Sound propagation over inhomogeneous ground including a sound velocity prole

Andrew T Peplow and Svante Finnveden

Marcus Wallenberg Laboratory for Sound and Vibration Research, Department of Aeronautical and Vehicle Engineering, Royal Institute of Technology, Stockholm, 100 44, Sweden Abstract. The atmospheric prole whose sound speed varies linearly with height is simple in concept, but leads to complications when solving for the sound pressure. Its effects are commonly approximated by a similar prole whose squared refractive index is a linear function of height. In this paper, the validity of the approximation has been examined for sound propagation above an impedance ground and a computational approximation is given. The method relies upon a family of radiation boundary conditions for the wave equation derived by truncating a summation function approximation of a corresponding plane wave reection coefcient representation. It is demonstrated how these boundary conditions can be formulated in terms of a nite element approach. Numerical examples illustrate results with four coefcients included at the upper ctitious boundary conditions for the case of sound propagation above a grassstrip and sound propagation over the grassstrip with an atmospheric prole. Keywords: Nonreecting boundary conditions, Sound propagation, Finite elements PACS: 43.20 Mv

INTRODUCTION
The atmospheric prole whose sound speed varies linearly with height is simple in concept, but leads to complications when solving for the sound pressure. Its effects are commonly approximated by a similar prole whose squared refractive index is a linear function of height. In this paper, the validity of the approximation has been examined for sound propagation above an impedance ground and a numerical approximation is given. The prediction of sound propagation above an impedance ground is an important topic in the area of airborne sound propagation. The atmospheric prole whose squared refraction index varies linearly with height is a commonly used replacement for an atmosphere whose sound speed is a linear function of height. The solutions using residue theory were provided Rasmussen, Pierce and later Berry [1], [2], [3] and computed through a parabolic equation method by White et. al [4] These solutions have been used to study and predict sound propagation within an upward refracting atmosphere. These works have also been extended to the downward refracting atmosphere. All of the solutions rest on an assumption that a linear variation of the sound-speed prole with height can be approximated by a squared refraction index prole that is also linear in height. A variational formulation for the twodimensional Helmholtz equation in an acoustic waveguide is presented. The objective is to calculate sound propagation over absorbing ground within atmospheric perturbations. The case considered is that of wave propagation from a monofrequency point source in a halfspace. Characterisation of an ab-

sorbing boundary is modelled by a locally reacting impedance condition and an upper radiation condition is characterised by a high order local boundary condition. Parameters governing the uid wavespeed are allowed to vary linearly in the vertical direction. The proposed approach can provide accurate solutions over domains of arbitrary length in a certain direction using a small number of degrees of freedom. The problems studied here are specic to halfspaces whose uid domains only vary in height. For general domains there has been a signicant effort in improving the numerical modelling capability for linear steadystate acoustic problems in two and three dimensions for interior or exterior acoustic domains. Numerical methods for solving reduced wave differential equations can be categorized according to their local or global nature. The most general local approaches are the nite element method (FEM) and the boundary element method (BEM). These discrete methods rely on spatial discretizations that are small compared to the wavelength of the problem. For the new spectral element method presented in detail in [5] the basis functions inherently possess some of the wave nature of the acoustic eld. Consequently, for acoustic problems posed for a simple geometry the spectral methods require considerably less degrees of freedom than the discrete methods. For linear, constant coefcient, time-dependent, wave propagation problems, there are non-reecting boundary conditions available, which work well for some specic problems. Examples of such problems are e.g. Maxwells equations, the linearized Euler equations, the elastic wave equation, and the acoustic wave equation for which the perfectly matched layer (PML) method is used today, with satisfactory results, where articial boundaries are closed and nite. However in this work the wave equation is dened over a domain with innite boundaries where an exact articial nonlocal boundary condition may be employed. Exact boundary conditions for the wave equation was considered in the classic paper by Engquist and Majda [6]; giving rise to the Dirichlet-to-Neumann (DtN) method. The DtN map is an operator relating Dirichlet datum to Neumann datum on the boundary, enforcing desired asymptotic behaviour of the solution at innity. In a computational setting, used here, the DtN map may be reduced to a set of higher order local boundary conditions. Optimal local higher order boundary conditions have been discussed comprehensively by Givoli in [7] and [8], but the ideas used here differ in that a reection coefcient is minimised for almost all angles of plane wave incidence. The spectral nite element method (SFEM) applied to waveguide problems, referenced in [9] [10], and [11] can be viewed as a merger of the dynamic stiffness method and the nite element method. Specically the method is based on a variational formulation for non-conservative motion in the frequency domain. The SFEM has been used to study vibration in beam frameworks [9], beamstiffened railway cars [10] and for uid-lled pipes [11]. Finnveden used the spectral pipe elements for various cases from assessment of approximate theory to experimental SEA calculations, [12]. Use of a variational formulation for the spectral method provides a natural basis for approximations and a simple tool for combination with standard nite elements. In the following spectral nite elements are combined to solve a waveguide problem, and in particular a possible optimal local boundary condition is presented that approximates the true non-local boundary condition at an interface for a half-space.

SPECTRAL FINITE ELEMENT MODELLING

Figure 1 shows a typical spectral element geometry where j , j = 1, . . ., 4 represent typical crosssections. In the usual nite element discretization the crosssection is divided so that material parameters are constant within each interval, here this is extended so a linear prole is possible. In the superspectral method this section will dene one waveguide nite element which sets the technique apart from standard nite element practise. On a sublocal level the midintervals are not subdivided as in the h-version nite element method, but high-order two-noded polynomials, i.e. the p-version nite element method, are used. Within one superspectral nite element, j for example, the approximate solution p(.) may be represented by an expression of the form p(x, z) = [N1 (z), . . ., NM (z)]{W (x)} = [N(z)] {W (x)} (1)

using M piecewise polynomial shape functions, compactly supported in the vertical direction. Shape functions, [ N j (z) ], are dened sublocally over the crosssection and elemental shape functions {W (x)} are described later and in more detail in [5]. That is, the problem may be reformulated conveniently as a functional : L (p, pa ) =

1 2 a 2 k pp p pa d + 2 2

2 ik a p p d. (2) 2

The sound-speed prole of the atmosphere we seek to model is c(z) = c0 (1 + az), 0 z H, with c0 = 343 ms2 representing the sound speed at the surface and a is a real parameter with unit m1 . We take a = 0.01 here. The index of refraction for this prole is n(z) = c0 /c(z) = (1 + az)1 . For small values of (az), the squared refraction index 2 prole was approximated by a linear function of height k2 (z) = 2 /c2 (z) = k0 n2 (z) k0 (1 2az), where k0 = /c0 . A similar approximation was used for the density prole (z) noting that characteristic acoustic impedance took same value at z = 0 and z = H. It is worthwhile to mention here that the exact proles could have been incorporated in the nite element model with integration performed by numerical quadrature. However, the linear approximations for material parameters, also used in [4], were particularly attractive in the existing exact nite element formulation for waveguides [5]. The superscripted symbol pa denotes the complex conjugate of the sound pressure in the adjoint system, similar to the original system with negative damping. As shown in reference [9] and discussed by Finnveden in [11], the losses 1 , 2 may be accounted by employing a variational principle similar to Hamiltons principle. With this approach the mechanisms of variational methods are retained for nonconservative systems. Henceforth the notation for losses are not shown for ease of exposition. Consider the functional L j , (2), for a single arbitrary region j = d c. Substitution of expression (1) into the resulting form, where subscript .x denotes xderivative, yields the approximation L j =
d

{W a }T [ K1 ]{W } dx {W } [ K3 ]{W } dx
a T

d d

{W a }T [ K2 ]{W } dx {Wxa }T [ K4 ]{Wx } dx,

+
d

r Source z0 a
a

Receiver z b x
c b

FIGURE 1. Illustration of geometry and four superspectral elements characterised by the admittance boundaries a , a , b , and c

where, noting that the entries of matrix K3 (of order n n) are explicitly stated assuming admittance L , local submatrices have dened in detail in [5] but are briey presented here [ K1 ] = and (q, r)th element [ K3 ] [ K4 ]
q,r

1 2

1 k2 [ N N ] dz, [ K2 ] = 2 c

1 dN dN dz dz

= =

ik(H)L 1 2

1 1 + ik0 a,b,c q1 r1 , 2 qn rn (H) (0) 2

1 [ NN ] dz. c

The EulerLagrange equation which corresponds to (3) may be found by taking an appropriate rst variation: [K4 ] d2 {W } + [K1 ] {W } [K2 ] {W } + [K3 ] {W } = 0 dx2 (6)

and yields the spectral wave functions W (x), see [5].

Analysis of the reection coefcient

The application of the spectral nite element method on unbounded exterior domains involves a domain decomposition of the exterior region. On an entirely ctitious boundary or upward radiating boundary, , Figure1 it is possible to prescribe

Upper Fictitious Boundary

dz,

(3)

(4) (5)

10

10

Absolute value of reflection coefficient, |R|

10

Absolute value of reflection coefficient, |R|

Zero coefficients in expansion One coefficient Two coefficients Three coefficients Four coefficients

Zero coefficients in expansion One coefficient Two coefficients Three coefficients Four coefficients
10
1

10

10

10

10

10

10

20

30

Angle of incidence,

40

50

60

70

80

90

10

10

20

30

Angle of incidence,

40

50

60

70

80

90

C (right) minimising average reected sound power from = 0o , . . . , 72o

FIGURE 2. Reection coefcient from Model B (left), minimising about = 60o and model

articial boundary conditions (ABC) that approximate the exact nonlocal boundary condition [7] and [8] Consider the total acoustic eld due to an incident plane wave, pI = A exp(i k x){ exp(i t)} p(x, z) = pI (x, z) + RM ( ) pR (x, z). For the wellknown transmitting boundary condition

p cos 1 ik = p R0 ( ) = , n cos + 1

(7)

taking the specic surface impedance ( = 1) gives perfect transmission for the normally incident plane wave. Now there are many varieties of extensions to the reection coefcient, here though we choose the following : RM ( ) = cos 1 + M am sin2m m=1 cos + 1 + M am sin2m m=1 (8)

Minimizing under a suitable norm it is possible to nd rational values for the unknown coefcients. Model A. Take a series expansion around = 0. For example take one term, M = 1, and immediately a minimum for RM yields a1 = 1 . 2 Model B. Minimise RM ( ) around various incident angles other than = 0 above. Model C. Minimise "average reected sound power", i.e. integrate R2 cos2 over M a certain angle range.

Explicit calculations for the coefcients in Model A are possible from Taylor series expansions and higher order terms may be incorporated sequentially, as shown in Table 1. However, expanding the reection coefcient expansion (8) about = 60o does not produce a similar common sequence, although values of coefcients are rationals.

In turn the resulting coefcients may be incorporated into the nite element model (4), through a modied boundary condition
m=M am 2m p p = (ik)p + , n (ik)2m s2m m=1 p where s is the tangential variable, since for plane waves s2m = (ik)2m sin2m ( )p, and hence the matrix element corresponding to the upper radiation boundary condition in (4) becomes
2m

n,n [ K3 ]

1 = (H) 2

m=M

(ik)

m=1

am (ik)2m

2m N(z) + k2 N(z) 2m z

z=H

TABLE 1. Coefcients for reection coefcient expansion a1 a2 Minimise about , = 0 (Model A) 1/2 1/2 1/2 1/2 2/3 1/3 3/4 1/24 0.6932 0.3681 0.5876 -1/8 -1/8 -1/8 -4/9 2/3 -13/6 -0.4266 0.2987

a3

a4

= 600 (Model B)

1/16 1/16

-5/128

20/27 -82/27

-136/81

Minimise average reected power (Model C)

0.5459

In the following calculations Model C was employed with three coefcients and the upper articial boundary condition was located around six wavelengths above the ground oor.

RESULTS
We show in this section some results illustrating the use of the numerical method described in the previous section. It will be recalled that the numerical model assumes a coherent line source of sound. For the results shown four superelements were used. Height of the region was 9.3 m with a total of 348 degrees of freedom at frequency 200 Hz studied. Figure3 shows a comparison between the spectral method and the boundary element along a xed height with respect to excess attenuation and exact solution for rigid ground. The number of unknowns in the boundary element method totalled for 420 at this frequency. For the numerical results the admittance b = 0 for the absorbing ground is calculated using the Delany & Bazley formulae [13] which give the normalised admittance (G ) and complex wavenumber (kG ) of a porous medium as functions of / f , where is an effective ow resistivity and f is the frequency. Modelling the ground as a porous layer of thickness D on top of a rigid half-space, and assuming the refractive index |kG /k| >> 1 so that the ground is locally reacting with

5 0 Excess attenuation, dB. (z0 = 1.0 m, z = 5.0 m) 5 BEM for absorbing ground: 1040m Exact solution rigid ground Spectral FE for absorbing ground

10

11

10

12

15 20 25

13

14

30 35 40 20

15
0 20 40 60 Horizontal distance, x (m) 80 100

10

20

30 40 Horizontal distance, x (m)

50

60

FIGURE 3. Excess attenuation along level z = 5.0 m for rigid ground (dashed), grassstrip between 1040 m using BEM model (dashed-dots) and Spectral FEM model (solid)

a ow resistance of = 250, 000 Nsm4 and depth, D = 0.1 m, values appropriate to grassland. Values taken outside strip were for rigid ground a = c = 0. Results are compared with predictions from the boundary element method described in [14]. The following calculations were carried out in terms of excess attenuation which is dened by EA = 20 log10 p(x) dB pFF (x) (9)

where p, pFF denote actual and free-eld acoustic pressure respectively. In free-eld conditions, the whole of space is lled with the medium of propagation and there are no boundaries.
5 0 5 10
12

BEM for absorbing ground :1040 m Exact solution rigid ground Spectral FE. Absorbing and 1% varying wavespeed

10

11

15 20 25
14 13

30 35 40 20
15 0 10 20 30 40 50 60

20

40

60

80

100

Horizontal distance, x (m)

FIGURE 4. Excess attenuation along level z = 5.0 m for rigid ground (dashed); and grassstrip between 1040 m using BEM model (dashed-dots) and Spectral FEM model (solid) for varying sound prole a = 0.01.

Figure4(b) clearly shows a small increase in sound level in the immediate vicinity of the grassstrip discontinuity due to the sound velocty prole. An increasing sound

speed prole when a positive gradient is common. This is caused by the rapid cooling of air at the surface as heat is now absorbed by the ground. This is called an inversion or negative lapse and the sound waves are bent downward. This phenomena explains why sound sometimes travels much better at night, because it is focused along the ground instead of radiating upward.

CONCLUSIONS
A general discussion has been given of radiating boundary conditions for the wave equation within a variational framework. The ability to permit passage of waves with nonnormal incidence is obtained by a formal factorization of the wave equation, followed by an approximate summation representation of the reection coefcient corresponding to outgoing waves. The concept with selected angles of ideal transmission is discussed, and it is demonstrated that any of these schemes can be formulated in terms of higherorder differential operators in the co-ordinates of the radiating boundary. The method was demonstrated with a classic example from sound propagation over an embedded grassstrip in an atmospheric prole. Results show some promise in this direction in terms of accuracy in the proximity of the grassstrip discontinuities.

ACKNOWLEDGMENTS
The authors wish to acknowledge support from a Marie Curie Individual Fellowship (HPMF-CT-2000-01080).

REFERENCES
1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. K. B. Rasmussen, J. Sound Vib. 104, 321335 (1986). A. D. Pierce, Acoustics: An Introduction to Its Physical Principles and Applications, Acoustical Society of America, New York, 1989. A. Berry, and G. A. Daigle, J. Acoust. Soc. Am. 83, 20472058 (1988). M. J. White, and K. E. Gilbert, Appl. Acoust. 27, 227238 (1989). A. T. Peplow, and S. Finnveden, J. Acoust. Soc. Am. 116, 13891400 (2004). B. Engquist, and A. Majda, Math. Comp. 31, 629647 (1977). D. Givoli, J. Comp. Phys. 94, 129 (1991). D. Givoli, Appl. Num. Math. 33, 327340 (2000). S. Finnveden, Acta Acustica 2, 461482 (1994). S. Finnveden, Acustica / Acta Acustica 82, 479497 (1996). S. Finnveden, J. Sound and Vib. 199, 125154 (1997). S. Finnveden, J. Sound and Vib. 208, 685703 (1997). M. E. Delany, and E. N. Bazley, Appl. Acoust. 3, 105116 (1970). A. T. Peplow, and S. N. Chandler-Wilde, J. Sound Vib. 146, 303322 (1999).