Acoustic Bedload Velocity Estimates Using a Broadband Pulse
Transcription
Acoustic Bedload Velocity Estimates Using a Broadband Pulse
Acoustic bedload velocity estimates using a broadband pulse-pulse time correlation technique Don W. Sutton and Jules S. Jaffe Scripps Institution ofOceanography, MarinePhj•sical Laboratory, La Jolla,California92093 (Received10December1991;acceptedfor publication4 May 1992) A novelnoninvasive instrumentfor measuringbedloadvelocityhasbeendevelopedusinga narrowbeam,high-frequency underwatersonarsystem.Pulsedwidebandsignalscenteredat 2.2 MHz are usedto insonifya bedloadat a 30-deggrazingangle.Under theseconditionsthe finestructureof thereturnedsignalsis dependent on the inter-particlespatialrelationof the bedloadparticleswithin theacousticbeam.A crosscorrelationof two consecutive backscattered waveformsis usedto obtainan estimateof the inter-pulsetranslationof the bed. Thispermits a maximum likelihood estimate oftheve!ocity ofthebed19ad. Theresults froma seriesof staticexperimentsin whichagarembeddedsandwastranslateda precisedistance showedan excellentcorrelationbetweenmeasured andestimatedtranslations usingthesignal processing method.In a setof dynamicexperiments, the velocityof a jet-drivenbedloadwas estimatedwith a standarddeviationin the estimatesof approximately5%. Thesewerefound to be in goodagreementwith independentvideocameraestimates. PACS numbers: 43.30.Pc, 43.30.Ma INTRODUCTION pling frequencies.Determination of particle velocities, Existingmethodsusedto measurebedloadtransport havenot adequatelymet the needsof thosestudyingthese transport processes.Bottom sedimentedparticle, or bedload, transportdrivenby a fluid shearfield is an important fundamentalissuein sedimentology, hydrology,andcoastal studies. Relevant bedload issues in these areas include alter- ationsto river topologies(bottoms,courses,and mouths), along with coastalerosion including beach erosion, and blockageof harbormouths.Satisfactorybedloadtransport measurementtechniqueswould allow for monitoringof theseenvironments, aswell as,providinga meansto fundamentallyresearchbedloadtransportby aidingin the developmentandsupportof mathematicalmodelsfor thesetransport processes. Pastmeasurement techniques havereliedmainlyoncollectionsamplingand opticalincludingphotographic techniques.Both methodsare handicappedby critical limitations. Collectionmethodsare invasive,tendingto alter the transport process duringsampling. ,,2Opticalmethods consistingof photographicor laserDoppler techniquesare difficult to setup andtime consumingto process. They are also both limited by the surroundinggeometryand fluid turbidity nearthebedloadregime. 3'4 An acousticsensingmethodhasa numberof desirable characteristics in regardto bedloadtransportmeasurement. It can provide a noninvasivetechnique.The neededhardware is simplistic,and its useis not limited by the interferenceof smallsuspended particles.Furthermore,signalprocessing methods can be developed to yield real-time however, has not been addressed. Thecurrentarticledescribes a newlydeveloped acoustic bedloadvelocitymeter (ABVM), whichusespulsedbroadband waveformsand a temporalcorrelationtechniqueto providean instantaneous estimateof the bedloadvelocity.A mathematicalmodel is presentedthat characterizesthe transmittedandreturnedwaveformsignals,andoutlinesthe methodsused to processthem. The resultsof laboratory measurementsmade under static and dynamic bedloadconditions to test the model and determine the characteristics of the ABVM are alsopresented. I. SIGNAL PROCESSING Our processing methodis basedon a cross-correlation techniquethat utilizescoherentchangesin bedloadparticle positionsbetweensuccessive pulsesto yield a velocityestimate. Recently, similar techniqueshave been proposedto measure bloodflowvelocities, 7-9however, toourknowledge this techniquehas not been appliedto oceanography. An essentialrequirementof this techniqueis that the bedload particlediametersareof thesameorderof magnitudein size asthat of the transmittedwavelength.This resultsin a strong scatter signal that has a fine structure unique to the inter- particlespatialrelationof that portionof the bedloadsheet that intersectsthe acousticbeam.By insonifingthe bedload at a pulserepetitionratethat is rapidenoughto ensurethat this inter-particlespatialrelationdoesnot changesignificantlybetweenpulses,yetallowsfor a significanttranslation (i.e., sheet flow), similarities in the fine structure of succes- clesviabackandsidescatter returns. 5'6In thesestudies par- sivereturn signalscan be usedto yield a time shift correspondingto thistranslation.As demonstrated, thistimeshift canthenbeusedto provideanestimateof thevelocityof the ticle sizesare detectedby analyzingreturnsat varioussam- bedload. information. Related acousticmethodshave focusedmainly on sizeandconcentrationmeasurements of suspended parti- 1692 J. Acoust.Soc. Am. 92 (3), September1992 0001-4966/92/091692-07500.80 ¸ 1992 AcousticalSocietyof America 1692 In general,bedloadparticlevelocitiesare not uniform and the correlationmethodyieldsa maximum likelihood estimateof spatiallyaveraged particlesvelocities. The lower bound of the variance of this estimate is also discussed. thebeam.Thiscanbemodeled by a factorf(d•(t)). If z,ø.is theith particleposition at t = O,thentheparticleposition at timet isz•(t) = • + v•cos•t, where,o• is the inter-pulse velocityfor particlei with positivevaluestakento be away fromthe receiver.Ifg(t) is transmitted at time t = r, then thesignalwill bereceived fromparticlei, 2z•(r + t')/c later, where z• (r + t') is the particle positionwhenhit by sound The geometryof the area consistingof the acoustic leaving the transmitter at r and t' isthetimeforthesoundto hardwareand bedloadis shownin Fig. 1. The bedloadis travel from the transmitter to the particle: insonifiedusinga narrowbeamtransducersuspended above thebedloadat a grazingangle6. The receivingtransducer is zi(r + t') =z,Oq- (r + t')v• cos4•. parallelto the transmittingtransducerand at an equaldisSincect' isequivalenttoz•(r + t' ), t' canbeexpressed as tancefrom thebedload.The particlesareassumed to beuniz,ø. cos formlydistributedoverthebedloadwith the ith particleat a distancezi from the receivingtransducer,and at a radial c - " cos0 c - v,.cos distancedi from the receivingbeamaxis.Comparedto the Therefore,after somealgebra,the delay 2z•(7 q- t ')/c beamwidthat the bedload,zi is assumedto be large. The can be shownto be equalto receivedscatteredsignalsareassumedto befrom singlescatA. Model for velocity estimates ter eventsonly. Assumingthatg(t) is the transmittedsignalfilteredby theimpulseresponse of thetransmitting andreceiving transducers,the analogfilters,and remainingelectronics,and 1o ß (t,z,qb,a)the scatterfunctionfor a particleof radiusa, then, the scatteredsignals(t) can be representedas g(t) convolvedwith •(t,z,(b,a). In general,the receivedsignal r(t) will dependontheradialpositionof theparticleswithin 2ziø cos + c - vi cos c-vcos Thisimplies thatforasingle particle, thereceived signal canbe representedas ( Yi 7 q- C-- Ui COS • ) {- AZi =f(di(7, t'))s(7). where AZ i -- c -- ui COS • ' Now, considerthe completesignalas a sumof the receivedsignalsoverall particles.Then,the response fromtwo consecutive pulsesignals,which havebeenspacedat time period T apart, can be representedas N N r,(t) = • r,(t)= • f(d,(7,t'))s(7), i::-0 i=0 (1) N+b r2(t)= j-b• ri(t-- T(I N = • f(di(r-- T,t'))s(r-- T), i ::= 0 (2) where (b) 2vi cos• O•i -- C -- /3i COS• and t=7(1 +ai) +AZ•. Althoughexactlythe samenumberof particlesmay not move into the beam as move out, here, the total number of [] FIG. 1. A schematicdiagramof the acoustichardwareand bedloadarea. (a) is a side view, (b) is from above. Three transducersare used,transmitting 1, receiving2, and reference3. Both I and 2 arc at a grazing angle • and 3040 cm abovethe bedload.The referencetransduceroperatesat a delay synchronized to the transmittingtransducer. A bedloadparticleis shown with velocityvi at a distancezi from the receivingtransducer,and at a distancer• from its beamaxis. 1693 J. Acoust.Sec. Am.,Vol.92, No. 3, September1992 particlesNcontributingto a returnisassumed to beconstant forbothpulses. The actualparticlescontributing a returnfor eachpulse,however,will changeslightlyas a resultof the bedloadtransport.This hasbeenmodeledsuchthat for r2, particlesindexed0 to b havemovedout of the beamat one end,whileparticleN to N + b havemovedintothebeamat the other end during the time T. D.W. Suttonand J. S. Jaffe:Bedloadvelocityestimates 1693 A crosscorrelation of the two (finite duration) returns is givenby Q.... (7-) =f rl(t)r2(t d-•')dt. (3) If the pulserate is chosensuchthat the inter-pulsedistancetranslatedby thebedloadparticlesis smallcompared to the beamwidth, then, the sameparticlesconstitutethe majorityof thereturnfor thetwopulses, i.e.,b•N. Furthermore,for thisconditionthe effectof the beamsensitivityfor a particularparticlewill not changesignificantlyduring a pulse period, i.e., fldi(r,t'))=fldi(r-T,t')). Under thesecircumstances, r2(t), the signalreceivedfrom a sound sentT seconds later, canbe summedfrom j = 0 to N and reduced to M/2 is the noisespectraldensity,E is the energyof signal r(t), andR (•o) theFouriertransformof r(t) forfrequencies co.This equationcanbe reducedto a moreconvenientform assuming theautospectrum istwosidedwithcenterfrequency•oo,bandwidthIF, anda signal-to-noise ratioSN:13 o'>•.... 1 I 1 1 /3•x/•Sx/g•COo • 1 (10) x/1d-W2/(12cog ) In particular,theminimumvariance,or maximumreso- lution,isproportional to theinversesquarerootof thetimebandwidthproductandthe centerfrequency. Thus,widebandsignals havea smallerlowerbound.Furthermore, fora givenbandwildth signals witha highercenterfrequency and thusfewercycles perpulsewidth will performbetter. N r2(t)= j=o• r•(t-- T(I +atj)}. (4) II. EXPERIMENTS Furthermore,if all particleswithinthebeamhavea constantand equalvelocity,then this can be expressed more simplyin termsof r• as r2(t) •- r•(t- T- aT) (5) so that Q.... (7-) =;ri(t)r•(t - T-aT+ 7-)dt, whichis the autocorrelation functionof r• that hasbeen shiftedby an amount T + a T or Qr,r_, (7-)-•-Q....(7-- T- aT). (6) Staticand dynamiclaboratoryexperiments were performedto aidin thedevelopment of theABVM andto assess its performancecharacteristics.Statictests,wherea simulated bedload was translated in a controlled manner between acousticpulses,wereusedto assess the earlydevelopments of thecorrelation methods, theeffectof thebeamsensitivity, andthe decorrelation of the returnsignalsasa functionof translationdistance.Dynamictestswerecarriedout to demonstratethefeasibilityof thesignalprocessing methodsand overalltechnology, and to characterize the performance of the ABVM in a more realistic environment. This function has a maximum at 7-= T + Ta. In the absence of the Dopplereffect,the shiftwouldsimplybe T. A. Methods Thus the Doppler effectresultsin a time shift of the cross- 1. Systems hardware correlated signals byanadditionalamountAt = T2v cos•/ A broadband transmitted signalwasgenerated usinga (c -- v cos•). This timediferentialcanbeusedto determine waveform generator (Wavetek, San Diego, CA), which conthebedloadvelocityby expanding terms(assuming v• cos• sisted of 2 cycles at 2.2 MHz and pulsed at 100-400/zs at 10 ,•c) and to a first-orderapproximation canbe represented V peak-to-peak. The recieved bedloadscattersignals were as preamplified (AnalogModulesInc.,Longwood, FL), bandv = Atc/2Tcos q•. (7) passfiltered(600kHz-6 MHz), andsubsequently digitized Duarte,CA). Thesesignals In processing the signals,the cross-correlation function at 40MHz, 8 bits(Markenrich, for a consecutive pair of returnsis normalizedby the maxi- weredirectlywritten to data fileson a PC (IBM PC/AT, mum values of the autocorrelation functions of the two indi- BocaRaton,FL) andlatertransferred for processing to a vidual returns: semi-mainframe(Sun Microsystems).All hardware was Q....(•')= triggered usinga pulsegenerator(Hp, PaloAlto,CA). All of thetransducers usedhadcenterfrequencies of 2.2 MHz with 6-mmdiameters, andproduced narrowbeamswith a Q....(7-) x/max(Q ....(r) )x/max(Qr, r,(7-)) If T is chosensuchthat a significant peakresultsfrom thenormalized cross-correlation (i.e.,b• N), thenEqs.(7) and (8) can be used to obtain a maximum likelihood esti- mate for the bedloadvelocity. a smalloffsetnecessary for thebeamsto intersectat the bed- B. Minimum error analysis A lowerboundfor the variance,•r, in the velocityestimateobtainedfrom Eqs. (7) and (8) canbeobtainedusing theCramer-Raoinequalilty. ••'12 aa> • YIR(co)12dco /3 2E J'co2[R(co)[2 dco' where/• = (2 T cosq)/c, 1694 J. Acoust.Sec. Am., Vol. 92, No. 3, September 1992 -- 3-dBattenuation at 2 deg( Panametrics, Waltham,MA). The transmitting andreceiving transducers werefixedin a planethatintersected thebedloadat a 30-deggrazingangle, andat a distance of 30to 40 cmabovethebedload(Fig. 1). Theywerepositioned nearlyparalleltoeachotherexcept for (9) loadsurface. A thirdtransducer, placed15cmawayfrom,, andaimedat, the receiving transducer generated a signal synchronized with the transmitter.This provideda referencepoint on the receivedsignalsto be usedas a meansto accuratelydeterminethe actualpulserate.Time shiftsbetweentheautocorrelated andcross-correlation signals were determined from this reference. D. W. Suttonand d. S. Jaffe: Bedloadvelocityestimates 1694 2. Static tests A testenvironmentwasassembled that allowedfor precisestatictranslationsof the acousticsystemrelativeto the bedload. Thisapparatus consisted ofa 0.7-m3tankwitha 2D translatingstagesupportingthe acoustichardwarefixed above the tank. Translations were made via a lead screw with of0.01mmfor distances translated upto 9 mm,witha 4.9% standarddeviation(Fig. 2). The static results were also able to characterize the de- correlationof the signalsas a functionof the translateddis- tance(Fig. 3). It wasfoundthat peakheightsfor the normalized cross correlations had values near 0.90 for translationsup to 1.5 mm, fallingto 0.5 near6 mm. a 1.264-mm/revpitch.Threesimulated bedloads consisting Returnsfrom controlsof sand-freeagar trayswereunof separatenarrowgradientsof sandsizes(0. l, 0.5, and 1 detectablefrom backgroundnoise,indicatinga negligible mm diameter) fixedin traysof agar solutionwereusedas scatterfrom the agar mediumand the bottomof the sand acoustictargets.A seriesof translationswere madewith trays, and no detectablecontributionfrom the transducer acousticreturnsgeneratedat each translatedincrement. sidelobes. Normalized cross correlations were used to estimate the translation distance ß by equating the expression (cos•c) to the time shiftAt, the differencebetweenthe peak locationsof cross-correlated signals,and T, thepulseperiod. Estimatesfor ß werethencomparedto actualtranslations. 3. Dynamic tests A 5-mportionof a longchannel(30 m in length)witha wettedcrosssectionof 60 by 60 cm anda bottomconsisting of 1-mm-diamsilicasandwasusedasan experimental bedload. A stationaryarrangementof the acoustichardware similarto thatof thestaticexperiments wassuspended above thebedload(Fig. 1). A fluidshearfieldprovidingtheenergy At 2.2 MHz it wasnot known asto what bedloaddepth particleswereableto contributea significantreturn.Experimentsusinga buriedreceivingtransducersensingtransmissionsfrom 30-deggrazingangle,showedthat the -- 3-dB cutofffor soundpenetrationinto a stationarybedloadconsistingof l-mm-diamsandwaslessthan3 mm in depth. As an exampleof typicalreturnsobtainedfrom the dynamicbedload,twoconsecutive signalsobtainedfromthejet drivenbedloadare shownin Fig. 4. Onsetof thesesignalsis determinedby the return from the referencetransducer. Some similar features in their fine structure between the two signalscan be seen.Normalizedautocorrelatedand crosscorrelatedwaveformsof thesesignalsare shownin Fig. 5, with thecorresponding timeshiftAt betweentheautocorrepeaksshown. From this time totransport thebedload wasgenerated usinga high-pressure lated and cross-correlated flowemanating froma specially fabricated nozzleconsisting shift,an estimateof the velocityis arrivedat usingEq. (7), of a 3-by 15-cmfacewith approximately100closelyspaced for this case a value of 67 cm/s. holes (2 mm diameter). The nozzle was positioned5 cm Our systemwassetup suchthat data signalsconsisting abovethe bedloadat a 5-deggrazingangle.Acousticdata of 32kpointswereobtained,eachcontainedapproximately5 weretakenduringshort5-sburstsof thejet witha resurfac- pulsereturns.TableI showsthe timeshiftsandcorrespond- ingof thebedloadbetween samples. The samples consisted of 32 k datapointsandincluded4 to 8 pulsereturnswhen pulsedat 100to 200/rs.In someexperiments anunderwater videocamerawasusedto simultaneously recordthebedload motion. Particles that could be uniquely identified on successive videoframeswereusedto obtainan independent velocityestimate. The onsetof returnsfrom individualpulseswithin the 10 - 1.0 mm send - 0.5 mm send /- - 0.1 mm sand 32k datapointsamples wereestablished usingthereference signalasa starting point.Theseindividual signals werethen digitallyfiltered,autocorrelated, andcrosscorrelated with nearestneighbors. The shiftsfromcenter,or autocorrelated peaklocation,for thecross-correlated peaks,labeledasAt, werethenusedfollowingEq. (7) to estimatethe velocity.A resolutionbelowthe40-MHz (0.025-/•s) samplingrate was obtainedviaa quadraticcurvefittingto establish timescorresponding to peak locations.Heuristically,consistent resultsindicatethat by thismethoda 0.005-/•sresolutionwas possible. B. Results 2 lation. Estimatesusingthe correlationtechniqueshoweda linearresponse to actualtranslations for all threesandsizes. For the 1-mm sand,the correlationmethodhad a resolution 1695 J. Acoust.Sac. Am.. Vol. 92. No. 3. September1992 • • 6 8 Act,soltranslateddistonce(ram) Resultsfrom the staticexperiments indicatethat the time eorelationtechniqueusingacousticbroadbandsignal returnsis an accuratemeansof estimatingthe bedloadtrans- • 4 FIG. 2. Actual translateddistancesplotted againstestimatesusingthe cross-correlation technique[Eqs.(7) and(9) ] fromthestaticexperiments. The data are obtainedfrom three sandsizesof separatenarrow gradients (0.1, 0.5, and 1.0 ram) set in an agar. The standard deviation from the dashed 11ineis 4.9%. D.W. Suttonand J. S. Jaffa: Redloadvelocityestimates 1695 1.0 •.o• (a) 0.8 [] [] El -1-0 1-0 0.0 0 • • • • • 2 4 6 8 10 Translateddistance (ram) FIG. 3. Valuesfor the normalizedcorrelationplottedagainstthedistance translatedby the bedload.Data are obtainedfrom the staticexperiments usingthe 1.0-mmsand. (a) -1-0 t (izsec) FIG. 5. Magnified viewsofthenormalized autocorrelation of thesignalin Fig.4(a), (a), andthenormalized cross correlation ofthesignals inFig.4, (b). For thisexample, thetimeshiftbetween thetwopeaks,At, is0.155ps (or aboutsixsamplingunits,at 40 MHz) corresponding to a 67-cm/sbedload velocity. I I I I (b) ing velocitiesfrom nearestneighborcorrelationsfor 4 setsof these32k length signals.Useful comparisonscan only be madewithineach32 k lengthsample,sincefor our laboratory setupthe bedloadbecomes deformedby thejet beforethe systemis readyto accepta subsequent 32k file. A resolution of 5 ns (one fifth of the 40-MHz samplinginterval) was obtainedfor establishingcorrelationand triggerreference peak locationsby useof the quadraticcurve fitting scheme. The pulseperiodbetweensignalsvariedby about0.3 sampling intervalunits (7.5 ns at 40 MHz), however,this error waseliminatedby matchingexactpulseperiodsto their correspondingcorrelationestimte.As outlinedin Table I, the standarddeviationsfor the velocity estimatesare around 5%. I o I 5o 1 oo t (Izsec) FIG. 4. Examplesof twoconsecutive returnsobtainedat a pulserepetition rateof 200/•s fromthejet shearfieldexperiments. Signal(b) follows(a). The triggerreferences are at the far left. 1696 J. Acoust.Sec.Am.,Vol.92, No.3, September1992 The estimatesfrom the video analysiswere consistent with thoseobtainedusingtheABVM. Averagevelocityvaluesfor 5 returnswithin a 32k signalwerewithin 15% of the videoframeanalysisestimate.This 15% discrepancy iscomparable to the error found when usingthe video method, whichis handicapped by the videolimitationof 30 sampling framesper second. D.W. SuttonandJ. S. Jaffe:Bedloadvelocityestimates 1696 TABLE I. Velocityestimates obtained fromfoursignals consisting of 32 k datapoints,eachwithfivepulsereturns. Measurements aremadebetween autocorrelation andnearest-neighbor cross-correlation peaks, withthetimeshifts reported insampling interval unitsat40MHz (0.025#s). Pulse periods for thesignals ranged between 130to 200/•s. Sample1 Sample2 Velocity Time shift cm/s Time shift 4.20 69.8 4.78 79.4 4.46 4.89 74.1 81.2 76.1 + 4.5 Sample3 Velocity Sample4 Velocity cm/s Time shift 4.83 76.8 4.68 74.4 5.01 5.23 79.7 83.2 Velocity cm/s Time shift cm/s 3.28 52.2 5.69 65.2 3.59 57.1 5.81 66.6 3.39 2.95 53.9 46.9 5.18 5.31 59.3 60.8 78.5 + 3.3 52.5 _+3.7 63.0 + 3.0 III. DISCUSSION ticles and the transmission attenuation are relevant consid- Analysisof Doppler-shiftedsignalscan be a useful meansof determining targetvelocityprovidedthefractional changesin timeor frequencyare significant.Bedloadvelocitiesare ableto produceonly slightshiftsin the carrierfrequencyof signalsat MHz frequencies. CarrierDopplershifts resultingfrom transportof suspended particleshave been determinedby averagingindividualestimtesover long per- erations in choosing this frequency. The scattering coefficient from a rigid sphereis, in general,dependent on iods.,4.•sAsanalternate means thecarriercanbepulsed. By 1-mm sand. pulsingthesignal,thelowerfrequencyof thepulseratewill havea moresignificantDopplershift.This scheme,known ascoherentor pulse-pulsecoherentDoppler,hasbeenproposedandcarriedout in a numberof oceanographic applica- Success of the overallmethoddependson the ability to obtaina strongcrosscorrelationbetweensuccessive pulse returns.The digitizationscheme,and the geometricoutlay of theoverallsystem,are importantissues that playa rolein determiningthe decorrelation of the signalsasa functionof tions.•6'17 The majorityof thesefavora narrow-band approachand usea frequencydomainaveragingtechniqueto estimatethe velocityby meansof a phaseshift of the envelopeof the returnedsignal. Our approachhas two distinct advantagesover this method.By utilizinga time domainprocessing technique ambiguitiesin phasealiasingare eliminated.Secondly,the traditionallyusednarrow-bandedsignalresultsin a higher valuefor the lower boundof the variance[Eq. (10)]. A narrow-bandedsignalwill havea wideautocorrelationfunction and henceits peak,whichis usedto detectthe velocity estimate,will havea greateruncertaintyin position.In past coherentDopplermethodsan averagingtechniquehasbeen carriedout to compromise this.Usinga broadbandsignal,a narrowerestimateof the velocitycanbe madefrom a single correlation,reducingthe needto average.Also, for widebandsignalsvariancesin the estimateswill more likely refleettheactualdistributionof theparticlevelocities. Results from the staticexperimentsshowedthat the width of the crosscorrelationsremainedconstantthroughoutthe range of translations(Fig. 3), indicatingthat any spreadseenin the dynamicestimatewill likely reflectthe velocitydistribution. The higherresolutionand fasterestimationcapability obtainedfrom broadbandsignalswill allow the systemto detecthigherfrequencyoscillationsin particlemotions,a featurethat is especiallyimportantin coastalstudyapplications.Broadbandschemes haverecentlybeenproposedfor acoustic currentprofllers.'8 An importantdesignparameterof thesystemisthecarrier frequency.The scatteringcoefficient of thebedloadpar1697 J. Acoust.$oc. Am_.Vol. 92. No. 3. September 1992 trigonometric functions of oddpowers of ka.•oHighvalues of the wavenumberk, however,needto becompromised by other factorsinfluencedby k includingbeamwidth,overall beamfield,and signalattenuationin water.Considerations of these issues led to our choise ofa 2.2-MHz carrier for the the translated distance of the bedload. Characterization of theseeffectswasthe primary concernof the statictests.Results of these tests indicate that a decorrelation of 10% oc- cursfor translationsup to 1.5 mm (Fig. 3). This is mainly dueto the 40 MHz digitizationrateof the 2.2-MHz carrier, resultingin a 5.5% uncertaintyof thephase,whichleadsto the 10% reductionof the normalizedcross-correlated peak height.Beyond1.5-mmtranslations, furtherreductionsin the peakheightcanbeattributedto an increasein the number of incoherentreturnsfrom new particlesenteringthe beambetweenpulses.For our system,decorrelationresulted in peak heightsthat matchedthe backgroundnoiselevelat about 9-ram translations. Thus, for translations within this limit velocityestimatescan be obtainedusingEqs. (7) and (8) provideda puresheetflowof the bedloadismaintained. Using the relationderivedfrom the Cramer-Rao inequality[Eq. (10) ], wecanpredictthe lowerboundof the resolutionof our system.The relevantparametersare a 2.2MHz centerfrequency,a signal-to-noise ratio of about30, a bandwildthof I MHz, anda pulsewidth of 50/rs.Theseyield a lower bound of approximately7 mm/s for the velocity variance.Paststudieson Doppler systemshaveshownthat actualvariancesalthoughcloseto their respectiveCramerRao lower bound do in fact exceed them. 17To achieve this resolutionwith the 40-MHz samplingrate and quadratic curvefittinga pulserate greaterthan 600/•s wouldhaveto be used.For our jet-driven bedloadwith a velocityof about 70 cm/s, consecutive return signalswoulddecorrelateto the levelof backgroundnoiseat abouta 200-/zspulserate. D.W. Suttonand J. S. Jaffe: Bedloadvelocityestimates 1697 Selectionof the optimalpulserate is dependent on the bedloadvelocityandthesamplingcapabilities of thesystem. The upperlimit of 9 mm for inter-pulsetranslationsin the statictestsis feasiblefor dynamicsituationsonly if incoherentvariationslateralto thebeamaxisaresmallcomparedto the particlesize.Ideally, the pulserate shouldbe selected just long enoughto allow for a significanttime shift in the peaklocationof the normalizedcorrelation,yetslowenough to maintaina highpeakvalue.If Tis too small,a smallshift will be detected,yieldinga poorresolutionin the estimate. On the other hand, largevaluesfor Twill resultin a significantdecorrelation of the signals. This occursat our 200-/zs limit for thejet flow. Factorssuchas the carrierfrequency anddigitizationrate arealsoof consequence in selectingT. The resolutionof the systemismostdependenton the digitizationrate.Usinga quadraticcurvefittingtechniqueon the 40-MHz digitizationwe wereableto achievea 0.0075-mm resolution.Our dynamic testswith the jet-driven bedload resultedin inter-pulsetranslations of 0.07-0.4 mm for pulse repetitionratesof 10-5 kHz. At thesevaluesthe signal-tonoiseratiowasabout40. Decorrelationto the pointof backgroundnoisewasseenat about1.5-mmtranslations. The bedloadparticlevelocitydistributionas a function of depthis currentlynot addressed by our ABVM. At 2.2 MHz we foundthat thesignalattenuationwithin thestationary bedloadto be -- 3 dB at a 3-ramdepth.Thusthe majority of the powerof the bedloadreturn is generatedfrom a shallowdepth.The estimatefrom our correlationmethod canbethoughtof asa spatialaverageof mainlysurfaceparti- directionof a uniformfluidshearfieldactingon thebedload. In general,the directionof shearmay not beknown,or perhapsbe variable.In order to assess the transportin these situations,a pairof orthogonaldevicessimilarto ourABVM will be needed.Further developments may alsoincludearraysof thesedevicesto givea spatialcharacterization of the transportwhich is importantfor the studyof bedloadsdriven by oscillatoryor other complexshearfields. ACKNOWLEDGMENTS This work wassupportedby the Army Corpsof Engineers under contract number DACW39-89-K-0023. Fund- ingwasfrom thecoastalengineering researchprogramproject STILE (Sediment Transport Instrumentationfor the Littoral Environment)directedby ThomasE. White. The authorswouldlike to thankItzhackLevyfor hisengineering assistance, Edith Gallagherfor her laboratoryhelp,and Richard Seymourfor his adviceon the project. •M. J. Crickmore, andR. F. Aked,"Pump•amplerfor measuring sand transport in tidalwaves,"Conference onInstrumentating Oceanography, Proc.No. 32, Bangor,England(1975). 2A.P.Salkield, G. P. LeGood, andR. A. Soulsby, "Amimpact sensor for measuringsuspended sandconcentration,"in Proceedings of the ConferenceonOceanTechnology (IERE, London,England,1981), pp. 37-47. •I'. G. Drake,R. L. Shreve, W. E.Dietrich, P. L Whiting, andL. B.Leopoid,"Bedloadtransportof finegravelobserved by motionpicturephotography,"J. Fluid Mech. 192, 193-217 (1988). cle velocities over the area of the intersection of acoustic 4R.L. Soulsby, A. P.Salkield, R. A. Haine,andB.Wainwright, "Observa- beamat thebedload,for our experimentsan areaofapprosil- tionsof the turbulentfluxesof suspended sandnearthe sea-bed,"Euromech,192,Munich, Germany(1985). mately50 cm2. •M. J.Crickmore, I.E. Shepherd, andP.M. Dore,"A fieldinstrument for andsizeof finesandsuspensions," Int. Conf. Resultsfrom theseexperiments indicatethat estimates measuringtheconcentration MeasuringTechniques of HydraulicPhenomena in Offshore,Coastal,and from singlepulsepairsare an accurateassessment of the InlandWaters,London,England(1986), pp.425-442. velocityasseenby the low deviationin successive estimates t,•. H. Flammer, "Ultrasonic measurements ofsuspended sediment," U.S. (Table I). Thus theneedto time averageiseliminatedallow- ingfor aninstantaneous estimateof thevelocityandtheability to observedynamiceventsoccurringat higherfrequencies such as those due to wave motion and turbulence. Geol. SurveyBull., 1141-A (1962). ?K.W. FerraraandV. R. Algazi,"A wideband spreadtargetmaximum likelihoodestimatorfor bloodvelocityestimation--partI: Theory,"IEEE Trans.Ultrason.Ferroelec.Freq. Control•8, 1-16 (1990). sO.Bonnefous andP. Pesque, "Timedomainformulation of a Doppler Our experiments haveonlyaddressed the performance ultrasoundandbloodvelocityestimationby crosscorrelation,"Ultrason. of the systemin a laboratoryenvironment.In considering Imag. 8, 75-85 (1986). •S.C[.Foster, P.M. Embree, andW. D. O'Brien,"Flowvelocity profilevia the ability of the deviceto measurebedloadtransportin a time-domaincorrelation:erroranalysisandcomputersimulation,"IEEE more realistic environment, such as the surf zone, several issuedneedto beaddressed. It isclearfromour experiments that if an incoherentlateral translationof the inter-pulse particletransportapproaches that of the wavelength of the carrier,thenthesignalsfrom that particlewill not correlate. Thus findingthe idealpulserate may not be trivial in complexfluidshearfields.In manybedloadregimesthetopology of thebottomisconstantlychanging.This affectsthe geometry surroundingthe acoustics, and hencethe return times and strengthof the signals.The systemwill thus haveto Trans.Ultrason.Fcrroelee.Freq.Control37, 164-175 (1990). lop.M. MorseandK. U. Ingard,Theoretical Acoustics (Princeton U.P., Princeton,NJ, 1996), Chap. 8.2. •H. L. VanTrees,Detection, Estimation, andModulationTheory, Part1 (Wiley, New York, 1968}. •P. M. Woodward, Probability andInformation Theory, With•4pplications to Radar (Pergamon,New York, 1953). 'SA.H. Quazi,"An overview onthetimedelayestimate in activeandpassivesystems for targetlocalization,"IEEE Trans.Acoust.SpeechSignal Process.29, 527-533 (1981). •4J.L. Christensen, "A newacoustic Dopplercurrentprofiler,"SeaTechnol. 24, 10-13 (1983). •ZR.Pinkel,"Observations on nonlinear motionin theopenseausinga adaptto suchchanges. Also,thepossibility of interference range-gatedDopplersonar,"]. Phys.Ocean.9, 675-680 ( 1979}. fromsuspended particles will havetobeaddressed including •R. Lhcrmitte andR. Serafin, "Pulsetopulsecoherent Dopplersonarsigits role in determiningan idealcarrierfrequency. Our ABVM wasdeveloped to demonstrate thefeasibility of themethodandthereforeconsisted of onlyonedirectionaldegreeformeasuring bedloadvelocities. For laboratory environments thiscanbeusedbyaligningthedeviceto the 1698 J. Acoust.Sec.Am.,Vol.92, No.3, September1992 nal proceasing techniques," J. Atmos.OceanicTech. 1, 293-308 (1984). '7R.Pinkel,"OntheuseofDoppler sonarforinternal wavemeasurement," Deep-SeaRes. 28A, 269-289 ( 1981). 'aB.H. Brumley,K. G. Cabrera,K. L. Denies,andE. A. Terray,"Performanceof a broad-bandacousticDopplercurrentprofiler,"IEEE J. Ocean Eng. 16, 402-407 (1991). D.W. SuttonandJ. S. Jaffe:Bedloadvelocityestimates 1698