You are on page 1of 14

Acoustic shock wave propagation in a heterogeneous medium: A numerical simulation

beyond the parabolic approximation


Franck Dagrau, Mathieu Rénier, Régis Marchiano, and François Coulouvrat

Citation: The Journal of the Acoustical Society of America 130, 20 (2011); doi: 10.1121/1.3583549
View online: https://doi.org/10.1121/1.3583549
View Table of Contents: https://asa.scitation.org/toc/jas/130/1
Published by the Acoustical Society of America

ARTICLES YOU MAY BE INTERESTED IN

Modeling the propagation of nonlinear three-dimensional acoustic beams in inhomogeneous media


The Journal of the Acoustical Society of America 122, 1352 (2007); https://doi.org/10.1121/1.2767420

Propagation of finite amplitude sound through turbulence: Modeling with geometrical acoustics and the parabolic
approximation
The Journal of the Acoustical Society of America 111, 487 (2002); https://doi.org/10.1121/1.1404378

Time-domain modeling of pulsed finite-amplitude sound beams


The Journal of the Acoustical Society of America 97, 906 (1995); https://doi.org/10.1121/1.412135

Nonlinear and diffraction effects in propagation of N-waves in randomly inhomogeneous moving media
The Journal of the Acoustical Society of America 129, 1760 (2011); https://doi.org/10.1121/1.3557034

One-way approximation for the simulation of weak shock wave propagation in atmospheric flows
The Journal of the Acoustical Society of America 135, 2559 (2014); https://doi.org/10.1121/1.4869685

A generalized Westervelt equation for nonlinear medical ultrasound


The Journal of the Acoustical Society of America 109, 1329 (2001); https://doi.org/10.1121/1.1344157
Acoustic shock wave propagation in a heterogeneous medium:
A numerical simulation beyond the parabolic approximation
Franck Dagrau,a) Mathieu Rénier, Régis Marchiano,b) and François Coulouvrat
Université Pierre et Marie Curie - Paris 6, Institut Jean le Rond d’Alembert (UMR UPMC/CNRS 7190),
4 place Jussieu, 75252 Paris Cedex 05, France

(Received 4 February 2010; revised 4 April 2011; accepted 6 April 2011)


Numerical simulation of nonlinear acoustics and shock waves in a weakly heterogeneous and
lossless medium is considered. The wave equation is formulated so as to separate homogeneous
diffraction, heterogeneous effects, and nonlinearities. A numerical method called heterogeneous
one-way approximation for resolution of diffraction (HOWARD) is developed, that solves the
homogeneous part of the equation in the spectral domain (both in time and space) through a one-
way approximation neglecting backscattering. A second-order parabolic approximation is per-
formed but only on the small, heterogeneous part. So the resulting equation is more precise than
the usual standard or wide-angle parabolic approximation. It has the same dispersion equation as
the exact wave equation for all forward propagating waves, including evanescent waves. Finally,
nonlinear terms are treated through an analytical, shock-fitting method. Several validation tests
are performed through comparisons with analytical solutions in the linear case and outputs of the
standard or wide-angle parabolic approximation in the nonlinear case. Numerical convergence
tests and physical analysis are finally performed in the fully heterogeneous and nonlinear case of
shock wave focusing through an acoustical lens. V C 2011 Acoustical Society of America.

[DOI: 10.1121/1.3583549]
PACS number(s): 43.25.Cb, 43.28.Js [ROC] Pages: 20–32

I. INTRODUCTION Neumann reflection,20,21 or phase singularities.22 Successful


extensions include cases with heterogeneities23,24 or flow
Although acoustic waves are usually small amplitude
motion.25
perturbations, nonlinear effects play a significant role in
Numerical solvers26 for KZK-type equations generally
many applications, such as in aeroacoustics (with for
rely on the split-step method,27 which separates diffraction,
instance sonic boom,1 fan noise,2 or jet noise3), underwater
attenuation effects, and quasi one-dimensional nonlinear
acoustics4,5 and various biomedical applications,6 including
propagation. One main class of numerical solvers handles
tissue harmonic imaging,7 lithotripsy,8,9 or high intensity
linear diffraction in the frequency domain, while spatial var-
focused ultrasound (HIFU).10 In all these cases, propagation
iations are handled either by finite differences28,29 or spectral
is nonlinear and occurs in a weakly heterogeneous medium.
Fourier decomposition.22 The alternative is to treat linear
The range of relative sound speed variations rarely reaches
diffraction in the time domain.17,30 The nonlinear part is also
over 10%. However, these weak variations may induce sig-
treated either in the frequency domain31 or, more frequently,
nificant effects, such as focusing aberrations, caustics occur-
in the time domain. One can use either various shock captur-
rence, shadow zone formations, or sound scattering.
ing algorithms (for a review of such algorithms, see Ref. 32),
The classical modeling for nonlinear diffracting waves is
or analytical solutions based on the Poisson solution for the
based on the nonlinear version of the standard parabolic
smooth case30 and its so-called Burgers–Hayes generaliza-
approximation, known as the KZ equation11 or KZK equa-
tion33 in case of shocks.
tion12 when thermoviscous absorption is included. It assumes
However, parabolic (or paraxial) approximation
wave propagation deviates only weakly from the plane case
remains an approximation, valid typically within an angu-
under the influence of diffraction and nonlinearities. This non-
lar range of about 6 15 of deviation from the main axis.
linear parabolic approximation has been extraordinarily useful
While this approximation is perfectly suitable for highly
for modeling and simulating highly collimated beams used ei-
directional beams, it is not sufficient in the case of low
ther in underwater acoustics13 or for various biomedical devi-
directional or highly focused sources, for sound scattered
ces using HIFU.14–17 Fundamental aspects of nonlinear wave
away from the main direction by inhomogeneities, or in
physics have also been investigated on the basis of this para-
the presence of sheared flows. Improved approximations
xial or parabolic approximation, such as caustics,18,19 von
can be obtained in the linear34 or nonlinear regime35 by
using parabolic (or equivalently Fresnel) approximations
a)
based on non-plane wavefronts, but this requires knowl-
Also at: Dassault-Aviation 78 quai Marcel Dassault 92552 Saint-Cloud edge of the suitable wavefront a priori, which can be
Cedex 300, France
b)
Author to whom correspondence should be addressed. Electronic-mail: done in practice only in a few cases, mostly in homogene-
regis.marchiano@upmc.fr ous media. An alternative approach is to use higher order

20 J. Acoust. Soc. Am. 130 (1), July 2011 0001-4966/2011/130(1)/20/13/$30.00 C 2011 Acoustical Society of America
V
parabolic approximations, which have been successfully where the function /ðx; tÞ is related to the acoustic pressure
developed for linear wave propagation, starting from the pa ðx; tÞ by the following relation:
so-called wide-angle approximation.36 However, in the
nonlinear regime, few works have gone in that direc- @/
tion.37 A recent one38 shows that the wide angle approxi- pa ¼ : (2)
@t
mation, while improving the nearfield simulation of a
piston, does not provide a substantial benefit for the prob- Note that /ðx; tÞ is not exactly the potential as used classi-
lem of shock wave scattering by a heterogeneity. cally in acoustics but is equal to minus the potential times
The objective of the present work is therefore to investi- the density.
gate qualitatively and quantitatively the benefit of an alterna- In Eq. (1), c0 ðxÞ is the medium sound speed, q0 ðxÞ
tive way to the usual parabolic approximation in nonlinear is the ambient density, B/A is the fluid nonlinearity pa-
acoustics. One particular motivation is the sonic boom appli- rameter (that may also depend on position), and
cation, where sound propagation is neither preferably hori- bðxÞ ¼ 1 þ B=2A. Equation (1) is the usual heterogene-
zontal nor vertical and for which diffraction is important in ous wave equation, augmented by the usual nonlinear
various situations like caustics,39 shadow zones,40 or random term of the Lighthill–Westervelt equation. It is valid as
scattering in the turbulent layer near the ground.23,41,42 The long as both nonlinear and inhomogeneous terms remain
so-called heterogeneous one-way approximation for the re- small.49 Indeed linear acoustical effects are on the order
solution of diffraction method (HOWARD) we propose here of the acoustical Mach number  ¼ Ua =c0 (expressed as
is indeed a generalization of the work of Christopher and the acoustical velocity Ua over the ambient sound
Parker.43 That method was based on a phenomenological speed), while nonlinearities are of order e2 . The relative
split-step between linear diffraction modeled by the full sound speed and density fluctuations are of order
wave equation instead of the paraxial approximation and g ¼ ðc0  c0 Þ=c0 where c0 is the mean sound speed.
nonlinear propagation described by the inviscid Burgers’ Cubic nonlinear terms of order 3 and nonlinear hetero-
equation. The linear diffraction of a piston source was geneous terms of order g2 are neglected. Nonlinear
handled numerically in the frequency domain by an efficient effects are decoupled from heterogeneous ones, so that
spectral method (discrete Hankel transform). In a similar the usual linear wave equation in a heterogeneous me-
approach, Tavakkoli et al.44 used the formalism of Rayleigh dium is recovered on the left-hand side with the quad-
integral to handle linear diffraction, an approach that has ratic nonlinear term of the homogeneous Lighthill–
been compared to parabolic approach.45 However, this for- Westervelt equation on the right-hand side. Typical val-
malism is relatively costly from a numerical point of view, ues for sonic boom or HIFU are  ¼ 103 and g ¼ 102 .
so that Zemp et al.46 replace it with the angular spectrum Hence the ratio of the first order neglected nonlinear het-
technique.47 The present work aims at investigating that erogeneous terms g2 to the dominant linear terms  is
approach further. typically about 105 , a pretty small value that motivates
In a first part, this manuscipt describes the asymptotic the use of Eq. (1).
process to formalize the phenomenological decomposition Without any loss of generality, sound speed and density
between linear diffraction and non linearities. That pro- can be decomposed into a mean value (symbolized by an
cess is generalized to include weak heterogeneities. It is overbar), independent on the observation point, and a fluctu-
shown that the process consists of applying an approxima- ation (symbolized by a prime) that is position dependent:
tion of the paraxial type, but only on the heterogeneous
diffraction terms instead of the full diffraction. This dif- c0 ðxÞ ¼ c0 þ c00 ðxÞ
ference with the usual (standard or wide-angle) parabolic (3)
q0 ðxÞ ¼ q0 þ q00 ðxÞ:
approximation is discussed in the second part. Then the
numerical scheme is briefly presented. Validation tests
In the following, to get more concise expressions, we note:
demonstrate the benefits of the method compared to the
parabolic approximation, either for a piston source or for
a plane wave scattered by a heterogeneity. Finally, a nu- c20 ðxÞ ¼ c20 þ cðxÞ; (4)
merical simulation of the focusing of shock waves by an
acoustic lens illustrates the coupling between diffraction, where c ¼ c02
0 þ 2c0 c00 can be negative,
heterogeneity and nonlinearities. Now with a view to one-way approximation, we
select one particular (so-called “axial”) direction x, while
y and z denote the transverse ones. In a way similar to
II. THE GENERALIZED LIGHTHILL–WESTERVELT the one required to establish the KZ equation, we intro-
EQUATION duce the retarded time in the x direction, corresponding
Let us start from the generalized Lighthill–Westervelt to the time of flight of a plane wave propagating in the
equation:48 x > 0 direction in a homogeneous medium with the
"  # mean sound speed s ¼ t  x= c0 . Shifting from ðx; y; z; tÞ
  variables to ðx; y; z; sÞ, the generalized Lighthill–Wester-
@2/ 2 r/ bðxÞ @ @/ 2
c ðxÞq0 ðxÞr: ¼ (1) velt equation is transformed without any approximation
@t2 0 q0 ðxÞ q0 ðxÞc20 ðxÞ@t @t
into:

J. Acoust. Soc. Am., Vol. 130, No. 1, July 2011 Dagrau et al.: Acoustic shock wave propagation 21
 2 
@2/ 2 @ / @2/ @2/ C. The wide-angle parabolic approximation
c0
2  c0
 þ þ 2
@x@s @x2 @y2 @z A higher-order parabolic approximation can be obtained
 2 2

@ / 2 @ / 1 @ / @2/ @2/
2
in the following way. Instead of neglecting completely the
c  þ þ þ 2
@x2 c0 @x@s c20 @s2 @y2 @z second order derivative with respect to x, an approximation
    of it can be obtained by using the first order approximation
2
c0 @/ 1 @/ @q0 @/ @q0 @/ @q00
0 0
þ  þ þ of Eq. (8). Dropping nonlinear terms (which are three orders
q0 @x c0 @s @x @y @y @z @z
"  # of magnitude smaller than propagation ones), one gets:
2
b @ @/  
¼ : (5)
q0 c20 @s @s @3/ c0 @ @ 2 / @ 2 /
¼ þ 2 : (9)
@x2 @s 2 @x @y2 @z
The first two terms of the preceding equation correspond to
the linear wave equation in a homogeneous medium, the and substituting this into the time derivative of Eq. (6)
third one includes sound speed fluctuations, then comes den- yields:
sity fluctuations and, on the right-hand side, nonlinear   
effects. When comparing magnitude orders, these terms are, @3/ @ c0 @ @2/ @2/
2c0  c20 þ þ 2
respectively, of order , g, g, and 2 . @x@s2 @s 2 @x @y2 @z
"  #
b @2 @/ 2
III. ONE-WAY VERSUS PARABOLIC ¼ 2 : (10)
c0 q0 @s2 @s
APPROXIMATIONS
A. The one-way approximation When dropping nonlinear terms, that equation is the so-
We first consider the homogeneous case (c00
¼ 0 and called Claerbout or wide-angle approximation.36 Note it is
q00 ¼ 0) for which the nonlinear wave equation, Eq. (5), runs: possible to iterate the asymptotic process and get higher-
order parabolic approximations.
 2  "  #
@2/ @ / @ 2
/ @ 2
/ b @ @/ 2
2c0  c20 þ þ ¼ : D. Dispersion relations
@x@s @x2 @y2 @z2 q0 c20 @s @s
(6) Now we can compare, in the homogeneous linear re-
gime, the dispersion relations of the wave equation, Eq. (6),
The linear part of Eq. (6) is the exact wave equation in a ho- the standard parabolic equation, Eq. (8), and the wide-angle
mogeneous medium. Nevertheless the numerical method, parabolic equation, Eq. (10), for a plane wave
used for the resolution of this equation introduced in Sec. V,
needs to progress plane to plane. Therefore, a one-way / ¼ A exp½iðkx x þ ky y þ kz z  x0 t
approximation, not appearing explicitely in Eq. (6) but in its ¼ A exp½iððkx  k0 Þx þ k? :x?  x0 s: (11)
solution, is used in order to neglect the backscattered field
and to retain only the exact solution propagating toward with x? ¼ ð0; y; zÞ and k? ¼ ð0; ky ; kq
z Þ. ffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Normalizing
ffi the
x > 0. wave vector by ðkx ¼ kx =k0 ; k? ¼ ky2 þ kz2 =k0 Þ with
k0 ¼ x0 =c0 the wave number, one gets the following disper-
sion relations:
B. The standard parabolic approximation and the KZ
equation 8 qffiffiffiffiffiffiffiffiffiffiffiffiffi
>
> Wave equation kx ¼ 6 1  k?
2
>
>
The standard parabolic (or paraxial) approximation >
> qffiffiffiffiffiffiffiffiffiffiffiffiffi
assumes that axial amplitude variations are slow compared >
> 2
>
> One  way approximation kx ¼ þ 1  k?
to the ones due to the phase: >
< 2
k? (12)
> Standard parabolic approximation kx ¼ 1 
1 @/ @/ >
> 2
 ; (7) >
> 2
>
> k?
c0 @s @x >
> Wide angle parabolic approximation kx ¼ 1  12 :
>
> 2
: k?
1 4
so that the second order derivative with respect to the axial
variable x can be neglected. Then Eq. (6) reduces to the
These dispersion relations are plotted on Fig. 1. For the
well-known KZ equation:
wave equation, it is a circle of radius 1. Note that for trans-
 2  "  # verse wave numbers jk? j larger than 1, the axial wave num-
@2/ 2 @ / @2/ b @ @/ 2 ber becomes imaginary (evanescent waves). This is
2c0  c0 þ ¼ : (8)
@x@s @y2 @z2 q0 c20 @s @s represented on the curve by the two black dots. The one-way
approximation is identical to the wave equation but selects
Each term is associated to one distinct physical phenom- only the wave numbers with positive real or imaginary part,
enon: from left to right, axial propagation, transverse diffrac- e.g., waves propagating only toward x > 0 (hence its name)
tion, and nonlinearities. or waves exponentially decaying in that same direction. In

22 J. Acoust. Soc. Am., Vol. 130, No. 1, July 2011 Dagrau et al.: Acoustic shock wave propagation
of magnitude smaller), that approximation will be much bet-
ter than when it is performed on the full wave equation. At
first order, one can neglect the second order derivative with
respect to the axial variable x that appears in the heterogene-
ous term of Eq. (5). A higher-order “wide angle” approxima-
tion on the heterogeneous term can be obtained by noting
that, according to the dominant part of the equation (the lin-
ear homogeneous wave equation), one has:

@2/ 2 @2/ @2/ @2/


’   2; (13)
@x2 c0 @x@s @y2 @z

which, after substitution into the term associated to sound


speed fluctuations, yields:
 2 
@2/ 2 @ / @2/ @2/ c @2/
FIG. 1. Normalized dispersion relations for the wave equation (gray solid 2c0  c0 þ þ 
line), the one-way approximation (solid line), the standard (dashed line) and @x@s @x2 @y2 @z2 c20 @s2
wide-angle parabolic (dash-dotted line) approximations.    
c2 @/ 1 @/ @q00 @/ @q00 @/ @q00
þ 0  þ þ
the next section, we will see how to implement numerically q0 @x c0 @s @x @y @y @z @z
"  #
that one-way selection. The dispersion curve of the parabolic b @ @/ 2
approximation is the parabola osculating the circle at the ¼ 2
: (14)
q0 c0 @s @s
point ðkx ¼ 1; k? ¼ 0Þ corresponding to the x direction of
propagation. It is also the first order Taylor expansion of the
exact one. The dispersion relation of the wide-angle approxi- Finally, a further improvement can be made by noting that,
mation is the (1,1) order Padé expansion of the exact rela- in the case where c is constant, q00 ¼ 0 and nonlinear terms
tion. It is closer to the circle near the plane wave but are omitted, the dispersion relation of the preceding equation
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
deviates more for large transverse wave numbers. Note that 2
is (in the one-way approximation) kx ¼ 1  c=c20  k?
the wave equation and the one-way approximation admit qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
evanescent waves as solutions. On the contrary, for parabolic while the exact dispersion relation is kx ¼ c20 =c20  k? . In
approximations, all wave numbers are either propagating the preceding two relations, the wave numbers are scaled by
ones or even counter-propagating ones (for large transverse the mean wave number x0 =c0 . Consequently, the exact dis-
wave numbers), which is not physical. Also the wide-angle persion relation can be recovered by redefining c as:
parabolic approximation is singular for k? ¼ 2. So Fig. 1
 
clearly shows that the one-way approximation is indeed a c20
better one than any of the two parabolic approximations. cðxÞ ¼ c20 1 2 : (15)
c0 ðxÞ
Obviously it cannot simulate backscattered waves (neither
can any of the parabolic approximations), but it is exact for This finally leads to the equation we suggest calling the
forward propagation and even for evanescent waves. In prac- HOWARD equation:
tice, the standard parabolic approximation is considered to
 2    2
be valid for wave directions within a 6 15 angular deviation @2/ @ / @2/ @2/ c20 @ /
from the plane wave, while the wide-angle extends that c0
2  c20 þ þ  1 
@x@s @x2 @y2 @z2 c20 ðxÞ @s2
range to 6 30 . In comparison, the one-way approximation   
is valid up to 6 90 . c2 ðxÞ @/ 1 @/ @q0 @/ @q0 @/ @q0
þ 0  þ þ
q0 ðxÞ @x c0 @s @x @y @y @z @z
"  #
2
IV. GENERALIZATION TO THE HETEROGENEOUS bðxÞ @ @/
CASE ¼ : (16)
q0 ðxÞc20 ðxÞ @s @s
The results of the preceding section can now be general-
ized to the heterogeneous case. The basic idea is to separate Note that each term of the HOWARD Eq. (16) is associated
in Eq. (5) homogeneous diffraction from heterogeneous to one distinct physical phenomenon: from left to right,
terms. The homogeneous case will be treated with the same propagation, homogeneous diffraction, sound speed hetero-
one-way approximation that relies on explicit solutions of geneity, density heterogeneity, and nonlinearities. Let us
the wave equation. Exact solutions do not exist for the heter- recall finally the approximations we have made to obtain
ogeneous case. Indeed, heterogeneities will create backscat- that equation. First, cubic nonlinear and quadratic nonlinear
tering that prevent an exact one-way approach anyway. heterogeneous terms are omitted; second, a quasi plane wave
Hence an additional approximation has to be performed by (or high frequency geometrical) approximation is performed
making a wide-angle parabolic approximation not on the full on the nonlinear term; third, a wide-angle parabolic approxi-
diffraction terms but only on the heterogeneous part of it. As mation is performed on the sound speed fluctuation term.
heterogeneity is assumed to be weak (in practice two orders The HOWARD equation is exact in the homogeneous case

J. Acoust. Soc. Am., Vol. 130, No. 1, July 2011 Dagrau et al.: Acoustic shock wave propagation 23
and also handles density fluctuations exactly for the for- B. Marching algorithm
warded waves. Concerning sound speed fluctuations,
Operator splitting27 was chosen to solve Eq. (17) in a rec-
although the dispersion relation is exact, the final equation
tangular computational domain: ½X1 ; XNX   ½Y1 ; YNY  with
remains nevertheless approximated to the case of weak con-
regular spacing Xn ¼ X1 þ ðn  1ÞDXðn ¼ 1toNx Þ and
trasts. Also not appearing per se in the HOWARD equation
Yj ¼ Y1 þ ðj  1ÞDYðj ¼ 1to Ny Þ. Here Nx and Ny are the
but required in the numerical resolution, is the one-way
number of points in, respectively, the x and y directions. The
approximation, which amounts to neglect the backscattered
equation naturally splits into the following three subequations:
field. Indeed we will see later on that last approximation is
1. Linear equation for the homogeneous diffraction
by far the most constraining one.
effects:

V. THE NUMERICAL ONE-WAY ALGORITHM @2U @2U @2U


¼D 2þ 2; (18)
@X@T @X @Y
A. Dimensionless form of the HOWARD equation
In two dimensions, the dimensionless form of the 2. Linear equation for the heterogeneous effects:
HOWARD equation is:
@2U @2U @U @U
¼ CðXÞ 2 þ RX ðXÞ þ RY ðXÞ
@2U @2U @2U @2U @U @X@T @T @X @Y
¼ D 2 þ 2 þ CðXÞ 2 þ RX ðXÞ @U
@X@T @X @Y @T @X þ RT ðXÞ ; (19)
  @T
@U @U @ @U 2
þ RY ðXÞ þ RT ðXÞ þ lðXÞ ; (17)
@Y @T @T @T 3. Equation for the nonlinear effects:

where  
@2U @ @U 2
¼ lðXÞ : (20)
(a) U ¼ /=/0 is the dimensionless potential and /0 the @X@T @T @T
characteristic value of the potential,
(b) P ¼ pa =P0 is the dimensionless acoustic pressure with The advancement term @ 2 U=@X@T is the common term ena-
P0 ¼ x0 /0 is for instance the peak overpressure at the bling the split-step algorithm to couple all the preceding
source, three equations in a suitable way to recover the initial
(c) Y ¼ 2y=a is the dimensionless transverse variable and a HOWARD equation. Note this choice is not unique and
is one characteristic length of the problem [for instance other splittings could be used. Nevertheless, following the
either the source (or scatterer) radius (or diameter)], physical effects is frequently an efficient way to develop nu-
(d) X ¼ 2x=ðk0 a2 Þ is the dimensionless variable in the direc- merical algorithms. Moreover, it allows a versatile use of the
tion of propagation; if a is chosen as the radius, it is final software that can be used in various situations, includ-
scaled by the Rayleigh distance R chosen as the charac- ing or not some of the effects present in the HOWARD
teristic length in the longitudinal direction, equation.
(e) X ¼ ðX; YÞ is the dimensionless position vector,
(f) T ¼ x0 s is the dimensionless time, C. Resolution of the diffraction equation
(g) D ¼ 1=k02 a2 ¼ k=4pR is a measurement of the wave-
length relative
 to the Rayleigh
 distance, The resolution of the diffraction equation relies on a
(h) CðXÞ ¼ 1  c20 =c20 ðXÞ k02 a2 = 4 is the coefficient meas- classical analytical solution in the spectral domain. By intro-
uring the effect of the sound speed heterogeneities rela- ducing the Fourier transform operators in time and in space:
tive to the diffraction effects,
 
(i) RX ðXÞ ¼ @q0 ðXÞ=@X c20 ðXÞ=q0 ðXÞ c20 k02 a2 is the coef- 
F T fUðX; Y; TÞg ¼ UðX; Y; xÞ
ficient measuring the effects of density heterogeneities in ð þ1
1
the x direction, ¼ pffiffiffiffiffiffi UðX; Y; TÞ expðixTÞdT; (21)
  2p 1
(j) RY ðXÞ ¼ @q0 ðXÞ=@Y c20 ðXÞ=q0 ðXÞ c20 is the coeffi-
cient measuring the effect of density heterogeneities in
the y direction, 
F Y fUðX; ^
Y; xÞg ¼ UðX; KY ; xÞ
(k) RT ðXÞ ¼ @q0 ðXÞ=@Xðc0 ðXÞ=2q0 ðXÞ c0 Þ is the coefficient ð þ1
1 
measuring the effects of density heterogeneities in the re- ¼ pffiffiffiffiffiffi UðX; Y; xÞexpðiKY YÞdY; (22)
tarded time direction, 2p 1
(l) lðXÞ ¼ bðXÞU0 k03 c0 a2 =4q0 ðXÞc20 ðXÞ is the coefficient of
the equation of diffraction Eq. (18) becomes:
nonlinear effects
Note that, when parameter D is small, diffraction effects ^
@2U ^
@U
D þ ix ^ ¼ 0:
 KY2 U (23)
are small, the sound beam is directive, and the parabolic @X2 @X
approximation is satisfied. The main benefits of the present
method appear when D is of order 1 because of either the The latter equation is an ordinary differential equation of
source size or the presence of one (or several) scatterers. second order. The solution is the superposition of two waves

24 J. Acoust. Soc. Am., Vol. 130, No. 1, July 2011 Dagrau et al.: Acoustic shock wave propagation
propagating in opposite directions in X. To be consistent of X is kept. That choice explains the reason why the method
with the marching algorithm, only the part of the solution is only valid in the one-way approximation. The solution of
that corresponds to propagation toward the increasing values Eq. (23) is then:

8  pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 
>
> ^ ixþi x2 4DKY2
>
< UðX; K Y ; xÞ exp 2D DX if 4DKY2 < x2
^
 
UðX þ DX; KY ; xÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  (24)
>
> ^ ix 4DKY2 x2
>
: UðX; K Y ; xÞ exp DX if 4DKY2  x2 :
2D

In the preceding solution, the term proportional to ix=D sionless width of the physical computational domain, and
comes out from the fact that the observer is moving with the 2ðH þ LÞ being the total thickness of the computational do-
mean sound speed (retarded time). The second case main in the Y direction. The value A0 ¼ 29:4 c0 =x0 is the
4DKY2 > x2 corresponds to evanescent waves that are fully dimensionless amplitude factor of the absorbing boundary
taken into account by the method and satisfy the exact dis- layers, tuned to give best performances. ABLs are known to
persion relation, as for propagating waves. The solution in suppress reflections for waves perpendicular to the lateral
the physical space ðX; Y; TÞ is then recovered by taking the boundaries and to significantly reduce reflections for other
inverse (in time and space) Fourier transforms of solution directions. More sophisticated techniques such as perfectly
Eq. (24). Hence the numerical method combines the accu- matched layers51 could be implemented, but it was not the
racy of spectral solutions, which is exact in the Fourier do- point of the present study to go in that direction, and the cur-
main, and the speed of computation due to standard fast rently implemented method was simple and sufficient for the
Fourier transform (FFT) algorithms. studied cases.
Note the same numerical method could allow us to
D. Absorbing boundary layers incorporate physical absorption (such as thermoviscosity or
relaxation) into the modeling. However, we have chosen not
To reduce the size of the computational domain, artifi-
to introduce an additional physical mechanism to be dis-
cial reflections on lateral boundaries are reduced by incorpo-
cussed as our study focused on heterogeneities.
rating a nonphysical absorbing boundary layer (ABL) within
the HOWARD modeling. This is achieved by introducing a
new term into the heterogeneous equation. The time Fourier
transform of this equation (Eq. 19) is: E. Resolution of the heterogeneous equation
To solve the equation taking into account the heteroge-
 
@U  
ðix þ RX ðXÞÞ ¼ x2 CðXÞ þ ixRT ðXÞ U  RY ðXÞ @ U : neous effects modified to incorporate the absorbing bound-
@X @Y ary layers (Eq. 26), a change of unknown function is
(25) performed, from U  That change permits to solve ana-
 to W.
lytically the effects due to the heterogeneity of the sound
Absorbing boundary layers can be added, in a heuristic way, speed for a plane wave:
by modifying that equation in the following manner:

UðX; 
Y; xÞ ¼ WðX; Y; xÞ expðCðX; Y; xÞÞ; (28)
 
@U 
ðix þ RX ðXÞÞ ¼ x2 CðXÞ þ ixRT ðXÞ þ ixAðYÞ ÐX
@X where CðX; Y; xÞ ¼ 0 ½x2 CðXÞ þ ixRT ðXÞ þ ixAðY; xÞ=
 ½ix þ RX ðXÞdX.
  RY ðXÞ @ U :
U (26)
@Y The new equation now to solve incorporates all the het-
erogeneous effects related to any deviation from a plane
The new term AðYÞ, depends only on Y and is chosen to wave:
increase progressively within the absorbing layer to damp  
more and more the waves inside that non-physical layer 
@W RY ðXÞ @C  RY ðXÞ @C @ W
¼ W : (29)
while avoiding reflections at the boundary with the physical @X ix þ RX ðXÞ @Y ix þ RX ðXÞ @Y @Y
domain. The functional form employed here is50:
It is numerically solved with a standard implicit and finite
2
AðYÞ ¼ A0 ðjYj  ðL  HÞÞ if jYj > L  H differences scheme. We note W  n ; Yj ; xÞ. The first
 n ¼ WðX
(27) j
AðYÞ ¼ 0 if jYj < L  H order derivative with respect to X is approximated by a first
order upwind finite differences stencil, while the first order
with H being the dimensionless width of the absorbing derivative with respect to Y is approximated by a second
boundary layer on each side (coming back with dimensions order centered finite differences stencil. Function C is com-
its typical value is several wavelengths), 2L being the dimen- puted similarly by the standard, second-order, trapezoidal

J. Acoust. Soc. Am., Vol. 130, No. 1, July 2011 Dagrau et al.: Acoustic shock wave propagation 25
rule, and its derivative by second order centered finite differ- That interpolation can be of any order, but in practice, linear
ences. The numerical approximation of Eq. (29) is then: interpolation turns out to be sufficient. That interpolation
     allows comparison of the different values of the potential
DX RY ðXn ; Yj1 Þ  n RY ðXn ; Yjþ1 Þ  n and selection of the maximum one to satisfy the weak shock
Wj1 þ Wjþ1
2DY ixþ RX ðXn ; Yj1 Þ ixþRX ðXn ; Yjþ1 Þ theory and the second principle of thermodynamics. That nu-
  merical step is easily implemented and makes the overall
DX RY ðXn ; Yj ÞDX
þ 1þ ðCðXn ; Yjþ1 Þ CðXn ; Yj1 ÞÞ procedure numerically efficient. The details of the numerical
2DY ixþ RX ðXn ; Yj Þ
method and the extension to other nonlinear equations can
 n
W ¼W  n1
(30)
j j1 be found elsewhere.33 Various examples of application in
the framework of KZ equation can be found, for instance in
This is valid for n > 1 and j 2 ½2; NY  1. It is assumed that the case of shock focusing at caustic cusps18 or irregular
the solution at n ¼ 1 is known (this is the input boundary (von Neumann) reflexion of shock waves.20
condition). At the lateral boundary conditions j ¼ 1 and
j ¼ NY , a zero field condition Wn ¼ W  n ¼ 0 is imposed to VI. VALIDATION TESTS
1 NY
be consistent with the use of the spectral method for the re-
solution of the homogeneous diffraction equation. Indeed the A. Linear homogeneous case: The piston source
spectral method used to solve that equation implies a perio- To illustrate the benefit brought by the present one-way
dicity of the field in the y direction. Consequently, pressure approach in comparison to the usual parabolic one, two se-
or potential in Y ¼ Y1 and Y ¼ YNY have to be the same. For ries of validation tests have been performed through compar-
sake of simplicity, we choose this value to be equal to zero. isons with analytical solutions in the linear case. Nonlinear
Moreover, this choice is consistent with the introduction of cases, for which no practical analytical solution is known,
absorbing boundary conditions. will be examined in the following subsection. We first inves-
For each frequency, the linear tridiagonal system Eq. tigate the homogeneous case, illustrated by a “piston” of size
(30) is solved with the standard Thomas’ algorithm. As it is 2a equal to four wavelengths, radiating a pure tone into a
an implicit scheme, it is unconditionnally stable, there is no two-dimensional fluid. We use the term “piston” even if we
associated CFL condition. Once the solution of this linear impose the pressure at the boundary of the computational do-
system is computed, the Fourier transform of the potential, main and not the normal velocity. That difference is mean-

UðX; Y; xÞ is recovered by applying Eq. (28). Finally the ingless in the standard parabolic approximation but has a
potential UðX; Y; TÞ in the physical domain is computed by sense for higher order approximations where imposing a uni-
using the inverse Fourier transform (in time). form pressure is not equivalent to a uniform velocity. The
pressure amplitude along the initial line x ¼ 0 is assumed to
F. Resolution of the nonlinear equation be equal to one over the piston surface 2a and zero other-
wise. The numerical domain is equal to 15 wavelengths in
The nonlinear equation, Eq. (20), is computed with the
the axial direction and 60 wavelengths in the transverse one
quasi-analytical Burgers-Hayes method.52 This method relies
with an absorbing layer starting laterally at 15 wavelengths
on the analytical but implicit Poisson solution of Eq. (20).
on each side. The discretization is equal to 40 points per
Physically, that solution is valid while the solution remains sin-
wavelength in the axial direction and 22 in the transverse
gle valued (no shock case) but is not appropriate if the solution
one. One time period is discretized by 2 048 points. The grid
is multivalued26 because of shock formation. In the latter case,
the weak shock theory has to be used in addition to the Poisson
solution. For the pressure field, Landau showed the physical
solution could be determined according to the law of equal
areas.53 For the potential field, if the nonlinearities are quad-
ratic (as they are here), the physical solution can be shown to
correspond to the maximum of the potential.54 This is inti-
mately related to the increase of entropy through the shocks,
which can be only compression ones (e.g., the pressure after
the shock Pþ is always larger than the pressure before the
shock P < Pþ ). Hence, the analytical solution of Eq. (20) at
step X þ DX can be deduced from the solution assumed to be
known at previous step X through the formula:
"  2 #
@U
UðX þ DX; Y; TÞ ¼ max UðX; Y; hÞ  lDX ðX; Y; hÞ
@h
@U
T ¼ h  lDX ðX; Y; hÞ (31) FIG. 2. Pressure amplitude (gray levels in dB, relative to input pressure)
@h versus axial and transverse distances normalized by wavelength, radiated by
a pure tone piston source of size four wavelengths at 2D: (a) analytical solu-
Numerically, the multivaluated analytical solution Eq. (31) tion, (b) HOWARD approximation, (c) standard parabolic approximation,
requires interpolation on the initial discrete points in time T. (d) wide-angle parabolic approximation.

26 J. Acoust. Soc. Am., Vol. 130, No. 1, July 2011 Dagrau et al.: Acoustic shock wave propagation
was made finer than necessary to reduce any convergence
problem and to compare the precision of the various approxi-
mations. Figure 2 shows the overall pressure field of the ana-
lytical solution compared to the three approximations
(one-way, standard, and wide-angle parabolic). Figure 2 is
presented in decibel scale in gray levels to better visualize
the field dynamic. Parabolic approximations Eqs. (8) and
(10) are solved in the frequency domain by a finite difference
scheme, centered and second order in the transverse direc-
tion, and upwind first order in the axial one (with the same
discretization as used with HOWARD). The reader is
referred to Ref. 20 and subsequent references for more infor-
mation and validation tests of the present KZ numerical
solver. Note that the upwind scheme naturally introduces nu-
FIG. 3. Normalized axial pressure amplitude radiated by a pure tone piston
merical dissipation, thus the large transverse wavenumbers source of radius equal to two wavelengths at 2D in the nonlinear regime:
that strongly deviate from the exact dispersion relation are HOWARD approximation (solid line), standard parabolic approximation
damped. This is especially important for those wave numbers (dashed line). From left to right and top to bottom: fundamental, second,
third, and fourth harmonics.
that are propagating according to the parabolic approxima-
tions, while they are evanescent according to the exact dis- are visible for the higher order harmonics, but they are less
persion relation. An energy-preserving scheme in the axial visible because the generation of harmonics is progressive
direction (like Crank-Nicholson) would produce artificial and therefore they have low amplitude in the very nearfield.
side lobes in direction up to 6 90 that would then reflect on The transverse field is visible on Fig. 4 at the distance
the domain boundary and scramble the result. Note that this x=k ¼ 14:6, or 1.16 Rayleigh distance from the source, again
problem is automatically avoided by the HOWARD method, for the first four harmonics. For the fundamental, as for the
which does not introduce numerical dissipation in the homo- linear case, the main lobe (up to y ¼ 3k) is well described by
geneous case. Indeed it conserves the energy while preserv- both the KZ and HOWARD approaches. Beyond y ¼ 3k, the
ing the dispersion relation. Clearly visible on Fig. 2 is the phase error of the paraxial approximation induces a significant
benefit brought by the one-way method, that is, almost exact mismatch for the side lobes. Beyond the point y ¼ 15k, the
(except the very small error due to discretization) as the numerical dissipation introduced by the numerical scheme
physics of wave propagation is here indeed fully one way. used to solve the KZ equation makes the field to decrease
All the details of field variations are precisely captured. It is exponentially. Indeed the numerical scheme is based on an
also interesting to see the benefit brought by the wide-angle implicit finite differences scheme known to introduce artificial
approximation compared to the standard one: it better cap- numerical attenuation. The algorithm used to solve the
tures the main details of the nearfield and the first side lobe HOWARD equation conserves energy (in the homogeneous
in the farfield. However, because of the inaccuracy of the case) and still shows side lobes. For the harmonics, the trend
dispersion relation, it cannot capture higher lobes. As is similar. However, when increasing the harmonic number,
expected, the standard parabolic approximation performs the side lobe oscillations are reduced, so that the domain of
less satisfactorily as it captures only the main lobe of the validity of the paraxial approximation slightly increases (up to
field and is inadequate in the nearfield. Other validation tests y ¼ 5k for the fourth harmonic). As a counterpart, due to the
(not shown) by varying the piston size lead to the same overestimated lateral damping induced by the numerical
conclusions.

B. Nonlinear homogeneous case: The piston source


The same piston source is now simulated in a nonlinear
case with the same numerical parameters. The acoustic
Mach number  ¼ 0:0098 is chosen so that the shock forma-
tion distance LS ¼ 1=ðbk0 Þ is comparable to the Rayleigh
distance LR ¼ k0 a2 =2, the ratio between these two distances
being here equal to 1.077 for air (b ¼ 1:2). This case has
been chosen so that significant nonlinear effects take place
in the near field.
Figure 3 presents the axial pressure amplitude for the first
four harmonics with the HOWARD simulation compared to
the KZ one. As expected, results are in good agreement
because the KZ equation is most valid near the x axis. The FIG. 4. Normalized transverse pressure amplitude (in dB, relative to input
pressure) radiated by a pure tone piston source of radius equal to two wave-
largest difference occurs in the nearfield of the fundamental, lengths at 2D in the nonlinear regime: HOWARD approximation (solid
where, as for the linear case, diffraction effects are not fully line), standard parabolic approximation (dashed line). From left to right and
described by the parabolic approximation. Some differences top to bottom: fundamental, second, third, and fourth harmonics.

J. Acoust. Soc. Am., Vol. 130, No. 1, July 2011 Dagrau et al.: Acoustic shock wave propagation 27
scheme associated with the KZ equation, the pressure ampli-
tude decreases much too fast laterally.
Other numerical tests have been performed (not shown)
by varying the size of the source relative to the wavelength.
As expected, when decreasing the source size, the mis-
matches between the HOWARD and KZ equations increase
as the source gets less and less directive and deviate more
and more from the paraxial approximation of the KZ equa-
tion. Differences get visible for the axial field of the harmon-
ics, and, in the farfield, side lobes are less and less correctly
described by the KZ equation (mostly in phase but also in
amplitude). On the contrary, for large sources with directive
beam fields, the agreement between the two methods is good
except for the very nearfield and for very lateral side lobes. FIG. 6. Normalized transverse pressure amplitude for a plane wave scat-
tered by a cylinder of diameter two wavelengths and 5% sound speed con-
trast at 2D: wave equation (gray solid line), the HOWARD one-way
C. Linear scattering by an heterogeneous cylinder approximation (solid line) and the standard (dashed line), and wide-angle
parabolic (dash-dotted line) approximations. Simulations at respectively (a)
The second case deals with the forward scattering of a 1, (b) 3, (c) 5, and (d) 7.5 wavelengths from the cylinder center.
harmonic plane wave incident on a cylinder of radius one
wavelength with the same density as the ambient fluid but solutions can capture the (small) axial field oscillations
with a sound speed, that is, 5% larger. The cylindrical case before the cylinder due to the backscattered field as they are
was chosen because an analytical solution is available, under all based on a forward marching method. Of course, this
the form of a Bessel series.55 Morevover, it will not induce error will induce an error on the forward scattered field even
any sharp diffraction effects at corners as would a square or for the HOWARD method even though this one satisfies the
rectangular inclusion. Note that for atmospheric applications, exact dispersion relation. That error is due to the neglected
5% sound speed fluctuations correspond to about 30 C tem- backscattered field. It is observable on the axial field, where
perature fluctuations, which is considerable. The numerical the wide-angle parabolic approximation slightly improves
domain is equal to 15 wavelengths in the axial direction and results compared to standard parabolic and is comparable
60 wavelengths in the transverse one with an absorbing layer with HOWARD method (which is slightly better when ob-
starting laterally at 15 wavelengths on each side. The discre- servation distance increases).
tization is equal to 34 points per wavelength, both in the The transverse fields (Fig. 6) clearly show the
axial and in the transverse direction. Such a fine discretiza- HOWARD method almost perfectly captures the phase of
tion is needed here to capture the cylinder geometry using the transverse oscillations as expected from theory when
cartesian coordinates. Coarser grids would induce artificial examining the dispersion relation. However, because it does
scattering by the corners of the discretized cylinder and not take into account the energy of the backscattered field, a
would produce spurious field oscillations. small error occurs for the amplitude (on the order of 0.01,
Figures 5 and 6 show the axial and transverse pressure that is, less than 1% of the incident amplitude, slightly more
fields, respectively, of the analytical solution compared to very close to the cylinder). As expected, standard parabolic
the three numerical approximations (HOWARD, standard, approximation, which has a large error on dispersion rela-
and wide-angle parabolic). The scale of the figures is linear. tion, induces significant errors for the second and higher
Clearly visible on Fig. 5 is that none of the three numerical fringes of interferences. The wide-angle parabolic approxi-
mation significantly improves the precision49 but is not as ef-
ficient as the HOWARD method. Other simulations (not
shown) lead to similar conclusions for weaker heterogene-
ities (1%) more comparable to what is expected for atmos-
phere with of course a higher precision as the contrast is
weaker. For stronger heterogeneities (25%), none of the
three numerical methods provide fully satisfactory results as
the sound speed contrast is too large to satisfy the assump-
tion of weak backscattering. However, such strong sound
speed fluctuations are unrealistic in terms of applications for
propagation in air, ocean, or biological media.

VII. SHOCK WAVE FOCUSING BY AN ACOUSTICAL


LENS
FIG. 5. Normalized axial pressure amplitude for a plane wave scattered by
a cylinder of diameter two wavelengths and 5% sound speed contrast at 2D: A. Numerical convergence
wave equation (gray solid line), the HOWARD one-way approximation
(solid line) and the standard (dashed line) and wide-angle parabolic (dash- We finally investigate the nonlinear and heterogeneous
dotted line) approximations. case. We selected the case of shock wave focusing by an

28 J. Acoust. Soc. Am., Vol. 130, No. 1, July 2011 Dagrau et al.: Acoustic shock wave propagation
acoustical lens (due for instance to temperature fluctuations) this point. Convergence in pressure amplitude is obtained for
and producing a cusp caustic. This is a demanding case from a time discretization of about 1 024 points. Figure 8 illus-
a numerical point of view as it is known that the amplitude trates the behavior of the time waveform at the geometrical
of sharp shocks at caustics results from a balance between focus with a zoom around the peak overpressure given Fig.
nonlinear and diffraction effects.56,57 Moreover, diffraction 9. One can observe convergence in pressure amplitude,
transforms a shock wave into a spike wave around the caus- including the peak overpressure, for a time discretization of
tic39,58 (because of the p=2 phase shift), which is more diffi- about 1 024 points. Note, however, that because no physical
cult to capture numerically.18 That is the reason why a absorption has been introduced in the problem, no conver-
numerical convergence is a good test. gence is obtained for the shock rise time as this one is con-
We consider the case of a plane shock wave incident on trolled by absorption, which is here purely numerical and
a medium with smooth variations of the sound speed under hence remains dependent on the numerical discretization.
the form of a spherical converging lens, with sound speed: When observing the whole signal, one can see the wave

h
i spiking characteristic for nonlinear focusing. When observ-
c0 ¼ c0 1  a exp  ðx  x0 Þ2 þ y2 =L2h ; (32) ing the zoom, the difference between the linear and nonlin-
ear cases are very visible. In the linear case, the signal is
very spiky. Note that, theoretically, in an inviscid medium, a
where a ¼ 0:07 so that the minimum sound speed at the cen-
focused shock is unbounded.56,58 Here, it remains artificially
ter of the lens is 7% lower than the ambient sound speed and
bounded only due to the numerical dissipation. In the nonlin-
with Lh ¼ 5k. This will produce a caustic whose geometrical
ear case, the peak amplitude is clearly bounded (as proved
cusp is located at about 22:6k from the lens center and 32
by the convergence in amplitude) so that the shock now
wavelengths from the initial plane. The lens is located at dis-
looks locally closer to a step shock. This limitation arises
tance x0 ¼ 9:775k so that heterogeneous effects are negligi-
from a balance between diffraction effects (that tend to focus
ble in the initial plane. The acoustic Mach number is chosen
more efficiently high frequencies) and nonlinear ones (that
equal to  ¼ 9:765 104 .
act more efficiently on the parts of the signal with the highest
In a first step, we check the numerical convergence by
amplitudes). In the ideal case of a purely incident step shock,
computing the case of an incident periodic saw-tooth wave,
this balance leads to a similitude law, according to which the
which is better suited for a convergence test. The numerical
peak overpressure varies as the power 2=3 of the incident
domain is 300 wavelengths wide in both directions with
amplitude (see Ref. 56 for theoretical demonstration and
ABL thickness 15 wavelengths on each side. In the trans-
Ref. 18 for numerical validation). Note finally the peak over-
verse direction, 8 192 points are used and 4 096 points in the
pressure occurs earlier in time in the nonlinear case because
axial one. Results are most sensitive to the time discretiza-
of the dependance of the sound speed with pressure.
tion, which varies from 128 to 8 192 points. The conver-
gence test of the pressure amplitude at the geometrical focus
is illustrated by Fig. 7. One clearly observes the continuous
amplitude increase with mesh refinement in the linear case,
while nonlinear effects make the focused shock bounded. In
this case, the amplification factor is about 6. Note that in the
nonlinear case, the point of maximum overpressure is not
right at the focus as nonlinear effects tend to slightly shift

FIG. 7. Normalized pressure amplitude at the geometrical focus versus FIG. 8. Normalized peak overpressure of the time waveforms at the geo-
number of points N of the time discretization per period. Upper curve (þ): metrical focus for various number N of points of the time discretization per
linear case, lower curve (h): nonlinear case. period. Upper figure: linear case, lower figure: nonlinear case.

J. Acoust. Soc. Am., Vol. 130, No. 1, July 2011 Dagrau et al.: Acoustic shock wave propagation 29
FIG. 11. Pressure (in Pascals) waveform versus time (in seconds) along the
axis at (a) x ¼ 24k (geometrical focus), (b) x ¼ 39k, and (c) x ¼ 95k wave-
lengths from the lens center and (d) off axis (x ¼ 95k and y ¼ 0:9k). Axial
distances are measured from the cylinder center.

is 68 m for a sound speed in air of 340 m/s. Figure 10


presents the two-dimensional pressure plots (pressure in gray
level versus time and transverse direction) at different dis-
FIG. 9. Zoom around the shock for the normalized peak overpressure of the tances, respectively, at 0 [lens center, Fig. 10(a)], 24 [almost
time waveforms at the geometrical focus for various number N of points of geometrical focus, Fig. 10(b)], 39 [Fig. 10(c)], and 95 [Fig.
the time discretization per period. Upper figure: linear case, lower figure:
10(d)] wavelengths from the lens center. One can clearly see
nonlinear case.
on the first plot the two wavefront (head and tail shocks) cur-
vatures due to the lens in the prefocal region. At the focus,
B. N-wave focusing
the wavefront is locally singular, and the signal is highly
This final section deals with the case of an N-wave as amplified (by a factor almost 4) within a tiny region. In the
could be observed for a sonic boom. Parameters for the N- postfocal region, the wavefronts clearly display a swallow-
wave are chosen similar to those of the sonic boom from a tail pattern typical for cusp caustics and in agreement with
conventional supersonic aircraft with an amplitude of 100 Pa catastrophes theory,56,59 which expands in space as one
and a duration of 0.2 s. According to the previous case, we propagates away. Axial time waveform signals are shown on
can estimate about 1 000 points are necessary within that Fig. 11 for the same positions (except the first one, not
time interval for convergence. However, because the signal shown, for which the N-wave is almost undistorted). One
is now transient, the time window of the numerical domain can clearly observe the typical U-wave at the focus,18,19,56
has to be chosen much longer especially after the signal then the progressive separation between the N- and U-waves,
(because the sound propagates slower in the lens, it will depending on the part of the signal that has already (1 ray-U
arrive later). Hence the time window is chosen about 8 times wave) or not yet (2 rays-N-wave) tangented the caustic. Off
longer, and the number of points in time is thus 8 192. Other axis, the two rays that have not yet tangented the caustic
physical (acoustical lens) and numerical parameters are arrive at different times and hence the N-wave itself splits
selected identical to the previous case. The wavelength here into two. The resulting signal exhibits six small amplitude
shocks (3 at the front and 3 at the tail). In a random medium,
this mechanism of wavefront focusing and folding has been
identified as the key feature for shock wave scattering and
increase of the rise time.23,60

VII. CONCLUSION AND OUTLOOKS


The objective of the present work was to develop a nu-
merical approach to handle shock wave scattering in a weakly
heterogeneous medium. Weak heterogeneities are present in
all realistic media: air, ocean, or biological tissues. Paraxial
approximations (either KZ equation or wide-angle parabolic
approximation) are very efficient one-way approaches, but
they nevertheless have some unavoidable limitations and
remain restricted to regions near the axis of propagation. In a
FIG. 10. Pressure (in Pascals) versus time (in seconds) and transverse direc-
heterogeneous medium, scattering will occur at any angle,
tion (normalized by wavelength) at (a) 0 (lens center), (b) 24 (geometrical and the need for a method going beyond these limitations is
focus), (c) 39, and (d) 95 wavelengths from the lens center. obvious. Such ad hoc numerical approaches had already been

30 J. Acoust. Soc. Am., Vol. 130, No. 1, July 2011 Dagrau et al.: Acoustic shock wave propagation
successfully performed by several authors, splitting the linear ACKNOWLEDGMENTS
forward propagation part and the nonlinear one. However,
The present work has been realized through a partner-
they handled only the homogeneous case and did not provide
ship between Dassault Aviation and Université Pierre et
an explicit mathematical formalism. In this paper, this formal-
Marie Curie/Institut Jean Le Rond d’Alembert/CNRS (UMR
ism is established and extended to a weakly heterogeneous
7190). Most of the results obtained during this partnership
case. Heterogeneities are handled through a wide-angle para-
are summarized in PhD dissertation of F.D.49 The authors
bolic approximation but performed only on the small hetero-
are grateful to Nicolas Héron (Dassault Aviation) for useful
geneous perturbation instead of the whole diffraction term. As
discussions.
a consequence, the resulting equation is more precise and
recovers the same dispersion equation as the exact wave equa- 1
D. J. Maglieri and K. J. Plotkin “Sonic boom,” Aeroacoustics of Flight
tion but for forward propagating waves only (including evan-
Vehicles, edited by H. H. Hubbard (Acoustical Society of America, New
escent waves). Hence the main approximation that underlies York, 1995), vol. 1, pp. 519–561
2
the present theory is the one-way approximation, e.g., the fact A. McAlpine and M. J. Fisher, “On the prediction of buzz-saw noise in
that the backscattered field is neglected. aero-engine inlet ducts,” J. Sound Vibrat. 248, 123–149 (2001).
3
K. L. Gee, V. W. Sparrow, M. M. James, J. M. Downing, C. M. Hobbs, T.
The so-called “HOWARD” numerical approach is then B. Gabrielson, and A. A. Atchley, “The role of nonlinear effects in the
implemented, combining through a three level split-step pro- propagation of noise from high-power jet aircraft,” J. Acoust. Soc. Am.
cess several classical approaches: angular spectrum in the 123, 4082–4093 (2008).
4
frequency domain for forward propagation in the homogene- J. N. Ambrosiano, D. Plante, B. E. McDonald, and W. A. Kuperman,
“Nonlinear propagation in an ocean waveguide,” J. Acoust. Soc. Am. 87,
ous case, finite differences in the frequency domain for the 1473–1481 (1990).
heterogeneous part, and the shock fitting Burgers-Hayes 5
K. Castor, P. Gerstoft, Ph. Roux, W. A. Kuperman, and B. E. McDonald,
method in time domain for nonlinearities. Various numerical “Long-range propagation of finite-amplitude acoustic waves in an ocean
validations have been presented for comparisons with either waveguide,” J. Acoust. Soc. Am. 78, 672–681 (2004).
6
M. R. Bailey, V. A. Khokhlova, O. A. Sapozhnikov, S. G. Kargl, and L.
analytical solutions or numerical simulations of the KZ A. Crum, “Physical mechanisms of the therapeutic effect of ultrasound - a
equation or its wide-angle extension. It shows the parabolic review,” Acoust. Phys. 49, 369–388 (2003).
7
approximations provide a precise output where they are sup- F. A. Duck, “Nonlinear acoustics in diagnostic ultrasound,” Ultrasound
posed to do (near the axis), but the present method really Med. Biol. 28, 1–18 (2002).
8
A. J. Coleman and J. E. Saunders, “A review of the physical properties
improves the accuracy of the simulation in the very nearfield and biological effects of the high amplitude acoustic fields used in extra-
of a piston source or far from the axis (side lobes). Espe- corporeal lithotripsy,” Ultrasonics 31, 75–89 (1993).
9
cially, the HOWARD method preserves the correct spatial M. Delius, “History of shock wave lithotripsy,” Nonlinear Acoustics at the
Turn of the Millenium, Proc. 15th Int. Symposium on Nonlinear Acoustics
phase of the signal. For a harmonic source, compared to par-
1999, edited by W. Lauterborn and T. Kurz (AIP, Melville, NY, 1999),
abolic approximations, the benefit of the HOWARD pp. 23–32.
10
approach is higher for the fundamental frequency as har- M. S. Canney, M. R. Bailey, L. A. Crum, V. A. Khokhlova, and O. A. Sap-
monic fields better satisfy the paraxial conditions, but, far ozhnikov, “Acoustic characterization of high intensity focused ultrasound
fields: A combined measurement and modeling approach.” J. Acoust. Soc.
from the axis, the difference is nevertheless significant. In Am. 124, 2406–2420 (2008).
the heterogeneous case, the method provides a very good 11
E. A. Zabolotskaya and R. V. Khokhlov, “Quasi–plane waves in the non-
result as long as the heterogeneity is small (in practice up to linear acoustics of confined beams,” Sov. Phys. Acoust. 15, 35–40 (1969).
12
a 10% sound speed contrast). The backscattered field of V. P. Kuznetsov, “Equations of nonlinear acoustics,” Sov. Phys. Acoust.
16, 467–470 (1970).
course is not captured, but more side lobes of the forward- 13
B. K. Novikov, O. V. Rudenko, and V. I. Timoshenko, Nonlinear Under-
scattered field can be satisfactorily simulated compared to water Acoustics. Acoustical Society of America (Translation Books, New
parabolic approximations. York, 1987), pp. 109–118.
14
In the strongly nonlinear case of shock wave focusing M. A. Averkiou and R. O. Cleveland. “Modeling of an electrohydraulic lith-
otripter with the KZK equation,” J. Acoust. Soc. Am. 106, 102–112 (1999).
through a smooth spherical lens, numerical convergence 15
A. Bouakaz, E. Closset, and D. Cathignol, “Harmonic ultrasonic field of
tests and physical analysis prove the ability of the method to medical phased arrays: Simulations and experiments,” IEEE Trans. Ultra-
simulate with an improved accuracy the shock wave propa- 16
son. Ferroelectr. Freq. Control 50(6), 730–735 (2003).
gation through heterogeneous media. The well-known phe- V. F. Humphrey, “Nonlinear propagation in ultrasonic fields: measure-
ments, modelling and harmonic imaging,” Ultrasonics 38, 267–272
nomena of caustics formation, wave front folding, pressure (2000).
local amplification, waveform transformation, and shock 17
X. Yang and R. O. Cleveland, “Time domain simulation of nonlinear
breaking are all recovered. acoustic beams generated by rectangular pistons with application to har-
monic imaging,” J. Acoust. Soc. Am. 117, 113–123 (2004).
Prospects for future extension of this work include nu- 18
R. Marchiano, F. Coulouvrat, and J.-L. Thomas, “Nonlinear focusing of
merical implementation of adapted perfectly matched layers acoustic shock waves at a caustic cusp,” J. Acoust. Soc. Am. 117, 566–
(PML) for the HOWARD method that are more efficient 577 (2005).
19
than the presently used absorbing layers. Development of a A. A. Piacsek, “Atmospheric turbulence conditions leading to focused and
folded sonic boom wave fronts,” J. Acoust. Soc. Am. 111, 520–529 (2002).
three-dimensional (3D) version will be a numerical chal- 20
S. Baskar, F. Coulouvrat, and R. Marchiano, “Nonlinear reflection of graz-
lenge, but the efficiency of the angular spectrum to treat 3D ing acoustical shock waves: unsteady transition from von Neumann to
problems is promising. Absorption can easily be imple- Mach to Snell-Descartes reflections,” J. Fluid Mech. 575, 27–55 (2007).
21
mented as the linear part of the equation is treated in the fre- M. Brio and J. K. Hunter, “Mach reflection for the two-dimensional Bur-
gers’ equation,” Phys. D 60, 194–207 (1992).
quency domain. For atmospheric applications, the method 22
R. Marchiano, F. Coulouvrat, and J.-L. Thomas, “Numerical investigation
has to be extended to include slow flow effects with Mach of the properties of nonlinear acoustical vortices through weakly heteroge-
numbers typically of order 0.1 or less. neous media,” Phys. Rev. E, 77:016605(1-11), (2008).

J. Acoust. Soc. Am., Vol. 130, No. 1, July 2011 Dagrau et al.: Acoustic shock wave propagation 31
23 42
Ph. Blanc-Benon, B. Lipkens, L. L. Dallois, M. F. Hamilton, and D. T. M. K. Kelly, R. Raspet, and H. E. Bass, “Scattering of sonic booms by ani-
Blackstock, “Propagation of finite amplitude sound through turbulence: sotropic turbules,” J. Acoust. Soc. Am. 107, 3059–3064 (2000).
43
Modeling with geometrical acoustics and the parabolic approximation,” T. Christopher and K. J. Parker, “New approaches to nonlinear diffractive
J. Acoust. Soc. Am. 111, 487–498 (2002). field propagation,” J. Acoust. Soc. Am. 90, 488–499 (1991).
24 44
Y. Jing and R. O. Cleveland, “Modeling the propagation of nonlinear J. Tavakkoli, D. Cathignol, R. Souchon, and O. A. Sapozhnikov,
three-dimensional acoustic beams in inhomogeneous media,” J. Acoust. “Modeling of pulsed finite-amplitude focused sound beams in time
Soc. Am. 122, 1352–1364 (2007). domain,” J. Acoust. Soc. Am. 104, 2061–2072 (1998).
25 45
M. V. Aver’yanov, V. A. Khokhlova, O. A. Sapozhnikov, Ph. Blanc- V. A. Khokhlova, R. Souchon, J. Tavakkoli, O. A. Sapozhnikov, and D.
Benon, and R. O. Cleveland, “Parabolic equation for nonlinear acoustic Cathignol, “Numerical modeling of finite-amplitude sound beams: Shock
wave propagation in homogeneous moving media,” Acoust. Phys. 52, formation in the near field of a cw plane piston source,” J. Acoust. Soc.
623–632 (2006). Am. 110, 95–108 (2001).
26 46
M. F. Hamilton and D. T. Blackstock, Nonlinear Acoustics (Academic R. J. Zemp, J. Tavakkoli, and R. S. Cobbold, “Modeling of nonlinear ultra-
Press, San Diego, 1988), pp. 309–341. sound propagation in tissue from array transducers,” J. Acoust. Soc. Am.
27
W. F. Ames, Numerical Methods for Partial Differential Equations (Aca- 113, 139–152 (2003).
47
demic Press, San Diego, 1977), pp. 307–310. J. W. Goodman, Introduction to Fourier Optics (McGraw-Hill, New-
28
N. S. Bakhvalov, Ya. M. Zhileikin, E. A. Zabolotskaya, and R. V. Khokh- York, 1968), pp. 45–50.
48
lov, “Nonlinear propagation of a sound beam in a nondissipative G. Taraldsen, “Derivation of a generalized Westervelt equation for nonlin-
medium,” Sov. Phys. Acoust. 22, 272–274 (1976). ear medical ultrasound,” J. Acoust. Soc. Am. 109, 1329–1333 (2001).
29 49
V. A. Khokhlova, A. E. Ponomarev, M. A. Averkiou, and L. A. Crum, F. Dagrau, “Simulation de la propagation du bang sonique: de la CFD à
“Nonlinear pulsed ultrasound beams radiated by a rectangular focused l’acoustique non linéaire (Numerical simulation of the sonic boom propa-
diagnostic transducers,” Acoust. Phys. 53, 481–489 (2006). gation: From CFD to nonlinear acoustics),” Ph.D. dissertation, Université
30
Y. S. Lee and M. F. Hamilton, “Time-domain modeling of pulsed finite- Pierre et Marie Curie, Paris, 2009.
50
amplitude sound beams,” J. Acoust. Soc. Am. 97, 906–917 (1995). G. Cohen, Higher-Order Numerical Methods for Transient Wave Equa-
31
S. I. Aanonsen, T. Barkve, J. Naze Tjøtta, and S. Tjøtta, “Distortion and tions (Springer Verlag, Berlin, 2002), pp. 307–330.
51
harmonic generation in the nearfield of a finite amplitude sound beam,” F. Collino, “Perfectly matched absorbing layers for the paraxial equa-
J. Acoust. Soc. Am. 75, 749–768 (1984). tions,” J. Comp. Phys. 131, 164–180 (1997).
32 52
R. J. Leveque. Numerical Methods for Conservation Laws (Birkhauser, W. D. Hayes, R. C. Haefeli, and H. E. Kulsrud, “Sonic boom propagation
Basel, 1992), pp. 114–157. in a stratified atmosphere with computer program,” Technical Report,
33
F. Coulouvrat, “A quasi exact shock fitting algorithm for general nonlinear NASA CR-1299 (1969).
53
progressive waves,” Wave Motion 49, 97–107 (2009). L. Landau, “On shock waves at large distances from the place of their ori-
34
T. D. Mast, “Fresnel approximations for acoustics fields of rectangularly gin,” J. Phys. USSR 9, 496–500 (1945).
54
symmetric sources.” J. Acoust. Soc. Am. 121, 3311–3322 (2007). J. M. Burgers, “Further statistical problems connected with the solution of
35
F. Coulouvrat, “Nonlinear parabolic wave equation in ray coordinates for a simple nonlinear partial differential equation,” Proc. Kon. Nederlandse
high frequency acoustical propagation in a inhomogeneous and high speed Akad. van Wet., Ser. B 57, 159–169 (1954).
55
moving fluid,” Wave Motion 45, 804–820 (2008). P. M. Morse, Vibration and Sound (McGraw-Hill, New York, 1948), pp.
36
J. F. Claerbout, Fundamentals of Geophysical Data Processing (McGraw- 347–350.
56
Hill, New York, 1976), pp. 194–207. F. Coulouvrat, “Focusing of weak acoustic shock waves at a caustic cusp,”
37
B. E. McDonald, “High-angle formulation for the nonlinear progressive- Wave Motion 32, 233–245 (2000).
57
wave equation model,” Wave Motion 31, 165–171 (2000). J.-P. Guiraud, “Acoustique géométrique, bruit balistique des avions super-
38
L. Ganjehi, “Ondes de choc acoustiques en milieu hétérogène, des ultra- soniques et focalisation (Geometrical acoustics, ballistic noise of super-
sons au bang sonique (Acousical shock waves in heterogeneous media: sonic aircraft and focusing),” J. Mécan. 4, 215–267 (1965).
58
From ultrasonic waves to sonic boom),” Ph.D. dissertation, Université A. D. Pierce, Acoustics: An Introduction to its Physical Principles and
Pierre et Marie Curie, Paris, 2008. Applications (Acoustical Society of America, New York, 1989), pp. 460–
39
Th. Auger and F. Coulouvrat, “Numerical simulation of sonic boom 469.
59
focusing.” AIAA J. 40, 1726–1734 (2002). P. L.Marston, “Geometrical and catastrophe optics methods in scattering,”
40
F. Coulouvrat, “Sonic boom in the shadow zone: A geometrical theory of High Frequency and Pulse Scattering. Physical Acoustics, edited by A. D.
diffraction,” J. Acoust. Soc. Am. 111, 499–508 (2002). Pierce and R. N. Thurston (Academic Press, San Diego, 1992), Vol. XXI,
41
P. Boulanger, R. Raspet, and H. E. Bass, “Sonic boom propagation pages 1–234
60
through realistic turbulence,” J. Acoust. Soc. Am. 98, 3412–3417 A. D. Pierce and D. J. Maglieri, “Effects of atmospheric irregularities on
(1995). sonic-boom propagation,” J. Acoust. Soc. Am. 51, 702–721 (1972).

32 J. Acoust. Soc. Am., Vol. 130, No. 1, July 2011 Dagrau et al.: Acoustic shock wave propagation

You might also like