翻译基于DSP的单芯片多处理器的通信应用的解决方案.docx

上传人:b****8 文档编号:9686072 上传时间:2023-02-05 格式:DOCX 页数:13 大小:27.44KB
下载 相关 举报
翻译基于DSP的单芯片多处理器的通信应用的解决方案.docx_第1页
第1页 / 共13页
翻译基于DSP的单芯片多处理器的通信应用的解决方案.docx_第2页
第2页 / 共13页
翻译基于DSP的单芯片多处理器的通信应用的解决方案.docx_第3页
第3页 / 共13页
翻译基于DSP的单芯片多处理器的通信应用的解决方案.docx_第4页
第4页 / 共13页
翻译基于DSP的单芯片多处理器的通信应用的解决方案.docx_第5页
第5页 / 共13页
点击查看更多>>
下载资源
资源描述

翻译基于DSP的单芯片多处理器的通信应用的解决方案.docx

《翻译基于DSP的单芯片多处理器的通信应用的解决方案.docx》由会员分享,可在线阅读,更多相关《翻译基于DSP的单芯片多处理器的通信应用的解决方案.docx(13页珍藏版)》请在冰豆网上搜索。

翻译基于DSP的单芯片多处理器的通信应用的解决方案.docx

翻译基于DSP的单芯片多处理器的通信应用的解决方案

ASINGLE-CHIPMULTIPROCESSORDSPSOLUTIONFOR

COMMUNICATIONAPPLICATIONS

DavidRegenold

IntelCorporation

6505W.ChandlerBlvd.,M/SCH11-91

Chandler,Arizona85226

ABSTRACT

ThispaperpresentsanoverviewofanarchitectureforasinglechipmultiprocessorDSPsolutionforcommunicationsapplications.Theintegratedcircuitwasdesignedtohandletheprotocolanddata-pumpfunctionsnecessarytoimplementhigh-speedmodemandaudiotasks.Thechipconsistsofa186microcontrollerwithtwoDigitalSignalProcessing(DSPs)coprocessors.Itinterfaceswithastandard186busandhasaportforcommunicatingwithacustomCODEC,the80127.

INTRODUCTION

Computerconnectivityisaneverexpandingandpervasivemarket.StandardPOTS-lineMODEMScontinuetobepushedtohigherfrequenciesevenastheneedforhandlingsimultaneousvoiceemerges.Atthesametime,wirelesscapabilitiesareemergingandtheconceptofwirelessemailisbecomingarealitywiththeadventofpacketradionetworkssuchasMobitex,ArdisandCellularDigitalPacketData(CDPD).

Toaddressthesemarkets,theCommunicationsProductsDivisionofIntelhasdevelopedachip-setspecificallyarchitectedtomeetthegrowingMIPSdemandsoftheemergingcommunicationsapplications.Thechipsetconsistsofthe80127CODEC,whichwaspresentedatlastyear'sASICconference,andthe8OC186CPcontroller/DSPwhichwillbepresentedhere.

DetailedArchitecture

The80C186CPisahighlyintegratedmicrocontrollerderivedfromthe80C186EBandprimarilytargetedtowardshighenddatacommunicationapplicationslikehighspeedmodems(v.34rates)andaudioprocess-ing.Toafirstorderitissoftwarecompatiblewiththe80C186EB.However,the8OC186CPenhancestheEBwiththeadditionofahighperformanceDigital-Signal-Processingcoprocessorandaversatileandflexibleperipheralsetdesignedtoeffectivelyinterfacewiththenecessarycommunicationchannels,suchasana-logCODECs,andISAorPCMCIAbuses.

Figure1showstheblockdiagramofthe8OC186CParchitecture.Asthediagramindicates,the80C186CP

isbuiltaroundtwointernalbusesknownastheF-busandtheE-bus.TheF-busisthestandardinternalbusfoundonall80C186Exproliferationsandisusedfordatatransfersbetweenthe80C186core,theexternalAddress/Databusandthe80C186peripherals.Itisa20-bitbuswhich,liketheexternalbus,multiplexesthe16-bitdatawith20-bitsofaddress.

ThemajorityoftheperipheralsattachedtotheF-busareidenticaltothosefoundonthe80C186EB.TheyincludetheEB'schip-selectunit,timerunit,interruptunit,parallelportunit,andserial-portunit(theEBhadtwo).Theclockpower-managementunitissimilartothatfoundonthe80Cl86EB,butithasbeensubstantiallymodifiedaswillbediscussedlater.TheEXcoreandallmodulesontheF-busareclockedat20.736MHzinthetargetapplications.

NewF-busperipheralsnotfoundonthestandard80C186EBincludeahost-interfaceunitandanHDLCaccelerator.Thehost-interfaceunitallowsthe80C186CPtobeattachedtoeitheranISAoraPCM-CIAbus.Itcontainsa16550emulator,whichmakesthe8OC186CPappeartothehostasmerelya16550UART.Inactuality,thedatatransferredbetweenthe16550emulatorisnotserializedasinarealUART,buteithertakenfromthecoresinthe8OC186CPorpresentedtothemfordatamanipulationasrequiredbythecommunicationstandardbeingimplemented.WheninPCMCIAmode,theunitsupportsaccessestotheCardInformationServices(CIS)memoryandcontainstheotherspecialfunctionregistersdefinedbythePCMCIAstandard.

DatabeingtransferredacrossthehostinterfacemustsometimesbeHDLCencodedordecodeddependingagainonthestandardbeingimplemented.Forthispurpose,anHDLCacceleratorisprovidedtoalleviatetheEXcorefromtheintensebitmanipulationthatwouldotherwiseberequired.UnlikeatypicalHDLCcontrollerthatreceivesandtransmitsdataserially,thisunitreceivesdatafromtheF-businparallelformsincetheEXcoreisdealingexclusivelywithparalleldata.TheHDLCacceleratorviewsthedataasaserialbitstreamandperformsthetasksofCRCgeneration,zerobitinsertionordeletion,anderrordetection.

AnothermoduleontheF-busnotfoundinother80C186EXproductsisthecoprocessor-interfaceblock(orarbiter).ThisunitisdiscussedaftertheDSPcoprocessor.Sufficeittosayfornowthatthismoduleplaysapivotalroleinthe8OC186CPchipandactsasthecentralpointofcommunicationbetweenthe80C186sideofthecontrollerandtheDSPcoprocessor.TheDSPcoprocessorconsistsprimarilyoftwoDSPcoresknownastheEP+cores,severalmemorybanksforprogramanddatastorage,andsomeperipherals,suchasaCODECinterfaceandaViterbiaccelerator.

AblockdiagramofanEP+isshowninFigure2.TheseDSPcoresaresimpleMultiplyandAccumulate(MAC)enginescapableofexecutingapeakrateofoneMACperclockcycle.Forthetargetapplication,the80C186CPisclockedat20.736MHz,whichresultsinapeakthroughputofmorethan40millionMACSpersecond.

TheEP+ALUisfedfromtwointernal256-wordRAMsthatcontaindataandcoefficientsrequiredbythealgorithms.One16-bitwordcanbefetchedfromeachSRAMeveryclockcycleinordertofeedtheMACengine.Themultiplierproducesa32-bitproductthatisaccumulatedina32-bitregister.Additionalhardwareisavailabletoassistdivision.

Accesstomemorybeyondtheinternal512wordsisachievedthroughamemory-expansionunitthatsitsonan8-wordparallelportoftheEP+calledtheEXTbus.Throughthisexpansionport,theEP+canaccessupto64kwordsofadditionaldata(althoughonly3.5kisactuallypresentinthisimplementation).TheadditionaldatamaybepresentonaprivatebusoftheparticularEP+ormayresideonthesharedE-bus.DatacanbeaccessedthroughthisportatapeakrateofonewordperclockcycleandreplacesoneoftheDSP’sinternalSRAMswheninuse.

TheDSPsaresuppliedtheirprogramcodeanddatathroughseveralbanksofSRAM.SincetheyareofaHarvardarchitecture,theyeachhavetheirownseparateprogrambusadseparateprivatedatabus.Four512-wordSRAMsareprovidedforprogrampurposesandfour256-wordSRAMsareprovidedfordatapurposes.AnyofthefourprogramRAMscanbeconnectedtotheprogrambusofeitherEP+whileanyofthefourdataRAMscanbeconnectedtotheprivatedatabusofeitherEP+.

TheE-busisasingle-clock-cycle-transferbusconsistingof16addressbitsand16databits.ItmaybeaccessedbyeitherEP+coreortheEXcore.TheDSPscantransferdataacrossthebusatarateofonewordperclockcycle.However,theEXcoreisstilllimitedtoonewordeveryfourclockcyclesduetothelimitationsoftheF-bus.AnyoftheprogramanddatamemorybankscanbeconnectedtotheE-bussothatanyofthethreecorescanarbitrateforaccesstothem.

ProgrammingofthememoryconfigurationoccursthroughtheIPCBblock.Thismodulealsoactsasacentral-switchingofficeforinterrupts.Interruptsignalsfromalldifferentpartsofthesystemconvergeonthisunitandaresenttoanyoneofthethreeprocessorsaccordingtoprogrammingofaregisterwithintheunit.

AttachedtooneoftheDSPsisaViterbiAccelerator.TheViterbialgorithmisencounteredfrequentlyincommunicationsbuthaslittletodowithMultiply/AccumulatesandthusisnotimplementedefficientlybytheEP+architecture.TheEP+usesthismodulebysimplywritingaseriesofdatapairstoitandthenreadingtheresultingmaximumorminimumofthesumsofthedatapairsandthecorrespondingindex.

Datareceivedacrossacommunicationsmedia(eitherwiredorwireless)comesintothe8OC186CPfromanAnalogFrontEnd(the80127AFE)viatheHigh-Speed-Synchronous-SerialPort(HSSSP).ThisunitusesaninterfacefoundonmanyindustrystandardDSPstocommunicatewiththeAFE.TheHSSSPhasaccesstotwomore256-wordSRAMsbetweenwhichitfetchesandstoresdata.ThisallowstheHSSSPtotransmitandreceivetwoblocksof256wordsinfull-duplexmodewithoutinterventionfromanyoftheprocessorsonchip.ThemoreusualcaseistohavetheHSSSPtransmitandreceivetwoblocksof128wordsinfullduplexmodefromasingleSRAMwhiletheEP+workswiththedatapreviouslyreceivedorsoontobetransmittedintheotherSRAM.

Thelastmajorblockonthe80C186CPisthecoprocessorinterfacemodule.Thisblockismultifunctional.ItallowstheEXcoretoaccessallofthememoryandSpecialFunctionRegistersresidingontheE-busasaregionintheC186's1MByteaddressrange.Italsoactsasthearbitrationunitbetweenthethreeon-chipprocessorsforaccesstotheE-busanditcontainsastatusregisterthroughwhichtheEXcorecanmonitoractivityontheE-bus.Butitsmostimportantfunctionisthatofafly-byDMAunittotransferdatafromtheF-bustotheE-busorvisa-versa.

ThecodeanddatamemoryrequirementsoftheDSPalgorithmsrunningonthetwoEPsareusuallylargerthanthefeasibleonchipmemory.Therefore,afastandlowoverheadmethodofdownloadinghploadingcodeanddatatotheonchipmemoriesfromoff-chipprogrammemoryisessential.Thisisaccomplishedviathefly-byDMAwhichresidesinthecoprocessor-interfaceunit.AnyofthethreeE-busmasterscanutilizetheDMAbyprogrammingthenecessaryparameters,suchasthestartingaddress,endingaddress,andwordcount.

Theprogrammingofthefly-byDMAtakesadvantageofthefactthattheDSPfirmwarerunningontheDSPsisfairlypredictable.Thatis,thecurrenttasksrunningontheDSPsshouldknowwhichsoftwaremodulesaremostlikelytobeneedednext.Thecur-renttask,throughawritetoaregister,informstheDMAunitthatitwill

展开阅读全文
相关资源
猜你喜欢
相关搜索
资源标签

当前位置:首页 > 人文社科 > 视频讲堂

copyright@ 2008-2022 冰豆网网站版权所有

经营许可证编号:鄂ICP备2022015515号-1