Shuvra S. Bhattacharyya Ed F. Deprettere · Rainer Leupers Jarmo Takala Editors Handbook of Signal Processing Systems Third Edition Handbook of Signal Processing Systems Shuvra S. Bhattacharyya • Ed F. Deprettere Rainer Leupers (cid:129) Jarmo Takala Editors Handbook of Signal Processing Systems Third Edition Foreword by S.Y. Kung 123 Editors ShuvraS.Bhattacharyya EdF.Deprettere DepartmentofECEandUMIACS LeidenEmbeddedResearchCenter UniversityofMaryland LeidenUniversityLeidenInstituteAdvanced CollegePark,MD,USA ComputerScience Leiden,TheNetherlands LaboratoryforPervasiveComputing TampereUniversityofTechnology JarmoTakala Tampere,Finland DepartmentofPervasiveComputing TampereUniversityofTechnology RainerLeupers Tampere,Finland RWTHAachenUniversitySoftware forSystemsonSilicon Aachen,Germany ISBN978-3-319-91733-7 ISBN978-3-319-91734-4 (eBook) https://doi.org/10.1007/978-3-319-91734-4 LibraryofCongressControlNumber:2018953763 ©SpringerInternationalPublishingAG,partofSpringerNature2019 1stedition:©SpringerScience+BusinessMedia,LLC2010 2ndedition:©SpringerScience+BusinessMedia,LLC2013 Thisworkissubjecttocopyright.AllrightsarereservedbythePublisher,whetherthewholeorpartof thematerialisconcerned,specificallytherightsoftranslation,reprinting,reuseofillustrations,recitation, broadcasting,reproductiononmicrofilmsorinanyotherphysicalway,andtransmissionorinformation storageandretrieval,electronicadaptation,computersoftware,orbysimilarordissimilarmethodology nowknownorhereafterdeveloped. Theuseofgeneraldescriptivenames,registerednames,trademarks,servicemarks,etc.inthispublication doesnotimply,evenintheabsenceofaspecificstatement,thatsuchnamesareexemptfromtherelevant protectivelawsandregulationsandthereforefreeforgeneraluse. Thepublisher,theauthorsandtheeditorsaresafetoassumethattheadviceandinformationinthisbook arebelievedtobetrueandaccurateatthedateofpublication.Neitherthepublishernortheauthorsor theeditorsgiveawarranty,expressorimplied,withrespecttothematerialcontainedhereinorforany errorsoromissionsthatmayhavebeenmade.Thepublisherremainsneutralwithregardtojurisdictional claimsinpublishedmapsandinstitutionalaffiliations. ThisSpringerimprintispublishedbytheregisteredcompanySpringerNatureSwitzerlandAG Theregisteredcompanyaddressis:Gewerbestrasse11,6330Cham,Switzerland To Milu ShuvraBhattacharyya To Deirdre EdDeprettere To Bettina RainerLeupers To Auli JarmoTakala Foreword Itgivesmeimmensepleasuretoreintroducethishandbooktotheresearch/develop- ment communitiesin the field of signal processing systems (SPS). The handbook representsthefirstofitskindtoprovideacomprehensivecoverageonstate ofthe artsofthisfield.Thefactthatitisalreadynowthethirdeditionisaclearattestation of the high demand from all the related professional communities. It is truly an influentialandtimelycontributiontothefieldofSPS. Thedrivingforcebehindinformationtechnologies(IT)hingescriticallyuponthe major advances in both componentintegration and system integration. The major breakthroughfortheformerisundoubtedlytheinventionofICinthe1950sbyJack S. Kilby,the NobelPrize Laureatein Physicsin 2000.In an integratedcircuit, all components were made of the same semiconductor material. Beginning with the pocketcalculatorin1964,therehavebeenmanyincreasinglycomplexapplications followed.In fact, processinggates and memory storage on a chip have since then grown at an exponential rate, following Moore’s Law. (Moore himself admitted that Moore’s Law had turned out to be more accurate, longer lasting, and deeper in impact than he ever imagined.) With greater device integration, various signal processing systems have been realized for many killer IT applications. Further breakthroughsin computersciences and Internettechnologieshave also catalyzed large-scalesystemintegration.Allthesehaveledtotoday’sITrevolutionwhichhas profoundimpacts on our lifestyle and overall prospect of humanity. (It is hard to imaginelifetodaywithoutmobilesortheInternet!) ThesuccessofSPSrequiresawell-concertedintegratedapproachfrommultiple disciplines,suchasdevice,design,andapplication.Itisimportanttorecognizethat system integration means much more than simply squeezing components onto a chip and, morespecifically, there is a symbiotic relationshipbetweenapplications andtechnologies.Emergingapplications,e.g.,5Gcommunication,bigdataanalysis, machinelearning,and thetrendyAI,will promptmodernsystem requirementson performance and power consumption, thus inspiring new intellectual challenges. Therefore, the new paradigm of SPS architectures must be amenable to various designfacetssuchasoverallsystemperformance,flexibility,andscalability,pow- er/thermalmanagement,hardware-softwarepartition,andalgorithmdevelopments. vii viii Foreword With greaterintegration,system designs becomemore complexand there exists a huge gapbetween what can be theoretically designedand what can be practically implemented. It is critical to consider, for instance, how to deploy in concert an everincreasingnumberoftransistorswithacceptablepowerconsumptionandhow to make hardware effective for applications and yet friendly to the users (easy to program). Therefore, major advances in SPS must arise from close collaboration betweenapplication,hardware/architecture,algorithm,CAD,andsystemdesign. ItisonlyfittingforSpringer/Naturetoproducethistimelyhandbook.Springer/- Nature has long played a major role in academic publication on SPS, many of them have been in close cooperation with IEEE’s signal processing, circuits and systems,andcomputersocieties.Fornearly30years,Ihavebeentheeditor-in-chief ofSpringer’sJournalofSignalProcessingSystems,consideredbymanyasamajor forum for the SPS researchers. Nevertheless, the idea has been around for years thatasingle-volumereferencebookwouldveryeffectivelycomplementthejournal in serving this technical community. Then, during the 2008 IEEE Workshop on SignalProcessingSystems,WashingtonD.C.,JenniferEvansfromSpringerandthe editorialteamledbyProf.ShuvraBhattacharyyamettobrainstormimplementation of such idea. The result was this series of right-on-time handbooks. Especially, this edition has collected a vast pool of leaders/pioneers to cover architectures; compilers,programmingandsimulationtools;anddesigntoolsandmethodologies. Indeed, the handbook offers a comprehensive and up-to-date treatment of the drivingforcesbehindSPS,currentarchitectures,andnewdesigntrends.Itprovides a solid foundation for several imminent technical areas, for instance, scalable, reusable,andreliablesystemarchitectures,energy-efficienthigh-performancearchi- tectures, IPdeploymentand integration,system-on-chip,memoryhierarchies,and future cloud computing. Moreover, it covers a wide spectrum of applications, including wireless/radio signal processing, image/video/multimedia processing, control and communication, video coding, stereo vision, computer vision, data mining,andmachinelearning. Lookingintothe(near)future,wenotethatmodernAItoolshavebecomeheavily data-drivenanddata-intensive.Asofnow,onthedaily basis,as manyas1 billion photos and 10 billion messages are being handled by a single Internet company and, moreover, such dazzling numbers are rapidly growing on par with Moore’s law.Inordertounravelusefulinformationhiddeninbigdata,itwillrequirenovel (and possibly parallel processing) algorithmic designs which in turn will call for specialhardware/softwaretechnologiesadvocatedhere.Inthissense,thehandbook isactuallywellpositionedtosupporttheincreasinglydata-drivenAItechnologies. With the utmost enthusiasm, my sincere congratulationsgo to the authors and editorsforputtingtogethersuchanoutstandingcontribution. DepartmentofElectricalEngineering S.Y.Kung PrincetonUniversity Princeton,NJ,USA Preface In this new edition of the Handbook of Signal Processing Systems, many of the chapters from the previous editions have been updated, and several new chapters havebeenadded.Thenewcontributionsincludechaptersonsignalprocessingmeth- ods for light field displays, throughputanalysis of dataflow graphs, modeling for reconfigurablesignalprocessingsystems,fastFouriertransformarchitectures,deep neural networks, programmable architectures for histogram of oriented gradients processing,highdynamicrangevideocoding,system-on-chiparchitecturesfordata analytics,analysisoffiniteword-lengtheffectsinfixed-pointsystems,andmodels ofarchitecture. We hope that this updated edition of the handbook will continue to serve as a useful reference to engineering practitioners, graduate students, and researchers workinginthebroadareaofsignalprocessingsystems.Selectedchaptersfromthe bookcanbeusedascorereadingsforseminar-orproject-orientedgraduatecourses in signalprocessingsystems. Given the wide rangeof topicscoveredin the book, instructors have significant flexibility to orient such a course towards particular themesorlevelsofabstractionthattheywouldliketoemphasize. This new edition of the handbook is organized in three parts. Part I motivates representativeapplicationsthatdriveandapplystate-of-the-artmethodsfordesign and implementation of signal processing systems; Part II discusses architectures for implementingthese applications;and PartIII focuseson compilers,as well as models of computation and their associated design tools and methodologies. The chapters are ordered alphabetically by the first author’s last name in Parts I and III,whiletheyareorderedinPartIIstartingwithchaptersthatcovermoregeneral topics,andfollowedbychaptersthataremoreapplication-specific. Weareverygratefultoalloftheauthorsfortheirvaluablecontributions,andfor thetimeandefforttheyhavedevotedtopreparingthechapters.Wewouldalsolike ix x Preface to thank Courtney Clark, Caroline Flanagan, and Jennifer Evans for their support andpatiencethroughouttheentiredevelopmentprocessofthehandbook. CollegePark,MD,USA ShuvraS.Bhattacharyya Leiden,TheNetherlands EdF.Deprettere Aachen,Germany RainerLeupers Tampere,Finland JarmoTakala 13January2018 Contents VolumeI PartI Applications SignalProcessingMethodsforLightFieldDisplays....................... 3 RobertBregovic,ErdemSahin,SurenVagharshakyan,andAtanasGotchev InertialSensorsandTheirApplications..................................... 51 Jussi Collin, Pavel Davidson, Martti Kirkko-Jaakkola, andHelenaLeppäkoski FindingItNow:NetworkedClassifiersinReal-TimeStreamMining Systems.......................................................................... 87 RaphaelDucasse,CemTekin,andMihaelavanderSchaar DeepNeuralNetworks:ASignalProcessingPerspective.................. 133 HeikkiHuttunen HighDynamicRangeVideoCoding ......................................... 165 KonstantinosKonstantinides,Guan-MingSu,andNeerajGadgil SignalProcessingforControl................................................. 193 WilliamS.Levine MPEGReconfigurableVideoCoding........................................ 213 MarcoMattavelli,JornW.Janneck,andMickaëlRaulet SignalProcessingforWirelessTransceivers ................................ 251 MarkkuRenfors,MarkkuJuntti,andMikkoValkama SignalProcessingforRadioAstronomy..................................... 311 Alle-JanvanderVeen,StefanJ.Wijnholds,andAhmadMouriSardarabadi DistributedSmartCamerasandDistributedComputerVision .......... 361 MarilynWolfandJasonSchlessman xi