Anda di halaman 1dari 11

On estimating the amplitude of harmonic vibration from the Doppler spectrum of reflected signals

Sung-Rung Huang
DeartmentElectrical of Engineering, University Rochester, of Rochester, York1427 New
Robert M. Lerner

DeartmentRadiology, of University Rochester, of Rochester, York1442 New


Kevin J. Parker

Rochester Center Biomedical for Ultrasound, University ofRochester, Rochester, York14627 New

(Received8 May 1990;accepted publication August 1990) for 6

The Dopplerspectrum echoes of from a sinusoidally vibratingscatterer discrete has spectral linesweighted Bessel by functions the firstkind. Because signaland spectrum of the are complicated functions the vibration of amplitude, numberof different a approaches been have tried in the pastto estimate vibrationamplitude, the givena received signal.Here, a new and simple relationship between spread variance)of the Dopplerspectrum the the (or and vibrationamplitude derived.A methodof estimating vibrationamplitudeis proposed is the based thisrelationanda noise on compensation procedure alsodemonstrated. is The performance the estimators studied of is throughsimulations. High accuracy predicted is underpropersampling conditions evenwhenthe signal-to-noise is poor.Slightdeviations ratio from single-frequency oscillation, would be caused nonlinearor nonidealmediumor as by
sourceeffects,are found to have little contribution to the total estimation error.
PACS numbers: 43.60.Gk, 43.20.Fn, 43.30.Es

INTRODUCTION

phones, and determined vibrationamplitude fitting the by the theoretical spectrum with an unweighted least-squares

The general problemof Dopplershifts from objects with time-varying velocityin an inhomogeneous layeredmedior um isquitecomplex.It is still a subject controversy of involv-

approximation. Lerneret al.7 in their newtechnique for


medical imaging elastic of properties tissue of called"sonoelasticityimaging,"havesuggested estimation vibrathe of tion amplitudeby calculating ratio of the two largest the harmonics. Jarzynski al.8 undertake similarestimation et a for precision measurement the soundfieldswith laser of Dopplerby comparing ratioof carrierandfundamental the
harmonics. Similar estimation had also been made by Cox

inglinearandnonlinear derivations. However, 1,2 whenthe


scattering objectis vibratingslowlysoasto produce w. a avelength much larger than the geometrical dimensions the of scattereritself,the Doppler spectrum the signals of returning from sinusoidally oscillating structures similarto that is

of a pure-tone frequency modulation (FM) process. 3 This


spectrumis a Fourier series with spectrallineslying above and belowthe carrierfrequency. The spacing between spectral harmonics equalto the vibrationfrequency, is and the amplitudes harmonics givenby different of are orders Besof

andRogersto study vibrational 9 the motion auditory of organs fish.Observing the ratiosof the adjacent in that Bessel

coefficients increase monotonically, Yamakoshi al.1 et


cameup with an estimator comparing relative by the magnitude of the adjacentharmonics their studyof tissuemoin tion. They alsoderivethe vibrationphasefrom the fundamental spectral components twoquadrature of channels. All of these techniques bebroadlyclassified thesame can in category,or approach estimation the vibrational to of parametersusingsomeratio of amplitudes. One of the disadvantages theratiomethods thattheyrequire of is eitherintensive computation large look-uptablesof theoreticalBessel or functions comparison for with the measured data. Besides, ratiomethods work well onlywhenthe argument the Besof selfunctionis small,which poses severe a limitation on the rangeof estimation. Furthermore,in practice,the performanceof the ratio methods highlydegraded is since almost all Doppler spectra sufferfrom poor signal-to-noise ratio. Finally, a sophisticated algorithmis requiredto determine the bestselection the harmonicpair to be compared. of
@ 1990Acoustical Society America of 2702

selfunctions the firstkind. A number applications of 3 of in


acoustics, optics,and radio haveled to research extracton ing the vibrationparameters from a measured Dopplerspectrum. Amplitude, phase,and frequencyof the oscillating structureare the mosttypical parameters be estimated. to Many techniques have beenproposed estimatethe to

vibrational parameters. Holenetal.4measured vibration the


frequencyof oscillatingheart valvesby looking visuallyat the spacing betweenharmonicsin the ultrasoundDoppler spectrogram. vibrationamplitudeisestimated countThe by ing the number of significantharmonicsunder a certain threshold.This procedure relativelycoarse is but is related to the observation FM that the bandwidthis roughlyproin portional to the modulationparameter,or amplitudeof os-

cillation. Taylor studied lasercalibration micro3 5'6 the of


2702 J. Acoust. Soc.Am.88 (6), December 1990

0001-4966/90/122702-11500.80

Downloaded 17 Oct 2012 to 152.14.136.96. Redistribution subject to ASA license or copyright; see http://asadl.org/terms

whereAois the wavelength associated with the wave of frequencycoo propagationspeed and Co. Thus, giventhe Doppler spectrumas described above, the estimation thevibrationamplitude equivalent the of is to estimationof the Besselargument/3. The exact spectral I. THEORY shape the Dopplersignal complicated dependent of is and on A. Derivation of the Doppler spectrum the parameter/3. Examplesof Doppler spectrafrom low, of/3 andthe two quadrature comSince FM spectrum well known,we brieflysum- medium,andhighvalues the is marize results in this section and introduce our notation. ponentsof the corresponding Doppler signalare given in Given a measured spectrum,it When a movingobjectis illuminatedwith an incidentlaser, Figs. 1 and 2, respectively. canbeseen that the backward estimation is not straightof/3 radio,or acoustic wave,the detected backscattered signals case.Experimental time-frefromthat moving object demonstratefrequency will a shift forward,evenin the noise-free quencydisplaydata from an ultrasoundB-scaninstrument known theDoppler as shift.If thescatterer oscillating is with are thevibration velocity muchslower thanthewavespeed and with Doppler capabilities givenin Fig. 3. The vibration frequency fixed, while the vibration amplitude,which is is the vibration frequency muchless thanthe cartier (incident wave) frequency,the spectrumof the detectedscattered proportional to the parameter/3, is increasedfrom left to More sidebands aliasing plus showup wavewill besimilarto that of a pure-tone process FM since right, thendecreased. whenthe parameter/3 increases. shownin Fig. 2, the two As the instantaneous frequencyof the scattered waveshas a of Dopplershiftproportional thevibration to velocity. Assume quadraturecomponents the Doppler signalcorresponding to thosein Fig. 1 are alsocomplex.Thus a time domain that the transmitted incident or signal is estimationapproachis not obvious. The estimationis even st (t) = cos(coot), ( 1) more difficultwhen the backgroundnoiseis mixed with the andthe scatterers vibratingwith the form are signal,as shownin Fig. 4.

Therefore,this work presents simpleand noise-immune a algorithmfor vibration estimation. The problemof vibrationamplitudeestimation apis proached throughthe measurement the spectral of spread (or variance)of the Dopplerspectrum. The proposed estimationtechniques be implemented can without difficulties. For instance, clinicalapplications Dopplerultrasound, in of onecouldobtainthe necessary parameters with slightmodifications existinginstruments. of Significant improvement on estimationaccuracycan be further achievedwith a noise correction algorithm.The theoreticalderivations the estiof mationandresults simulation shown the following of are in section. The effects noise,sampling, of and nonlinearityon the estimator performanceare also demonstratedsubsequently.

Usingtrigonometric identities, (7) canbe replaced Eq. by theseries 3

s,(t) =,4

J. (/Y)cos[coot + n(coLt q) (8) + ],

where the modulationindex or the argumentof the Bessel functions/3 directlyrelatedto the vibrationamplitudeof is the velocityor displacement field as follows:
/3=

Aco.,
coL

2v.,coo 0 cos
coL C0

2'.,coo 0 cos
C0

= 4rr cos 0, /0
(9)

g(t) = 'msin(co,.t q), or

(2)

v(t) =(t) = v,,cos(co,.t rq),

(3)

B. Vibration estimation from Doppler spectral spread

wheree(t) is the displacement the vibration,v(t) is the of

velocity thevibration, is thevibration of co,. frequency, is q the vibrationphase,g,, is the vibrationamplitudeof the displacement and v, = co,.e,, the vibration field, is amplitude of the velocityfield. The instantaneous frequencyof the receivedor scattered waves will be shifted to

Two Dopplerspectral parameters, spectral variance(or spectral spread)and meanDoppler frequency, usually are
defined as:

0'2o, =
and

(co )2S(co)dco --

S(co)dco

(10)

coo Acoa, +

(4)

coS(co)dco

S(co)dco

( 11)

AcoaAco,,cos t + v = (col ),
Aw,, = 2v,,Wo 0/Co, cos

(5)
(6)

where istheDoppler c% spectral spread isthevariance (a2o


or secondmoment), is the mean frequencyshift of the Doppler spectrum (the first moment), and S(co) is the Doppler power spectrum downshiftedto baseband.Note that the mean Doppler frequency shift is not necessarily zero sincethe Doppler signalis generallycomplexand the Doppler spectrum S(co) can be asymmetric. If the scattereris vibrating, the Doppler power spectrum that canbederivedfrom Eq. (8) in theprevious section
is

wherecotis the instantaneous frequency the scattered of waves, a istheDopplershift,Co thepropagation Aco is speed of illuminating waveat frequency and 0 is the angle coo, between wavepropagation the and the vibrationvectors. Therefbre, received scattered the or waves be writcan
ten as

s,(t) =A cos[coo (Aco,,/coL)sin(cot ] t+ + qv)

(7)

since instantaneous the frequency by definition, is, givenby the time derivativeof the argumentof the carrier cosine
wave.

S(co) 2r -n---

J (/3)6(co nco -),

(12)
2703

2703

J. Acoust. Soc. Am., Vol. 88, No. 6, December 1990

Huang eta/.' Amplitudeof harmonicvibration

Downloaded 17 Oct 2012 to 152.14.136.96. Redistribution subject to ASA license or copyright; see http://asadl.org/terms

#=3

#=o

o o

'

'

'

'

'

'

'

'

'

'

0 (a)

16

32

4.8

64

16

32

48

64

Normalized Frequency

Id

Normalized Frequency

p=3

p=30

16

32

48

64

16

32

48

64-

Normalized Frequency

(e)

Normalized Frequency

/ = 30

16

32

48

64

16

48

64

Normalized Frequency

()

Normalized Frequency

FIG. 1.Examples noise-free of Doppler spectra low/5' 3: (a) linearand(b) logscale; for = medium/5' 10:(c) linearand (d) logscale; = high/5' 30: (e) = linearand (f) logscale. Normalized fold-over frequency = 64, normalized flora segment lengthTFET= 4, for all cases.
2704 J. Acoust.$oc. Am., Vol. 88, No. 6, December 1990 Huang eta/.' Amplitudeof harmonicvibration 2704

Downloaded 17 Oct 2012 to 152.14.136.96. Redistribution subject to ASA license or copyright; see http://asadl.org/terms

-,-, c-

fi= 10

0.0

0.2

0.4

0.6

0.8

1.0

0.0

0.2

0.4

0.6

0.8

1.0

Normolized Time

(d)

Normalized Time

((-

/= 30

E
o
(Y) o

T 0.0
lb)

0.2

0.4

0.6

0.8

1.0

0.0

0.2

0.4

0.6

0.8

1.0

Norm01ized Time

()

Normalized Time

/= 10

- c-

/=30

08

o.o
t)

0.2

o. o. o. NormalizedTime

to

If)

o.o

0.2

o. NormalizedTime

FIG. 2. Examples twoquadrature of components noise-free of Doppler signals low/3= 3: (a) in-phase (b) quadrature-phase for and component; medium /3= 10:(c) in-phase (d) quadrature-phase and component; high/3-- 30:(e) in-phase (f) quadrature-phase and component. thefunction periodic, Since is
only onecycleis shownin the graphs.

2705

J. Acoust. Soc.Am.,Vol.88, No.6, December 1990

Huang eta/.' Amplitude harmonic of vibration

2705

Downloaded 17 Oct 2012 to 152.14.136.96. Redistribution subject to ASA license or copyright; see http://asadl.org/terms

(8)

fi=10

0.0

16.0

32.0

48.0

64.0

Norrnolized Frequency

(b)
c:)

ll= 10

FIG. 3. Experimental observation Doppler spectrum of from vibrating structures underclinical B-scanultrasound. The lower portion is a time ( horizontal -frequency ) (vertical) display data from the regionselected of by the cursorasshownon the upperportion.Vibrationfrequency held is fixed,whilevibration amplitude increased thendecreased is and fromleftto right.

wherethe powerspectrum beendownshifted zero frehas to quency,asby quadraturedetection. For thisparticularBessel spectrum, meanfrequency the is zerosince J_, (/3) = ( -- 1) "J, (/3) and the powerspectrum is thereforesymmetricaboutzerofrequency. Thus, the spectralspreadcanbe calculatedfrom the zeroth and second momentsof the spectrumdefinedas

0.0

115.0

32.0
Normolized Frequency

4.0

64.0

mo=
m2 = Z

O2,,(/3), r
(no ) J (/3), L

(13)
(14)

FIG. 4. Exampleof noisyDoppler spectrum with/g = 10, signal-to-noise ratio SNR = 20 dB, normalizedfold-overfrequency ftoa = 64, and normalizedsegment lengthTFF --4, (a) linearscale,(b) log scale. T

wheremk is the k th momentof the Doppler spectrum. However, the zeroth moment is the total energyof the signaland is equalto unity. One can easilyshowthis using the followingmathematicalidentity:

Integratingthe aboveequationover one period (0 = 0 to 2rr), and usingorthogonality exponential of functions for
m//,

eia 0= sin
term, we have

Jn ei"O. ()

( 15 )

1--

j2.(/3)__mo,

(17)

Squaring above the equation replacing with -- 0 in one and 0


1 eily Oeil ( - o) sin sin

a result that has been noted in the literature. The secondmoment can be derived in the sameway by

takingthe firstderivative the Eq. (15) with respect 0, of to

Z
m= --c n=

Z
--oo

Jm ()Jn() ei(mn)O. -

(16)

/ COSi/3sin Z Oe 0__
n

n ()einO. J,

(18)
2706

2706

J. Acoust.Soc. Am., Vol. 88, No. 6, December 1990

Huang ot a/.' Amplitudeof harmonicvibration

Downloaded 17 Oct 2012 to 152.14.136.96. Redistribution subject to ASA license or copyright; see http://asadl.org/terms

Squaring from the above equation before, as

2COS sin sin- o) 20eifi Oeifi (

j 2,, - nr)+Ncoco (,8)6(0 ( ))d co


(19)

=
m

--oo n

mnJ, (/)e (/)J, i('-

J,(l)6(co ncoL N(CO) -)+ do


(25)

Then, integratingover oneperiodagain,


--

= (mzs + m2.v)/(mo,s mo,v) +

2 --

= [oo,s (1/SNR)o,,v]/[1 (1/SNR)], + +

nJ(/5')=m2.

(20)

where is the meanDoppler shift of the noisysignalgiven


by

In general, moments the Bessel all of spectrum be can calculated from derivatives Eq. (15) by properdifferentiof ation,squaring, integration. and Low-order moments the of Bessel spectrum givenin Table I. are Thesemomentsof the Bessel spectrum can alsobe calculatedfrom the generating functionof Bessel function:

=
X

co

J (/5')6(0 -- ncot. + N(co) dco )


j2 (/5')6(0- not.) + N(co) do
(26)

/(,Z)

zngn()=e (B/2)[z-l/z)l

(21)

ink, andink,v thek th moment signal noise s are of and about thecorresponding frequencies mean respectively, isthe rro,s Doppler spectralspreadof vibration only, rro,v the is
Dopplerspectral spread noiseonly,andSNR isthe signalof to-noiseratio givenby

which can be found in a standard handbook on bessel func-

tions. By takingthe 1stthrough k th derivative the 11 the of


generating functionwith respect z and substituting = 1 to z into the resultingexpressions, momentsof the Bessel all spectrum, functions theparameter/5', thenbecalcuas of can lated from lower-ordermoments squaring by and simplealgebraicmanipulation. From thispointof view,the Bessel spectrum actually is a one-parameter function.Therefore,the second moment serves a goodestimator the spectrum. as of One can estimatethe vibrationfrom the Doppler spectral spreadas

SNR =
X

j 2 (,8)6 (co-- not-) do


N(co)dco

=mo,s/mo,v 1/mo, = s.

(27)

As long asthe noiseis stationary,the momentsof noise power spectrum can be estimated when the vibrationis removed or halted. Once the noise moments have been esti-

oao_, = (m -- m )/mo = [m -- () ]/mo, 2

(22)

mated, the noise-free vibrational Doppler spectralspread can then be estimated from the noisysignalas

and for this caseof the Doppler spectrum, = O,thus

cr2o,,s [ 1 + (1/SNR)] -- (1/SNR)a2o,,v (28) = cr2, .


In someapplications, vibrationis inherentand canthe not be controlled externally.The noisecompensation, this in case,can be doneby estimatingthe signal-to-noise ratio as well as the Doppler spectralspreadof the noisefrom the finite bandwidthwhite noiseassumption follows: as

or

/5'= x/(rro/cot).

(24)

This indicatesthat the amplitudeparameter/5'can be estimatedfrom the standarddeviationof the power spectrum. This straightforward resulthas,apparentlynot been previously derivedfor the caseof FM broadcast Doppler or spectrum from vibratingobjects.

mo,v =
--B

Nodo = 2NOB,

(29)

in2,N :

co2N = odo '"2NoB3 3 B '

(30)

o;o,,v(2NOB = 3/3)/2NOB B 2/3, =


B is the one-sided bandwidth of the white noise.

(31)

whereNoisthe powerspectral density thewhitenoise, of and

C. Noise correction algorithm

In practicalsituations, noisepresents problemsin parameterestimation. The Dopplersignals tendto be 30-50 dB lower than the carrier in many applications, therefore,the signal-to-noise ratio for Doppler signalis usuallypoor. Additive, stationary,and uncorrelatednoisecan be removed from the Dopplerspectral spread vibrationestimator. staIf tionary uncorrelatednoisewith power spectrumN(co)is added in the received backscatteredsignal, the noisy Doppler spectralspreadrro can be written as
2707 J. Acoust.Soc. Am., Vol. 88, No. 6, December1990

TABLE I. Low-ordermomentsof the Bessel spectrum.


All odd moments 0th moment 0 1

2ndmoment 4thmoment 6thmoment

/2 /2__384 /2_4 __66

Huangeta/.' Amplitude harmonicvibration of

2707

Downloaded 17 Oct 2012 to 152.14.136.96. Redistribution subject to ASA license or copyright; see http://asadl.org/terms

Even when the noiseis not white, the noisecompensation is still possible longasnoise as powerandnoisespectral spread can be estimated priori by statistical a techniques.
II. RESULTS AND DISCUSSIONS

A. Effects

of noise

SimulatedDoppler spectraare obtainedby taking fast Fourier transform (FFT) of finite segments that are composed the two quadrature of components Eq. (7). Figures of
5-9 are families of the estimation errors

[ (/-/)//] X 100% asfunctions variousparameters of in the estimationprocess, where/ is the estimated vibration parameterand/ is the true parameter.

The estimation errorsare plottedasfunctions signalof to-noiseratio from 0-40 dB in Fig. 5 (a). White Gaussian noise was added independently into the two quadrature components. normalized The sampling frequency (defined assampling frequency divided thevibration by frequency) is 128. Or equivalently, the normalizedfold-overfrequency (definedas fold-overfrequency,or aliasingfrequency,or half-sampling frequency divided thevibration by frequency) is 64. A high sampling rate wasusedto reduce aliasing the

Noise-Dee

og
k'v
o

;,, .--.-/:o \
"% ""--... .................
2;10 0
Normalized Fold Over Frequency

0.0

1.0

20.0

30.0

40.0
SNR = 40 dB

(a)

Signal-to-Noise Ratio

o /=1
o
,
o.

#=5

__/=

'
4-

/=o
20

o.o

'

0.0

'

20.0

30.0

40.0

2,1
lb)

(b)

Signal-to-Noise Ratio (dB)

Normalized FoldOver Frequency

FIG. 5. Plot of estimation errorsin percentage against signal-to-noise ratio from 0-40 dB at normalizedfold-overfrequencYfro.d 64 and normalized = segment length TFrr -- 4, (a) withoutnoise correction(theoretical prediction and resultsfrom one simulation), (b) with noisecorrection(average overfivesimulations).Note theexpanded scale (b) showing in drasticerror reductionachieved the noisereductionprocedure. by

FIG. 6. Plot of estimationerrorswithout noisecorrectionagainstnormalizedfold-over frequency with TFrr = 4, (a) noise-free, signal-to-noise (b) ratio SNR = 40 dB. As samplingfrequencyincreases, underestimations originated from aliasing decrease all cases. (b), noisespectral in In spread causes large errors at high samplingratessincewidebandwhite noiseis employed simulations. in

2708

J. Acoust.Sac. Am., Vol. 88, No. 6, December 1990

Huang eta/.' Amplitudeof harmonicvibration

2708

Downloaded 17 Oct 2012 to 152.14.136.96. Redistribution subject to ASA license or copyright; see http://asadl.org/terms

.............. /5'=5 ....... /5'=10 .-/ = :o --- -/ = 50

ageestimation errorsover five simulations. The estimation errors are swingingback and forth around zero as the parameter/3 changes. This indicatesthat the estimatoris unbiased.The maximum estimationerror for poor signal-tonoise ratio at 0 dB is still within 4%. The estimation error of

high/3 is muchlessthan that of low/3, asexplained earlier.

B. Effects of sampling

The aliasingeffectcan be seriouswhen the fold-over frequency sampling or frequency not proper.Figure 6 (a) is
shows the noise-free estimation error at normalized fold-

o.

0.0

1.0

' 2.0 ' 3.0 ' 4.0 ' 5.0 ' 6.0 ' 7.0 ' 8.0 ' 9.0 ' ' ' ' ' ' ' ' ' Normolized Segment Length

10.0

over frequency expressed a powerof 2. The effectof the as impropersegmentation minimalheresince normalized is the total lengthisfour, aswill beexplained later. The estimation errorsblow up at the point when the parameter/3is larger thana threshold. sampling This criterion beexpressed can as
X ---2ffold /2timaxiL, (32)

FIG. 7. Plot of noise-freeestimationerrors againstnormalized segment length TFrr -- 4. The finite-length effectproduces overestimation. When thesegment lengthisanintegral multiple vibration of period, estimation the error dropsto a minimum.

wheref is the sampling frequency, froldis the fold-overfrequency, fL is the vibrationfrequency Hz. and in It is interesting notethat the aliasing to alwaysresults in underestimation. The reasonis that harmonicshigher than fold-overfrequencyare foldedback into lower frequency

contents the co or n2 term associated the higher and 2 with


error resulting from the finiterepresentation the infinite of spectrum. The normalizedsegment length (definedas segmentlength divided thevibration by period)is4. Two sets of curvesare shownin Fig. 5 (a). The rapidly fluctuatingone comes from a singlesimulation, whilethe smooth oneis the theoreticalpredictionof the deviationof the estimation withoutanynoise correction using Eq. (28). The simulation results agreewith the theoretical prediction. Theseresults show that thesignal-to-noise mustbehigherthanabout ratio 30 dB to achieveacceptable accuracy estimation of unless noisecorrection procedure performed. is From Eq. (25), the performance the vibrationestiof mator without noisecorrectionwill be degradedwhen the signal-to-noise ratio is low or the spectral spread noise of only is largecompared that of vibrationonly. When the to parameter/3 small,the spectral is spread the Bessel of spectrum is narrow sincethe bandwidth (or spectralspread) is proportional theparameter/3from Eq. (23). In thiscase to of low/3, the spectral spread noiseis comparable even of or largerthan that of vibrationonly, thusthe performance is highlydegraded the additionof the noise. by When the parameter/3is large as in wideband FM, the performance of the estimator fairly goodaslong as/3is still smallerthan is normalized fold-overfrequency. The overestimation due is
to the subtractionin Eq. (28).

harmonics calculating second in the momentis smallerthan what it shouldbe in the theoreticalexpressions (10), (14),
and (23).

Figure6 (b) isthesame simulation with 40 dB signal-tonoise ratio. Interestingly,as the fold-over frequencyincreases, the estimation errors increase as a second-order

Figure 5(b) is a plot of the estimation error with the noisecorrectionalgorithm.The vibrationwas removedto estimatethe momentsof the noise-onlyDoppler spectrum, andthentheparameter/3 estimated is from the noisysignal byEq. (28). Sampling conditions thesame thatin Fig. are as 5 (a). Note the expanded verticalscale Fig. 5 (b). Obviousis ly, dramaticerror reductionis achieved throughthe noise correction procedure. The results Fig. 5(b) are the averof
2709 J. Acoust. Soc. Am., Vol. 88, No. 6, December 1990

polynomial.This resultsfrom the property of the white Gaussian noiseusedin simulation.Sincethe spectralspread of a uniformspectrum proportional the second is to powerof the bandwidth,asshownearlierin Eq. (31 ), the estimation error increases approximately a second-order as polynomial in normalizedfold-overfrequency whenno filteringprocess is involved.Therefore,filtering shouldbe taken if the sampling frequency increased avoid the aliasingerror. In is to practice,filterscan be designed optimizethe estimation to according the actualspectral to moments signal of andnoise. Figure7 is the plot of noise-free estimation error against normalizedtotal sampling segment length Tvrr. To exclude the effect aliasing of error,the normalized fold-overfrequency is setto 64. The estimation error is minimal whenever the normalizedtotal sampling lengthis equalto an integer.The largeerror of nonintegral normalizedtotal sampling length is due to the sharpdiscontinuity the time domainsignal of from the inherentperiodicity whenFFT analysis exploited is without windowing.This indicatesthat windowingor synchronizationis requiredin practicalanalysis. Note that the finite samplinglength alwayscauses overestimation. This happens expected as sincethe sharpdiscontinuity time in domaingenerates significant sidebands that can subsequently increasethe spectralspread.
C. Effects of nonlinearity

If the vibration is not perfectlysinusoidal due to some medium or vibration sourcenonlinearity, the Doppler specHuang otaL' Amplitudeof harmonicvibration 2709

Downloaded 17 Oct 2012 to 152.14.136.96. Redistribution subject to ASA license or copyright; see http://asadl.org/terms

/ = 10, Nonlinearity = 2 %

= 10, Nonlinearity = 10%


(c)

.o
Normalized Frequency

.o

4.0

0.0

16.0

.o
Normalized Frequency

.o

= 10, Nonlineariy = 2 %

= 10, Nonlinearity = 10 %

(d)

I' 0.0 .......

16.0

.0
NormalizedFrequency

,.0

6,.0

0.0

,.o

.o Normolized Frequency

.o

FIG. 8. Examplesof noise-free Doppler spectrum 2% and 10% nonlinearitywith fundamental for vibrationamplitude/3= 10, normalizedfold-over frequencyffoa 64 andnormalized = segment lengthTFvr = 4. Note that thepeaks thespectra of shiftfrom thecase no nonlinearity of (Fig. 1). The relative ratiosof harmonics alsovary asa functionofnonlinearity,vibrationamplitude,and fold-overfrequency: 2% nonlinearity, (a) linearscale,(b) 2% nonlinearity, log scale,(c) 10% nonlinearity, linearscale,(d) 10% nonlinearity, scale. log

tral shapewill deviatefrom the one parameterBessel spectrum. Assumingthe vibrationis periodicand can therefore be represented a Fourier series, by both the ratio of harmonics and the spectralspreadwill be differentfrom that of a pure sinusoidal vibration. Assumingthat only the fundamental and secondharmonicsare significantin the vibration, the returnedsignalnow canbe written as

tal and second harmonicsof the vibration, respectively.

An expression the Fourier series for expansion the of abovesignalcanbe obtained from the analysis multitone of

FM 3withslight modification follows: as

$r(t) =

J ([3)J. ([32)
(34)

S (t) = A COS -t-/ sin Lt -t- 1 r [ COot (CO )


-t-/2sin(2COL %)], t -t(33)
where/3and/32are the modulationindices the fundamenof
2710

Xcos[COo m(COt qo) n(2COt q02) t+ + + + ].


The nonlinearity is definedas
Huang eta/.' Amplitudeof harmonicvibration 2710

J. Acoust. Soc. Am., Vol. 88, No. 6, December 1990

Downloaded 17 Oct 2012 to 152.14.136.96. Redistribution subject to ASA license or copyright; see http://asadl.org/terms

error comesfrom the nonlinearity.If the nonlinearityis removed, estimator without noise correction gives accurate estimationwithin 0.01%, while the accuracyis increasedto
0.0001% after noise correction.

If wideband whitenoise added no filteringprocess is but is used (same parameters,exceptffold = 64 fL ), then the
estimation error for estimator without noise correction rises

to 30.4%, while the error remains as small as 1.75% after

5.0

10.0

Nonlineority (%)

FIG. 9. Plot of noise-free estimationerrorsagainstnonlinearity.The errors are almost the samefor all valuesof/' from 0.1-50. The errors due to nonlinearity are lessthan 2% when nonlinearityis lessthan 10%.

noisecorrection.The large error is due to the large wideband-noise spectralspreadas predictedin Eq. (31 ). This suggests either noisecorrectionor filteringis necessary that for noise reduction when the samplingfrequencyis increased reducethe aliasing to error in practicalimplementation. If filtering is applied, the effectson signal spectral spreadmust be taken into accountduring the filter-design phase,sincehigherharmonics the signals removedas of are well. It shouldbe notedthat the performance the estimaof tor with noise correction beimprovedby increasing can sampling frequency without filtering.
III. CONCLUSION

Nonlinearity = N2 (/3://3) X 100%.

(35)

Figure 8 a-d showsthe spectra 2% and 10% nonlinof earityfor the caseof/3 = 10.The spectral shapes peaks and are differentfrom that of pure tonevibration (Fig. 1). Surprisingly, this doesnot affect the Doppler spectralspread much.Estimationfor noise-free, smallaliasing, andproperly sampled signalhasbeenperformedas shownin Fig. 9. The resultsshowthat nonlinearitylessthan 10% contributesless than 2% estimationerror. In comparison, conventional all estimators that utilizeamplituderatioswouldbedramatically affected 10% nonlinearity.This can be appreciated by by comparing the patterns of peak amplitudes in Fig. 1(c) againstthosein Fig. 8 (c).
D. Combined effects

We haveanalyzedthe signalreflected from a sinusoidally vibrating object. An estimator of vibration amplitude based on the derived relation between Doppler spectral spreadand vibrationamplitudehasbeenproposed. noise A correctionalgorithm is also proposed, which can improve the estimation accuracy dramatically. Simulations show good results within 4% error given signal-to-noise ratios rangingfrom 0-40 dB. Adequate sampling frequency, givas en in the text, must be satisfied accordingto the expected maximum vibration amplitude. Proper filtering can reduce the spectralspreadof noiseand then reducethe estimation error. Windowingor synchronization requiredin practical is implementationto reducethe finite length effect.The proposedestimatorsurvives throughvibrationnonlinearityless
than 10%, whereas other known estimators that make use of

A case that is typicalin "sonoelasticity imaging, "7'12


with the followingparameters, was performedto showthe
combined error:

Co 1.5X 10 cm/s, =
fo = Wo/2rr= 7.5 MHz (i.e., ,go= 0.2 mm),
0 = 100,

f = co/2r = 200 Hz,

/3 = 10 (i.e., m = 0.16 mm, Vm-- 20.3 cm/s),


SNR = 20 dB,

the harmonicratio of amplitudes not tolerateslightnondo linearities.Overall, the proposed estimatoris robustin the presence noisecomparedto earlier estimators. of The estimation can be obtainedvia simplecalculations from the parametersin someexistingDoppler instruments,especially thoseavailablein clinical Doppler ultrasound.To displaya realtime vibration image, existing faster time domain Doppler spectralspreadestimators(e.g., Refs. 13, 14) can be appliedwith modifications (e.g., changingscanning patterns and samplingfrequencyto reduce,theeffectof vibration phase)and/or synchronization (with low-frequency vibration) to display the vibration amplitude. The results shouldbe relevant to echocardiology, sonoelasticity imaging, lasercalibrationof soundfieldsand vibration,and other radio, radar, and sonarapplications.

fold= 12.8fL = 2560 Hz (i.e., fs = 5120 Hz), TFFT-- 10 TL = 50 ms (i.e., LFrr ---256) l,
N2= 10%,
ACKNOWLEDGMENTS

where Tr is the segment lengthfor FFT analysis ms, in and L r is the numberof samples Tr. The estimation in
error without noise correction is 3.39%, while the estimation is reduced to 1.97% after noise removal. Most of the
2711 J. Acoust. Soc. Am., Vol. 88, No. 6, December 1990

The support encouragement Dr. R. Gramiakand and of Dr. J. Holen are gratefullyacknowledged. This work was supported the Departmentof ElectricalEngineering, by
University of Rochester.
Huang eta/.' Amplitudeof harmonicvibration 2711

Downloaded 17 Oct 2012 to 152.14.136.96. Redistribution subject to ASA license or copyright; see http://asadl.org/terms

D. Censor, "Acoustical Doppler effect analysis--Is a validmethod?," it J.


Acoust. Soc. Am. 83, 1223-1230 (1988).

IX, Proceeding SPIE Conference Optics, of on Electro-Optics, Sensors, and SPIE 925, 250-254 (4-6 Apr. 1988).

2j. C. Piquette, L. Van Buren, P. H. Rogers, A. and "Censor's acoustical


Doppler effectanalysis!s it a valid method?," Acoust.Soc.Am. 83, J.
1681-1682 (1988).

9M.CoxandP. H. Rogers, "Automated noninvasive motion measurement


of auditoryorgans fishusing in ultrasound," Vib. Acoust. Rel. Des. J. Str.
109, 55-59 (1987).

3A. B. Carlson,Communications Systems (McGraw-Hill, New York,


1986), Chap. 8, pp. 221-227.

Joy. Yamakoshi, J.Sato, T. Sato, and "Ultrasonic imaging internal of vibration insideof soft tissueunder forcedvibration," IEEE Trans. Ultrason. Ferroelec. Freq. ControlUFFC-37, 45-53 (1990).

4j. Holen, C. Waag, R. Gramiak, R. and "Representations ofrapidly oscillatingstructures theDopplerdisplay,"UltrasoundMed. Biol. 11,267on
272 (1985).

G. N. Watson, Treatise theTheory Bessel on of Functions (MacMillan,


New York, 1945), Chap.2, pp. 14-15, 31.

5K. J. Taylor,"Absolute measurement acoustic of particle velocity," J.


Acoust. Soc. Am. 59, 691-694 (1976).

2R.M. Lerner, J. Parker, Holen, Gramiak, R. C. Waag, K. J. R. and


"Sono-elasticity: Medicalelasticity images derived fromultrasound signalsin mechanically vibrated targets," Acous. Imag.16,317-327(1988).

6K.J. Taylor,"Absolute calibration microphone a laser of by Doppler," J.


Acoust. Soc. Am. 70, 939-945 (1981).

7R.M. Lerner, R. Huang, K. J. Parker, S. and "Sonoelasticity images derived from ultrasoundsignalsin mechanicallyvibrated tissues,"Ultrasound Med. Biol. 16, 231-239 (1990).

3C.Kasai,K. Namekawa, Koyano, R. Omoto, A. and "Real-time twodimensional bloodflow imaging usingan autocorrelation technique,"
IEEE Trans. SonicsUltrason. SU-32, 458-463 (1985).

4K. Kristoffersen, "Time-domain estimationthecenter of frequency and


spread Dopplerspectra diagnostic of in ultrasound," IEEE Trans.Ultrason.Ferroelec. Freq. ControlUFFC-35, 484-497 (1988).

8j. Jarzynski, Lee,J.Vignola, H. Berthelot, A.D. Pierce, D. Y. and "Fiber


opticsDopplersystems remotesensing fluid flow," in OceanOptics for of

2712

J. Acoust. Soc.Am.,Vol.88, No. 6, December 1990

Huangotal.: Amplitude harmonic of vibration

2712

Downloaded 17 Oct 2012 to 152.14.136.96. Redistribution subject to ASA license or copyright; see http://asadl.org/terms

Anda mungkin juga menyukai