rle , illav %'iav46 i'i · 2019-11-20 · ingalgorithmsforspeech,image,and...
TRANSCRIPT
RLEI'I_,_Ill AV %'I AV 46
Volume 2, Number 2 " June 1989
The Research Laboratory of Electronics at the Massachusetts Institute of Technology
DIGITAL SIGNAL PROCESSING AT RLE:Beyond the Realm of the Compact Disc
For audiophiles, perhaps the most
commercially visible application ofmodern digital signal processing is the
highly acclaimed compact disc. Digitalinformation is stored in microscopicbumps andsmooth areas on the surfaceof an aluminized plastic disc. The discitself measures only 120 millimeterswide and 1.2 millimeters thick, and its
500-megabyte capacity can store up to74 minutes a side. The laser beammechanism contained in the compactdisc player is equivalent to the conven-tional record player stylus. Behind the
technology used to produce these su-
perior recordings are the digital signalprocessing techniques developed byscientists to dramatically enhancesound. Butthese techniques are not
just limited to acoustic applications inthe home entertainment market. Whatis going on behind the advances thathave become so familiar to us as today'shigh-tech consumers?
Digital signal processing is the ma-
nipulation of digitally represented sig-nals by means of combining sophisti-cated algorithms implemented inmodern integrated-circuit digital com-
puter technology. By sampling data
(from speech or image sources, for ex-
ample) and applying signal processingprocedures (such as spectral analysisor filtering), researchers can estimateand transform a signal's features. Some
techniques are used to measure charac-teristics which suggest further analysisof a specific system. Other techniques,such as filtering, remove noise or re-
cover meaningful data from degradedsignals. Once processed into a more
workable form, researchers can then
study the features, components, orsource of a signal.
The Basic Science of SignalsSignals can he represented in either
analog or digital form. Analog signals
(continued on pg. 2)
F7*V"__Ow",
Onegoal ofthe research activities in RJ.Es Digital Signal l'rocLcsnig Group G the
development ofan algorithm design environmentbasedon the symbolic rearrange-ment ofsignaiprocessing expressions. The current version ofasystem, demon-strated above by Professor Alan V. Oppenheim, has been designed and implement-ed on a LISP machine bygraduate student Michele M. (Jovell aspart ofherdoctoral thesis.
Director's Message
Building on long-standing ex-
pertise in signal processing, an in-
creasing experience with contem-
porary digital computers, and
strong ties to industrial researchlabs and MIT Lincoln Laboratory, itwas natural for RLE investigators toassume a central role in the earlyand continuing evolution of digitalsignal processing. Although the
digital computers of the 'GUs werenot very powerful by today's stan-dards, RLE researchers were quickto exploit their capabilities, and tobenefit from the relative ease of
experimenting with a variety of sig-nal processing systems in many ap-plication areas. Without the needto build analog signal processingsystems from physical compo-nents, systems could he readilysimulated. Problems of componenttolerance and aging, as well as
parametric variations with tem-
perature and voltage, disappearedand new skills for programmingsignal processing algorithms wereneeded. Spurred by the burgeon-ing digital electronic technology,newalgorithms in efficient formsfound many uses, and digital pro-cessing became an essential part ofan electrical engineer's training.
Since the beginning of digitalsignal processing research, algo-rithms and implementation archi-tectures have grown in a synergis-tic way. Fruitful connections with
computer science and moderncontrol theory have expanded therich set of techniques and capabili-ties, and there has been steadygrowth in a wide variety of applica-tions. Spectral estimation, vocod-
ing, imageprocessing, seismic andunderwater signal processing,knowledge-based signal process-ing, signal recovery from degraded
ProlessorJonathan Allen, Director
Research laboratory ofElectronics
forms, fault-tolerant signal pro-cessing architectures, and innova-tive, highly parallel computing sys-tems are some of the areas wherethere has been intense research ac-
tivity and correspondingly signifi-cant results. The ambientcomput-ing technology has provided not
only the high-performance frame-work fornewand innovative algo-rithms, buthas also led to theconsideration of new design tech-
niques that efficiently explore the
large space of design alternativesafforded by space-time trade-offs.
RLE is proud of its outstandingexpertise in the digital signal pro-cessing area, and the increasingimpact that our research is havingon many ubiquitously found sys-tems. We look forward to new and
powerful results in the future, andan increasing presence of these
techniques throughout the Labo-
ratory's activities.
DIGITAL SIGNAL(continued)
are continuous and canbe illustrated asa curve which indicates the amplitudeofa signal as a function ofa continuous,
independent variable such as time.Some analog signals repeat with a cer-tain frequency, measured in hertz, andeach repeated cycle has a duration,known as its period. Analog signalsarise from awide variety of natural andman-made sources, such as earth-
quakes (seismic) or radio broadcasttransmitters (electromagnetic). Manyanalog signals are essentially band-limited, that is, all their energy is con-tained below a certain cut-off frequen-cy. In these cases, all of the signal'sinformation canbe represented by ase-
quence of numbers obtained by sam-
pling it at a frequency that is at or abovetwice its highest frequency. This is a
consequence of the fundamental sam-
pling theorem that allows scientists to
easily make transformations betweenthe analog and digital domains.
When engineers and scientists
sample analog signals such as speech,they must first transform the analog sig-nals into a series of correspondingnumbers which accurately representtheshape of the analog waveform. The
sampling process quantizes the con-tinuous analog signal values into alimited set of discrete binary numbers,each of which comprisesa string of bi-
nary digits, or hits. Once an analog sig-nal is converted into the digital do-main, computers can analyze the fre-
quency spectra and other character-istics ofasignal by using awide varietyof digital signal processing algorithms.
Algorithms are effective proce-dures, usually represented as computerprograms, whichare used to solve sig-nal processing problems such as digitalfiltering, correlation, and signal param-eter estimation. Many applications re-
quire the input sampled sequence tohe transformed to the frequency do-main, and this is accomplished throughthe use of the discrete Fourier trans-form (DFT) algorithm. In the early his-
tory of digital signal processing, the
computational expense of the DFT wasenormous, but in 1965, James CooleyandJohn Tukey invented the fast Four-ier transform (FFT) algorithm, which
computed the same output values astheDFT but at amuch faster rate. Inter-
estingly, the FFT can be divided into
many smaller processes that can he
Source Frequenc
RECONSTRUCTION FROM THRESHOLD CROSSINGS
ORIGINAL
Under appropriate constraints, an im-
age can be exactly represented by its
crossings ofa single threshold Shownabove are the original image ofanangiogram (upper left) and The thresh-
old crossing representation (upper
right). The recovered image (below) is
obtained by applying a recently devel-
oped reconstruction algorithm to the
threshold crossing. In general, signal
processing techniques have played an
important role in the development ofmedical technology. Research atRLEhas includeda variety ofimage repre-sentation andprocessing techniquesapplied to medical images such as theestimation of coronary artery bound-aries in angiograms. Other medical ap-plicationsfor signalprocessing devel-opedat RLE include the measurement ofbloodflow characteristics by analyzing the
spectrum ofheartsounds using an active ultrasonic measurement system.
THRESHOLD CROSSINGS
RECOVERED
Signalprocessing algorithms are being
developed at RLE toprocessperiodicgravitational wave datafor which thesource location andfrequency are un-
known. Gravitational waves are gener-ated when matter in the universe
moves, causing a change in its gravita-tionalfield The algorithms, developed
byProfessor Alan V. Oppenheim and
graduate student Tae H. joo, employtwo-dimensional data processing in lo-
cation-frequency space, and combinemeasures ofcoherence in multichan-
nd data. This work has application tomoregeneralproblems ofnarrowband
signal detection in wideband noise,
particularly when there is relative mo-tion between transmitter and receiver.
Thisproject is carried out in collabora-tion with Professor Rainer Weiss ofMIT's Physics Department, who built the1. 5-meterprotot-pe interferometricgravitational wave detector.
computed simultaneously. This has ledto the design of special-purpose com-
puters for digital signal processing thatare substantially faster than common
single-sequence computers. Althoughsome signal processing algorithmsare very flexible computer programsthat can be used for a variety of appli-cations, new algorithms must constant-
ly be developed since there are manydifferent signal processing techniques,and a single algorithm cannot be usedin all cases.
In the mid-'60s, the held of digitalsignal processing evolved rapidly when
computers and digital circuitry becamefast enough to process large amounts ofdata. The advent of digital computersfor signal processing provided the
flexibility to simulate signal processingsystems before their implementationon analog hardware, thus introducinggreater accuracy and cost effectiveness.
Widespread signal processing applica-tions resulted from the increased
speed of computers, the introductionof fast array-processor peripherals, andthe development of the microproces-sor.
Many signal processing applica-tions that require high sampling rates,
high production volumes, or low costcan only be satisfied by special-purposedigital hardware. The useof these sys-tems did notbecome widespread untilcustom VLSI circuits made them an at-tractive alternative to existing tech-
nologies. Digital signal processing sys-tems are well-suited to VLSI imple-mentation since they have highlyparallel algorithmic structure, local
connectivity, and circuit modularity.
Practical Applications of SignalProcessingIn the 1950s, signal processing meth-ods for seismic imagingwere devel-
oped by the oil industry to determinethe presence of oil andgas depositsdeep in the earth. One imagingmethod
exploded underground dynamitecharges and recorded the vibrations asseismic traces on a seismograph. Theseismic traces were then examined forarrival times and strength of the vibra-tions to determine the structure of
sedimentary rock layers and whetheroil was present. Seismic data analysis isalso used in earthquake measurementsand nuclear test monitoring.
Today, signal processing has manypractical applications:"Speech signals can he synthesizedfrom discrete-time models of the hu-man vocal tract using digital signal pro-cessing. The speech synthesizer is a
time-varying digital filter which modelsthe response of the articulators
(mouth, throat, etc.) to excitation fromthe vocal cords and air turbulence. Thisresults in quality voice synthesis for
computers, and data reduction for the
digital voice communications used in
talking consumer products. Analysis of
speech production can also be per-formed by examining thespeech sig-nal's spectrum and its related proper-ties. These techniques can be used todetermine the spectral properties of
speech (such as the voiced sounds invowels or the unvoiced sounds in frica-tives) andthe distance between pitchpulses in voiced sounds. Sounds canalso be analyzed by examining their
spectra. Once obtained, this informa-tion can he useful in the developmentof an automatic speech recognition or
compression system." Signal processing is central to manydefense-related systems, and plays a
major role in automatic control, naviga-tion, and target tracking. Radar and so-nar systems rely heavily on the applica-tion of signal processing technology." In neurophysiology, signal processingtechniques are used to mathematicallycharacterize signals obtained by elec-
troencephalography from the scalp of a
subject. Frequency spectrum analysis isused to reveal the possible presenceof energy prominences at certain fre-
quencies, whichare important to physi-cians for diagnostic purposes.*Two-dimensional or image process-ing is used to analyze visual data suchas x-rays, aerial photographs, and satel-lite transmissions from space. With theadvent of high-definition television,new image processing algorithms are
being developed to enhance picturequality, while new techniques are be-
ing investigated to improve the quality
of existing transmissions. The potentialof high-definition television coupledwith image processing technology has
opened thewayfor the all-digital televi-sion sets of the future.
RLE's Digital Signal ProcessingGroupDuring the '60s, many areas of research
capitalized on the development of
computers by redefining traditional
disciplines within a new computationalenvironment.An example of this evolu-tion was the broadening of studies incontinuous linear and nonlinear sys-tems to discrete systems. At RLE, the
rapid growth ofdigital signal process-ing led to new techniques in speechsuch as spectral estimation.
Today, RLE's Digital Signal Pro-
cessing Group pursues basic researchin signal processing and algorithm de-
sign with applications to speechprocessing, image andvideo process-ing, and underwater acoustics. One ofthe group's primary focuses is to devel-
op newsignal processing algorithmsand to explore their characteristicswhen applied to real and synthetic sig-nals. These include algorithms fornoise suppression and bandwidth re-duction, robust model estimation, timeand frequency scale modifications, ho-
momorphic analysis, and the recoveryof signals from limited information. Al-
though the development of new algo-rithms is a major focus, the group alsostrives to maintain close ties to practi-
Estimaic of Pitch (dog) Estimate of Yaw (dog)
-to
0
S
In side-scan sonar imaging, 4ypi-cally the transmitter and receiverundergopitch, roll, andyawasthey are towed through the water,which results in distortion ofThedesired image. Professor Alan VOppenheimandgraduate studentDaniel F. cobra, in collaborationwith scientists at the WoodsHoleOceanographic institution, devel-op andapply algorithms to esti-mate andthen correctfor this un-wantedmotion. Shown is an
original side-scan image taken inVineyard Sound(photograph) withestimates ofpitch andyaw(graphs)that were obtained directlyfromthe image.
Fractalsignatures ofstate evolution in a chaotic digitalfilter are illustratedabove. In certainfinite-precision environments; the implementation ofdigitalfil-tencan exhibit this type ofchaotic behavior. ProfessorAlan V. Oppenheim andgraduatestudent GregoryW. Wornellexamine the various applications ofbothchaotic dynamical system theory andfractalprocess theory to signalprocessing.This research couldpotentially lead to new modelsfor quantization, speech gen-eration, andsignal coding.
cal applications and implementation is-sues, because algorithm efficiency not
only depends on the number of re-
quired operations in a signal process-ing procedure, but also the algorithm'ssuitability to the computer architectureit will eventually run on.
Professor Alan V. Oppenheim isconcerned with fundamental issues in
signal processing. His primary interestis in the development of signal process-ing algorithms for speech, image, andseismic data. Onemajor focus of his re-search is geophysical signal processingapplications, including the develop-ment of new algorithms for wave
propagation, analysis of data collected
by seismic arrays, and the solution ofinverse problems to infer underlyinggeological or oceanographic struc-tures. In earlier work, Professor Op-penheim, in collaboration with theWoods Hole Oceanographic Institu-tion, developed an algorithm that esti-mates the acoustic reflection coeffi-cient at the ocean bottom. His studiesin seismic digital signal processinghave led to the development of several
significant new techniques for seismicdata analysis, such as cepstral analysisand homomorphic deconvolution.
Currently, he is workingon thetransformation of side-scan sonar datain order to extract topographic infor-
mation from sonographs. Side-scan so-nar, invented by Professor Harold Ed-gerton of MIT, has been used since the
early '60s as an important tool in under-water exploration. It has enabled ma-rine geologists to survey the oceanfloor, to locate objects at the bottom ofthe sea, and to prospect for mineral de-
posits. The side-scan sonar collects re-flected underwater soundwaves andproduces agraphic record (sono-graph) of topographic features and the
reflectivity of materials on the sea bed.But, since various distortions preventthe sonographs from presenting an ac-curate depiction of the topology, digitalsignal processing techniques are beingapplied to the sonographs to enhancethe data.
Although signal processing has
historically emphasized numerical
techniques, knowledge-based signalprocessing is a new direction whichcombines the heuristic knowledge of
speech and other signals with moretraditional signal processing tools and
concepts. Recent work at RLE hasfo-cused on knowledge-based pitch detec-tion, signal enhancement, and distrib-uted sensor networks. One areaaddressed by Professor Oppenheini isthe symbolic representation and ma-
nipulation of knowledge and expres-sions in signal processing. The issuesassociated with this work include theuniform representation of knowledge,derivation of new knowledge from
knowledge that has already been pro-vided, and strategies to control the useof this knowledge. Also being investi-
gated is the problem ofsignal matchingusing multiple levels of description inthe context ofvector coding of speechsignals. This work seeks to reduce
matching complexities by using multi-level signal representations that aresimplified by abstraction of detail. Po-tential applications for knowledge-based signal processing are "intelligentsignal processing" oriented toward sig-nal understanding and characteriza-tion, "expert" systems capable of supe-rior performance, image storage,pattern recognition, and signal pro-cessing system design environments.
In addition, Professor Oppenheimhas applied newdigital signal process-ing techniques to speech processing.This work is aimed at low-cost, high-quality compression and enhancementof degraded speech, and algorithmdevelopment for robust speech com-
pression in the presence of additivenoise.He has also investigated adaptive
ProfessorJaeS. Lim (standing) and
graduate studentMatthew M. Bacedemonstrate a newmethod thatwas recently developed to reducechannel degradation in theNa-tional Television Standards Com-mittee television system. The
receiver-compatible method to re-duce channel degradation is illus-trated usinga conventionaltransmitter andreceiver (figure 1),anew transmitter andconven-tional receiver (figure 2), andanewtransmitter andnewreceiver(figure3). This methodhaspoten-tial application to high-definitiontelevision systems.
Figure 1
noise cancellation techniques in multi-
ple microphone environments.At RLE, an intense effort is devoted
to multidimensional signal processingwhich focuses on signal estimation andreconstruction from incomplete or
noisy data, powerspectrum estimation,andspeech enhancement. Image ortwo-dimensional signal processing isalso examined, including image recon-struction from a single threshold cross-
ing.Anew image restoration systemhasbeen developed to improve de-
graded images, and a newset of condi-tions hasbeen introduced underwhich
Figure 2
an image canbe restored from its Four-ier transform phase, or magnitudealone.
l'rofessorJae S. Lim's research in-terests include multidimensional signalprocessing, signal reconstruction andenhancement, and speech and imageprocessing. lie has developed a tech-
nique for very low bit rate video con-
ferencing by transforming an imageinto a hi-level image, and then codingthe difference between the consecutive
image frames. Professor I.im has also
investigated topics associated with im-
age coding, where an image is repre-
Figure 3
sented in as few hits as possible while
preserving its quality and minimizingtransmission costs. Image coding has
practical applications in video tele-
phones, facsimile machines, digitaltelevision, and medical image storage.
Currently, Professor Lim is explor-ing various image processing issues.Ile is investigating new techniques formotion estimation whichcan be usedfor video interpolation, enhancement,andcoding. lIe is also developing amotion compensation algorithm thatwill stabilize video images from under-seacameras used in exploration, re-
connaissance, and salvage operations.Since undersea cameras are hamperedby the effects of wave motion andocean currents, they cannot monitor
images at a constant depth with acon-stant angle. Although it maybe impossi-ble to mechanically stabilize these cam-
eras, recent image processing researchat RLE maymake it possible to processthe video images, and thus, make thecamera appear stable. Professor Lim isalso studying signal processing issuesrelated to the improvement of currenttelevision systems and the design of
N®raF-1
.T-W
Professor Bruce R. Musicus usesplaying cards to demonstrate a newconcept infault-tolerant signalprocessor architectures. A smallamount of redundant com-putation is introduced into the system, andthe workload is spread across manyprocessors. Asimple test can detect which processors havefailed, andcanquicklycorrect thefailedprocessors' outputs without repeating their calculations.
In a worldwide cooperative e[ft.rt u'itI eral
academic andscientific institu-tions, Professor Arthur B. Baggeroer conductsfield experiments at Fram Straits,Svafbard, Norway (82°N, 7°F) on techniquesformodelling long-range sound
propagation in the Arctic Marginal Ice Zone. These experimentsseek to measureacoustic transmission andscattering, andseismic refraction in the Arctic Ocean.Campsites have beenset up at various times over the lustten years usingahorizon-tal receiving array, a multichannel data handling andstorage system, andadigi-tally operated data analysis system.
high-definition television systems. In arecent effort, he developed a receiver-
compatible system that reduces trans-mission channel degradation such as
multipath echoes, noise, and interfer-ence. The new system requires a sim-
ple transmitter modification, and even
though the transmission channel deg-radation is not reduced with an existingreceiver, it is reduced significantly witha new receiver.
In addition to image processingresearch, Professor Lim has developeda new model-based speech analysis sys-tem for high-quality speech produc-tion. This research explores methodsto apply anewspeech model, known asthe multi-hand excitation speech mod-el, in areas such as speech coding, en-hancement, and time scale modifica-tion. Arobust, good-quality speechcoding system has been developedfrom this model with potential applica-tions to mobile telephones.
Professor Bruce R. Musicus ex-
plores fast, efficient, highly parallelalgorithms and special-purpose com-
puter architectures for digital signalprocessing. Thegoal of his work is to
develop more robust andsensitive al-
gorithms that will extract informationfrom digitally sampled data, and to de-vise unconventional architectures thatare matched to specific algorithms with
improvements in processing speeds.He has developed an inexpensive,high-performance cellular architecture
design for these applications. ProfessorMusicus has also studied architecturaltransformations in signal processingsystems, andhas developed methods to
manipulate signal flow graphs into sev-eral task-invariant architectural forms.Hisother research interests includestochastic estimation and signal recon-struction and enhancement.
Professor Arthur B. Baggeroer pur-sues studies in underwater acoustics,multidimensional signal estimation,and array processing. A major focus ofhis work involves geophysical andacoustic field experiments in the Arctic
marginal ice zone to collect sonar and
oceanographic data for signal pro-cessing. The results of this work charac-terize the acoustic reflections from theocean bottom. Professor Baggeroer hasalso conducted research at the WoodsHole Oceanographic Institution on anunderwater communication systemused for the telemetry of data from un-tethered oceanographic sensors on thesea floor.
Dorothy A. Fleischer
FACULTY PROFILE
FACULTYPROFILE:Alan V. Oppenheim
After receiving his doctoratefromMITin 7,964, Alan V. Oppenheimjoinedthe Institutefaculty, wherehe iscurrently Professor ofElectrical Engi-neeringand computer Science. His re-search interests are in the general areaofsignalprocessingand its applica-tions to speech, image, andgeophysicalsignaiprocessing. He is co-author ofthewidely used textbooks Discrete-Time
Signal Processing, Digital Signal Pro-
cessing, and Signals and Systems. He isalso the editor ofseveral advancedbooks on signalprocessing.
Professor Oppenheirn is a member
ofthe NationalAcademyofEngineer-ingandaFellow ofthe IEEE. He hasbeen a Guggenheimfellow andaSacklerfellow, andhas received manyawarchforoutstanding research and
leaching, including the 1988 IEEEEducation Medal, the IEEE CentennialAward, andthe Society and TechnicalAchievement awards ofthe IEEE Societyon Acoustics, Speech andSignal Pro-
cessing.
" Who influencedyoumost in the
early stages ofyour career?
As a graduate student, I worked withManuel Cerrillo in RLE during the earlyphases of my doctoral research. Cer-nIb was a captivating and inspiringperson who had many beautiful andsometimes strange ideas. Amongother
things, he interested me in modern al-
gebra, and that eventually led to theideas in my doctoral thesis. I alsoworked closely with Amar &se as a
graduate teaching assistant, and he wasan inspiration to me in terms of stan-dards, ideals, creativity, and teachingmethodology. Another influential per-son, during my time as a graduate stu-dent and afterwards, was Tom Stock-ham. He was tremendously excited and
encouraging about the nonlinear the-or)' that I was developing, and later
played a major role in its application toaudio and image processing.
Perhaps the person who had the
greatest impact on me after I joined the
ProfessorAlan V. Oppenheim
faculty in 1964 was Ben Gold fromLin-colnLaboratory. I first met Ben whenhe was a visitor in Ken Stevens' speechgroup. Ben has a friendly, informal, and
supportive style, in addition to deep in-
sights and broad experience. Becauseofthese qualities, I and many others
sought him out for discussion and con-sultation. Ben and I worked togetherformany years, and are still very close
personal friends.
" What was thenature ofyour workat LincolnLabfrom 1967-69?
I initially became involved at Lincoln
Laboratory as a consultant, primarilythrough my interactions with Ben GoldandTom Stockham. In 1967, I decidedto take a two-year leave of absencefrom the faculty and spend full timethere as a staff member. During that pe-riod, I wasfortunate to work closelywith Benand Tom, as well as with Char-lie Rader on a variety ofspeech and im-
age processing problems, and on sev-eral theoretical issues related to the
emerging field of digital signal process-ing. The landmark paperon the fastFourier transform had just been pub-lished by Jim Cooley andJohn Tukey,and there was a rapidly growing inter-est in exploiting the newly discovered
flexibility, efficiency, and accuracy of
processing signals on acomputer orwith digital hardware. At Lincoln, therewere many signal processing appli-cations that took advantage of this new
technology, enthusiastic and creative
colleagues, and a long list of interestingproblems. It was a unique and specialopportunity to work there during thattime. My involvement with Lincoln has
since continued at various levels, in-
cluding two extremely enjoyable andeducational years as half-time AssociateHead oftheData Systems Division un-derAl McLaughlin.
'How didRLE's research in signalprocessing evolve?
Signal processing has a long, rich histo-
ry dating back to the 17th century, and
perhaps before. Wherever signals arise,
signal processing plays a role in someform. At RLE in the '50s and '60s, therewas considerable activity in analog sig-nal processing in the context ofcommunication theory (NorhertWeiner, Claude Shannon, YukWingLee), speech processing (Ken Stevens),video (Bill Schreiber), biomedical pro-cessing (Murray Eden), andaudio
(AmarBose and Manuel Cerrillo). Inthe early '60s, Amar Bose, Tom Stock-ham, andJim Bruce used computerprocessing to simulate room acoustics,and Ken Stevens was using the TX-0
computer for speech analysis. How-ever, at that time, a typical viewpoint atMIT and elsewhere was that processingsignals on a computer was inherentlynon-real-time, and was principally in-tended to approximate or simulate an
analog system. It wasn't until after thedisclosure of the fast Fourier trans-form, and the associated developmentsthat it spawned, that an appreciationdeveloped for the advantages of imple-menting real-time signal processingsystems digitally. Many of these devel-
opments happened at Lincoln Labora-
tory and Bell Labs. As aconsequence of
having been part of the Lincoln activity,when I returned to MIT in 1969, I start-ed a research group in RLE and offeredwhat I believe was the first digital signalprocessing course in the country.
*Peoplehave saidyour research
group is the strongest of its kind inthe world What areyourfeelings onthat?
To the extent that it's true, I'm verypleased. Over the years, there havebeen several factors that have contrib-uted to ourgroup's good fortune. Our
relationship with Lincoln has been im-
portant, not only in terms ofthe peoplewe interact with, but also in terms ofthe technical problems that they dealwith. I feel strongly about this becausein order to do good research, you need
coupling to the real world, and this was
F A C U L T_Y P R 0 F I I. E
one of the most important aspects of
my Lincoln experience.Another factor has been our rela-
tionship with industry.When the chem-
istry is right, the relationship canbe
tremendously productive and exhila-
rating. It's a great opportunity for tech-
nology transfer.We are introduced to
important real problems, which moti-vates us to think in certain directions;the companies see the results of ourwork, and that motivates them to applyour research. The two companies thatwe've had perhaps the most successful
relationships with are Sanders Associ-ates and Schlumberger. Recently, we'vealso been working with a small com-
pany called Atlantic Aerospace. These
companies all have real problems that
they're trying to solve, and they'relooking for innovative solutions. AtRLE, we have exciting solutions that are
looking for problems. If we can matchthese two situations in the right way,everybody wins.
Also, the MIT students are extraor-
dinary. Fortunately, ourgroup has hadthe resources to develop an environ-ment that attracts some ofthe best stu-dents.We've also been extremely fortu-nate in terms of the support and
encouragement that we've receivedfrom our research sponsors.
"Thefacultymembersinyourgrouphave diverse talents andresearch in-terests What's the "glue" that bringsyou together?
Our common culture is signal process-ing and a deep interest in the develop-ment of innovative algorithms. Also,our individual interests in applicationsstrongly overlap. For example, I'm con-cerned with speech processing, imageprocessing, geophysical signal process-ing, and underwater acoustics. Art Bag-geroer, who is partially affiliated withourgroup, has a strong background instochastic signal processing and is inti-
mately involved with communication
systems, geophysical signal processing,and underwater acoustics. JaeI.im has
focused heavily on speech, image, andvideo processing. BruceMusicus is theindividual in our group who is most
significantly involved with hardware.He is concerned with signal processingarchitecture issues and the theoretical
aspects ofsignal processing algorithms.All of us come from a common intellec-tual background, and bothJae and
Bruce did their doctoral research inthis group.
*Is there a currentarea of researchthatyou're excitedabout?
Therearemany areas that we're excitedabout, and as usual, more ideas to pur-suethan the time to pursue them.Someof these areas are applications-orient-ed, and others are highly speculativeideas which, if they lead somewhere,are likely to be in the category of "solu-tions in search of problems." An exam-
ple on the applications side is our workon removing distortions in side-scansonar images. Ideally, when the sonar
equipment is towed through the water,it should move in a straight line. How-ever, the ocean's environment and
ship's movement introduce distortionssuch as the pitch, roll, and yawof thesonar.We areexploring the processingof side-scan sonar imagery to estimatethe pitch, roll, andyaw from the data,andthen use the estimates to removethe distortions.
Another applications-orientedarea is noise or signal cancellation,
wherebywe estimate and then subtractout the unwanted components from a
composite signal. These techniques of-ten involve making measurements inreal-time, and based on those measure-ments, adaptively subtracting outwhat's not wanted; essentially relyingon destructive interference. Tech-
niques of this type have been used for
multiple microphone speech enhance-ment, in noisy industrial environments,and in other controlled environments.But, it's in the less controlled environ-mentswherethe problems become sig-nificantly more exciting and difficult.
We're also excited about the broadarea of multidimensional signal pro-cessing, on whichJae and I haveworked for many years, and on whichhe has recently published a new text-book. One problem area involves thereconstruction of multidimensional
signals from only partial informationand some known constraints. Forex-
ample, we have shown that, under cer-tain general constraints, an image canbe recreated from its crossings of onlya single threshold. The procedure be-comes increasingly robust as morethresholds are used, andwhile this the-or)' can be thought ofas one of thosesolutions looking for the right prob-lem, potentially it maybe the basis for
some innovative image coding algo-rithms. Jae is also heavily involved intwo-dimensional image processing andenhancement, and in video or three-di-mensional processing, including mo-tion estimation and high-definitiontelevision.
Somewhat more speculative, but
very exciting, is our work on symbolicsignal processing. The notion of manip-ulating signals andsignal processingsystems symbolically rather than nu-
merically requires a very different kindof signal processing. Several years ago,we became seriously involved with
combining numeric and symbolic sig-nal processing. Our interest in this was
originally motivated by Bob Kahn whenhe wasat DARPA, whosuggested thatwe incorporate some of the principlesofartificial intelligence into our re-search.We foundthe symbolic process-ing technology used in artificial intelli-
gence and other disciplines to be of
particular interest. We pursued this inseveral directions. Onehas been in
combining classical numerical signalprocessing with rule-based systems touse more qualitative or subjective in-formation and constraints. We'veworked closely with Sanders Associateson combining numerical processingwith rule-based systems, and throughSanders, several ofour ideas have been
successfully incorporated into practicalsignal processing systems.
We've also found the symbolicprocessing technology used in artificial
intelligence and other disciplines to beofparticular interest and importance in
developing a signal processing algo-rithm design environment. In a some-what similar manner to the way inwhich the MacSymasystem manipu-lates mathematical expressions sym-bolically, our system manipulates sig-nals and signal processing expressionssymbolically to achieve semi-automat-ed or user-interactive algorithm de-
sign. Typically, in signal processingsystem design, an algorithm is first
specified at a relatively high level.Then, the system designer manipulatesand rearranges the algorithm using awell-formulated set of rules, proper-ties, or identities that are blended withformal or informal cost measures, ex-
perience, or intuition. Parts of this pro-cess can he automated, and often a
computer can sort through many more
options than a designer. Also, a designenvironmentcan potentially incor-
FACULTY PROFILE
porate more expertise into this processthan a relatively inexperienced systemdesigner. Bruce Musicus' work in map-ping signal processing algorithms ontoVLSI architectures also couples nicelyinto this work. In the somewhat distantfuture, onecould imagine a sophisticat-ed signal processing design environ-ment that can symbolically explorealgorithm rearrangements and imple-mentations based on cost measures as-sociated with VLSI implementation,and can suggest details of the imple-mentation. Probably not this year,though.
In a very different direction, we've
recently become interested in chaos
theory and fractals, and howthey relateto image modelling, spread-spectrumcommunications, and signal coding.We're exploring a variety of excitingideas, but there's nothing specific that Ican report on right now. Mentioningchaos and fractals brings to mind an is-sue that I tend to be sensitive to andsomewhat nervous about in my re-search. There's always -a certain set of
topics or areas that receive a lot of at-tention in the technical or popularpress, and tend to develop into fads. In-
evitably, expectations quickly outrun
reality. A lead article somewhere willdescribe how a particular science or
technique is going to solve some majorproblem, then, what they'd really liketo show in next week's issue is how itdid solve the problem. If it didn't, often
disappointment and disinterest follow.For this reason, I prefer to maintain alowprofile on highly speculative ideasor directions.
"What is the relationship between
digital signalprocessing researchandcomputer hardware technology?
It's really a symbiotic relationship. Infact, in my opinion, much of the vitalityofthe digital signal processing fieldstems from the close coupling between
applications, algorithm development,and the technologies available for
signal processing. Ten years ago,nanosecond signal processing micro-
processors with floating-point arithme-tic (which are commercially availablenow) would have seemed totally far-fetched. Similarly, a decade ago, the fastFourier transform algorithm in real-time, at any reasonable data rates, wasconfined to supercomputers or racks of
special-purpose digital hardware.
There are now special-purpose chipsets, and even a single-chip fast Fouriertransform processor is available. The
pace of technology development is ex-
traordinary, and there's certainly no in-dication that it's slowing down. Ofcourse, that gives those of us who are
primarily involved in algorithm devel-
opment considerable confidence, thatifwe were to come up with a robust,
intellectually sound innovative algo-rithm, it will eventually be implementa-ble at the size, speed, or cost needed tomake it practical. Undoubtedly, this is anaive assumption, but so far, thingshave worked out that way. In a similar
way, those whodevelop the technologymake the assumption that algorithmsand applications will require and makeuse of the new technologies.
It is worth mentioning that manyof the developments referred to underthe heading of digital signal processingare not really tied to computer or even
digital hardware. In the revision of mybook, Digital Signal Processing, co-
authored with Ron Schafer, we changedthe title to Discrete-Time Signal Pro-
cessing in recognition of the fact that itis the discrete-time nature of the tech-
niques and technology that we are fo-cusing on. Digital techniques will con-tinue to be extremely important, butthere are many other discrete-time
technologies such as surface-acoustic-wave and charge transport devices.And, of course, there will always benew continuous-time signal processingtechnologies.
.Inyour opinion, who has made themost substantial contributions to
digital signaiprocessing?
One of the important early people in-volved in digital signal processing and
digital filtering wasJim Kaiser.Jim hadworked in the MIT Electronics SystemsLab on sampled data control systems asa graduate student. After MIT, he wentto Bell Labs and became involved in
digital filter design for vocoder simula-tion. Jim is clearly one of the pioneersin digital filtering, and he recognizedvery early the potential of filtering sig-nals with computers and special-pur-pose digital hardware.
At Lincoln Laboratory, Ben GoldandCharlie Rader were also using digi-tal filters for vocoder simulation andmade many substantial contributions
through their work on speech process-
ing and processor architectures; as didTom Stockham with his work on digitalaudio and image processing. Larry Ra-biner and Ron Schafer (both former
graduate students in RLE) contributed
significantly with their work on speechprocessing at Bell Labs. Bill Lang of IBM
played an important role in promotingthe IEEE group on Audio and Electro-acoustics as the "home" for digital sig-nal processing. Eventually, the group'sname was changed to Acoustics,
Speech, and Signal Processing.Clearly, the contributions ofJim
Coolcy andJohn Tukey in identifyingthe fast Fourier transform algorithmwere seminal. It was probably the most
significant event in launching digitalsignal processing in its current form.
" What do youseeforthefuture ofsignalprocessing?
Fortunately, there will always be sig-nals that need to be processed; ad-vances in technology will continue tooffer new opportunities for sophisticat-ed systems; and this, in turn, energizescreativity in algorithm developmentand refinement for practical systems.Frankly, I don't ever see an end to this
cycle. There are many areas of theoreti-cal research that will likely see signifi-cantly more activity in the future than
they have so far. One that stands out in
my mind is multidimensional signalprocessing as it arises, for example, in
imageand video processing, and in a
variety of array processing problems.It's a difficult area theoretically, in partbecause multidimensional polynomialsdon't have the nice properties of one-dimensional polynomials. Also multidi-mensional data sets of any reasonablesize in each dimension result in verylarge amounts ofdata to be processed.Consequently, current practical algo-rithms tend to be relatively simple andconservative. As the technology offers
expanded opportunities for computa-tional speed, data storage, and access-
ing, I believe that some dramatic new
algorithms will emerge.Another area in which there is con-
siderable need for innovation and newideas is nonlinear theory and nonlinear
signal processing. This has always beena difficult area because very simplenonlinear systems can generate appar-ently complex behavior. Ofcourse, this
might be part of the promise ifthat
complex behavior can be understood.
FACULTY PROFI LE
Complex behavior ofsimple nonlinear
systems is intimately connected to cha-os theory, where there is nowconsid-erable interest and activity. Perhaps de-
velopments in this area will eventuallylead to novel nonlinear signal process-ing systems, and to a better understand-
ing of the behavior in current systems.With regard to applications, so-
phisticated signal processing systemsin the military context will continue t
play an important role, and in fact,there is considerable evidence that the
importance of signal processing will
expand significantly in areas such as"smart weapons," surveillance, intelli-
gence, communications, andbattlefield
management. In theconsumer arena,
signal processing will certainly contin-ue to play a significant role. Signal pro-cessing is so much a part of our every-day lives that we almost don't notice itin ourhome entertainment and com-munication systems. In the not-too-distant future, a television receiver will
undoubtedly he a special-purpose digi-tal computer because of today's majordevelopments and interest in signalprocessing technology for digital and
high-definition television. Other typesof very sophisticated consumer-orient-ed digital signal processing are on thehorizon. The telephone system is clear-
ly going digital, and voice mail is in-
creasingly common.The difficulties ofvoice mail present interesting opportu-nities for digital signal processing re-search and its applications. Currently,for example, no answering machine or
voice mail system can scan messagesfor key words. The notion of addingword spotting andvery limited speechrecognition to voice mail systems re-
quires sophisticated signal processingthat incorporates processing algo-rithms and the application-specificwork done by people like Victor Zue.Other applications, such as adaptiveaudio systems in which the equalizerfrequency response adjusts auto-
matically, are do-able, but the current
technology doesn't yet make it cost-effective or fast enough for it to be
practical.
" What is the nature ofyour researchat the WoodsHole OceanographicInstitution?
As you know, I spend summerswork-
ing there as part of thejoint MIT/WoodsHole Program. One project that I have
Discussing the resultsfrom their signalprocessing research are (from lefi) graduatestudent Daniel T. Cobra, ProfessorAlan V. Oppenheim, graduate students TheH. loo,MicheleM. Govell, andGregory W. Wornell.
worked on, in collaboration with
George Frisk, a physicist at WoodsHole, involves the study of ocean bot-tom acoustics. George ran a series of
experiments to collect acousticmeasurements, and I've worked withhim to interpret the data and developtheories on howto extract model pa-rameters of the ocean bottom from it.Oneof the interesting things that cameoutof this work was the so-called fastIlankel transform algorithm. More re-
cently, I've worked with JulesJaffee on
problems related to correcting distor-tions in side-scan sonar data.
Both the intellectual and recrea-tional aspects ofWoodsHole make it an
extremely enjoyable place to spendtime.Thetown ofWoodshole, which is
really part of Falmouth, has a fairly aca-demic flavor because theOceano-
graphic Institution, the Marine Biologi-cal Laboratory, and the U.S. GeologicalSurvey are located there. A magazinearticle once commented that, in thesummertime, there's a beach at WoodsHole with the highest density of Nobellaureates of any beach in the world.The town itself hasa unique style that's
perhaps best summarized on a tee-shirtthat says on oneside, "Isn't Falmouthnice," andon the other side, "Isn'tWoods Hole weird." I find the institu-tion exciting because many large-scaleexperiments there involve data collec-tion in a natural environment. I alsohave a special love ofthe ocean, andmy
extracurricular activities include wind-
surfing and sailing. It's nice to integratethose activities into an environmentwhereI can also carry out my research.
"Why didyou choose a teachingcareer?
That question triggers many memoriesfor me. I became a graduate teachingassistant because I liked tutoring as an
undergraduate, and I needed the finan-cial support. After my first fewweeks asa teaching assistant, I felt that a newworld had opened up for me. Teachingforces you to understand things at a
very fundamental level, in part, be-cause in preparing for class, there's justa natural motivation to want to be fullyprepared. Even so, no matter howcare-
fully you think you understand some-
thing, it's likely a student will ask a
question that you hadn't anticipated. A
strong side benefit to this is under-
standing something better and at a
deeper level than you thought possible.Another attraction of teaching, for me,is the thrill of a "successful" perfor-mance. Of course, not every class goesthat way.And the ones you thought did,often didn't-and vice versa.
As I said earlier, Arnar Bose had atremendous impact on my view of
teaching. I recall Amar commenting onthe enormous leverage and responsi-bility that teaching provides in terms of
propagating your standards and ideals.
If those standards and ideals are trans-mitted to just a small number of stu-dentswho go into teaching careers,
they will likely do the same for theirstudents, and this expands through asuccession of generations. There are
many events that continually remindme of this far-reaching and often un-seen impact, and one in particular hap-penedat a recent technical conference.I had just gotten into an elevator withone other person,andwhen the doorsclosed, he looked at me and said, "Idon't know if you remember me, but Itook your videotape course several
years ago." Seriously.
" What's the most challengingpartofyour career-scientist, teacher, orauthor?
I enjoy the interrelationship of thethree, so I don't think they should be
separated. Of course, the role of ateacher isn't just limited to a classroom,
you are also a thesis advisor and men-tor. I strongly believe that in order tohe an effective teacher, either in or out-side the classroom, you must be in-volved in research because it keepsideas fresh, exciting, and on the cuttingedge. By the same token, teachingbenefits research by providing amechanism to convey and refine ideas.
Every semester that I teach the digitalsignal processing course, new insightsemerge about the material that end upaffecting my research. As a researcher,it's important to present your ideas in
writing, in a classroom, in conferences,or in whatever wayforces you to com-municate your ideas and present themfor scrutiny. Many people hate to writebecause the idea of putting one's ideasdown on paper for peer review can be
terrifying. Although I don't enjoy writ-
ing, it's always nice to have written.Once it's done, it's a thrill to see it. But,for most of us, it's an agonizing pro-cess.
" What's the most rewarding aspectfyour work?
The individual interaction with my stu-dents. My hope is that by the time theygraduate, each of my students will havean international reputation, and that weare personal friends and professionalcolleagues. I can't achieve this with ev-
ery student, but watching the relation-
ship evolve from one of student-teach-er to one of close friend and peer is
extraordinarily gratifying. That is myreward.
ett %a'tII U DAt ttIflt 0
Dr. A. NihatBerker, Professor of Phys-ics, was elected to Fellow of the Ameri-can Physical Society in February 1989.The Society confers the title ofFellowto those members whohave contribut-ed to the advancement of physics byindependent, original research, or whohave rendered some other specialservice to the cause of the sciences.Professor Berker was cited for the de-
velopment of the position space renor-
malization-group technique and its ap-plication to studies of phase transitionsin physisorbed systems and liquidcrystals.
Recently appointed White House Chiefof Staff, Dr.JohnH Sununu (S.D. '61,S.M. '63, Ph.D. '66) continues the RLEtradition of serving as the President's
"right-hand man." Coincidentally, Dr.Sununu was agraduate thesis student atRLEwhen then-ELE Director Jerome B.Wiesnerwassummoned to Washingtonto serve as PresidentJohn F. Kennedy'sscience advisor. Dr. Sununu investigat-ed measurements of cesium vapor ther-mal conductivity in RLE's Plasma Mag-netohydrodynamics Research Group in
1961, when the above photo was taken.
Dr. Francis F. Lee (S.D. '50, SM. '51,Ph.D. '66), Professor Emeritus of Elec-trical Engineering and Computer Sci-ence, has retired from the Institute.Since beginning his association withRLEas a research assistant in 1952, Pro-fessor Lee focused his research on digi-tal systems applications andsensoryaids for the handicapped. In the mid-
'60s, he worked on character recogni-tion and speech production. Followinghis studies on the relationship betweenwritten language and phonetic tran-
scription, Professor Lee developed a
computer-driven system to synthesizespeech by rule, which contributed to a
reading machine for the blind.
With recent experiments that suggestthe possibility of "cold fusion," Dr.PeterL Rage/stein (B,S. '76, MS. '76,Ph.D. '81), Associate Professor of Elec-trical Engineering and Computer Sci-ence, has advanced a speculative the-
orywhich seeks to explain the natureof these processes. Professor Flagel-stein presented an overview of these
developments to interested colleaguesat MIT on April 14, 1989.
Dr Henry I. Smith, Professor of Elec-trical Engineering and Computer Sci-ence, and Senior Research ScientistDr.Robert H. Redike'r were among the
ninety outstanding engineers electedas members of the National Academy of
Engineering in February 1989. Electionto theAcademy is among the highestprofessional honors awarded to engi-neers. Academy membership distin-
guishes those individuals who havemade important contributions to engi-
Dr. Xiao Dong Pang was appointedResearch Scientist in RLEs SensoryCommUnication Group in March 1989.
Previously, Dr. Pang wasa Fairchild Fe!-low at MIT. In his new position, Dr.
Pang will conduct research on aids forthe hearing impaired, on the abilities ofthe human hand to sense and operateupon the environment, and on multi-model human interfaces for teleopera-tor and virtual environment systems.Dr. Pang received his B.S. ('81) in Elec-trical Engineering from Zhejiang Uni-
versity, an M.S. ('83) in Biomedical En-
gineering from Vanderbilt University,and Ph.D. ('88) in Electrical Engi-neering and Computer Science fromMIT.
neering theory and practice, and havedemonstrated unusual accomplish-ment in newand developing fields of
technology. Professor Smithwas citedfor his contributions to submicronstructure technology and research, andhis leadership in teaching and promot-ing submicron structures. Dr. Redikerwas honored for his pioneering contri-buiions and leadership in constructingsemiconductor compound light emit-ters and lasers.
Dr. Mandayam A. Srinivasan was ap-pointed Research Scientist in RLE's Sen-
sory Communication Group in Febru-
ary 1989. Dr. Srinivasan will conductbasic research on hand hiomcchanicsand the sense of touch, as well as re-search directed towards the use ofmanual sensing and control in tele-
operator systems. Dr. Srinivasan cameto MITin 1988 as a Fairchild postdoc-toral fellow. He holds a B.E. in Civil En-
gineering ('75) from Bangalore Univer-
sity, an M.E. in Aeronautical Engi-neering ('77) from the Indian Instituteof Science, and an M.S. in Applied Me-chanics ('77) and Ph.D. in Mechanical
Engineering ('8/i), both from Yale Uni-
versity.
In Memoriam
Dr. Dennis H. K1at 50, died De-cember 30, 1988, at the MITMedi-cal Department, following a longbattle with cancer. Dr. Klatt wasaSenior Research Scientist in RLE's
Speech Communication Group.After obtaining his undergraduatedegree in 1961 from Purdue Uni-
versity, Dr. Klatt received his doc-torate in communication sciencesfrom the University of Michigan in1964. He joined MIT in 1965 as anAssistant Professor, and became aSenior Research Scientist in 1978.
Dr. Klatt made significant con-tributions to the areas of speechproduction and synthesis. In col-laboration with Professor KennethN. Stevens, he conducted researchin vocal tract modelling and furth-ered the understanding of how in-dividual speech sounds are pro-duced. Dr. Klatt also developeda rule system for constructingspeech waveforms from abstract
linguistic specifications. He soughtto incorporate naturalness into his
speech synthesis systems, and hiswork was applied commercially bythe Digital Equipment Corporationin DECtalk. Dr. Klatt also pursuedresearch in speech perception andthe specification of acoustic prop-erties for various phonetic distinc-tions. As the author of more than60 scientific papers and a co-au-thor ofFrom Text to Speech: TheMITa1k System, he was working ona book at the time of his death.
Dr. Klatt received the SilverMedal in Speech Communicationfrom the Acoustical Society ofAmerica, and theJohn Price Weth-erhill Medal from the Franklin In-stitute. Photo byJohn F. Cook
History of Digital SignalProcessing at RLE
1953
Professor ErnstA. Guillemin (left) andDr. Manuel V (Jerrillo discuss
filler theory. Professor Guillemin was well-knownfor his commit-ment to teaching andresearch in circuit theory, andDr. Cerrillo wasdevoted to signalprocessing research in man andmachine usingboth auditory andvisual signals as media. Together, Professor Guil-leminandDr. Cerrillo collaborated onproblems in time-domainnetwork synthesis.
"The statistical methods-of linearfiltering andprediction evolved
largely byProfessor Norbert Wiener were used to some extent duringWorld WarII, but thefield ofstatistical communication theory actu-
ally blossomedfurther after the War. Having worked with ProfessorWienerformany years, Professor Y WLeebrought to the Laboratorythe mathematical basisfor the theory. Within a very short time, the
potential significance ofcorrelation techniques hadfired the imagi-nation ofall communication engineers. To a very important extentthegeneral enthusiasm wasdue to experimental evidence (providedby Lee, Wiesner, andCheatham) that weak signals couldhe recov-
ered in thepresence ofnoise by using correlation techniques. Fromthatpoint thefield evolved very rapidly. "-ProfessorHenryJ. Zim-
mermann, ELFDirector, in Twelve Years of Basic Research: ABriefI listory of RLE, 1958
Ca 1948From Aft: ProfessorsJerome B. Wiesnerand Yuk WingLee with his
first doctoral student Thomas P. Cheatham, Jr.) andProfessor Nor-
bert Wiener. Cheatham built thefirst electronic (analog) correlator,shown in thephotograph. Thefinal version ofthe machine represent-ed the waveform as a sequence ofsample values, andmultipliedparts ofsamples by generating apulse with aheight proportional toonesample anda width proportional to the other. Cheatham 's ana-
log correlatorpavedthewayforHenryH Singleton's digital correla-tor in 1949, which rapidlyperformedstorage, multiplication and
integration by a binary digitalprocess.
Professors Yuk Wing Lee (left) andAmar G. Bose con-vincedNorbert Wiener topresentaseries oflectures onnonlinearproblems in random theory. Profissor Lee
tape-recorded the lectures, andphotographed hundredc
ofcomplex equations diagramed on the blackboard byWiener. In collaboration with ProjSorBose, Lee com-
piled this material into thefirst book on the subject in
English, Nonlinear Problems in Random Theory,pub-lished in 1958.
"With thegrowth ofcomputing in RLEin the early 1960s, it wasnatural that many areasofresearch wouldexploit this new
capability, and in many cases newresearch groups evolved
through coupling traditional disciplines with anewcomputa-tional emphasis. One example is the extension ofthe earlierstudy ofcontinuous systems, both linear andnonlinear, to a
study ofdiscrete systems, thusforminga newarea now calleldigitalsignalprocessing. Professor Alan Oppenheim built up i
very stronggroupwithin RIB, initially addressing problems iiispeech such as spectral estimation andcoding techniques, b/ifin more recentyears becoming involved in two-dimensional
image-processing techniques."-KarlL Wi/des aridNib A
Lindgren, A Century of EIcctricd EnncLring :ulLI (1/T(-1-Science at MIT, 1882-192
1960
RLEc Oeech Communication Group, under the direction ofProlessorKenneth N. Stevens (far right), developedsignalprecessingprograins on the 7X-O computerforspeech applica-tions. Gathered around the console ofthe TX-O (from left) areHiroyaB. Fu/isak:, Paul T. "Pete"Brady, Jr. (seated), GordonM.Bell, andKenneth N. Stevens.
1963
Variation on a theme(from left): graduatestudent Thomas G.Kincaid, violinist George Humphrey ofthe Boston SymphonyOrchestra, andgraduate studentsJohn F McDonald andAlanV Oppenheim. This ensemble worked with Dr. Manuel V. Cer-rillo on the application ofgrouptheory to generate syntheticStrauss waltzes.
L)UU (JUC 000 Don
-
ii04:-mU
1975
During the '60s and '70s, researchersfrom RIBused theTX-2 computer at Lincoln Laboratoryfora variety qfappli-cation.s including simulation, imaging, time-sharing re-search, andthe investigation of neural nets. Seated at the7X-2's console are Omar Wheeler ofLincoln Laboratoryfright.) andan unidentified scientist.
UPDATE:CommunicationsPublicationsRLE has recently published the follow-
ing technical reports:
Finding Acoustic Regularities in
Speech: Applications to Phonetic
Recognition, byJames Robert Glass.RLE TR No. 536. December 1988.152 pp. $10.00.
Formalized Knowledge Used in Spec-trogram Reading:Acoustic andPer-
ceptual Evidencefrom Stops, by LoriFaith Lamel. RLE TR No. 537. December1988. RLE TR No. 537.180 pp. $12.00.
Levinson andFast CholeskiAlgo-rithmsforToeplitz andAlmost Toe-
plitz Matrices, by Bruce R. Musicus.RLETR No. 538. December 1988. 95 pp.$8.00.
The OKJAdvancedArrayProcessor
(AAP)-DevelopmentSoftwareMan-ual, by Bruce R. Musicus. RLE TR No.
539. December 1988. 242 pp. $17.00.
A Theoretical andExperimentalStudyofNoise andDistortion in the
Reception ofFM Signals, by Amar G.Bose and William R. Short. RLETR No.540. January 1989. 56 pp. $7.50.
Adaptive Array ProcessingforMulti-
pleMicrophone HearingAids, by Pat-rick M. Peterson. RLETR No. 541. Feb-
ruary 1989. 125 pp. $9.00.
The annual RLEProgress ReportNo.
131 will he available in mid-June.RLE
Progress ReportNo. 131, covering the
periodJanuary through December1988, contains both a statement of re-search objectives and a summary of re-search efforts for each group in RLE.
Faculty, staff, and students who have
participated in these projects andfund-
ing sources are identified within each
chapter. ThePIEProgress Report alsocontains a list of current RLE personnel.
TheResearch Laboratory of Electronicswelcomes inquiries regarding ourre-search and publications. Please contact:Barbara PasseroCommunications OfficerResearch Laboratory of ElectronicsRoom 36-412Massachusetts Institute of TechnologyCambridge, MA 02139
telephone: (617) 253-2566telefax: (617) 258-7864
The Future of FM Broadcasting
In a recent RLE Technical Report(TR No. 540), Dr. Amar G. Bose, Profes-sor of Electrical Engineering and Com-
puter Science, and Dr. William R. Shortof Bose Corporation Research Staffhave published results of a research
program on the effects of multipath onFM broadcasting. Specifically, the re-
port focuses on the newly developedFMX technology which claims to dou-ble the stereo reception area of FM
broadcasting. TheFMXsystem seeks to
overcome the multipath effects of stan-
dard FM broadcasts by transmitting anextra channel of stereo informationwhich eliminates background noise. Intheir research findings, Drs. Bose andShort have demonstrated that the mul-
tipath effects on current FM stereo
equipment were greater with FMXthanwith current FM stereo transmissions.Thereport contains mathematical mod-els, analysis, and computer simulations
Dr. Will/mn R, Short aszc/ Professor
Amar G.Bose
basedon field tests conducted at MIT'sradio station WMBR. The report con-cludes that FMX actually degradesreception and does not increase thestereo broadcast range. To orderacopyofTechnical Report No. 540, pleasecontact the RLE Communications
Group.
UPDATE:RLE CollegiumThe ELE Collegiumwas established in
1987 to promote innovative relation-
ships between the Laboratory and busi-ness organizations. Thegoal of RE's
Collegium is to increase communica-tion between RLE research and indus-trial professionals in electronics andrelated fields.
Collegium members have the op-portunity to develop close affiliationswith the Laboratory's research staff, andcan quickly access emerging resultsand scientific directions. Collegiumbenefits include access to a wide rangeof publications, educational video pro-grams, RLE patent disclosures, semi-nars, and laboratory visits.
The RLE Collegium membershipfee is $20,000 annually. Members ofMIT's Industrial Liaison Program canelect to transfer 25% of their ILP mem-
bership fee to theRLECollegium. Mem-
bership benefits are supported by the
Collegium fee. In addition, these fundswill encourage new research initiativesand build new laboratory facilitieswithin RLE.
For more information on the RLE
Collegium, please contact RLE or theIndustrial Liaison Program at MIT.
RLE currentsRLE currents is a biannual publication of'he Research Laboratory of Electronics atthe Massachusetts Institute ofTechnology.
Jonathan Allen, ..............
John F. Cook.............
Everett Design ..............
DorothyA. Fleischer
uarbaraJ. Passero..........
DonnaMaria Tiechi.
l-ienryJ. zimmermann
Editor-in-Chief
Photography.
DesignStaff Writerand Editor.
Productionand Circulation.
ManagingEditor.
Advice and
Perspective
The staffof currents would like to thankProfessor Alan v. Oppenheim for histechnical guidance during the preparationof this issue.
Inquiries maybeaddressed to ELFcurrents, 36-412, Research Laboratory ofElectronics, Massachusetts Institute of
Technology, 77 Massachusetts Avenue,
Cambridge, MA 02139.
(155N 1040-2012)
RLE© 1989. Massachusetts Institute ofTechnology.
All rights reserved worldwide.