You are on page 1of 9

Applied Acoustics 150 (2019) 227–235

Contents lists available at ScienceDirect

Applied Acoustics
journal homepage: www.elsevier.com/locate/apacoust

Acoustic modelling of a transient source in shallow water


Roy L. Deavenport a,⇑, Matthew J. Gilchrest a, David J. Thomson b
a
Naval Undersea Warfare Center, Newport, RI 02841, United States
b
733 Lomax Road, Victoria, BC V9C 4A4, Canada

a r t i c l e i n f o a b s t r a c t

Article history: A desirable requirement for underwater acoustic models is the ability to simulate a transient signal prop-
Received 4 June 2018 agating in an ocean waveguide. Relevant broadband signals are generated, for example, by airguns used
Received in revised form 7 November 2018 in seismic exploration, transducers used for underwater communication systems, and vocalizations pro-
Accepted 24 January 2019
duced by marine mammals. Single-frequency acoustic models based on ray, mode, wavenumber integra-
tion or parabolic equation representations are often used for this purpose by employing a Fourier
synthesis method. In this case, the multiple-frequency, bandlimited medium transfer function is filtered
Keywords:
by the spectrum of the source and then inverse Fourier transformed to yield the time-domain waveform.
Convolution
Bandlimited
In contrast, direct convolution of a source signal with the medium’s impulse response can be carried out
Fourier synthesis in the time domain. In this paper we describe a convolution-based approach using a ray model represen-
Pekeris waveguide tation of the propagation whereby each eigenray amplitude is assumed to be independent of frequency
Ricker wavelet but its phase variation is taken into account. As a result, only a single-frequency calculation is required of
Image method the ray model at each range. The two methods are subsequently used to simulate a bandlimited transient
signal propagating in a shallow-water, lossy-bottom, Pekeris waveguide.
Published by Elsevier Ltd.

1. Introduction approach (see e.g. [7]). Since the effective spectral bandwidth is
determined by the frequency content of the source waveform, a
Current numerical models for simulating coherent underwater number of single-frequency acoustic predictions must be carried
sound propagation in realistic shallow-water waveguides have out that span this bandwidth using a frequency-sampling interval
evolved to the point of being routinely used for marine mammal determined by the time spread of the multipath arrivals at the
mitigation surveys, source localization based on matched-field range of interest. The resulting medium transfer function is then
processing (MFP) concepts [1] and geoacoustic inversion using multiplied by the spectral content of the source and the product
nonlinear, MFP-based, global optimization procedures [2]. Propa- is inverse Fourier transformed to yield the transient waveform. A
gation models that are based on the four standard representations real time-domain signal is obtained by ensuring that the real and
for the acoustic field have been developed into reliable and effi- imaginary parts of the complex frequency-domain product exhibit
cient computer codes, e.g., ray-based [3], wavenumber integral even and odd symmetry, respectively. Jensen et al. [8] used the
[4], normal mode [5], and parabolic equation [6] (see also [7]). Fourier synthesis method in this way to simulate ray, modal and
For each of these representations of the acoustic field, a time- parabolic equation broadband propagation for a Ricker wavelet
harmonic, (eixt ), single-frequency point source is considered. This source [9,10] in a 2D shallow-water waveguide with upslope/-
source derives from the historical motivation that underlies each of downslope bathymetry. Their study was aimed at assessing the
these representations, namely, to provide single-frequency esti- accuracy of the range-dependent modelling as well as the effi-
mates of transmission loss (TL) for use in navy sonar performance ciency of the various computer codes. Excellent inter-model agree-
prediction applications. ment was obtained for all of the acoustic models used except for
Any of the four single-frequency acoustic model representa- the ray-based code that had difficulty modelling the early arrivals
tions given above can be used to generate transient signals in a that were refracted from the near-surface sound speed profile. In
range-independent ocean waveguide by using a Fourier synthesis addition, for the modal codes, which made use of the adiabatic
approximation, an upgrade to coupled-modes was recommended.
In contrast to the 2D range-dependent waveguide used by Jensen
⇑ Corresponding author. et al., Sturm [11] has used the Fourier synthesis method to study
E-mail addresses: roy.deavenport@navy.mil (R.L. Deavenport), matthew. time-domain propagation in a range-, depth- and azimuth-
gilchrest@navy.mil (M.J. Gilchrest), drdjt@shaw.ca (D.J. Thomson).

https://doi.org/10.1016/j.apacoust.2019.01.028
0003-682X/Published by Elsevier Ltd.
228 R.L. Deavenport et al. / Applied Acoustics 150 (2019) 227–235

dependent upslope environment using a 3D parabolic equation a summary of the formulae used for computing the acoustic field in
model. terms of an approximate image theory for a shallow-water waveg-
In the time-domain, the Fourier synthesis method is equivalent uide with a penetrable bottom. In Section 3 we provide a numerical
to evaluating the convolution of the source wavelet with the med- example of propagation in a Pekeris waveguide and compare the
ium’s impulse response. This direct approach to modelling propa- single-frequency transmission losses between the approximate
gation in the time domain was recently carried out using a image theory and a wavenumber integral code [4] at the peak
modal representation for an isovelocity waveguide with perfectly frequencies (100 Hz and 200 Hz) of a Ricker wavelet. This is
reflecting boundaries [12]. The general causal solution of the inho- followed by a comparison between the Fourier synthesis and con-
mogeneous wave equation was obtained analytically from two volution methods for the time-domain Ricker wavelet signals at
independent solutions of the (incompletely) separated wave equa- ranges of 5 km and 10 km in this Pekeris waveguide. Section 3 con-
tion. Numerical examples were carried out for both cosine and cludes with a discussion of ray displacement and complex effective
gaussian source waveforms. Single-frequency transmission loss depth extensions to the standard image theory results. Our conclu-
comparisons were made between the exact normal mode solution sions and some discussions for further work are provided in
and a standard numerical mode model [5]. In the time domain, the Section 4.
normal mode solution was compared to an exact image method
solution. Excellent agreement was obtained for all model
2. Theory
comparisons.
An alternative convolution procedure for simulating a transient
A treatment of the acoustic wave equation for harmonic and
signal was developed in [13] and is based on a ray-based represen-
transient sources can be found, for example, in [7] and [19, p. 436].
tation of shallow-water propagation. In this previous work, the
In the following, we briefly review some basic concepts.
impulse response of the medium was convolved with a nonlinear
explosive source waveform whose peak pressure level is given
semi-empirically in terms of a scaled range (geometric similitude). 2.1. Broadband modelling
A recent experiment [14] involving explosive sources in shallow
water (< 15 m) used similar formulas based on measurements The time-dependent equation for the pressure field in the ocean
reported in [15] to analyze near-field (< 1 km) propagation to a in cylindrical coordinates ðr; zÞ is given by (assuming rotational
vertical line array (VLA) for determining peak and sound exposure symmetry),
levels. An important assumption that underlies the use of the polar !
1 @2 dðr Þdðz  z0 Þ
form of the ray theory model used in [13] is that each eigenray r2  Pðr; z; t Þ ¼ Sðt Þ : ð1Þ
amplitude is assumed to be independent of frequency. As a result, c2 ðr; zÞ @t 2 2pr
the phase variation of the nth eigenray is readily handled solely via
where
x in the factor eixT n where T n is its delay time (see Eq. (16)) prior to
summing the eigenray contributions in the calculation of the med-  
1 @ @ @2
ium’s transfer function at a given range. After inverse transforming r2 ¼ r þ 2; ð2Þ
r @r @r @z
to the time domain (via an FFT), the resulting impulse response is
convolved with the source waveform to generate the simulated and cðr; zÞ is the sound speed as a function of range r and depth z.
transient signal in the waveguide. Here SðtÞ is a real, linear point-radiated signal [units MT2] at the
It is worth mentioning that a nonlinear progressive wave equa- channel (filter) input and the instantaneous pressure at the output
tion based on the parabolic approximation (NPE) was developed by is given by Pðr; z; t Þ. The pressure is defined dimensionally by
McDonald et al. [16] to solve the nonlinear Euler equations numer- [ML1T2]. A point source is located at ðr; zÞ ¼ ð0; z0 Þ. In our numer-
ically. They used the NPE model to estimate the effects of shock- ical work, Sðt Þ is given explicitly by the Ricker wavelet [9,10]
wave propagation in an ocean waveguide. In addition, Cotaras    
2 2
et al. [17] developed a nonlinear ray theory for finite amplitude SðsÞ ¼ 1  2p2 f p s2 exp p2 f p s2 : ð3Þ
propagation. A summary of this ray-based progressive wave prop-
agation in the context of nonlinear acoustics can be found in [18]. where s ¼ t  t 0 and f p denotes the peak frequency of its spectrum,
In this paper we use the convolution method developed in [13] i.e.,
to propagate a Ricker wavelet [9,10] in a shallow-water waveguide
f
2  
similar to the one analyzed previously by Jensen et al. [8]. We 2 2
Sð f Þ ¼ pffiffiffiffi 3 exp f =f p : ð4Þ
introduce a simplified version of their shallow-water waveguide, p pf p
i.e. we replace their downward-refracting sound speed profile by
an isovelocity layer and the range-varying bathymetry by a flat In Fig. 1 (top) we show a Ricker wavelet for a peak frequency
bottom. These modifications transform their waveguide into one f p ¼ 200 Hz. Its normalized asymmetric spectrum is presented in
of Pekeris type. This range-independent environment allows us Fig. 1 (bottom) and is seen to have an effective bandwidth between
to use the wavenumber integration code SAFARI [4] to provide 10 Hz and 450 Hz.
time-domain simulations computed using the Fourier synthesis
method to use for comparison. In addition, it was convenient to 2.2. Fourier synthesis
model the ray representation in terms of image theory where each
bottom-interacting image is modified by the plane-wave reflection In underwater acoustics, Fourier synthesis is traditionally used
coefficient at the appropriate grazing angle. Numerical justification to determine time-dependent solutions for broadband transient
for using image theory in this context is provided in the section on signals. The reason for this is that any of the usual numerical
numerical results. approaches developed for a single-frequency source (ray,
In Section 2 we briefly review the theory underlying two wavenumber integral, normal mode or parabolic equation) can
methods for computing the time response in a shallow-water be run (without modification) for a number of frequencies that
waveguide due to a transient source: Fourier synthesis and convo- span the bandwidth of the transient source. From the convolution
lution. This is followed by a description of the bandlimited impulse theorem, the resulting transfer function can be multiplied by the
function in its polar (ray-based) form. This section concludes with spectrum of the source and inverse Fourier transformed to yield
R.L. Deavenport et al. / Applied Acoustics 150 (2019) 227–235 229

Fig. 1. (top) Ricker wavelet signal for a peak frequency of 200 Hz. (bottom) Normalized spectrum.

the time-domain signal. To see this, we can form a Fourier trans- extensively by seismologists for treating transients in the earth.
form on Eq. (1) to obtain A modern treatment of the Cagniard method is given by Aki and
Z 1 Richards [23]. Our convolution procedure is more general than
pðr; z; xÞ ¼ Pðr; z; t Þ eixt dt; ð5Þ the Cagniard method in that it does not require homogeneous lay-
1
ers. However in both methods one convolves an arbitrary source
where pðr; z; xÞ satisfies the following equation with the impulse function of the medium.
h i dðr Þdðz  z0 Þ
In the convolution method Eq. (1) can be written symbolically
r2 þ k2 ðr; zÞ pðr; z; xÞ ¼ SðxÞ ; ð6Þ in the form
2pr
dðr Þdðz  z0 Þ
and k ¼ x=cðr; zÞ is the total wavenumber. Knowledge of the har- LPðr; z; tÞ ¼ Sðt Þ ; ð10Þ
monic Green’s function Gðr; z; xÞ, which is a solution to the 2pr
single-frequency, point-source Helmholtz equation [20], namely where the linear differential operator L is defined by
h i !
dðrÞdðz  z0 Þ 1 @2
r2 þ k2 ðr; zÞ Gðr; z; xÞ ¼  ; ð7Þ L¼ r  2
2
: ð11Þ
2pr c ðr; zÞ @t 2
leads to the time-domain representation
Z 1
From Eq. (9), the retarded time-domain Green’s function impulse
1 hðr; z; t  t 0 Þ for a point source in space and time is
Pðr; z; t Þ ¼ Gðr; z; xÞSðxÞ eixt dx ð8Þ
2p 1
dðrÞdðz  z0 Þdðt  t 0 Þ
which provides the desired Fourier synthesis in terms of the pro- Lhðr; z; t  t0 Þ ¼ ; ð12Þ
2p r
duct of the source and medium transfer functions. In the next sec-
tion, we require the time-domain impulse response hðr; z; t Þ defined where t 0 is the initial response time. Here the impulse function
as the inverse Fourier transform of the medium transfer function, hðr; z; t  t 0 Þ depends only on relative time ðt  t 0 Þ. Note that all
i.e., terms in equation Eq. (12) must be dimensionally homogeneous,
Z 1
with each term of dimension L3T1 and the three-dimensional
1 time-dependent impulse has dimensions of L1T1.
hðr; z; tÞ ¼ Gðr; z; xÞ eixt dx: ð9Þ
2p 1 Multiplying both sides of Eq. (12) by Sðt 0 Þ and integrating with
respect to t 0 yields
2.3. Convolution Z 1
0 dðrÞdðz  z0 Þ
Lhðr; z; t  t 0 ÞSðt 0 Þdt ¼ SðtÞ : ð13Þ
0 2pr
There exist situations where the Fourier synthesis approach can
be computationally difficult, such as when the number of frequen- If all initial conditions are assumed to be zero, then the order of dif-
cies required is large or when the source signal does not possess a ferentiation and integration in Eq. (13) can be interchanged since L
Fourier transform (see [21] for a discussion of these and other lim- was defined to be linear. Consequently, Eq. (13) can be put in the
itations). In these cases, alternate methods for modelling broad- equivalent form,
band propagation need to be considered. Direct convolution in Z 1
0 dðr Þdðz  z0 Þ
the time domain is a method that does not involve Fourier trans- L hðr; z; t  t 0 ÞSðt 0 Þdt ¼ Sðt Þ : ð14Þ
0 2pr
forming the signal, that may be a problem for initial value prob-
lems. Cagniard [22] was one of the first to use convolution to Comparison of Eq. (14) with Eq. (10) reveals the convolution inte-
solve time-dependent initial value problems. It has been used gral for the instantaneous pressure P ðr; z; t Þ, namely
230 R.L. Deavenport et al. / Applied Acoustics 150 (2019) 227–235

Z 1
0 the eigenray information in Eq. (16) that is needed to determine
Pðr; z; t Þ ¼ hðr; z; t  t 0 ÞSðt 0 Þdt : ð15Þ
0 the bandlimited transfer function. Each eigenray follows a
sequence of straight-line segments that reflect from the surface
Eq. (15) gives the pressure solution to Eq. (1) in terms of a convolu-
and bottom of the waveguide (i.e. there is no refraction). The
tion of the bandlimited linear impulse response hðr; z; t  t 0 Þ with a
surface- and bottom-reflected eigenrays are modified by the
real point-radiated signal SðtÞ. This is analogous to using linear sys-
plane-wave reflection coefficients V s and V b respectively. Although
tem theory, but with a linear source that gets propagated and dis-
this representation is an approximation when V b ¼ V b ðhÞ, it can be
persed in a linear medium via convolution with a bandlimited
quite accurate in some situations [26, p. 346],[27, p. 139]. For
impulse function.
cb > c, the representation can be improved by taking ray displace-
ments at the sea-bottom interface into account for those grazing
2.4. Bandlimited impulse function
angles smaller than the critical ray angle cos1 ðc=cb Þ [28–30]. In
the present work, however, we have not implemented the ray
The spatial Green’s function Gðr; z; xÞ that satisfies the elliptic
displacement contributions (for reasons given in the discussion
Helmholtz equation in Eq. (7) has dimensions of L1 where the har-
following the presentation of numerical results).
monic time dependence expðixt Þ has been suppressed. The
The single-frequency image solution for the Pekeris
wavenumber kðr; zÞ ¼ x=cðr; zÞ where x ¼ 2pf is the angular fre-
waveguide can be cast in the form (adapted from the treatment
quency of the CW source. For an arbitrary angular frequency
in [26, p. 346]),
x; Gðr; z; xÞ can be expressed in terms of traveling waves using a
ray-based model representation in the form [24,25],  
expðikR01 Þ expðikR02 Þ X 1 X 4
  exp ikR‘j
pðr; zÞ ¼ þ Vs þ V ‘j h‘j
X
N R01 R02 ‘¼1 j¼1
R‘j
Gðr; z; xÞ ¼ An expðixT n þ i/n Þ; ð16Þ
n¼1 ð17Þ
where An is the path amplitude, T n is the delay travel time of the nth qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
eigenray path, and /n is the eigenray phase shift associated with where R01 ¼ r 2 þ ðz  z0 Þ2 and R02 ¼ r 2 þ ðz þ z0 Þ2 are the slant
boundary and/or caustic interactions. Eq. (16) is recognized as a ranges for the direct path (first term) and surface-image path
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
general form for the travelling wave Green’s function solution to (second term), respectively. R‘j ¼ r 2 þ z2‘j ; j ¼ 1; . . . 4 are the slant
the wave equation.
ranges for the ‘th set of four bottom-bounce image paths with
Gðr; z; xÞ is determined from the eigenray amplitudes and
reflection-coefficient products V ‘j given by
phases for a finite range of frequencies. With this representation,
a significant simplification occurs if the amplitude of each eigenray
V ‘1 ¼ V ‘1 ‘
s V b; h‘1 ¼ tan1 ðz‘1 =r Þ; z‘1 ¼ 2‘h  z  z0 ;
is independent of frequency. In this case, the only frequency-
dependent part of Eq. (16) occurs in the expðixT n Þ factor. As a V ‘2 ¼ V ‘s V ‘b ; h‘2 ¼ tan1 ðz‘2 =r Þ; z‘2 ¼ 2‘h  z þ z0 ;
ð18Þ
result, only a single-frequency calculation is needed by the V ‘3 ¼ V ‘s V ‘b ; h‘3 ¼ tan1 ðz‘3 =r Þ; z‘3 ¼ 2‘h þ z  z0 ;
image/ray model, and the bandlimited transfer function can be
V ‘4 ¼ V ‘þ1
s V b;

h‘4 ¼ tan1 ðz‘4 =r Þ; z‘4 ¼ 2‘h þ z þ z0 :
assembled very efficiently. It remains to transform Gðr; z; xÞ for a
receiver at each range and depth via an inverse FFT to obtain the
Here k ¼ 2pf =c is the wavenumber in 0 < z < h. For the pressure-
bandlimited impulse response which is then convolved directly
release sea-surface used here, V s ¼ 1. The slant ranges of the
with the time-domain source signature. This provides the instanta-
eigenrays R‘j have grazing angles h‘j at the receiver. The
neous pressure time series as given by Eq. (15).
frequency-independent fluid–fluid bottom losses V b are given by
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2.5. Image theory for the Pekeris waveguide   m sin h‘j  n2  cos2 h‘j
V b h‘j ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ; ð19Þ
m sin h‘j þ n2  cos2 h‘j
The lossy Pekeris waveguide used in the numerical calculations
is shown in Fig. 2. It consists of a uniform ocean of depth where m ¼ qb =q and n ¼ c=ðcb þ ibÞ with b ¼ a=ð40plog10 eÞ Np/m.
h ¼ 200 m, water sound speed c ¼ 1475 m/s, density q ¼ 1 g/cm3, Since kR‘j ¼ xR‘j =c, each eigenray term of Eq. (17) can be put in
and absorption a ¼ 0 dB/k overlying a fluid half-space bottom of the polar form of Eq. (16) in terms of its travel time T ‘j ¼ R‘j =c,
sound speed cb ¼ 1600 m/s, density qb ¼ 2 g/cm3 and absorption namely,
ab ¼ 0:5 dB/k. Here k is the acoustic wavelength in the relevant  
p‘j ¼ jV ‘j =R‘j j exp ixT ‘j þ i arg V ‘j : ð20Þ
medium. For this configuration, image theory can be used provide
The phase T ‘j is the same as the delay time that Chapman [24],
[31, p. 19] uses in his WKBJ formulation of seismic wave propaga-
tion. Moreover, it is analogous to Hamilton’s principal function of
classical mechanics as described by Whittaker [32] in his develop-
ment of a quantum mechanical principal function involving non-
commuting operators.

3. Numerical examples

3.1. Ricker wavelet (f p ¼ 200 Hz)

To provide some justification for using the approximate image


representation given above, at least for the specific material prop-
erties of the Pekeris waveguide in Fig. 2, we compare single-
frequency transmission losses versus range between the image
Fig. 2. Environmental parameters for the Pekeris waveguide. model and the SAFARI wavenumber integration model at the peak
R.L. Deavenport et al. / Applied Acoustics 150 (2019) 227–235 231

frequency f p ¼ 200 Hz of the Ricker wavelet. The source and recei-


ver depths are at z0 ¼ 100 m and z ¼ 20 m respectively. Here trans-
mission loss (TL) in dB re 1 m is defined by TL ¼ 20log10 jp=p0 j
where p0 is a reference pressure at a distance of 1 m. The coherent
TL comparisons are shown in Fig. 3 for the range interval 0–10 km.
Even without ray displacement corrections, the agreement
between the two models is seen to be excellent in this case.
The waveguide parameters considered here are very similar to
the ones used by Jensen et al. [8] in their broadband signal propa-
gation study with the exception of the sound speed profile in the
water column. To evaluate the Fourier synthesis versus convolu-
tion procedures in this waveguide, we compare the signal wave-
forms computed for the receiver at ranges of 5 km and 10 km,
respectively.
Fig. 4 shows the Fourier synthesis results compared to the
results obtained by convolution at a range of 5 km. From Fig. 1
(bottom) we observe that the effective bandwidth of the Ricker
wavelet for a peak frequency of 200 Hz is between 10 Hz and
450 Hz. At 10 km range, the multipath spread DT in the source sig-
nal is between 0.5 and 1.0 s so that a sufficient frequency sampling Fig. 4. Comparison of convolution and Fourier synthesis predictions of the Ricker
is satisfied by taking Df ¼ 1 Hz [8]. For the Fourier synthesis proce- wavelet signals (f p ¼ 200 Hz) in the Pekeris waveguide at a range of 5 km.

dure, the (complex) waveguide transfer response is first computed


by SAFARI at 441 frequencies that span the 440-Hz bandwidth. The resembles the one generated by the Fourier synthesis approach
transfer function is then multiplied by the (complex) source spec- at a range of 5 km.
trum and the product is inverse Fourier transformed to yield the To illustrate this comparison more closely, in Fig. 5 we show the
required time signature in the waveguide. A reduced sound speed signal characteristics of the Fourier synthesis and convolution
of 1475 m/s was used to confine the multipath signals to a time methods for an expanded time interval between 0.0 and 0.1 s,
window of 0.5 s at the 5-km range. This result is to be compared again for a receiver at a range of 5 km. For this early arrival portion
to the time signature computed using the convolution method of the time signatures, the agreement between the multipath
which is also shown in Fig. 4. For the convolution calculation, the structure between the two methods is seen to be excellent in both
amplitude of each eigenray evaluated by the image model is amplitude and arrival time.
assumed to be independent of frequency over the bandwidth of To further examine the capability of the convolution procedure,
the Ricker wavelet. As a result, the waveguide transfer function in Fig. 6 we compare both propagated waveforms at a range of
is easily modelled by simply populating the frequency-dependent 10 km. At this range, the pulses in the time window 0.0–0.5 s are
term eixT across the full x-bandwidth for each image contribution, seen to contain more multipath arrivals than those observed in
summing over the images at each frequency, and then taking the Fig. 4 for the receiver at the 5-km range. The agreement between
inverse transform to the time domain. The resulting impulse the different numerical approaches is again excellent for the early
response is then convolved with the source waveform to produce arrivals, but slight differences in amplitude between the two meth-
the signal that is propagated in the waveguide. It is evident from ods are noticeable for those pulses arriving at later times. The
Fig. 4 that this procedure results in a time signature that closely detailed agreement for the early arrivals at the 10-km range is

Fig. 3. Coherent transmission loss (TL) comparison at 200 Hz between the wavenumber integration model SAFARI and the image model for the Pekeris waveguide.
232 R.L. Deavenport et al. / Applied Acoustics 150 (2019) 227–235

Fig. 5. Early arrival comparison between convolution and Fourier synthesis Fig. 6. Comparison of convolution and Fourier synthesis predictions of the Ricker
predictions of Ricker wavelet signals (f p ¼ 200 Hz) in the Pekeris waveguide at a wavelet signals (f p ¼ 200 Hz) in the Pekeris waveguide at a range of 10 km.
range of 5 km.

easier to see in Fig. 7 where the comparisons are shown on an


expanded time interval of 0.0–0.1 s.

3.2. Ricker wavelet (f p ¼ 100 Hz)

Next, we examine the convolution and Fourier synthesis meth-


ods at the same 5- and 10-km receiver ranges but for a lower-
frequency signal. In Fig. 8 we display the TL versus range curve
comparison between SAFARI and image solution for a Ricker wavelet
with a peak frequency f p ¼ 100 Hz. The effective bandwidth is
reduced from that of the f p ¼ 200-Hz spectrum to between 10 Hz
and 250 Hz. At this peak frequency, the coherent TL interference
pattern shows less variation with range than was evident for the
200-Hz case. The agreement between TL estimates is still very good
but small differences can be noted.
Following the same procedures described above for the 200-Hz
case, we show in Fig. 9 the time-signature comparison for a 0.0–
0.6 s time window. Again the reduced velocity of 1475 m/s was
used to follow the signals to the 5-km range. The early arrivals
are in very good agreement between the two methods, but the Fig. 7. Early arrival comparison between convolution and Fourier synthesis
predictions of Ricker wavelet signals (f p ¼ 200 Hz) in the Pekeris waveguide at a
later arrivals indicate some amplitude differences. This is easier
range of 10 km.
to see in Fig. 10 which shows the results for an expanded 0.0–
0.2 s time window.
Finally, we show in Fig. 11 the time-signature comparison for [26,28–30,33–35]). For cb > c, this comes about by noting that for
the 100-Hz peak frequency wavelet at a range of 10 km. While grazing angles less than the critical angle, cos1 ðc=cb Þ, the ampli-
the initial portion of the signatures still show excellent agreement tude of the reflection coefficient V b is near unity and only its phase
in relative amplitudes and arrival times, the trailing signals show varies significantly there. As a result, one can replace V b in the inte-
some amplitude deviations. Fig. 12 shows the initial arrivals for grand by jV b j expði arg V b Þ and incorporate this phase into the sta-
an expanded 0.0–0.2 s time interval. At this scale, it is seen that tionary phase evaluation (for a detailed analysis see [34]). As
the later pulses are also beginning to exhibit slight differences in pointed out by Tindle [29] however, this maneuver introduces
arrival times as well as amplitudes. caustics into the region 0 < z < h to which caustic corrections must
be applied in order to obtain accurate estimates of the propagated
3.3. Ray displacement and complex effective depth field, especially at low frequencies.
For image theory, it is more natural to approach ray displace-
Each term in the image sum in Eq. (17) can be derived by apply- ment in the context of the effective depth approximation [35,36].
ing a stationary phase evaluation to its exact integral representa- This concept involves replacing reflection off a penetrable sea-
tion (after replacing the Hankel function factor in the integrand bottom with reflection off a perfect reflector located below the
by its asymptotic approximation). The angles h‘j are the specular sea-bottom in order to adjust the locations of the images to take
grazing angles of the boundary-interacting rays on which the slant ray displacement into account.
ranges R‘j and reflection coefficients V ‘j depend. Improvements to Consider the geometry illustrated in Fig. 13. A ray at grazing
this approximation can be obtained by taking into account the angle h is reflected from the sea-bottom with a phase change
effects of ray displacement at the sea-bottom boundary (e.g., wðhÞ. This ray represents an infinite plane wave. The phase of the
R.L. Deavenport et al. / Applied Acoustics 150 (2019) 227–235 233

Fig. 8. Coherent transmission loss (TL) comparison at 100 Hz between the wavenumber integration model SAFARI and the image model for the Pekeris waveguide.

Fig. 10. Early arrival comparison between convolution and Fourier synthesis
Fig. 9. Comparison of convolution and Fourier synthesis predictions of the Ricker
predictions of Ricker wavelet signals (f p ¼ 100 Hz) in the Pekeris waveguide at a
wavelet signals (f p ¼ 100 Hz) in the Pekeris waveguide at a range of 5 km.
range of 5 km.

reflected wave front AB can be exactly reproduced by placing a  i log V b ðhÞ þ p


Dh ðhÞ ¼ : ð22Þ
perfectly-reflecting, pressure-release boundary at a depth Dh 2k sin h
below the actual sea-bottom boundary where it undergoes a phase Fig. 13 suggests that the effective depth of the ‘jth image is
change of p. By straightforward geometry we find accommodated by setting
wðhÞ þ p 
DhðhÞ ¼ ; Dr ðhÞ ¼ 2 cot h DhðhÞ: ð21Þ z‘j ¼ z‘j þ 2jDh‘j : ð23Þ
2k sin h
The effective depth Dh in Eq. (20) gives rise to a ray displacement Dr One outcome of this representation is that for a lossy sea-bottom
as shown in Fig. 13. Weston [35] compares this lateral ‘‘wave shift” this effective depth becomes a complex number (denoted here by
to the ‘‘beam shift” resulting from incorporating arg V b into the sta- an asterisk). As discussed by Zhang and Tindle [36] this is necessary
tionary phase approximation for a specific Pekeris waveguide and for correct determination of the complex normal mode eigenvalues
shows numerically that although they are similar in magnitude (both propagating and leaky) by iteration for the Pekeris problem.
they differ slightly for grazing angles below the critical angle. Because we are dealing with a point source and its images, the
Zhang and Tindle [36] generalize the effective depth approxi- angle-dependence of the reflected field to a given receiver from
mation to include the loss encountered during the reflection pro- the complex boundary is not the same as the one depicted in
cess by introducing a complex phase wðhÞ ¼ i log V b ðhÞ so that Fig. 13. An iterative procedure for determining the true angle of
the effective depth becomes the reflected field for each image at the appropriate effective depth
234 R.L. Deavenport et al. / Applied Acoustics 150 (2019) 227–235

can be carried out in the following way. Starting with the specular
angle h‘j for the ‘jth image, one can use Eq. (22) to determine the

first approximation to Dh‘j to yield a new (complex) depth in Eq.
(23). From this (complex) image depth one can determine a new
(complex) grazing angle
 
h‘j ¼ tan1 z‘j =r ; ð24Þ

that in turn allows us to use Eq. (22) again to determine a new effec-
tive depth via Eq. (23) and grazing angle in Eq. (24). This process can
be repeated until an appropriate tolerance has been met. In our
experience, only a few iterations are needed to accomplish this. In
this context, the iteration procedure for determining the effective
depth for each image is the eigenray analog of the iteration proce-
dure used by [36] for determining the eigenvalues (horizontal
wavenumbers) of the normal modes.
Once this procedure has been carried out for each range r of
interest, the field there can be computed by summing the images
as before but now with each bottom-interacting image (at a com-
Fig. 11. Comparison of convolution and Fourier synthesis predictions of the Ricker plex depth) reflecting off a pressure-release boundary. As a result,
wavelet signals (f p ¼ 100 Hz) in the Pekeris waveguide at a range of 10 km. the waveguide field pðr; zÞ is now computed using Eq. (17) with R‘j
replaced with
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
R‘j ¼ r 2 þ z2 ‘j ; ð25Þ

and with V b ¼ V s ¼ 1.


Unlike the ray displacement obtained asymptotically, the lat-
eral shift induced by the effective depth approximation does not
become singular at the critical angle. Moreover, no caustics are
introduced by this now range-dependent, perfectly-reflecting
interface. For single-frequency calculations, we have found this
procedure always improves the TL estimates. One does have to deal
with complex angles, complex depths, complex slant ranges and
complex times, however. Unfortunately, it is clear from Eq. (22)

that the complex depth Dh‘j for each image is frequency depen-
dent. This violates the constant-amplitude assumption for our con-
volution method as angles, depths, slant ranges and travel times
for the effective depth approximation vary with frequency at each
receiver range r.
Consequently, we choose not to pursue incorporating the
broadband effects of ray displacement for our calculations in this
paper. This is partly due to the fact that the standard image
treatment seems to account for the propagation in our Pekeris
Fig. 12. Early arrival comparison between convolution and Fourier synthesis
predictions of Ricker wavelet signals (f p ¼ 100 Hz) in the Pekeris waveguide at a waveguide with acceptable accuracy for the 100- and 200-Hz
range of 10 km. peak-frequency Ricker wavelets. Addressing the minor disagree-
ments with the Fourier synthesis results is beyond the scope of this
paper. We find that the complex effective depth treatment,
however, sufficiently interesting to merit further consideration in
future work. The main point of comparing the Fourier synthesis
method to the convolution method is to demonstrate that, for
the Pekeris waveguide at least, comparable results can be obtained
with an equivalent convolution approach that needs much less
computational effort.
As for comparing the computing times involved in the above
calculations, we note that the Fourier synthesis method required
441 single-frequency calculations for the f p ¼ 200 Hz Ricker wave-
let source and 241 such calculations for the f p ¼ 100 Hz source.
These are to be compared to just one single-frequency calculation
for the image model at each range. It is difficult to provide an exact
meaningful comparison of computing times as the SAFARI pulse
module is provided in compiled FORTRAN while the image convolu-
tion is executed in GNU OCTAVE (compatible with MATLABÒ). Both
models were run on a 8-year-old laptop computer. The SAFARI code
Fig. 13. Geometry for plane wave reflection and the effective depth of a perfectly took 20 s for the 441 frequencies while the image code took a frac-
reflecting boundary. tion of a second.
R.L. Deavenport et al. / Applied Acoustics 150 (2019) 227–235 235

There is one final point to make on our image calculations. We [4] Schmidt H. SAFARI: Seismo-Acoustic Fast field Algorithm for Range-
Independent environments. User’s Guide, Rep. SR-113. San Bartolomeo,
did compare our image-based transmission-loss estimates to those
Italy: SACLANT Undersea Research Centre; 1988.
obtained with the gaussian beam model BELLHOP [37]. At low graz- [5] Porter M. The KRAKEN Normal Mode Program, Memo. SM-245. San Bartolomeo,
ing angles (h < 25 ), the TL versus r curves agreed within a line Italy: SACLANT Undersea Research Centre; 1991.
thickness for the 200-Hz peak-frequency case. This implies that [6] Collins M. The split-step Padé solution for the parabolic equation method. J
Acoust Soc Am 1993;93:1736–42.
BELLHOP does not account for ray displacement in its normal docu- [7] Jensen F, Kuperman W, Porter M, Schmidt H. Computational ocean acoustics.
mented mode of operation. Consequently, we are confident that 2nd ed. Springer; 2011.
the image method approximation as used in this paper is an [8] Jensen F, Ferla C, Nielsen P, Martinelli G. Broadband signal simulation in
shallow water. J Comput Acoust 2003;11:577–91.
acceptable ray-based model for validating the convolution method [9] Ricker N. The form and laws of propagation of seismic wavelets. Geophysics
as described herein for the specific Pekeris waveguide shown in 1953;18:10–40.
Fig. 2. [10] Wang Y. The Ricker wavelet and the Lambert W function. Geophys J Int
2015;200:111–5.
[11] Sturm F. Numerical study of broadband sound pulse propagation in three-
4. Conclusions dimensional oceanic waveguides. J Acoust Soc Am 2005;117:1058–79.
[12] Sertlek H, Aksoy S. Analytical time domain normal mode solution of an
acoustic waveguide with perfectly reflecting walls. J Comput Acoust
In this paper, a convolution-based, time-domain procedure was 2013;21:1250026.
described for simulating a transient signal propagating in a [13] Deavenport R, Gilchrest M. Time-dependent modeling of underwater
shallow-water waveguide. In the frequency domain, the signal explosions by convolving similitude source with bandlimited impulse from
the CASS/GRAB model, Tech. Rep. 12,176, Naval Undersea Warfare Center,
was taken to have the form of a sum of eigenray arrivals whose Newport, RI (June 30 2015).
amplitudes could be taken to be constant over the effective fre- [14] Soloway A, Dahl P. Peak sound pressure and sound exposure level from
quency band of the source. In this case, only a single-frequency underwater explosions in shallow water. J Acoust Soc Am 2014;136:EL218–23.
[15] Chapman N. Measurement of the waveform parameters of shallow explosive
propagation run is required at the range of the receiver: the charges. J Acoust Soc Am 1985;78:672–81.
frequency-dependent part of the total signal is contained in the [16] McDonald B, Kuperman W. Time-domain formulation for pulse propagation
time phasor and is readily incorporated into the medium’s impulse including nonlinear behavior at a caustic. J Acoust Soc Am 1987;81:1406–17.
[17] Cotaras F. Nonlinear effects in long range underwater acoustic propagation,
response before inverting via an inverse Fourier transform. This is Tech. Rep. ARL-TR-85-32. Appl Res Lab, Univ Texas Austin 1985.
in contrast to the usual Fourier synthesis method of generating [18] Morfey C, Cotaras F. Propagation in inhomogeneous media (ray theory). In:
bandlimited signals in the time domain: by running a single- Hamilton M, Blackstock D, editors. Nonlinear Acoustics, Acoust. Soc. of Amer..
p. 361–70.
frequency model many times at a set of frequencies across the
[19] Tolstoy I. Wave propagation. McGraw-Hill; 1973.
band, filtering by the source spectrum, and then taking an inverse [20] Deavenport R. A normal mode theory of an underwater acoustic duct by means
Fourier transform. Numerical examples were provided to compare of Green’s function. Radio Sci 1966;1:709–24.
[21] Sertlek H, Askoy S. Benchmarking of acoustic pulse propagation problems in an
time signatures at receiver ranges of 5 km and 10 km for two
isovelocity waveguide by an analytical time domain normal mode method.
Ricker wavelet sources having peak frequencies of 100 Hz and Proc Mtgs Acoust 2013;17: 070103.
200 Hz, respectively. A numerical comparison of the two proce- [22] Cagniard L. Reflection and refraction of progressive seismic waves. McGraw-
dures was shown to exhibit favourable agreement between these Hill; 1962.
[23] Aki K, Richards P. Quantitative seismology: theory and methods, vol. 1. W.H.
synthetic signals for a Ricker wavelet in a particular lossy- Freeman; 1980.
bottom, shallow-water Pekeris waveguide. [24] Chapman C. Ray theory and its extensions: WKBJ and Maslov seismograms. J
Future work will address the effects of a variable sound speed Geophys 1985;58:27–43.
[25] Brown M. A Maslov-Chapman wavefield representation for wide-angle one-
profile in the water column as well as frequency-dependent eigen- way propagation. Geophys J Int 1994;116:513–26.
ray amplitudes. The former will require the use of a more capable [26] Brekhovskikh L. Waves in layered media. New York: Academic Press; 1960.
ray-based computer code e.g. [3,30,37,38]. For the latter we intend [27] Tolstoy I, Clay C. Ocean acoustics—theory and experiment in underwater
sound. McGraw-Hill; 1966.
to investigate sub-band processing (e.g. third-octave bands), where [28] Tindle C, Bold G. Improved ray calculations in shallow water. J Acoust Soc Am
the eigenray amplitudes in each sub-band can be considered to be 1981;70:813–9.
approximately independent of frequency. Such frequency depen- [29] Tindle C. Ray calculations with beam displacement. J Acoust Soc Am
1983;73:1581–6.
dence can arise, for example, for a multilayered subbottom
[30] Hovem J, Dong H, Li X. A forward model for geoacoustic inversion based on ray
structure. tracing and plane-wave reflection coefficients. In: Caiti A, Chapman N,
Hermand J-P, Jesus S, editors. Acoustic sensing techniques for the shallow
water environment. The Netherlands: Springer; 2006. p. 241–55.
Acknowledgments
[31] Chapman C. Fundamentals of seismic wave propagation. Cambridge:
Cambridge Univ. Press; 2004.
The authors gratefully acknowledge the support of Dawn Vail- [32] Whittaker E. On Hamilton’s principal function in quantum mechanics. Proc R
lancourt, J. Hamilton, P.H. Hulton, and T.B. Carroll for funding this Soc Edin A 1941;61:1–19.
[33] Ainslie M. Caustics and beam displacements due to the reflection of spherical
research. Also thanks to Dr. Joseph Fayton for his careful review waves from a layered half-space. J Acoust Soc Am 1994;96:2506–15.
and to J.M. Tattersall (formerly of NUWC Division Newport) for [34] Frisk G. Ocean and seabed acoustics—a theory of wave propagation. Prentice
introducing us to the bandlimited impulse response method. Hall; 1994.
[35] Weston D. Wave shifts, beam shifts, and their role in modal and adiabatic
propagation. J Acoust Soc Am 1994;96:406–16.
References [36] Zhang Z, Tindle C. Complex effective depth of the ocean bottom. J Acoust Soc
Am 1993;93:205–13.
[1] Porter M, Tolstoy A. The matched field processing benchmark problems. J [37] Porter M. The BELLHOP Manual and User’s Guide, Available online at http://oalib.
Comput Acoust 1994;2:161–85. hlsresearch.com/Rays/HLS-2010-1.pdf, last accessed on May 18, 2018; 2011.
[2] Tolstoy A, Chapman N, Brooke G. Benchmarking for geoacoustic inversion in [38] Rodríguez O, Collis J, Simpson H, Ey E, Schneiderwind J, Felisberto P. Seismo-
shallow water. J Comput Acoust 1998;6:1–28. acoustic ray model benchmarking against experimental tank data. J Acoust Soc
[3] Weinberg H, Keenan R. Gaussian ray bundles for modeling high-frequency loss Am 2012;132:709–17.
under shallow-water conditions. J Acoust Soc Am 1996;100:1421–31.

You might also like