You are on page 1of 6

Isakson et al.

: JASA Express Letters

[http://dx.doi.org/10.1121/1.4890195]

Published Online 5 August 2014

A three-dimensional, longitudinally-invariant finite
element model for acoustic propagation in shallow
water waveguides
Marcia J. Isakson, Benjamin Goldsberry, and Nicholas P. Chotiros
Applied Research Laboratories, The University of Texas at Austin, Austin,
Texas 78713-8029
misakson@arlut.utexas.edu, goldsbe@arlut.utexas.edu, chotiros@arlut.utexas.edu

Abstract: A three-dimensional, longitudinally-invariant finite element
(FE) model for shallow water acoustic propagation is constructed
through a cosine transform of a series of two-dimensional FE models at
different values of the out-of-plane wavenumber. An innovative wavenumber sampling method is developed that efficiently captures the
essential components of the integral as the out-of-plane wave number
approaches the water wavenumber. The method is validated by comparison with benchmark solutions of two shallow water waveguide environments: a flat range independent case and a benchmark wedge.
C 2014 Acoustical Society of America
V

PACS numbers: 43.30.Dr [AL]
Date Received: December 20, 2013

Date Accepted: June 24, 2014

1. Introduction
High fidelity models of acoustic propagation in shallow water waveguides are necessary for
sonar performance predictions and acoustic communication. However, the waveguides are
often very complex including interface roughness, sediment patchiness, sound speed profiles, and complex bathymetry. The finite element method (FEM) provides an excellent
tool to model these environments because it is fully customizable and provides a full wave
solution. One drawback of the method is its intense computational load. There are currently no fully three-dimensional finite element shallow water waveguide acoustic propagation models although two-dimensional models do exist.1 Because two-dimensional models
cannot be compared with experimental data, a longitudinally-invariant three-dimensional
model is proposed to bridge the gap. In this scheme, a cosine transform is performed on
the three-dimensional Helmholtz equation, producing a sum of two-dimensional components that can be solved using current methods. Although this produces a fully threedimensional model, one constraint is that the geometry must be invariant in one direction.
Many environments including continental slopes, ridges, and canyons can be approximated as longitudinally invariant. Within this constraint, the method can still include range
dependent effects such as sediment and sound speed variation.
2. The longitudinally-invariant finite element model
In this FEM, the variational form of the Helmholtz equation is solved over small subdomains or elements using polynomial basis sets. The solution is obtained by considering the boundary conditions between the elements that lead to a linear system of equations. The collection of interconnected elements, know as a mesh, represents the total
field. A rigorous mathematical description of the finite element method is given in
Ref. 2, while a description of its application to underwater acoustics is given in Ref. 3.
The FEM makes no approximations to the Helmholtz equation, and the system geometry can be expressed accurately to the size of the element. This makes the model
extremely versatile. For the calculations in this paper, a commercially available finite
element program, COMSOL, is used for meshing and solving.4
Full three-dimensional FEMs for ocean waveguides are still beyond the computation capabilities available. However, the physics of three-dimensional propagation

EL206 J. Acoust. Soc. Am. 136 (3), September 2014

C 2014 Acoustical Society of America
V

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 103.23.224.131 On: Wed, 03 Sep 2014 03:04:54

Both the variable and the constant discretization schemes used 300 points.23. the two-dimensional (2D) pressure fields are computed for a series of out-of-plane wavenumbers using 2D FE and summed to determine the 3D field. ky . In this study. Following Ref. To determine these parameters. yo. Two different sampling schemes are shown in Fig. The effect of the different discretization schemes is shown in Fig. 1(d) where the results of the two schemes are compared with the analytic solution. The structure of the integrand for an ocean waveguide is much more complex. x. k ¼ x/c(x. This evanescent part of the integral describes the curving wavefronts near the source and is only significant at short ranges. Two parameters in this derivation must be addressed: The limits of the sum in Eq. (xo. 1(c). y. zÞ ¼ kX ymax  Pðx. z.org/10. it is helpful to consider the canonical problem of a point source in an unbounded medium at 500 Hz. However. Download to IP: 103. x. y. If the sound speed and other environmental factors are only functions of range. The constant scheme did not resolve the integrand near ky ¼ k and results in an unstable solution. 136 (3). Soc. 1(a) is the integrand of Eq.doi. (2).1121/1. The source frequency spectrum is s(x). 03 Sep 2014 03:04:54 . (2) is complex. For ky > k the samples are discretized much more coarsely because the integrand is smooth. Note that the value of the integrand goes to zero smoothly past ky ¼ k where the effective wavenumber in Eq.: JASA Express Letters [http://dx. zÞ cos ðky yÞdky : (4) ky ¼0  x. 1(b). This suggests that the integrand should be sampled more finely near ky ¼ k. September 2014 Isakson et al. (4) and the discretization of the sum. Shown in Fig. ky . only single frequencies are considered. a cosine transform is performed on the three-dimensional Helmholtz equations along one spatial axis. y.4890195] Published Online 5 August 2014 can be captured if the geometry modeled is invariant in one direction. Using Eq. This results in a much finer sampling near ky ¼ k for the same number of samples. zo) is the point source location. (3) as a function of the normalized out-of-plane wavenumber and range from the source. r22D P (2) y 2 where ð1   Pðx. In this scheme. results in a near perfect solution that deviates less than the width of the analytic solution line. zÞ ¼ 0 The integral can be written discretely as  Pðx. ky . The variable discretization scheme. x. z) where x is the radial acoustic frequency and c is the sound speed that can vary with position. Acoust.131 On: Wed. the derivation begins with the three-dimensional Helmholtz equation. k is the acoustic wavenumber. Note how the variable discretization samples the integrand more finely near ky ¼ k where the function is more variable. zÞ cos ðky yÞdky : (3) Pðx. and depth. 5.224. it retains many of the same qualities as the free space solution. x. The sampling scheme is shown for the integral at range 10 m from the source in Fig. on the other hand. appropriate for a time harmonic acoustic propagation problem. This example illustrates how a variable sampling scheme can provide a stable solution with fewer samples. note that the integrand is smooth as ky ! 0. Also. field. see http://acousticalsociety.Isakson et al. the three-dimensional Helmholtz equation can be reduced to    þ k2  k2 P  ¼  1 sðxÞdðx  xo Þdðz  zo Þ.: Finite element propagation model EL207 Redistribution subject to ASA license or copyright. zÞ is the transformed Here ky is the out-of-plane wavenumber and Pðx. The resulting two-dimensional pressure components can be calculated using two-dimensional FEM. Am. The straight line has constant discretization in ky. The curved line uses a slightly offset gamma cumulative distribution function (CDF) coupled with a linear function just after ky ¼ k.org/content/terms. Last.  þ k2 P  ¼ sðxÞdðx  xo Þdðy  yo Þdðz  zo Þ: r2 P (1)  Here P is the pressure field. x. It approaches J.

4890195] Published Online 5 August 2014 Fig. The authors of Ref. 03 Sep 2014 03:04:54 .: JASA Express Letters [http://dx.Isakson et al. 7 rejected quadrature scheme in favor of the simple equidistant discretization.23. the discretization based on the cumulative CDF has been found to be the most robust and efficient for this problem. running the fully parallelized model on the Texas Advanced Computing Center LoneStar Cluster took 3 days with the variable discretization method compared with a projected 90 days using the constant method. 136 (3).org/10. this is impractical for this problem. 4. with respect to the evaluation of the inverse Hankel transform for axially symmetric problems. However. zero smoothly for values of ky > k and only has significant values for the evanescent wavenumbers at short ranges. In particular. see http://acousticalsociety. (Color online) (a) The integrand for the cosine transform of a point source in free space. Am.131 On: Wed. Through trial and error. as discussed previously.doi. it may seem that the integral could be evaluated with quadrature schemes.2. the integrand suffers from many of the same problems discussed by Jensen. September 2014 Isakson et al. Porter. (c) An example of how the integral is sampled for a range of 10 m.1121/1. For propagation in a range dependent waveguide described in the following text. (d) A comparison of the two different sampling methods and the analytic solution. Acoust. the oscillatory nature of the integral is highly dependent on range and environment requiring different quadrature points for each instance. Download to IP: 103. Kuperman.224. 7. EL208 J. (b) Two different methods of sampling the out-of-plane wavenumber.6 At first glance.5. Sec. 1. and Schmidt in Ref. It is much more complex near ky ¼ k and becomes smooth as ky ! 0. To put this in perspective. Therefore a similar scheme in which the discretization is determined by the gamma CDF can be used.org/content/terms. Soc. However.: Finite element propagation model Redistribution subject to ASA license or copyright. the variable discretization method resulted in a 30-fold decrease in the number of evaluations of the 2D model relative to the equivalent constant method.

5 2. see http://acousticalsociety. 136 (3). 4.Isakson et al. 2. Download to IP: 103.224. Waveguide depth (m) Source depth (m) Frequency (Hz) Receiver depth (m) Water sound speed (m/s) Water density (kg/m3) Water attenuation (dB/k) Sediment sound speed (m/s) Sediment density (kg/m3) Sediment attenuation (dB/k) Wedge angle ( ) Range independent Range dependent 200 150 100 150 1500 1024 0 1700 1500 0. J. Waveguide parameters for validation. 03 Sep 2014 03:04:54 .org/content/terms. (Color online) A comparison of the finite element approach and the wavenumber integration approach for transmission loss in a range independent shallow water waveguide.23. Fig.1121/1. The black line is the ocean bottom. September 2014 Isakson et al.doi. The gray line denotes the cut shown in Fig. Am. (Color online) Transmission loss calculated using finite elements for a wedge environment. Acoust.4890195] Published Online 5 August 2014 Table 1.86 Fig.: Finite element propagation model EL209 Redistribution subject to ASA license or copyright.5 – 200-0 100 25 100 1500 1000 0 1700 1500 0. 3. Soc.: JASA Express Letters [http://dx.131 On: Wed.org/10.

Many of the same physical features such as shadow zones and mode cut-offs are evident. 3. (Color online) The out-of-plane transmission loss for the wedge environment along the cut shown in Fig.doi. 03 Sep 2014 03:04:54 . The range dependent case was a wedge environment appropriate for propagation near shore. Because the FEM solves the Helmholtz equation exactly to the degree of the discretization and is fully customizable. with a lossy penetrable bottom. The model parameters are given in Table I. In both instances. An inverse cosine transform yields the final solution. 7 of Ref. It should be noted that unlike the calculations provided by Jensen and Ferla in Ref. 2.9 This case required 6100 evaluations of the 2D field components. the source was considerably closer to the wedge apex. Am.4890195] Published Online 5 August 2014 Fig.: JASA Express Letters [http://dx. Conclusion A 2D finite element propagation model has been extended to three dimensions through a cosine transform that is appropriate for scenarios in which the geometry along one spatial coordinate is invariant. This can be compared qualitatively with Fig. The range independent case consisted of a 200 m deep waveguide with a sandy bottom. 9. The models agree with excellent precision. 9.1121/1. the FE model includes all orders of coupling and scattering making it an excellent benchmark solution. 10. This can be directly compared with Fig. Soc.8 Results are shown in Fig. Acoust. Validation To validate the model. case III. 10 although in Ref. 4. Download to IP: 103. 3 and shown in Fig. It was determined that a constant sampling interval scheme to compute the transform was computationally prohibitive. The range dependent case is the ASA benchmark wedge. it can be applied to complex range dependent EL210 J. the finite element model agreed well with available solutions. A non-uniform spacing based on an offset gamma cumulative distribution function was found to be robust and efficient.: Finite element propagation model Redistribution subject to ASA license or copyright.org/10. The model requires that a 2D field be calculated for each out-of-plane wavenumber. appropriate for a continental shelf environment. 3.224. 3.org/content/terms.131 On: Wed. see http://acousticalsociety. Note the modal cut-offs as the field approaches the wedge apex. 4.Isakson et al.23. The solution was compared with a wavenumber integration solution provided by OASES (Ocean Acoustic and Seismic Exploration Synthesis). September 2014 Isakson et al. The pressure field along the source axis up the wedge is shown in Fig. 4. The model was validated by comparison with range independent and range dependent benchmark problems. To display the out-of-plane features of the model. 136 (3). a cut was taken along the gray line in Fig. This problem required 2500 evaluations of the 2D field components. range independent and dependent test cases were selected to compare with existing models. The range independent case was a shallow water waveguide of water over sand. 8 in Ref.

“Composite boundary-valued solution of the 2.org/content/terms. Acoust.23. Chap. Computational Ocean Acoustics. Jensen and C. Soc. 2012). J. Cambridge. Computational Ocean Acoustics.1121/1. “Finite element modeling of reverberation and transmission loss in shallow water waveguides with rough boundaries. Am. OASES Version 2. range dependent sediment parameters. 136 (3). Therefore it can be used for both propagation and reverberation modeling. Porter. Kuperman. 3 F.: Finite element propagation model EL211 Redistribution subject to ASA license or copyright. Acoust. Schmidt. Soc. P. Am. New York.” J. New York. Soc.” Geophysics 63(5). Jensen. 4. An Introduction to the Finite Element Model (McGraw-Hill. 9 F.5-d green’s function for arbitrary acoustic media. 03 Sep 2014 03:04:54 . Acknowledgments This work was supported by ONR. Chotiros. 10 G. Zhou and S. Jensen. 5 B. Massachusetts Institute of Technology.: JASA Express Letters [http://dx. 87(4). Ocean Acoustics under the direction of Robert Headrick. Kuperman. Greenhalgh.131 On: Wed. and water column sound speed profiles are important. Soc. (Springer. Am. September 2014 Isakson et al. B. “Numerical solutions of range-dependent benchmark problems in ocean acoustics. 2 J. Schmidt. A. 1997). A. B. Am. 2nd ed. it provides the entire pressure field.tacc. 4 Information available on Comsol Multi-Physics at http://www. 8 H. 93(3). 2006). and H. 1499–1510 (1990). 1273–1279 (2011). 1319–1328 (1993). 2nd ed. Buckingham. MA. Acoust.” J. Porter. 7 F. see http://acousticalsociety. Isakson and N. Unlike many other models. B. J. W.org/10. Chap. Schmidt.edu (Last viewed December 17. M. 2011). 6 Information on Texas Advanced Computing Center Lonestar Cluster available at https:// www.1 User Guide and Reference Manual (Department of Ocean Engineering. 7. “An analysis of the three-dimensional sound field in a penetrable wedge with a stratified fluid or elastic basement. A.” J.224. and H. W. 129(3).utexas. 1813–1823 (1998).com/ (Last viewed June 4. 2013). 2011). B.doi. J. both forward and backward propagation. Deane and M. B.4890195] Published Online 5 August 2014 environments such as continental shelf regions where interface roughness. New York. References and links 1 M. (Springer. B. Reddy.Isakson et al.comsol. N. M. Download to IP: 103. M. Acoust. Ferla.