You are on page 1of 7

An improved ray-tracing algorithm for predicting sound propagation outdoors

Kai Ming Li,a) Shahram Taherzadeh, and Keith Attenborough


Engineering Mechanics Discipline, Faculty of Technology, The Open University, Milton Keynes MK7 6AA, United Kingdom

Received 14 May 1997; revised 18 May 1998; accepted 8 July 1998 A new ray-tracing method for predicting the sound eld near a at impedance ground in a refracting atmosphere is developed that includes the effect of vector wind and turbulence explicitly. Improvements on previous ray-tracing schemes include the use of a generalized Snells law, an integration by means of a Gaussian quadrature, and a bracketing method for nding the ray paths. For the turbulence calculations, the interray coherence is determined from turning point heights rather than from integrals along the ray paths. A more efcient algorithm is developed for computing the sound eld above an impedance plane in a realistic atmosphere. Despite the inherent limitations of the ray-trace approach, particularly in respect of logarithmic wind proles, the algorithm is valid over a wide range of practical situations. 1998 Acoustical Society of America. S0001-49669804010-7 PACS numbers: 43.28.Fp, 43.20.Fn LCS

INTRODUCTION

The heuristic ray-tracing method1 is a convenient and computationally undemanding way of predicting the sound propagation outdoors in the presence of wind and temperature gradients above an impedance plane. In a recent study,2 Li has shown that the ground and surface wave terms can be included rigorously by using an asymptotic analysis. Raspet et al.3 have demonstrated that the sound eld predicted by the heuristic ray-tracing model,4 using an effective sound speed gradient to represent the combined effects of wind and temperature, agrees remarkably well with that calculated by the fast eld program FFP at long ranges. Furthermore, Salomons5 has included atmospheric turbulence in his ray model for the computation of sound eld in a temperature stratied medium. The purpose of this paper is to present an efcient numerical implementation of the theory developed in Ref. 2 in a way that includes the effect of vector wind and turbulence explicitly. It should be noted that the analytical solution for the sound eld above an extended-reaction plane is derived in Ref. 2. However, most of the outdoor ground surfaces are locally reacting. Hence, although it is straightforward to derive a prediction for propagation over an extended reaction plane, we consider only the locally reacting plane. The ray-tracing method used in the present paper differs from the traditional one.6,7 First, when tracing the ray paths, we use a numerical integration algorithm instead of the Euler method. Alternatively, a fourth-order RungeKutta algorithm with adjustable step sizes may be used. Second, to search for the eigenrays, we use a bracketing method in place of the hit-and-miss approach. The traditional accuracy of computation of the phase angle is improved by using the modied Snells law to calculate the ray angle along the ray
a

path, and by using numerical integration to compute the acoustical path length. Turbulence is included in the formulation. However, the interray coherence coefcient is computed from the turning point heights rather than from detailed tracing of the ray paths. The inherent limitations of the ray-trace method for predicting the sound eld above an impedance plane under logarithmic sound speed gradients are explored. In Sec. I, the theoretical basis for the ray-trace implementation is outlined. Section II details the new computational features. Section III describes the method for including turbulence. Section IV presents results of computations and comparisons with FFP predictions. Finally, in Sec. V, we offer some concluding remarks.
I. MODIFIED WEYLVAN DER POL FORMULA

In an earlier study, Li2 has demonstrated that the sound eld due to a monopole source above a porous ground in the presence of temperature and wind gradients can be written in form that is recognizably similar to the WeylVan der Pol formula. The theory may be summarized as follows. In a moving stratied medium, the index of refraction m ( z ) is dened as m z n z , 1 M z cos w sin 1

Corresponding author: present address: Department of Mechanical Engineering, Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong. J. Acoust. Soc. Am. 104 (4), October 1998

where n ( z ) is the index of refraction in an otherwise homogeneous medium at height z above the impedance plane. The symbols, u and w , are the magnitude and direction of wind, ( z ) and are the polar and azimuthal angles of a wavefront normal, c ( z ) is the speed of sound, and M ( z ) u / c is the Mach number. We note that the azimuthal angle is constant for a given wavefront normal, but the polar angle varies as a function of height in a moving stratied medium. Using the index of refraction in a moving medium, we can derive the modied Snells Law for a wavefront normal as
1998 Acoustical Society of America 2077

2077

0001-4966/98/104(4)/2077/7/$15.00

m sin sin 0 constant,

F w 1 i we w erfc iw ,
2 1 2 ik 0 R w 2 2 2 cos 0 e ,

7b 7c

where 0 is a reference polar angle of a ray at z 0, for example. In other words, we can determine the wavefront normal of a particular ray by specifying a polar angle at a given height and its corresponding azimuthal angle. As pointed out by Ostashev,8 the trajectory of the wavefront normal should be distinguished from the sound ray that travels from the source to receiver. In essence, the wavefront normal is not coincident with the sound ray in a moving medium. To aid theoretical and numerical analyses, we nd it more convenient to dene the acoustical path length, R L ( , ), the Doppler factor D ( z ), the stratication factor S ( , ), and the Jacobian factor J ( , ) as follows: R L ,

where erfc( ) is the complementary error function, e is the effective admittance of the porous ground given by
2 e e e 2 n sin 0 ,

m cos dz r sin 0 cos ,

3 4

D z 1 M cos w sin 1 , S , and J , D z D zs


z zs
2

cos 0 m cos

cos 0 , m s cos s 5a
2


1 sin 0

2R L 2R L 2R L 2 2 0 0

5b

where is the density of air which is a function of height z and r is the horizontal separation between the source and receiver. The subscripts and denote the variables to be evaluated at z and z , respectively, where z and z are the lesser and the larger of the source and receiver height. In Ref. 2, an elaborate analysis has been used to derive an expression for the sound eld above an extended reaction ground in a stratied medium. However, a heuristic model for the inclusion of the ground wave is found to be adequate for most practical situations in a homogeneous medium.9 Since the interaction between the sound waves and ground should remain unaffected by the presence of wind and temperature gradients, it is proposed to use the same approach for the case of a stratied medium. Suppose that the polar and azimuthal angles of the wavefront normals of the direct and the reected waves are , , , and , respectively. Then, the sound eld, above an extended reaction ground surface, can be approximated according to the heuristic model by2,9 p r ,0,z
1 S d e ik 0 R

e ( 0 / 2 ) and e n ( k 2 / k 0 ) are, respectively, the ratios of densities and wave numbers for the upper and lower media. The subscript 2 denotes the parameters of the lower medium, i.e., the porous ground, whilst the subscript 0 denotes those for the upper medium with the vertical height z 0. As shown in Eq. 6, k 0 R 1 and k 0 R 2 are the total phase changes for the direct and reected waves of the sound propagated from the source to the receiver. Hence, R L ( , ) and, R 2 R L ( R1 , ) may be regarded as the corresponding acoustical path lengths. The quantities S d S ( , ) , J d J ( , ) , S i S ( , ), and J i J ( , ) are the corresponding stratication and Jacobian factors for the direct and reected rays. Also, it is of interest to note that the receiver position is given in the cylindrical polar coordinate system with r as the range and z the height of the receiver. Without loss of generality, the azimuthal angle of the receiver is set at zero. For a locally reacting surface, we have e n 1. Hence, it follows from Eq. 8 that the effective admittance becomes

e e e n

0c 0 , 2c 2

4 J d

R 0 1 R 0 F w

S i e ik 0 R 2

4 J i

where k 0 is the reference wave number, R ( 0 ) is the plane wave reection coefcient, F ( w ) is the boundary loss factor, and w is known as the numerical distance. These parameters can be determined according to

0 R
2078

0 e cos , cos 0 e

7a

which is the specic normalized admittance of the ground. Furthermore, in the special case of a stationary medium where M 0, Eq. 6 can be reduced to the form given in Ref. 11 see his Eq. 58. The above analysis is based on the method of Fourier transformation and the well-known WKB approximation in which the sound eld can be expressed in terms of an integral representation. By evaluating the integral asymptotically, we can conrm the validity of the heuristic approximation used in the previous analyses2 which include the ground wave term explicitly the second term of the curly bracket in Eq. 6. In addition, the previous analysis is valid for a relatively short separation between the source and receiver. This implies that, in an upward refracting medium, the analysis is invalid in the shadow zone and in close proximity to the shadow boundary. Also, in a downward refracting medium, the analysis is only satisfactory for a close range such that the reected wave suffers only a single bounce. Also the receiver must not be close to a caustic where the Jacobian factor vanishes. In another study,10 Li has extended his previous work11 to allow for multiple bounces in a temperature stratied but downward refracting medium. Although the analysis is based on a monotonically increasing function of c ( z ), the principle should apply equally for other more intricate proles that may include regions, where dc ( z )/ dz is zero. This earlier work can be extended readily and straightforwardly to take account of the wind effects, but the details are not shown
Li et al.: An improved ray-tracing scheme 2078

J. Acoust. Soc. Am., Vol. 104, No. 4, October 1998

here. The asymptotic solution for the total eld above a locally reacting impedance plane can be derived to yield p r ,0,z exp i k 0 R 1 1 4Rd

Qj

j j exp i k 0 R 2 2

j 4R r

10

where the subscript j represents the corresponding parameters of the reected wave that hits the ground j-times before reaching the receiver. Since the sound eld is expressed in an analogous form as the WeylVan der Pol formula, R d j) ( j) ( j) ( J d / S d ) and R ( r ( J i / S i ), which have the dimension of length, may be treated as the effective path lengths for the direct and reected waves. These terms are introduced for the ease of reference. The quantity Q j is the spherical wave reection coefcient for the reected wave:
j j j j Q j 0 1 0 F w 0

ally involves some form of hit-and-miss approach. A ray is launched in a given direction at the source and then it is determined whether the launched ray hits the target at the receiver location. To reduce computational time, the target is often restricted to a nite area. The numerical accuracy is controlled by setting the size of the target as well as the step size in tracing the ray path. The computational time increases as the step length and target are made smaller. In this paper, we propose a bracketing scheme that avoids the need for blind shooting. Instead of using the eikonal equations as our starting point, we determine stationary points , for the direct wave and ( , ) for the reected wave for R L ( , ) in Eq. 3. The eigenray can then be determined by solving a pair of nonlinear equations for and as follows: r sin and r cos

11

M sin w dz cos

14

and, for an impedance plane, the plane wave reection coefj) ( cient ( 0 ) is given by
j cos 0 j , 0 cos j 0

12

sin M cos w dz . cos

15

where the specic normalized admittance is used in favor of the effective admittance e . The boundary loss factor is determined by using Eq. 7b with the numerical distance w j given by
j j 1 j cos w 0 2 ik 0 R 2 0 .

13

It should be noted that one has to nd all possible rays or the so-called eigenrays for a given j. Each eigenray, which j) hits the ground j-th time, has different values of ( 0 , wj , j , and Q j . Equation 10 describes the solution as the sum of contributions from all possible rays linking the source and receiver. We remark that there is at most one possible direct ray where the combined wind and temperature gradients give ( j) rise to monotonic proles. When calculating R , 1 , Rd , R 2 ( j) and R r , it is crucial to include all possible branches of the ray traces. Furthermore, there are additional phase shifts, 1 j) and ( 2 for the direct and reected waves as a result of the ray grazing the caustic. There will be a phase reduction of /2 each time the ray touches a caustic.12
II. AN IMPROVED SCHEME FOR RAY TRACING

To determine an eigenray, it is sufcient to specify the polar angle at a given height and the azimuthal angle of a wavefront normal because the polar angle at other heights can be found by using the modied Snells law i.e., Eq. 2. The small ratio between wind and sound speed ( M 0.03) in the normal atmospheric environment means that the azimuthal angle of the wavefront normal is very close to the azimuthal angle of the receiver, and, therefore, 0 may be used as the rst approximation. We can determine ( z ) from Eqs. 2 and 15 for a given range r. A second approximation for can be found by using ( z ) and Eq. 14 to give

tan

sin w z M /cos dz

cos

z w z M /cos

dz r

16

In the past, Thompsons method6,7 and its derivative13 were commonly used as the basis for the ray-tracing algorithm in a moving stratied medium. These previous methods are based, in turn, on the theory derived by Blokhintzev.14 Nevertheless, we have shown2,15 earlier that the theoretical results of our present approach are identical to that of the previous methods. These earlier methods often involve a pair of rst-order differential equations which can be solved either by numerical integration13 or by the construction of a nite-step wavefront i.e., by the Euler method.6,7 More importantly, the search for an eigenray usu2079 J. Acoust. Soc. Am., Vol. 104, No. 4, October 1998

The integration in Eq. 16 can be carried out numerically and hence evaluated. This iterative process in determining and ( z ) can be repeated until the required accuracy is achieved. In practice, is very small and one only needs two or three iterations for most situations. Next, we describe an improved method of nding ( z ) for a given . We restrict our analysis to a monotonically increasing function of m ( z ) and describe a sure-hit method in contrast to the conventional approach of trial-anderror. As mentioned earlier, it is sufcient just to specify 0 (0) , for example, as the unknown variable in Eq. 15. The polar angle at other heights can be determined through the use of Eq. 2. Determination of 0 involves the evaluation of the integral given in Eq. 15. There is an integrable singularity at the turning point where the polar angle is /2 or the slope of the wavefront normal is zero. Substituting Eq. 14 into 15, we obtain

tan dz

r sin w . sin w

17
2079

Li et al.: An improved ray-tracing scheme

Using partial integration, we can remove the integrable singularity of the left side of Eq. 17 and make the subsequent numerical analysis somewhat simpler. In this case, Eq. 17 can be rearranged in a more convenient form as I 0 where I 0 m sin cos r sin w sin w m

,
z

18

sin cos 1 mm / m 2 dz ,

19

and the primes denote the derivatives with respect to z. The above analysis is valid as long as the derivative of the index of refraction m does not vanish at any point along the ray path. Physically, this situation corresponds to the case of an innitely narrow sound channel where a ray will be trapped. Numerical integration is required in general to calculate I ( 0 ) in Eq. 19. On the other hand, the computation of denite integrals may be regarded as an initial value problem. By differentiating both sides of Eq. 19 with respect to z, we transform it into the following rst-order differential equation: dI mm sin cos 1 , dz m2

FIG. 1. Schematic diagram to show a ray is launched at an angle ( z ). a ( z ) u : Sound ray cannot reach the receiver at height greater than z . b ( z ) u : Critical ray, the ray reaches the receiver at height z at the turning point. c ( z ) u : Sound ray can reach the receiver at height z either at the ascending or descending branches.

20

with the initial condition of I 0 at z z . We wish to compute the function I at z z with due consideration for tracing all branches for a complete ray path. It is adequate to just calculate the integral from z to the turning point, from z to the turning point, and from the ground surface to the turning point. The appropriate number of these portions are then added together to trace all eigenrays in all possible ways. A typical approach involves the use of a nite step size ( z ). The increment of I can be computed by multiplying the right side of Eq. 20 by z for each step. This method is sometimes called the Euler method. However, the Euler method is not recommended for practical use because other more sophisticated numerical methods provide a more accurate and stable solution.16 Since a high degree of accuracy is required in computing I ( 0 ), a fourth-order RungeKutta or the socalled adaptive BulirchStoer technique may be used; see for example, Ref. 16. Nevertheless, we have also used a direct numerical integration method to compute I ( 0 ). We nd that a 10 point Gaussian quadrature16 is sufcient for most practical purposes. It is also of interest to note that the height of the turning point z max can be determined by using Eqs. 1 and 2, and by noting ( z max)0. The nonlinear equation for z max can be solved readily by means of a standard NewtonRaphson method.16 To nd all solutions for Eq. 18, we introduce the eigenray error function, E 0 m sin cos r sin w I 0. sin w m

21

The eigenrays are then determined by minimizing E ( 0 ) for a given source/receiver geometry and wind/temperature proles. The error function E ( 0 ) is sampled at a range of polar angles specied by the user. This allows the determi2080 J. Acoust. Soc. Am., Vol. 104, No. 4, October 1998

nation of zero crossings, which, in turn, provides information about a pair of bracketing polar angles. The spacing of the samples 0 is also set by the user. A smaller 0 and a large range of polar angles increase the probability that all eigenrays will be found at the expense of higher computational time. The Brent root nding algorithm16 is used to nd the eigenray solution when zero crossings are found. It is of interest to note that an initial examination of the wind velocity and sound speed proles will provide some useful insight in the choice of upper and lower bounds of the polar angles. For example, suppose a ray launches upward at some polar angle u that has its turning point at height z . In this case, it is clear that any ray launched upwards at a polar angle greater than u cannot reach a receiver situated at a height greater than z . Figure 1 shows the schematic diagram for the typical rays launching at different polar angles. Hence, it is obvious u provides a useful upper bound for the sampling range for the reected rays. On the other hand, the choice of the sampling range may also be decided on a trial and error basis. We also note that, in a downward refracting medium, a ray launched at an angle ( u ) has the same characteristic parameters as the ray launched upward with the polar angle u . Therefore, the down-going rays can be searched at the same time. Once the polar angle ( z ) of the wavefront normal has been determined, its value is substituted into Eq. 16, and that allows to be found by a simple iteration procedure as described earlier. We wish to point out that the principle described above can be generalized to other more intricate proles, but further developments are beyond the scope of the present paper. On nding the polar and azimuthal angles of all eigenj) ( j) rays, it is straightforward to compute R , J d , and J ( 1 , R 2 i by direct numerical integration. Making use of Eq. 5a, the j) stratication factors, S d and S ( can be determined, and i j) hence the effective path lengths R d and R ( r can be computed. These parameters are used, in turn, to compute the total sound eld using Eq. 10. The determination of J d , j) ( j) J( i and the reduction in phases 1 and 2 can be facilitated by noting

2R L cos2 0 2

n2 dz , m 3 cos3

22a
2080

Li et al.: An improved ray-tracing scheme

j) tions for replacing R d and R ( r in Eq. 10 when the receiver is situated far from the caustics.

III. ALLOWANCE FOR TURBULENCE

A quantity that is of particular interest in studies of outdoor sound propagation near the ground is the excess ground attenuation in the presence of turbulence ( A e ). Using a modied form of Eq. 10 and Ref. 4 Eq. 35, A c may be dened by A e 10 lg p 2 / p 2 d , where
FIG. 2. Plots of the geometrical path length and the effective path length of the direct wave versus height from the ground up to the turning point. The solid line represents the geometrical path length calculated according to Eq. 24 and the circles represent the effective path length calculated by J d / S d .
N

25
i1

p 2

q i 2

i0

2R L sin2 0 2
and

z z

cos k 0 R j R i Arg

R2 i

i1 j0

q i q j R iR j


qj qi

26

n m cos

M 2 sin2 w sin2 0 dz , m 3 cos3

22b

2R L cos 0 sin 0

nM sin w sin 0 dz . m 3 cos3

and p d is the free eld term. Note that Eqs. 25 and 26 are based on a study by Clifford and Lataitis17 but also see Eq. p 2 is the total sum of 35 of Ref. 4. The mean pressure, interactions between any two rays of a total of N 1 rays including the direct wave. The variables, R i and R i , represent the appropriate geometrical and acoustical path lengths. The term q i is the appropriate spherical wave reection coefcient, with q 0 1 which represents the direct wave term. Suppose that the i-th ray hits the ground l i times, then the reection coefcient is given by see Eq. 11 q i i i 1 i F w i l i , 27 where i is the angle of incidence of the reected wave, and ( i ) and w ( i ) are, respectively, the plane wave reection coefcient and the numerical distance see Eqs. 12 and 13. The variable is a turbulence parameter allowing for the destruction of coherence between the rays. The turbulence parameter is dened by exp 2 1 ,
2

22c

The equation of the caustic can be found by setting the Jacobian factor J, given by J

2R L 2R L 2R L 2 2 0 0

2 1/2

23

to zero,1 and it is then possible to determine 1 and ( j ) by examining whether the ray grazes the caustic. In view of the fact that ray theory breaks down near caustics and it is computationally more expensive to compute the Jacobian factor, therefore, it is found more convenient to use the geometrical path lengths in favor of the effective path lengths of the direct and reected waves, R d and j) 4 R( r . The geometrical path length, R g ( , ) is dened as the length of the ray trajectory linking the source and receiver in a vertically stratied medium. From the geometrical consideration, it can be calculated by R g ,

28

where and are the variance of phase uctuation along a path and the covariance between paired rays for equal source and receiver heights, respectively. They are dened as

2
and

n 2 k 2 0 RL 0

29a

L 0
2h

erf

h , L0

29b

dz . cos

24

Consequently, the geometrical path lengths of the direct and reected waves can be determined according to R 1 , ) , respectively. Their nu R g ( , ) and R 2 R g ( merical values can be computed efciently by a standard numerical integration routine. Figure 2 compares R 1 with the effective path length R d ( J d / S d ). It demonstrates that the corresponding geometrical path lengths are good approxima2081 J. Acoust. Soc. Am., Vol. 104, No. 4, October 1998

where erf is the error function with a real argument, n 2 is the refractive index variance, L 0 is the outer scale of turbulence, h is the maximum separation between paired rays, 1 if L 0 d / k 0 , 0.5 otherwise, and d is the distance between source and the receiver. The effect of atmospheric absorption is ignored in the present calculations, but it is straightforward to include it in our analysis. In previous raytracing methods,4 h may be computed from predictions of the complete ray paths. Here we follow the method of Raspet and Wu18 and dene it as one-half of the vertical distance
Li et al.: An improved ray-tracing scheme 2081

FIG. 3. Effect of turbulence on excess attenuation under downwind conditions with varying degrees of turbulence. Source and receiver heights are 1.5 and 1.8 m, respectively, and the horizontal separation is 1000 m. The assumed linear sound speed gradient is 10 4 s1 and the ground impedance is calculated using the two-parameter model with e 300 kPa s m2 and e 20 m1: solid line indicates n 2 0, dashed line: n 2 1.0 10 7 ; dashdot line: n 2 5.0 10 7 ; and dotted line: n 2 1.0 10 6 .

FIG. 4. Comparison between ray-trace and FFP calculations as a function of range at 1000 Hz under downward refraction conditions. The source and receiver heights and the ground impedance values are as for Fig. 2. The sound speed gradient is 0.1 s1.

between the turning points of the two ray paths. Since the locations of the turning points are calculated in the routine anyway, no extra computation is necessary.
IV. NUMERICAL RESULTS AND DISCUSSIONS

In the following numerical calculations, the specic surface impedance of the boundary is represented by a twoparameter model which assumes a rigid porous ground in which the porosity decreases exponentially with depth. The specic normalized admittance is given by a twoparameter model19

1 , 0.436 1 i e / f 0.5 19.48i e / f

30

where e and e are, respectively, the effective ow resistivity and the effective rate of change of porosity with depth. The values chosen for these parameters here are 300 kPa s m2 and 20 m1, respectively. They are typical of a grass-covered ground. Figure 3 demonstrates the result of including turbulence in the manner described in Sec. III for a source and receiver at 1.5 and 1.8 m heights, respectively, horizontal separation 1000 m, a linear sound speed gradient of 10 1 s1, and n 2 varying between zero and 10 6 . This gure essentially reproduces Fig. 8 in Ref. 4. Figure 4 compares predictions of transmission loss against range at a frequency of 1 kHz obtained from our ray-trace procedure with results of FFP calculations in the presence of a linear sound speed gradient of 0.1 s1. In this numerical example, 16 384 integration points and 1000 layers, each 0.5 m thick, were used to obtain the FFP results. However, the FFP calculations results have been plotted at 100 point intervals since ner interference structure than that shown will be destroyed by turbulence and has no practical signicance. The ray-trace predictions are in good agreement with the smoothed structure predicted by the FFP out to the 4-km range.
2082 J. Acoust. Soc. Am., Vol. 104, No. 4, October 1998

A comparison of excess attenuation calculations, at shorter ranges and at 1 kHz, indicates signicant discrepancies beyond 100 m Figs. 4 and 5. This comparison is similar to that in Fig. 9 of Ref. 4 where the ray-trace predictions included turbulence whereas the FFP calculations did not. Here the ray-trace predictions are made both with and without turbulence thus showing that the discrepancy is the result of the ray-trace approximation. Calculating the excess attenuation for sound downwind of a source in an atmosphere with a logarithmic wind speed prole u 1.73 ln(12.3z ) at a range of 250 m and with source and receiver heights 1.5 and 1.8 m, respectively shows clearly that the ray-trace method is decient when compared with an accurate FFP calculation Fig. 6. In this example, 1000 layers with thickness increasing with height were used in the FFP calculation.20 The thickness of each layer near the ground plane was of order of a millimeter for accurate discretization of the logarithmic sound speed prole. The problem for the ray-trace method as for the WKB Airy formulation of the FFP20 arises from violation of the approximation that requires small change in sound/wind ve-

FIG. 5. Comparison of FFP and the ray-trace predictions at 1 kHz and short ranges including the effect of adding turbulence. The source and receiver heights are as for Fig. 2 and the positive sound speed gradient is 0.34 10 1 s1. Li et al.: An improved ray-tracing scheme 2082

J. S. Lamancusa and P. A. Daroux, Ray tracing in a moving medium with two-dimensional sound-speed variation and application to sound propagation over terrain discontinuities, J. Acoust. Soc. Am. 93, 1716 1726 1993. 2 K. M. Li, A high frequency approximation of sound propagation in a stratied moving atmosphere above a porous ground surface, J. Acoust. Soc. Am. 95, 18401852 1994. Two typographical errors are noted. z The rst equation of Eq. 55 should be r z cos()tan dz and Eq. n a s dR a . R L( , ) Rm a 58 should be R 1
R

FIG. 6. Plot of excess attenuation for sound downwind of a source in an atmosphere with a logarithmic wind speed prole u 1.73 ln(12.3z ) . The solid line is the prediction by the ray method, while the line with circles is that of the FFP program. A maximum of 1000 layers was used in the FFP calculations and the thickness of each layer increased with height. The layer thicknesses near the ground plane were of order of a millimeter for accurate discretization of a logarithmic sound speed prole. The range is 250.0 m and the source and receiver heights were 1.5 and 1.8 m, respectively.

locity per wavelength. However, for the logarithmic prole, the change is the steepest for regions close to the ground. This renders the use of ray-trace method to be inappropriate for the logarithmic prole.
V. CONCLUDING REMARKS

The heuristic ray-based method for calculating the sound eld near an impedance plane beneath a refracting atmosphere has been extended to include the effect of vector wind and turbulence. Moreover the ray-based scheme has been improved computationally by using a generalized Snells law, a ten-point Gaussian quadrature, and a bracketing method for nding the ray paths. For the turbulence calculations, the interray coherence has been determined from turning point heights rather than from integrals along the ray paths. The new ray-trace method should be accurate and efcient for predicting sound elds at short range in any monotonically and slowly increasing wind prole. However, despite the computational improvements, it has been demonstrated that the new ray-trace method is inaccurate when used to predict sound elds in logarithmic wind proles.
ACKNOWLEDGMENT

This work was supported by the EPSRC, U.K. through Grant No. Ref. GR/J42052.

R. Raspet, A. LEsperance, and G. A. Daigle, The effect of realistic ground impedance on the accuracy of ray tracing, J. Acoust. Soc. Am. 97, 154158 1995. 4 A. LEsperence, P. Herzog, G. A. Daigle, and J. R. Nicolas, Heuristic model for outdoor sound propagation based on an extension of the geometrical ray theory in the case of a linear sound speed prole, Appl. Acoust. 37, 111139 1992. 5 E. M. Salomons, Downwind propagation of sound in an atmosphere with a realistic sound-speed prole: A semianalytical ray model, J. Acoust. Soc. Am. 95, 24252436 1994. 6 R. J. Thompson, Ray theory for an inhomogeneous moving medium, J. Acoust. Soc. Am. 51, 16751682 1972. 7 M. M. Boone and E. A. Vermaas, A new ray-tracing algorithm for arbitrary inhomogeneous and moving media including caustics, J. Acoust. Soc. Am. 90, 21092117 1991. 8 V. E. Ostashev, Ray acoustics of moving media, Izvestiya Atmos. Oceanic Phys. 25, 661673 1989. 9 K. M. Li, T. Waters-Fuller, and K. Attenborough, Sound propagation from a point source over extended-reaction ground, J. Acoust. Soc. Am. 104, 679685 1998. 10 K. M. Li, Propagation of sound above an impedance plane in a downward refracting atmosphere, J. Acoust. Soc. Am. 99, 746754 1996. 11 K. M. Li, On the validity of heuristic ray-trace based modication to the WeylVan der Pol formula, J. Acoust. Soc. Am. 93, 17271735 1993. 12 L. M. Brekhovskikh, Waves in Layered Media Academic, New York, 1980, 2nd ed., pp. 389395. 13 F. Walkden and M. West, Prediction of enhancement factor for small explosive sources in a stratied moving atmosphere, J. Acoust. Soc. Am. 84, 321326 1988. 14 D. I. Blokhintzev, Acoustics of a nonhomogeneous moving medium, NACA Tech. Memo 1399, Washington, DC English translation 1956. 15 K. M. Li, S. Taherzadeh, and K. Attenborough, A new approach in predicting sound propagation outdoors, Proc. 6th International Symposium on Long-Range Sound Propagation Ottawa, Canada, 1994, pp. 168180. 16 W. H. Press et al., Numerical Recipes in FORTRAN Cambridge U.P., London, 1992, 2nd ed., Chap. 4 for Gaussian quadrature, Chap. 9 for the Brent method, and Chap. 16 for the Bulirsch-Stoer method. 17 S. F. Clifford and R. J. Lataitis, Turbulence effects on acoustic wave propagation over a smooth surface, J. Acoust. Soc. Am. 73, 15451550 1983. 18 R. Raspet and W. Wu, Calculation of average turbulence effects on sound propagation based on the fast eld program formulation, J. Acoust. Soc. Am. 97, 147153 1995. 19 K. Attenborough, Ground parameter information for propagation modeling, J. Acoust. Soc. Am. 92, 418427 1992. 20 S. Taherzadeh, K. M. Li, and K. Attenborough, Some practical considerations for predicting outdoor sound propagation in the presence of wind and temperature gradients, Appl. Acoust. 54, 2744 1998.

2083

J. Acoust. Soc. Am., Vol. 104, No. 4, October 1998

Li et al.: An improved ray-tracing scheme

2083

You might also like