfono 2
+
FIGURA 3. El cancelador de eco, colocado en el extremo cercano de la lnea, no
bloquea
el eco, ni atenuarla. En su lugar, se sintetiza el eco y la resta, lo que hace
la seal resultante libre de eco. (Reproducida despus figura en la pg. 31 de
Charles WK
Gritton y David W. Lin, "Echo algoritmos de cancelacin" (IEEE Revista
ASSP, vol.
1 (1984), no. 2, pp. 30-38).)
ser diseado de acuerdo a las prescripciones en el tiempo domain.21In 1976 Alain
Croisier, Daniel Esteban, y Claude Galand introdujo espejo de cuadratura
filtros (QMF) 0,22 Sidney Darlington y Maurice Bellanger fue pionera mul ..
tirate filters.23Bede Liu, Toyohisa Kaneko, Alan Oppenheim, Hans Wil ..
timn Schuessler, Clifford Weinstein, y otros hechos importantes contribu ..
2IC. Sidney Burrus y Thomas W. Parks, "Diseo de dominio Tiempo de filtros
digitales recursivos"
(IEEE Transacciones sobre Audio y Electroacstica, vol. 18 (1970), pp. 137-141).
22Perifl: kolam P. Vaidyanathan ", bancos de filtros de espejo en cuadratura,
extensiones M..band y por ..
tcnicas fect..reconstruction" (IEEE Revista ASSP, vol. 4 (1987), no. 3, pp. 4-20).
23Sidney Darlington, "En moduladores single..sideband
digital" (IEEE Transacciones en Circuito
Theory, vol. 17 (1970), pp. 409-414), y Maurice Bellanger ", la tasa de Computacin y stor ..
estimacin de la edad en el filtrado digital de mltiples frecuencias con filtros
half..band" (IEEE Transacciones sobre
Acoustics, Speech, and Signal Processing, vol. 25 (1977), pp. 344-346).
Pgina 94
98
ciones en el anlisis de la exactitud de la tecnologa digital filters.24Thomas
Huang fue pionera
en el desarrollo de filtros para una imagen processing.25State ... mtodos
espaciales y relacionados
Se desarrollaron tcnicas matemticas, que se introdujeron ms tarde en
campos tales como el diseo de filtros, procesamiento de matriz, procesamiento
de imgenes, y adaptativa
filtering.26FFT theory was extended to finite fields and used in areas such as
coding theory.27
Work on digital filters and advanced DSP technologies was stimulated
and disseminated by the second, third, and fourth Arden House Work...
shops, held in 1970, 1972, and 1974 respectively. These workshops at...
tracted researchers from around the world, and the proceedings made up
three special issues of the transactions of the IEEE Group on Audio and
Electroacoustics.28
24See, for example, Bede Liu, "Effect of finite word length on the accuracy of
digital filters-
a review" (IEEE Transactions on Circuit Theory, vol. 18 (1971), pp. 670-677); Bede Liu and
Toyohisa Kaneko, "Error analysis of digital filters realized with floating..point
arithmetic"
(Proceedings of the IEEE, vol. 57 (1969), pp. 1735-1747); Alan V. Oppenheim and Clifford
J. Weinstein, "Effects of finite register length in digital filtering and the fast
Fourier trans..
form" (Proceedings of the IEEE, vol. 60 (1972), pp. 957-976); and Hans Wilhelm Schuessler,
"On the approximation problem in the design of digital filters" (Proceedings of the Fifth
An..
nual Princeton Conference on Information Sciences and Systems 1971, pp. 54-
63). Other impor..
tant papers in this area were Paul M. Ebert, James E. Mazo, and Michael G.
Taylor, "Over..
flow oscillations in digital filters" (Bell System Technical ] oumal, vol. 48 (1969), pp.
2999-3020) and Leland B. Jackson, "Roundoff..noise analysis for fixed..point digital filters re..
alized in cascade or parallel form" (IEEE Transactions on Audio and Electroacoustics, vol. 18
(1970), pp. 107-122).
25Thomas S. Huang, "Stability of two..dimensional recursive
filters" (IEEE Transactions on
Audio and Electroacoustics, vol. 20 (1972), pp. 158-163); Thomas S. Huang, "Two..dimen..
sional windows" (IEEE Transactions on Audio and Electroacoustics, vol. 20 (1972)); and
Thomas S. Huang, James Burnett, and Andrew Deczky, "The importance of
phase in image
processing filters" (IEEE Transactions on Acoustics, Speech, and Signal Processing, vol. 23
(1975), pp. 529-542).
26See, for example, Alfred Bruckstein and Thomas Kailath, "An inverse
scattering frame..
work for several problems in signal processing" (IEEE ASSP Magazine, vol. 4 (1987), no. 1,
pp. 6-20); Patrick Dewilde, "Advanced digital filters" (Thomas Kailath,
ed., Modem Signal
Processing (Washington, DC: Hemisphere Publishing, 1985), pp. 169-209); and Richard A.
Roberts and Clifford T. Mullis, Digital Signal Processing (Reading, MA:
Addison..Wesley,
1987).
27See, for example, Richard E. Blahut, "Algebraic fields, signal processing, and
error control"
(Proceedings of the IEEE, vol. 73 (1985), pp. 874-893).
28Lawrence R. Rabiner, "The Acoustics, Speech, and Signal Processing Society-
A Histori..
cal Perspective" (IEEE ASSP Magazine, January 1984, pp. 4-10). The special issues of the
transactions appeared in June 1970, October 1972, and June 1975.
Pgina 95
Chapter 5 DSP Comes ofAge: The 19705
99
THOMAS HUANG:
At Purdue [in the mid 1970s] we looked into some of the
nonlinear filters, especially the so-called median filters. The median filter is for
reducing noise in the image. The conventional way is to replace each point by
the local average in the neighborhood to smooth out the noise. But it's not
very effective when the noise is a spike type-salt and pepper-which is very
common in digital transmission. It turns out that the median filter is much bet-
ter. For a given point, you take the neighborhood around it, and instead of re-
placing the middle point by the mean, you replace it with the median gray
nivel. It takes out spikes very easily and also has the nice property of keeping
edges sharp. If you use the mean, the edges smear. We looked into this and
found a very efficient algorithm. It became very popular; many of the software
packages today use this algorithm. 1
FUMITADA ITAKURA:
I continued my hobby of visiting mathematics libraries and
accidentally found an interesting paper which transformed the autocorrelation
funcin. It didn't say autocorrelation at all, but that a positive definite function
could be expanded using line-spectrum type of transformation. The language
was completely mathematical. But if I interpreted it in my engineering ap-
proach, the autocorrelation function could be expanded using minimum line-
spectrum and the frequency and amplitude combination. That was the so-
called LSP [line spectrum pair] theory. So I was quite lucky to find good
mathematics-a very good mathematician paved the way for the speech
scientist. 2
2Fumitada Itakura oral-history interview 22 April 1997, p. 20.
The concepts of probability theory and statistics had long been part of
signal processing, as, for example, in Bernard Widrow's work in the 1950s
on quantization noise. 29 Throughout the 1960s, and later, Thomas Kailath
and Enders Robinson published many papers that applied statistical theory,
operator theory, and state..space techniques to a variety of different prob..
29Bemard Widrow, "A study of rough amplitude quantization by means of
Nyquist sampling
theory" (IRE Transactions on Circuit Theory, vol. CT~.J (1956), pp. 266-276), and Bernard
Widrow oral...history interview 14 March 1997. Even earlier, Norbert Wiener's
work on pre...
diction and filtering involved statistical concepts.
Pgina 96
100
lems.30In about 1970 the theory of Markov chains began to be applied to
problems in speech processing, as in the work on speech recognition by
James K. Baker and Fred Jelinek at IBM.3 ! By the mid 1980s this type of sta~
tistical pattern recognition had found a wide range of speech applications.32
The 1970s saw the initiation of a great deal of work on facilitating com~
munication between computers and people. At Bell Labs a system able to
speak messages from a stored vocabulary was developed, and it found use in
speaking the instructions for wiring telephone equipment (since the techni~
cian could work more rapidly and accurately if he did not need to look away
from the work to read instructions).33 (See Figure 4.) There was work on au~
tomatic speaker~recognition, either to identify speakers or to verify that a
voice belongs to the person alleged. In the 1970s systems capable of high suc~
cess rates were demonstrated, yet little practical use was found for automatic
speaker~recognition, a situation which continued into the 1990s.34
The challenge of automatic speech recognition was also taken up, the
hope being that people could dial a number by speaking it or even give oral
instructions to computers. A landmark advance was Fumitado Itakura's in~
troduction of a distance metric, between the utterance and templates of
speech within the computer, that became widely used in the field.35In the
early 1970s researchers at Lincoln Lab began a line of work destined to as..
sume great importance when they invented techniques for sending speech
over packet networks.36
30See, for example, Thomas Kailath, Lectures on Kalman and Wiener Filtering Theory (New
York: Springer...Yerlag, 1981); Thomas Kailath, Linear Systems (Englewood Cliffs, NJ:
Pren...
tice...Hall, 1980); Enders A. Robinson, Random Wavelets and Cybernetic Systems (London:
Charles Griffin and Company, 1962); and Enders A. Robinson, Statistical Communication and
Detection with Special Reference to Digital Data Processing of Radar and
Seismic Signals (London:
Charles Griffin and Company, 1967).
31Steve Young, "A review of large...vocabulary continuous...speech recognition"
(IEEE Signal
Processing Magazine, vol. 13 (1996), no. 5, pp. 45-57).
32Lawrence R. Rabiner and BH Juang, "An introduction to hidden Markov
models" (IEEE
ASSP Magazine, vol. 3 (1986), no. 1, pp. 4-16).
33Millman op. cit., pp. 129-131, and Manfred Schroeder oral...history
interview 2 August
1994, p. 35.
34Herbert Gish and Michael Schmidt, "Text...independent speaker recognition"
(IEEE Signal
Processing Magazine, vol. 11 (1994), no. 4, pp. 18-32), and Douglas O'Shaughnessy,
"Speaker recognition" (IEEE ASSP Magazine, vol. 3 (1986), no. 4, pp. 4-17).
36Eva C. Freeman, ed., MIT Lincoln Laboratory: Technology in the National Interest (Lexing...
ton, MA: MIT Lincoln Laboratory, 1995), p. 231. An important publication in this
area was
Gold's 1977 paper on "Digital speech networks" (Proceedings of the IEEE, vol. 65 (1977), pp.
1636-1658).
Pgina 97
101
FIGURE 4. This diagram shows the basic design of the Bell Labs computer
voice-
response system that was used for speaking the instructions for wiring telephone
equipo. (Redrawn after figure on p. 130 of Sidney Millman, ed., A History of En-
gineering and Science in the Bell System: Communication Sciences (1925-1980)
(AT&T Bell Telephone Laboratories, 1984).)
The 1970s were a watershed in the history of signal processing for the
development of signal,processing hardware. The decade opened with the
completion of what may be the first real,time DSP computer: the Lincoln
Fast Digital Processor (FDP).37 (See Figure 5.) A special,purpose, parallel,
processing computer, the FDP performed signal,processing tasks about a
hundred times as fast as the general,purpose computers of the time.38There
followed in 1974 the Lincoln Digital Voice Terminal, a computer built to
carry out a variety of speech,compression algorithms.39A more powerful
version, the Lincoln Digital Signal Processor (LDSP), was built in four
copies and remained the main research tool for the Signal Processing Group
at Lincoln Lab into the 1980s. 40
The future, however, belonged to single,chip signal processors. Through,
out the 1960s integrated,circuit techniques advanced, and, in the form of cal,
culators and watches, integrated circuits came into the hands and onto the
3700n Johnson, "Rewarding the pioneers" (IEEE Signal Processing Magazine, vol. 14 (1997),
no. 2, pp. 20-22).
38Freeman op. cit., p. 228. The FOP contained 10,000 separate integrated..circuits and
had a
multiply time of 900 nanoseconds.
4
0 Ben Gold personal communication 30 January 1998.
Pgina 98
102
FIGURE 5. This photograph shows the Lincoln Fast Digital Processor (FOP),
built
under Charles Rader's direction at Lincoln Lab. (Lincoln Lab photo reproduced
by
permission.)
wrists of a great many people. Going beyond a calculator chip (having hard..
wired functions), Intel produced the first (programmable) microprocessor
chip, the Intel 4004 in 1971. 41 The following year Intel introduced the 8008, a
microprocessor that operated on 8..bit, rather than 4.. bit, words. 42 Microproces..
sors were used in some DSP devices, such as the LPC vocoder, completed at
Lincoln Lab in 1977. 43 Because the usual signal..processing operations of filter..
ing and transforming are calculation..intensive, involving huge numbers of
multiplications, the introduction in 1976 by TRW of a 16..bit by 16..bit single..
chip multiplier stimulated the development of real..time DSP machines. 44
41At the time, the chip was called a 'micro..programmable computer on a chip';
the term 'mi..
croprocessor' came into use in 1972 [Ernest Braun and Stuart
Macdonald, Revolution in
Miniature: The History and Impact of Semiconductor Electronics, second edition
(Cambridge:
Cambridge University Press, 1982), p. 108]. The designer of Intel's first
microprocessor,
Marcian E. "Ted" Hoff, was a graduate student of signal..processing pioneer
Bernard Widrow;
Hoff was also the designer of the first Intel programmable DSP chip, the
Intel 2920 (men..
tioned below) [Widrow personal communication 2 February 1998].
4 2 Braun and Macdonald op. cit., pp. 108-109.
(IEEE Signal Process...
ing Magazine, vol. 14 (1997), no. 5, pp. 51-57). The TRW multiplier (the MPY.. 16) was in..
troduced in 1976 and advertised in the 4 July 1976 issue of Electronics [Earl
Swartzlander per..
sonal communication 30 January 1998].
Pgina 99
103
THOMAS KAILATH:
Through a student from industry who came to us, we got in-
terested in an area of antenna array processing. You have signals coming
from different directions. You have an antenna array, and you have got to sep-
arate these signals. The traditional methods for doing that are really equiva-
lent to taking the FFT of the data. It's a spatial FFT rather than a temporal
FFT, but it's similar. If you have sinusoidal waveforms in noise, you would
tend to get peaks of the FFT at the sinusoidal frequencies. Similarly, when
you take the spatial FFT of signals coming from different directions, you tend
to get peaks in those directions. This is non-parametric, and it doesn't take ac-
count of the fact, for example, that these are plane waves coming in, or that
there are only two or three of them.
The FFT doesn't care about those things. You have numbers, you take
the FFT. So, one of our students, Ralph Schmidt, said, "Well, even if you have
no noise, that method can't solve the problem exactly. But, if you add the in-
formation that you believe there are only, say, three plane waves coming in
from a few directions, then we can get an exact solution without noise. Y
with noise present, we can deduce an efficient algorithm that gives good esti-
mates of the direction." That idea launched a new field of high-resolution
model-based sensor array processing. And I had six or seven students work
in that area. It was timely because those were the days of SOl, and that was
one of the funding sources for this, because they were interested in determin-
ing the direction of incoming missiles. 1
1Thomas Kailath oral-history interview 13 March 1997, pp. 6-7.
Chips designed specifically for signal processing began to appear at the
end of the decade. The Speak & Spell toy contained a speech synthesis
chip (TMC0281),45 and, as mentioned above, in 1979 Bell Labs completed
a fully...integrated single chip adaptive echo...canceler.46In February 1979
Intel introduced a single chip DSP computer, the Intel 2920, but its lack of
a hardware multiplier limited its speed, and its 9...bit analog...to...digital and
digital...to...analog converters limited its accuracy.47 As the decade came to a
close, a number of companies-AMI, AT&T, Intel, Matsushita, Motorola,
Instruments
TechnicalJournal, vol. 13 (1996), no. 2, pp. 5-16).
46Gritton and Lin Ope cit.
4 7 Harvey G. Cragon, "The early days of the TMS320 family" (Texas Instruments Technical
Journal, vol. 13 (1996), no. 2, pp. 17-26), and Edward A. Lee, "Programmable DSP architec..
tures: part I" (IEEE ASSP Magazine, vol. 5 (1988), no. 4, pp. 4-19).
Pgina 100
104
Chapter 5 DSP Comes ofAge: The 19705
JAMES KAISER:
Hank McDonald came into my office-I remember this very
well-one Friday afternoon and he said, "Jim, look, I've got this idea." Then he
described to me in detail his idea on multiplexing. He showed me how he
wanted to do it, and he was doing serial arithmetic, not parallel arithmetic.
That had the advantage of using only one wire with a number of sequential
bits on that single line. That way, when you switch things around to do the
multiplexing you are only switching one line, which is a whole lot easier than
switching 12 lines all at exactly the same time for a 12-bit number. So Hank,
after he outlined his multiplexing scheme asked, "What's wrong with it?" I just
sat and looked at his scheme and finally said, "Hank, I can't see anything
wrong with the idea." We discussed it more, talked over what we thought the
properties of the design were, and I said, "I just cannot see any reason why
that shouldn't work. In fact, that structure will let us do both recursive filter im-
plementations and non-recursive filter implementation, to whatever order we
want. That is an absolutely beautiful structure."
So at that point, Hank started to do a detailed hardware design. He had
a number of projects also going on, however, so he enlisted the help of Leland
Jackson, a student we had with us who was beginning to work on a Ph.D. a
Stevens Institute of Technology. Jackson's work at Stevens was on analyzing
quantization effects in digital filters, both correlated noise (the limit cycle prob-
lem), and then uncorrelated noise-Le., how you sequence the computations
you are going to do to minimize noise when you are building a complicated fil-
ter. Under Hank's tutelage, Leland did the hard work of implementing Hank's
estructura. We completed that implementation in '67 and reported on it at the
IEEE Convention in New York. That was McDonald and Jackson, and me, but
I was by far the least important contributor to the team. I mean, Hank had the
basic multiplexing idea, Leland basically built it, and I helped with just a few of
the examples and acted as a sounding board. That was my part in it. That im-
plementation paper really got things going. 1
1James Kaiser oral-history interview 11 February 1997, pp. 16-17. The paper
was Le-
land B. Jackson, James F. Kaiser, and Henry S. McDonald, "An approach to the
imple-
mentation of digital filters" (IEEE Transactions on Audio and
Electroacoustics, vol. 16
(1968), pp. 413-421).
NEe, and Texas Instruments-were working energetically to design and
build single..chip DSPs.48 As we will see in the next chapter, success came
in the early 1980s.
4
8 Cragon op. cit. The AMI S2811 was announced in 1978, but not delivered until later; a
Bell Labs, the single...chip DSP1 was completed in 1979, but was not marketed
outside the
company [Lee op. ciL].
Pgina 101
105
JAMES KAISER:
It was at that point-1973 or so-I wrote another conference
papel. This was for the IEEE Symposium on Circuits and Systems, ISCAS
'74. It was a 4-page conference paper, and it talks about 1 sinh and it lays out
FIR design by the window function method. It's very complete, all in just one
little 4-page paper. That made a little more of an impact, but still, I think when
engineers read "Bessel functions," they don't want to know about it. Ellos
don't want to know about Bessel functions at all, unless they're in electromag-
netic theory. Then it's second nature to them, but otherwise, their attitude
about Bessel functions is, "I'll take your word for it, but don't bother me with
the details." So people don't use it, simply because they haven't taken that
extra little 5% effort that's required to understand what's going on. I call this
the "Bessel function syndrome.,,1
1 James Kaiser oral-history interview 11 February 1997, p. 43.
In the late 1960s and early 1970s, analog...to...digital converters
(ADCs) were getting faster and faster, the speed doubling roughly every
two years; and in the early 1970s ADCs and DACs (digital...to...analog con...
verters) with 16...bits of dynamic range were available. This further stimu...
lated the development of digital electronics in instruments, such as the dig...
ital oscilloscope, one of the first being the 1090A Explorer put on the
market in 1972 by Nicolet Instruments. 49 Another new technology of the
1970s, fiber optics (discussed below), also improved instruments, as fiber...
optic sensors came to be widely used in measuring devices.
5o
Large...scale...integrated electronics was certainly the most momentous
new technology of the 1970s, but there were two other new semiconductor
technologies of great value to signal processing. These were charge...coupled
devices (CCDs) and surface...acoustic...wave devices (SAW devices), both of
which appeared about 1970. CCDs were often used as optical sensors, as in
video cameras, but they were also used for time delay, as in echo generation
for music or ghost cancellation in television. 51 SAWs tum an electrical sig'"
49Karen Fitzgerald, "Digital scopes" (IEEE Spectrum, vol. 25 (1988), no. 11, pp. 82-86).
50Gerhard M. Sessler, "What's new in electroacoustic transducers" (IEEE
ASSP Magazine,
vol. 1 (1984), no. 3, pp. 3-11).
51Robert W. Broderson and Richard M. White, "New technologies for signal
processing"
(Science, vol. 195 (1977), pp. 1216-1222).
Pgina 102
106
Chapter 5 DSP Comes of Age: The 1970s
HANS GEORG MUSMANN:
At that time [in the mid 1970s] the problem was the
digitization of a video signal, which is required for realizing such a visual com-
munication system, and reducing the bit rate for moving images, which is al-
most three thousand times that of a speech signal. I always said our goal
would be to cut down the bit rate of a video signal to that of a speech signal.
Otherwise it's too expensive to use it.
INTERVIEWER:
The factor of three thousand is for television resolution?
MUSMANN:
S. Of course, you can use a smaller picture. That was also pro-
posed later on. But even if you take an image one fourth as large, then you'll
still have a factor of eight hundred or something like that. You can reduce the
frame frequency from fifty hertz to ten hertz. Then you come down to a hun-
dred times the rate of the speech signal. Nobody thought at that time that this
compression factor could be achieved by coding....
This indicated to me that if we wanted to create a visual communication
system, we would have to cut down the bit rate for visual information to that of
speech-or be close to it. Otherwise, nobody would be able to afford it. So we
studied some techniques. And, in 1979, two years after the facsimile work, we
demonstrated in the United States a transmission of moving images requiring
just 64 kilobits per second....
There was a big interest. The communications industry saw that it might
be possible to transmit moving images via speech channels. But they were
still hesitating, waiting to see if the problems with motion could be solved in
el futuro. We transmitted mainly those parts of an image which had changed,
and we took the other parts from the stored preceding image in the memory of
the receiver.
INTERVIEWER:
So the receiver had to have a frame memory?
MUSMANN:
Yes, and this memory was a problem at that time. The memories
of computers were not transistor memories, but magnetic-core memories.
Each core was one bit. I needed a frame memory: 400,000 picture elements
and 8 bits per picture element. The price was 250,000 D-marks!1
ALAN OPPENHEIM:
The reason why it was so significant was that, prior to the
Speak-&-Spell, you could basically think of digital signal processing as high-
end, funded largely by the military or by high-end industry like the seismic in-
dustry, because it was expensive to do. 2
1 Hans Georg Musmann oral-history interview 30 August 1994, pp. 16-18.
2Alan Oppenheim entrevista de historia oral 28 de febrero de 1997, p. 36.
Pgina 103
107
LAWRENCE RABINER:
Las personas y empresas saltaron en: TI, Motorola y
Rockwell, salt en muy rpidamente. Academia salt en muy rpidamente. AI
Op
penheim Schafer y Ron tena un libro por ah que se centr en la aca-
audiencia demia. Ben Gold y yo tuvimos otro libro por ah con ms de una
enfoque de ingeniera. Ben y Charlie [Rader] tenan este libro temprano, pero era,
como todos los libros de la primera, un poco demasiado pronto. Incluso en el '75,
cuando los libros salieron, yo
pensar en la primera lnea de la materia fue bastante bien desarrollada, y que era
la verdadera clave ....
Haba una enorme demanda acumulada de estas cosas. Puede salir y
hacer las cosas con una computadora, y despus de un tiempo se convirti en el
hardware digital y
luego se convirti en real. 1
CHARLES RADER:
Ahora bien, al mismo tiempo que haba encontrado una manera de expresar una
Transformada de Fourier como una convolucin, un colega mo, con el nombre
de Leo
Bluestein, que estaba entonces en Sylvania, pero que en realidad haba
compartido una oficina
conmigo unos aos antes, cuando estaba en el Lincoln Lab, lleg alrededor de
uno
da y dijo: "Tengo este resultado interesante." El dijo: "Yo puedo hacer de
Fourier
transformar, donde el nmero de puntos es un cuadrado perfecto, y tengo esta al-
gorithm." Y su algoritmo no era muy interesante en s mismo, sino que forma
parte de la
camino a travs de la explicacin del algoritmo, que haba hecho algunas
manipulaciones a
cambiar una transformada de Fourier en una convolucin, en todava otra
forma. Tena noth-
ing que ver con la teora de nmeros. Tena que ver con la multiplicacin de la
forma de onda de entrada
por lo que los ingenieros llamaran una forma de onda de chirrido. Era una
sinusoide cuya fre-
cuencia aumenta continuamente a medida que el tiempo avanza. Y luego, si lo
hara
de acuerdo despus de un-multiplicar la transformada de Fourier por un chirrido,
lo que
encontrado en el medio se le fueron convolucin con un chirrido. Por lo tanto, la
multiplicacin in-
CORRE PELIGRO, post-multiplicacin para deshacer, era otra manera de
convertir una Fourier
transformar a una convolucin.
Yo le dije, "Leo, podemos utilizar la FFT hacer circunvoluciones, por lo olvide
sobre su pequeo algoritmo inteligente para hacer la parte media. Vamos a usar
la FFT
para eso ". La ventaja de ello es que el algoritmo chirrido funcionara para
cualquier secuencia de longitud, y por lo que poda hacer cualquier secuencia de
longitud utilizando cualquier
longitud FFT. Eso fue transformar el llamado chirrido Z. 2
1 Lawrence Rabiner entrevista de historia oral 13 de noviembre de 1996 p. 20.
2Charles Rader entrevista de historia oral 27 de de febrero de 1997 pp. 15-16.
nal en una seal acstica y de vuelta; aplicaciones incluyen pulso ex ...
pansion, compresin de impulsos, y los filtros transversales. 52
52Broderson y negro op. cit., y Thomas Kailath, ed., procesamiento de seales del mdem (Wash
..
ington, DC: Hemisferio Publishing, 1985), pp. 374-375.
Pgina 104
108
Al igual que en la dcada anterior, las fotografas desde el espacio llamaron la
la atencin del pblico en la dcada de 1970. La Venera 7 sovitica envi el
primer pic ~
turas de la superficie de Venus en 1970; la Mariner 10 alcanz Mer ~
cury en 1973; y nos Viking 1 y 2, lanzado en 1975, enviados de vuelta de la
primera
imgenes de la superficie de Marte el ao siguiente. Tambin estaba el
mapeo de radar de Venus por la sonda Magellan Estados unidos en 1980.
Seal pro ~
cessing estaba involucrado en la codificacin, la reconstruccin, y la mejora de
estas imgenes, y jug un papel importante en todas las comunicaciones con los
satlites
y el espacio probes.53
En la dcada de 1970 el procesamiento de seal digital se aplica cada vez ms a
los radares
y las imgenes de sonar. Investigadores de ERIM y Lincoln laboratorio
mostraron que el DSP
era til en la formacin de imgenes de radar de objetivos, determinar la
velocidad de los objetivos, y
eliminando clutter.54In la dcada de 1970 sistemas acsticos submarinos
alcanzaron un alto
nivel de rendimiento, y el problema de separar y clasificar seales
came to the fore, as attention shifted from signal acquisition to signal process~
ing.55And there were important advances in geophysical signal processing,
such as Richard Lacoss's work on adaptive filters for seismic applications.56
Communications entered a new era in the 1970s with the develop~
ment of fiber~optic technology. Light, because of its high frequency,
promised great bandwidth for communications. Required, though, would be
a controllable and intense source of light and a low~loss transmission
medium. Rather suddenly, in about 1970, both of these became feasible:
Corning Glass demonstrated highly transparent fibers, and Bell Laborato~
ries demonstrated semiconductor lasers that could operate at room tempera~
ture. (See Figure 6.) In 1977 fiber~optic telephone systems were put into
service both by General Telephone & Electronics (in Santa Monica, Cali~
fomia) and AT&T (in Chicago).57 Just over a decade later the first fiber~
53Much of the communications was handled by the National Aeronautics and
Space Ad..
ministration's Deep Space Network, operated by the Jet Propulsion
Laboratory U ohn R.
Pierce and A. Michael Noll, Signals: The Science of Telecommunications (New York: Scien..
tific American Library, 1990), p. 198].
54Dale A. Ausherman, Adam Kozma, Jack L. Walker, Harrison M. Jones, and
Enrico C. Pog..
gio, "Developments in radar imaging" (IEEE Transactions in Aerospace and Electronic Systems,
vol. 20 (1984), pp. 363-400); and Freeman op. cit., p. 228.
55William S. Burdic, Underwater Acoustic Signal Analysis (Englewood Cliffs, NJ: Prentice..
Hall, 1984), p. 14.
56Richard Lacoss, "Data adaptive spectral analysis methods" (Geophysics, vol. 36
(1971), pp.
661-675).
57Trudy E. Bell, "Fiber optics" (IEEE Spectrum, vol. 25 (1988), no. 11, pp. 97-102):
Pgina 105
Voice
Transducer
109
Encoder
. f\-----.D------+I~
\...J
Elctrico
Voice Signal
Pulse-Code
Modulated Signal
Glass Fiber
Lente
Optical
Pulse
Electro-optic
Shutter
Laser
I ~------... Pulse-Code
Modulated
Seal
-----
Elctrico
Voice
Seal
Repeater
Station
Detector
Decoder
Transducer
Voice
FIGURE 6. Fiber-optic cable is typically employed as shown. The analog signal
is
converted to electrical pulses, which control a shutter in front of a laser beam di-
rected down the cable. En route the optical signal may be strengthened at repeater
stations. At the receiving end, the optical pulses are converted to electrical
pulses,
from which is recovered the original analog signal. (Redrawn after figure on p.
51 of
Britannica Yearbook of Science and the Future 1975 (Chicago: Encyclopaedia
Bri-
tannica, 1974).)
optic cable across the Atlantic went into service; its capacity exceeded all
previous cables combined. 58
Signal processing came to public attention also through music. los
early music synthesizers, such as Robert Moog's, were analog devices, with
oscillators, voltage regulators, filters, and other components to produce and
manipulate the waveforms that became music (or at least output). En el
1970s there appeared synthesizers capable of digital sampling, so that any
58Britannica Yearbook of Science and the Future 1991 (Chicago: Encyclopaedia Britannica,
1990), p. 325.
Pgina 106
110
CHARLES RADER:
In the mid-'70s there was a congressional committee on as-
sassinations, the assassination of President Kennedy and Martin Luther King,
among others. That committee was given testimony from some acoustics ex-
perts who believed that one of the police radio channels in Dallas had
recorded the gunshot sounds. And by a further analysis, they thought they
could prove that one of what they thought were four shots had originated not
from the school depository building, but from the so-called grassy knoll. ...
So, the Justice Department asked the National Academy of Engineering
to put together a team and look at this data and technique and judge its valid-
ity Along with several Nobel Prize winners, like Luis Alvarez, I was on this
committee. I played a significant role in proving the negative conclusion that
the sounds that were being identified as gunshots were in fact distorted
human speech. I knew what the speech was, and could associate it with a
time, about a whole minute after the assassination, because it was a Dallas
sheriff talking about moving men around to check out the possible site where
somebody had seen somebody suspicious. We were able to figure out a lot
about this horribly distorted data. 1
1 Charles Rader oral-history interview 27 February 1997, pp. 52-53. Another signal-pro-
cessing pioneer, Bernard Widrow, was involved with this investigation. The
Dallas po-
lice had recorded the sound from a police motorcycle with a stuck transmit-
switch that
was believed to have been located near the site of the assassination. A method
pub-
lished shortly before by B. Widrow et al. ("Adaptive noise cancelling: principles
and ap-
plications", Proceedings of the IEEE, vol. 63 (1975), pp. 1692-1716) was used to
re-
move the noise of the idling motor of the motorcycle. [Widrow personal
communication
2 February 1998.]
instrument, or any sound at all, could be recorded, modified, and replayed. 59
In the 1970s in Paris, IRCAM (Institut de Recherche et Coordination
AcoustiquefMusique) embarked on its mission of integrating traditional
musical media with the new medium provided by the computer. Giuseppe
Di Giugno at IRCAM designed a device, the 4X, for electronically modify...
ing sounds of traditional instruments. The 4X has been used in perfor...
mances, as, for example, of the 1981 composition Repons by Pierre Boulez
for "six instrumental soloists, chamber orchestra and real...time digital...signal
processors" .60
59Lubar op. cit., p. 193.
60Pierre Boulez and Andrew Gerzso, "Computers in music" (Scientific American, vol.
258
(1988), no. 4, pp. 44-50).
Pgina 107
111
MANFRED SCHROEDER:
We were working on speech synthesis because by tak-
ing speech apart and synthesizing it again at the other end we could com-
press its bandwidth by a factor of 5 or more.
... [But] it was not just bandwidth compression. Another reason we were
interested in speech synthesis was to read documents, as in reading ma-
chines for the blind. Optical scanners were already in existence in the '60s,
but it was not easy to get speech from such devices. We wanted to do that,
even if that was not immediate telephone business.
There was a beautiful application made by Western Electric, for the guys
who wire these complicated circuits. Here's a complicated circuit-chart. En
wiring it, they would often solder a wrong connection. So someone... wrote an
automatic program that translated the wiring chart into a code that we trans-
lated into spoken instructions. The wiring man at Western Electric used ear-
phones. He never had to turn his eyes off what he was doing. The earphones
would tell him to connect the green wire to terminal 47, that kind of thing. Ese
was, I think, the first application of synthetic speech within the Bell System. 1
HANS SCHUESSLER:
Very early Alfred Fettweis started this idea to transform an
analog selective system into a wave-digital filter and to obtain just a digital fil-
ter. The main point is that in that case the starting point is and has to be an
analog filter. But the knowledge about the design of analog selective systems
is more or less gone. And that's why he did not have so much success in the
States, as far as I see it. 2
ROBERT W. LUCKY:
One of the more practical engineers in the audience [at a
workshop about 20 years ago] stood to tell people how he had made a custom
VLSI chip that did some powerful signal-proce.ssing algorithms. En ese tiempo,
such a feat was very unusual and the people in the audience mostly held the
belief that VLSI chips were the exclusive province of a small number of de-
signers chained in the basement at Intel or somewhere like that.
I remember distinctly how the idea that an ordinary engineer could
make a custom chip for mathematical algorithms created an unwelcome
thought in the audience. People averted their glances and stirred in their
seats.... It wasn't that the mathematicians and engineers present at this work-
shop were actively against VLSI or anything like that. It was just that this new
capability introduced a new element into their world-an element that they did
not understand and one that they personally would be unlikely to master. 3
1 Manfred Schroeder oral-history interview 2 August 1994, p. 62.
2Hans Wilhelm Schuessler oral-history interview 21 April 1997, p. diecisis.
3Robert W. Lucky, Lucky Strike ... Una vez ms (Nueva York: IEEE
Press, 1993), p. 229.
Pgina 108
112
Captulo 5 DSP viene de edad: La dcada de 1970
El grupo IEEE sobre Audio y Electroacstica, que, como ya se ha
mencionado, se haba establecido un comit tcnico para el proceso de seal
digital ..
ING, hizo mucho para avanzar en la nueva tecnologa. Desde DSP era un campo
nuevo
y sus practicantes tenan bastante diferentes orgenes, la terminologa era
no fijo, con diferentes palabras que se usan para lo mismo y lo mismo
palabra que se usa de manera ligeramente diferente. El G..AE poda ayudar a
remediar esta
situacin con la publicacin en 1972 de recomendaciones detalladas para la
uso de unos 200 terms.61The G..AE trabaj con la recin creada
IEEE Press a publicar en la dcada de 1970 varios volmenes de reimpresiones-
muchas de las
importantes trabajos pioneros en DSP eran de otro modo difcilmente disponible
y
una bibliografa DSP en 1972, con una versin actualizada en 1975.62The G..AE
trabajado con el estndar IEEE Pulse para retener a cabo en 1979 un conjunto de
algoritmos DSP, ex ..
presionado en FORTRAN, que era ampliamente influentia1.63The IEEE Press
tambin
publicado en la dcada de 1970 varios volmenes de reimpresin de documentos
sobre el habla comu ..
nications.64
Otro caso notable de la dcada fue el inicio de la Inter ..
Conferencia Nacional sobre Acstica, Habla y Procesamiento de Seales
(ICASSP) en 1976, y ICASSP se ha celebrado cada ao desde entonces. dos
nombre
cambios fueron significativos: en 1974 el grupo cambi su nombre de Audio
y Electroacstica de Acstica, el habla y procesamiento de seales, y en
1976 alcanzaron IEEE Sociedad de estado y se convirti as en la acstica,
Discurso, y Procesamiento de Seales Society.65
A travs de estas y otras innumerables actividades, el procesamiento de la seal
era
emerge como una disciplina reconocida. Hubo una asociacin profesional, la
61Lawrence R. Rabiner, James W. Cooley, Howard D. Helms, Leland B.
Jackson, James F.
Kaiser, Charles M. Rader, Ronald W. Schafer, Kenneth Steiglitz, y Clifford J.
Weinstein,
"Terminologa en el procesamiento de seales digitales"
(IEEE Transacciones sobre Audio y Electroacstica,
vol. 20 (1972), pp. 322-337).
62The dos primeros volmenes de reimpresin, que fueron muy ampliamente
utilizados, eran Lawrence R. Rabiner y
Charles M. Rader, eds,. Procesamiento de Seal Digital (Nueva York: IEEE Press, 1972), y
Digital
Comit de Procesamiento de Seales, eds,. Papeles seleccionados en Digital Signal Processing II
(Nueva York:
IEEE Press, 1976). Los dos bibliografas eran Howard D. Helms y Lawrence R.
Rabiner,
Literatura en Procesamiento Digital de Seales: Terminologa y permutado
ndice Ttulo (Nueva York: IEEE
Press, 1972), y Howard D. Helms, James F. Kaiser, y Lawrence R. Rabiner, Literatura
en
Procesamiento Digital de Seales: El autor y ttulo ndice permutado, edicin revisada y
ampliada (Nueva
York: IEEE Press, 1975).
Comit 63Digital Procesamiento de Seales, eds,. Los programas de procesamiento de seales
digitales (Nueva York:
IEEE Press, 1979).
64Rabiner op. cit. (nota 28).
65Rabiner op. cit.
Pgina 109
113
RICHARD HAMMING:
As que me di a un amigo, Jim Kaiser (JF Kaiser), que fue
uno de los expertos mundiales en filtros digitales en ese momento, y sugiri que
debe detener su investigacin actual y escribir un libro sobre filtros-libro digital
ESCRITO
ing para resumir su trabajo fue una etapa natural en el desarrollo de una cien-
TIST. Despus de una cierta presin a la que accedi a escribir el libro, por lo
que se salv, por lo que
thought..But seguimiento de lo que estaba haciendo revel que estaba escribiendo
nada.
Para rescatar a mi plan ofrec, si me educar a ms de almuerzos en el
restaurante (se obtiene ms tiempo para pensar all que en la cafetera), para
ayudar a escribir
el libro de forma conjunta (sobre todo la primera parte), y podramos llamarlo
Kaiser y del jamn
ming. Convenido!
A medida que pasaba el tiempo que estaba recibiendo una buena educacin de l,
y me dieron mi
primera parte del libro va pero segua escribiendo nada. As que un da le dije:
"Si no se escribe ms vamos a terminar llamndolo Hamming y Kaiser." - y
el acepto. An ms tarde cuando haba completado sobre todo lo escrito y no
tena
Todava nada escrito, dije que poda gracias en el prlogo, pero debe ser
llama Hamming, y estuvo de acuerdo-y estamos siendo buenos amigos! 1
* Richard W. Hamming, el arte de hacer Ciencia e Ingeniera: aprender a aprender
(Amsterdam: Gordon y Breach, 1997), pp 164-165..
publicaciones, conferencias y talleres. Haba textos clsicos, en particular
Alan Oppenheim y Ronald Schafer procesamiento de seales digitales y
Lawrence Rabiner and Ben Gold's Theory and Application of Digital Signal
Processing, which formed the foundation for courses at major universities. 66
And both techniques and applications were being rapidly developed.
66Alan V. Oppenheim and Ronald W. Schafer, Digital Signal Processing (Englewood
Cliffs,
NJ: Prentice..Hall, 1975 and 1988), and Lawrence R. Rabiner and Ben
Gold, Theory and Ap..
plication of Digital Signal Processing (Englewood Cliffs, NJ: Prentice..Hall, 1975). Large num..
bers of DSP engineers first learned the subject from these two books; the
Oppenheim and
Schafer text was the standard introduction, while the Rabiner and Gold text had
more of an
engineering focus. Perhaps the first DSP course given at a university was one
given at MIT
in 1965; the material taught in this course formed a large part of Ben Gold and
Charlie
Rader's Digital Processing of Signals (New York: McGraw..Hill, 1969) (as stated on p. vii of
this book).
Pgina 110
Captulo 6
Etched in
Silicon:
The 1980s
Amadeus
ET: The Extra Terrestrial
Platoon
Rainman
A Room With a View
"The Cosby Show"
"Dallas"
"Hill Street Blues"
Bobby McFerrin's "Don't Worry, Be Happy"
In Search of Excellence by Thomas Peters and
Robert Waterman
Garrison Keillor's Lake Wobegon Days
115
Pgina 111
I n much of the world, the 1980s was the decade in which japanese con..
sumer electronics became commonplace: stereos, televisions, VCRs,
camcorders, CD players, video games, the Walkman, "boom boxes", per..
sonal computers, fax machines, cellular phones, and many other products
manufactured by Hitachi, JVC, Mitsubishi, NEC, Nintendo, Panasonic,
Sanyo, Sony, Toshiba, and other companies. The japanese economy had
grown vigorously after World War II: output increased 9.5 percent a year in
the 1950s and 10.5 percent a year in the 1960s. In the 1970s the Ministry of
International Trade and Industry (MITI) led a reorganization of the econ..
omy away from steel, ships, and chemicals and toward electronics and preci..
sion machinery. This helped bring about the "second economic miracle" of
the 1980s. 1 japanese companies pioneered many of the new electronic tech..
nologies, such as VCRs, videodisks, the Walkman, and compact disks, and
the first commercial cellular..phone system anywhere in the world was the
one put into service in Tokyo in 1979.
The burgeoning of consumer electronics owed much, of course, to the
appeal of new products, such as personal computers and camcorders, and
the new capabilities of already standard devices, such as stereo systems and
telfonos. 2 Tal vez lo ms importante fue el xito de los fabricantes
. lRichard Overy, ed, Hammond Atlas del siglo 20 (Londres: Times Books, 1996), pp.
156-157.
2 Otro factor fue la generalmente alta fiabilidad de los productos de consumo ... electrnica. En el
A mediados de 1990 una encuesta de Harris encontr que Sony fue la marca ms
respetada entre nosotros con ...
consumidores [Tiempo de revistas, 17 de noviembre de 1997, p. 56].
Pgina 112
Captulo 6 Grabado en silicio: La dcada de 1980
11 7
FIGURA 1. El IBM PC, introducida en 1981, se convirti rpidamente en el estndar de la industria.
(IBM foto reproducido con autorizacin.)
en la prestacin de ms por dlar del cliente (o yenes o marca). En los 1970s
y 1980, mientras que el ndice de precios al consumidor en los Estados Unidos se
triplicaron
(Es decir, bienes de costo promedio tres veces ms que al final de esta pe--
RIOD), el precio de la mayora de productos de electrnica de consumo, ya sea
permaneci con--
constante o cay. Su costo 1990 en trminos reales, a continuacin, fue uno -
tercio de la de
1970. 3
La dcada de 1980 fue tambin la dcada en la que los ordenadores personales se
convirtieron
ubicuo. Cuando se abri la dcada, mucha gente pensaba de personal
computadoras como algo slo para los aficionados. En 1981 IBM present sus
per--
Sonal equipo, y su sistema operativo de disco (DOS) se convirti rpidamente en
el
estndar en la industria. 4 (Ver Figura 1.) El siguiente ao apareci el primer IBM
- PC
3El US Consumer Electronics Industry in Review: 94 Edicin (Washington, DC: Electrnico
Asociacin de Industrias, 1994), pp. 16, 24. El precio de los aparatos de televisin se
redujo sustancialmente en este
periodo, a pesar de una mejora considerable en el producto.
4IBM dio una pequea empresa, Microsoft, el contrato para producir el sistema
operativo; Mio'
crosoft obtuvo una pieza adecuada del software de otra empresa pequea,
Software Seattle
Productos, por $ 30.000; era MSo'DOS, provisto de casi todos los ordenadores
personales de IBM
y mquinas compatibles, que se convirtieron en el estndar de la industria
[Martin y Campbello'Kelly
William Aspray, informticos: Una historia de la Informacin de la mquina (Nueva York: Libros
bsicos,
1996), pp. 254-255].
Pgina 113
118
Captulo 6 grabados en silicio: El 19805
ENTREVISTADOR:
Lo que pasa con el enfoque basado en modelos es que se necesita AC-
modelos coadjutor. que ha sido un problema?
THOMAS KAILATH:
S, esto siempre ha sido un obstculo. La gente dice que no lo hace
tienen los modelos, o que los modelos son muy pobres. Sin embargo, esto no
alcanza la
punto de modelizacin matemtica. Se ha demostrado innumerables veces ms
(en
fsica, qumica, y la ingeniera) que los modelos muy simples, que puede ser
bastante inexacta, resulta que tienen poder predictivo tremenda. As que no lo
hace
realmente necesitan modelos muy elegantes. Los modelos simples harn. 1
JAMES KAISER:
Soy una de esas personas que, cuando he hecho algo de trabajo, al-
maneras dice "Ch, alguien est obligado a ya han hecho esto antes." yo
recuerde, me pregunt por qu, con todo el trabajo que se hizo en la muestra
sistemas de datos, con Bill Linvill, Ron Howard, y Bob Sittler, que estaban en el
MIT,
Lotfi Zadeh y John y Ragazzini en Columbia (aquellos eran los dos grandes
muestreado escuelas de datos), por qu esos tipos no hicieron ms con el filtro-
digitales
En g. Creo que el problema fue que en ambos lugares, ellos no tienen la
econmica
medios para implementar de forma importante los sistemas de datos muestreados,
las cuales
fueron motivados principalmente por el sistema de control de problemas-
principalmente en radar, donde
usted tiene seales peridicas derivadas de la exploracin de la antena, la
exploracin circular.
Despus de todo, el circuito integrado an no haba llegado a lo largo.
Tras el desarrollo del circuito integrado, sin embargo, se convirti en posi-
bles para que la gente empiece a pensar en la implementacin en serio. Eso es lo
DSP hizo tomar OffL 2
1Thomas Kailath entrevista de historia oral 13 marzo de 1997, p. 32.

2 James Kaiser-entrevista de historia oral 11 febrero de 1997, p. 13.
clon, fabricado por Compaq. En 1983 lleg el ratn y mens pull..down,
con (comercialmente fracasado) Lisa de Apple, y ese mismo ao la IBM
PC..XT se convirti en el primer ordenador personal con una unidad built..in
hard..disk.
Apple introdujo el Macintosh en 1984, y su user..friendliness era
copiado en el software de Microsoft Windows, lanzado en octubre de 1985. La
nmero de ordenadores personales en todo el mundo aument de 1,3 millones en
1980 a 115 millones en 1990. 5 Las aplicaciones informticas se convirti en una
industria importante,
crecimiento de las ventas anuales de $ 140 millones en 1981 a US $ 1,6 mil
trescientos
SWilliam Gates, "Computadoras personales en la era de la informacin",
en Britannica Anuario de Ciencia ~
ence and the Future 1992 (Chicago: Encyclopaedia Britannica, 1991), pp. 144-153.
Pgina 114
Chapter 6 Etched in Silicon: The 1980s
119
JAMES KAISER:
Think of the designer of elements of a communication system:
he or she has what I call the big parts box next to his or her desk. The parts
box has a couple of different compartments, labeled "cheapest element," "next
cheapest element," and "most expensive element." Remember, economics is
always the name of the game. The engineer is going to try to use as much of
the cheap stuff as possible. Well, for continuous filters design, the cheapest
thing was wire. That's the cheapest thing going. Next were resistors and ca-
pacitors. Inductors were fairly expensive. The most expensive thing, however,
was gain. With the vacuum tube supplying gain, you had to provide the fila-
ment supply and the plate supply (called the "C+") and so on. Consequently,
filters were implemented primarily out of RLCs. Oh, maybe a little later you
could get some active filters using miniature tubes, but it was mainly RLCs.
These components cost money, so the design techniques for continuous fil-
ters were set up so that you were always trying to get the filter of minimum
order to meet your specifications. Minimum order meant a minimum number
of parts.
When we got the integrated circuit, it was like somebody completely
changed the rules. The cheapest thing around now was gain-supplied by the
transistor. You could lay down the resistors, that was easy. Capacitors were
fairly easy. Inductors were still pretty hard, but you didn't really need those.
You could leave those out of an instrument. The most expensive thing around,
however, was the interconnection-the wire! All of the sudden, the tables
were completely turned around. 1
1 James Kaiser oral-history interview 11 February 1997, pp. 13-14.
years later. 6 In the late 1980s email became common as personal computers
connected to networks such as the Internet. (The number of Internet host
computers grew from a thousand in 1984 to 150,000 in 1989.)7
The economic and technological advances took place in a fairly stable
political setting: Ronald Reagan (followed by his former running~mate
George Bush), Margaret Thatcher, Helmut Kohl, and Fran~ois Mitterand
led their countries through most of the 1980s and into the 1990s. The Eu~
ropean Economic Community, which expanded by the addition of three
nations in the 1970s and three more in the 1980s, rivaled the superpowers
6Campbell..Kelly and Aspray op. cit., p. 260.
7Campbell..Kelly and Aspray op. cit., p. 297.
Pgina 115
120
in population and economic strength. 8 The decade ended, however, with
the dissolution of the Soviet Union and the downfall of most Communist
regimes in eastern Europe, epitomized for many by the opening of the
Berlin Wall in November 1989. There were far...reaching changes in politi...
cal and economic institutions in China, though the crushing of a protest at
Peking's Tiananmen Square in June 1989 was a set...back for political free...
dom. 9 The first launch of the Space Shuttle Columbia in 1981 came to be
overshadowed for most people by the explosion, shortly after launch, of the
Shuttle Challenger in 1986. The 1984 breakup of AT&T into a long...dis...
tance company and seven regional operating companies ("Baby Bells") af...
fected almost everyone in the United States.10
One of the exciting new technologies at the start of the decade was
videodisk, available in a variety of systems. They were of two types: systems
in which an electrode sensed differences in capacitance, pioneered in the
RCA VideoDisc, and systems with optically encoded information read by a
laser, pioneered in Philips Laservision.11None of these achieved great mar...
ket success,12 though development of the latter type contributed to one of
the most successful consumer products of all time, the CD player.
1983 was the year Philips and Sony, rival companies that collaborated
in bringing the new technology to market, began selling CD players. Sus
technical features, some of which were described in Chapter 1, drew on sev...
eral decades of advances in signal processing, using, for example, the type of
error...correction coding invented by Irving S. Reed and Gustave Solomon
in 1960 and incorporating a 16...bit analog...to...digital converter and digital
filters of various sorts. 13 Besides setting a new standard for audio fidelity-
8David Reynolds, "Europe divided and reunited, 1945-1995" (in TCW Blanning,
ed., The
Oxford Illustrated History of Modem Europe (Oxford: Oxford University Press, 1996), pp.
279-304).
9Signal...processing engineers may have noted that the fax machine was an
important tool of
the student revolt in China in 1989 [Steven Lubar, Infoculture: The Smithsonian Book of In...
formation Age Inventions (Boston: Houghton Mifflin, 1993), p. 10].
lOOne effect was to make the telephone, for the first time, a consumer
electronics product;
until then telephones had been leased from the phone company.
llMargaret BW Graham, The Business of Research: RCA and the VideoDisc (Cambridge:
Cambridge University Press, 1986), pp. 14-24.
120ne cause of the market failure of the videodisk in the early 1980s was that its
introduc...
tion came just at the time that VCRs were catching on with consumers
[Graham op. cit., pp.
213-219].
UPeter J. Bloom, "High...quality digital audio in the entertainment industry: an
overview of
achievements and challenges" (IEEE ASSP Magazine, voL 2 (1985), no. 4, pp. 2-25), and
Fred Guterl, "Compact disc" (IEEE Spectrum, voL 25 (1988), no. 11, pp. 102-108).
Pgina 116
121
BEOE Llu: In 1967 or so, I proposed to NASA a way to implement digital fil-
ters using delta modulation, without the multiplier, because in delta modula-
tion, it is easy to do multiplication by one bit. That was not funded. But, two or
three years later, I was working with Abe Peled [a graduate student] and I
asked him to look at it again. He came back a couple of days later and said,
"This is a great idea. It works." We continued working on it and he said, "Oh
yeah, you can do this sort of thing; that will be very good." It is now the
scheme most people call distributed arithmetic.
... It turned out some others thought of the same idea just around that
time, maybe even a little bit ahead of us, but for some reason they did not re-
ceive wide publicity. Stan White at Rockwell was very interested in this. l
dubbed the scheme the "Princeton Multiplier.,,1
consumers now expect CD,quality sound from other products-this tech,
nology contributed to the development of a variety of other optical storage
media, such as the CD,ROM (Read Only Memory optical,disk for comput,
ers), which was introduced by Philips and Sony in 1984, and the Digital
Versatile Disk (DVD), which began to be marketed in the late 1990s. An,
other form of digital audio, digital audio tapes, began to be marketed in the
mid 1980s, but were slow in gaining popularity.
The CD player, which for every second of music performed prodigious
computations, would not have been economically feasible before the era of
integrated circuits and solid,state lasers,14 and no doubt the biggest story in
signal processing in the 1980s was the design and production of integrated
circuits, especially single,chip DSPs. It was, of course, possible to use gen,
eral,purpose chips, but a chip designed specifically for signal,processing op'
erations could do the job much faster. And for real,time signal processing,
where a signal must be processed as fast as it arrives, speed is not just desir,
able, but requisite.
A number of factors determine how rapidly a single,chip DSP carries
out an algorithm. Processor speed, measured in millions of instructions per
second (MIPS), is obviously important.'So is the use of dedicated multiplier
units, which perform multiplications in a single cycle, not, as in traditional
14The first CD players offered by Philips contained 20 ICs each, though a long--
term design
goal was to reduce this to a single IC [Guterl op. cit.].
Pgina 117
122
computer architectures, through a series of shift...and...add cycles. Still
greater speed comes from the use of special instructions that involve several
operations in a single cycle, such as multiplying two numbers and simulta...
neously adding the previous product to a total. Also, there is the use of
pipelined architectures, which permit the simultaneous execution of differ...
ent stages of a process.15As a result of these and other design features, the
performance of single...chip DSPs has consistently exceeded that of micro...
processors with arithmetic co"'processors by more than an order of magni...
tude.16
In the early 1980s there appeared quite a few single...chip DSPs. Partic...
ularly successful were the AMI S2811, the Intel 2920, the NEe muPD7720,
and TMS3201 from Texas Instruments.17The last named was the first in
the highly successful and continuing TMS320 series. (See Figure 2.) In
1984 AT&T began marketing DSPs with the DSP32, which was the first
32...bit floating...point DSP.1 8 Particularly successful in audio products was
the Motorola 56000, introduced in 1985.19
As the applications of signal processing increased, great efforts were
made to improve the performance of DSPs. Innovative algorithms were in...
troduced, such as the fast Hartley transform, developed by Ronald
Bracewell in 1984,20 and the arithmetic Fourier transform, introduced by
Donald Tufts and G. Sadasiv.21Some algorithms were designed specifically
to match hardware capabilities, such as the Winograd discrete Fourier
transform, which reduced multiplications at the expense of increased addi...
Instruments
Technical]ournal, vol. 13 (1996), no. 2, pp. 5-16).
16Edward A. Lee, "Programmable DSP architectures: part I" (IEEE ASSP Magazine, vol.
5
(1988), no. 4, pp. 4-19).
17Craig Marven and Gillian Ewers, A Simple Approach to Digital Signal Processing (New York:
John Wiley & Sons, 1996), p. II.
18Lee op. cit.
(IEEE Signal Process...
ing Magazine, vol. 14 (1997), no. 5, pp. 51-57).
2oRonald N. Bracewell, "The Fourier transform" (Scientific American, vol. 260 (1989),
no. 6,
pp. 86-95). The original Hartley transform was conceived by Western Electric
researcher
Ralph Hartley in 1942 as an alternative to the Fourier transform in analyzing a
function in
terms of sine and cosine functions.
21Donald W. Tufts and G. Sadasiv, "The arithmetic Fourier transform"
(IEEE ASSP Maga...
zine, vol. 5 (1988),no.l,pp.13-17).
Pgina 118
123
FIGURE 2. This was the first single-chip DSP manufactured by Texas
Instruments.
(Texas Instruments photo reproduced with permission.)
tions.22The decade of the 1980s also saw much work on adaptive IIR (infi;
nite;impulse;response) filters and on nonlinear filters.23
A great deal of attention was given to alternative ways of carrying out
multiplication, such as cellular arrays, memory intensive policies, homo;
morphic systems, and modular arithmetic, which led in the late 1980s to
the design of dedicated multiply, multiply/accumulate, and numeric proces;
sor chips.24 For algorithms involving trigonometric, logarithmic, or expo;
nential functions, the CORDIC (COordinate Rotation DIgital Computer)
algorithm received renewed attention as an alternative to the conventional
multiply;and;add hardwares.25Another alternative architecture that at;
22David J. DeFatta, Joseph G. Lucas, and William S. Hodgkiss, Digital Signal
Processing: A
System Design Approach (New York: John Wiley, 1988), p. 3.
23John J. Shynk, "Adaptive IIR filtering" (IEEE ASSP Magazine, voL 6 (1989), no. 2, pp.
4-21), and V. John Mathews, "Adaptive polynomial filters" (IEEE Signal Processing Maga..
zine, voL 8 (1991), no. 3, pp. 10-26).
24Gin..Kou Ma and Fred J. Taylor, "Multiplier policies for digital signal
processing" (IEEE
ASSP Magazine, voL 7 (1990), no. 1, pp. 6-20).
25yu Hen Hu, "CORDIC..based VLSI architectures for digital signal
processing" (IEEE Sig..
ool Processing Magazine, voL 9 (1992), no. 3, pp. 16-35), and Jack E. VoIder, "The CORDIC
trigonometric computing technique" (IEEE Transactions on Electronic Computers, voL 8
(1959), pp. 330-334).
Pgina 119
124
INTERVIEWER:
What happened when these chips became available?
BEOE Llu: People underestimated how difficult it was to program them. Ellos
did not realize that you need a whole support structure to do it. It's not that
you can plug it in, twiddle a few knobs and it works. Assembly-language pro-
gramming was not widely taught, there were no compilers. You could not write
in high-level language and then compile. That came much later. There were
some transitions to go through. I think something like the Intel 2920 was actu-
ally very useful for a while, but they did not realize-this is the story I heard-
that each of these products required multi-million-dollar support. So some of
these failed. I think Texas Instruments succeeded because of this. Once the
chip is out, they have a developing board, an interface board; they have a
whole system to use. Those were in the mid-'80s-it took roughly five years
for those to become useful. 1

tracted interest in the 1980s was distributed arithmetic. Invented indepen,
dently in about 1970 by Alain Croisier and colleagues in France and by
Shalhav Zohar in the United States, distributed arithmetic is well suited to
the usual sum,of,products computation.26The 1980s also saw increased in,
terest in various schemes of parallel computation. Parallel processing for the
FFT had been developed in the mid 1970s,27 and in the 1980s considerable
work on parallel processing led to the adoption of new algorithms.28
Signal processing played an important part in what became quite ap'
parent in the 1980s: an increasingly globalized culture. People in countries
26Stanley A. White, "Applications of distributed arithmetic to digital signal
processing: a tu"
torial review" (IEEE ASSP Magazine, vol. 6 (1989), no. 3, pp. 4-19). The name
'distributed
arithmetic' was introduced by Hans Wilhelm Schuessler; see Manfred Buettner
and Hans
Wilhelm Schuessler, "On structures for the implementation of the distributed
arithmetic"
(Nachrichtentechnische Zeitschrift, vol. 29 (1976), pp. 472-477).
27Bernard Gold and Theodore Bially, "Parallelism in fast Fourier transform
hardware" (IEEE
Transactions on Audio and Electroacoustics, vol. AU..21 (1973), pp. 5-16).
28See, for example, Alfred Bruckstein and Thomas Kailath, "An inverse
scattering frame..
work for several problems in signal processing" (IEEE ASSP Magazine, vol. 4 (1987),
no. 1,
pp. 6-20); Peter Strobach, "New forms of Levinson and Schur algorithms"
(IEEE Signal Pro..
cessing Magazine, vol. 8 (1991), no. 1, pp. 12-36); and Thomas Kailath and Ali H. Sayed,
"Displacement structure: theory and applications" (SIAM Review, 1995, pp. 297-
386).
Pgina 120
125
around the world very often experienced the same television programming,
the same movies, the same music, and the same computer games, and all of
these products involved signal processing. Television became even more in,
fluential: in 1992 there were 153 television sets for every 1000 people
worldwide, 490 per thousand in developed countries. In 1986 television was
played, on average, eight hours per day in every US home. 29 Some 700
million people watched the marriage of Prince Charles and Lady Diana
Spencer in 1981, and the next year three or four times that many watched
the World Cup soccer final.
The decade opened with enthusiasm for videotex, which was to put
inexpensive computer terminals in the homes of millions of people, who
could use the terminals, connected to large computers through telephone
lines, to view news, sports, and other information, as well as do shopping
and banking. In the late 1970s many national PTTs (post, telegraph, and
telephone administrations) and dozens of private companies designed
videotex systems. There were problems, notably the high cost of a terminal
(or a decoder to attach to one's television set) and the slowness with which
high,resolution graphics appeared on the screen. The first videotex system
was British Telecom's Prestel. Launched in 1979, it had only 48,000 sub,
scribers five years later. Of the European videotex systems, the only one to
become at all popular in the 1980s was the French Teletel, which used a
terminal called "Le Minitel", provided to users free of charge.
3o
In the late
1990s personal computers attached to the Internet achieve many of the
purposes of videotex, but one should remember that PCs were hardly
known when the videotex systems were first planned.
The effort to devise more efficient coding for videotex graphics exem,
plifies the increased attention in the 1980s to image processing. Other ap'
plications included fax machines, camcorders, radar, enhancement of satel,
lite photos, and medical imaging. For example, CT (computerized
tomography) scanners, introduced in the 1970s, became faster and more
economical, and MRI (magnetic resonance imaging) scanners began to be
utilizado en los hospitales a principios de 1980. 31 tcnicas de procesamiento de
seales eran
Tambin se usa para hacer ecografas (imgenes de ultrasonido de diagnstico)
ms tiles. 32
290very op. cit., p. 188.
30Carol Fletcher, "videotex: volver compromiso" (IEEE Spectrum, Vol. 22
(1985), no. 10, pp.
34-38).
31IEEE Spectrum, vol. 25 (1988), no. 11, p. 73.
32Roman Kuc, "Procesamiento de diagnstico seales de ultrasonido" (Revista IEEE
ASSP, vol. 1
(1984), no. 1, pp. 19-26).
Pgina 121
126
Captulo 6 grabados en silicio: La dcada de 1980
Hans GEORG MUSMANN:
Hemos aprendido que al ir digital y luego la reduccin
la tasa de bits, se puede reducir el tiempo de transmisin, y por lo tanto el
costo. En el
Mientras tanto, las mquinas de fax fueron aumentando en la resolucin a pesar
de
an estaban analgica. Pero una pgina requiere unos quince minutos para trans-
misin. Creo que en ese momento el costo de transmisin a los Estados Unidos
era
alrededor de veinte a treinta D-marcas, por lo que era muy caro para transmitir
una
imagen.
ENTREVISTADOR:
Se utiliza esta comercialmente en ese momento?
Musmann:
Slo para la polica y la industria, ya que era demasiado caro para la mayora
gente. Tambin las mquinas eran muy caros: Diez mil a thou- de veinte
arena D-marcas. Pero nos dimos cuenta de que haba potencial en el desarrollo
facsmil, que estaba relacionado con el desarrollo de impresoras para
ordenadores.
Las computadoras requieren impresoras y mquinas de fax son similares a una
impresora. Asi que
haba una sinergia entre estos dos acontecimientos.
... La compaa del Dr. Hell [Rudolf Hell GmbH], que estaba en Kiel, con-
structed mquinas de facsmil para la polica, y estbamos en contacto con ellos.
Hemos encontrado que es posible reducir el nmero de bits para presentar una
imagen en un factor de diez. Esto era relativamente complicado; era el
Doctor en Filosofa. obra de Dieter Preuss, y todava se utiliza como referencia
en la actualidad. Haba
un desarrollo paralelo, tanto en Estados Unidos y Japn.
... Era la industria japonesa, que reconoci el potencial detrs
el facsmil y empuj esto para hacer productos fuera de l. Hubo una especial
la necesidad de Japn; los pases occidentales ya tenan tlex. 1
1 Hans Georg Musmann entrevista de historia oral 30 de de agosto de 1994, pp. 10-13.
A mediados de 1980 pelculas black..and..white comenzaron a tener el mismo
color con el
ayuda de un system.33 signal..processing de vdeo (Aunque deplorado por la
mayora
cineastas, el mtodo sin duda aument el nmero de pelculas antiguas se
muestra
en la televisin.) 34 Hacia el final de la dcada, muchos reproductores de video
contenidos digi ..
Tal circuitos de memoria para generar still..frames y otros effects.35 especiales
Fue en la dcada de 1980 que la mquina de fax se hizo comn en oficinas.
Esto dio como resultado en parte de una codificacin ms eficiente, menos
costoso hardware,
33El sistema construido por los sistemas de color Tecnologa contena ms de 20
microprocesadores
trabajando en paralelo [IEEE Spectrum, vol. 22 (1985), no. 10, p. 30].
34Studies mostraron que los televidentes probar diferentes canales para algo que
ver Were
mucho ms probable que se detenga para una pelcula de color que no sea una
pelcula black..and..white.
3SU.S. Industria de Electrnica de Consumo: 94 op Edicin. cit., p. 23.
Pgina 122
127
y la adopcin de estndares internacionales. Ya en 1968 hubo una
norma internacional para la transmisin de fax. Esta norma, para la so..called
Grupo 1 mquinas, y la siguiente, adoptada en 1976 para el Grupo 2 mquinas,
eran para la codificacin analgica. En 1980 lleg el primer Stan digital
internacional ..
dard, para el grupo 3 mquinas. Y. Wakahari y T. Yamada en Japn y
Thomas S. Huang en los Estados Unidos se encuentran entre los contribuyentes a
la
Grupo 2 y Grupo 3 de compresin estndares. Sigui en 1984 la
estndar para Grupo 4 mquinas, que era para acomodar la nueva ISDN
(see below). From 1980 to 1992 the cost of a digital fax machine fell by a
factor of 30, and in the United States the market for fax machines grew
from a half million units (annual sales) in 1985 to six million in 1991. 36
In the 1980s image processing became a recognized technical spe..
cialty. The first books on image coding-such as Computer Techniques in
Image Processing by Harry Andrews, Picture Bandwidth Compression edited by
Thomas Huang and Oleh Tretiak, and Image Transmission Techniques edited
by William K. Pratt-appeared in the 1970s. 37 John Limb and Art Murphy
showed how to measure the speed of moving objects, a result later used by
Arun N. Netravali and John D. Robbins to introduce motion..compensated
television coding. 38 Hans Georg Musmann demonstrated video transmis..
sion using only one 64..kbps speech channel and introduced a number of
techniques for video transmission at low bit..rates, including scene analysis
and object..dependent parameter coding. 39 Roger Y. Tsai and Thomas
36Henry Petroski, Invention by Design: How Engineers Get from Thought to Thing (Cambridge,
MA: Harvard University Press, 1996), pp. 113-118.
37Harry C. Andrews, Computer Techniques in Image Processing (New York: Academic Press,
1970); Thomas S. Huang and Oleh J. Tretiak, eds., Picture Bandwidth Compression (New
York: Gordon and Breach, 1972), and William K. Pratt, ed., Image Transmission Techniques
(New York: Academic Press, 1979). Another important book of the 1970s was
Harry C. An..
drews and Bobby R. Hunt, Digital Image Restoration (Englewood Cliffs, NJ:
Prentice..Hall,
1977). See Hans Georg Musmann oral..history interview 30 August 1994.
38John O. Limb and H. Arthur Murphy, "Measuring the speed of moving objects
from televi..
sion signals" (IEEE Transactions on Communications, vol. 23 (1975, and Arun N.
Netravali
and John D. Robbins, "Motion..compensated television coding: part I" (Bell System
Technical
Journal, vol. 58 (1979), pp. 631-669).
39Hans Georg Musmann and Juergen Klie, "TV transmission using a 64 kbit/s
transmission
rate" (International Conference on Communications, 1979, pp. 23.3.1-23.3.5) and Hans Georg
Musmann, "Predictive image coding" (in William K. Pratt, ed., Image Transmission Tech..
niques (New York: Academic Press, 1979), pp. 73-112). See also Hans Georg Musmann,
Michael Hotter, and Jorn Ostermann, "Object oriented analysis synthesis coding
of moving
images" (Signal Processing; Image Communication, vol. 1 (1989), pp. 117-138) and Hans
Georg Musmann, "A layered coding system for very low bit rate video
coding" (Signal Pro..
cessing: Image Communication, vol. 7 (1995), pp. 267-278).
Pgina 123
128
Huang achieved valuable results in estimating 3... dimensional motion from
2...dimensional images.4oIn 1984 appeared the influential text Multidimen...
sional Digital Signal Processing by Dan Dudgeon and Russell Mersereau, and
in 1988 Arun Netravali and Barry Haskell published a comprehensive pre...
sentation of the theoretical and practical aspects of image coding.41
The improvement of image processing was facilitated by the use of
common test images, which made it easier to compare the performance of
different algorithms or image processing systems. These test images, four of
which are shown in Figure 3, were chosen for their inclusion of a wide
range of visual features such as textures, shadings, and colors. For the study
of motion"'picture systems there are standard video sequences. (Similarly,
the speech...processing community has employed standard samples of
speech.)42
A major achievement of the 1980s was JPEG, the international stan...
dard for digitizing and compressing still pictures. The Joint Photographic
Experts Group, under the auspices of the International Organization for
Standardization and the International Electrotechnical Commission,
elicited the views of experts from around the world and reached agreement
on a standard that has been widely adopted. The success of JPEG inspired
efforts to reach standards for moving images, which, as we will see in Chap...
ter 7, was achieved in the 1990s in the form of MPEG1 and MPEG2.43
Automated image...recognition is one of the tasks signal...processing en...
gineers have tackled. There are numerous contexts in which this would be
useful, as in detecting missile...sites in aerial photographs, recognizing a per...
son from a video image, or finding pictures of interest in a large photograph
collection. It has, however, proved extremely difficult to write traditional
computer"'programs to analyze images. A different approach is to construct
neural networks that can be trained to do the job.
4 Roger Y. Tsai and Thomas S. Huang, "Estimating three..dimensional motion
0
parameters of
a planar patch" (IEEE Transactions on Acoustics, Speech, and Signal Processing, vol. 29 (1981),
pp.1147-1152).
4 Dan E. Dudgeon and Russell M. Mersereau, Multidimensional Digital Signal Processing (En..
1
glewood Cliffs, Nl: Prentice..Hall, 1984), and Arun N. Netravali and Barry G.
Haskell, Digi..
tal Pictures: Representation and Compression (New York: Plenum Press, 1988).
42An example from the early 1990s is the Oregon Graduate Institute
Multi..language Tele..
phone Speech Corpus, designed for language identification research [Yeshwant
Muthusamy,
Etienne Barnard, and Ronald Cole, "Reviewing automatic language
identification" (IEEE
Signal Processing Magazine, vol. 11 (1994), no. 4, pp. 33-41)].
43William Sweet, "Chiariglione and the birth of MPEG" (IEEE Spectrum, vol. 34
(1997), no.
9,pp.70-77)
Pgina 124
129
FIGURE 3. These are four common test images: Lena (or Lenna), the MIT
camera-
man, Stripes, and the mandrill. The last has been used especially in studying the
processing of color images. (Lena image reproduced by permission of Playboy
mag-
azine.)
Work on neural networks, which began in the 1940s, was motivated
by the wish to emulate the way the human brain works. While a traditional
computer consists of a central processing unit carrying out a specified set of
operations sequentially, a neural network consists of a large number of pro..
cessing units working simultaneously and interconnected by multiple links.
The basic unit, called a neuron, may have the structure shown in Figure 4.
Two milestone achievements of the 1950s were Frank Rosenblatt's
perceptron (a simple neural network that decides which of two classes a
given pattern belongs to) and Bernard Widrow's adaline (adaptive linear el..
Pgina 125
130
Activation
Function
uk
q>(')
Yk
Salida
Summing
1
Junction
9k
Threshold
Entrada
Seales
Synaptic
Weights
FIGURE 4. The neuron receives input signals and produces an output signal. por
each input there is a weighting factor, and the activation of the neuron may be
influ-
enced by a threshold signal. (Redrawn after figure on p. 8 of Simon Haykin, Neural
Networks: A Comprehensive Foundation (New York: Macmillan, 1994).)
ement). Engineering interest in neural networks flagged in the 1960s and
1970s, but was strongly revived in the 1980s. Among the causes were new
algorithms, the new capability to implement neural networks in very,large,
scale integrated (VLSI) circuits, and the belief that massive parallelism was
needed for speech and image recognition. Some of the most important new
algorithms appeared first in two landmark publications: John J. Hopfield's
1982 paper on recurrent networks, and a 2,volume text, edited by David E.
Rumelhart and James L. McClelland and published in 1986, on parallel dis,
procesamiento buido. 44 Tambin podra sealarse que el 1987 tutorial ar,
44Simon Haykin, "Las redes neuronales se expanden los horizontes de SP"
(IEEE Revista de Procesamiento de Seales,
vol. 13 (1996), no. 2, pp. 24-49), y Richard P. Lippmann, "Una introduccin a la
computacin
con neural redes"(IEEE Revista ASSP, vol. 4 (1987), no hay. 2, pp. 4-22). Las publicaciones
a que se refiere son John J. Hopfield, "Redes neuronales y sistemas fsicos con
col emergente ..
capacidades computacionales lectivas" (Actas de la Academia Nacional de Ciencias de EE.UU., vol.
70
(1982), pp. 2554-2558) y David E. Rumelhart y James L. McClelland, eds., Paralelo Dis ...
Procesamiento buido: Exploraciones en la microestructura de la
cognicin (Cambridge, MA: MIT
Press, 1986). Rumelhart y McClelland describen el algoritmo back..propagation que
era
inventado por Paul Werbos y publicado por primera vez en su economa
Ph.D. tesis en Harvard; esto es
el algoritmo ms utilizado en el campo de las redes neuronales hoy [Widrow com
personales ..
municacin 2 de febrero de 1998].
Pgina 126
Captulo 6 Grabado en Silicio: Los 1980
131
ENTREVISTADOR:
A continuacin, de alguna manera tiene que codificar la forma del objeto?
Hans GEORG MUSMANN:
S, y ese es el problema que estamos trabajando en la actualidad.
En 1989 se sugiri la codificacin basada en objetos en lugar de codificacin
basada en bloques. Nosotros
representado cada objeto que se mueve por tres conjuntos de parmetros: la
forma, la
movimiento, y el color del objeto. Si se transmite estos parmetros, se puede
sintetizar la imagen en el receptor. El codificador calcula el movimiento de un
objeto, y tambin trata de estimar la forma tridimensional. La forma es
representado por un marco de alambre. El color se proyecta en la parte superior
de la estructura de alambre
superficie.
Hemos desarrollado algoritmos que calculan automticamente el tres
forma tridimensional y el movimiento tridimensional de un objeto. Entonces
nosotros
mover estos objetos modelo y calcular una proyeccin de la escena cambi.
As generamos una secuencia de imgenes en movimiento, que se utiliza para la
prediccin,
paralela a la real. Mediante esta tcnica, hemos reducido las reas de pre
errores de diccin a slo cuatro por ciento de cada imagen. El noventa y seis por
ciento es pre-
pronosticadas correctamente y no necesita ninguna informacin transmitida. Esto
hace que sea posible
para transmitir imgenes en movimiento con una velocidad de bits en el intervalo
de 8 kbitls a 64 kbitls.
Esto significa que puede transmitir imgenes en movimiento en el sistema de
telefona mvil. 1
1Hans Georg Musmann entrevista-historia oral 30 agosto de 1994, pp. 26-27.
tculo escrito por Richard Lippmann y publicado en el IEEE ASSP Maga,
zine se convirti en uno de los papeles ms referenciados en la litera de redes neuronales,
ture. 45
El campo fue ampliado en varias formas en la dcada de 1980. Sobre la base de
la
obra de Rosenblatt, numerosos investigadores idearon sofisticada mltiples, capa
perceptrn (MLP) redes. Un tipo diferente, el radial funcin de base "
(FSR) de red, lleg a ser muy popular en la dcada de 1980. Tanto el MLP y
redes RBF eran estticas, en el sentido de que las unidades de procesamiento, o
nodos,
estaban sin memoria, por lo que la red podra basarse en nica entrada actual.
redes dinmicas, por el contrario, podran basarse en las entradas pasadas o
futuras o hacia fuera,
pone. (El uso de lneas de retardo hace que las entradas o salidas disponibles en
el futuro.). los
so'called red de Hopfield se convirti en el tipo ms ampliamente utilizado de
dinmica
red. 46
45Don R. silencio y Bill G. Inicio, "El progreso en las redes neuronales
supervisadas" (IEEE Seal
46Hush y Home, op. cit.
Pgina 127
132
Captulo 6 grabados en silicio: La dcada de 1980
FIGURA 5. dos chips neural de red desarrollado en el Laboratorio Lincoln en
1988.
El de la izquierda es un clasificador perceptrn y el de la derecha es un feature-
map quantizer, or perhaps the other way around. (Lincoln Lab photo reproduced
with permission.)
Quite a few potential applications of neural networks in automatic
speech recognition, automatic target recognition, robotic vision, and other
areas were shown to be at least feasible, if not yet practica1.47The use of
VLSI reached fruition in 1988, when neural...network chips were developed
both at Bellcore and Lincoln Laboratory.48 Figure 5 shows two chips devel...
oped at Lincoln Lab that were charged...couple devices and that used analog
signal processing for vector multiplication.49
The 1980s saw the evolution of the Integrated Services Digital Net...
work (ISDN) from concept to large...scale implementation. The overall aim
of ISDN was to maximize the capabilities of the public switched telephone
network, allowing transmission of voice, data, and image on a single net...
trabajo. 50 Digital transmission throughout, from subscriber to subscriber,
would make this possible, and the standards adopted would be independent
of the transmission medium or distance involved. Agreement on standards
was reached through the International Telecommunication Union, with an
important set of recommendations appearing in 1985. Among the areas of
digital signal processing that were vital to ISDN were digital filters, adap...
4 7 Freeman op. cit., pp. 232-233.
48John R. Pierce and A. Michael Noll, Signals: The Science ofTelecommunications (New York:
Scientific American Library, 1990), p. 4, and Eva C. Freeman, ed., MIT Lincoln
Laboratory:
Technology in the National Interest (Lexington, MA: MIT Lincoln Laboratory, 1995), p. 233.
4 9 Freeman op. cit., p. 233.
1990), p. 325.
Pgina 128
133
LAWRENCE RABINER:
I think the technology is so mature that it's second nature
to everybody. Everyone you hire is absolutely steeped in the technology,
knows it inside out, and uses it without thinking.
INTERVIEWER:
Did this maturing take place in the early 80s?
RABINER:
Probably in the early- to mid-'80s. When people come in now, they
are just so good with it, it's just fundamental. It's being taught to undergradu-
ates, throughout almost the entire world. People don't even think about it, it's
like programming. 1
CHARLES RADER:
A very important development in the '70s and '80s was the
importance of the relation between algorithms and architectures....
In specialized signal processing, where you know what the algorithm is
and you're willing to commit it to hardware, you can have a lot of multiplica-
tions, additions, registers, divisions, and square roots, and so on, all working
together, all working at the same time. And then one of the issues is how do
these things all get connected together so the results of one computation
could get to where they're needed next, and things can all be kept busy at the
same time. 2
1 Lawrence Rabiner oral-history interview 13 November 1996, p. 41.

tive filters, and signal analysis and recognition. 51 In 1997 an ISDN line aI,
lowed a transmission rate of 64,000 bits per second. 52
Some achievements of the 1980s might be mentioned. Blind linear
equalization moved from concept into practice. 53 The concepts of multirate
signal processing (decimation and interpolation of signals) became much
better and more widely understood through the 1983 text Multirate Digital
Signal Processing by Ronald Crochiere and Larry Rabiner. 54 In 1984 Bishnu
51Maurice Bellanger, "New applications of digital signal processing in
communications"
(IEEE ASSP Magazine, vol. 3 (1986), no. 3, pp. 6-11).
52Limitations within the telephone network, however, sometimes mean that data
speeds are
limited to 56,000 bits per second (IEEE Spectrum, vol. 32 (1995), no. 6, p. 25).
53John R. Treichler, I. Fijalkow, and C. Richard Johnson, Jr., "Fractionally
spaced equalizers"
(IEEE Signal Processing Magazine, vol. 13 (1996), no. 3, pp. 65-81).
54Ronald E. Crochiere and Lawrence R. Rabiner, Multirate Digital Signal
Processing (Engle...
wood Cliffs, NJ: Prentice...Hall, 1983).
Pgina 129
134
Atal and Manfred Schroeder published an important paper on code...excited
linear prediction (CELP), which today is ubiquitous for speech coding at 4
to 16 kbps.55 In 1988 Robert McAulay and Thomas Quatieri introduced the
sinusoidal transform coder (STC), which outperforms LPC in midrate
vocoding.56
Many advances in digital signal processing were stimulated by the
satellite...communications business and by space programs. Packet...speech
conferencing over a satellite network was demonstrated in 1982.57Space
communications relied on a whole range of DSP techniques, and NASA's
Deep Space Network, arrays of antennas located around the world, contin...
ued to perform prodigies of signal extraction (the 1989 Voyager images of
Neptune, for example, traveling across almost three billion miles of
space).58
Electroacoustic transducers appeared to be a mature technology in
1970, yet the 1970s and 1980s were years of rapid progress. Fiber...optic sen...
sors saw great improvement and were used in a variety of instruments be...
sides in communications. (See Figure 6.) Silicon transducers, that is, sen...
sors on silicon chips, were developed with the aim of being able to place
both sensor and signal...processing circuitry on a single chip. Esto era
achieved in 1983 when M. Royer and coworkers at Honeywell demon...
strated a fully integrated acoustic sensor.59The electret microphone, first
commercialized by Sony in 1968, was improved, and other types of electret
transducers were developed. A breakthrough of the 1970s was the commer...
cialization of piezopolymer transducers, first by Pioneer Corporation in
}apan.60
Audio engineering saw many advances in the 1980s. In the United
States stereo was added to television broadcasting in 1985. Though digital
audio tape was not a great success in the consumer marketplace, it was
adopted by sound engineers in audio, television, and movie studios, as
55Biing..Hwang luang, "Speech, audio, and acoustic processing for multimedia"
(IEEE Signal
Processing Magazine, vol. 14 (1997), no. 4, pp. 34-36). The paper by Atal and Schroeder is
"Stochastic coding of speech at very low bit rate" (Proceedings of the International Conference
on Communications, Amsterdam, 1984, pp. 1610-1613).
56Preemanop. cit., p. 23l.
57Preeman op. cit., p. 231.
58Pierce and Noll op. cit., p. 198.
59Gerhard M. Sessler, "What's new in electroacoustic transducers" (IEEE ASSP
Magazine,
vol. 1 (1984), no. 4, pp. 3-11).
6Sessler op. cit.
Pgina 130
135
Laser
Fuente
Photo Detector
Reference Fiber
Modulator
Air
Agua
rrr
Underwater
Sound Wave
FIGURE 6. A fiber-optic interferometric sensor works by combining the light
from a
reference fiber and the light from fiber exposed to sound. Since the sound wave
changes the phase of the light in the sensor fiber, the interference pattern of the
re-
combined beam changes, allowing detection of the sound. 51 (Redrawn after
figure
on p. 3 of Gerhard M. Sessler, "What's new in electroacoustic transducers" (IEEE
ASSP Magazine, vol. 1 (1984), no. 4, pp. 3-11).)
digital mixing consoles were available as early as 1984. 62In 1983 the man..
ufacturers of electronic musical instruments sanctioned a standard digital
interface called MIDI (Musical Instrument Digital Interface);63 MIDI al..
lowed keyboards, synthesizers, and other electronic instruments to com..
municate with each other and with computers. By 1990 most synthesizers
61Sessler op. cit.
62Bloom op. cit. A technical advantage of digital audio, that a copy may be made
that is
identical to the original, has been seen as a disadvantage by the recording
industry, since it
makes piracy easier. Digital audio has special advantages with telephone
answering ma..
chines, as messages may be easily stored, duplicated, annotated, and forwarded.
63U.S. Consumer Electronics Industry Today (Arlington, VA: Consumer Electronics Manu..
facturers Association, 1997), p. 60.
Pgina 131
136
were fully digital, using microprocessors to manipulate the coded signa1.64
Synthesizers enable sound designers for cinema or stage to combine exist...
ing sounds and newly created sounds and to modify them in numerous
ways (as changing tempo or adding or removing harmonics).65
In the 1980s the IEEE Acoustics, Speech, and Signal Processing Soci...
ety continued its steady growth in membership, in publication, and in
scope. The Society played a large role in the development of VLSI signal
tratamiento. There was, for example, an important workshop of this topic,
organized by Sun Yuan Kung, in 1982.66Research, using techniques of
graph theory, linear programming, and scheduling theory, was directed to...
ward systematic procedures for the design of systolic arrays and regular itera...
tive arrays.67
In 1983 the Society established a new technical committee on
VLSI. Its concern was "the unique nature of VLSI technology as applied
to signal processing, where the algorithm and its implementation strongly
influence each other, and where the accumulation of algorithms available
for implementation and the rapid evolution of new algorithms demand
accelerated design techniques to rapidly develop VLSI...based systems for
implementation."68 The other five technical committees at that time were
Digital Signal Processing, Speech Processing, Underwater Acoustics,
Multidimensional Signal Processing, and Spectral Estimation. In 1972
Transactions began appearing six, rather than four, times a year, and in
1987 it began monthly publication. Toward the end of the decade the So...
ciety joined with several other IEEE Societies in the Neural Networks
Council, which began publication of the IEEE Transactions on Neural
Networks in 1990.
Over the past century many types of electrical technologies have be...
come ubiquitous. For example, in the middle decades of this century, small
64Prank E. Wukitsch and John Culbert, "Technology on stage" (1991 Yearbook of
Science and
the Future (Chicago: Encyclopaedia Britannica, 1990); pp. 200-223).
65Wukitsch and Culbert op. cit.
66Papers from that workshop became a book: Sun Yuan Kung, Harper J.
Whitehouse, and
Thomas Kailath, eds,. VLSI y Procesamiento de Seales Modem (Englewood Cliffs, Nueva
Jersey: Prentice,
Hall, 1985).
67See, por ejemplo, Sailesh K. Rao y Thomas Kailath, "algoritmos iterativos y
regulares
su aplicacin en matrices de procesadores" (Actas de la IEEE, vol. 76 (1988), pp.
259-282).
68IEEE ASSP Magazine, vol. 1 (1984), no. 1, p. 32 (noticia por John Ackenhusen).
Pgina 132
137
motores elctricos se abrieron camino en herramientas, equipo de oficina, apli
casa ...
ANCES, y los vehculos, muy a menudo sin que la gente sea consciente de sus
pres ...
cia. Del mismo modo, en la dcada de 1980 los microprocesadores se dirigieron a
la tienda,
la oficina, la sala de estar, la cocina, el lavadero, y el coche. Como
veremos en el siguiente captulo, el procesamiento de la seal era cada vez ms la
rea ...
hijo de este crecimiento que inervan (algunos diran diseminacin cancerosa) de
mi ...
croelectronics.
Pgina 133
Captulo 7
Fugarse:
los aos 1990
QU FUE pasando en el 1990
Pelculas personas estaban viendo:
Forrest Gump
Parque jursico
Rey Leon
Pulp Fiction
TV muestra personas estaban viendo:
"The Oprah Winfrey Show"
"Seinfeld"
"Los archivos x"
gente de la msica estaban escuchando:
(Y bailando a) la Macarena
spice Girls
libros la gente estaba leyendo:
Robert James Waller Puentes de Madison
Condado
Stephen Covey Los 7 Hbitos de la altamente
Las personas efectivas
139
Pgina 134
T l dcada de 1990 se inici una nueva era en las relaciones internacionales. La guerra fria
had dominated the years from 1945 to 1990, as that struggle be..
tween the superpowers was often superimposed on regional and local
conflicts. In the 1990s it was economic relations-in an increasingly glob..
alized marketplace-that took center stage.! World trade had increased 10..
fold from 1950 to the early 1990s, and in the mid 1990s there were some
37,000 multinational enterprises with some 200,000 foreign affiliates world..
amplio. One of the essential factors leading to this growth of international
trade and multinational enterprise was the modern system of global com..
munications that facilitated transnational operation. 2
The 1990s may be remembered as the decade the human race made its
greatest strides toward interconnection. The worldwide telephone network
continued to grow in extent, especially in the less industrially developed parts
of the world, but rather unexpectedly it grew spectacularly in density, as it
connected modems, pagers, cellular phones, fax machines, and the second and
third telephones that many families found they wanted. To traditional broad..
casting.. and cable..television was added, for millions of people, television
broadcast from satellites. Even more so than in previous decades, much of the
IJ.AS Grenville, A History of the World in the Twentieth Century (Cambridge, MA: Harvard
University Press, 1994), pp. 925-926. In the 1990s the states of central and
eastern Europe
faced huge problems in converting from a Communist command...economy to a
free...market
economy.
2Richard Overy, ed., Hammond Atlas of the 20th Century (London: Times Books, 1996), pp.
172-173.
Pgina 135
Chapter 7 Breakout: The 1990s
141
world experienced through television the same events at the same time, such
as the Persian Gulf War in 1991 and the funeral of Princess Diana in 1997.
Added to this was the effect of the Internet, as personal computers, in addition
to being stand..alone devices, became tools of access to this network, and mil..
lions of people all around the world viewed the same Web sites and joined in
virtual communities concerned with almost any conceivable subject. 3
Signal processing played an important part in all of these changes. So it
is not surprising that, after decades of steady growth and the cultivation of
niche markets, signal processing achieved something of a breakout in the
1990s. In 1985 there were only three large commercial markets for DSP
chips-speech coding, video compression, and modems-which, together,
were a $50..million business. The growth of these three applications and the
appearance of many new ones, such as cellular phones, sound boards, hard..disk
drives, scientific instruments, and robotics resulted in a DSP business totaling
$2.2 billion in 1995, as programmable and function..specific DSP achieved
large markets. 4
The breakout for signal processing came with continuing growth of the
consumer electronics market. Some of the products that have become popular
in the 1990s, besides those already mentioned, are multimedia pes, laptop
computers, scanners, projection TVs, widescreen TVs, broadcast satellite an..
tennas, camcorders, digital cameras, fax machines, cordless telephones, and
tapeless answering machines. Most of these products contained DSP chips,
and signal processing thus provided a significant push to consumer electronics.
The advance of integrated circuit technology continued in the 1990s,
as the density of components on chips continued to follow Moore's law,
doubling every 18 months. 5 An important benchmark was the Pentium
3The number of Web servers increased from about 10 in 1991 to 100,000 in 1994 [Scientific
American, vol. 277 (1997), no. 6, p. 36]. The number of hosts attached to the Internet in..
creased from fewer than 1 million in 1992 to almost 20 million in
July 1997 [IEEE Spectrum,
vol. 35 (1998), no. 1, p. 37].
Instruments
Technical]ournal, vol. 13 (1996), no. 2, pp. 5-16).
5The course of development of most new technologies describes an S curve
when some measure
of performance is graphed against time. There is, roughly speaking, exponential
growth in the
first half of this S curve, followed by a decreasing growth rate. What is
remarkable for microelec..
tronics-and perhaps without precedent in the history of technology-is the length
of this pe..
riod of exponential growth (measured in number of consecutive doublings of
performance). los
exponential growth in component density (Moore's law) has been, not
surprisingly, reflected in
exponential growth in many other aspects of the computing. For example, the
software com..
pany Microsoft has doubled in size every two years, and judging by value of
shares, Microsoft is
worth more than the three large us carmakers put
together [Economist 31 January 1998, p. 65].
Pgina 136
142
Chapter 7 Breakout: The 19905
Half Rate
Cellular
Phones
Modems
mil novecientos ochenta y dos
1994
2002
FIGURE 1. This graph shows the increase since 1982 in DSP speeds, measured
en millones de instrucciones por segundo (MIPS), para dos aplicaciones,
telfonos celulares
y mdems. (Redibujado despus de la figura 11 del gen de Frantz y Panos
Papamichalis,
"Introduccin a soluciones DSP" (Texas Instruments Technical Journal, vol. 13 (1996),
no. 2, pp. 5-16).)
viruta, introducida en 1994, que contena cinco millones de transistores. 6 DSP
virutas, tambin, alcanzado una mayor densidad de componentes, y los de Texas
Instruments
Chip DSP TMS320C805, presentado el mismo ao que el Pentium, tambin
contenida cinco millones de transistores. Tambin hubo reducciones en la energa
IC
disipacin, y se convirti en comn incluir funciones analgicas en un DSP
chip.7
Tales avances, combinados con la produccin en masa, llevados por la
costo de procesamiento de seal digital. En la dcada de 1990 podra ser menos
de $ 1 por
MIPS (millones de instrucciones por segundo), mientras que en la dcada de
1960 la ma ~
lomos capaces de costo de procesamiento de seal digital tal vez $ 1000 por
MIPS. 8
(Vase la figura 1.) Este solo hecho va mucho en la explicacin del crecimiento
explosivo
de aplicaciones DSP. Con la tecnologa IC, al igual que con los productos
electrnicos de consumo
6Frantz y Papamichalis op. cit.
8Frantz y Papamichalis op. cit. Segn James Kaiser [comunicacin personal 12
De febrero de 1998] el coste por MIPS en la dcada de 1960 fue
considerablemente mayor que $ 1000.
Pgina 137
Captulo 7 Breakout: La dcada de 1990
143
la cmara
Motor de control
HDTV
Settop
Radio DSP
CD de vdeo
Multimedia
La cancelacin de ruido
Active Suspension
Imgenes en tiempo real
Sin cable
Telfono
Videoconferencia
3-D Grficos
10
1K
100
10K
1980
1985
1990
1995
2000
Figura 2. El grfico muestra el aumento en la velocidad (medida en millones de
opera-
ciones por segundo) para DSPs de un solo chip y microcontroladores
respectivamente. Newappli-
reas de cationes se colocan de acuerdo con la fecha y la velocidad de
procesamiento. (Redibujado despus de
La Figura 14 del gen de Frantz y Panos Papamichalis, "Introduccin a las
soluciones DSP"
(Texas Instruments Technical Journal, vol. 13 (1996), no. 2, pp. 5-16).)
mercado, procesamiento de seales, as como dio recibido: aplicaciones DSP son
hoy tan importante como la industria de la computacin en la conduccin de la
tecnologa de
high..speed, circuitos integrados low..power. 9
El menor coste de clculo sin duda abre nuevas reas de apli ..
cin. Lo mismo ocurri con el aumento de velocidad, especialmente con
procesamiento de seal real..time.
Esto se debi a circuitos integrados ms rpidos y, ms importante an, mejores
algoritmos. Higo..
Ure 2 se refiere el rendimiento DSP y nuevas aplicaciones. La proliferacin de
aplicaciones condujeron a una mayor variedad de chips DSP: a finales de 1990
Mo ..
Torola ofreci una lnea compatibles de 16 .. 24 .. y 32..bit DSP; Tejas Instru ..
mentos ofrecido ms de 150 DSP, sobre todo los de su TMS320
familia; y Analog Devices, IBM, Lucent, NEC, y muchos otros fabri ..
Turers ofreci un gran nmero de chips DSP ms.
Extremadamente importante en el crecimiento de las aplicaciones DSP ha sido la
disponibilidad de soporte de software. programas Filter..design se remontan
varios
dcadas, tales como el programa Parks..McClellan desde principios de 1970, la
cual
ha sido tal vez el ms ampliamente utilizado. 10 Ms recientemente, dos clases de
suave ..
9Paul G. Flikkema, "tcnicas Spread..spectrum para comunicaciones
inalmbricas" (IEEE Seal
lOBede Liu comunicacin personal 27 de de enero de de 1998.
Pgina 138
144
cermica han sido de gran importancia: las herramientas de cdigo ... de
generacin y herramientas para sys ...
tem integracin y depuracin. El primero permite a un programador para usar
una
lenguaje de alto nivel. Por ejemplo, la optimizacin C ... compilador produce
como ...
blea cdigo de idioma que es casi tan eficaz como la mano ... montado pro ...
gramos. Las herramientas de integracin de sistemas y depuracin, tales como el
software simula ...
tores, emuladores de hardware y herramientas de evaluacin del sistema, han
simplificado el
process.II desarrollo
Entre los muchos avances en las comunicaciones en la dcada de 1990 fueron los
rapid development of mobile communications, as cellular phones became
commonplace. The transmission capacity of optical fiber continued to fol...
low its own law of exponential increase: 10...fold every four years since
1975. 12 By 1990 virtually all of the 140 largest metropolitan areas in the
United States were connected by optical fiber. 13 Videoconferencing did be...
come easier, but is still not very much used.I4
There was also a continuing movement toward the digitization of
communications and information storage. Digital cellular phones and PCS
(Personal Communication Services), also digital, became popular. Digital
standard television was introduced. Image scanners became common. Ellos
allowed the digitization of documents of all sorts, especially those with
graphs, charts, diagrams, and pictures, and they were made more useful by
the use of software for optical character recognition.
Until the 1990s telephone answering machines incorporated an ana...
log tape recorder. In the mid 1990s digital telephone answering devices
(DTADs) began to claim a large share of the market. DTADs offer many
advantages, such as message manipulation capability (skip, save, selective
delete, and so on), increased reliability (not having the tape recording and
reading mechanism), and the ability to add value...enhancing features (such
as caller...ID, full...duplex speakerphone, and speech recognition) with small
addition to the unit cost. DTADs are, needless to say, highly dependent
upon DSP technology.Is
llRose Marie Piedra and Andy Fritsch, "Digital signal processing comes of age"
(IEEE Spec--
trum, vol. 33 (1996), no. 5, pp. 70-74).
12Emmanuel Desurvire, "Lightwave communications: the fifth
generation" (Scientific Ameri--
can, vol. 266 (1992), no. 1, pp. 114-121).
1989), p. 323.
14W. Wayt Gibbs, "Taking computers to task" (Scientific American, vol. 277 (1997), no.
1,
pp.82-89).
1STandhoni S. Rao, "DSP software technology for digital telephone answering
devices"
(Texas Instruments Technical]ournal, vol. 13, no. 2 (March..April1996), pp. 57-64).
Pgina 139
145
MAURICE BELLANGER:
S. One important aspect for the acceptance of more
sophisticated, intelligent terminals and appliances is that they must be
friendly, easy to manipulate. That's certainly one area where signal process-
ing has a lot more potential. We need more flexible terminals, and in computer
science we can see the agent technology being developed. I believe that the
concept of agent might apply to signal processing as well. By agent, I mean
an autonomous subset which can find its own way of operating and just report
to the user. If we take again the mobile radio receiver, one can imagine a re-
ceiver which would help the user-it would select the best available channel,
best modulation, do all that by itself, and just report its decisions to the user.
Some kind of internal intelligence. Definitely the man-machine interface is re-
ally critical for the near future. 1
MAURICE BELLANGER:
We feel indeed that in Europe we have always had
some excellent laboratories, certainly at the same level. We had, for example,
excellent background in mathematics and in the basics, which prepared the
young people to develop these techniques. The advantage of being in the
US is certainly the availability of the technology, which undoubtedly came at
that time from US companies, particularly California companies. Another dif-
ference is that in Europe the students and engineers are not forced to publish.
It's perfectly feasible to present a thesis without having published internation-
ally. Local publication and local appreciation may be enough. There is also
the language issue, so there are a number of reasons why the signal process-
ing field might appear more developed in the US than in Europe, but I don't
think that's the case. We could draw a history of European contributions to
signal processing quite easily.2
1Maurice Bellanger oral-history interview 22 April 1997, p. 26.
2Maurice Bellanger oral-history interview 22 April 1997, pp. 21-22.
There has been progress in speech recognition. An important text by
Larry Rabiner and Biing..Hwang luang, Fundamentals of Speech Recognition
(1993), covered both theory and practical aspects. diecisis
Some systems now
have error rates below five percent for most speakers. 17 Apple Computer of..
16Lawrence R. Rabiner and Biing..Hwang Juang, Fundamentals of Speech
Recognition (Engle..
wood Cliffs, NJ: Prentice..Hall, 1993).
17Steve Young, "A review of large..vocabulary continuous..speech
recognition" (IEEE Signal
Pgina 140
146
fers software for voice dictation of Mandarin Chinese.18In 1992 AT&T in,
troduced an automated operator service called VRCP (Voice Recognition
Call Processing), which can handle call processing without operator assis,
tance; VRCP currently handles more than a billion voice,recognitions per
year.19MIT researchers have built a system (GALAXY) that can under,
stand and answer spoken questions about weather, airline flights, and the
city of Boston; the system, however, often makes mistakes. Tambin hay una
speech,recognition system that takes dictation from radiologists with 97
percent accuracy, yet a study found that radiologists completed their reports
faster without the dictation system. Widespread use of speech recognition
remains an elusive goal, as does the even more challenging goal of building
reliable language,understanding systems.20
As we have seen in the preceding two chapters, digital means have
gradually displaced analog means for recording, processing, transmission,
and reproduction of audio. In the 1990s two new standards, MPEG1 and
MPEG2 (discussed further below) made efficient coding of audio widely
used, as they allowed CD,like quality with six,to,one or twelve,to,one
compression ratios.21In the early and mid 1990s the film industry switched
from analog to digital formats.22In the 1990s there has been much work on
microphone arrays, especially for teleconferencing, such as self,steering ar,
rays (that electronically aim at the person speaking at that moment) and
adaptive beamforming (in which the beamforming program adjusts its para,
meters on the basis of received data). Also notable has been work on
recording and reproducing a 3,dimensional sound,field,23 and on DSP,as,
sisted loudspeakers (which have built,in digital filters to correct for the
shortcomings of speaker performance).24
Since the 1930s engineers have worked on active noise control
(ANC), the intentional generation of sound to interfere destructively with
unwanted sound. But it was not until the 1990s that signal processing tech,
18Lin..Shan Lee, "Voice dictation of Mandarin Chinese" (IEEE Signal Processing
Magazine,
vol. 14 (1997), no. 4, pp. 63-101).
19Lawrence Rabiner personal communication 30 January 1998.
20Gibbs op. cit.
21 Marina Bosi, "Perceptual audio coding" (IEEE Signal Processing Magazine, vol. 14 (1997),
no. 5, pp. 43-49).
22Bosi op. cit.
23Gary Elko, "Electroacoustic transducers" (IEEE Signal Processing Magazine, vol. 14 (1997),
no. 5, pp. 31-36).
24Ken Kantor, "DSP in audio" (Audio, December 1996, pp. 27-32).
Pgina 141
Communications
Real-time encode and decode
Low delay
147
Information/Entertainment
Real-time decode
Delay not critical
1990
1995
H.261
H.263
MPEG 1
MPEG2
1990
1993
FIGURE 3. Four important standards for video compression. (Redrawn after
Figure
2, with dates of adoption added, of Karen Oehler, Raj Talluri, Yuji Itoh, and Fritz
Whittington, "Digital video compression" (Texas Instruments Technical
Journal, vol.
13, no. 2 (March-April 1996), pp. 27-40).)
niques were advanced enough to build commercially viable systems. Esta
work is related to, but distinct from, adaptive noise..cancellation, where the
cancellation takes place in the electrical system, as with the echo..cancelers
used in telephony. ANC systems have been commercialized for the control
of propeller noise in aircraft and for the control of noise in ventilation
ducts. 25
In the 1990s image processing became an even more prominent
branch of signal processing. Since video in uncompressed form requires
great bandwidth, effective means of image compression are required for
many transmission and storage technologies to be practical. Here, as with
many other areas of DSP, international standards have been vital to rapid
technological and commercial development. Figure 3 depicts the relation..
ships among four important compression standards: H.261 and H.263 de..
signed for real..time encoding and decoding with very low delay, as for
videophone communication; and MPEG1 and MPEG2 for multimedia PC
and television where some delay is acceptable. 26
25Stephen J. Elliot, "Active noise control" (IEEE Signal Processing Magazine, vol. 14 (1997),
no. 5, pp. 36-39).
26Karen Oehler, Raj Talluri, Yuji Itoh, and Fritz Whittington, "Digital video
compression"
(Texas Instruments Technical Journal, vol. 13, no. 2 (March..April1996), pp. 27-40).
Pgina 142
148
Chapter 7 Breakout: The 19905
THOMAS HUANG:
In the good-old-days it was more fun, but now things have
changed. We are more application-driven now, even if our main interest is still
in basic research. At that time I was more romantic-I did things just for fun
and didn't worry too much about the application. 1
THOMAS HUANG:
First, there was differential PCM, next was transform coding,
and then came fractal coding, a really novel idea in image coding. After trans-
form coding, nothing really new happened in image coding, people just tried to
improve it, at least until Barnsley's idea of fractal coding. Michael Barnsley
was a mathematician at Georgia Tech, then he left to form his own company.
The wavelet method is still a traditional signal processing approach, but it's
not that different from transform coding. Fractal coding is completely new.
The idea is, you have an image and you try to find a mathematical func-
tion-a system-for which this image is an attractor, what in mathematics
they call a fixed point. You have a system for the input. If you can find an input
which, when fed into the system, looks exactly the same when it is output,
then this input is called a fixed-point of the system.
... The interesting thing is that for most systems, if the input is not a fixed
point, and you repeatedly feed its output back into the system, it will approach
a fixed point. With any real, positive number, if you take the square root, and
the square root of that, again and again, it will approach one. So, the idea of
fractal coding is to take an image and find a system for which this image be-
comes a fixed point. The description of the system is your compressed data.
Once you have this system, you can take any image, feed it in iteratively, and
eventually you will get the original image back. You try to make this system
simple, like linear transformation. Fractal coding is a completely radical idea. 2
2Thomas S. Huang oral-history interview 20 March 1997, pp. 15-16.
As a standard for videophones the International Telecommunications
Union developed H.261; approved in 1990, it was intended for use on
ISDN lines at 64 kbps or a multiple of that rate. (See Figure 4.) For lower
bit~rates, H.263 was defined in 1995. The Moving Pictures Expert Group
(MPEG) reached agreement in 1990 on a standard for CD video and audio,
known as MPEG1, then went on to define a more general syntax that could
be used for broadcast video. MPEG2, adopted in 1993, can apply to digital
standard television, digital high~definition television, and the DVD (digital
Pgina 143
149
Video
Entrada
+
Anterior
Imagen
Coefficient
Valores
Varia
S re
IDCT
+
Displacement
Vectors
FIGURE 4. The encoding scheme of H.261 is shown. (Redrawn after Figure
6.35
of Craig Marven and Gillian Ewers, A Simple Approach to Digital Signal
Processing
(New York: John Wiley & Sons, 1996).)
versatile disk).27 MPEG4, still under development, will apply to video cod,
ing at low bit,rates, such as are typical for Internet connections.28
MPEG2 has been a great achievement from several points of view. A
its meetings, which began in 1990, hundreds of experts from around the
world contributed their knowledge and opinions. It became a very broad
standard: it includes MPEG 1 as a subset, and its original goal of defining
coding for standard digital television expanded to include HDTV, thus ob,
viating the planned MPEG3. The standard was reached swiftly, as MPEG
was meant to be an anticipatory standard. And it has been of enormous
commercial importance: it has been adopted for direct,broadcast and cable
television, and it is the core of the so,called Grand Alliance standard for
HDTV. The driving force behind MPEG has been Leonardo Chiariglione
of the Centro Studi e Laboratori Telecomunicazioni SpA (CSELT) in
Turin, Italy. In 1996 the uS broadcasting industry awarded MPEG, along
with JPEG, an Emmy.29
270ehler et al. op. cit.
28William Sweet, "Chiariglione and the birth of MPEG" (IEEE Spectrum, vol. 34 (1997), no.
9, pp. 70-77).
29Sweet op. cit.
Pgina 144
150
THOMAS HUANG:
I guess one way of looking at this is to look at the different
goals: compression, enhancement, restoration, and analysis. In compression,
los hitos son bastante claras: PCM diferencial, codificacin de transformacin, a
continuacin,
wavelets, fractales y la compresin ahora quiz basado en modelos. En el en-
rea hancement, un hito importante de la reconstruccin de la imagen que
tenemos
completamente ignorado en esta discusin es la tomografa computerizada. Eso
es realmente
uno de los principales logros en la mejora de la imagen y la reconstruccin.
... Ms recientemente, por supuesto, tenemos imgenes de resonancia
magntica. por
anlisis, los hitos son ms difciles de decir, hay tantas facetas diferentes.
Reconocimiento de objetos en imgenes ha estado ocurriendo durante mucho
tiempo, pero es difcil de
identificar hitos. Reconocimiento de objetos es tal vez demasiado vaga-tenemos
que estar
mas especifico. Me gustara pensar OCR, huellas dactilares, y ms recientemente,
muchas perso-
PLE estn entrando en el reconocimiento de caras. 1
THOMAS HUANG:
La transformada de Fourier para untar cosas por todas partes. Si tienes un
pequeo objeto en la imagen, su transformada de Fourier se extiende sobre toda
la fre-
cuencia de dominio. Por lo tanto, tiene ventajas y desventajas. Si quieres
buscar el objeto, no puede hacerlo en el dominio de frecuencia, porque se
propaga
por todas partes. Adems, si usted cometer errores o tener errores en el dominio
de la frecuencia,
que se extienden por toda la imagen. Por otro lado, la transformada wavelet es
concentrado tanto en frecuencia como en el dominio espacial. La transformada
do-
principal tiene varias capas diferentes con diferentes frecuencias en los
componentes.
En cada capa, el objeto original permanece concentrado, no hacia fuera.
Una aplicacin para la representacin de la imagen est en
recuperacin. Quieres volver
trieve imgenes, pero mientras tanto se desea representar la imagen en su Data-
de base de una manera eficiente. Por lo tanto, desea comprimir, pero an as ser
capaz de recuperar
diferentes objetos. Si se utiliza la transformada de Fourier tiene que
descomprimir BE-
Fore puede buscar su objeto, pero si ests en la representacin wavelet,
puede buscar objetos directamente (aunque esto no est completamente
terminado todava) 0,2
1Thomas S. Huang entrevista-historia oral 20 de marzo de 1997 pp. 28-29.
2Thomas S. Huang entrevista-historia oral 20 de marzo de 1997 pp. 11-12.
La dcada de 1990 han sido aos de cambio rpido para la televisin. El 3 de
junio
1989 Japn inici emisin de programas de televisin de alta definicin; aunque
la fuente
seal se define en un formato digital, la transmisin es analgica. Ahora ap"
las peras que Japn, al igual que el resto del mundo, adoptar HDTV totalmente
digital. 30
30Britannica Anuario de la Ciencia y el Futuro 1991 (Chicago: Enciclopedia Britnica,
1990), p. 231, e Hiroshi Watanabe e Hiroshi Yasuda, "La obtencin de una
resolucin ms alta en
"Asia (IEEE Revista Signal Processing, vol. 14 (1997), no. 4, pp. 52-62).
Pgina 145
151
Los sistemas digitales de satlite, la radiodifusin de televisin directamente a los
hogares, se han
convertido rpidamente en un gran negocio; en 1996 ms de tres millones directa
- a--
sistemas domsticos fueron vendidos en los Estados Unidos. 31 de la televisin por
cable digital
comenz a estar disponible en algunas reas. En los Estados Unidos desde 1993
nueva
aparatos de televisin con 13 - pulgadas o pantallas ms grandes han contenido
subttulos DE--
codificadores. 32 Y aunque 4 - sonido del canal no haba podido alcanzar accep--
consumidor
tancia cuando se present bajo el nombre de 'sonido cuadrafnico' en
1969, de audio multicanal tuvo xito en la dcada de 1990 en forma de
"envolvente
sonido" o 'home -. los sistemas de cine' 33
Otra rea de aplicacin del DSP ha sido la fotografa digital. Por
1997 casi todos los principales fabricantes de cmara estaba ofreciendo cmaras
digitales,
que permiten la edicin y mejora de las imgenes en las propias cmaras
y una fcil transferencia de imgenes a los ordenadores empresariales y
privados. 34 Tcnicas
de mejora de la imagen digital de incluir la mejora del contraste, sharp-- borde
Ening, la pseudo - mejora del color (debido a que las discrimina de los ojos
humanos
muchos ms colores que lo hace tonos de gris), y la eliminacin de los efectos de
fenmenos degradantes (tales como la aberracin del objetivo), y tales tcnicas
tienen
sido de gran importancia en la ciencia, siendo utilizado por los arquelogos,
como-
tronomers, eclogos, gelogos, meteorlogos, oceangrafos, y oth--
ers. El trabajo en la compresin de imgenes multiespectrales ha sido importante
para satel--
lite de sistemas de deteccin, ya que tales sistemas estn limitados por enlace
descendente
ancho de banda de las comunicaciones. 35
La dcada de 1990 han visto la aplicacin generalizada de procesamiento de
seales
en imgenes mdicas, tales como x - ray tomografa computarizada, positrones -
emisin
tomografa, tomografa de impedancia elctrica, formacin de imgenes pticas,
cardiaco
31U.S. Industria Electrnica de Consumo Hoy (Arlington, VA: Electrnica de Consumo Manu ..
fabri- Asociacin, 1997), p. 27.
32U.S. Op Industria Electrnica del Consumidor hoy. cit., p. 21.
33El U. S. Electrnica de Consumo Industria de la opinin: 94 Edicin (Washington, DC:
Electrnico
Asociacin de Industrias, 1994), pp. 19-20. Dolby Pro Logic es un sistema de 4
ehannel: un centro
canal, izquierda y derecha front..channels, y una rear..channel (normalmente
transportan a dos hablan ..
ERS, uno a la izquierda y otro a la derecha, en la parte trasera de la sala).
34U.S. Op Industria Electrnica del Consumidor hoy. cit., p. 59. Tambin hay muchas empresas
la fabricacin de cmaras digitales que no tenan cmaras film..based. Las
cmaras digitales tienen
ha utilizado durante algunos aos en el negocio de la recoleccin de noticias; lo
que es nuevo es su generalizada
mrketing.
35Val D. Vaughn y Timothy S. Wilkinson ", consideraciones Sistemas para
multiespectral
diseos de compresin de imagen"(IEEE Revista procesamiento de seales, vol. 12 (1995),
no. 1, pp.
19-31).
Pgina 146
152
Figura 5. Este ejemplo de la tomografa computarizada es una seccin pro-
abdominal
ducido por CT reconstruccin de haz en abanico. (Reproducido con permiso
del IEEE sig-
Revista nal Processing, marzo de 1997, p. 56.)
formacin de imgenes elctrica, formacin de imgenes por resonancia
magntica, imgenes de ultrasonido. 36 (Vase
Figura 5.) tcnicas de restauracin Image eran importantes en el procesamiento
im--
las edades, desde el telescopio espacial Hubble.
Procesamiento de la seal ha contribuido a una variedad de tcnicas para ex--
tracting informacin de seales. En interferome-- acstica de frecuencia de
barrido
intentar, por ejemplo, una seal acstica se utiliza para observar el interior de los
objetos de metal, tales
como proyectiles de artillera. 37 Basndose en trabajos anteriores, Didier
Massonnet y su col--
ligas de la agencia espacial francesa (CNES) han demostrado que la planta mo--
ciones de unos pocos milmetros pueden ser detectados por las franjas de
interferencia cre--
ado cuando dos imgenes, tomadas desde la misma posicin en diferentes
momentos, son
superpuesto. Esta tcnica, llamada interferometra de radar por satlite, tiene al--
listos demostrado su utilidad en el mapeo de fallas geolgicas y montaas
volcnicas. 38
36See, por ejemplo, cuatro artculos en el nmero de enero de 1997 cuestin de la
IEEE Revista de Procesamiento de Seales
en, respectivamente, las imgenes elctrica del corazn, la tomografa
positron..emission, magnticos
la resonancia, y de imgenes por ultrasonido.
37Scientific Americana, vol. 277 (1997), no. 6, p. 42.
38Didier Massonnet, "interferometra radar Satlite" (Scientific American, vol. 276 (1997),
no. 2, pp. 46-53).
Pgina 147
153
THOMAS KAILATH:
Aqu est un ejemplo de procesamiento de seales en nuestra semiconductores
trabajo. Calentamos una oblea a 1.000 grados muy rpido en una cmara caliente
con halgeno
lmparas. Cmo se mide la temperatura? No se puede tocar la oblea con una
investigar, porque contamina la oblea. Uno tiene que ser indirecta. As que lo que
se dice
is there's radiation coming off the wafer. If you can count the number of photons,
then that's related to the temperature by Planck's law, which says how many
photons are emitted at a given temperature. However, the number of photons is
random, so you get a count that's random. You're interested in something else,
which is the temperature of that wafer as it changes in form. You've got to infer
knowledge about one signal, which is not observable, from another signal, which
is observable. Both of these are random (in the latter case, because of the errors
in measurement), but there's some dependence between them, some statistical
dependence. That's signal processing in its fundamental or generic form. It's ex-
tracting information from signals for other purposes. 1
INTERVIEWER:
Can you tell us about particularly important advances in the last
ten years or so?
JAMES KAISER:
Two words. Chaos theory, and wavelets. There are a lot of
books out on wavelets now, 30 or 40 at least. Wavelets is a wonderful tech-
nique, but it's an analytic technique. It's a very efficient way to represent quite
a broad range of signals, but it hasn't to my knowledge increased our physical
understanding of those signals. It's another tool, but the danger is expecting
that tool to increase your physical understanding of what's going on. 2
JAMES KAISER:
I also want to stress again this is an engineering agenda ver-
sus science agenda issue. It has become so easy to do so much computation
using computers that people will press keys on the keyboard without thinking
what they are doing. I think that there has got to be a resurgence of work on
the scientific end of things. It's so easy to generate a tremendous amount of
garbage that you've got to understand what it is you're doing. So it is very im-
portant that we get back to basic understanding, get a much better grounding
of what science underlies the phenomenon we are looking at. We have got to
be able to do that. I mean, this world is not an ideal world. It's time-varying
and nonlinear. That's the first message.
The next message is that the young peopl~r anybody, really-who are
using these tools have got to thoroughly understand what assumptions underlie
the tool that they are using. That will tell them what they can expect to get out.
You want to know what the guts of the filter are so you can know what you have
filtered out and what will pass through that filter. You've got to know that-is-
sues such as quantization, and, even more so, linearity versus nonlinearity.3
1Thomas Kailath oral-history interview 13 March 1997, p. 22.
2James Kaiser oral-history interview 11 February 1997, pp. 57-58.
3 James Kaiser oral-history interview 11 February 1997, p. 54.
Pgina 148
154
The field of signal processing has continued to make heavy use of
computers, and in the 1990s it became common for computers to assist with
analytical, as well as numerical problems, through symbolic processing and
object,oriented programming. 39 In the 1990s the debt was more than repaid
en la forma DSP aument el rendimiento de los ordenadores personales y
estaciones de trabajo. De hecho, Leonardo Chiariglione ha argumentado que la
seal de pro,
campo cessing y el campo de procesamiento de datos estn convergiendo, como
programable
seal, los procesadores se han convertido en comn y que los equipos utilizan sig
compleja,
procesamiento nal para la entrada, salida y almacenamiento. 40
Por ejemplo, en la dcada de 1990 chips DSP se convirti en una parte estndar
de disco,
unidades de disco y CD, ROM, que se utilizan para controlar el disco y el
la cabeza de lectura. 41 En un primer momento se tom una docena de chips DSP
para ejecutar una unidad de disco. los
nmero se redujo a la mitad a finales de 1990, y los fabricantes ahora objetivo
al ser capaz de ejecutar una unidad de disco con un solo chip.
Este es uno de muchos ejemplos de la utilizacin de DSP en sistemas de control.
Mucho ms importante han sido las mejoras en la computadora entre otras,
caras, sobre todo en los ordenadores multimedia de la dcada de 1990, cuando
chips DSP
manejar la informacin audiovisual y rpidamente, hacindose cargo de esta
tarea desde el
microprocesador principal. 42 La gente ha comenzado a utilizar la voz, software
de reconocimiento
para el control de la computadora. Y procesamiento de seales ha jugado un
papel central en la
crecimiento de Internet y la World Wide Web, como, mdems de alta velocidad
y datos, tcnicas de compresin ampliar enormemente las posibilidades de inter,
computadoras conectadas.
En la dcada de 1990 ha habido un mayor inters en las redes neuronales para
procesamiento de la seal. Ya que ha demostrado ser muy difcil para un
programa para detectar
ojos, narices y bocas con precisin, Takeo Kanade y sus colegas
Carnegie Mellon University han entrenado a un programa de red neural (usando
una gran coleccin de imgenes no impliquen la cara y) para reconocer a estas
caractersticas,
gether en una configuracin apropiada, es decir, de reconocer las caras. 43 La ca,
39Benjamin Monderer, "asistido por ordenador herramientas de procesamiento
de seales para la investigacin" (IEEE ASSP
Magazine, vol. 7 (1990), no. 2, p. 4.
4oLeonardo Chiariglione, "proceso de seales y la estandarizacin (IEEE Signal
Processing
42William Gates, "Computadoras personales en la era de la informacin",
en Britannica Anuario de Ciencia ...
ENCE y el Futuro 1992 (Chicago: Enciclopedia Britnica, 1991), pp. 144-153.
43David Forsythe, Jitendra Malik, y Robert Wilensky, "Bsqueda de imgenes
digitales" (Sci ...
cien- Americana, vol. 276 (1997), no. 6, pp. 88-93).
Pgina 149
155
CHARLES RADER:
Volver algn lugar antes de los aos 80, estaba pensando en los trminos
de salir del laboratorio y convertirse en un profesor en alguna parte, y cuando
se hace eso, les gusta que usted pueda dar una charla. Mont una charla en el
campo de la digi-
procesamiento de la seal tal. Y yo era capaz de poner en una hoja de papel de un
conjunto de lminas superior
ICS y conectar a todos juntos con lneas, mostrando lo que haba llevado a lo
y lo que se relaciona con lo que. Yo no creo que nadie puede hacer eso. primero
sobre todo, si se trat de poner todo en una hoja de papel, no importa cun
grande, hay
sera tantas lneas que se cruzan entre s que no se poda seguirla. 1
ENDERS ROBINSON:
Gran parte del tratamiento de la seal actual se orienta hacia
procesamiento de imgenes. La gente no quiere ver slo texto, que quieren ver
una imagen.
Puede ser que sea esttica, porque bsicamente ests usando el viejo
procesamiento de seales
algoritmos, pero con mucho ms detalle. Pero de alguna manera todo lo que le da
una mejor imagen no es cosmtica. Es algo as como cuando la lupa o el
microscopio o el telescopio entraron. Se poda ver la luna antes, pero cuando
Galileo vio que era una luna diferente. Y eso es una especie de lo que est
sucediendo hoy en da.
Se poda ver imgenes de la tierra, pero ahora se pueden ver casi
como mirar a travs de un telescopio. As que es un juego diferente. 2
HANS SCHUESSLER:
Bueno, para m el ncleo terico era y todava es la esen-
punto de TiAl. Como ingeniero, tengo la suerte de que todo esto se puede
aplicar. Y la in-
fluencia debe ser visto no slo en las regiones industriales, sino para el pblico
tambin.
La primera aplicacin fue un disco compacto de msica. Fue hecho ya 15
Hace aos, un poco ms. Como ingeniero me alegro de ello. Pero la principal
razn para m estar interesado y todava trabajo en el procesamiento de la seal
es slo el
ncleo intelectual en ella. 3
1Charles Rader entrevista de historia oral 27 de de febrero de 1997 pp. 48-49.
2Enders Robinson entrevista de historia oral 6 de marzo de 1997 p. 64.
3Hans Wilhelm Schuessler entrevista-historia oral 21 de abril de 1997 pp. 34-35.
pability de la formacin por ejemplo, para que una red neuronal puede reconocer
una
objeto no es fcilmente definible o identificar una funcin desconocida
basndose en un conjunto
de datos de entrenamiento, que es valioso en muchos contextos. 44
44Neural Redes para el Comit de Procesamiento de Seales, "El pasado,
presente y futuro de los nervios
redes para proceso de seales"(IEEE Revista Signal Processing, vol. 14 (1997), no. 6, pp.
28-48).
Pgina 150
156
Las redes neuronales han demostrado ser tiles para rechazo de interferencias en
alambre ...
less communications.45Neural networks for aircraft control and for automo...
bile...engine control are being developed.46Simon Haykin has demonstrated
several ways neural networks can improve the processing of radar images.47
Neural network techniques have helped in image restoration and, because of
their highly parallel nature, led to efficient VLSI architectures for image
restoration.48The VLSI implementation of neural networks is exemplified
also by the work of Carver Mead and others in building models of biological
signal...processors, such as the retina and the cochlea; a silicon retina built by
Misha Mahowald and Carver Mead displayed behavior quite similar to that
of the human retina, including susceptibility to certain optical illusions.49
Among the most exciting developments of the past decade have been
the application by SP engineers of the concepts of fractals, chaos, and
wavelets. Fractals are objects in geometry that have non...integral dimen...
sion. Fractal coding has been extensively applied in image compression.50
(See Figure 6.)
In physics, a chaotic system is a mathematically defined system that
evolves in a seemingly random way, neither approaching a steady state nor
cycling through a sequence of states.51First recognized in meteorology in
the 1960s,52 chaotic systems have since been discerned in astronomy, bioI...
45Jeff D. Laster and Jeffrey H. Reed, "Interference rejection in digital wireless
communica..
tions" (IEEE Signal Processing Magazine, vol. 14 (1997), no. 3, pp. 37-62).
46Neural Networks for Signal Processing Committee op. cit.
4 7 Simon Haykin, "Neural networks expand SP's horizons" (IEEE Signal Processing
Magazine,
vol. 13 (1996), no. 2, pp. 24-49).
48Mark R. Banham and Aggelos K. Katsaggelos, "Digital image
restoration" (IEEE Signal
4 9 Misha Mahowald and Carver Mead, "The silicon retina" (Scientific American, vol. 264
(1991), no. 5, pp. 76-82).
50See, for example, Arnaud E. Jacquin, "Image coding based on a fractal theory
of iterated
contractive image transformations" (IEEE Transactions on Image Processing, vol. 1 (1992),
pp. 18-30), and Tim Bedford, F. Michel Dekking, Marcel Breeuwer, Michael S.
Keane, and
Daan van Schooneveld, "Fractal coding of monochrome images" (Signal Processing:
Image
Communication, vol. 6 (1994), pp. 405-419).
51Fractals and chaos are related. The behavior of a dynamical system may be
described by a
trajectory in state space, and that trajectory delimits a mathematical form called
an attrac..
tor. A system is chaotic exactly when its attractor is a fractal, in which case it is
called a
"strange attractor".
Academic Press, 1995), pp. 188-194.
Pgina 151
157
r= 128, N= 27
FIGURE 6. A fractal that has found application in signal processing is
Barnsley's
fern.
ogy, chemistry, and other sciences. Recently, chaotic models have been em..
ployed in signal processing. For example, Simon Haykin modeled sea clut..
ter (that is, the radar backscattering from an ocean surface) as a chaotic
process in order to build a neural network for use in canceling sea clutter. 53
Wavelets are a mathematical decomposition technique, which may be
viewed as an extension of Fourier analysis. The theory underwent consider..
able development in the mid 1980s, and it was soon applied to digital signal
processing by Ingrid Daubechies and Stefan Mallat. 54 Besides stimulating
some new approaches to signal processing, wavelet theory provides a uni..
fied framework for techniques already developed for particular SP applica..
53Simon Haykin op. cit.
54Ingrid Daubechies, "Orthonormal bases of compactly supported
wavelets" (Communications
in Pure and Applied Mathematics, vol. 41 (1988), pp. 909-996), Stefan MaHat, "A theory for
multiresolution signal decomposition: the wavelet representation"
(IEEE Transactions on
Pattern Analysis and Machine Intelligence, vol. 11 (1989), pp. 674-693), and Stefan MaHat,
"Multifrequency channel decomposition of images and wavelet models"
(IEEE Transactions
on Acoustics, Speech, and Signal Processing, vol. 37 (1989), pp. 2091-2110).
Pgina 152
158
ciones. 55 Among the areas where wavelets have been successfully applied are
image processing, speech processing, data compression, subband coding,
multiresolution analysis, and wideband correlation processing. 56
There have been other new departures for signal processing in the
1990s. One example is work, which has led to commercial application, on
so...called fuzzy algorithms, which use fuzzy logic. This type of logic, in...
vented by Lotfi Zadeh in the 1960s, is based on the concept of a fuzzy set,
for which membership is expressed in varying degrees. 57 Another example is
work on so...called genetic algorithms, which use a searching process mod...
eled on the laws of genetics and natural selection. Genetic algorithms have
been applied as an optimization tool in many areas of signal processing. 58
The widespread use of object...oriented programming for DSP might also be
mentioned. 59
One sign of the fecundity of signal processing is its growth into other
areas of engineering, such as control systems, imaging systems, and instru...
mentacin. De hecho, la amplia aplicabilidad de sus conceptos y tcnicas es
erosionando las fronteras entre las disciplinas. 6o teora de la informacin, la
codificacin de la ...
ORY, teora de la comunicacin, y la teora de redes neuronales estn
estrechamente re ...
lated para sealar la teora de procesamiento.
550livier Rioul y Martin Vetterli, "Wavelets y procesamiento de seales"
(IEEE proceso-- Seal
ING Magazine, vol. 8 (1991), no. 4, pp. 14-38).
56Lora G. Weiss, "Wavelets y el procesamiento de correlacin de banda ancha"
(IEEE Signal Processing
57Craig Marven y Gillian Ewers, un enfoque sencillo para Procesamiento Digital de
Seales (Nueva York:
John Wiley & Sons, 1996), p. 7.
58Kit S. Tang, F. Kim Man, Sam T. Kwong, y P. l, "algoritmos genticos y sus
apli ..
cationes"(IEEE Revista Signal Processing, vol. 13 (1996), no. 6, pp. 22-37). En 1996 la
IEEE estableci un diario, IEEE Transacciones sobre la computacin evolutiva, en esta rea.
59Matti Karjalainen, "la integracin de software de programacin de DSP por
object..oriented: un caso
estudio de "QuickSig (IEEE Revista Procesamiento de Seales, vol. 7 (1990), no. 2, pp. 21-31).
60See, por ejemplo, Arogyaswami Paulraj, Vwani Roychowdhury, y Charles D.
Schaper,
eds,. Comunicaciones, Cmputo, Control y Procesamiento de Seales: Un tributo a Thomas
Kailath (Boston: Kluwer Academic Publishers, 1997). Un ejemplo es el uso de la mini ..
formulacin mximo introducido en la teora de control hace una dcada, que
recientemente se ha utilizado para
mostrar la optimalidad de la ampliamente utilizada, pero generalmente
considerado como subptima-least..mean ..
cuadrados algoritmo de filtrado adaptativo de Widrow y Hoff [Babak Hassibi,
Ali H. Sayed, y
Thomas Kailath, "optimalidad H..infinity del algoritmo LMS"
(IEEE Transacciones sobre la seal
Processing, vol. 44 (1996), pp. 267-280)]. Otro ejemplo es el uso de un Kalman estocstico
formulacin de filtrado para unificar el campo de rpido crecimiento de filtrado
adaptativo determinista; ver
Ali H. Sayed y Thomas Kailath, "Un enfoque state..space al filtrado adaptativo
RLS"
(IEEE Revista Signal Processing, vol. 13 (1996), no. 1, pp. 18-60).
Pgina 153
159
Otro signo de la vitalidad de procesamiento de seales es la reciente prolifera ~
cin de siglas. Los autores de ~ procesamiento de seales textos han sentido la necesidad de
proporcionar una lista de acrnimos, como lo han hecho, incluso el artculo ~
escritores de vez en cuando. (Ver
Figura 7.) Hay incluso segundos ~ orden acrnimos, tales como FAM para "FFT
Acumulacin Mtodo" o AAL para 'capa de adaptacin ATM' (ATM
bienestar modo de transferencia asncrono). (Los acrnimos son un 20 ~ phe siglo ~
Nomenon, provocado por la multiplicacin de los organismos humanos, ya sea por
lo ~
cial, como los muchos commissions y juntas de nuevo de Franklin Roosevelt
Deal, o tecnolgico, como dispositivos o procedimientos.) 61
El 1 de enero de 1990, la Acstica, Habla y Procesamiento de Seales As ~
ciedad cambi oficialmente su nombre por el procesamiento de seal Society.62
Su
continu el crecimiento, en nmero de miembros, en la publicacin, y en las
actividades.
Su publicacin principal, Transacciones, se dividi en tres: Transacciones en seal
Procesamiento, Transacciones en procesamiento de
imgenes, y Transacciones en Habla y
Procesamiento de audio. Esto dio lugar a un aumento en el nmero de pginas pub ~
cado y una reduccin en la acumulacin de artculos en espera publication.63The
La sociedad adquiri un ~ a tiempo completo del personal profesional, y la Misericordia
fue Kowalczyk
contratado como Delegado. La Sociedad se hizo cada vez ms internacional,
con muchos ms miembros y muchas ms conferencias y talleres en
Europa y Asia.
En la dcada de 1990 la Sociedad ha llevado a cabo muchas nuevas
empresas. Una seg ~
ond importante serie de conferencias internacionales, la Conferencia
Internacional sobre
Procesamiento de Imgenes (ICIP), comenz en 1994. Ese mismo ao la
Sociedad de la ini ~
ATED cartas Procesamiento De Seales para la rpida publicacin de los resultados
importantes. Un nuevo
61The New Deal trajo siglas en conocimiento del pblico-NRA (Recuperacin
Nacional
Administracin), la CCC (Cuerpo de Conservacin Civil), la TVA (Tennessee
Valley
Autoridad), la REA (Administracin de Electrificacin Rural), el WPA (progreso
de trabajos anuncio ..
ministracin), y muchos otros, y se llam al gobierno alphabet..soup. Vale la
pena
sealando que los cientficos manejan la plenitud de los mundos biolgicos y
geolgicos con
nombres de tipo tradicional. Se podra argumentar que, a diferencia de los objetos
presentes en la naturaleza y
materiales, instrumentos inventados por los seres humanos son bastante ilimitada
en nmero, de ah el
la necesidad de acrnimos para evitar nombres inconvenientemente largas.
procesamiento 62Image se haba convertido en un campo grande dentro de la
Sociedad y se sigue creciendo
rpidamente, pero aadiendo 'imgenes', a un nombre ya larga no parece
prudente. Entonces el
Sociedad decidi simplificar el nombre, el uso de 'proceso de seales' como un
trmino all..inclusive.
[Comunicacin personal David Munson 5 de febrero de 1988.]
63Tariq Durrani, "el mensaje del presidente" (IEEE Revista de Procesamiento de
Seales, vol. 11 (1994), no.
4, pp. 8-10). Inicialmente, la circulacin se redujo porque antes Transacciones haban
llegado auto ..
ticamente con miembros de la Sociedad; con la divisin de la revista, la
suscripcin al dif ..
rentes transacciones se hizo opcional.
Pgina 154
160
Tabla 1. Abreviaturas utilizadas en este artculo
Aida . . . . . . . . . . . . . . . . . . . Anlogo a digital
ACIA
La interferencia de canal adyacente
ADC
Conversor analgico a digital
ADFA. . . . . . . . . . . . . . . . AdaptiveDigitalFilter
AEQ. . . . . . . . . . . . . . AdaptiveLinearEqualizer
AGC. . . . . . . . . . . . . . . . Control de ganancia automtica
AIC. . . . . . . . . . . . AdaptiveInterferenceCanccler
Alea. . . . . . . . . . . . . . . Adaptativo Lnea Enhancer
AM. . . . . . . . . . . . . . . . . Amplitud modulada
AMPS. . . . . . . . . . Sistema Avanzado de Telefona Mvil
ANC. . . . . . . . . . . . NonlinearConverter adaptativo
ANLE. . . . . . . . . . . adaptativo NonlinearEqualizer
ARA. . . . . . . . . . . . . . . . . . . . . autorregresivo
ATFA. . . . . . . . . . . . . . AdaptiveTime-Frecuencia
AWGN. . . . . . . . . . AdditiveWhite ruido gaussiano
B-CDMA. . . . . . . . . . . . . . .. BroadbandCDMA
BER. . . . . . . . . . . . . . . . . . . . . Tasa de error de bit
PAB. . . . . . . . . . . . . . . . . . . . Filtro de paso de banda
CCI. . . . . . . . . . . . . . . . La interferencia cocanal
CDMA. . . . . . . . . . codigo de DIVISION DE ACCESO multiple
ACMA. .
. . . ConstantModulus Algoritmo
CNNDFF
Complejo Neural-Network-Based
Adapive filtro DF
Cofa . . .
. . . Filtro de cdigo ortogonalizacin
CPMA. . . . . . . . . . . . La modulacin de fase continua
CWA. . . . . . . . . . . . . . . . . . . Ola continua
DEDS. . . . . . . . . . Sistema dinmico de eventos discretos
DFA. . . . . . . . . . . . . . . . . . . DecisionFeedback
DFE. . . . . . . . . . . . . DecisionFeedbackEqualizer
DPSK
Diferencial de modulacin por desplazamiento de fase
DS. . . . . . . . . . . . . . . . . . . . . DirectSequence
DSSS . . . . . . . . . . . . . . . . . DS Spread Spectrum
FDMA. . . . . . . . . . Frecuencia-DivisionMultiplexing
FDMA. . . . . . . Frecuencia-DivisionMultipleAccess
Fera. . . . . . . . . . . . . . . . . . . FrameErrorRate
FFH. . . . . . . . . . . . . . . FastFrequency Hopping
FH. . . . . . . . . . . . . . . . . '. . El salto de frecuencia
Fimma . . . . . . . . . . Interactuar rpida Modelo mltiple
FIR. . . . . . . . . . . . . . . . Respuesta de impulso finito
FLAA. . . . . . . . '. . . . . . . FastLearning Algoritmo
FM. . . . . . . . . . . . . . . . . Modulacin de frecuencia
Fresha. . . . . . . . . . . . . . . . . desplazamiento de frecuencia
FSBLP. . . '. . . Fraccionalmente Spaced BilinearPerceptron
FSDFMLP
Fraccionalmente Spaced DF
Perceptrn multicapa
FSE. . . . . . . . . '' . . Fraccionadamente-SpacedEqualizer
FSRPP. . . ... . . . . . . . . . . fraccionadamente Spaced
Recursiva polinmica Perceptrn
FSK
Frecuencia de modulacin por desplazamiento
GFC. . . . . . . . . . . Gradiente-SearchFastConverging
PRVG. . . . . . . . . GeneralizedLikelihood-RatioTest
GMSK. . . . . . . . . GaussianMinimum ShiftKeying
GPSA. . . . . . . . . . . . . . Sistema de Posicionamiento Global
Televisin de alta definicin. . . . . . . . . . . . . -Alta DefinitionTelevision
HOS
Estadsticas de orden superior
ICEA. . . . . . . . . . . Interferencia-CancelingEqualizer
IMM. . . . . . . . . . . . . . InteractingMultipleModel
IPAA. . . . . . . . . . Anlisis de la perturbacin infinitesimal
IS-54
Medio Estndar - 54
lsia. . . . . . . . . . . . . . . . La interferencia entre smbolos
J/S
Relacin de emisin de seal-a-
LCCM. . . . . . linealmente ConstrainedConstantModulus
LEO. . . . . . . . . . . . . . . . . . Bajo EarthOrbiting
LFSE. . . . . . . . LinearFractionally SpacedEqualizer
LMS
Menos cuadrada media
LO. . . . . . . . . . . . . . . . . . . . . localmente ptima
LPPA. . . . . . . . . . . . LatticePolynomial Perceptrn
LPF. . . . . . . . . . . . . . . . . . . . . Filtro de paso bajo
LS. . . . . . . . . . . . . . . . . . . . . . mnimos cuadrados
LTEA. . . . . . . . . . . . . LinearTransversalEqualizer
MAA. . . . . . . . . . . . . . . . . . . . Acceso multiple
MAl. . . . . . . . . . . . . Interferencia de Acceso Mltiple
MF. . . . . . . . . . . . . . . . . . MisadjustmentFilter
MLA. . . . . . . . . . . . . . . . . Mxima verosimilitud
MLSE. . . . . . . . . . MaximumLikelihoodSequence
Estimacion
MMSE. . . . . . . . . . . MinimumMean SquaredError
M-QAMA . . . . . . . . . . . . . . . . . QAM multinivel
NBIA. . . . . . . . . . . . . . . NarrowbandInterference
Neda
Deteccin de envolvente normalizada
NTSC. . . . . . . National Television System Committee
Osa
Estadsticas de orden
OTDR. . . . . . . . . Tiempo ptimo-DependentReceiver
PCS. . . . . . . . . . PersonalCommunicationsSystems
PSKA. . . . . . . . . . . . . . . . . . Phase Shift Keying
PN
Pseudo-ruido
pp.. . . . . . . . . . . . . . . . . polinmica Perceptrn
QAMA. . . . . . . . . . . . . . . . . . . QuadratureAM
QPRS. . . . . . . QuadraturePartial respuesta de sealizacin
QPSK. . . . . . . . . . . . . . . . . . . QuadraturePSK
RBF. . . . . . . . . . . . . . ... Funcion de base radial
SPI. . . . . . . . . . . . . . . . RecursiveLeastSquares
SDR. . . . . . . . . . . Simtrica de reduccin de dimensiones
SEOR . . . . . . . . . . . . . . Relacin seal a interferencia
SINR. . . . . . . . . . Relacin Seal-a-InterferenceNoise
SNR. . . . . . . . . . . . . . . . . Seal-a-NoiseRatio
Sol
Seal de intereses
SNOI. . . . . . . . . . . . . . . . No seal de intereses
Soma . . . . . . . . . . . . Auto-OrganizingFeatureMap
SPREIS. . . . . . . . . Explotando SpectralRedundancy
Supresor de interferencias
SSMA. . . . . . . . . Spread Spectrum Acceso Mltiple
SSMF. . . . . . . . . . SpreadSpectrumMatchedFilter
SSA
Espectro ensanchado
TDAF
Dependiente del tiempo de adaptacin FilterTDL
Aprovechado-Delay Line
TDMA. . . . . . . . . . Tiempo de Acceso Mltiple por Divisin
TFDA
Las distribuciones tiempo-frecuencia
LA A . . . . . . . . . . . . . . . . . . ThresholdExcision
VSIE. . . . . . . . . Vector-SpaceInterference Escisin
WF. . . . . . . . . . . . . . . . . . . . . . Filtro de Wiener
WHT. . . . . . . . . . . . . Walsh-HadamardTransform
FIGURA 7. Jeff D. Laster y Jeffrey H. Reed proporcionado este listado de acrnimos para
su artculo sobre "el rechazo de interferencias en las comunicaciones
inalmbricas digitales" puede ser .64It
considerado como un ejemplo de codificacin de voz especfico de la disciplina.
64Laster y Reed op. cit.
Pgina 155
Captulo 7 Breakout: El 19905
161
comit tcnico, el procesamiento de seales multimedia, se estableci en
1996. Y la sociedad se ha embarcado en la publicacin electrnica, tanto en lnea
y en CD, ROM.
La nuestra es una era de la informacin. Gran parte de la economa y una gran
parte de
La experiencia de un individuo depende de forma electrnica, informa mediada,
cin fluye. Incluso en trminos ms generales, el rpido ritmo y el alcance
mundial de todos
mdem-economa de la vida, la poltica, el entretenimiento, la ciencia - dependen
de elec,
comunicaciones Tronic. Y virtualmente todos estos flujos de informacin en,
Volve el trabajo de los ingenieros de seales, procesamiento. Lo que es ms, ya
no lo es
procesamiento de seales de una especialidad tcnica oscura. En la dcada de
1990, Stephen
Hawking (un fsico que, a causa de una enfermedad neuromotora, es capaz de
hablar
slo a travs de un sintetizador de voz) y sin cintas contestadores automticos
hacen escala en,
tention para sntesis de voz. Los anunciantes promocionan el nmero de pxeles
en imgenes
desde cmaras digitales. Las personas que usan Internet discuten las tasas de
mdem y
compresin de datos. Y software ampliamente disponible permite a cualquier
persona
para manipular la msica o mejorar las imgenes. En resumen, despus de 50
aos de desarrollo,
cin, procesamiento de seales ha asumido un papel grande y prominente en el
mdem
vida

Libro de Pds Traducido

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Libro de Pds Traducido

Uploaded by

Copyright:

Available Formats

Pgina 1

1James Kaiser, entrevista de historia oral, 11 de febrero de 1997, p. 36.

23Norbert Wiener, extrapolacin, interpolacin y suavizado de Estacionaria serie de tiempo con

marca y Columbia hicieron importantes innovaciones tecnolgicas, incluyendo

1Ben Gold oral-history interview 15 March 1997, p. 3.

1 Enders Robinson oral-history interview 6 March 1997, p. 18.

Going Digital: The 1960s

x[1] o--~-o-----I---(,)--"",,--o---A........--(::>--........ ----<:>--t----t-.o x[4]

lJames Kaiser oral-history interview 11 February 1997, pp. 6-7.

1 Hans Georg Musmann entrevista de historia oral 30 de agosto de 1994, p. 8.

1 Lawrence Rabiner oral-history interview 13 November 1996, pp. 8-9.

1 Charles Rader oral-history interview 27 February 1997, p. 4.

"Mara tena un cordero"

1Thomas Kailath entrevista de historia oral 13 marzo de 1997, p. 32.

1Bede Liu oral-history interview 10 April 1997, pp. 25-26.

1 Lawrence Rabiner oral-history interview 13 November 1996, p. 41.

You might also like