Sie sind auf Seite 1von 8

Finite-element modeling of depth and range dependent acoustic

propagation in oceanic waveguides


Chiruvai P. Vendhan,
a
Ganesh C. Diwan, and Subrata K. Bhattacharyya
Department of Ocean Engineering, Indian Institute of Technology Madras, Chennai 600036, India
Received 8 December 2008; revised 3 March 2010; accepted 11 March 2010
Finite-element models FEMs of ocean acoustic waveguides are capable of predicting the full wave
solution including the effect of inhomogeneities and interfaces. However the method appears
computationally feasible at present only for low to intermediate frequencies. The FEM discussed in
this paper treats the radiation boundary condition involving multiple propagating modes using a
penalty function approach. The effect of a point source has been represented as a pressure boundary
condition on a small boundary surrounding the source. The FEM has been validated with a few
examples of ideal waveguides. The FE results for a range and depth dependent parallel waveguide
an ASA benchmark problem F. B. Jensen and C. M. Ferla, J. Acoust. Soc. Am. 87, 14991520
1990compare well with published results.
2010 Acoustical Society of America. DOI: 10.1121/1.3392440
PACS numbers: 43.20.Mv, 43.30.Bp KGF Pages: 33193326
I. INTRODUCTION
A variety of computational techniques have been devel-
oped over the years for the solution of ocean acoustic wave-
guide problems.
1
Buckingham
2
has presented a comprehen-
sive review of ocean acoustic propagation models including
some that account for geoacoustic effects. Among the vari-
ous methods available, the coupled-mode model
3
COUPLE
using the normal mode solution of the Helmholtz equation is
considered a benchmark method to evaluate approximate
waveguide formulations based on a variety of parabolic
equation PE models, as well as the accuracy of full wave
solutions obtained using other numerical methods such as the
nite-difference FD and nite-element methods FEMs.
The latter is a versatile tool for modeling waveguides with
complex inhomogeneities and interfaces. The objective of
this paper is to revisit the nite-element approach to obtain-
ing the full wave solution to range and depth dependent un-
derwater acoustic waveguide problems and explore the use
of a penalty function method for imposing a multimode form
of the radiation boundary condition BC.
When using domain discretization techniques such as
the FE and FD methods to obtain numerical models for wave
propagation in an unbounded domain, the problem domain
needs to be truncated and suitable nonreecting/absorbing
boundary condition must be imposed on the ctitious/
articial truncation surface. A variety of approximate radia-
tion boundary conditions, both local and global, have been
devised for this purpose, for harmonic waves, in particular.
4,5
These absorbing boundary conditions directly depend on the
medium wavenumber k of the harmonic wave problem. For
the purpose of modeling an ocean acoustic waveguide, which
is bounded at the top and bottom, but unbounded in the
radial/range direction, Fix and Marin
6
observed that the far-
eld radiation/outow condition should provide for the
propagation of many modes, each having a different wave-
number, unlike in the case of unbounded domain problems,
which use only a single wavenumber in the radiation bound-
ary condition. They have presented a FE model for the wave-
guide problem wherein the normal mode solution at the far-
eld boundary is matched with the FE solution there, using
the orthogonality property of the depth eigenfunctions. Fol-
lowing this approach, Bayliss et al.
7
presented a FE model
for the ocean waveguide problem and solved the nite-
element equations using a preconditioned conjugate-gradient
method.
Murphy and Chin-Bing
8,9
have developed a nite-
element model called FOAM for ocean acoustic propagation
ignoring seismoacoustic effects. Later they
10
presented a
seismoacoustic FE model called SAFE. When solving the
standard wedge problem, they introduced a wide-angle radia-
tion boundary condition to reduce spurious bottom reec-
tions. They have not addressed the issue of providing for the
radiation of multiple propagating modes at the far-eld
boundary of parallel waveguides. Li et al.
11
have considered
this aspect and proposed the use of numerically derived nor-
mal modes for a vertically heterogeneous waveguide, follow-
ing the approach of Fix and Marin
6
to impose the radiation
condition; they evaluated the accuracy of three models using
PE solutions. Buckingham
2
has reviewed the features of a FE
code for underwater acoustics developed by Pack.
12
In this
work, innite elements have been employed to model wave
transmission across bottom layers. Kampanis and Dougalis
13
have described a FE computation capability, again following
the Fix and Marin
6
approach, for axially symmetric
waveguides; they used standard linear iterative solvers. Sub-
sequently, Athanassoulis et al.
14
have presented extensive
comparisons of FE results based on their FENL code and re-
sults based on a coupled-mode model.
3
Isakson et al.
15
have
presented both time and frequency domain solutions to a few
canonical and two-way range dependent waveguide prob-
lems. They employed an existing all-purpose FE code.
Isakson
16
has developed an FE model using a commercial
a
Author to whom correspondence should be addressed. Electronic mail:
vendhan@iitm.ac.in
J. Acoust. Soc. Am. 127 6, June 2010 2010 Acoustical Society of America 3319 0001-4966/2010/1276/3319/8/$25.00
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 129.234.0.0 On: Wed, 25 Jun 2014 08:16:21
code for solving a two-dimensional shallow water waveguide
problem with rough interfaces.
Four essential aspects in FE modeling of waveguide
problems are, 1 modeling of the radiation boundary condi-
tion in the far eld; 2 modeling of the radiation boundary
condition at a penetrable bottom; 3 modeling of a point
source; and 4 efcient solution of the FE equations for
practical frequencies and ranges of interest. While the rst
aspect has been studied in detail following the work of Fix
and Marin,
6
the second aspect seems to have received rather
limited attention. Buckingham
2
discusses the use of innite
elements for this purpose. The fact that one needs to provide
for the passage of many vertical modes, each having a dif-
ferent wavenumber, poses modeling difculties, which needs
further study. Representation of a point source in the context
of FE modeling appears to be a crucial issue that has not
been specically highlighted in the literature. Regarding the
solution of the FE equations, while standard Gauss solvers
can readily be used, they require large computer storage
when considering long ranges and intermediate and higher
frequencies. In this context, Bayliss et al.
7
and Athanassoulis
et al.
14
discuss the use of iterative methods based on the
conjugate-gradient method. However, block Gauss solvers in
conjunction with standard superelement formulation adopted
in general nite-element modeling would partly obviate the
above problems.
8
The potential of this approach remains
largely unexplored.
It is the purpose of this investigation to formulate a FE
model for acoustic waveguides with a provision for multiple
propagating modes at the truncation boundary, following an
approach somewhat different from that used by Fix and
Marin.
6
A brief presentation of the Helmholtz equation in an
inhomogeneous parallel waveguide is made with specic ref-
erence to the outow boundary condition in the far eld. The
salient features of the standard FE idealization are presented
and attention is focused on imposing the radiation boundary
condition as well as modeling of the point source. A few
numerical examples are used to validate the computation and
an ASA benchmark problem has been solved for comparison.
The major highlight of the present work is the formulation
and implementation of the penalty function technique to im-
pose multimode radiation boundary condition in the FE mod-
eling of underwater acoustic waveguides.
II. GOVERNING EQUATIONS AND BOUNDARY
CONDITIONS
The uid domain =
I
+
O
Fig. 1 of the waveguide
problem consists of the inner domain
I
truncated by the
articial radiation boundary S
R
, and the outer domain/far-
eld
o
. The waveguide is assumed to be axially symmetric
about the vertical z- axis containing a source at depth z
s
,
with r denoting the radial coordinate or the range. It is
bounded at the top by the z=0 plane, which is the air-sea
interface S
F
, and at the bottom by a seabed of arbitrary
topography S
B
. The waveguide is assumed to have un-
bounded range. For time harmonic linear acoustic waves,
assuming p z, r, t =pr, ze
it
, being the circular fre-
quency of the source, the governing equation is given by
1

1

p + k
2
p =
f
0
rz z
s

2r
, 1
where is the gradient operator, the density of the acoustic
uid, k the acoustic wavenumber kr, z =/ cr, z, c the
local speed of sound, and f
0
is related to the point source at
r=0 and z=z
s
, which is represented in Eq. 1 by using the
Dirac delta functions r and zz
s
.
Considering the large impedance mismatch between air
and water, a pressure release boundary condition may be
used at the free surface. Thus,
p = 0 on S
F
. 2
As the waves encounter the seabed, there is partial reection
and the remaining energy is transmitted into the seabed. A
part of the transmitted waves may be coupled back into the
water column because of refraction through the sediment
layers. However in the present study a rigid bottom is as-
sumed, for which
p
n
= 0 on S
B
, 3
where S
B
denotes the sea bottom with normal n .
For the purpose of FE modeling, the waveguide, which
is unbounded in range, is truncated at r=r
b
, and the trunca-
tion boundary is treated as the radiation boundary S
R
, on
which a suitable approximate radiation condition should be
imposed. Here, the well-known boundary damper approach
due to Bayliss et al.
4
has been adopted. The rst order cylin-
drical damper equation may be written as
p
m
r
+
m
p
m
= 0, m = 1, 2, . . . , M on S
R
, 4
where M denotes the number of propagating modes, and the
damper coefcient
m
associated with the mth mode is given
by

m
=
1
2r
ik
rm
, 5
where k
rm
denotes a horizontal wavenumber. It may be noted
that Eq. 5 is exact for the asymptotic form of a cylindri-
cally symmetric wave. On the truncation boundary S
R
, the
FIG. 1. Denition sketch of waveguide and computational domain for the
FE model.
3320 J. Acoust. Soc. Am., Vol. 127, No. 6, June 2010 Vendhan et al.: Ocean acoustic waveguide
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 129.234.0.0 On: Wed, 25 Jun 2014 08:16:21
acoustic pressure may be expressed as a sum of normal pres-
sure modes as
pz =

m=1
M
p
m
z, 6
where p
m
z, m=1, 2, . . ., are the normal modes of propaga-
tion for the problem in Eq. 1. Following Fix and Marin,
6
the radiation boundary condition for the waveguide problem
may be written, using Eqs. 4 and 6, as
p
r
+

m=1
M

m
p
m
= 0 on S
R
. 7
Denoting a depth eigenfunction or normal mode by f
m
z,
which is associated with the mth propagating mode eigen-
value, the pressure modes in Eq. 6 may be written as
p
m
z = a
m
f
m
z, m = 1, 2, . . . , M, 8
where a
m
denotes a modal participation factor. Then the ra-
diation boundary condition in Eq. 7 may be rewritten as
p
r
+

m=1
M
a
m

m
f
m
z = 0 on S
R
, 9
where the constants a
m
are obtained by using the
1/ z-orthogonality of the normal modes. It has tacitly
been assumed here that the waveguide has constant water
depth and range independent but depth dependent sound
speed in the vicinity of the truncation boundary S
R
and be-
yond, so that the depth eigenproblem corresponding to the
problem in Eq. 1 could be solved at least numerically for
example, see Ref. 11. If the radiation boundary is at a loca-
tion where one or more evanescent modes contribute signi-
cantly, the required number of such modes can readily be
included by extending the summation in Eq. 9.
Note that while the radiation condition in Eq. 4 on an
individual mode is local, the radiation condition in Eq. 9 is
global, meaning that nodes of an element on the truncation
boundary are linked to other elements there in view of the
coefcients a
m
.
A. Constraints
In view of Eq. 8, Eq. 6 may be written as
Cpz, a
1
, a
2
, . . . a
M

T
= 0 on S
R
, 10
where
C = 1, f
1
z, f
2
z, . . . . f
M
z, 11
and the companion vector in Eq. 10 is unknown. Equation
10 will be treated as a constraint in the following FE
model.
III. VARIATIONAL FORMULATION
For the purpose of nite-element modeling, it would be
convenient to construct a variational formulation. In the
present study, in order to avoid possible numerical difcul-
ties with handling a point source, a small uid domain
S
surrounding the source has been excluded so that the com-
putational domain is

I
=
I

S
. Consider the following
axisymmetric functional Ip dened in the cylindrical coor-
dinate system r, z Fig. 1:
Ip =
1
2

I
1

p
r

2
+
p
z

rdrdz

1
2

I
k
2

p
2
rdrdz +
1
2

m=1
M

s
R
1

m
p
m
2
rdz

S
D
+S
N
1

p
n
pdS, 12
where S
D
denotes the surface on which a Dirichlet BC is
prescribed and S
N
the surface with prescribed Neumann BC,
and the other domains of integration are identied in Fig. 1.
It can readily be shown that the variational condition
I = 0 13
leads to the governing differential equation in Eq. 1 and the
boundary conditions in Eqs. 24. Thus Eqs. 12 and 13
can be used to develop a FE model using the well-known
RayleighRitz approximation. But the resulting solution
should also obey the constraints in Eq. 10, which will en-
sure the imposition of the radiation boundary condition as
discussed above. This will be achieved by modifying the
discrete approximation to the functional in Eq. 12.
IV. FINITE-ELEMENT MODEL
The nite uid domain

I
which excludes the source
of the axisymmetric waveguide in Fig. 1 may be discretized
using ring elements of quadrilateral cross section with
C
0
-continuity and the well-known isoparametric formula-
tion.
17
The computational domain is discretized into a mesh
of nite elements, which are connected at both the grid in-
tersection points and at one or more points between the grid
intersections. Such points constitute the nodes at which the
eld variable is treated as unknown. The nite-element ap-
proximation for the eld variable p may then be written as
pr, z

j=1
n
p
ej
N
j
, = N
T
p
e
, 14
where n denotes the number of element nodes 4 or 8 in the
present study, p
ej
the nodal pressure variable/degrees of
freedom dofs and N
j
, the polynomial shape function in
the parametric coordinates , in the r, z-plane for de-
tails see Ref. 17. Substituting Eq. 14 into Eq. 12 yields
the following discrete form:
Ip
e

1
2
p
e

T
K
e
M
e
p
e

+
1
2

m=1
M
p
em

T
R

em
p
em
p
e

T
f
e
, 15
where p
em
denotes nodal pressure on the radiation bound-
ary due to the mth mode and the various matrices above will
be identied subsequently. The stationary condition of the
potential Ip
e
above should be sought subject to the con-
J. Acoust. Soc. Am., Vol. 127, No. 6, June 2010 Vendhan et al.: Ocean acoustic waveguide 3321
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 129.234.0.0 On: Wed, 25 Jun 2014 08:16:21
straint in Eq. 10. There are two ways of implementing this,
one the classical Lagrangian multiplier approach and the
other the penalty function approach; the latter, which is com-
monly used in the context of nite-element analysis
17,18
is
adopted in the present work. To achieve this, a modied
potential I may be dened as
I = I +
1
2
p
e

T
C
e

P
C
e
p
e
, 16a
where C
e
denotes the constraint matrix in Eq. 11 specic
to an element. The penalty coefcient matrix
P
above
may be chosen to be diagonal for convenience, with
Pm
denoting the penalty parameter associated with the mth
mode. Equation 16a may be expanded as
Ip
e
=
1
2
p
e

T
K
e
M
e
+ R
e
+ C
e

P
C
e
p
e

p
e

T
f
e
, 16b
where the enlarged element dof vector is dened as
p
e
=

p
e

. 17
The enlarged stiffness, mass and damping matrices, and load
vector in Eq. 16b, consistent with p
e
in Eq. 17, are
given by
K
e
=

K
e
0
0 0

; M
e
=

M
e
0
0 0

,
R
e
=

0 0
0 R
e

, f
e
=

f
e

0

. 18
The matrices K
e
, M
e
, and R
e
in Eq. 18 are tradition-
ally called, respectively, the element stiffness, mass and ra-
diation damping matrices, and f
e
the load vector. They are
given as follows:
K
e
=

e
1

N
T
Nd, 19a
M
e
=

e
k
2

N
T
Nd, 19b
R
e
= diagR
e1
, R
e2
, . . . , R
eM
, 19c
R
em
= f
zm

T
R

em
f
zm
, R

em
=
m
S
Re
1

N
T
NdS,
19d
f
zm
= f
zm
z
1
, f
zm
z
2
, . . . , f
zm
z
n

T
, 19e
f
e
=

S
Ne
1

N
T
dS, 19f
where N denotes the shape function matrix see Eq. 14
and f
zm
z
j
denotes the jth nodal value of the mth mode on a
nite element in contact with the radiation boundary S
Re
. The
steps required to derive Eq. 19c are outlined in the Appen-
dix. The load vector in Eq. 19f originates from the Neu-
mann condition p

=p/ n at the element boundary S


Ne
. It is
of interest to note that the radiation damping matrix R
e
in
Eq. 19c implies uncoupled modal participation. However,
the constraint term involving the matrix C
e
in Eq. 16b
brings about modal coupling.
The various integrals above are dened over relevant
nite-element domains. The stationary condition of the po-
tential I in Eq. 16b is obtained by setting
I
p
e

= 0. 20
Equation 20 leads to general element equations of the form
K
e
M
e
+ R
e
+ C
e

P
C
e
p
e
= f
e
. 21
It may be noted that if the penalty matrix
P
=0 in Eq. 21,
the constraints are ignored; as the penalty parameter values
increase, the error in satisfying the constraint equations de-
creases, and for very high values of penalty, the numerical
solution may break down. Hence a judicious choice of the
penalty parameters is essential. For the nite elements not in
contact with the radiation boundary, the radiation damping
matrix R
e
and the constraint matrix C
e
in Eq. 21 are
trivial. Hence, for this case, the FE equation may be deduced
from Eq. 18 and Eq. 21 as
K
e
M
e
p
e
= f
e
. 22
The radiation damping matrix R
e
in Eq. 19c, which is
complex in view of Eq. 5, is dened only for elements that
share one or more of their boundaries with the articial
boundary S
R
. Also, in Eq. 12, p

=p/ n , and the integral in


Eq. 19f will contribute to the external load vector only
when an element boundary has a known Neumann boundary
condition; otherwise such terms are eliminated while carry-
ing out the nite-element assemblage operation, which yields
the following global nite-element equations:
K M + R + Cp = f, 23
where the global solution vector p
consists of all the pres-
sure dof in the computational domain as well as the unknown
vector a in Eq. 8.
The global nite-element matrices in Eq. 23 may for-
mally be written as
K =

e
K
e
, M =

e
M
e
, R =

e
R
e
,
C =

e
C
e

P
C
e
, f =

e
f
e
, 24
where
e
denotes the standard nite-element assemblage
operation.
17
A. Modeling of a point source
When the inhomogeneous Helmholtz equation in Eq. 1
is employed in the FE model, the source term involving the
delta function, as the other terms of the differential equation,
is satised only approximately over the nite elements in
contact with the point source. Of course, the error is expected
to decrease with mesh renement. The present FE formula-
3322 J. Acoust. Soc. Am., Vol. 127, No. 6, June 2010 Vendhan et al.: Ocean acoustic waveguide
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 129.234.0.0 On: Wed, 25 Jun 2014 08:16:21
tion uses the complex pressure p as the eld variable. Hence
a kinematic/Dirichlet boundary condition in terms of p
would be satised exactly at the nite-element nodes. In
light of this, it would be interesting to see whether the effect
of the source could be modeled as a kinematic boundary
condition. To facilitate this, the computational domain em-
ployed above see Eq. 12 excludes the source. This is
achieved conveniently by matching a nite-element node
with the source and excluding all the nite elements that are
in contact with the source node see Fig. 1. Then the free
eld pressure due to the source on the periphery of the ex-
cluded domain is imposed as a kinematic BC in the nite-
element model. Premkumar et al.
19
employed this strategy to
solve both unbounded and waveguide-type acoustic prob-
lems. It may be argued that the pressure distribution on the
excluded domain boundary is not the actual one, which
would be known only after solving the FE equations. How-
ever, the following argument justies the approach. It is
known that for small volume sources, the pressure in the far
eld is not affected by the individual shape of a source, as
long as the source strengths are equal. Thus, this justies the
use of a computational domain that excludes a small FE
domain around a point source. In the present study, the size
of the excluded domain has been kept at about a tenth of the
wavelength. Comparison of the FE results with an analytical
solution indicates that such a choice is satisfactory.
V. SOLUTION OF FE EQUATIONS
The global FE equation in Eq. 23 may be written for
brevity as
Ap = f. 25
It may be noted that for an acoustic medium with real sound
speed, the coefcient matrix A above is complex because
of the radiation damping matrix R
e
see Eqs. 19c and
19d. For a lossy medium modeled with complex sound
speed, the second term in Eq. 23 also becomes complex.
Although A is non-self-adjoint, it is a complex symmetric
matrix and hence the Gauss solver employed here to obtain
the solution to Eq. 25 exploits the attendant computational
advantage. Since such solvers for FE equations are coded as
block solvers with compact storage scheme, large nite-
element models can be handled even with modest computer
storage. Of course such a solution strategy involves overhead
in the form of read/write operations on secondary storage
devices. At this juncture, it may be recalled that Bayliss et
al.
7
and Athanassoulis et al.
14
have used iterative methods
based on the conjugate-gradient technique.
20
Since the present FE model adopts a penalty function
approach to impose the radiation boundary condition with
multiple radiating modes, the choice of suitable penalty pa-
rameter
Pm
is important. This can be resolved only through
numerical experiments. The penalty parameter was obtained
by prescribing a scale factor on the average value of the
diagonals of the coefcient matrix A in Eq. 25; i.e.,

Pm
=
s

i=1
n
A
ii
/n, 26
where n denotes the total number of FE equations/dof and

s
a user-specied penalty scale factor. Computations indi-
cate that the results are stable over a wide range of
s
values.
21
The results reported here have been obtained using

s
=10
2
.
VI. NUMERICAL EXAMPLES
A. Isovelocity examples
The nite-element method for the solution of inhomoge-
neous ocean acoustic waveguide problems is validated rst
with analytical results for isovelocity waveguides. A cylin-
drically symmetric parallel waveguide of depth 100 m with a
point source is shown in Fig. 2. The nite-element model
consists of a uniform grid of isoparametric quadrilateral ele-
ments, with the element length being about / 10, where
denotes the wavelength. As discussed previously, a domain
of two elements has been excluded to remove the source
from the truncated domain Fig. 1. The FE model is rst
solved assuming pressure release boundary conditions both
at the top and bottom of the waveguide, and a point source of
frequency 20 Hz located at a depth of 36 m see Tables I and
II for the data. The FE mesh consists of 400 elements in
range and 50 elements in depth. Computed acoustic pressure
along the range at the depth of the source is compared in Fig.
3 with the normal mode solution
1
with 50 modes, of which
only the rst two are propagating. The normal mode solution
is also compared with the FE solution obtained with the
wavenumber k as the propagating wavenumber in Fig. 4.
FIG. 2. Idealized ocean waveguide.
TABLE I. Waveguide parameters.
Parameter
Examples in
Figure 3 Figure 5 Figure 7
Sound speed c m/s 1500 1500 c
1
=1500, c
2
=2000
Density kg/ m
3
1000 1000
1
=1000,
2
=2000
Depth of source z
s
m 36 25 25
Receiver depth m 36 25 25
Wavenumber k m
1
0.08377 0.4189 0.1466
Source frequency Hz 20 100 35
Thickness of water layer h m 100 100 100
J. Acoust. Soc. Am., Vol. 127, No. 6, June 2010 Vendhan et al.: Ocean acoustic waveguide 3323
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 129.234.0.0 On: Wed, 25 Jun 2014 08:16:21
Since Fig. 4 refers to the FE model that does not use the
radial wavenumbers k
rm
in the radiation boundary condition,
the results are expected to be in error. The comparisons in
Figs. 3 and 4 emphasize the need to use the multimode ra-
diation boundary condition. The FE solution has also been
obtained assuming a pressure release boundary at the top and
a rigid bottom with a 100 Hz source located at a depth of 25
m see Tables I and II for the data; the results are compared
with the analytical solution in Fig. 5. The FE mesh used in
this case has 1500 elements in range and 60 elements in
depth. For computational purposes, the range of the uid
domain is truncated at 3 km for both the cases. The source is
normalized to yield a unit amplitude of absolute pressure at
r=1. The comparisons are deemed very good.
B. Two layer waveguide
Figure 6 shows a two layer waveguide with pressure
release surface and rigid bottom. A harmonic point source
S of unit strength and frequency f =35 Hz is located at z
s
=25 m see Tables I and II for the data. The radiation con-
dition is imposed at a range of 3 km. The medium in the
upper layer is sea water with sound speed c
1
=1500 m/ s and
density
1
=1000 kg/ m
3
and has thickness h=100 m. The
bottom layer has sound speed c
2
=2000 m/ s, density
2
=2000 kg/ m
3
, and thickness H=50 m. The FE grid consists
of 750 isoparametric quadrilateral elements in range and 30
elements in depth. The FE results obtained are shown to
agree with the normal mode solution
22
in Fig. 7.
C. Range and depth dependent waveguide
To assess the validity of the FE model for an inhomoge-
neous waveguide, the ASA benchmark problem with range
and depth dependent sound speed is utilized.
23
The medium
is bounded above by a pressure release surface and below by
a rigid sea oor. Aharmonic point source of unit strength and
frequency f =25 Hz is placed at z
s
=250 m along with a re-
ceiver, in a water depth of D=500 m. The sea water has a
mean density of 1000 kg/ m
3
and the sound speed is given
by see Fig. 8
cr, z
c
0
=

1 +
l
1
D

2
e
2r/D
+
2l
2
D

2
e
4r/D

2l
1
D

1
2l
2
D
e
2r/D

cos
z
D
e
r/D

4l
2
D
cos
2z
D
e
2r/D

1/2
, 27
where c
0
=1500 m/ s is the reference sound speed, l
1
/ D
=0.032 and l
2
/ D=0.016. The computational domain is trun-
cated at a range of 4 km. The sound speed in Eq. 27 is
virtually constant well before the articial radiation bound-
ary S
R
r
b
=4 km. A 500 m deep isospeed channel with c
=1500 m/ s has 17 propagating modes. The FE grid consists
of 667 elements in range and 100 elements in depth. The
range and depth dependent transmission loss TL in dB is
dened as
TLr, r
s
= 20 log

pr, r
s

p
0
r
s

, 28
where pr, r
s
is the acoustic pressure at range r due to a
simple point source located at range r
s
, and p
0
r
s
is the
pressure produced at a distance of 1 m from the source in an
TABLE II. Propagation wavenumbers for waveguide examples.
Mode, m
k
rm
m
1
for the examples in
Figure 3 Figure 5 Figure 7
1 0.07766 0.4186 0.1442
2 0.05541 0.4162 0.1364
5 0.3943 0.0182
11 0.2582
12 0.2120
13 0.1458
0.5 1 1.5 2 2.5 3
0
0.01
0.02
0.03
0.04
Range (km)
M
o
d
u
l
u
s
o
f
p
r
e
s
s
u
r
e
,
|
p
|
(
N
/
m
2
)
Normal mode solution (50 terms in series)
FEM with 2 propagating modes
FIG. 3. Color online Pressure along range for an isovelocity waveguide
with pressure release surface and bottom see Tables I and II.
0 0.5 1 1.5 2 2.5 3
0
0.01
0.02
0.03
0.04
Range (km)
M
o
d
u
l
u
s
o
f
p
r
e
s
s
u
r
e
,
|
p
|
(
N
/
m
2
)
Normal mode solution (50 terms in series)
FEM with k as propagating wavenumber
FIG. 4. Pressure along range for the isovelocity waveguide in Fig. 3, but
with wavenumber k used in the radiation condition.
FIG. 5. Color online Pressure along range for an isovelocity waveguide
with pressure release surface and rigid bottom see Tables I and II.
3324 J. Acoust. Soc. Am., Vol. 127, No. 6, June 2010 Vendhan et al.: Ocean acoustic waveguide
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 129.234.0.0 On: Wed, 25 Jun 2014 08:16:21
unbounded, homogenous medium with density r
s
.
Figure 9 gives a comparison of the TL along the range at
the receiver depth obtained using the present FE model and
the one-way COUPLE results.
23
The comparison is deemed
good.
VII. CONCLUSION
A linear FE model for the solution of range and depth
dependent ocean acoustic waveguides has been discussed.
While other related works impose the multimode radiation
boundary condition following Fix and Marin
6
and obtain the
FE approximation using the Galerkin technique, the present
work adopts a variational formulation in which the radiation
boundary condition is treated as a constraint, which has been
imposed using the penalty function approach. The choice of
penalty factor could easily be established through a few nu-
merical experiments. The modeling of a point source adopted
here provides a simple approach which has been justied by
appealing to the equivalence of small volume sources having
different shapes but the same source strength, when comput-
ing the far-eld acoustics. The examples conrm the accu-
racy of the FE model. The computer code developed has the
potential to solve waveguide problems with complex inho-
mogeneities and interfaces. The method could become com-
petitive for intermediate to high frequencies by using the
superelement concept in the solution of long range problems,
as indicated by Murphy and Chin-Bing.
8
ACKNOWLEDGMENT
The authors are grateful to the reviewers for their valu-
able comments.
APPENDIX: DERIVATION OF MULTIMODE RADIATION
DAMPING MATRIX
Consider the functional in Eq. 12. The contribution,
I
R
p
e
, from the radiation boundary of a nite element is
represented by the third integral in that equation; i.e.,
I
R
p
e
=
1
2

m=1
M

s
Re
1

m
p
m
2
rdz, A1
where M denotes the number of propagating modes,
m
the
damper coefcient associated with the mth mode see Eq.
5, p
m
z the pressure associated with the mth normal mode
and S
Re
the element surface on the articial radiation bound-
ary see Fig. 1.
The modal pressure on the radiation boundary is given
by Eq. 8:
p
m
z = a
m
f
m
z, m = 1, 2, . . . , M, A2
where f
m
z denotes a normal mode function and a
m
the
modal participation factor. Using the nite-element represen-
tation, the modal pressure on the radiation boundary may be
written as
FIG. 6. Two layer waveguide with pressure release top and rigid bottom
see Table I.
FIG. 7. Color online Pressure along range for a two layer waveguide see
Table II.
FIG. 8. Color online Sound speed contours for a range and depth depen-
dent waveguide source: Ref. 23.
FIG. 9. Color online Transmission loss along range at receiver depth of
250 m for a range and depth dependent problem see Fig. 8.
J. Acoust. Soc. Am., Vol. 127, No. 6, June 2010 Vendhan et al.: Ocean acoustic waveguide 3325
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 129.234.0.0 On: Wed, 25 Jun 2014 08:16:21
p
m
z

e
N
T
p
em
, A3
where N denotes the shape functions and p
me
the nodal
pressure vector on an element edge on the radiation bound-
ary due to the mth mode. The summation symbol is used to
indicate that Eq. A3 is a piecewise polynomial representa-
tion over the entire depth of the waveguide. Using Eqs. A2
and A3, Eq. A1 may be written in a discrete form for a
nite element as also see Eq. 15
I
R
p
e
=
1
2

m=1
M
p
em

T
R

em
p
em
, A4
where
R

em
=
m
S
Re
1

N
T
NdS. A5
In view of Eq. A2, the vector of modal pressure at the
nodes of an element in Eq. A3 may be written as
p
me
= a
m
f
zm
z
1
, f
zm
z
2
, . . . f
zm
z
n

T
= a
m
f
zm
, A6
where f
zm
z
j
denotes the jth nodal value of the mth eigen-
mode on a nite element in contact with the radiation bound-
ary.
Now, using Eq. A5 and A6, the functional in Eq.
A4 may be written as
I
R
=
1
2

m=1
M
a
m
2
R
em
, A7
where
R
em
= f
zm

T
R

em
f
zm
. A8
The foregoing steps form the basis for Eq. 19c.
1
F. B. Jensen, W. A. Kuperman, M. B. Porter, and H. Schmidt, Computa-
tional Ocean Acoustics American Institute of Physics, Melville, New
York, 2000.
2
M. J. Buckingham, Ocean-acoustic propagation models, J. Acoust. 11,
223287 1992.
3
R. B. Evans, A coupled mode solution for acoustic propagation in a
waveguide with stepwise depth variations of a penetrable bottom, J.
Acoust. Soc. Am. 74, 188195 1983.
4
A. Bayliss, M. Gunzburger, and E. Turkel, Boundary conditions for the
numerical solution of elliptic equations in exterior regions, SIAM J.
Appl. Math. 42, 430451 1982.
5
D. Givoli, Nonreecting boundary conditions, J. Comput. Phys. 94,
129 1991.
6
G. J. Fix and S. P. Marin, Variational methods for underwater acoustic
problems, J. Comput. Phys. 28, 253270 1978.
7
A. Bayliss, C. I. Goldstein, and E. Turkel, The numerical solution of the
Helmholtz equation for wave propagation problems in underwater acous-
tics, Comput. Math. Appl. 11, 655665 1985.
8
J. E. Murphy and S. A. Chin-Bing, A nite-element model for ocean
acoustic propagation, Math. Comput. Modell. 11, 7074 1988.
9
J. E. Murphy and S. A. Chin-Bing, A nite element model for ocean
acoustic propagation and scattering, J. Acoust. Soc. Am. 86, 14781483
1989.
10
J. E. Murphy and S. A. Chin-Bing, A seismo-acoustic nite element
model for underwater acoustic propagation, in Shear Waves in Marine
Sediments, edited by J. M. Hovem, M. D. Richardson, and R. D. Stoll
Kluwer Academic, The Netherlands, 1991, pp. 463470.
11
G. Li, J. E. Murphy, and S. A. Chin-Bing, Radiation boundary conditions
for vertically heterogeneous acoustic media, J. Comput. Acoust. 1, 321
333 1993.
12
P. M. W. Pack, The nite element method in underwater acoustics, Ph.D.
thesis, Institute of Sound and Vibration Research, University of Southamp-
ton, UK 1986.
13
N. A. Kampanis and V. A. Dougalis, A nite element code for the nu-
merical solution of the Helmholtz equation in axially symmetric
waveguides with interfaces, J. Comput. Acoust. 7, 83110 1999.
14
G. A. Athanassoulis, K. A. Belibassakis, D. A. Mitsoudis, N. A. Kampa-
nis, and V. A. Dougalis, Coupled mode and nite element approximations
of underwater sound propagation problems in general stratied environ-
ments, J. Comput. Acoust. 16, 83116 2008.
15
M. J. Isakson, R. A. Yarbrough, and P. S. Wilson, Finite-element model-
ing of long range, range-dependent acoustic propagation in shallow wa-
ter, J. Acoust. Soc. Am. 122, 3074 2007.
16
M. Isakson, A nite element model for acoustic propagation in shallow
water waveguides A, J. Acoust. Soc. Am. 125, 2501 2009.
17
R. D. Cook, D. S. Malkus, M. E. Plesha, and R. J. Witt, Concepts and
Applications of Finite Element Analysis, 4th ed. Wiley, New York, 2002.
18
O. C. Zienkiewicz and R. L. Taylor, The Finite Element Method: Basic
Formulation and Linear Problems, Vol. 1 McGraw-Hill, New York,
1989.
19
R. J. Premkumar, T. Venkatesh, C. P. Vendhan, and S. K. Bhattacharyya,
Harmonic sound wave propagation in oceanic waveguides by nite ele-
ment method, in Proceedings of the ICONS 2002, International Confer-
ence on Sonar-Sensors and Systems Naval Physical Oceanographic Labo-
ratory, Kochi, India, 2002.
20
Solvers based on the conjugate-gradient method have been found much
more efcient than Gauss solvers when the size of the matrix equation is
very large, say, several tens of thousands of equations, and hence they hold
promise for high frequency FE models.
21
G. C. Diwan, Low frequency acoustic propagation in underwater
waveguides with nite element method, MS thesis, Ocean Engineering
Department, IIT Madras, India 2009.
22
G. V. Frisk, Ocean and Seabed Acoustics: A Theory of Wave Propagation
Prentice-Hall, Englewood Cliffs, NJ, 1994.
23
F. B. Jensen and C. M. Ferla, Numerical solutions of range-dependent
benchmark problems in ocean acoustics, J. Acoust. Soc. Am. 87, 1499
1520 1990.
3326 J. Acoust. Soc. Am., Vol. 127, No. 6, June 2010 Vendhan et al.: Ocean acoustic waveguide
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 129.234.0.0 On: Wed, 25 Jun 2014 08:16:21

Das könnte Ihnen auch gefallen