Extension of the angular spectrum approach to curved radiators

Ping Wu and Tadeusz Stepinski

Uppsala University, Department of Materials Science, Signals and Systems, Box 528, S-751 20 Uppsala,
Sweden

~Received 22 September 1998; accepted for publication 7 January 1999!

The angular spectrum approach ~ASA! is conventionally applied to the evaluation of acoustic fields
from planar radiators because it is usually based on the 2-D Fourier transform ~or the zero-order
Hankel transform in the axisymmetrical case! which is implemented only in a plane. The present
paper is intended to extend the ASA to more general cases where radiators have curved surfaces. For
this purpose, two approaches are developed. The first one is the extended ASA and is derived in a
general way. From this approach, the angular spectrum of a curved radiator is given by a double
integral that does not take the 2-D Fourier transform form, and thus cannot be implemented using
2-D fast Fourier transform ~FFT! but by numerical integration. The second approach is the indirect
ASA that gives the angular spectrum via 2-D Fourier transforming an initial field pre-calculated in
a plane. The method for calculating the initial field is proposed based on the method developed by
Ocheltree and Frizzell for planar sources. An example is given of a linear array with a cylindrically
concave surface, and in this case, the angular spectrum ~the double integral! from the extended ASA
reduces to a single integral. The angular spectra of the array are calculated using both approaches
and compared. The comparison has shown that the angular spectra obtained from both approaches
are in excellent agreement. The accuracy and efficiency of the two approaches are studied in the
numerical implementation. In this example, the extended ASA has been shown to be more efficient
and more accurate than the latter approach. Both approaches can be applied to arbitrarily curved
transducers. In the general case where the double integral cannot be reduced to a single integral, the
latter approach can be more efficient. © 1999 Acoustical Society of America.
@S0001-4966~99!01304-1#
PACS numbers: 43.20.Rz, 43.20.Bi @DEC#

INTRODUCTION in our present interest. Concave transducers and arrays with

concave surfaces have become more and more common
The angular spectrum approach ~ASA! has found wide nowadays since piezocomposite materials have been avail-
applications in modeling wave propagating fields in optics,1,2 able. Among the most commonly used are spherically con-
electromagnetics,3 acoustics and ultrasonics.4–18 The ASA is cave and cylindrically concave transducers, as well as vari-
usually implemented based on the two-dimensional ~2-D! ous ultrasonic arrays which have either linearly, or
spatial Fourier transform, or Hankel transform in the axisym- cylindrically, or curvilinearly aligned elements. The arrays,
metrical case. It is well known that the 2-D Fourier transform in addition to performing electronic focusing in azimuth, of-
~or the Hankel transform! is implemented in a plane. Thus ten have cylindrically curved surfaces for geometrical focus-
the ASA is usually restricted to dealing with the radiation of ing in elevation. Acoustic fields from curved transducers
planar radiators,1–8 and the backpropagation of ultrasonic
have been studied extensively by means of different meth-
fields in planes.9,10 In this paper, we extend the ASA to more
ods, but most of the methods were used to treat spherically
general cases where the radiators are curved. The motivation
concave transducers11–14,19–26 and focused transducers with
of the present work is our intention to use the ASA to simu-
arbitrary axisymmetric surface velocity distributions.27 The
late elastic fields in immersed solids ~e.g., copper! radiated
ASA is one of the methods used, modeling spherically
by a linear array that we have in our laboratory and use in
immersion test15 because this approach is easily applied to curved transducers, however, mostly in the applications of
the immersion case.16 The main target of the paper is the acoustic microscopy, e.g., in Refs. 11–14, where the angular
angular spectra of curved radiators because an angular spec- spectra were obtained only in the focal plane. Moreover, due
trum of a source is the basis that is necessary for synthesiz- to axisymmetry of the transducers, the 2-D problem could be
ing the elastic fields radiated by the source into an immersed simplified into the 1-D problem. However, it is not possible
solid. for curved transducers that are not axisymmetrical, e.g., cy-
Ultrasonic beam focusing is commonly employed to at- lindrically concave transducers and linear or phased arrays
tain fine spatial resolution and is usually realized by means with cylindrically curved surfaces. The ASA has also been
of either curved transducers or planar transducers coupled to used to the transducers coupled to focusing lenses. Two rel-
focusing lenses. A large body of literature is available deal- evant papers by Guyomar and Powers,17,18 are worth men-
ing with the radiation of the focused transducers of both tioning here. In these two papers, they studied and extended
types. Our attention is focused on transducers of the first the angular spectrum approach to the case of transient fields
type, i.e., transducers with curved surfaces, because they are from curved transducers. It should be pointed out that the

2618 J. Acoust. Soc. Am. 105 (5), May 1999 0001-4966/99/105(5)/2618/10/$15.00 © 1999 Acoustical Society of America 2618

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 132.241.216.43 On: Mon, 15 Dec 2014 06:55:16
 extension made was the time extension of the angular spec- spherical waves, and in the limiting case where S is a plane
trum giving the time impulse response. The curved transduc- radiator with an infinite baffled plane, the equation is rigor-
ers were essentially treated as planar transducers having ously correct. If S is curved, waves radiated from any part of
time-delay excitations or, equivalently, coupled to lenses the curved surface are diffracted by other parts of this sur-
with various focusing surfaces such as spherical concave, face, but the integral in Eq. ~1! neglects this fact. When the
spherical convex, conical, parabolic, and cylindrical surfaces. surface S is only slightly curved, this secondary diffraction
The thin lens approximation was implicitly employed in the becomes relatively unimportant. The resultant effects due to
formulation. Thus the 2-D Fourier transform was applicable the secondary disturbances which originate at different parts
and used in their case. In our case the transducers and the of S will be further reduced by phase differences if the extent
linear arrays studied are cylindrically curved. To our knowl- of S is large relative to the wavelength. Also, with this re-
edge, little work has been reported exploring the ASA for the striction on the dimension, most of the energy will be radi-
calculation of acoustic fields from such curved transducers ated in a central beam, which will be affected very little by
and arrays. the presence or the absence of a baffle around the radiator.25
In the present work two angular spectrum approaches Hence, if the diameter or breadth of a slightly curved source
are developed to calculate the angular spectrum of a curved surface S is large compared with the wavelength, the veloc-
transducer. The first one is the extended ASA, that is, the ity potential at points in or near the main beam will be rep-
conventional ASA extended in space from the planar sources resented approximately by the integral in Eq. ~1!. This is
to the curved sources. The starting point is the Rayleigh in- often called O’Neil theory. This theory has been used by
tegral describing the radiation from a curved transducer, and many researchers dealing with curved transducers, such as
then a general form of the angular spectrum is derived that is Penttinen and Luukkala,20 Arditi et al.,21 Fink and
a double integral not possessing the 2-D Fourier transform Cardoso,25 Schmerr,26 etc. A number of experimental studies
form, which means the angular spectrum cannot be imple- ~e.g., Refs. 22 and 23! and numerical investigations24 have
mented using 2-D fast Fourier transform ~FFT! but by nu- demonstrated that the O’Neil theory agrees very well with
merical integration. The second approach is the one that in- experiments and with more exact models for curved trans-
directly gives the angular spectrum via 2-D Fourier ducers. An important conclusion from Coulouvrat’s work24
transforming an initial field pre-calculated in a plane, and is states that the conditions of validity of the O’Neil theory are
therefore referred to as indirect ASA in this paper. Theories much less stringent than expected, and this work also derived
of these approaches will be presented in Sec. I. The method more precise conditions for the applicability of the O’Neil
used for calculating ultrasonic fields from a linear array with theory. Thus the O’Neil theory does seem to be a good basis
a cylindrical concave surface ~referred to as Ocheltree-based for evaluating focused transducers. From the relation of pres-
method in sequel! was developed based on the idea proposed sure with velocity potential, p̃(r)52 r] f (r)/ ] t
by Lee and Benkeser28 who extended to the circular array 5 jk r c f (r) where r is the density, c is the sound velocity,
case the method developed by Ocheltree and Frizzell for pla- and k5 v /c is the wave number, one obtains the Rayleigh
nar sources,29 and the Ocheltree-based method is given in integral in terms of pressure in the following manner:
detail in the Appendix. In Sec. II, both approaches are ap-
plied to the specific case of a linear array with a cylindrically
concave surface. In Sec. III, calculations and comparison of p̃ ~ r! 52
jk r c
2p
EE S
ṽ n ~ r8 !
exp~ jkr s !
rs
dS. ~2!
the angular spectra of the linear array are made using both
approaches.
Two angular spectrum approaches, the extended ASA and
the indirect ASA, are derived from this Rayleigh integral.
I. THEORY
A. General consideration
A curved transducer having a rigid baffle and radiating B. The extended angular spectrum approach
acoustic fields into lossless fluids is considered here. It is Consider a source with surface S that is represented by
well known that a sound field from a rigid baffled planar z5 f (x,y) under the Cartesian coordinates. Assuming that z
piston source in a fluid can be rigorously depicted by the 5 f (x,y) is continuous and has the first-order derivatives, the
Rayleigh integral. For a curved source the Rayleigh integral elementary surface dS can be expressed by dS
can approximately represent the radiation of the source under 5dx 8 dy 8 /cos uz , where cos uz is the direction cosine of the
certain conditions.19,24,25 Following O’Neil,19 the velocity normal with respect to the z-axis. cos uz can be obtained
potential can be approximately expressed by the Rayleigh from
integral if the normal velocity of the source on the curved
surface S is represented by ṽ n (r) at frequency v,
1
f ~ r! 52
1
2p
EE S
ṽ n ~ r8 !
exp~ jkr s !
rs
dS, ~1!
cos u z 5
A~ f x ~ x,y !! 2
1 ~ f y ~ x,y !! 2 11
, ~3!

where r s 5 u r2r8 u is the distance from source point r8 on where f x (x,y) and f y (x,y) are the partial derivatives with
surface S to field point r in the medium. Equation ~1! treats respect to x and y, respectively. From Eqs. ~2! and ~3!, the
each surface element dS as a point source of strictly hemi- pressure in the Cartesian coordinates can be easily derived as

2619 J. Acoust. Soc. Am., Vol. 105, No. 5, May 1999 P. Wu and T. Stepinski: Extension of the angular spectrum approach 2619

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 132.241.216.43 On: Mon, 15 Dec 2014 06:55:16
 p̃ ~ x,y,z ! 52
jk r c
2p
EE S xy
ṽ n ~ x 8 ,y 8 , f ~ x 8 ,y 8 !!
where the condition z 1 >max@ f(x8,y8)# must be met in order
to ensure that V(k x ,k y ;z5z 1 ) is always finite for all k x and
k y . Equation ~6! is the final form for the angular spectrum of
exp~ jkr s ! dx 8 dy 8 a curved source and is derived, based on the Rayleigh inte-
3 gral of curved surface ~the O’Neil’s theory! and the ASA in
rs cos u z
the case of planar sources, without any approximation. Ob-
52
jk r c
2p
EE S xy
ṽ z ~ x 8 ,y 8 , f ~ x 8 ,y 8 !!
viously, the angular spectrum of the curved source is not of
the form of the 2-D spatial Fourier transform. Since k z
5 Ak 2 2k 2x 2k 2y 5k A12n 2x 2n 2y , evanescent waves show up
exp~ jkr s ! when k 2x 1k 2y .k 2 or n 2x 1n 2y .1. When z 1 is chosen to be so
3 dx 8 dy 8 , ~4!
rs large that z 1 >max@ f(x8,y8)# plus a few wavelengths, e.g.,
z 1 >max@ f(x8,y8)#12l, the evanescent waves are strongly at-
where S xy is the area of the projection of the surface S
tenuated and thus can be neglected without loss of
onto the x-y plane, ṽ z 5 ṽ n /cos uz , and rs
accuracy.4–9 Therefore, V(k x ,k y ;z5z 1 ) in the range of k 2x
5 A(x2x 8 ) 2 1(y2y 8 ) 2 1 @ z2 f (x 8 ,y 8 ) # 2 .
1k 2y <k 2 is usually used for synthesizing the acoustic field.
In the case of a planar piston source, its angular spec-
In general, the double integral in Eq. ~6! is very difficult to
trum can be obtained by applying the convolution theorem to
directly solve. But, in some special cases, the integral can be
the Rayleigh integral1,2 and it is given by a 2-D spatial Fou-
simplified to a significant extent. One of the cases is linear
rier transform of the normal velocity on the source
arrays with cylindrically curved surfaces, which will be stud-
surface.4–8 For a planar piston source located in the z5z 0
ied in the following section.
plane, its angular spectrum is denoted by V(k x ,k y ;z5z 0 ).
For an acoustic field from the planar source propagating for- C. The indirect angular spectrum approach
ward in the 1z-direction to a plane at z 1 (>z 0 ), the angular
In the indirect ASA, the angular spectrum of a curved
spectrum of the field can be derived from the relation1
transducer is obtained via 2-D Fourier transforming an initial
V ~ k x ,k y ;z5z 1 ! 5V ~ k x ,k y ;z5z 0 ! exp~ j ~ z 1 2z 0 ! k z ! , ~5! field pre-calculated in a plane. Since the 2-D Fourier trans-
form can be implemented very efficiently using FFT, the
where k x 5kn x , k y 5kn y , k z 5 Ak 2 2k 2x 2k 2y 5k A12n 2x 2n 2y method for calculating the initial field is most essential to the
are the spatial frequencies in the x-, y-, and z-directions in indirect ASA. For a linear array with a cylindrically curved
the medium, respectively. However, the above relation does surface, we have developed the Ocheltree-based method for
not apply to a curved source. Now we divide the surface S of calculating its ultrasonic fields, based on the method devel-
the curved source into a set of such small elementary sources oped by Ocheltree and Frizzell for planar sources29 and the
with surface d S that each of them can be approximated by a appropriate coordinate transformation. This Ocheltree-based
plane. Then we obtain the angular spectrum of the whole method is derived from the Rayleigh integral in Eq. ~2! and
source by the superposition of the angular spectra of elemen- the derivation is presented in the Appendix. Although it is
tary sources. Supposing that d S xy is the projection of d S developed for the linear array with a cylindrically curved
onto the x-y plane, the angular spectrum of the small el- surface, the method can be adapted to the cases of circular
ementary source on surface S at (x 8 ,y 8 ,z 8 ) where z 8 arrays28 and transducers with conical and spherical
5 f (x 8 ,y 8 ) can be approximated by surfaces,30 or perhaps the case of a transducer with an arbi-

EE
trarily curved surface. Consider a source with a curved sur-
d V ~ k x ,k y ;z 8 ! 5 ṽ z ~ x 8 ,y 8 ,z 8 ! face S, represented by z5 f (x,y) under the Cartesian coor-
d S xy dinates, as in Sec. I B. An initial velocity field of
z-component radiated by the source is calculated in a plane
3exp@ 2 j ~ x 8 k x 1y 8 k y !# dx 8 dy 8
at z5z 1 , and represented by ṽ 0z (x,y,z 1 ). Performing the
' ṽ z ~ x 8 ,y 8 ,z 8 ! exp@ 2 j ~ x 8 k x 1y 8 k y !# d S xy . 2-D Fourier transform on the initial field that can be thought
of as a secondary source with normal velocity ṽ n (x,y,z 1 )
Thus at the plane z5z 1 (>max@ f(x8,y8)#) the angular spec- 5 ṽ 0z (x,y,z 1 ), we obtain the angular spectrum of the curved
trum of the whole source can be obtained from superposition source at the plane z5z 1 ~refer to the Appendix for the pro-
of the spectra of all elementary sources in the following man- cedure!.
ner:

V ~ k x ,k y ;z5z 1 ! 5 lim ( d V ~ k x ,k y ;z 8 !
d S xy →0 S xy
II. THE ANGULAR SPECTRUM AND THE ACOUSTIC
FIELD OF A LINEAR ARRAY WITH A
CYLINDRICALLY CONCAVE SURFACE
3exp~ j ~ z 1 2z 8 ! k z !

EE
The ultrasonic linear array considered here has a cylin-
5 ṽ z ~ x 8 ,y 8 , f ~ x 8 ,y 8 !! drical concave surface performing geometrical focusing in
S xy elevation, i.e., in the y-z plane ~see Fig. 1!. Calculation of
3exp@ 2 j ~ x 8 k x 1y 8 k y !# the angular spectrum and the fields from this type of array is
presented in this section. For the indirect ASA, the angular
3exp@ j ~ z 1 2 f ~ x 8 ,y 8 !! k z # dx 8 dy 8 , ~6! spectrum of the array is just the 2-D Fourier transformed

2620 J. Acoust. Soc. Am., Vol. 105, No. 5, May 1999 P. Wu and T. Stepinski: Extension of the angular spectrum approach 2620

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 132.241.216.43 On: Mon, 15 Dec 2014 06:55:16
 lindrically concave and expressed by the following equation:

y 2 1 ~ z2R ! 2 5R 2 , or z5R2 AR 2 2y 2 ,

$ x i 2a<x<x i 1a, 2b<y<b % , ~7!

where R is the radius of the curvature, 2a and 2b define
the width and length of each element, respectively, x i
5(i21/2)d is the position of the center of the ith element
on the x axis, i52N 8 /211, . . . ,0,1,2, . . . ,N 8 /2, and z5e
for y56b. The direction cosine of the normal of the array
surface with respect to the z-axis is easily found from Eqs.
~2! and ~7! as follows:

cos u z 5 AR 2 2y 2 /R. ~8!

Consider the array with all the elements having uniform nor-
mal velocity distribution on the surface. For the ith element,
the normal velocity excitation at frequency v is supposed to
be
FIG. 1. Geometry and notation of the array with a cylindrically concave
surface. ṽ ni ~ x,y,z ! 5 ṽ i ~ v ! exp~ j v t i ! , ~9!
where ṽ i is the complex amplitude, and t i is the delay time,
initial field in a plane radiated by the array. The initial field is which, in the time-harmonic case, creates phase shift v t i for
calculated by the Ocheltree-based method that has been pre- beam steering and focusing.16 Inserting Eqs. ~7!–~9! into Eq.
sented in the Appendix. Here we only consider the extended ~6!, we obtain the angular spectrum for the ith element of the
ASA used for the linear array case. array centering at (x,y,z)5 $ (i21/2)d,0,0 % and having the
Referring to Fig. 1, we assume that the array has N 8 dimension $(i21/2)d2a<x<(i21/2)d1a, 2b<y<b%,
elements (N 8 is assumed to be even! and its surface is cy- as follows:

V i ~ k x ,k y ;z5z 1 ! 5 ṽ i ~ v ! exp~ j v t i ! E ~ i21/2! d1a

~ i21/2! d2a
exp~ 2 jx 8 k x ! dx 8 E b

2b
exp@ 2 j ~ y 8 k y 1 ~ R2 AR 2 2y 8 2 2z 1 ! k z !#
AR 2 2y 8 2 /R
dy 8

5 ṽ i ~ v ! exp@ j v t i 2 j ~ i21/2! dk x # 2a sinc~ ak x ! E 0

b 2R cos~ y 8 k y ! exp~ j ~ z 1 2R1 AR 2 2y 8 2 ! k z !
AR 2 2y 2
dy 8 ,

~ z 1 >e ! , ~10!

where sinc(x)5sin(x)/x. From Eq. ~10! and by superposition,

the angular spectrum of the array can be obtained, p̃ ~ x,y,z ! 5k r c E E`

2`
`

2`
V ~ k x ,k y ,z5z 1 !

exp@ j ~ xk x 1yk y 1 ~ z2z 1 ! k z !#

N 8 /2 3 dk x dk y ,
kz
V ~ k x ,k y ;z5z 1 ! 5 ( V i ~ k x ,k y ;z5z 1 ! . ~11!
i52N 8 /211 ~ z>z 1 ! . ~12!
In the transient case, the pulse normal velocity of the
Inserting Eq. ~10! into Eq. ~11!, it is easy to find that ith element is represented by v ni (x,y,z;t) and its temporal
2a sinc(ak x ) and the integral can be factored out from the Fourier transform is ṽ ni (x,y,z) defined by Eq. ~9!. The
sum in Eq. ~11!. Therefore, in the cases of a cylindrically transient pressure field can, thus, be obtained by the
concave, rectangular transducer, and an array consisting of inverse Fourier transform, that is, p(x,y,z;t)
such linearly aligned transducers, the double integral of the 5(1/2p ) * `2` p̃(x,y,z)exp(2jvt)dv.
angular spectrum can be reduced into a single integral. The
single integral will be numerically implemented, which will III. NUMERICAL CALCULATION AND COMPARISON
IN THE CASE OF THE LINEAR ARRAY
be further studied in next section ~Sec. III!.
Based on the angular spectrum in Eq. ~11!, the pressure In this section, we present some calculations of angular
field from the array can be synthesized, spectra using the extended and the indirect ASA for a spe-

2621 J. Acoust. Soc. Am., Vol. 105, No. 5, May 1999 P. Wu and T. Stepinski: Extension of the angular spectrum approach 2621

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 132.241.216.43 On: Mon, 15 Dec 2014 06:55:16
 cific linear array similar to the one used in our lab for the array, only one-fourth of the initial field is calculated that is
inspection of electron beam welds in copper.15 The programs in the first quadrant (x>0, y>0) of the plane at z 1 . To
for both approaches were written in C and run on a PC with reduce the aliasing error in the transform domain, a spatial
Pentium 133 MHz. The array has the geometrical param- interval of half-wavelength (l/2) ~Nyquist interval! is used.
eters: 2a51 mm, 2b530 mm, R5100 mm, and N 8 522. Concerning the initial field size, the field is calculated only in
No gaps between the adjacent elements is assumed. The me- the first quadrant with size of 1293129 points that makes
dium in which waves propagate is water whose sound speed 32 mm332 mm, and then it is mirrored to the other three
and density are 1500 m/s and 1000 kg/m3, respectively. Here quadrants so as to form a 2573257 matrix. The amplitudes
the considered array is not electronically focused. In this of the field points along the four sides of the square matrix
case, both the angular spectrum and the radiated field are are less than 0.55% maximum of the field. Finally, zero-
symmetrical with respect to the x- and the y-axis, respec- padding is done to the 2573257 field so as to create a 512
tively. 3512 matrix for the 2-D FFT. For obtaining such a 512
From the extended ASA, that is, using Eqs. ~10! and 3512 angular spectrum the whole computational time was
~11!, we calculate the angular spectrum for z 1 5e15 mm 877 s. The angular spectrum obtained from the indirect ASA
~where e5R2 AR 2 2b 2 51.13 mm). The evanescent waves ~solid curves! is also shown in Fig. 2 in comparison with the
are neglected in this case without affecting the accuracy ~see one obtained from the extended ASA ~bold dotted curves!.
Sec. I! and thus V(k x ,k y ;z5z 1 ) in the range of 21<n x , The comparison in Fig. 2 shows that in the on-axis case
n y <1 is of interest. The computational accuracy of the ap- @see Fig. 2 ~a!–~d!# the results from the two approaches agree
proach completely depends on the numerical integration, excellently except that some small discrepancies exist in the
which obviously depends on the integration interval dy ~or range of large n x and n y ~around 0.95–1!, and that in the
spatial sampling interval!, that is, the smaller the dy the off-axis case @Fig. 2~e!–~h!# the results agree very well ex-
more accurate the integration and meanwhile the more time cept that some small discrepancies exist in the range of small
consuming the computation. With consideration of both ac- n x and n y ~around 0! and in the range of large n x and n y
curacy and efficiency for the numerical integration in Eq. ~around 0.5!.
~10!, the most important thing is choosing an appropriate To figure out the reason for these discrepancies, we used
spatial sampling interval. Since R@b ~the upper limit of y 8 ), the indirect ASA to re-calculate the angular spectrum with
1/AR 2 2y 8 2 varies very slowly in a small range from 1/R to the initial field size increased to 64 mm364 mm, that re-
1/AR 2 2b 2 as y 8 changes from 0 to b, and either does R sulted in 2573257 field points. Then the 2573257 field was
2 AR 2 2y 8 2 in the range from 0 to R2 AR 2 2b 2 mirrored to the other three quadrants so as to form a 512
('b 2 /(2R)). Due to the small-range variation of R 3512 matrix for 2-D FFT. The computational time in this
2 AR 2 2y 8 2 , exp(j(z12R1AR 2 2y 2 )k z ) oscillates only with case was 2850 s, almost four times long as for the computa-
a few cycles even for k z 5k. Thus cos(y8ky) in the integral is tion of the 32 mm332 mm initial field. The results calculated
the dominant term for choosing an appropriate sampling in- in such a way have been shown to be closer to those obtained
terval for the integration. Taking into account that the spatial from the extended ASA. This can be demonstrated by com-
frequency k y 5n y 2 p /l varies from 0 to 2 p /l as n y goes paring the normalized root-mean-square ~rms! difference be-
from 0 to 1, the three different sampling intervals are used tween the results obtained from the extended ASA and the
for the integration that are l/5 for n y in the range 0–0.33, indirect ASA in the above two cases, respectively. The nor-
l/10 for n y in the range 0.33–0.66, and l/15 for n y in the malized rms difference is here defined as
range 0.66–1. In this way, the angular spectrum with dimen-
sion of 2573257 in the first quadrant (0<n x , n y <1) was $ ( m51
M
@ V ind~ m ! 2V ext~ m !# 2 /M % 1/2
calculated with consideration of its symmetry, and the com- D rms5 ,
$ ( m51
M
@ V ext~ m !# 2 /M % 1/2
putational time used was 99 s. The calculated angular spec-
trum is shown in Fig. 2 as a bold dotted curve.
Using the indirect ASA, i.e., using the Ocheltree-based where V ext(m) and V ind(m) are the real or imaginary part of
method to calculate an initial field in the plane at z5e the discrete angular spectra from the extended ASA and the
15 mm and then performing the 2-D FFT of the initial field, indirect ASA for given n x or n y . Take the real part of the
we obtained the angular spectrum of the array in the plane at angular spectrum for n x 50 @like the one in Fig. 2~a!# as an
z5e15 mm. The accuracy of this approach depends on the example. In the case of the 32 mm332 mm initial field, D rms
accuracy of the initial field, the spatial interval of the field for the results in Fig. 2~a! is 0.0387, and in the case of the
points, and the size ~the spatial truncation! of the field. For a 64 mm364 mm initial field, D rms reduces to 0.0297. This re-
good accuracy of the initial field, the condition is taken into veals that the extended ASA is more accurate than the indi-
account that determines the limitation on the width and rect ASA, and that the spatial truncation of the initial field
length of each subelement into which each array element is has an effect on the computation accuracy of the indirect
divided @see Eq. ~A7! in the Appendix#. Setting F580 and ASA, i.e., the larger the initial field size the better the com-
z55 mm in A4lz/F and then taking appropriate integer putational accuracy. Obviously, computation of an initial
numbers @refer to Eq. ~A3!#, each element in the array is field with a larger size requires much longer computational
divided into 3385 subelements whose widths and lengths time.
satisfy Eq. ~A7!. Thus the 22-element array is divided into From the above calculation, comparison, and analysis, it
66385 subelements altogether. Due to the symmetry of the is concluded that the ASA can be extended to directly solve

2622 J. Acoust. Soc. Am., Vol. 105, No. 5, May 1999 P. Wu and T. Stepinski: Extension of the angular spectrum approach 2622

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 132.241.216.43 On: Mon, 15 Dec 2014 06:55:16
 FIG. 2. The angular spectra from the extended ASA ~dotted line! and the indirect ASA ~solid line!. ~a! The real and ~b! the imaginary part for n x 50; ~c! the
real and ~d! the imaginary part for n y 50; ~e! the real and ~f! the imaginary part for n x 50.855; ~g! the real and ~h! the imaginary part at n y 50.855. Note that
the vertical axes are different in scale.

2623 J. Acoust. Soc. Am., Vol. 105, No. 5, May 1999 P. Wu and T. Stepinski: Extension of the angular spectrum approach 2623

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 132.241.216.43 On: Mon, 15 Dec 2014 06:55:16
 FIG. 2. ~Continued.!

the fields from curved sources based on Eq. ~6!. In the case IV. CONCLUSIONS
of a linear array with a cylindrically concave surface, the
double integral in Eq. ~6! reduces to the single integral in Eq. The ASA which is conventionally applicable to the case
~10!. In this case, the extended ASA has been shown to be of planar transducers has been extended to treat the case of
more accurate and more efficient than the indirect ASA ~the curved transducers, and thus two approaches, namely, the
efficiency of the extended ASA is better by a factor of 8 or extended ASA and indirect ASA, have been developed.
more!, and for the similar computational accuracy, the ex- The extended ASA has been derived without any ap-
tended ASA is much more efficient than the indirect ASA proximation in a general way starting from the Rayleigh in-
~the efficiency of the extended ASA is better by a factor of tegral of curved surface ~O’Neil’s theory! and using the con-
28 or more!. ventional ASA theory for the case of planar sources. A
However, in the general case where the double integral general form of the angular spectrum has been obtained that
of the extended ASA cannot be reduced to a single integral, is given by a double integral that does not have the 2-D
the double integral must be evaluated, and the extended ASA Fourier transform form. Therefore, the extended ASA is
may not be so efficient as the indirect ASA. For example, if implemented by numerical integration instead of using the
we had directly evaluated the double integral in Eq. ~6! for 2-D FFT. This approach has been applied to the case of a
the above-studied linear array case using the sampling inter- linear array with a cylindrically curved surface, and it has
val dx5l/5 for all n x ~so as to make 110 integration inter- been shown that in this case the double integral reduces to a
vals in the x-direction!, the computational time could have single integral.
been about 100 times the computational time ~99 s! for the The indirect ASA gives the angular spectrum of a
evaluation of the single integral in Eq. ~11!, instead of 28 curved source via the 2-D FFT of an initial field pre-
times 99 s—the computational time for the indirect ASA. calculated in a plane. The method for calculating the initial
Therefore, in this general case, the indirect ASA can be a field from a linear array with a cylindrically concave surface
better alternative. has been established based on the method developed by Och-
Using the angular spectrum calculated from the ex- eltree and Frizzell.29 This Ocheltree-based method is the
tended ASA and Eq. ~12!, we simulated an acoustic field most essential part of the indirect ASA.
from the array ~see Fig. 3!. The acoustic field is in the y-z A specific example has been given of a linear array with
plane at x50 and its focal zone is seen around 95 mm. a cylindrically concave surface similar to that used in our lab

FIG. 3. Simulated acoustic field ~pressure magnitude!

in water from the linear array with a cylindrically con-
cave surface, based on the extended ASA. The field is
normalized by multiplying by the acoustic impedance
of water.

2624 J. Acoust. Soc. Am., Vol. 105, No. 5, May 1999 P. Wu and T. Stepinski: Extension of the angular spectrum approach 2624

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 132.241.216.43 On: Mon, 15 Dec 2014 06:55:16
 for the inspection of weld in copper. An efficient way for the case the method developed by Ocheltree and Frizzell for pla-
numerical implementation of the extended ASA has been nar sources.29 Unlike the circular array dealt with by Lee and
proposed. The angular spectra of the array were calculated Benkeser, the array in our case is a linear array with a cylin-
using the two approaches and compared. The comparison has drically concave surface, their idea has been adapted to our
shown that the angular spectra calculated from both ap- case. Therefore, we name it Ocheltree-based method.
proaches are in excellent agreement. To facilitate the calculation of fields from such a linear
The accuracy and efficiency of both approaches have array using Ocheltree-based method, we re-arrange Fig. 1
been discussed in detail. From the example of the linear ar- and use two coordinate systems ~see Fig. A1!. The first one
ray, the extended ASA has been shown to be more efficient is the x-y-z coordinate system which is the same as in Fig. 1,
and more accurate than the indirect ASA, especially in the and the other is x m -y m -z m coordinate system which is a ro-
range of large direction cosines (n x and/or n y ). However, in tating system. Their relation will be given below.
the general case where the double integral of the extended The half angle, a 0 , subtended by chord 2b ~see Fig. A1!
ASA cannot be reduced to a single integral, the double inte- can easily be found to be
gral must be evaluated, and the extended ASA may not be as
a 0 5arcsin~ b/R ! . ~A1!
efficient as the indirect ASA. Therefore, in this general case,
the indirect ASA can be a better alternative. Thus, the arc length spanning over chord 2b is
In principle, both approaches can be applied to arbi-
& 52 a 0 R.
L ~A2!
trarily curved transducers provided that the secondary dif-
fraction due to the curvature of a transducer does not yield When each element is divided into such M x 3M u small sub-
significant errors in calculation. In practice, the secondary elements with width Dh and length DL & ~see Fig. A1! that the
diffraction is usually unimportant and its effect can be ne- curvature of the subelements can be neglected, the subele-
glected, as stated in Sec. I. ments can be thought of as plane piston sources. The subele-
ments are indicated by (m x ,m u ), where m x 51, . . . ,M x and
ACKNOWLEDGMENTS m u 51, . . . ,M u . Thus the width and the length of each sub-
This work was sponsored by the Swedish Nuclear Fuel element are
and Waste Management Co. ~SKB! under Contract No. & 5D a R52 a 0 R/M u ,
Dh52a/M x , DL ~A3!
61230-97-52. The authors are thankful to the anonymous re-
viewers for their valuable comments on this paper. respectively, where
D a 52 a 0 /M u ~A4!
APPENDIX: THE METHOD FOR THE CALCULATION
OF FIELDS FROM A LINEAR ARRAY WITH A is the angle subtended by the arc spanning the centers of
CYLINDRICALLY CONCAVE SURFACE adjacent subelements in the y-direction. The angle subtended
This method was developed based on the idea proposed by the arc from the center of subelement (m x ,m u ) to the
by Lee and Benkeser28 who extended to the circular array z-axis is
u m u 52 a 0 1D a ~ m u 21/2! , ~ m u 51, . . . ,M u ! . ~A5!

To facilitate the calculation, we shall use the second set of

coordinates, (x m ,y m ,z m ), for all subelements with u m u . The
(x m ,y m ,z m ) coordinates are a system rotating around point
(x,y,z)5(0,0,R) and are related to the x-y-z coordinates

FGF GF G
through the following transformation:
xm 1 0 0 x
ym 5 0 cos u m u sin u m u y . ~A6!
zm 0 2sin u m u cos u m u z2R
Equation ~A6! means that the x m -y m -z m coordinate system is
obtained in such a way that the x-y-z coordinates first trans-
late by R in the z-direction, and then rotate clockwise by an
angle u m u .
Following Ocheltree and Frizzell for planar sources,29
each subelement has such a small size DA5Dh3DL & that
Dh and DL & meet the condition

& < A4lz/F,

Dh, DL ~A7!
in which the constant F represents the distance from the
FIG. A1. Geometry for calculating acoustic field from the linear array with source to the field point relative to the distance to the near-
a cylindrically concave surface based on Ocheltree’s method ~Ref. 29!. field–far-field transition for a source of size DA5Dh

2625 J. Acoust. Soc. Am., Vol. 105, No. 5, May 1999 P. Wu and T. Stepinski: Extension of the angular spectrum approach 2625

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 132.241.216.43 On: Mon, 15 Dec 2014 06:55:16
 3DL &, and thus, the Fraunhofer approximation can be ap- N 8 /2 N 8 /2
plied. p̃ ~ x,y,z ! 5 (
i52N 8 /211
p̃ i ~ x,y,z ! 5 (
i52N 8 /211
p̃ 00~ x1x i ,y,z ! ,
For such a small subelement that has a uniform normal
~A12!
velocity v 0 and is located at point ~0, 0, 0! under the x-y-z
coordinates, corresponding to point ~0, 0, 2R) under the where x i 5(i21/2)d. Choosing a plane z5z 1 parallel to
x m -y m -z m coordinates with u m u 50, the acoustic field from the x-y plane and using the relation ṽ z (x,y,z 1 )
the subelement can be approximated by29 52(1/jk r c) ] p̃(x,y,z)/ ] z u z5z 1 , we calculate an initial ve-
locity field of z-component, denoted by ṽ 0z (x,y,z 1 ), in the
d p 0,0~ x 0 ,y 0 ,z 0 ! plane z5z 1 via Eq. ~A12!. Based on the initial field, which

EE e 2 jkr s0
jrc can be thought of as a secondary source with normal velocity
52 v n ~ x 80 ,y 80 ,z 80 ! dS 0 ṽ n (x,y,z 1 )5 ṽ 0z (x,y,z 1 ), the angular spectrum of the array
l DA r s0
at the plane z5z 1 can be obtained using the 2-D FFT.

'
j r cDA v 0
l
sinc
2R s0S
kx 0 Dh
sinc
2R s0
&
ky 0 DL
,D S D ~A8!
Although the method is established for the case of a
cylindrically curved, rectangular transducer, and a linear ar-
ray consisting of such transducers, it can be extended to the
where dS 0 5dx 08 dy 08 , case of transducers with conical and spherical surfaces30 or
perhaps the general case where a transducer has an arbitrarily
r s0 5 A~ x 2x 8 ! 2 1 ~ y 2y 8 ! 2 1 ~ z 1R ! 2 curved surface.
0 0 0 0 0

and R s0 5 Ax 2 1y 2 1(z 1R) 2 .

0 0 0 1
J. W. Goodman, Introduction to Fourier Optics ~McGraw-Hill, New
For such a subelement that is located at point (x,y,z) York, 1968!.
5(x m x ,R sin umu,R2R cos umu), where 2
G. C. Sherman, ‘‘Generalization of the angular spectrum of plane waves
and the diffraction transform,’’ J. Opt. Soc. Am. 59, 146–156 ~1969!.
3
x m x 52a1Dh ~ m x 21/2! ~ m x 51, . . . ,M x ! , ~A9! P. C. Clemmow, The Plane Wave Spectrum Representation of Electro-
magnetic Fields ~Pergamon, London, 1966!.
4
E. G. Williams and J. D. Maynard, ‘‘Numerical evaluation of the Rayleigh
we can obtain the same formula as in Eq. ~A8! by transform- integral for planar radiators using the FFT,’’ J. Acoust. Soc. Am. 72,
ing the x-y-z coordinates into the x m -y m -z m coordinates by 2020–2030 ~1982!.
means of Eq. ~A6!. The pressure field from subelement 5
D. P. Orofino and P. C. Pedersen, ‘‘Efficient angular spectrum decompo-
sition of acoustic sources—Part I. Theory, and Part II. Results,’’ IEEE
(m x ,m u ) under the x m -y m -z m coordinates is expressed as
Trans. Ultrason. Ferroelectr. Freq. Control 40, 238–257 ~1993!.
6
P. T. Christopher and K. J. Parker, ‘‘New approaches to the linear propa-
d p m x ,m u ~ x m ,y m ,z m ! gation of acoustic fields,’’ J. Acoust. Soc. Am. 90, 507–521 ~1991!.

EE
7
P. Wu, R. Kazys, and T. Stepinski, ‘‘Analysis of the numerically imple-
jrc e 2 jkr sm mented angular spectrum approach based on the evaluation of two-
52 8 ,y m8 ,z m8 !
v n~ x m dS m
l DA r sm dimensional acoustic fields—Part I: Errors due to the discrete Fourier

S D S D
transform and discretization, and Part II: Characteristics as a function of
k ~ x m 2x m x ! Dh & angular range,’’ J. Acoust. Soc. Am. 99, 1339–1359 ~1996!.
j r cDA v 0 ky m DL 8
P. Wu, R. Kazys, and T. Stepinski, ‘‘Optimal selection of parameters for
' sinc sinc ,
l 2R sm 2R sm the angular spectrum approach to numerically evaluate acoustic fields,’’ J.
Acoust. Soc. Am. 101, 125–134 ~1997!.
~A10! 9
D.-L. Liu and R. C. Waag, ‘‘Propagation and backpropagation for ultra-
sonic wavefront design,’’ IEEE Trans. Ultrason. Ferroelectr. Freq. Control
8 dy m8 ,
where dS m 5dx m 44, 1–13 ~1997!.
10
M. E. Schafer, ‘‘Transducer characterization using the angular spectrum
r sm 5 A~ x 2x 8 2x ! 2 1 ~ y 2y 8 ! 2 1 ~ z 1R ! 2 , method,’’ J. Acoust. Soc. Am. 85, 2202–2214 ~1996!.
m m mx m m m 0 11
K. K. Liang, G. S. Kino, and B. T. Khuri-Yakub, ‘‘Material characteriza-
tion by the inversion of V(z),’’ IEEE Trans. Sonics Ultrason. SU-32,
and R sm 5 A(x 2x ) 2 1y 2 1(z 1R ) 2 .
m mx m m 0 213–225 ~1985!.
12
The pressure field from an element centered at (x,y,z) C.-H. Chou and G. S. Kino, ‘‘The evaluation of V(z) in a type II reflec-
5(0,0,0) is the summation of d p m x ,m u (x m ,y m ,z m ), that is, tion microscope,’’ IEEE Trans. Ultrason. Ferroelectr. Freq. Control 34,
341–345 ~1987!.
13
Mu Mx V. S. Ardebili and A. N. Sinclair, ‘‘A new angular spectrum approach for

( (
modelling the acoustic microscope response with high attenuation cou-
p̃ 00~ x,y,z ! 5 d p m x ,m u ~ x m ,y m ,z m ! pling fluids,’’ Proc. 1995 IEEE Ultrason. Symp. ~IEEE Cat. No.
m u 51 m y 51
95CH35844!, Vol. 2, 873–876 ~1995!.

S D
14
Y. Sugara, J. Kushibiki, and N. Chubachi, ‘‘Performance of concave
j r cDA v 0 u
M My k ~ x m 2x m x ! Dh
5
l m u 51
( m(51 sinc
x
2R sm 15
transducers in acoustic microscopy,’’ Proc. 1988 IEEE Ultrason. Symp.
~IEEE Cat. No. 88CH2578-3!, Vol. 2, 751–756 ~1988!.
T. Stepinski and P. Wu, Inspection of Copper Canisters for Spent Nuclear

3sinc S 2R sm
&
ky m DL
, D ~A11! 16
Fuel by Means of Ultrasonic Array System, SKB Projektrapport 97-08
~SKB, Stockholm, August 1997!.
P. Wu and T. Stepinski, ‘‘Elastic fields in immersed isotropic solids from
phased arrays: The time harmonic case,’’ Res. Nondestruct. Eval. 10,
where x m , y m , z m depend on m u via Eq. ~A6!. The pressure 185–204 ~1998!.
17
D. Guyomar and J. P. Powers, ‘‘Transient fields radiated by curved
field from the array with N 8 elements is, therefore, obtained surfaces—Application to focusing,’’ J. Acoust. Soc. Am. 76, 1564–1572
by the summation of p̃ i (x,y,z), the field from the ith array ~1984!.
18
element, in the manner D. Guyomar and J. Powers, ‘‘Transient radiation from focused transduc-

2626 J. Acoust. Soc. Am., Vol. 105, No. 5, May 1999 P. Wu and T. Stepinski: Extension of the angular spectrum approach 2626

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 132.241.216.43 On: Mon, 15 Dec 2014 06:55:16
 ers,’’ Proc. IEEE 1984 Ultrason. Symp. ~IEEE Cat. No. 84CH2112-1!, 25
M. A. Fink and J.-F. Cardoso, ‘‘Diffraction effects in pulse-echo measure-
Vol. 2, 979–982 ~1984!. ment,’’ IEEE Trans. Sonics Ultrason. SU-31, 313–329 ~1984!.
19 26
H. T. O’Neil, ‘‘Theory of focusing radiators,’’ J. Acoust. Soc. Am. 21, L. W. Schmerr, Jr., A. Sedov, and T. P. Lerch, ‘‘A boundary diffraction
516–526 ~1949!. wave model for a spherically focused ultrasonic transducer,’’ J. Acoust.
20
A. Penttinen and M Luukkala, ‘‘The impulse response and pressure Soc. Am. 101, 1269–1277 ~1997!.
nearfield of a curved radiator,’’ J. Phys. D 9, 1547–1557 ~1976!. 27
W. A. Verhoef, M. J. T. M. Cloostermans, and J. M. Thijsssen, ‘‘The
21
M. Arditi, F. S. Foster, and J. M. Hunt, ‘‘Transient fields of concave impulse-response of a focused source with an arbitrary axisymmetric sur-
annular arrays,’’ Ultrason. Imaging 3, 37–61 ~1981!. face velocity distribution,’’ J. Acoust. Soc. Am. 75, 1716–1720 ~1984!.
22 28
E. L. Madsen, M. M. Goodsitt, and J. A. Zagzebski, ‘‘Continuous waves C. Lee and P. J. Benkeser, ‘‘Computationally efficient sound field calcu-
generated by focused radiators,’’ J. Acoust. Soc. Am. 70, 1508–1517 lations for a circular array transducer,’’ IEEE Trans. Ultrason. Ferroelectr.
~1981!. Freq. Control 39, 43–47 ~1992!.
23 29
H. Djelouah, J. C. Baboux, and M. Perdix, ‘‘Theoretical and experimental K. B. Ocheltree and L. A. Frizzell, ‘‘Sound field calculation for rectangu-
study of the field radiated by ultrasonic focused transducer,’’ Ultrasonics lar sources,’’ IEEE Trans. Ultrason. Ferroelectr. Freq. Control 36, 242–
29, 188–200 ~1991!. 248 ~1989!.
24 30
F. Coulouvrat, ‘‘Continuous field radiated by a geometrically focused C. Lee and P. J. Benkeser, ‘‘A computationally efficient method for the
transducer: Numerical investigation and comparison with an approximate calculation of the transient field of acoustic radiators,’’ J. Acoust. Soc.
model,’’ J. Acoust. Soc. Am. 94, 1663–1675 ~1993!. Am. 96, 545–551 ~1994!.

2627 J. Acoust. Soc. Am., Vol. 105, No. 5, May 1999 P. Wu and T. Stepinski: Extension of the angular spectrum approach 2627

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 132.241.216.43 On: Mon, 15 Dec 2014 06:55:16