Escolar Documentos
Profissional Documentos
Cultura Documentos
RobertA. Dalrymple
Ocean Engineering Group, Department of CivilEngineering, Unioersity ofDelaware,NewarlL
Delaware 19716
Levsiri C. Munasinghe
Ocean Engineering Group, Department of CivilEngineering, University ofDelaware, Newark,
Delaware 19716
David H. Wood
James T. Kirby
Department of CivilEngineering, University ofDelaware, Neumrk, De/aware 19716
A parabolic model, valid forwide angles (out to90* from the assumed propagation direction
for a homogeneous environment),is presented. Numericalcomputations for the modelare donealmostentirelyin the Fourierdomain,and the modelcanbeshownto betheoretically exactfor a homogeneous ocean.In addition,the modelcan implicifiy handlerange-dependent
sound-speed profiles. An erroranalysis indicates that themodelis moreaccurate thanthe standard parabolic equation (SPE) andthemodified-wide-angle parabolic equation (MWAPE) for constant perturbations of the indexof refraction. The accuracy of the modelis examined by comparison of computed solutions with exactsolutions for rangeindependent cases. Several idealized range-dependent cases are alsoexamined.
PACS numbers: 43.30,Bp
INTRODUCTION
-where
ar
+ koZQ2p = 0,
(3)
Q: isan operator = n: + -- -k
environment results in
(4)
(r+ikoQ)(r-ikoQ=O. (5)
If we are concerned with waves propagating in the positive directionand neglect the influence of the waves ih the oppositedirection(the backscattered waves),thenour governing equationis, in parabolicform,
(1)
It is the representation of Q that determines how well the parabolic model functions at largepropagation angles.
ko is a characteristic wavenumber; n(r,z) = Co/C(r,z) = k(r,z)/ko = indexof refraction; Co is thesound speed associated with ko; C(r) is the soundspeed; (2) k(r,z) = to/C(r,z) is the wavenumber;
to is the angular frequency; is the rangedirection;
Q = x/1+ p + E,
where
(7)
= kaz
e = n2 -- 1.
0001-4966/90/101863-14500.80 1990Acoustical Society of America
(8)
(9)
1863
0p=0 onz=h.
(16)
(11)
(12)
o2p 1 0 ---+k2p=0' ar 2 r
(17)
Q= 41 +it + e = x/1
+ e/(1 +it),
where k(r,z) is the wave number governing the acoustical wavemotion.A representative wavenumber, ko(r), isintroducedvia the indexof refractionn and a parameter,v(r,z),
which is a measure of the deviation of the index of refraction
then we can approximate Q in a variety of ways. From a binomialexpansion, neglecting termsof order, andhigher, andneglecting termsinvolving products of and/, we have,
correct to order 6,
from unity:
=kon =k(l--).
Now, our equationis
(18)
Q.-x/l+ It + e/2.
(13)
Tappert suggests that thismodel is not an improvementover the standard parabolic equation. Alternatively,reversing the rolesof 6 andIt in (12), we can approximateQ as
O:'p +r 3r
3r
+ (k--kgd)p = 0. (9)
Finally,the reduc fore 0f the Helmholtzuation is obtainedby removingthe radial spreadingby assuming that
Q/I
+ e +it/(2/1 + e),
(14)
p(r,z) =p(r)/.
Substituting Eq. (20) into .
(20)
(19) andneglecting thete
asin Tappert andTappert andLee. s with 1/(4), i.e.,a far-field approximation, weobtain Thomson andChapman, 4 utilizing the methodology developed by FeitandFleck 6 in optics, introduced thefollowingrepresentation:
Or
(15)
definedsymbolicallyas
P(r,v,, ) = (P).
Explicitly, the Fourier transfore paim are
(22)
acoustics, Greene 7 for example, hasused a ( l, 1) Pad6approximant, whileKnightlyet alfi extend themethod to a
(2,2) approximant, with the numericaldisadvantage of introducing mixed fifth-order derivativesin the governing equation.
(23)
(24)
(25)
Heimholtz
y=(m+)(,/h),
I. THE WIDE-ANGLE MODEL
m = 0,1,2 .....m.
equation,. (21), transforms the equationto one solely dependent on the r erdinate;
In thispaper,the problem is Fouriertransformed in the cross-range directionprior to obtaining the parabolic equation form. This will permitan exactrepresentation of Q for isovelocity oceans, andveryaccurate results using step-wise integrationsfor each of the Fourier modes,which are then inverse transformed to find the sound-pressure profile.This
ly model surface water wave propagation over a uniform
(26)
To obtaina parabolic mel from thissecond-order equation, we split the uation, assuming in Fouder space that the pressure, )(r,y ) can be split into a forward-propagattechnique was used byDairymple andKirby to successful- ing wave and a backscattered wave, p(r,y)
= p + (r, y,, ) + p- (r,y), which satisfythe followingpair of equations:
slope andbyDalrymple etal.l tomodel water wave propagation over irregular bathymetry. A cylindrical coordinate systemwill be usedto describe
the domain,with r in the rangedirection,z pointingdownward and 0 asthe azimuthalangle.We will assume that the
(27)
azimuthal variation isveryslight andthatthedependence of pressurep on theazimuthalangle0 canbeneglected. At the surface, we will assume a pressure release boundary condition, p(t:,0) = 0, and, at the constantdepth bottom, a noflow boundarycondition,
1864 J. Acoust. Soc.Am.,VoL88, No.4, October 1990
(28)
whereG(r,y ) isunknown apriori. [If thelasttermin s. (27) and (28) werezero, i.e., for a homogeneous environment,thenthe parabolic splittingwouldbe exact. ] SubstiDalrymple otaL:Wide-angie acoustic model 1864
that tuting these equations into (26) and assuming [P+1>> IP- ] results in anexpression for G(r,y, ):
The factthat thearguments of theexponentials in Eqs. (36) and ( 38) arenotthesame leads to phase errors, such as are oftenobserved in modeling wavepropagation by using Eq. (27), whichhasonly first derivatives with respect to
range.
2o2-- /2rn
(29)
(30)
An estimate of the phase error inherentin our method can be determined and compared to similarerrorsin two othertypesof parabolic waveequations, the standard parabolicequation(SPE), and themodifiedwide-angle parabolic equation (MWAPE). The phaseerror in our method is
due to the use of
Equation(27) with either (29) or (30) serves asa parabolicmodel.If p(r,y,.) is defined suchthat
(39)
(40)
(31)
OeXp(ix0 --T,,, r)
___ _ t9( o2--,.)/Or
24ko _ 7m
Equation (32) is better suitedfor numericalsolutions
(41)
Similarly, the corresponding approximationused in the modified-wide-angle parabolic equation as given by St.
/,oe 2
r 2ko
8
(42)
It is possible to rewritekt
II. ERROR ESTIMATES
...., =4(-r)+
--
It is possible to estimate the accuracy of Eq. (32) asan approximate solution to Eq. (21), the Helmholtzequation, fora range-independent sound-speed profile. A perturbation s(z) fromn2(z) -- 1 of ordere canbedefined asfollows:
2x/(o _ rm)
+ h.o.t. (43)
8(,o - )4( g- )
andthefollowing errorestimates for a constt perturbation
mn then be made:
2 = __[n2(z) _ 1] = -- es(z).
(33)
To analyze the effect of perturbing the indexof refraction by a term of order e, consider the casewheres(z) = 1
and e is a smallconstant. Then, an exactsolutionto Eq. (32) can be formally statedas follows:
(34)
+h.o.t. )
P* = o sin(y,z)
Curves of 6 vssin- (v/k o) for s. {)-(46) are givenin Fig. 1 as a functionof the aolute vues of
(k...., -- k, )/k o, (k,, -- k, )/k o, and -- k )/ko, restively. Since /k o isthesine of the where p + is theacoustic pressure of the forward-propagatangleassociat with the mth mode,notethat, for angles
ing wave. (37)
associat with small valuesof m, the MWAPE can handle a
l(k,,,--k)/ko[
thanthe
p+ = bo sin(y.z) exp(ix/k -- r + koer) (38) error](k.... -- kmp , )/kol of theMWAPE islger than
1865 J. Acoust.Soc. Am., Vol. 88, No. 4, October 1990 Dairytopicota/.: Wide-angieacousticmodel 1865
", o
\
",.'"<Z.
.-/-. %- '
--/
\o.
scheme,the equationsfor a simple second-order RungeKutta scheme with regardto bare shownin the Appendix. The modelwastested by computation of sound~pressure transmission lossasa functionof rangein an idealizedocean environment. For comparison, reference solutions were re-
.'
1-:", K..'I',
-"\ .....
.....
0.007.
tionsby a fast-field programalgorithm(FFP) and a highfrequency approximation based on an exactfactorization of the Helmholtz equation for range-independent environments.
FIG. L Crvesofe vssin- ' (7,./ko) i detees asa function ofte absolmc
values of (k..., -- lq. )lk., (k. , -- k. )tk., and( k..... -- k.,..r )lko
for the WA, $PE, and WAPE models, respectively:--,WA - - -, $PE,
The testoceanenvironment, as givenin Table I, representsthree differentbilinear sound-speed profiles,casesA, B, and C, respectively. These idealizedoceanmodelswere
theerror [(ke., -- kv, )/ko[ for the VWA model for the
It is clearlyevidentfrom Eqs. (44)-(46) andFig. I that
the error of our method is much smaller than that for the
standard parabolic equation, whichhasan error evenwhen = 0 and is alsosmallerby an o() than the error incurred by the modified-wide-angle parabolicequation.However,in the latter case,the relative error magnitudes, as shownin Fig. 1, will depend on the sineof the propagating angleas well asthe magnitude of 6. Hence,our methodcanbecharacterized as accurate for all modes exceptpossibly one at zerothorderandcorrectto firstorderin (constant)perturbations of the index of refraction.
for testing thebeamwidth capabilities of solution algorithms for the Helmholtzequation. Cases A-C requirehalf-beamwidthcapabilities of 18 ,30,and40,respectively. The model depthandrangeincrements were10and20 m, respectively. A "false bottom" from 1500 m down to 2560 m with an exponential attenuationprofilewasusedto minimizereflections returned to the water column above from the bottom of
k /k o = n = no + ifi,
equal to 1.
(50)
whereno is the reference indexof refraction and is usually The phase of a planewavein the direction of propagation is then as follows:
If any modeis at cutoff,i.e., whenko = y,, our method andtheabove error analysis is invalid.To avoidthispossibility duringactualcomputations, it isnecessary to compare ko and y,, beforehand and shiftthe valueof ko suchthat it lies
betweenmodes.However, if the highest-ordermode is very
( 51)
(52)
c= soundtransmissionlossin dB/m/kHz,
(47)
(53)
a.,.,,, --
koZk(r,z)dz,
(48)
(49)
TABLE I. Idealized ocean environment andpropagation parameters. The soundspeed is a linear functionof depth between givenpoints;frequency = 25 Hz; source depth= 500m; receiver depth= 500m;andmaximum
attenuation = 2 dB/m/kHz.
Depth (m)
0 1000 1500 2560
CaseA
1500
Soundspeed(m/s) CaseB
1500
CaseC
1500
Test computations haveshownthat both the fourth-order Runge-Kutta as well as the fourth-order Adams-Bash-
forth-Moulton predictor-corrector algorithm (with a fourth-order Runge-Kutta startoff) can be successfully usedto numericallysolvethe resultingEqs. (32) for the
1866 J. Acoust. Soc.Am.,VoL88, No.4, October 1990
1866
using Kaiser-window techniques, anda high-angle PE forasound speed of 1600 m/s,shown inFig.4(b). Similar to source function by Greene 7 wereconsidered. All threeini- case B, thehigher sound speed contributed to a slight imtial fields gave model results for thetestcases thatdiffered provement indetail inthe10to20-km region. A benchmark very littlefrom each other onthegraphical scale oftheresul- for computational speed can be provided with the knowledge tant transmission losses. thateach of theabove cases was computed byone thousand Thereference sound speed for theinitialfieldwastaken 20-mrange sections using thefourth-order Runge-Kutta tobe1510 m/s,while themodel reference wave number ko scheme, which took100s of scalar processing onan IBM andhence the model reference sound speed wasobtained 3090.Since eachstepof the fourth-order Runge-Kutta
from
Fourier sine transform (see theAppendix) for thefourth(54) order scheme, eighttransforms are required. In addition,
oneinverse Fourier sine transform isrequired at theendof where theintegral was designed tominimize themagnitude theentirestep giving a totalof nineFouriersine transforms
ofkov inG(r,y,, ). This gave model reference sound speeds for eachrangesection or a total of 9000 Fouriersinetransof 1520, 1547, and1574 m/sforcases A, B, andC, respec- forms for 20 km.In comparison, theSPEimplemented by tively. It should benoted thattheintegral does nottakeinto thesplit-step Fourier transform algorithm took just25sfor accountthe highersoundspeedin the falsebottomand, thesame computations using anidentical range step. hence, the actual model reference sound speeds couldbe The model wasalsoused to compute sound-pressure higher. transmission losses in an idealized range-dependent ocean UsingThomson andChapman's initialfield,the com- environment, asshown inFig.5.Therange dependency was putational results ofthe transmission loss (dB)versus range provided by reducing the sound speed froman otherwise (km)are shown inFigs. 24. The overall agreement isgood. homogeneous 1500 m/sby0.2m/spermeter concentrically An interesting feature of themodel isshown in Fig.2 (case along theradiiof a circle of diameter 50.5kin,extending A), where theresultant detail generally agrees wellwiththe from a range of 5-15kmandreaching a maximum depth of FFP result inthe 10-to20-kinregion thanwithFishman and 500m at a range of 10km.Thisgave a sound speed of 1400
MeCoy'shigh-frequency modelresult. The modelwassome-
m/s at theocean surface at a rangeof 10km, anda vertical section through thisringof reduced sound speed isshown in
(b) formodel reference sound speeds of 1547 and 1600 m/s, Fig. 5. Similar to casesA-C, a false bottom from 1500 m respectively forcase B. Thehigher sound speed resulted in a downto 2560m withanexponential attenuation profile was small improvement of detail of thehightransmission losses used to minimize reflections fromthebottom of thegrid.A at 11, 15.5, and 17 km. Results for caseC were obtained for normal-mode solutionfor a homogeneous oceanwith a thesound speed of 1574 m/s,asshown inFig.4(a) aswellas source at a depth of 500m wasutilized to generate thestart-
z
I-
8O
9C
lOC
RANGE (kin)
1867
1867
7o
9c
10C
sultfromFishman andMcCoy; s__, Fishmanand McCoy'shigh-frequency model result.(b) Computational resultsfor caseB with a model
7O
80
go
awide-angle split-step Fourier transform algorithm, as out20 m, respectively. The results are shown as acontour map of lined byThomson and Chapman 4 byassuming negligible the transmission loss (dB) over the entire grid inFig. 6. range dependency, and the results are shown inFig. 7.The Asa comparison, thetransmission loss fortherange- similarity between Figs. 6 and7 is remarkable. Closer independent test environment was also computed bymeans of spection ofFig.7 reveals that thecontours donotquite
(i-
7O
9O
RANGE (kin)
!oc
RANGE (km)
matchthose of Fig. 6 after a rangeof 10km andare spatially shifted, suggestive of accumulated phase errors,whichare to be expected from the application. of the split-stepFourier transformalgorithm to a range-dependent environment. As a control,Fig. 8 gives a comparison of transmission
1869 J. Acoust. Soc.Am.,Vol.88, No.4, October1990
speedof 1500 m/s and with an exponentially attenuating falsebottom for a normal-modeinitial field. It is interesting to note that evena weakly range-dependent oceanenvironment suchas the ring shownin Fig. 5 considerably affects
Dalryrnple et al.: Wide-angle acoustic model 1869
RANGE ( kl ,
0.2
10
15
20
//
o o
underwatersoundpropagation. A more severetestwas providedby applyingthe model to the range-dependent oceanenvironment, asshownin Fig. 9. The contoursare similar to thosein Fig. 5 exceptthat the soundspeed variesfrom 1500m/s by 0.6 m/s/m concentri-
eally along the radii from 5 to 15 km and givinga sound speed of 1200m/s at the oceansurfaceat a rangeof 10 kin. A falsebottom,similarto that of the earlierexample,wasused
to minimize reflections from the bottom. A normal-mode
RANSE (k)
0.02
lO
15
20
o o
FIG. 6. Computationalresultsfor the very-wide-angle model applied to the idealized range-dependent environment with contours of the
o o
1870
1870
RANGE
10 15
(km)
2O
0.02
FIG. 7. Computational resultsfor the split-step Fouriertransform algorithm applied to the idealized range-dependent environment with
contours of the transmission loss
25 Hz at a depth of 500m was used to generate a starting field. Themodel range and depth increments were10and20 m,respectively. Figure 10shows the result foramodel reference sound speed obtained fromEq. (54), whileFig. 11
shows theresultfor a modelreference sound speed obtained
from
(55)
0.02
10
15
o o
1871
1871
0.2
lo
15
20
20 km between Figs. 10and 11.However,asexpected, there wasa strongsimilarityin the overallstructure. If a receiver wasplaced at a range of 20 km andat a depthof less than 500 m, it wouldbe importantto evaluate the modelreference soundspeed from an integralwith limits similar to thosein Eq. (55) and not from Eq. (54). However, if the receiver
For comparison purposes, the sametest casewas also modeled by the wide-angle split-step Fourier transformalgorithmwith a reference sound speed of 1500m/s. The contours of transmission loss (riB) are shownin Fig. 12. A1-
0.02
lO
15
2O
FIG. 10. Computational resultsfor the very-wide-angle model applied to the idealizedstronglyrange-dependent environment using a model reference sound speed computed througha depthof 1500m with contours of the transmissionloss (dB).
Contour interval is 5 dB.
1872
1872
0.02
o
10
15
20
FIG. 11. Computational results for the very-wide-angle model applied to the idealizedstronglyrange-dependent environment using a model reference sound speed computed through a depth
of 700 m with contours of the transmis-
o
o
thoughthereare similarities between Fig. 12 and Figs. I0 and 11,in the 10-20-km region, theyarenotasreadilyvisible asfor theprevious testcase wherea sound-speed variation of only 100m/s wasused. This is to beexpected fromthe splitstepalgorithmwith largedeviations ofk(r,z) from the referenceko(r), in additionto its inability to handlestrongly range-dependent environments. The phenomenon of wavefocusing for cases of waves
propagating through a refractive focus iswell known.Accurate reproduction of experimental results for this casehas
0.02
10
15
20
FIG. 12. Computational results for the split-step Fourier transform algorithm appliedto the idealized stronglyrangedependentenvironmentwith contours
of the transmission loss (dB). Contour
interval is 5 riB.
1873
Dalrympleeta/.: Wide-angleacousticmodel
1873
RANG;
5 250
(m)
500
FIG. 13. Idealizedoceanfor a rangedependentsound-speed environment to illustrate effects of focusing. Soundspeed contourintervalis 50 m/s.
We considered an idealizedrange-dependent oceanenunderwater sound propagation, a quitedifferent initialconsimilarto theonediscussed before except that the ditionis usedcorresponding to the form of a depth-depen- vironment dent delta function. We were interested to find an idealized rangedependency waslimitedto a semi-ellipse extending to oceantestcasecombinedwith an initial sound-pressure cona depthof 250m andsituated at a range of 100to 350m. The boundary of the semi-ellipse wasdefined according to dition that exhibitedstrongfocusing characteristics.
RAE
(m)
20
lO
500
1000
1500
2000
500
FIG. 14.Computational results for the very-wide-angle modelapplied to the idealized range-dependent environment usinga model refer-
lOOO
15oo
1874
1874
R 2+ Z 2= 1,
where
(56)
(57)
ronment. The algorithm hasbeen successfully applied to idealized ocean environments and is computationallywell suited for arrayprocessing, particularly whenthe AdamsBashforth-Mouitonpredictor-corrector schemeis implemented,alongwith vectorized computation of the discrete
fast Fourier transforms.
In the future, the algorithmwill be appliedto rangedependentoceanenvironments and will be extendedto ineludethe effects of strongdepth-dependent perturbations of
This idealized range-dependent oceanenvironment is shown in Fig. 13.The sound speed outside theellipse wasa homogeneous 1500m/s, whiletheinitialcondition of a continuoussourceof noisewas represented by an idealized square waveof magnitude unity and frequency 25 Hz extending from 100to 1400m in depth.This initial condition approximately represents the effectof a great numberof superimposed depth-dependent delta functions and hasan
exactanalyticFourier sinetransform.The contours of sound transmission loss(dB) for thiscaseis shownin Fig. 14 and
APPENDIX
3= - {f
ar
where
2i4/o: -- Ym exp(i,q/o -- m r)
= k0.
For the second-order Runge-Kutta, (r+ At) = b(r) + 0.5(k, + k2 ),
where
We havedescribed a very-wide-angle numerical method forsolving theHelmholtzequation for sound-pressure propagation for a constant density, range-dependent ocean envi-
k2
or
-- Ar-,( if-/- '[ b exp(/x/k 02 -- yz,,, r) ] } 2ik02exp(ix/k 02 _ -- Ar-s(Vl,f 1{(I//--[ kI )exp 02 - /2rn (r+ Ar)]}) 2io2-- ,. exp [ ix/k o -- y,(r+ Ar)]
(A4)
(AS)
Now,
(A7)
F. D. TapDeft and R. H. Hardin,"Computer simulations of long-range ocean acoustic propagation using theparabolic equation method," m Proceedings of the8th International Congrexs on.4coustics (Goldcrest, London, 1974), Vol. 2, p. 452.
F.D. Tappert, "Theparabolic approximation method," in Waoe Propagation and UnderwaterAcoustics, Lecture Notesin Physics, edited by J. B. Keller andJ. S. Papadakis (Springer,New York, 1977), Vol. 10,pp. 224287.
W. F. AmesandDing Lee,"Currentdevelopments in thenumerical treatment of oceanacoustic propagation," Appl. Num. Math. , 2547
{ 1987).
1875
1875
6M. D. Felt and J. A. Fleck, Jr., "Light propagation in graded-index fibers,"Appl. Opt. 17, 3990-3998 (1978). ?R. R. Greene, "Therational approximation to theacoustic wave equation
with bottom interaction," J. Acoust. Soc. Am. 76, 1764-1773 (1984).
waveequation," J. Comput.Phys.71, 304-315 (1987). E. C. Titchmarsh, Eigenfunction Expansions/lssociated withSecond-Order DifferentialEquations (Oxford [J.P., Oxford, England,1958), Part II, pp. 217-221. L. Fishman andJ. J. McCoy,"A newclass of propagation models based on a factorization of the Helmholtzequation," Geophys. J. R. Astron.
So:. 80, 439-461 (1985).
aJ.A. Davis,D. White,andR. C. Cavanagh, "NORDA Parabolic Equation Workshop, 31 March-3 April 1981,"NORDA Tech. Note 143
(1982); NTIS No. AD-121 932.
OR. A. Dalrymple, K. D. Sub,]. T. Kirby,andJ. W. Chae,"Models for verywideangle waterwaves andwave diffraction. Part2. Irregular bathymetry,"J. Fluid Mech. 201, 299-322 (1989).
S.T. McDaniel, "Propagation ofnormal modes in theparabolic approximation," J. Acoust. Soc.Am. 57, 307-311 (1975).
2j. S. Robertson, "An exactsolution to a modified wide-angle parabolic equation," J. Acoust. Soc.Am. 84, 1791-1793(1988).
A. C. Radder, "Ontheparabolic equation method forwater-wave propagation,"I. Fluid Mech.95( 1), 159-176(1979). P. L-F. Liu.andT-K. Tsay, "Refraction-diffraction modelfor weaklynonlinear water-waves," J. Fluid Mech. 141, 265--274 (1984).
1876
1876