Nearfield Acoustic Holography (NAH) II. Holographic Reconstruction Algorithms and Computer Implementation

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 17

Nearfield acoustic holography (NAH) II.

Holographic reconstruction algorithms and


computer implementation
W. A. Veronesi, and J. D. Maynard

Citation: The Journal of the Acoustical Society of America 81, 1307 (1987); doi: 10.1121/1.394536
View online: https://doi.org/10.1121/1.394536
View Table of Contents: http://asa.scitation.org/toc/jas/81/5
Published by the Acoustical Society of America

Articles you may be interested in


Nearfield acoustic holography: I. Theory of generalized holography and the development of NAH
The Journal of the Acoustical Society of America 78, 1395 (1985); 10.1121/1.392911

Method of superposition applied to patch near-field acoustic holography


The Journal of the Acoustical Society of America 118, 671 (2005); 10.1121/1.1945470

Hybrid near-field acoustic holography


The Journal of the Acoustical Society of America 115, 207 (2004); 10.1121/1.1631415

Sound source reconstructions using a microphone array


The Journal of the Acoustical Society of America 68, 340 (1980); 10.1121/1.384602

Application of BEM (boundary element method)-based acoustic holography to radiation analysis of sound
sources with arbitrarily shaped geometries
The Journal of the Acoustical Society of America 92, 533 (1992); 10.1121/1.404263

Helmholtz equation–least-squares method for reconstructing the acoustic pressure field


The Journal of the Acoustical Society of America 102, 2020 (1997); 10.1121/1.419691
Nearfield acousticholography (NAH) II. Holographic
reconstruction algorithms and computer implementation
W.A. Veronesiand J. D. Maynard
Department
ofPhysics,
ThePennsylvania
StateUniversity,
University
Park,Pennsylvania
16802
(Received28 October1985;acceptedfor publication31 December1986)
Thebasictheorytreatingsteady-state acoustic radiationproblems in thenearfieldhasbeen
presented
in severalarticleson nearfieldacousticholography. In thisarticle,the
approximations
andassumptions necessaryto reducethe infiniteand continuous convolution
integrals
encountered
in these
problems
toa finiteanddiscrete
form,suitable
forhigh-speed
numerical
processing,
areilluminated
theoretically
andtestednumerically.
To evaluatethe
convolution
integrals
twoassumptions aremade:First,theboundary
fieldmaybereplaced
witha patchwise
constant
fieldforreasonably
smallpatches;
and,second,
thefieldisnegligible
outsideof a finiteregion.With thesetwoassumptions,
theproblemreducesto oneof
representing theGreen's functions.Sixmethods
of sampling
or representing
theGreen's
functionsaredeveloped, andthesearecomparedtheoretically
andnumerically.
PACS numbers:43.20.Rz, 43.60.Sx, 43.35.Sx

INTRODUCTION
lemin othercoordinatesystems
isdiscussed
in thereviewof
An importantproblemin theoreticaland experimental NAH.2Thesolution
to theHelmholtzequation
( 1) forthe
soundradiationresearchis the numericalprocessing of con- different(D or N) boundaryconditionsis a two-dimension-
volution integrals.Evaluation of convolutionintegralsis al convolutionintegral,
most expedientin determiningthe acousticfield produced
by theoreticalor measuredboundaryconditions. • Inversion
of convolutionintegralsor deconvolutionis likewisedesir-
--ff_©½D,N
(X'y')GD,N
(X--X',y
--y',z)dx'
dy',
(2)
able for solvingproblemsin inversescatteringand inverse
sourcetheoryandin experimentalsoundradiationmeasure- wherethe Green'sfunctionsGo and GN for an infiniteplane
ments as discussed in an earlier review of nearfield acoustic boundaryareknownS:
holography(NAH). 2 Sincemostmodelsof radiationas- GD(x,y,z) = z( 1 -- ikR )eikR/2•rR 3, (3)
sumeinfinite and continuousfields,the integralsencoun-
GN (x,y,z) = ieikR/2•rR3, (4)
teredin the theory involvevariablesdefinedoverinfiniteand
where
continuousdomains.Experimentalmeasurementand digi-
tal processing requiretheseintegralsbe replacedwith finite R x/x2 + y2 _+_
22 '
and discreteoperations.However,the useof finite and dis-
creteoperationsin placeof the infiniteand continuousinte- The Neumann condition has the form
grations
canintroduce
significant
errors.
3Recently,wehave
developednew techniquesfor minimizingtheseerrorswhile (5)
maintainingshort digital processingtimes, a necessityin & z=0

nearfieldacousticholographyandotherapplications.In this
Historically,Eq. (2) with two conditions(D or N) are re-
article we presenta comprehensive
studyof the processing
ferredto asthefirstandsecond Rayleighintegrals.7
techniques,
includingsomedeveloped
earlier,2andexamine
For numericalprocessing, it iswellknownthat a convo-
the resultsin testcasescoveringa rangeof parameters.
lutionis morereadilyevaluatedby usingFouriertrans-
forms.8We shalldenotethecontinuous, infinitetwo-dimen-
I. FORMULATION OF THE PROBLEM
sionalFouriertransformof a functionf (x,y) by

As discussedin numeroustexts, and in the review of


NAH, radiationproblems involvefindingthesolution0 (a
complexfield)for theHelmholtzequation,4-6
.•(k•,,ky
)= ff._• -i(k•+•y.v)
f (x,y)e dxdy (6)

+ k = 0, and the inverseoperationwill be indicatedasF- •. The do-


mainof (kx,ky) will bereferredto ask space,
whilethatof
withfixedk -- 2•r/A,A beingthecharacteristic
wavelength (x,y) will bereferredto asrealspace.
of the radiation,andwith specified
Dirichlet0o or Neu- Applyingtheconvolution theorem9to Eq. (2) yields
mann 0$ conditionson someboundary.In this article we
restrictour discussion
to planeboundariesin a Cartesian O(x,Y,Z)
=F-I[•D,N(kx,ky)•D,N(k•,ky,z)]
. (7)
coordinate system
anduse0D(x,y) or 0$ (x,y) ontheplane TheFouriertransforms oftheGreen's
functionsGo andGN
z -- 0 astheboundary
condition.
Formulation of theprob- in Eqs.(3) and (4) maybefoundanalytically1ø:
1307 J. Acoust.Soc. Am. 81 (5), May 1987 0001-4966/87/051307-16500.80 @ 1987 AcousticalSociety of America 1307
k xd-ky
2 2•k2 ,
(8)
=texp( - + 2 _k •) , k 2 -1-k 2 2-
A -kx -k.v, k X2 d- k2<k 2'
GN(k•
,ky,z)
= exp(/zx/k -
- i exp( -2
k •
z• k2
- k y
2 d-k2
,, y
)/x/k2 2 22- k 2,
2-- k 2)/!]k,,2 + k.v kx
2 d-k.v
y

2 >k 2.
(9)

f
The circlek 2 + k 2 = k 2 is referredto as the radiationcir- z = Zoin processes
b andd, the process
shallbereferredto as
cle.Forpoints(kx,kx) insidetheradiation
circle,GD,Nrep- a transformation.
resentsthe z-directionphasechangeof propagatingplane In the theoreticalsectionof this article, Eq. (2) will be
w•aves,whileforpoints (kx,kx) outsidetheradiationcircle used primarily to simplifynotation.In the numericaltesting
GD.Nrepresents the rapidexponential decayof evanescent section,•N at z = 0 will be used to generatea test field
waves. By defining a complex function kz(k,,,k.v) •(x,y,z), and then the four processes will be examinedfor
= x/k2_ k 2 -- k 2 thenthetransformed Green'sfunc- various valuesof Zo and z.
tions may be written in the condensed forms
A A
II. FINITE AND DISCRETE OPERATIONS
Gt• = exp(ikzz) and GN = exp(ikzz)/kz.
For holographyor inversescatteringproblems,the use A. General assumptions and formulas
of Eq. (6) is essential,becausein this way the convolution As already discussed,numerical evaluation of the for-
integral (2) may be quickly inverted; that is, the field mulasin Eq. (2) or (6) requiresfinite, discreteoperations.
•(x,y,zn) may be deconvolvedto yield the sourcefield The limits of theseoperationsmay be establishedeither by
•D,N(x,Y). Assumingthat •(x,y,zn) is known,we have experimentaldata acquisitionor by computationtime and
from Eq. (6) capacity.In either caseseveralbasicassumptionsmust be
•D,N
(x,y)F-l[•l(kx,ky,ZH•
•' -1 (k•,,ky,zn)],
D,N (10) made. The first is that the sources of the wavefield are such

where• r•,$
-• (k• ,ky,zn), theinverseof theformulain Eqs. that the boundarydata •Pr•(x,y) or •N (x,y) are negligible
outsideof somefinite domain in real space.Denoting this
(7) or (8), providesthe reconstructionof both propagating
domainby - L/2 < x < L/2 and -- L/2 < y < L/2, Eq. (2)
andevanescent
waves.Oncethe sourcefield•Pr•,$hasbeen becomes
determined,all other propertiesof the field may be calculat-
ed as discussedin the earlier review article. 2 It should be
noted that, as far as the processingtheory is concerned,
•(x,y•z) '•'
f,2 L/2
•D,N (x',y')GD,N
½r•,$(x,Y) neednotcorrespond
to a physicalsourcesurface. X (x- x',y-y',z)dx' dy'. (12)
In fact, Eq. (2) may be generalizedto apply to the field and
its derivativebetweenany two planes,oneat z and the other
The secondassumptionis that the wavefield•pcanbe repre-
at Zolying betweenzero (the actual source)andz. The more
sentedsufficientlywell with a discrete,aswell asfinite, setof
generalequationsare
numbers.This setof numbersmay be a data setfrom experi-
mentalmeasurements at latticepointsin real spaceor coeffi-
cientsfor a superposition of basisfunctions.In thisarticlewe
adoptthe simplestcasein which it is assumedthat the wave-
X (x -- x',y --y',z -- zo)dx' dy', ( 1la) field•pisnot significantly
alteredif theboundaryfield•Pr•,$is
replacedwith a piecewiseconstantfield. That is, the L X L
realspacedomainis dividedinto N 2 patchesof size(L/
Ip(x,y,z
)= • (x',y'•Zo
)GN N) X (L/N), labeledwith integersl,m = 0,1,...,N- 1, and
thediscretesetof data•Pr•,$
(l,m) isassumed
to betheaver-
X (x-x',y-y',Z-Zo)dx' dy'. (11b) age of the actual boundaryfield over the patch. The field
The evaluationor inversionof thesetwo equationsrepre- radiatedfrom the piecewiseconstantboundaryfield can be
sentsfour operationsreferredto as: assumedto represent•pif N is sufficientlylarge.
(a) propagationof a wavefront,where•(x,y,z) isdeter- For the piecewiseconstantboundaryfield, Eq. (12) be-
comes
mined from • at %;
(b) transformationand propagation,where•b(x,y,z)is
determinedfrom c•/8z at %;
(c) reconstruction,where•(x,y,zo) is determinedfrom
•p(x,y,z) =
•1N•I
•D,N( l,m)
1= 0 m = 0 dx t -- A/2 m -- A/2
GD,N

•(x,y,z); and X (x- x',--y-y',z)dx' dy', (13)


(d) reconstruction and inverse transformation, where with
(•/8z) (x,y,zo)is determinedfrom •(x,y,z).
X1= (l + « -- N/2)A, (14a)
Sincez > Zo> 0 and zerois the actualsourceplane,then the Ym= (m + •--N/2)A, (14b)
reconstructionprocessesmay be referred to as the inverse where A = L/N.
propagationof a wavefront back toward the source.When Finally, it is assumedthat it is sufficientto examinethe

1308 d. A½oust.Soc. Am., Vol. 81, No. 5, May 1987 W.A. Veronesiand d. D. Maynard:Nearfieldacousticholography 1308
field•bin anyplanez at discrete
points(%,,yq) defined
asin discreteconvolutiontheorem must be rederived;the details
Eq. (14) for integersp,q = 0,...,N -- 1. By definingvariables of this procedureare presentedin AppendixA. Briefly, the
U= Xp-- X', V• yq-- y' wehave results are as follows.
Thesequence
•PD,N
(l,m) thatisdefinedonlyforintegers
•,+•/•Iy•+•/•GD,s (xp-- x',yq-- y',z)dx'dy',
dx ! -- A/2 d Ym -- A/2
(l,m) in (0, N- 1) must be extended over a (2N) X (2N)
domainby addingzeros.This newsequence
is definedwith
:
(;pp--l
q-
1/2)A
f(q--
-- 1- mq-
1/2)A
1/2)A d(q -- rn -- 1/2)A
GD,N (u,v,z)dudv
gbb.
N(l,m)
= [½'D'N
t0,
(l,m),ififN0 <l<Nand 0<m<N,
<l < 2N or N<m < 2N.
--=GD.•(p -- l,q -- re,z), (15) (17)
with integers l,m,p,q--O,...,N--1. ThediscreteLetting •b(p,q,z)
FouriertransformDFT { f } of anarray(l,m)
= •b(xe,yq,Z),
theexpression
fortheradiated
fieldbecomes with (l,m) in (0, 2N-1) is defined (for /z,v=O,
N--1N--1
1,...,2N- 1) as
•( p,q,z)= Z 1=0
Z •D.N(l,m)GD.N
m=O
(P-- l,q-- m,z),
2N-- 1 2N-- 1
(16)
DFT{f}•,,,,= Z Z f(l'm)e-(i'r/N)(O•+m")
which is the discrete,finite versionof the Rayleighintegral l=0 m =0

(2). (18a)
As with the infinite, continuousconvolution, the finite, and the inverse is indicated as
discreteconvolutionin Eq. (16) is most readily evaluated 1 2N-- 1 2N-- 1
and inverted using the convolutiontheorem and Fourier
transforms(in particular,the fast Fourier transformalgo-
IDFT(F)t,m
4N2 • •

/z-----0 • F(/z,v)e
v=0 +(,r/m)(t•,
+ .

(18b)
rithm, FFT). Although discussions of the discrete,finite
convolutiontheoremmay be foundin textbookson signal As shown in Appendix A the finite discreteconvolution
processing,most treatmentsassumethat both arrays to be (16) can be calculated with
convolvedare either periodicor zero for indicesoutsidethe
range 0, ....N-- 1. However, in the discrete convolution in •(p,q,z) = IDFT [ DFT(•b,N)gD,N
(Z)], (19)
Eq. (16) oneof thearrays,GD,N ( p -- l,q -- rn,z) [definedin where
Eq. (15) ] maybeevaluatedfor all integersl, rn,p,q, because
theGreen'sfunctionGD,N(U,V,Z)isknownanalytically over gD,$
(Z)•,,v-- DFT{• b,$(z))•,,v, (20)

an infinitedomain.Becauseof this exceptionalfeature,the with


I

•D N(l,m,z), if 0<I<N and 0<m<N,

-- •D'N
(l--2N,
m,z), if N<I < 2N and 0<m < N, (21)
Gb,N
(l,m,z)
-- •D'N
(l,m--2N,
z), if 0<l <N and N<m < 2N,
•D]N
(/- 2N,
m-2N,z), if N<I < 2N and N<m < 2N.
I

B. Convolution, wavefront propagation • D,N


(1) (r,s,z)= (L/N)2G D,N (rA, sA,z), (22)
Equations(19) and (20) form the computationalbasis wherer,s = - ( N -- 1) ," .,0,..,N.
ß Here or(1)
•SD,N is the discrete
for wavefrontpropagationin which•b(p,q,z) is determined transformof the analyticGreen'sfunctionsampledat dis-
in any plane z>0 given the data •bD,N(l,m), crete points in real space. The Green's function
(l,m) = 0,...,N -- 1.The datamanipulationsimplyinvolves GD.N(rA, sA,z) is singularwherer = s = z = 0, so when
adding zeros to the data sets to form an extended r--s = 0 and z is lessthan one-tenthof a wavelength,an
(2N) X (2N) domain, performinga two-dimensionalDFT analyticintegratedaverageof the Green'sfunctionis used.
on the extendeddata set, multiplying by a "propagator": This integrationis over a circulararea equalin sizeto the
gD,N (2), andthenperforming anIDFT. The computational samplelatticesquare area,A2;thedetailsarein AppendixB.
difficultyliesin evaluating
thepropagator
arraygD,N(2) We refer to this form as the "sampledreal-spaceGreen's
that must be accomplishedfor eachvalue of z and k (if the function."
radiation involvesdifferentwavelengthsA). If Eq. (20) is (2) ReplaceGD,N•
(U,V,Z)in theintegralofEq. (15) with
used,thenit is necessary to evaluatethe integralin Eq. (15) a polynomialexpansionabout the centerof the patch. A
whichdefinesGD,N.Sincethe integrationcannotbe per- procedurefor accomplishingthis is presentedin Appendix
formedanalytically,it mustbeevaluatedwith someapproxi- B. Theresulting
approximation
forGD,Nisgiventhesymbol
mation. Two easywaysof doingthis are asfollows. GD,N ,(2) (z) is
(2) (r,s,z)withr,s= - (N • 1)," .,0,N.Here•i).N
( 1) ReplaceGD,N (U,U,Z)in the integralin Eq. ( 15) by as definedby Eqs. (20) and (21) with GD,N replacedby
its value at the center of the integration patch; i.e., • (2). thisform
D,N, isreferred
toasthe"integrated
real-space
GD,N(U,V,Z)• GD,N[ ( P -- l) A,(g -- m) A ]. The resulting Green's function."
propagator ,,(1)
•D,N(Z)•,• is asdefined by Eqs.(20) and (21) In derivationofEqs. ( 19) and (20) thedefinitionsof the
with GD,N replacedwith real-spacesamplingpointsgivenby Eq. (14) wereused.Be-

1309 J. Acoust. Soc. Am., Vol. 81, No. 5, May 1987 W.A. Veronesiand J. D. Maynard: Nearfield acousticholography 1309
causetheDFT requiresan evennumberof samplingpoints,
the definitions(14) resultin an asymmetricdistributionof GD,N(kx,ky,2)= e GD,N(x,y;z)dxdy. (25)
pointsin the samplingof the Green'sfunction[as in Eq.
(15) ]; that is,thereisan oddpointat theorigin,N negative In the followingsteps,the right-handsideof Eq. (25) will be
values,and N- 1 positivevalues.By redefiningthe real-
approximated
and•mparedwithEq.(24), providing
a re-
spacesampling, a symmetricdistributionof pointsmaybe lationshipbetweenGD,N andgD,N'
obtained;the Green'sfunctionwill then be sampledover The integralof Eq. (25) isbrokenup into a sumof patch
patchescenteredat (rA d- A/2, sAd- A/2). This shifted integrals
withtheapproximation
thate- i(kxX
+k•) iscon-
form of Eq. ( 15) is A A

stantovereachpatch.The resultis GD,N • G •,N, where

D,N(kx, ,Z) = e
.•
Go,• ( r,s,z) --
f(r+
"--JrA 1)A
(fsA•
q-1)A
Go• ( u,v,z) du dr,
'
(23)

• G•,• (x,y,z)dxdy.
with r,s = -N,...,0, (N •- 1). If a sampledapproximation
to GO,Nis usedin Eq. (21), thenoneobtainsthe "shifted (26)
sampled
real-space
Green's
function,"
•,(3)
•SD,N (z)•,,v
ß There
is An estimateof error from approximatinge as
no explicitreasonwhy the symmetricsamplingmight pro- constantovereachpatchis shownin AppendixC to be
duce more accurate results, other than that it makes better A

useof the radial symmetryof the Green'sfunction.The fact • . (27)


isthat it doesproduceslightlybetterresults,asshownin Sec. •v,$(k•,ky,z) 4sin(k•A/2)sin(kyA/2)
IV. Whenan integratedapproximationGo,Nis usedin Eq. Thiserrorislimitedif the2Nvalues
ofk• andkyarechosen
(21 ), oneobtainsthe "shiftedintegratedreal-spaceGreen's so as to minimizethe distancefrom the origin in k space.
function,"
•,(•) (z)•,,v
•SD,N ß Useof theunshifted
Green's
func- Thisrestrictionalongwith considerationof theFouriercoef-
tions is equivalentto replacingthe field ½(x,y,z) with a ficientsin Eq. (24) motivatesthe definitions
patchwiseconstantfield where each samplingpoint is cen-
teredin a patch;useof the shiftedform corresponds to hav-
ing the samplingpoint at onecornerof its patch.If the same
k•• [t•(
•p/NA' if
p- 2N)/NA, 0<p
<<N,
ifN<p 2N, (28)
setof fieldvalues•Po,N
(l,m) isusedfor thesamplepointsin and similarlyky• in termsof •. Evaluating(k•,ky) at
both cases,the net resultis a shiftin the apparentlocationof (k•u,ky•) andusingEq.( 15) defining
GD,N,
Eq.(26) canbe
the representedfields.This shift shouldbe consideredwhen written
theoreticalfieldsareused,but it is of little consequencewith •, N--1 N=I
experimentally measuredfields; with an experimentally GD,N(kxI•'kY
•'z)• Z Z e--(•r/N)(l•r+w)
r= --Ns= --N
measuredfieldthereis alwaysan ambiguityof half a sample
spacingin thex andydirectionswhendecidinghowto repre-
sentthe field with a patchwiseconstantfield. Throughout X Z Z •D.N
this study, wavefrontspropagatedby the shifted Green's
X (r + 2mN, s + 2nN, z). (29)
functionsare comparedto theoreticalfieldsshifted - A/2
in the x andy directions.
In ordertodetermine
go,N(z)•,• fromEq.(20) it isnec- Thern= n = 0 termissimplygD,N(2)uv,SOthatEq. (29)
can be written as
essaryto performthe DFT operation.However,in the infi- A

niteandcontinuous
formulation
inE•. ( 6)-( 8), theFour- GD,
N(kxt
• ,ky•,2)
ier transformed
Green'sfunctionGo,N(k•,ky,z)has a N--1 N--I

known (fairly simple) closedform. If an approximationfor


A
•gD,N(J)U,
v d- Z r= --Ns=
Z e--(i•r/N)(ur+vx)
--N

go,• (z)•,• canbefoundin termsof Go,N(k•,ky,z) in Eqs.


(7) and (8), thengo.• (z)•,• couldbe calculated
without
having to perform a DFT. X •
m= --oo n=
• --•o
(JD,N
(r d-2mN,
s+ 2nN;z). (30)
In orderto derivesuchan approximation,Eq. (20) giv- (rn = n)%0
A

inggo,N(z)•,• isexpanded
usingEqs.( 15), ( 18), and(20), Oneshould
notethatif G(k,,•,ky•,z)replaced
g•,N(z)
andtheperiodicityof theFouriercoefficients,
exp( -- irrpl / in Eq. (19), only the re=n=0 term of Eq. (29) [the
N) = exp[ - irrp(l - 2N)/N ] is used.Equation(20) can g•.N(z)•v terminEq.(30) ] willyieldtheresultdesired
asin
then be written as
the discreteconvolutionof Eq. (16). The summationre-
mainingin Eq. (30) effectivelyextendsthe rangeof the sum-
N•I N•I
mationin Eq. (16) to includethe infiniteperiodicextension
gD'N
(2)tz•'= Z r= --Ns=
Z e--(irr/N)(t•r+ys)
--N of the finiteinput data field.This extensionmay be viewedas
representingan infinite set of fictitious image sources;an
X
f((rr--•l
1/21
)•f({S
S/1
lii)iGD,N(x,y•)& dy.
(24)
apparentcopyof eachfieldpointislocatedat all of thepoints
(x + 2mL,y d- 2nL ) for all pairsof nonzerointegers(re,n).
Thesefictitiousimagesourcescan producelarge errorsin
The Fouriertransformof G•,• is expressed
explicitlyas numericalcalculationsof propagation.

1310 J. Acoust.Soc. Am., Vol. 81, No. 5, May 1987 W.A. Veronesiand J. D. Maynard:Nearfieldacousticholography 1310
If GD,N (x,y,2) falls off rapidly with x and y, then f 1 Fkrq-kO
GD,N (x,y,z) also falls off rapidlyand the effectof image
sourceswill be small.In thiscasethe summationin Eq. (30)
12k'
kI &D,N
(k;,z)k
;dkj, kr>ko,

•may be ignoredand Eq. (30) then showsthat


GD,N(kx•,kyv,2) approximates
gD,N(Z)•v directly.Thuswe
have the "sampled k-space Green's function," labeled (34)
g(5>(z)• whichisdefined
D,N by
, wherek •2= k x•,
2 q-k ys.
2 Closed-form
evaluations
are pre-
sentedin AppendixB. The sixdifferentformsfor the discrete
g(5)(Z)•,,,•
D,N =•DN(kx•,,ky,•,Z) , ' (31) Green's function are summarized in Table I.
On the otherhand,if GD,N (x,y,z) doesnot fall off rapidly, A

thenthesumin Eq. (30) maybesignificant andGD,N maybe


C. Deconvolution, source field reconstruction
an unacceptable
approximation of gD,N'Inspectionof Eqs.
(25)-(30) shows that this sum could be reduced if For the caseof infinite and continuousfields,the convo-
GD,N (x,y•z) in Eq. (25) werereplacedby GD,N multiplied lutiontheoremprovidesan exactformulafor the deconvolu-
by a suitablewindow function W(x,y). This window func- tion or reconstruction
of ½•,$(x,y) from½(x,y,zH), asin
tion shouldhave a valuenear unity for x andy in the range Eq. (10). It wouldbe temptingto try the sametechnique
[ -- L,L ] and shouldfall off rapidlyoutsidethisrange.Such with the discreteconvolutiongivenin Eq. (19). One could
a window would reduce any fictitiousimage sourceerror write

without
in•troducing
anynewerror.
2A simple
closed
form •bb,N
(/,m)= IDFT[DFT(½(p,q,ZH)}/gD,N(ZH)
]. (35)
functionG •,N is soughtsuchthat
A
However, Eq. (19) can be derived (without approxima-
;;,N tions) only if the DFT is defined for a 2N X 2N domain.
Equation(19) wasfoundby assuming
that½b,$(l,m) was
= W(x,y)GD,N(X,y,z)e k•V>dx
dy, (32) zero at the extra pointsin the extendeddomain.In order to
evaluate Eq. (35), the (in general nonzero) values of
and the sameprocessoutlinedin Eqs. (25)-(30) yieldsan 0( p,q,zH) mustbeknownat theextrapointsin the extended
expressionanalogousto Eq. (20). domain.However, in holographythe "hologramdata" are
definedto betheN X Narray 0( p,q,z• ). AlthoughN canbe
A redefinedsothat the hologramdata coverthe (2N) X (2N)
G•,N (k•, ,ky•,z) domain, it is almost certain that the data measured outside
N--1 N--1
theN X N domaindonotcorrespond to theidealvaluescom-
=gD'N
(Z)/•'v
q- E r= --Ns=
E e--(irr/N)(l•r+vs)
--N
patiblewith the derivationof Eq. (35). Furthermore,one
mustconsiderthe possibilitythat for some(/t,v) the array
oo oo i
element gD,N(Z) may be zero; in the ideal case
X • m= --oo n=
• --oo
W(r+ 2mN,
s+ 2nN){JD,N DFT{½(p,q,z•)}•,,•willalsobezeroandthequotient
inEq.
(m =n)•O (35) will be indeterminate.If 0( p,q,z• ) are the data mea-
X (r + 2rnN, s + 2nN, z). (33) sured over the 2N X 2N domain, it is very unlikely that
DFT{½(p,q,ZH
)}•,,•willbeexactly
zero,sothatitsusein
Eq. (35) mayleadto seriouscomputationalerrors.Thus,for
Becauseof the windowfunctionthe multiple summationis
measuredhologramdata, Eq. (35) cannotbe useddirectly,
not as significantas that of Eq. (30) regardlessof how fast
and the sourcefield reconstructionproblemmust be refor-
GD,N(x,y,z) decreases with increasing x andy. mulated.
Ideally,the valueof thewindowfunctionwouldbeunity
Oneformulationof thesourcefieldreconstruction prob-
within the limited range(corresponding to rn -- n = 0) and
lem can be found from Eq. (16), which can be written in
zero outside,but neitherEq. (32) nor its equivalentconvo- matrix form:
lutionintegralin k spaceyieldsa simpleclosed-form expres- N--1N--1
sionfor sucha window.The goalhereis to reducethe com-
½(p,q,z•)= • • Gl"•
(D,$•pq
½D,N
(l,m), ( 36)
putationtimeof gD,N(Z)• in Eq. (20), anda windowfor l=0 m=O

which Eq. (32) can be evaluatedanalyticallyis necessary. where the elements of the N2XN 2 matrix Glm
(D,N)pq are
The "radial sinc function,",2J•(kor)/kor [J1(z) is the GD,N ( _]9
-- [, q -- m, zH ). The sourcefield reconstruction
first-order
Besselfunction,ko• rr/L, andr2 = (x2 + y2) ] is problemmay be solvedby invertingthe matrix G (D,N)pq ß
a suitablewindow function.It doesprovidea simpleand A
However,invertingthismatrixforlargeN mayexceedcom-
closedformexpression for G •,N. Thiswindowfunctionre- putationalcapabilitiesand will precludeeasy high-speed
duces
thesumin Eq. (33) sothatG•,N•k•,ky•,z) more dataprocessing.It may be possibleto usethe EFT to obtain
accurately
approximates
gD,N(Z)• thanGD,N
(k•,k•r,z). an approximateinversionof the matrix, and then iterate to
Usingtheconvolution
theorem,
it canbeshownthatG •,N improve the approximation.Further researchwith matrix
for this windowis approximatelyan integratedaverageof
A inversion
techniquesiscurrentlyin progress.
TM
G•,N overan annularregionaroundpoints(k•,ky). This
A Anotherapproachto the inversescatteringproblemis
functionG •,N isthe"integrated
k-space"Green'sfunction, to assumethatboth•,N (x,y) and•(x,y,z• ) arenegligible
labeled,(6> , which is definedas3
6D,N(J)ktv for sufficientlylargex andy. This is especiallyvalid in near-

1311 d. Acoust.Soc. Am., Vol. 81, No. 5, May 1987 W.A. Veronesiand d. D. Maynard:Nearfieldacousticholography 1311
TABLE I. Summaryof the sixdifferentformsfor the discreteGreen'sfunction.

Discretely sampled Integrated average

Defined in real space,


asymmetricdistribution g{1)
= FFT[G(xi,Y
j)]
of points,FFT to k space

Defined in real space,


symmetricdistribution
of points,FFT to k space
g(3)
__FFT[G(xid-A/2,y,+ A/2) ]
g(4)--'
FFT(f;;i
+a/2
G(x.y.dx
dy)+

Definedin k spacefrom
analyticFourier transform
g(6'
=f f•,•&(
k•,ky
)dk•
dky

fieldacousticholography,wherethe hologramdataare mea- 2N X 2N zeroaugmentedhologramarray. This extendedar-


suredascloselyaspossible to thesources,
whichareassumed ray islabeled•p'( p,q,zH) in analogywith •P•.N(l,m).
finite in size. The measuredhologram data are assumedto Basedon the aboveconsiderations,
Eq. (10) canbe ap-
comprise the NXN array •p(p,q,z,), for p,q proximatedasfollows:
= 0,1,...,N -- 1. As already mentioned,Eq. (10) provides
an exact deconvolution formula for infinite, continuous
lPD,
N(l,m)= IDFT{W(k•,•,
,ky,,
)
fields.This equationcanbe employedby approximatingthe XDFT[•p'
(p,q,zH
) ]&D.N
-1 (kx/•.kyv•.)).
infinite continuousFourier transformof the hologramfield (38)
•p(x,Y,ZH ). With the assumedfiniteextentof ½(x,y,zH) the where W(k•,,ky,,) is the suitablefilter functionand
DFT of the hologramdata•p(p,q,zH) is a logicalchoicefor •P'(p,q,zH) is the hologramdataaugmen.ted with zeros.For
suchan approximation,but the DFT of a sampledfieldwill the reconstructions presentedlater in thisarticle,the follow-
differ from the continuous field transform. The nature of this
ing filter functionwas used:
difference must be considered, and the DFT results must be
corrected. 1• e(kr/kc--
2
1)/a kr<kc,
In AppendixC it is shownthat the DFT of a piecewise (39)
constant field deviates from its continuous transform as IV(kx•
,kyv
)= (1--kr/kc)/a
,e kr > kc,
2 '
DFT{•p(x,y)
)•,v
wherek r2 = k x•,
2 + k yv2 andkcanda areadjustable param-
4 sin(kx•,A/2 ) sin(kyvA/2 )etersof the filter. Experiencehasshownthat a valueof kc
(37)
equal to 0.6 (Nrr/L) and a value of a equal to 0.2 yield
The field•p(x,y,zH) is assumedto beadequatelyrepresented consistently goodresults.In a particularexperimentit may
by a piecewiseconstantfieldwith thefieldvalueof eachpiece bedesirableto changetheseparameters,or possiblythefilter
givenby one correspondinghologramdatum value. Equa- function'sforms,basedon somea priori knowledgeof the
tion (37) is therefore relevant, and it indicates that, while source.However, any filtered source field reconstructions
the DFT of the hologramdata will providereasonableesti- presentedor comparedin this article havebeencomputed
matesof the transform
fieldfor smallkx•,andky•, it will withEq. (39) andthefilterparameters
kc= 0.6(kxu)max
overestimatethe higherspatialfrequencyregion. and a = 0.2.
Compoundingany error introduced by the DFT or
noise
intheinputsdata
istheexponential
growth
oftherecon- III. QUALITATIVE EXAMINATION OF THE GREEN'S
struction
kernelG D,N
- 1 (z) in Eq. (10) withspatialfrequency FUNCTIONS
and reconstructiondistance.This problem suggeststwo
A. Introduction
things:First, someform of filteringin k spacewill be neces-
sary;and,second,deconvolutions or sourcefieldreconstruc- In the precedingsectionsit was shownthat, although
tions which include the evanescent wave information must
analytic expressions exist for holographicprocessing with
be restricted to small z distances. uniquely specifiedGreen's functions,the use of finite and
Undersampling ink spac•eisanothersourceoferrorthat discreteexpressions,
A
asrequiredby data acquisitionor digi-
maymanifest itselfif either•p(x•,ky,ZH
) or GD.N
- 1 (kx.ky,Z) tal processing, introduceserrors.To reducetheseerrors,the
varies rapidly over the DFT sampling intervals exactanalytic Green'sfunctionsor their Fourier transforms
Akin,andAky•.Undersampling canbeavoidedbyreducing may be sampledor representedin variouswaysproducing
the k-spacesamplingintervals.When usingthe FFT, this improvedfinite and discreteexpressions. Coveringa wide
samplespacingis most easily reducedby augmentingthe range ofpossibilities,sixdifferent discrete
forms, ,,(1)(z)•,•
/SD,N

hologramdata ½( p,q,zH ) with zerosin the samemanneras through •,(6)(z)•,•, havebeenintroduced


/SD,N andsummarized
in ½•),N(l,rn) of Eq. (17) and performingthe FFT of the in Table I. In the nextsectiona realisticcomputermodelwill

1312 J. Acoust.Soc. Am., Vol. 81, No. 5, May 1987 W.A. Veronesiand J. D. Maynard' Nearfieldacousticholography 1312
be usedto find the bestdiscreteform for a given situation. with regularspacing;it becomesimpossibleasz goesto zero
Beforedoing a quantitativecomparison,it will be worth- and the functionbecomessingular.
whileto examinethe exactGreen'sfunctionsto gaina quali- Figure 1(a) and (b) suggests that for small distances
tative understanding of the sourceand correctionsfor the (z/A •<0.5) a direct sampleof the k-spaceGreen'sfunction
errorsencounteredin goingfrom the continuousanalytic will yield good results.This expectationis also consistent
expressionsto discreteandfiniteforms. withtherelationship
•,r•,(•)(z) togD,N(Z) developed
•'•D,N inSec.
II and summarizedby Eqs. (30) and ( 31). Sincemostof the
B. The Green's functions and their dependence on areaunder
thecurve
ofFig.1(b) iscontained
inareiatively
propagation distance small area around x- y = 0, the sum in Eq. (30), which
An important parameteraffectingthe nature of the quantifiesthe fictitious image source error, will be very
Green'sfunctionsis the propagationdistancez; the Green's small,andhenceg•.• (z) willberepresented
wellby,,•
•SD,N ß

functions,Eqs. (3) and (4), and their Fourier transforms, For small distancesz/A, conditions are not favorable for the
Eqs. (8) and (9), changesignificantlyas z variesfrom a direct samplingof the real-spacefunctiondue to the singu-
fractionof a wavelengthto severalwavelengths. This parti- larity at x -- y -- z -- 0. Insteadof a direct samplingof the
cular rangeof propagationdistanceis of great interestin real-spacefunction, a more sophisticatedapproximationof
NAH applications.
zIn thissubsection,
plotsoftheDirichlet the integration in Eq. (1•) is needed, such as
and Neumann Green's functions and their Fourier trans- g(D
2)(Z) or g•)(z).
formsarepresented for propagationdistances of a fractionof Figure 1(c) and (d) represents
plots of the analytic
a wavelengthand severalwavelengths. transformandreal-spaceform of the Dirichlet Green'sfunc-
The analyticFouriertransformof the Dirichlet Green's tion for z greaterthan threewavelengths(z/A = 3.1). Since
function,Eq. (8), is plottedin Fig. 1(a) for the smalldis- the analytic transform representsthe real-spaceGreen's
tancez/A = 0.13 and extendingover a typicalrangeof kx functionof infinite extent,this function doesnot fall off rap-
withky-- 0. Theslowlyvarying
natureof thefunction
al- idly in real space,and thisshowsup in k spaceasthe rapid
lows for easysamplingwith a uniform spacingof points oscillations of Fig. 1(c), whichare very difficultto sample
directlyasin g•) (z). TheGreen'sfunctionfor thisdistance
The real-spacefunction,Eq. (3), for the samedistance mightberepresented by theintegratedaveragek-spaceform
z/A isplottedin Fig. 1(b). Sincethe sourcefieldsconsidered g•>(z); however,the slowvariationevidentin Fig. l(d)
in this articleare assumedto be negligibleoutsidea region suggeststhat a directsamplingof thereal-spacefunctionwill
length L, the Green's functionsin real spaceneed only be be an adequaterepresentation of the propagator.Usingone
consideredin the range [ --L,L]. However, in contrastto of the real-spaceGreen'sfunctionforms,g(•>(z) through
the k-spaceform, the spikedshapeof the real-spacefunction g•4>(z), alongwithaugmenting
thedatafieldwithzerosasin
makesit difficult to samplewith a finite numberof points Eq. (16), will eliminatethe fictitiousimagesourceerror.

i i I 1

1.0 _

1.0 -

z - ß z/X=0.13 (b)

0.5 _

o-

I
2•
0.0 _

i i i i i i i i i
-5.0 -2.5 0.0 2.5 5.0 -2.0 0.0 2.0 4.0
FIG. 1. The real part of the Dirichlet
wovenumber dz sp 1acemerit pressureGreen's function plotted over
typicalranges(a) in k spaceand (b) in
I I ! I I
real spacefor small propagationdis-
1.0
_ _

1.0
tance,and (c) plottedin k spaceand (d)
z/X=3.1 (c) in real spacefor more than three wave-
0.5
_

0.5 lengthspropagationdistance.

0.0- 0.0

o_-0.5 - -0.5

i I I I I c-l.O • • •
-5.0 -2.5 0.0 2.5 5.0 -4.0 -2.0 0.0 2.0 4.0

wovenumber (k•/k) dz sp 1ocement

1313 J. Acoust. Soc. Am., Vol. 81, No. 5, May 1987 W.A. Veronesi and J. D. Maynard: Nearfield acousticholography 1313
The analytic transformof the Neumann Green'sfunc- tion, reconstruction,and inversetransformation.The com-
tion, Eq. (9), is plottedfor smallz/A in Fig. 2 (a), whilethe parisonsspan a suitable range of separationdistance
real-spacefunction,Eq. (4), is plottedin Fig. 2 (b). Unlike betweenthe input and output fieldsfor eachcase.To com-
the well-behavedfunctionplottedin Fig. 1(a), the singular- parethe sixforms,readilycalculated fields½ø(x,y,z)and
ity of the k-spaceNeumannGreen'sfunctionmakesit un- (8½ø/Sz)(x,y,z) of a tractable,theoretical
boundarycondi-
suitablefor directsampling
asg•> (z). Theintegrated
aver- tion ½$(x,y) are used.Thesefieldsare usedas simulated
ageg•> (z) isnecessary
to sampletheGreen'sfunctionfrom inputfields,or holograms,
andasabsolute reference
fieldsin
k space.While the k-spacefunctionis singularon the radi- comparisonsof the six output fields. An output field
2 = k 2 for all distancesz/A, the real-
ationcirclek 2 + k y •( i)(x,y,z) or (•( i)/•Z ) (x,y,z) isassociated witheachsam-
spacefunction is singularonly at the point x = y = z = 0. pledGreen'sfunctiong(i) D,N (g), i = 1," .,6, used in the nu-
This singularitypresentsno computationalproblemif either mericalprocessing of the input data.
g($
2)(Z) org(•)(z) is usedto represent
gN(z)' Sincethe processing problemsthat havebeendiscussed
The k-spaceform of the Green'sfunctionfor the Neu- have direct relevanceto nearfieldacousticsholographyit is
mannboundaryconditionsat the largerdistancez_• 32, de- natural to test and comparethe differentGreen'sfunctions
pictedin Fig. 2 (c), showsthe samerapid oscillationsasthe _(i). (z) in simulatedexperiments
formsgD,• with an acoustic
Dirichlet version,and it is alsosingularon the radiationcir- pressurefield,•(x,y,z) = p(x,y,z). The Neumannboundary
cle.The integrated
formg($
2)(Z) mustbe usedif a k-space conditionand (•3•A3z)(x,y,z) for anyz arespecified
in terms
Green'sfunctionis used,but again,the slowvariationof the of thez component
of theparticlevelocitydefined
by'2
real-spaceform, Fig. 2(d), suggestsusingone of the real-
spaceformsg• )(z) throughg•) (z). Vz(x,y,z) = -- •p(x,y,z), (40)
The eight plotsof the Green'sfunctionspresentedare pck •z
representativeof all the casesstudied.They are intendedto
wherepc is the characteristic
acousticimpedance.
serveasa basisfor understandingthe resultsof the computa-
tional experimentspresentedbelow. B. A theoretically tractable boundary condition: The
baffled piston
IV. QUANTITATIVE COMPARISON OF THE SAMPLED
GREEN'S FUNCTIONS The computationallytractableand relativelynonsingu-
lar acousticpressurefieldof an oscillatingpistonin an infi-
A. Introduction
nite baffleis usedfor the NAH simulations.In particular,
The purposeof this sectionis to presenta quantitative the chosensourceis a baffledpistonof radiusa = 3.44/k and
comparisonof the effectiveness of the six sampledGreen's surfacevelocityamplitudeUo.This sourceradiatesinto a
functionforms,.D,N
,,(' • (Z) through,,(6)
6D,N (z) , when they are mediumwith a characteristic impedance pc.
usedin eachof the four processes: propagation,transforma- Calculatingthe theoreticalfield for eachfieldpoint re-

I I i œ I I I I

1.0 o 1.0

z/X=0.1 o z/X=0.1 3 (b)


c

• 0.5
0.5 _

o
13.13
I
2•

-0.5 FIG. 2. The real part of the Neumann


-2.5 0.0 2.5 5.0 -4.0 -2.0 0.0 2.0 4.0 Green'sfunctionnormalizedto the larg-
est value plotted and displayed over
wovenumber (k•/k) dz sp 1ocement typical ranges(a) in k spaceand (b) in
z/X=3.1 (d) real space for short propagation dis-
I I I I I I I I I I
tance,and (c) in k spaceand (d) in real
1.0-
• •.0
_

C
0
spacefor a few wavelengths propagation
distance.

o
z/X=S.1 (c)
• 0.5
c 0.5-
q_

o
0.0-
•-0o5
i

i
2•
-0.5 - L •.0
i
-5.0 -2.5 0.0 2.5 5.0 -4.0 -2.0 0.0 2.0 4.0

wovenumber d.i sp 1clcement (x/X)

1314 J. Acoust.Soc. Am., Vol. 81, No. 5, May 1987 W.A. Veronesiand J. D. Maynard:Nearfieldacousticholography 1314
quires the evaluationof a one-dimensional integral with
fixedlimits. From Eqs. (2)-( 5), the fieldsproducedby the
pistonare givenby
! 6__•0
63 AR(i)X(i)
2
= 2rrUo
Pø(x'Y•) •
R r dr dO (41)
+ (A•
'/(')X(i)
2J
• )1/2
- ) 192 . (48)

The results of each case are summarized in a chart such as


and
that of Fig. 3(a). A bar is drawnfor eachfunctiongD,s
•o (Z),
i = 1 to 6, with the center drawn a distance A from the zero
o
v• = •Uo•0
(x,y,z) 2rr
•0
kaz
•-• (1--ika)½ikR
r
drdO, axiswhilethe top andbottomare drawn a distance+ • and
(42) - • from the center.
where To further illustrate the effectivenessof each method,
severalplots are presentedthat displaya central crosssec-
R •=(x •+y•)+•--2(x •+y•)•/•cos0 +z •.
tion of eitherthe real or imaginarypart of the outputfield
Equations
(41) and(42) canbereduced
•3tothefollow- •olm • represented
bycrosses,
against
thecorresponding
refer-
ing one-dimensional
integrals: encefieldp• or v• represented
by a solidline.
pO
(x,y,z
) pcuo
2•
• e • E. Test results for wavefront propagation
In this section,chartsand plots are presentedwhich
( bcosO--
1 )dO+a(b)ei•Z
• (43) quantify and illustratethe effectiveness
of the sampled
Green'sfunctionformswhenusedto propagatea pressure
o(x,y,z)= Uo z e•
z/),=O.006 (o)
x (.1 0 ' L
0.08 - _

whereb 2 + (x2 + y2)/a 2and o


L
L

0,b>
1,
-0.08 - _

a(b)= •, b= l, ,

1 2 3 4 5 6
2•, b<l.
momp
1i nS Form
C. Sampling the theoretical field z/X=O. 13 (b)
By numericalintegrationsof Eqs. (43) or (44) with a 0.001
o(x,y,z) areevaluated
fixedvalueofz, thefieldspø(x,y,z) or vz o 0
at eachof the coordinatelocations(x•,Ym), definedin Eq.
(14), yieldinga 64 X 64 array. Eachelementof the array is a
complexnumberspecifying the fieldamplitudeandphaseat 1 2 3 4 5 6
the point (x•,Ym,Z).Elementsof thesearraysare written as
momp
1i nS Form
P•mor V•mwith thevalueof z takenfromcontext.
z/),=0.97 (c)
D. Quantifying the error in the output fields 0.01 --
-- __
--

In orderto evaluatethe relativeaccuracyof the different o 0


ß (i)
Green's
functions,
theoutput
fields
•bJ•processed
withg D,N m -0.01 -- _

are comparedwith the referencefieldsP•mor V•m,and an


averagedifferenceand standarddeviationare determined. 1 2 3 4 5 6

All quantitiesarenormalizedby thelargestmagnitudeof the momp


1i nS Form
referencefield in the array.
The differencein the real (Re) and imaginary (Im) z/X=3.1 (d)
0.1_ - _
partsat eachpoint are definedas
L
R (i) ß
him• [Re(•r•) -- Re(½•
m)]/max(l•P?m
[) (45)
and
-0.1 _ --

AI (i) .
• = [Im(O•) - Im(O?•) ]/max(lO?•I). (46) 1 2 3 4 5 6

The averagedifferenceis definedas somp1i nS Form


FIG. 3. The normalizederror [definedin Eqs.(47) and (48) ] producedby
-- (6=•0
A(i)--
63'R(O !'I(i)
//
• (A•m+ A•m) 8192,
and the standard derivation is
m=O
(47) eachof thesixGreen'sfunctionsamplingmethodswhenusedto propagate
the pressurefield progressively
largerz distances,(a), (b), (c), and (d),
froman inputtheoreticalbaffledpistonpressure field.

1315 J. Acoust.Soc.Am.,Vol.81, No.5, May 1987 W.A. VeronesiandJ. D. Maynard:Nearfieldacousticholography 1315


wavefrontand to transformand propagatea z componentof functionsampling
formg• )(z) throughg•) (z) yieldsimi-
the particlevelocity.As definedin Sec.I, theseprocessesare lar, goodresultswhilethe relativeerrorof the the k-space
the numericalevaluationsofEqs. ( 1la) and ( 1lb) for fixed formsgrows,with g•) givingthe largesterror;thislarge
z- Zo> 0. Data for four differentvaluesof (z- Zo)/A are error is due to unattenuatedfictitiousimage sources.
presented[to simplifynotationin the figures(z- Zo)/A is The chartsof Fig. 3 suggestthat the shifted,integrated
written z/A ]. real-space
Green'sfunctiong•)(z) is thebestchoicefor a
The chart in Fig. 3 (a) graphicallydisplaysthe average singlefunctionto yield good resultsover the range of
differences•(i) andstandard
deviations
tr(i) of theoutput (z -- Zo)considered
here.The plotsof Figs.4 and5 further
fieldsPl(•
(z = 0.0062)fromthefieldp•m
(z = 0.0062)when illustratethe deficienciesof the other functionsin compari-
theinputfieldisthe pressureat thetheoreticalpistonsource sonwiththemeritsinherenting•) (z). In Fig.4 (a) a central
boundary, p•,,(z = 0). The k-spaceGreen'sfunction,Eq. cross sectionof the imaginary part of the output field
( 8), iswellbehaved
forsmail(z -- Zo)andsorequires
onlya p(3)(z
lm =0.006/l) is plotted with the imaginarypart of
simple representation;
Fig.(3a) shows thatg•) (z - Zo)and P•m(z = 0.006/l),andthesame isdoneforpt(,
4)(z -- 0.0062)
g•) (z - Zo)doyieldgoodresults forverysmallpropagation in Fig. 4 ( b ). In bothcasesthepropagationdistance(z -- Zo)
distance.The real-spaceGreen'sfunction,Eq. (3) is nearly was0.006/l.Figure4(a) shows thatpltm
3>underestimates the
singularfor small (z- Zo) and so requiresa sophisticated imaginary partofthefieldwhileFig.4(b) shows thatp•m
4>isa
representation;
Fig. 3(a) showsthat,of thereal-space
forms, muchbetterapproximation.As a first-orderapproximation
g•) (z- Zo)yieldsthe bestresultsfor verysmallpropaga- of•;D(Z),G• >(z) apparently
cannot
represent
thebehavior
tion distances. of the nearly singularreal-spaceGreen'sfunctionat very
The input field for the chart of Fig. 3(b) is again smalldistances.
The higher-order
approximation
G •)(z)
p• (Zo= 0) whilethesixoutputfieldsarep• (z = 0.13•). doesappearto representthe smalleffectof this behavior
At this distance,all six functionsyield goodresults. well.
For the chart in Fig. 3(c), the input field is Forpropagating overdistances
ofa wavelength
ormore
(Zo-- 0.13) andtheoutputfieldsarep/(•(z = 1.M). At this theeffectsof imagesources
shoulddegradetheperformance
distancethe error from any of the real-spaceformsis about ofthek-space
formsg•>(z) andg•) (z). Figure5(a) clearly
one-thirdthat of eitherk-spaceform. Thereis little differ- showsan interferenceeffectfrom theseimagesourcesin the
ence between the errors from the IOtll
•--- t •i:11-•pi:ll,•C:
-' •1011113.
.... 111 plot of p}•)(z= 3.2x) againstp•; (z = 3.2A), where
general,at distances(z- zo) •>/1,/2,the real-space
forms, (z- Zo)was3.1/l. The integratedreal-space formdoesnot
g• >(z) throughg•) (z), providebetterresultsthanthe k- sufferfrom a fictitiousimagesourceerror ascanbe seenin
spaceforms. Fig.5(b), wherep•
4)(z = 3.2/l) iscompared
withp•m
forthe
For a propagation distance
of severalwavelengths,the sameconditionsas in Fig. 5 (a).
chartin Fig. 3(d) indicates
thatall of thereal-space
Oreen's The chartsandplotsof Figs.3-5 comparetheutilityof
thesixGreen's
function
sampling
methods
whenusedinthe

0.4 •.4 I I I l_

0.2 • o.2

0 - g o
L

ø--0-2
•• • • • %0.2
-2.0 -1.0 0.0 1.0 2.0 -2.'0 - 1.•0 O.'0 1.'0 2.'0
di sp 1ocement (x/X) disp locememt (x/X)

I I I I I
I I I I
0.4 _ _

0.4 _ _

(b)

o
t_

•-0.2 - I I I I I
- , -0.2 -- _

-2.0 -1.0 0.0 1.0 2.0 I


- 1.0 i
0.0 1.I0 2.'0
disp1 ocom•mt (x/X) di $p 1cloomornt
FIG. 4. The imaginarypart of the pressure0.006 wavelengthsfrom the FIG. 5. The realpart of the pressure
3.12lfrom the planeof a theoretical
planeof a theoreticalbaffledpistonsourcegenerated
fromaninputpressure baffledpistonsourcegenerated froman inputpressure fieldsampletaken
fieldsample
takenatthesource
using(a): plussigns,
g(o3)(z),
and(b): plus 0.1321fromthesource using(a)' plussigns,
g•)(z), and (b)' plussigns,
signs,g•)(z) againstthetheoretical
field,solidline. g•(z). Theoryis thesolidline.

1316 J. Acoust. Soc. Am., Vol. 81, No. 5, May 1987 W.A. Veronesi and J. D. Maynard: Nearfield acousticholography 1316
numericalevaluationof Eq. ( 1la). The chartsand plotsof 0.4 _
I I ! I I
_

Figs.5-7 dothesametypeof comparison fornumericaleval-


uationsof Eq. (1 lb). The chart in Fig. 6(a) displaysthe
0.2
average
differences
•(i) andstandard
deviations
c(øof the
outputfieldspt•,•(z = 0) comparedto the referencefield L

p•,•(z = 0). The inputfieldis thenormalcomponent


of the •o
•o
surfacevelocityat the theoreticalpistonsourceboundary, L

V?m(Z0 = 0). Functions


g•) (z) andg•) (z) yieldthebestre- -0.2 _ _

sultsforthistransformation -(ø(z =0).


fromv?,•(z = 0) top•,• -2.'0 -1. 0.0 1.0 2.0
Unlikethefunctionsgg•(z), thereappears to benoval- di sp 1ocement (x/X)
ueof (z - z') whereall g• (z) producesimilarresults;this
canbeseenin Fig.6(a)-(d). Thefunction
g• (z) yieldsas 0.4 _
I I I I 1
_

goodresultsas any functionover the rangeof distances


(z-z') considered. The real-spaceforms g•(z) and
g• (z) areonlyslightlyinferiortog• (z) forusewhenthe
distance(z - Zo) issmall.Thereal-space
formg• •(z) starts L

out the worst of all six forms and steadilyimproveswith •o


•o
propagationdistance,whilethe k-spaceformsdecrease in L
performance with increasing propagationdistance. -0.2
Figure8(a) and (b) compares
withp• (z = 0) the -2.0
' -1.0
• O.lO 1.'0 2.'0
displacement (x/X)
FIG. 7. Therealpartof thepressure field3.22 fromtheinputbaffledpiston
z/X=O.O (o) z component of velocityat the sourceplaneas producedusing(a): plus
0.05 _ - - signs,
g•)(z), and(b)'plus signs, g•4)(z).Thesolidlinesplottheory.

0
BOo fieldsp•)(z = 0) andpt(m 4)(z = 0), respectively,
asproduced
-0.05 ,_ _ - withg•) (z = 0) andg(N4) (z = 0). Againstp•),
(z = 3.22), the
1 2 3 4 5 6 fieldsl-•,(5)(z
lrn = 3.22) and,,('•)(z = 3.22) asproducedwith
somp1i nS form g•(z- 3.2A) and g•(z- 3.2A), are comparedin Fig.
7 (a) and (b), respectively.
In both Figs.7 ( a ) and 8( a ), the
z/X=O.06 (b) effects
offictitious
images
introduced
byg• •(z) areseen.By
0.03
L
o
L 2.0
L
w
-0.03

1 2 3 4 5 6

sampiinS form
0.5
z/X=O.97 (c)
0.05 _
i

I
•nO.0
L -2.0 -1.0 0.0 1.0 2.0
o
L
L di sp1 ocement (x/X)
w
-0.05
1 2 3 4 5 6
2.0
sOreR
1i n9 form - (b)
1.5
z/),=3.2 (d)
0.2
i

L
o
L
•o5
L
w
-0.2
-2.0 -1.0 0.0 1.0 2.0
1 2 3 4 5 6
c•isp1 ooement (x/X)
somp1i nS form FIG. 8. The real part of the pressurein the planeof a baffledpistonsource
FIG.6.Thenormalized
error
produced
b'y
each
ofthesixGreen's
function alonga centerrow of the output generatedfield array producedusing (a):
samplingmethodswhenusedto propagatethe pressurefield progressively plussigns,g•) (z), and(b)' plussigns,
g•) (z) withthez component of the
largerdistances,(a), (b), (c), and (d), from an input theoreticalz compo- particlevelocityin the sameplaneasinputto theprocessingroutines.The
nent of the particlevelocityboundaryfield. solidlinesrepresentthe theoreticalvalue.

1317 J. Acoust. Soc. Am., Vol. 81, No. 5, May 1987 W.A. Veronesi and J. D. Maynard: Nearfield acoustic holography 1317
z/X=O.06 (=) functionfor inversion.The suitabilityof thereciprocals
of all
0.3 _

L
six Green's function forms as representatives of
0
L
G - •(k,,,ky,z)inEq.( 38) wastested.
Theresults
areshown
L
mmm
in the chartsof Fig. 9 (a) and (b) for the caseof determining
-0.3 thepressure fieldsp• (Zo= 0.065A)andz component ofve-
2 3 4 5 locityfieldsVim
.i• (Zo• 0.065A)usingP?m(Z -- 0.13A)asthe
•omp1i n9 •or'm inputfield.The integrated k-space
form•SD,N
,,•6• (z) is notin-
cludedin thesechartsasthe errorsproducedusingthisform
z/X=O.06 (•) are solargethat scalingthe figuresto accommodate them
4 obscuresany informationabout the others.The sampledk-
sp•eGreen's
function
,,•5•(z)- • isthebestrepresentative
•SD,N

ofG - •(kx,ky•z) in allreconstruction


cases
considered.
As discussedin Sec. II, someform of filteringmust be
2 3 4 5
includedin the reconstructionprocessusingDFTs. This
needfor filtering,theformof thefilterfunction,anditseffect
•omplin S •o•m on a numericalreconstructionare illustratedin Fig. 10.The
FIG. 9. The normalized error encountered with each method in the source error from the high spatialfrequencyapproximationasso-
fieldreconstructionof (a) the pressurefield0.065from the sourcefrom an ciatedwith the DFT andits amplificationby the reconstruc-
inputsamplearrayof thetheoretical pressure field0.13fromthesource,and tion process is apparentif the DFT of an unfilteredsolution,
(b) thez component of theparticlevelocityfield0.065fromthesourcefrom
the pressurefield0.132 from the source. p•5)(z) isplottedwiththeDFT ofp• (z) In Fig. 10(a) the
crossesindicatethe amplitudeof an unfilteredDFT of the
pressure
field•lm
• (z = 0.065A)asprocessed fromtheinput
contrastbothFigs.7 (b) and 8(b) showthe goodagreement
pressure
fieldp• (z = 0.13A).The solidlineplotstheDFT
obtained
withgff)(z).
ofp• (Zo= 0.065A).The differencebetweenthetwo plots
Over the full rangeof distances(z -- Zo)consideredand
clearlygrowsfor largermagnitudesof spatialfrequency.In
for bothpropagationproblemsandproblemsof transforma-
realspace,
thishigh-frequency
errorshows
upaslargeoscil-
tion and propagation,the shifted integrated real-space lations of the numerical solution about the actual solution.
Green'sfunctions ,•(4) (z) appearto yieldconsistently
6D,N good
This behavioris evidentin Fig. 10(b) which comparesthe
results.This consistency makesthis form idealfor computa-
tion.
unfilteredFlm
.(1) (Z0 • 0.0652) with P•m(2o• 0.0652)' The
numericalsolutionis vastly improvedby a multiplication
F. Reconstruction and inverse transformation test with the filter function,Eq. (39). This functionis plottedin
results Fig. 10(c) for k• = 0.6 N•/L and for threedifferentvalues
For reconstructionand inversionone usesthe k-space of a. Valuesof k• = 0.6 N•/L anda = 0.2 havebeenfound
formulationin Eq. (38) sincethereis no real-spaceGreen's empiricallyto givegoodresults,andthesevaluesareusedin

400 2.0

- (b) -
300

200

100
_ + + + + _
+ ++ ++ +
FIG. 10. (a) Plussigns,amplitudeof the
++ ++ --
0 DFT of a sampleof the theoreticalpres-
-8.0 -1.0 0.0 1.0 2.0 sure field 0.132 from the source multi-
pliedbytheG g- 1)(Z) toreconstruct
the
dl sp 1ocernemt pressureof the source.The solid line
plotsthe theoreticalvalues.(b) The un-
filteredfieldagainsttheory;(c) the filter
2.0


o 0 c)
I I I I I_ function for three values of a; and (d)
the filtered output usinga = 0.2 filter.
(d) -

• 0.5 o +

0.0 , , , , •
-8.0 -4.0 0.0 4.0 8.0 -1.0 0.0 1.0 2.0

wavenumber" (k./k) dl sp 1ocernemt

1318 J. Acoust. Soc. Am., Vol. 81, No. 5, May 1987 W.A. Veronesiand J. D. Maynard' Nearfield acoustic holography 1318
I I I i i

1.0
finiteregion;and, second,thisfieldcanbe adequatelyrepre-
- (o)
sentedby a patchwiseconstantfield for reasonablysmall
patches.If thesetwo assumptions are satisfied,it wasshown
that the problemwasto representthe known Green'sfunc-
0.5 tion in a discreteform and six methodsfor accomplishing
thisweredeveloped.A qualitativeexaminationof the actual
form of the Green'sfunctionwasgivenin Sec.III. This was
done to identify the difficultiesin representingthe Green's
-2.0 -1.0 0.0 1.0 2.0 functionsand to identify how these difficultiesmight be
d• sp ] acement (x/X) overcomeby one or more of the six methods.The resultsof
testingpresentedin Sec.IV E suggestthe useof the shifted
I I

+
I

+
I I
integrated real-spaceGreen's function, (4) (Z), for acoustic
gD,N
1.0 • - propagationproblems.The resultsshowthisform produced
(b)
a minimum of error for propagationdistancesfrom zero to
severalwavelengths.
0.5
For reconstructions,the discussionof Sec. II C and the
testresultspresentedin Sec.IV F indicatethat the reciprocal
ofthesampled
k-space
Green'sfunction (5) (z) witha spa-
gD,N
tial frequencyfilter is the bestinversepropagator.
0.0
-2.0 -1.O 0.0 1.O 2.0

d• sp } ooemernt
APPENDIX A: DISCRETE CONVOLUTION WITH THE
FIG. 11. (a) Real part of the reconstructedz componentof the particle DFT
velocityfieldat 0.062 fromthesourcestartingfromthe pressure0.132 from
the source.(b) Real part of the z componentof the particlevelocityas re- Discreteconvolutions,suchas Eq. (16) of Sec.II, can
constructedin the sourceplanefrom the input sourceplanepressure.The
beperformedexactlyandquicklywith theaid of the FFT to
reciprocal
ofg• >(z) withfilteringwasusedin both.In both,thesolidlineis
theory.
performDFT operations.
To introduce the DFT and IDFT into Eq. (16), Eq.
(16) is rewritten with the aid of the Kronecker delta func-
all filtered reconstructionsin this study;theseinclude those tion,
which providedata for Fig. 9 (a) and (b) in this study.Fig-
ure 10(d) shows the excellent fit of the filtered and actual
solutions.
=[1,
•rnn 0, m=n,
The result is
As further illustrationof the capabilitiesof the sampled
N--1N--1 N--1 N--1
k-spaceGreen'sfunctionwith filtering,Fig. 11(a) showsa
comparison plotof therealpartof -(•)
"'bn (Zo= 0.0652), where ½(v,q,z)=
E Z l=0 m=0 r=
E --Ns=
Z --N
½D,
$(l'm)
theinputfieldiSp•m (Z = 0.132), againstV•n(Z0= 0.0652);
Fig. 11(b) showsthe real part of, (•)(zo 0) wherethe X GD,N(r,s,z)CSr,
p_ lCSs,
q_m' (A1)
inputfieldisp• (z -- 0), againstv• (zo-- 0). Bothfigures The delta function can be written in terms of the Fourier

showgoodagreementbetweenthe numericaland actualso- coefficients as

lutions.Theseplotsalongwith Figs.9 and 10 indicatethat


= e -- (irr/N)tz(rn
-- n).
accurate solutions to reconstruction and inverse transforma-
tion problemsare attainableusing the sampledk-space
(•rn
n
Green'sfunction,S•,N
•(•) (z)-' , andthe filterfunctionof Eq. Substitutingthis expansionfor the delta functionsin Eq.
(39).
(A 1) and rearrangingthe summationsleave

V. SUMMARY AND CONCLUSIONS ½( p,q,z) =--


This study had two main objectives.The first was to
developand summarizethe theorybehindthe reductionof
the infiniteand continuousconvolutionintegralsof nearfield
acousticholography,as presentedin Sec.I, to a finite and
discreteform suitablefor high-speednumerical computa-
tion andto developcomputationaltechniques basedon this
theory;the secondgoalwasto evaluatethe levelof accuracy
-- t
possiblewith thesetechniques. Usingthefunctionslpl•,N andG D,Nasdefinedin Eqs. (17)
SectionII developedthe theoryfor approximatingthe and (21 ), the bracketed double sumsin (A2) can be rewrit-
infinite and continuousintegrals with finite and discrete ten as DFTs. By making use of the identity exp(irrrF/
forms.The approximationwas basedon two main assump- N) = exp[irr(4 -- 2N)F/N] to arrangethe secondbracket-
tions.First, the boundaryfield can be neglectedoutsideof a ed sum, Eq. (A2) becomes

1319 J. Acoust. Soc. Am., Vol. 81, No. 5, May 1987 W.A. Veronesi and J. D. Maynard: Nearfield acoustic holography 1319
1 2N-- 1 2N-- 1 givenin Eq. (B 1) exceptwhenthe point (x = 0,y-- 0) is
=•M2•=o
•b(p,q,z) Z v=o
Z ei(rr/N)(btP+vq) includedin the integrationpatchindicatedby l andm. Ap,
propdatevaluesfor XoandYoaregivenbyxt andYmof Eq.
(13) for•D,N
,.(2) (z)andby Xl'=Xl + A/2andy•, =Ym + A/2
X\ l=•O ½•,N
(l,m)e
- i(•/N)(lM
+m•) for 6D,
,,(4)
N (z) '

The point (x = 0,y = 0) is centeredin the integration


X GD.N(r,s,z)e i(=/•>(r•
+,•> .
)
(A3)
patchof• ,N(l • 0,m• 0,z)for6D,N
approximation
"(2) (z)' In thiscase,
an
for GD,N(l = 0,m= 0,z) is the sumof the
The summations
in.Eq. (A3) cannow be replacedby the appropriateintegralfrom Eqs. (B3) and (B4) plusfour
times (oncefor eachcomer of the patch) Eq. (B2), with
D• andID• operations
asdefined
in Eqs.(18a)and
R = A/2 andwith Xo= Yo= A/2 (to determinethe coeffi-
(18b). The final fern with thesesubstitutionsis given by
cientsa• ).
Eqs. (19) and (20).
Whencalculating "(4)
•SD,N (z) basedon Eq. (19) andthe
shiftedsamplingcoordinates x$ and y;,, the four integra-
tions,whichyieldGD,N(I,m,z)for (l,m) = (0,0), ( -- 1,0),
(0,- 1), (- 1,- 1), havethe point (x = 0,y = 0) at one
Integrations,
overrectangular patches,
of thereal-space cornerof their integrationpatches.Here G•,N(l,m,z) for
Green'sfunctionare approximated by a truncatedTaylor eachof thesefour indexpairsis givenby one-fourththe ap-
seriesexpansionof the functionabouteachpatch'scenter propriateintegralfromEqs.(B3) and (B4) plusthecorner
(xo,Yo)'
integration,Eq. (B2), with R = A andwithXo= Yo= A.
G(xo + x,yo+ y• - Zo) The specialform•SD,N
"(•) (z) is calculated fromEq. (19)
=f (x,y) with z = Zo const with GD,N(l,m,z) asdescribed
exceptwhenl -- m = 0. For
l = m = 0, the appropriateintegral,Dirichlet or Neumann,
fromEqs.(B2) and(B4) isusedwithR = (h2/7T)
1/2.
+ a•2y + a7xy2+ aaX2y
2, The coefficients/3•are

wherethea• aregivenbelow.Integrationof thisformover /30= 2.1460184X10-',


the patchyields
}•1= 4.793517•10-2,
)•./2f•y/2
&x/2
f (x,y)dx
.
• -- &y/2
/33= 6.2685032
X 10-3,
/34= 1.8252296
X 10-2,
=ao• &y+ (a4/12)(•)3&Y + (as/12)•(&Y) 3
/36= 1.5856291
• 10-3,
+ (a8/144)(•)3(&y) 3. (B1)
/38= 2.889426
X 10-4.
Alongthez axis,Xo= Yo= 0, a differentapproach is
used:The exactintegralovera circularregionaroundthe The expansion
polynomialcoefficients
a i (xo,Yo):
axisis performed, andthe integration overthe remaining a = kRo, b= kxo, d-- kyo, k =co/c,
region,to complete a square,
isapproximated byexpanding
aboutthesquare's corner.In termsof theexpansioncoeffi- + (z)
cienta•, a singlecornerintegration
yields
Velocity to pressure:

•a•a f(x',y'
)dx'
.'
R 2 _ x2) •/2
ao= - i( pck2/2//')(e•/a),
al = - i( pck3/2rr)b(ia-2 - a-3)eia,
= a• • + (a, + a=)•,R 3+ a•3R 4
Ot
2= -- i( pck3/2rr)d(ia-2 -- a-3)eia,
(B2) a3= - i( pck4/2rr)bd(- a-3 - 3ia-4 + 3a-5)ei•,
wherex,y, arefrom the centerof the square,whilex', y' are a4= - i( pck4/4rr)[ia-2 _ a-3
from oneof its corners.The exactintegrationover the cen-
tral circular regionis + b 2( _ a-3 _ 3ia-4 + 3a-5) ]ei•,
a5= - i( pck4/4rr)[ia-2 - a-3
•2=••G(r,zo)rdrdO + d2( -- a-3 _ 3ia-4 q- 3a-5) ]eia,
velocityto pressure(Neumannboundau conditions) Ot
6= -- i( pck5/4)d [ -- a-3 -- 3ia-4 or-3a-5
=pc(eii(•)-- e•i[•' + (•)=•"=) (B3) + b 2( _ ia-4 + 6a-5 + 15/a-6 _ 15a-7) ]ei•,
pressureto pressure(Difichlet boundau conditions)
a7= -- i( pck5/4rr)b[ -- a-3 -- 3ia-4 ør-3a-5
=e•i(z)
_ (g--gn)ein[•=+(z)=]
1/= q- c2( -- ia-4 q- 0a-5 q- 15ia-6-- 15a-7) ]ei•,
. [R2+(z)2]•/2' (B4) a8= -- i( pck6/8rr)[ ( -- a-3 -- 3ia-4 q-3a-5)
To calculate ,(2)
,•,$ 0(4) (Z) from •.
(Z) or *•,$ (19),
G •,$ (l,m•) isapproximatedby a polynomialexpansion
as + (b 2 + d 2) ( _ ia-4 q- 0a-5 q- 15ia-6

1320 J. Acoust. Sec. Am., Vol. 81, No. 5, May 1987 W.A. Veronesi and J. D. Maynard: Nearfield acoustic holography 1320
-- 15a-7) q- b 2d2(a-5 q- 10ia-6 - 45a-7 k2= (k2
r, -- k2)1/2= - i2pck
, whenZ1= 0,
(kl + k2)
_ 105ia-8 q- 105a-9) ]ei". Pressureto pressure(Dirichlet boundaryconditions)'
Pressureto pressure:
Radiation region
ao = [k 3(g)/2•'] (a -3 -- ia-2)e ia, ( 1 -- iklz)eik'z-- ( 1 -- ik2z)d
( =
al = [k 4(z)/2rr]b(a-3 q- 3ia-4-- 3a-5)eia, (k•z 2 -- k 2)
at2= [k 4(z)/2rr]d(a-3 q- 3ia-4- 3a-5)eia, 2)•/2, k2=(k 2-- 2 )1/2
kl= (k 2_ kr,
l•3 = [kS(z)/2rr]bd(ia-4 - 6a-5 _ 15ia-6 Mixed evanescentand propagatingregion

+ 15a-7)d •, ( 1 -- iklZ) (e•'• -- ( 1 q-k2z)e-


=
•t4 = [kS(z)/4rr] [ (a -3 + 3ia-4 _ 3a-5) (kl•z2 q-k•z 2)
kl= (k2-kr,2 )2/2, k = (k 2r2-- k2) 1/2
+ b 2(ia-4 -- 6a-5 -- 15/a-6 q- 15a-7) ]d•,
Evanescentregion
a5 = [kS(z)/4rr] [ (a -3 + 3ia-4 -- 3a-5)
( 1 + klz)e- •,,z_ ( 1 + k2z)e-- k2z

+ d 2(ia-4 -- 6a-5 -- 15/a-6 q- 15a-7) ]e% (L ,z) =


(k ]z2 + k •z2)
a6 = [k 6(z)/4rr] [d(ia-4 -- 6a-5 -- 15ia-6 q- 15a-7) 2 • k2)1/2.
kl=(k 2
rI -k2) 1/2
• k2• (kr•
+ b 2d( -- a -5 -- 10ia-6 q- 45a-7 q- 105ia-8
-- 105a-9) ]ei", APPENDIX C: DFT ERROR IN APPROXIMATING THE
CONTINUOUS FOURIER TRANSFORM
a7= [k6(z)/4rr][b(ia-4 - 6el
-5 _ 15ia
-6 q-15a-7)
The two-dimensionaltransformof a functionf (x,y) is
q-bd2( _ a-5 _ 10ia-6 q-45a-7 q- 105ia-8
givenexplicitly by
_ 105a-9) ]eia,
as= [k 7(z)/8rr][ (ia-4 - 6a-5 - 15ia-6 + 15a-7) F(k•,ky) = f (x,y)e •'•dxdy. (C1)
+ 2bd( -a -5 -- 10ia-6 q-45a-7 q- 105ia-8
It is assumedthat the functionf (x,y) is reasonablysmooth
_ 105a-9) q-b2d2( _ ia-6q- 15a-7 q- 105ia-8 andis knownat a sufficientlylargenumberof points(xr,ys)
so that one can write
_ 420a-9_ 945ia-lo+ 945a--11)]½ia.
The integratedk-spaceGreen'sfunction: f (x,y) =f (Xr,Ys) if X-Xrl<A/2 and ly-y•l<A/2,
The integrated
Green'sfunctions areobtainedby per- (C2)
formingan integrationaveragingof the analyticGreen's whereXr = rA andys = sA for r, s = 0, __+1, __+
2,....
functionin the area betweenkr,andkr•_,kr2-(k2 Equation (C 1) can then be written as
+ ky2)1/2+ Am
r andkr, -- (k 2 q-ky2)1/2__
x Akr In this A

study,Akr= x/•Ak,•wasused. F(k•,ky) -- Z Z f (Xr•Ys)


Thez velocityto pressure
(Neumannboundarycondi-
tions)'
X e dxdy. (C3)
Radiationregion dx r -- A/2 dy s -- A/2

The integralin Eq. (C3) is readilyevaluatedyielding


g•)(k•,ky,z)
- 2ipckz(d•'• - ei•'z)
( k • z• - k 2•z•)
,
'• -- i( kxxr q-my)
kl __(k2--kr,2 )1/2• F(k•,ky) = • •f (xr,y•)e

k2: (k2-- k2
r2)1/2
-- 2pck , whenzI = O.
(kl = k2) X k• 2 • 2 ' (C4)
Mixed evanescent
andpropagatingregion The discreteFourier transformoff (Xr•Ys) can be written
explicitly as
g(•)
(kx,ky,z)
= 2pckzl
( --ie'•"z'
+
+ie- •,2z')
,
kl = (k 2-- k 2 )1/2
•l(kx'ky
) = Z Z f (xr•ys)½-i(k•'xr+ky'vOA2.
(C5)

k2_ (mr
• __k2)1/2
= 2pck(kl-
ik2) when
z'-0. By dividingEq. (C5) by Eq. (C4), an estimateis obtained
Evanescentregion
form of an approximately piecewise constant function
f (x,y) by an appropriatediscretetransform.The resultis
g(N2)
(k,,,k.v,Z
l)= - i2pckz(
e- •:,"
_e- •:•
,') .
2 -- k2)1/2,
kl-• (kr• •'(k•
,ky
) 4sin
(kx
A/2)
sin(ky
A/2) (C6)
1321 J. Acoust.
Soc.Am.,Vo1.,81,
No.5, May1987 W.A.Veronesi
andJ. D. Maynard'
Nearfield
acoustic
holography 1321
•E. G. Williams and J. D. Maynard, Phys.Rev. Lett. 45, 554 (1980). 8E.O. Brigham,
TheFastFourierTransform
(Prentice-Hall,
Englewood
2j. D. Maynard,E.G. Williams,andY. Lee,J. Acoust.Soc.Am. 78, 1395 Cliffs, NJ, 1974).
(1985). 9j. W. Goodman,
Introduction
to FourierOptics(McGraw-Hill,New
3E.G. WilliamsandJ. D. Maynard, J. Acoust.Soc.Am. 72, 2020 (1982). York, 1968), p. 10.
4p. M. Morse and H. Feshbach,Methodsof TheoreticalPhysics(McGraw- •øI.S.Gradshteyn
andI. M. Ryzhik,Tableoflntegrals,
Series,
andProducts
Hill, New York, 1978), Vol. II, p. 1331. (Academic,New York, 1980), p. 736.
5p.M. MorseandK. U. Ingard,TheoreticalAcoustics(McGraw-Hill, New •W.A. Veronesi
andJ.D. Maynard,tobesubmitted
toJ.Acoust.Soc.Am.
York, 1968). •2L.E. KinslerandA. R. Frey,Fundamentals
ofAcoustics
(Wiley,New
6M.C. JungerandD. Feit,Sound,Structures,
andTheirIn teraction( MIT, York, 1962), p. 157 (note time dependence
difference).
Cambridge, 1972). 13j.h. Archer-Hall,A. I. Bashter,and A. J. Hazelwood,J. Acoust.Soc.
7LordRayleigh,Philos.Mag. 43, 259 (1897). Am. 65, 1568 (1979).

1322 J. Acoust.Soc. Am., Vol. 81, No. 5, May 1987 W.A. Veronesiand J. D. Maynard:Nearfieldacousticholography 1322

You might also like