While time-domain analysis and modeling have been the approach for studying low-frequency seismic wave propagation in geophysics, underwater acousticians have traditionally favored a spectral analysis technique because only the mean signal energy seemed to have a predictable behavior at sonar frequencies of interest. However, now we can observe a trend in sonar development towards lower frequency, which should lead to both higher signal stability and better predictability because low frequency signals are less affected by bottom attenuation, by rough interface scattering and by scattering from small scale inclusions. Thus, powerful time-series analysis is becoming a valuable tool for studying complex propagation situations in the ocean. Within the framework of linear acoustics there are fundamentally two approaches to this modeling problem. The first one is to solve the pulse propagation problem via the frequency domain by Fourier synthesis of CW (continuous wave) results:
(1)
where is the source spectrum and is the spatial transfer function. This approach is attractive since it
requires little programming effort.
Once one has developed a model for the solution of the CW problem, one
can generate the transfer function at a number of discrete frequencies within
the frequency band of interest. The
evaluation of the integral in Eq. (1) can be then done by an FFT at each
spatial position r for which the
pulse response is desired.
Alternatively, one can solve the problem directly in the time domain,
which, however, requires the development of a new set of propagation codes and
does not allow the incorporation of developments achieved earlier in underwater acoustics. In my dissertation I follow the Fourier
synthesis approach. The strength of
this method lies in its simplicity and versatility, so we can use the new CW
model presented in Chapter 3 to produce a pulse response and can also compare
the results with any existing CW computational codes, described later in this
chapter.
Since the coefficients of the two differential operators
in the wave equation are independent of time, the dimension of the equation can
be reduced to three by use of the frequency-time Fourier transform pair:
(2)
, (3)
leading to the
frequency-domain wave equation, or
Helmholtz equation,
, (4)
where is the medium wavenumber at angular frequency
and is the source spectral function. It should be pointed out that although the
Helmholtz equation (4) is simpler to solve than the full wave equation, this
simplification is achieved at the cost of having to evaluate the inverse
Fourier transform, Eq. (2), if we have a process which differs from sinusoidal
in its time dependence. However, many
ocean acoustic applications in the past were of narrow-band nature. Thus, the Helmholtz equation has formed the
theoretical basis for the most important numerical methods, including the Ray, Wavenumber Integration (WI), PE,
and NM approaches.
Since acoustic
sources are local in nature, most of the time we deal with the homogeneous
Helmholtz equation:
. (5)
In spite of the relative
simplicity of Eq. (5), there is no universal solution technique available. The actual solution technique that can be
applied depends on the following factors:
- Dimensionality of the
problem.
- Sound speed variation, .
- Boundary conditions.
- Source-receiver geometry.
- Frequency and bandwidth.
For some problems the
environment is so complex that only direct methods of the discretization of the
differential operator in (5), such as the Finite-Difference Method (FDM) and Finite-Element
Method (FEM), are applicable. In
general an optimal approach is a hybrid of analytical and numerical methods,
and all the computational methods described in the following are of this
category. Although all these methods
have the Helmholtz equation as the starting point, they differ in the degree to
which analytical and numerical
components are utilized in the solution scheme. Since analytical methods
are restricted to a few canonical problems with simple environmental and source
parameters, the computational methods with a large analytical component are
also restricted to problems where the actual environment is well represented by
a simple idealized model. In the
following a review of the main computational methods for the solution of the Helmholtz equation and their mathematical bases are
presented.