Matched field and matched beam source localization using a bottom moored horizontal array

(1)

MATCHED

FIELD

AND

MATCHED

BEAM SOURCE

LOCALIZATION

USING

A

BOTTOM MOORED

HORIZONTAL

ARRAY

(2)

Matched Field and Matched Beam Source

Localization Using a Bottom Moored Horizontal

Array

Captain R. Matthew S. Barlee, MSM, CD B. Sc., Royal Roads Military College, 1991

A Thesis submitted in Partial Fulfillment of the Requirements for the Degree of

MASTER

OF

SCIENCE

in the

SCHOOL OF EARTH AND OCEAN SCIENCE

@ Captain R. Matthew S. Barlee, MSM, CD, November 28, 2003 UNIVERSITY OF VICTORIA

(3)

Abstract

Presented is an analysis of the performance of a matched field and matched beam processor in localization of a shallow water source using a 63-element bottom-moored planar array. Two scenarios are considered: in the first, a stationary, submerged CW source is localized in a high SNR environment; in the second, the source is being towed in the presence of multiple ship-generated directional noise sources. Array element localization is carried out using regularized linear inversion on transients from light bulb implosions around the array. The method provides the simplest array shape solution while still fitting the data and initial hydrophone position estimates to a statistically appropriate level. Estimation of geoacoustic parameters is accomplished using a hybrid inversion algorithm producing a robust geoacoustic model which is subsequently used to produce replica acoustic fields for the full field localization methods. Localization proved highly successful for the first scenario, while increased noise and geomet- rical limitations led to moderate performance in the second scenario.

(4)

. . .

3.1.1 Matched Field Processing

. . . 3.1.2 Matched Beam Processing

. . . 3.1.3 Computational Load . . . 3.2 Performance Factors . . . 3.2.1 Array Geometry . . . 3.2.2 Mismatch

4 RDS-2: Rapidly Deployable Systems I1

. . . 4.1 Introduction

. . . 4.2 Site Environmentals

. . . 4.2.1 Sound Speed Profiles

. . . 4.2.2 Bathymetry

(5)

4.2.3 Initial Geoacoustic Parameter Estimates . . . 39

5 Array Element Localization 43

5.1 Introduction . . . 43 . . .

5.2 AEL Inversion Algorithm 45

. . . 5.3 AEL Procedure 48 . . . 5.4 Results 51 . . . 5.4.1 InversionSolution 51 5.4.2 Uncertainty Estimates . . . 53 6 Geoacoustic Inversion 57 . . . 6.1 Introduction 57 . . . 6.2 Downhill Simplex 59 . . . 6.3 Simulated Annealing 61 . . . 6.4 Adaptive Simplex Simulated Annealing 62

. . .

6.5 The RDS-2 Geoacoustic Model 64

7 Localization: MFP and MBP Results 70

. . .

7.1 Introduction 70 . . . 7.2 High SNR: Serial PD1 72

. . .

7.3 Low SNR: Serial MF3 82

. . .

7.4 SNR Assessment 88 8 Conclusion 92 . . . 8.1 Major Findings 92 . . . 8.2 Future Work 94

Appendix A AEL Inversion Algorithm 102

. . .

A.l Inverse Theory 102

. . .

(6)

List

of

Tables

4.1 Sound velocity sampling details . . . 35

. . .

4.2 RDS-2 geoacoustic model parameters 42

. . .

6.1 ASSA annealing schedule 65

. . .

6.2 ASSA geoacoustic model comparison 68

7.1 PD1 localization frequencies . . . 74 7.2 MF3 serial localization frequencies . . . 83

. . .

(7)

List of Figures

. . .

2.1 Simple Pekeris waveguide 9

. . .

2.2 80 Hz mode structure 14

3.1 Conventional Beamforming bias of ULITE array . . . 19 . . .

3.2 Full field processing schematic 21

3.3 Effective longitudinal and transverse array aperture

. . .

28 . . .

4.1 RDS-2 site location map 31

. . .

4.2 RDS-2 experimental setup 32

4.3 Sound velocity plots demonstrating stable velocity structure . . . 36

. . .

4.4 Sound velocity profile differences 37

. . .

4.5 Subsurface currents at RDS-2 site 38

. . .

4.6 RDS-2 site bathymetry 40

. . .

5.1 ULITE array hydrophone positions 49

. . .

5.2 Time series of AEL light bulb implosion 50

. . .

5.3 ULITE array structure plot 52

. . . 5.4 Absolute AEL positional uncertainties 54

. . .

5.5 Relative AEL positional uncertainties 55 5.6 Comparison of UVic, SPAWAR, and DSTO AEL solutions . . . 56

. . .

6.1 Geoacoustic model parameter diagram 58

. . .

6.2 Geometric steps in the downhill simplex 60 6.3 ASSA inversion plots for geoacoustic parameter estimation

. . .

66

. . . 6.4 Sensitivity plots for ASSA inversion 67

. . .

(8)

vii

7.1 Sedimentary layer waveguide phenomenon

. . .

71

7.2 Transmission loss curve for 80 and 195 Hz

. . .

72

7.3 Mode structure for 80 and 195 Hz . . . 73

7.4 PD1 source-receiver geometry . . . 74

7.5 PD1 frequency spectrum . . . 75

7.6 CBF output for PD1 . . . 76

7.7 MFP ambiguity surface for PD1 . . . 77

7.8 R vs Z ambiguity surface for PD1 MFP . . . 79

7.9 MBP ambiguity surface for PD1 . . . 80

7.10 R vs Z ambiguity surface for PD1 MBP

. . .

81

7.11 MF3 source-receiver geometry . . . 83

7.12 MF3 frequency spectrum . . . 84

7.13 CBF output for MF3 . . . 85

7.14 MFP ambiguity surface for MF3 . . . 86

7.15

R

vs

Z

ambiguity plot for MF3 MFP . . . 87

7.16 MBP ambiguity surface for MF3 . . . 89

7.17

R

vs Z ambiguity surface for MF3 MBP

. . .

90

(9)

Chapter

1 Introduction

My initial experience with underwater acoustics began in 1991 in a role decidedly different than conducting research into underwater acoustic processing methods. As an Acoustic Systems Operator aboard an Aurora maritime patrol aircraft I was the end user, employing equipment and techniques developed by research teams during the post-war era to locate and track submarines. With the end of the cold war, the game of Anti-Submarine Warfare (ASW) was changing significantly. The demise of the Soviet state resulted in the decommissioning and scrapping of the majority of their costly nuclear submarine fleet. At the same time, economic hardship forced the sale of many Soviet diesel-electric submarines to Iran, India, and various other ambitious countries. The combined result was a major shift in how and where ASW was to be conducted. The ASW focus had changed from open-ocean, 'blue water' nuclear subs capable of delivering ballistic missiles, to diesel-electric subs operating in littoral areas and the threat they pose to military and commercial shipping traffic. Compared to the open ocean, the littoral environment provides a much more complex acoustic environment, requiring advanced acoustic processing techniques to accurately localize underwater sound sources. Unfortunately, these techniques have only been applied experimentally, requiring the ASW community to continue relying on traditional blue water array processing techniques which degrade rapidly in shallow water environments, giving a decisive advantage to the submarine. This weakness provides the prime motivation behind this body of work. I hope that with this work, and further investigation into the performance of advanced array processing techniques using experimental data, future ASW will be conducted on a more level playing field.

(10)

Chapter 1: Introduction 2

1.1 Background

Since it's first recorded use in 1827 by the Swiss physicist Daniel Colladon to measure the sound velocity in Lake Geneva, underwater sound transmission has continued to be the primary tool in remote sensing of the world's oceans. The German submarine threat of WWI brought about the first military application of passive acoustic processing with the development of the 'eel,' a simple twelve-element hydrophone array that, when towed behind allied ships, allowed the operator to determine the bearing of a noisy submarine with an accuracy of 0.5" [Urick, 19751. From the advent of the 'eel,' military application has been the driving force behind the advancement of underwater acoustic processing, and until the late 19801s, that force was primarily directed toward deep water applications.

The most common method used in deep water acoustic localization is referred to as conventional beamforming (CBF). Based on the assumptions that: 1) sound received from the source is assumed t o have travelled far enough such that wavefront curvature is negligible (plane wave assumption) and; 2) that these waves are received at shallow arrival angles, CBF is a method in which the received acoustic field at each hydrophone is spatially filtered with respect to bearing (beamforming) and the amplitude of the resultant outputs, or beams, are correlated. The output with the highest amplitude then provides the source bearing estimate. In most open-ocean environments the above assumptions are valid and CBF is a robust and efficient method.' Conversely, in shallow water environments neither assumption holds and CBF can be significantly degraded. When considering multipath propagation of acoustic signals in a shallow, bounded waveguide, acoustic receptions can have both steep arrival angles and significant wavefront curvature. As arrival angle and wavefront curvature increase, the assumption-based CBF degrades, producing CBF bearing error, or bias. lit should be noted that CBF alone does not provide target localization. In practice, accurate target fixes require a minimum of two bearing lines separated by a minimum of 30•‹, and even then, the fix provides no depth information.

(11)

Chapter 1 : Introduction 3

How then, can this breakdown in CBF be overcome?

In 1976, Homer Bucker proposed an answer. Instead of ignoring the complex variability of the acoustic field as in CBF, he employed a method to exploit it. By linearly correlating the Fourier-transformed acoustic field received at the array with similarly transformed 'replica' fields produced by a forward propagation model for selected candidate source p o ~ i t i o n s , ~ Bucker estimated target location from the replica field that most closely matched the recorded field. He referred to the method as Matched Field Processing (MFP) and it provided the first truly passive means of 3-D target localization [Bucker, 19761. Many variations of MFP have since been pursued, some using different processors to correlate the fields, while others correlate derivations of the acoustic field such as modes or beams. A thorough review of MFP methods is presented by Baggeroer et al. [1999]. Matched Beam processing (MBP) is a variation in which MFP is conducted in beam space. In MBP, array data is first beamformed, but unlike CBF which correlates only beam amplitudes, MBP then correlates the complex beam data (both phase and amplitude information) with replica beams generated for candidate source positions.

In a theoretical experiment where no noise exists and the waveguide is perfectly modelled, both MFP and MBP would provide a correlation of one at the true source position. Inclusion of environmental noise and errors in waveguide modelling reduce the strength of the correlation and bring about the emergence of secondary correlation maxima, or sidelobes at false target positions. Reducing these negative impacts on processor performance is the crux of employing MFP and MBP in experimental application. Numerous simulation studies have been conducted to assess the various types of processors and their sensitivity to noise and modelling errors, but very few experimental applications have been conducted for horizontal bottom-moored arrays [D'Spain et al., 1999, Ozard, 19891.

2The propagation model used to produce replica fields uses a geoacoustic model that most closely approximates the experimental environment.

(12)

Chapter 1: Introduction 4

1.2 Purpose and Scope

The purpose of this thesis is two fold. First, it is hoped that this work will serve as a template for applying full field processing methods to experimental data from a bottom-moored horizontal array. The work is presented in a progressive order, outlining the necessary components or steps required to conduct source localization using MFP and MBP. Second, having carried out these steps, I then will assess and compare the performance of CBF, MFP, and MBP in localizing an underwater target in a high and low signal to noise environment. Data used in this thesis work was collected by a horizontal bottom moored planar array (ULITE) during the RDS-2 experiment conducted in the shallow waters of the Timor Sea, north of Australia.

Chapter 2 presents the fundamental acoustic theory upon which this thesis is based. The development of the normal mode solution to the linear wave equation is detailed for a simple, range-independent environment and the physical representation of this solution is discussed. ORCA, the acousto-elastic propagation model used for propagation modelling in this work, is then discussed in terms of a general derivation emphasizing the significant features of the model.

Chapter 3 examines the acoustic data processing methods evaluated in this research. The method of CBF is discussed and its limitations in a shallow water environment are presented through the results of a simulation study on CBF bias sensitivity. A schematic detailing the basic steps involved in full field processing (applicable to both MFP and MBP) is followed by detailed descriptions of the different correlation algorithms employed in MFP and MBP, and the computational loads associated with each algorithm. Next, the effect of array geometry on processor performance is examined and the chapter closes with a discussion of full field processing's most critical consideration - mismatch.

(13)

Chapter 1: Introduction 5

description of the experimental setup and the equipment involved, a przori environmental parameter estimates for the site are presented. Bathymetry, sound velocity, and bottom composition are each addressed and a composite geoacoustic model is provided as an initial model for the geoacoustic inversion described in Chapter 5.

Chapter 5 describes one of the major steps required for effective source localization, array element localization (AEL). Because the hydrophone array used in the RDS-2 work was surface deployed and lowered to the bottom, the position of each hydrophone was known with little accuracy. Because full field processing methods are extremely sensitive to sensor position errors, a method to accurately locate the hydrophones is required. The method to accomplish this, AEL, involves inverting the arrival times of array-recorded nearby transient sound sources to determine individual hydrophone positions in the array. The method is discussed first in terms of the algorithm used for the inversion, and then in terms of the procedure carried out to generate the inversion data. Results of the inversion are presented in the form of an overall array solution accompanied by individual uncertainty estimates for each hydrophone.

Chapter 6 deals with the second critical component in target localization using full field processing-determination of an effective geoacoustic model. The problem of inverting acoustic data to determine geoacoustic parameters is addressed and the primary methods used to solve the problem are discussed. Following an explanation of the downhill simplex, a local inversion search method, and of the global search method, simulated annealing, Adaptive Simplex Simulated Annealing (ASSA) is presented as a hybrid of these local and global methods, incorporating the desirable features of each. Using acoustic data recorded during two independent serials of the RDS-2 experiment, ASSA is used to determine an effective geoacoustic model for subsequent use in MFP and MBP source localization. Quality of the inversion, individual parameter sensitivities, and agreement between the geoacoustic models generated for two independent inversions are all addressed in the determination of the final geoacoustic model.

(14)

Chapter 1: Introduction 6

Chapter 7 presents the findings from the final, and most important stage of this research-source localization. After a discussion of propagation characteristics specific to the RDS-2 waveguide, results of CBF, MFP, and MBP are presented for two individual acoustic serials. Capabilities and limitations of all three methods are examined first in the localization of a stationary target in a high SNR environment, and then in the localization of a slowly moving target with a much lower SNR. For each serial, characteristics of the source signal (frequency, SNR, signal stability) are presented along with the CBF output and an assessment of CBF bearing bias. Out- puts, or ambiguity functions, generated by both MFP and MBP are then examined for the two serials. Both qualitative and quantitative descriptions of the correlations are provided as a means of measuring the quality of a given localization against not only that of the other processors, but against the other serial as well. Suspected factors behind processor degradation are then discussed, including an examination of the impact of increased SNR on source localization.

Finally, Chapter 8 summarizes the steps required for effective source localization using full field processing methods and recaps the major findings of this research. Suggestions for future work are provided in hopes that some of the many remaining unanswered questions will soon be addressed.

(15)

Chapter

2 Background Acoustic Theory

2.1 Acoustic Wave T h e o r y

A common starting point in the development of underwater acoustic models is the linear wave equation. As detailed in Jensen et al. [1994], the wave equation is derived from linear approximations1 of the hydrodynamic equations for the conservation of mass, Euler's equation, and the adiabatic equation of state, producing

where p is the fluid density, c the sound velocity in the fluid, and P, the acoustic pressure, is a function of spatial location r and time t. Assuming constant density in space reduces (2.1) to the standard wave equation in the absence of sources,

By the superposition principle, harmonic waves of the solution to (2.2). By applying the Fourier transform pair,

(2.2) form f (w)eciWt provide a

to (2.2), the frequency domain wave equation, or Helmholtz equation (HE), is obtained,

[V2 -t- k2 (r)] p(r

,

w) = 0,

'In the linear approximation only first order terms of the equations of state are retained, as second order terms are many orders of magnitude less and as such can be ignored.

(16)

Chapter 2: Background Acoustic Theory 8

where k(r) is the wavenumber at frequency w and spatial location r,

As detailed by Frisk [1994], acoustic sources, or forcing terms f (r, t), are commonly assumed as point sources, mathematically represented by the Dirac delta function, S(r - r,). Kinsler et al. [I9821 show that a forcing term is introduced by simply adding it t o the right hand side of the wave equation (2.2). Assuming harmonic time-dependence of both the acoustic field and the forcing mechanism, (2.5) can be modified to produce the inhomogeneous Helmholtz equation,

2.2 N o r m a l M o d e T h e o r y

The method of normal modes, which is used to calculate the acoustic fields studied in the body of this thesis, is employed extensively in underwater acoustic modelling. Its solution to the Helmholtz equation provides a good approximation to the acoustic pressure field in many underwater environments. Originally introduced in underwater acoustics by Pekeris in 1948, the method solves an unforced version of the Helmholtz equation with a summed set of modes of vibration which, in a range independent environment, fully describe the acoustic field. The set is comprised of discrete modes that propagate horizontally away from the source, and continuous or leaky modes that exponentially decay with range. Because the modal contributions from continuous modes are minimal in the RDS-2 experiment, they are ignored and acoustic field is approximated by the set of dominant discrete modes.

To describe modal propagation in a simple acoustic waveguide, it is instructive to first develop a general solution to the Helmholtz equation for propagation in a

(17)

Figure 2.1 Range independent environment used in development of normal mode theory. homogenous, range independent medium with perfectly reflective or pressure release boundaries (Fig. 2.1).

In this simple waveguide it is convenient to adopt a cylindrical coordinate system

r = (r, cp, z), for which the Laplacian operator

,

I d d I d 2 d2 ~2 = --r- + -- + -

r d r d r r2 dcp2 8z2'

when applied to (2.7), produces the Helmholtz equation in cylindrical coordinates,

where cp is the azimuthal angle and z, the source depth. By assuming cylindrical symmetry of the field about the source at ro = (O,O, ss), (2.9) will have no dependence on cp and can be reduced to

By employing separation of variables to the range independent, unforced Helmholtz equation (thus the right hand side of (2.10) equals zero), a solution is sought of the form,

p(r7 2) = @(r)Q(z). (2.11)

(18)

To separate the unforced HE into range and depth independent terms, first sub- stitute (2.11) into (2.10), and then divide through by @(r)Q(z), producing,

Because the first and second terms of (2.12) are independent, the equation will hold only if both terms equal a separation constant, denoted k:m. As such, (2.12) can be divided into range and depth independent solutions, respectively:

Equation (2.14), the modal equation, is in the form of a standard Sturm-Liouville equation with an infinite number of characteristic eigenfunction solutions, @,(z), satisfied by a discrete set of eigenvalues, k:m for each mode m. These orthogonal eigenfunctions, which represent the natural way in which the system vibrates, are referred to as normal modes and are analogous to the modes of a vibrating string. The eigenvalues are the horizontal components of the wave number and are analogous to the frequency of vibration. A complete discussion of the properties of the Sturm- Liouville problem is detailed by Frisk [1994]. Because these modes form a complete set, the pressure field can be represented as a summation of them,

(19)

Eq. (2.16) can be simplified by applying the modal equation (2.14) to the above square bracketed term, yielding

To solve for the range-dependent equation, the integral operator

~ , h

s d z is applied to (2.17), and due to the orthogonality of the eigenfunctions, when m =

n, (2.17) reduces to the Bessel equation

the depth independent solution of which is

where ~ : ' ~ ) ( k ~ , r ) are Oth order Hankel functions of the first and second kind,

Eq.s (2.20) and (2.21) represent cylindrical waves that asymptotically diverge and converge with range for krmr

>>

1 .

Substitution of the range dependent (2.19) into the equation for pressure (2.15), yields

In accordance with the Sommerfeld radiation condition stating that acoustic energy radiates outward as r t m, a Hankel function of the first kind ( H i ) is chosen,

as its positive sign convention, combined with the time dependence (eViwt) produces a cylindrically spreading radial wave that decays as r-* for far field approximations

(20)

(kTmr

>> 1). Correspondingly, substitution of (2.20) into (2.22) produces the equa-

tion for the acoustic pressure field in terms of the normal modes of the waveguide,

Through use of the simple waveguide described at the beginning of this section, specific boundary conditions are imposed. Specifically, a pressure-release surface dic- tat es

P ( ~ ) ~ Z = O = 0, (2.24) and a perfectly rigid bottom at z = h requires the normal component of particle velocity be zero,

Application of these boundary conditions and subsequent separation of variables reduce the modal equation (2.14) to

where the vertical wavenumber is,

When a trial solution of the form

is substituted into (2.26) and the boundary conditions are applied, it can be shown that Bm = 0 (from (2.24)), and A,,&, cos(k,,h) = 0 (from (2.24)). For a non-trivial solution, it can be seen cos(k,,h) = 0, or

(21)

and the depth dependent eigenfunction becomes3

Incorporation of (2.30) into (2.23) provides a complete equation for the normal modes of the waveguide,

which reveals the existence of nodes for each mode at the surface and periodically at depths znm =

(3)

for n = (1,2,3 . . .) throughout the entire waveguide.

2.2.1 Physical Description of the Field

By expressing the mode function, sin(k,,z), in exponential form e'k'm';~-zkzm'

,

(2-31) can be rewritten

where,

k,, = k sin Om, kTm = k cos Om.

Examination of the exponential terms in (2.32) reveals that each normal mode is both a standing wave in the vertical, and a travelling wave in the horizontal direction with grazing angle 0,. The positive and negative depth dependent exponentials represent waves travelling in the negative and positive z directions, respectively, and the interference pattern created from superposition of these plane waves creates standing waves in depth (modes), which are uniform in range. The solution for the radial part of the HE can be seen as an outwardly propagating wave of the form eibmr. Figure 2.2 depicts mode function amplitudes plotted as a function of depth for an 80 Hz signal in a waveguide modelling the RDS-2 site.

(22)

Chapter 2: Backgrozmd Acoustic Theory 14

2

3 4

Mode

Figure 2.2 Mode function amplitude vs. depth far an 80 Hz signal in the RDS-2 environment

2.3 Propagation Modelling

The prime factors influencing the choice of propagation model are the bathyrnetry of the experimental site, source-receiver ranges expected, and the computation time required to calculate the modelled acoustic field.

The RDS-2 test site was intentionally chosen for it's uniform bathymetry. With

seafloor slope varying less than one percent over the entire site, a propagation model is sought that behaves predictably and accurately in range-independent environments. An average water depth of 107 m and source-receiver ranges of two to four kilome- ters combine to produce multiple bottom interactions in this cylindrically spreading environment. As such, the ability of the model to represent acoustic propagation in multi-layered elastic ocean bottoms is a priority. Finally, because the propagation model would be called upon numerous times as part of an MFP algorithm, efficiency and speed of the model computations is also a desirable property.

The ORCA propagation model, designed by Westwood et al. [1996], satisfies all

of the above stated requirements. Not only has the model been successfully evaluated against numerous benchmark environments, but also with its use by this laboratory

(23)

in a number of acoustic modelling experiments, ORCA is a reliable and predictable acousto-elastic normal mode model.

2.3.1 ORCA General Derivation

To summarize the method used in ORCA for determining mode functions, an intermediate result of the normal mode derivation proposed by Bucker [I9761 is expressed in terms of up and down looking plane wave reflection coefficients, R1 and Ra These coefficients are obtained analytically and from their product, R1R2, eigenvalues, kn, are determined. Using these eigenvalues, the wave equation is solved for each layer, and from these solutions, compressional and shear wave mode functions are determined. As detailed in Westwood et al., ORCA begins with the Helmholtz equation in cylindrical coordinates (2.10), but instead of introducing the familiar solution p(r, z) = @,(r)q,(z) as in (2.11), ORCA introduces a solution of the form,

which allows for density variations, not accounted for in the previous ideal fluid normal mode derivation. Variation of parameters is then applied to express O(z) in terms of the functions U(z) and V(z), which satisfy the wave equation above and below the reference depth, G , re~pectively.~

Using the residue theorem, (2.34) is solved as a sum of normal modes,

where the ~ r o n s k i a n ,

4As the method treats the waveguide as a series of horizontally stratified layers, equation (2.34) is introduced as a solution for each. Thus, z, corresponds to the reference depth for each layer from which up and down looking reflection coefficients are determined.

(24)

Chapter 2: Background Acoustic Theory 16 Mode eigenvalues, k,, are subsequently determined as the horizontal wavenumbers, krm, for which the Wronskian equals zero. To do this, the depth functions, U(x) and V(x), are expressed in terms of up and downward looking reflection coefficients, R1 and R2:

where A and B are arbitrary constants and k,, is the vertical wavenumber where k: = (u2/c2) - k:.

R1 and R2 are analytically determined by sending unit-amplitude plane waves up and downwards from the reference depth, z,, and then computing their plane wave reflection coefficients at the upper and lower b~undaries.~ Substitution of (2.37) into the Wronskian, which for eigenvalue determination equals zero, results in the eigenvalue criterion,

Rl(kn)R2(kn) = 1. (2.39)

Thus, eigenvalues, Ic,, can be determined by searching the complex k-plane6 looking for points where (2.39) is satisfied. To simplify this search, the points, or eigenvalues, are sought along a contour where (R1R21 = 1. Eigenvalues are found at points along this contour where the phase of RlR2 is a multiple of 27r.

Once determined, modal eigenvalues are used to solve the respective wave equations for compressional (p) and shear (s) wave displacement potential,

$'(z)

+

[K;(Z) - k2] $ = 0, (2.41)

5For each layer at reference depth z,, half-spaces are assumed to exist at each boundary. 6To account for shear wave attenuation in bottom layers and transmission into and reflection off of the half space boundary, ORCA searches off the real axis to determine the complex eigenvalues associated with leaky modes that result from these boundary interactions.

(25)

where K p = w/cp and K , = w / c , are the p-wave and s-wave numbers for frequency w . From these, the associated vertical wavenumbers can be evaluated,

from which solutions for the mode functions can be determined. Using (2.42), the solution for (2.40) provides the compressional mode function in an isovelocity7 layer,

which, with the substitution of

P

for y, is also the solution for the s-wave mode functions.

Because U ( k n , z) and V ( k n , z) together form the un-normalized mode function, inclusion of these terms into (2.35) estimates the acoustic pressure field as a summation of all propagating modes in the waveguide. As earlier noted, this approximation excludes continuous modes, but because of the long ranges associated with this experiment, the contribution of these modes, which decay with range, is negligible.

7For brevity, the isovelocity case is presented. ORCA is also capable of determining mode functions in layers with

$

linear sound speed gradients. A description of such cases is detailed in [Westwood et al., 19961.

(26)

Chapter

3 Acoustic Data Processing

3.1 Full Field Processing

Hydrophone array processing began in the post World War I1 era with an increased interest in passive acoustics vice traditional active systems. With the advent of solid- state digital components, Anderson [I9601 produced the digital multibeam steering system (DIMSUS) in 1960, ushering in the use of beamforming in modern underwater acoustics. Since that time, CBF has been the primary means of processing acoustic array data for practical purposes.

Plane-wave beamforming is a one or two dimensional spatial filtering process in which weighting vectors of the array elements are correlated with array output providing an indication of target bearing. Because this method assumes plane-wave arrivals, it is most effective in deep water, long range applications and degrades in shallow, multi-path environments. In these environments the acoustic signal arrives from many different propagation angles and is split into multiple beams. Because boundary reflected arrivals can achieve much higher grazing angles than those of direct path arrivals from the source, acoustic energy associated with these angles will introduce errors into bearing estimation. Figure 3.1 depicts absolute bearing errors, or bias, as functions of source frequency, range, and depth for a simulation using the ULITE array in a waveguide representative of the RDS-2 site. Analysis of the plot provides the following general observations:

At short ranges where multi-path acoustic arrivals are still present, bearing bias can be very large.

(27)

Chapter 3: Acoustic Data Processing 19

Varying freq for fixed range (3 krn) and depth (50 m)

20 I I I I I

+

25 Hz 15

-

t 50 HZ - f 100Hz 10 - ir200 _{Hz $} h V)

A Varying range for fixed frequency ('I 00 Hz) and depth (50

m)

7

.

I I m

/"

I - c l k r n l 2

-

- 0

p

--ro * *

-

10 20 30 40 90

Varying depth for fixed range (3 km) and freauencv (1 00 Hzl

Source Bearing (degrees)

Figure 3.1 Absolute CBF bias of the ULITE array as a function of source bearing for various source frequencies, ranges, and depths. (Bias values of 20" indicate a

bim

2

209

The desire to overcome this weakness of CBF in such complex acoustic environments provided the impetus to develop the methods of full field processing which are employed in this thesis.

FulI field processing (FFP) techniques, such m matched field processing and matched beam processing, are three dimensional generalizations of plane-wave beam- formers that exploit the complex structure of the acoustic field, instead of being re- stricted by it, as in CBF. Full field methods take advantage of the coherence of modal and multipath acoustic field structure in environments where traditional methods

(28)

Chapter 3: Acoustic Data Processing 20

have faltered. By replacing the weighting vectors used in CBF with solutions to the wave equation, full field processing not only corrects aforementioned bearing bias, but can provide estimates of target range and depth by matching the acoustic field properties measured at the array with replica fields generated by a propagation model for possible source positions.

The steps involved in F F P source localization are best explained by the schematic illustrated in Fig. 3.2. A sound source is located at an unknown range, depth, and bearing in an acoustic waveguide in which hydro- and geoacoustic properties are reasonably estimated. This environmental information is included in a propagation model (ORCA, in this work) along with geometric information describing the position of the hydrophone array. The model is run for various candidate source positions within the waveguide and a modelled or replica acoustic field is produced for each candidate position. Each of these replica fields is then iteratively matched, or correlated with the actual field recorded at the array via a processor specific to the method being used. Matching of the full acoustic field is done in MFP; whereas other full field processing techniques, such as MBP or matched mode processing, seek to match a component or a quantity derived from the acoustic field [Jensen et al., 19941. The correlations for all iterations over candidate source positions are then presented as an

Ambiguity Surface1 that visually depicts the correlation as a function of range and

depth for selected bearings. The highest correlation on the surface indicates the best estimate of target position.

lThe term 'ambiguity surface' stems from the fact that there are often numerous high correlations for erroneous source positions (sidelobes) that lead to ambiguous determination of true source position.

(29)

Chapter 3: Acoustic Data Processing 21 Source 4 Ambiguity Surface

I

Shows match between P and P(m)

Figure 3.2 Illustration of the basic steps involved in source lo~alization using f d field processing.

3.1.1 Matched Field Processing

MFP can be considered the most direct method of FFP as matching is conducted using the pressure fields. Aside from the steps required for discrete Fourier transformation to the frequency domain, the field data is unaltered prior to it's correlation.

First applied to underwater acoustics by Bucker in 1976, matched field processing did not progress past theoretical assessment for seven years due to its taxing computational load. With the development of numerical normal mode propagation models (reviewed in Buckingharn [1992]) computation times were significantly decreased and in 1983 the method was first applied to data at a Naval Ocean Research and Devel- opment Agency (NORDA) workshop [Heitmeyer et al., 1983, Baggeroer et al., 19991.

(30)

From that time, MFP has emerged as the primary method of full field processing and continues to be actively researched.

Of all processors available in MFP [Tolstoy, 1993, Baggeroer et al., 19991 the simplest and most commonly used for processing experimental data is the linear, or Bartlett processor. Although lacking the resolution of more advanced processors, the Bartlett processor was selected for this research due to it's robustness in the presence of mismatch, as is demonstrated in an early paper by McDonough [1972]. Because of the scarcity of ground truth to determine geoacoustic parameter estimates of the RDS environment, the potential exists for significant mismatch (described in 3.2.2) and as such a robust processor is highly desirable.

For the MFP conducted in this work, the time domain acoustic signal is Fourier transformed and the resultant narrow band data are then matched using the Bartlett processor. The single frequency processor output, commonly referred to as 'power', is a measure of the square of the magnitude of the correlation between the measured and modelled acoustic fields and, using notation proposed by Tolstoy, is given by

whose variables are described as follows:

F = ( F ~ ,

F2,

... F ~ ) T is the replica acoustic field where

- F~ is the modelled acoustic field at frequency w on the nth hydrophone,

- _{denotes the transpose of a vector,}

F+ is the conjugate transpose of F,

C

= (FF+) is the mean value of the cross-spectral matrix (CSM) where

- F = ( F l , F2, . . . F ~ J ) ~ is the measured acoustic field at frequency w with F,

(31)

Chapter 3: Acoustic Data Processing 23

F (similarly F) has been normalized such that

J I

F~

l2

+

. . .

+

I

FN

l2

= 1. The primary feature contributing to the robustness of the linear processor is the inclusion of the mean cross-spectral matrix,

C.

Due to the inherent variability of ocean acoustic fields, significant phase and amplitude fluctuations occur as a function of time, thus producing dramatically different results in, for example, localization of an unknown source. By considering the mean behavior of the CSM, the undesirable effects of this variability can be strongly suppressed [Tolstoy, 19931. By calculating the CSM for L statistically independent data sets2 the mean CSM n, mth entry can be estimated as follows:

Here the frequency contribution for each time segment I, of T seconds length, for hydrophone n is

K

-iwtk

Fn,i (w) =

C

fn,i (tk)w(tk)e 7 (3.3)

k=O

where tk = kAt, At =

5,

K is the number of samples in time T, and w(t) is a window function applied to reduce Gibbs phenomenon associated with time series truncation. For all acoustic data analyzed in this research, spectal estimation is conducted by applying a K = 2048 point Fourier transform "c a series of time segments of length T = 4 s which are overlapped by 50%. The sampling frequency of the acoustic recorder is 510.621 Hz resulting in a frequency resolution approximately 0.25 Hz. The window function, w(t), is a 2048 point Hamming (cosine) window

Because averaging of independent time segments reduces the variance of the CSM estimate, noise that is uncorrelated across the segments will be suppressed, thus effectively increasing the signal to noise ratio. Additionally, Tolstoy [I9931 states that

21t is assumed that the signal of interest is steady state and the only variability in the acoustic field is the noise.

(32)

the effect of transitory phase disturbances that occur coherently across the array3 is minimized through the use of the mean CSM estimate.

3.1.2 Matched Beam Processing

Effectively a synthesis of CBF and MFP, matched beam processing is most eas- ily described as MFP carried out in the beam domain. A relatively new method,

MBP gained exposure in 1995 when Yang and Yates used it to invert for seafloor parameters in a shallow water environment. Subsequently, MBP has been applied to a number of simulations, including source localization using vertical and horizontal arrays [Yang and Yates, 1998, Yang et al., 1998, Yang and Yoo, 19981, but to this author's knowledge, no work has yet been published on the application of MBP to experimental data.

Like MFP, the 'power' or output of the matched beam processor is a measure of correlation, but instead of correlating the full pressure fields, MBP beamforms the field data prior to correlation. The complex data and replica beam outputs are calculated by multiplying the measured and replica fields by a steering angle, 0, and

summing the products over N hydrophones in the horizontal array,

where (xj, yj) is the coordinate of the j t h phone, k = is the wave number, and s denotes the candidate source position. Correlation of the above data and replica

beams is then carried out in the beam domain using a linear processor [Yang and

3Coherent phase disturbances are phase changes that occur equally at each hydrophone of the array. An example of this is the bobbing of a surface suspended vertical array.

(33)

Yates, 19981,

where O1 and O2 define the angle cutoffs for beam filtering and w(O) is the beam

weighting vector.

Like MFP, MBP corrects bearing bias, but it also offers some practical advantages that can potentially improve source localization. Because MBP correlates CBF outputs, effective signal processing methods can be applied to the beam data prior to correlation to increase processor fidelity. For example, in the case of an interfering coherent noise source, adaptive beamforming techniques can be applied to steer a null in the direction of the interfering source before correlation. Additionally, because MBP is a linear transformation, beams containing directional noise can simply be excluded from the correlation [Yang and Yates, 19981.

To increase the statistical reliability of source position estimates in this research, the ambiguity surfaces produced by both MFP and MBP are incoherently averaged over frequency. The normalized processor outputs (correlations) range in value from 0-1, with 1 being a perfect correlation.

3.1.3 Computational Load

Although both MFP and MBP offer distinct advantages over CBF, they come at the cost of a significantly increased processing load. A comparison of the relative processing load between MFP and MBP is presented for an array of N hydrophones. For source localization, each processor requires an ORCA call to determine wavenumber values. Then, for each candidate position, both processors require geometric calculations to determine the distance between the source and each hydrophone.

(34)

employing this inner product formulation of the Bartlett processor, 2N complex cal-

culations4 are required to conduct a correlation for each of the L time segments,

totalling 2LN complex calculations per candidate position, or Q M F P = 2LN. Di-

rect implementation of (3.1) would require 4N2 calculations, which is a much greater

number of calculations when N

>>

L.

Because MBP requires the extra step of beamforming, computational time is greater still. N complex calculations are required to form each one of the Nb beams

giving NNb calculations to convert to the beam domain. Correlation then demands 2Nb complex calculations per time slice, resulting in 2LNb correlation calculations.

Totalling the two steps, we have QnlBP = Nb(N

+

2L).

Sample values for the above variables demonstrate the load increase associated with MBP. An average of L = 5 slices from N = 63 phones are processed and with a beamwidth of 2", Nb = 180. Correspondingly, QhfFP = 630 complex MFP cal- culations are required per candidate position, compared to QMBP = 13140 MBP calculations for the same one candidate position, a twenty fold increase. Obviously

Q M B p can be significantly reduced by increasing beamwidth (but at the cost of res-

olution) and through the use of a priori information to limit the number of beams correlated, but even with such measures, the load is significantly higher.

For this comparison, the total number of computations has been examined, and although much of the data is vectorized in practice to reduce this total, MFP and MBP remain computationally intensive. If numerous I/O calls to ORCA are required, as in the case of inversion algorithms, processing time is further increased.

To practically demonstrate the computational times associated with running both MFP and MBP, the following are actual times associated with localization runs using RDS-2 data on a UNIX based P2 desktop computer. For a 4 frequency MFP localization searching 90' - 110' at

i0

bearing intervals, 1-5 km at 50 m range intervals, and 6-94 m at 8 m depth intervals, computation time is 183 s. Using the same frequencies,

(35)

MBP localization is conducted using 720

i0

beams requiring 2955 s processing time. If the number of

i0

beams is reduced by searching only between 80' - 120•‹, the time is reduced to 483 s.

3.2 Performance Factors 3.2.1 Array Geometry

Source-receiver geometry has a significant impact on the quality of the correlation and can be demonstrated by examining a linear horizontal array. Range, depth, and bearing discrimination are each dependent upon an effective aperture which determines to what extent the vertical and horizontal distributions of acoustic energy are sampled by the array [Bogart and Yang, 19941. In particular, bearing discrimination for the array depicted in Fig. 3.3 is dependent upon the transverse aperture, the horizontal projection of the array length transverse to the target direction. If the source is broadside (0") to the array, the transverse aperture is maximized, equalling the array length, and thus providing the best bearing estimate; whereas a source at endfire (90") minimizes the transverse aperture giving the worst bearing discrimination. Converse to bearing, range and depth discrimination are dependent on the array's longitudinal aperture, the horizontal projection of the array in the target direction. Therefore, range and depth estimates are most accurate at endfire where the vertical extent of the field 'seen' by the array is greatest. Thus, for a linear array, increased discrimination in bearing comes at the cost of range and depth discrimination, and vice versa.

To overcome this weakness associated with linear arrays, two arms of the ULITE array (shown in Chapter 4) are used in this work, providing a V-shaped array that has sufEicient transverse and longitudinal apertures for all source positions. Additionally, the planar geometry of the array removes the left-right ambiguity normally associated

(36)

Longitudinal

Aperture

Source

IPt

# 0 0 0 /

Array

Transverse

Aperture

Figure 3.3 Illustration depicting the transverse and longitudinal aperture of a linear array with linear arrays.

Array depth has an impact on the quality of fieldlbeam correlation. Because modal structure is fixed with depth, the performance of a horizontal array is dependent upon the depth at which it is located. If the array is at a depth corresponding to a zero crossing of a particular mode function, no acoustic information from that function will contribute to the pressure field, resulting in degradation of processor performance [Tantum and Nolte, 20001. For example, of the six modes associated with the 80 Hz signal in Fig. 2.2, modes two and six show zero crossings at or near 107 m, the array depth. Consequently, only four of the six modes will contribute to the field, and if the source-receiver range is large enough, the higher order modes will be stripped away by multiple boundary interactions, further reducing modal contributions and thus processor fidelity.

(37)

Chapter 3: Acoustic Data Processing 29

3.2.2 Mismatch

The most influential phenomenon in increasing the sidelobe levels of the ambiguity surface is termed mismatch. At best, mismatch introduces minor errors in the source position estimation, and at worst, it causes total processor failure. Mismatch describes the condition in which undesirable errors are introduced into the FFP method resulting in degradation of processor performance.

The most common type of mismatch is termed environmental mismatch and is a result of inaccuracies in modelling the acoustic waveguide. Because the ocean is not a static environment and because every measurement has inevitable uncertainties, environmental mismatch will always be present. To effectively reduce this prevalent type of mismatch, a knowledge of which parasmetric errors most profoundly affect performance is valuable:

Sound velocity profile (SVP) mismatch. Numerous studies [Bucker, 1976, Ham- son and Heitmeyer, 1989, Gingras, 19891 evaluating the impact of errors in modelling the sound velocity structure in both the water column and bottom layers agree that accurate sound speed determination is critical. Processor performance appears to be the most sensitive to this type of mismatch, and although the short ranges used in this research minimize range-dependent SVP changes, time dependent profile changes are still of concern.

Water depth mismatch. Delbalzo et al. [I9881 examined the effects of erroneous water depth modelling and report that minor depth errors are reflected by corresponding errors in source range (shallow estimates result in shorter source ranges, while deeper estimates produce erroneously long range estimates). How- ever, if depth estimate errors are greater, such that the model calculates an erroneous number of waterborne modes, random errors in source localization result [Tolstoy, 19931.

(38)

Geoacoustic parameter mismatch. This broad category of environmental mismatch includes errors in estimation of sedimentary and basement properties and is generally the most difficult to correct due to difficulties in obtaining ground truth data. Although mismatch is fairly insensitive to geoacoustic parameters such as sediment density and attenuation, its considerable sensitivity to the thickness and sound velocity of sedimentary layers requires care in modelling the ocean bottom [Shang and Wang, 1991, Klemm, 19811. This is most important in shallow water environments such as the Timor Sea, where higher order modes have strong interactions with the seafloor. Because a significant portion of acoustic energy is borne by the bottom in this case, incorrect estimation of parameters affecting this transmission has detrimental effects on acoustic field representation at the receiver [Porter et al., 19871.

Ranging from minor errors introduced by equipment generated electronic noise to significant errors due to incorrect sensor location estimation, system mismatch is used to describe errors associated with the receiver system. In this experiment the prime source of system error stems from uncertainties in array element location, as discussed in Chapter 6.

(39)

Chapter

4 RDS-2: Rapidly Deployable Systems

I1

4.1 Introduction

From 1-11 November, 1998, a multi-national experiment to study Rapidly Deployable array Systems, called m-2, was carried out in the southern Timor Sea 160 km west of Darwin, Australia (Fig. 4.1).

(40)

Chapter 4: RDS-2: Rapidly Deployable Systems II 32

Aboard the three vessels involved in the trial, the FRV Southern Surveyor, the TSMV Pacific Conquest, and the HMNZS Manawanui, were members of defense research laboratories from Australia, Canada, New Zealand, the United Kingdom, and the United States.

The aim of the experiment was to test and demonstrate advanced deployable array technologies, data recovery methods, and rapid array deployment techniques [Desharnais and Heard, 19991. To accomplish this aim, the team conducted 40 acoustic experiments (serials) over 10 days in which acoustic signals from an underwater projector were recorded by various hydrophone arrays.

(41)

Cl~apter 4: RDS-2: Rapidly Deployable Systems I1 33

The sound source used in the RDS-2 experiment was a Sonar Research Projector (SRP) that was towed by the Southern Surveyor at a nominal depth of 50 m, with depth verification provided by depth gauge recordings from the SRP's on board pressure/depth sensor. SRP position was calculated using the Southern Surveyor's P-code GPS1 and tow cable length data, also an SRP output. Using these SRP positions, source-receiver ranges are measured from the central node position of the hydrophone array discussed below. For serials analyzed in this research, the SRP emitted a series of narrow band, continuous wave (CW) tonals. Source levels, estimated by a monitor hydrophone close to the SRP, varied between 145 and 155 dB (re: 1pPa) depending

on the frequency, and the bandwidth of each tonal was less than 0.25 Hz.

To record the SRP signals, the team deployed three separate bottom-moored hydrophone arrays - a horizontal/vertical in-line conventional array (ULRICA), a circular planar array (OCTOPUS), and an ultra-light bottom-moored horizontal Y- shaped array (ULITE). All data analyzed in this document pertains t o two acoustic serials recorded by the ULITE array. The experimental setup is depicted in Figure 4.2.

Designed and deployed by the US Space and Naval Warfare System Center (SPAWAR), the ULITE planar seabed array consists of three 470 m long, flexible cable arms extending from a central node. Attached to the cable of each arm are 32 hydrophone elements asymmetrically nested in three distinct subarrays. For a fully extended arm, intended inter-hydrophone spacing is 7.8 m in the first subarray, 15.6 m in the second, and 31.3 m in the third, corresponding to design frequencies of 24, 48, and 96 H z 2

It should be noted that these are design spacings whereas spacing of the deployed array differed significantly. A localization experiment was carried out to determine the actual configuration and a detailed analysis of the ULITE geometry is discussed in Chapter 6.

IP-code GPS is the military encrypted code with an accuracy of approximately 15 m.

2Deszgn frequencies are those with wavelength equal t o twice the inter-hydrophone spacing (Aph =

X

(42)

Chapter 4: RDS-2: Rapidly Deployable Systems I1 34

According to Tolstoy [1993], inter-hydrophone spacing no greater than the wavelength of interest (Aph

<

A) allows for optimal full field processing. As such, the smallest subarray spacing of 7.8 m equates to a 198 Hz frequency, beyond which full field processing can be expected to degrade. In keeping with this theoretical limit, only source signals of frequency 195 Hz and below are processed for MFP and MBP. For CBF, a common rule of thumb states that Aph

<

$

to avoid spatial aliasing. To determine if the deployed ULITE array (shown in Fig. 5.1) conforms to this rule, minimum inter-phone spacings were assessed for all 360 incident angles. The maximum of all the 360 minimum spacings was 3 m. Because

$

= 3.84 m for the highest frequency used in localization (195 Hz), no aliasing can occur with the array.

Additionally, because left-right bearing ambiguity is a property of linear arrays, we conducted a study to examine such ambiguity in the deployed ULITE array. The CBF output for simulated plane wave arrivals from all incident angles demonstrated that left-right ambiguity has been effectively removed by the 2D planar structure of the array.

ULITE recorded data for two independent serials were used in this thesis work. The first data set, denoted 'PDl', was recorded at 06242 hrs on Nov. 3"d, while the

second, denoted 'MF3' was recorded at 00502 hrs on Nov. 4th.

4.2 Site Environmentals

4.2.1 Sound Speed Profiles

Echoing the findings of numerous studies, namely Feuillade et al. [1989, 19901 and Hamson and Heitmeyer [1989], accurate modelling of the sound velocity profile in the water is critical to MFP and MBP performance. As such, it is most desirable to sample the in situ sound velocity profile during each acoustic serial. Although concurrent sampling was not conducted for RDS-2, sound speed profiles were taken

(43)

11

Date GMT Latitude Longitude

I

XBT

Table 4.1 Details of sound speed profiles collected for days 3 to 6 of RDS-2. Highlighted profiles form part of the environmental model used in full field processing. within sufficient spatial and temporal proximity to allow for accurate modelling.

Sound velocity profiles of the water column were collected by two primary means- expendable bathythernlographs (XBT) and conductivity, temperature, depth profilers (CTD). Seven XBT profiles were collected by the Pacific Conquest and the Southern Surveyor deployed a further 17 XBTs as well as two CTD profiles. The details of profiles taken in the days relevant to this work are presented in Table 4.1.

Figure 4.3 displays a series of three XBT profiles collected at various times on 6 November. All profiles were taken within an 8 KM radius and by virtue of distance from land and consistent weather in the area, local effects are negligible. The profiles are representative of those collected throughout the experiment and display a strong

(44)

Chapter 4: RDS-2: Rapidly Deployable Systems II 36

positive gradient thermocline in the surface layer down to approximately 5 m. Ex- amination of salinity data in this region indicates the presence of a persistent layer of fresh water that was most likely maintained by the almost daily tropical rain showers noted in ship logs. Below the surface layer, an isovelocity mixed layer extends to 30-40 m, on top of a weak negative gradient thermocline to a depth of 70 m. Beyond this, pressure effects cause a slight positive sound speed increase extending to the bottom. These four distinct features of the profiles are present in all velocity profile data collected during RDS-2, indicating a stable, stratified water column.

Due to the limitation of sound speed data being collected primarily during daylight

1355 hrs 2045 hrs

1 20

1540 1545 1550 1540 1545 1550 1540 1545 1550

Sound Speed (rnls)

Figure 4.3 Sound speed profiles from XBT samples for 6 Nov. demonstrating stability of velocity structure. Times are local.

(45)

Chapter 4: RDS-2: Rapidly Deployable Systems II 37

hours, the full effect of diurnal temperature changes could not be assessed. However, analysis of all profile data shows little daily variation in the sound speed profile. Figure 4.4 displays the differences in sound speed between the profiles previously examined (Fig. 4.3) and reveals maximum differences of just over 2 m/s between morning, afternoon, and evening profiles. Expected variations in the surface layer temperature due to night time cooling and solar heating were moderated by the cooling effect of the recurring afternoon rain showers. Additionally, the absence of

1

120

-5 0 5 1201 -1 0

1

SVP Difference (mls)

Figure 4.4 Variability of sound speed profiles for 6 Nov.

a. SVP difference between 11:15 and 13:55; b. SVP difference between 11:15 and 20:45; c. SVP difference between 13:55 and 20:45.

(46)

any significant subsurface current was a further factor in the stability of the sound velocity profile. As can be seen in the representative current profiles shown in Fig. 4.5, average current speeds were less than 0.2 m/s or 0.4 knots, and as such were not strong enough to disrupt the sound velocity profile through vertical mixing.

0 0.2 0.4 0.6 0.8 1 [Current Speed (rn/s)l 0 0.2 0.4 0.6 0.8 1 /Current Speed (rn/s)l I: 1 . I .

-

0 100 200 300 Current Direction (T) 08502 0 100 200 300 Current Direction (T)

Figure 4.5 Current profiles collected on 04 Nov during a MBP serial. Bold lines represent low-pass filtered current speed and direction data.

4.2.2 Bathymetry

One of the primary objectives of RDS-2 was to assess array and processor performance in littoral environments. As such, the trial site was intentionally chosen for it's shallow depth and uniform bathymetry.

(47)

Site bathymetric data were not collected as part of RDS-2, and as a result, estimates of water depth and bottom features were obtained from historical data collected primarily from oil exploration in the Timor Sea. Examination of the site bathymetry, taken from the auSEABEP database, reveals a relatively featureless seafloor at an average water depth of 107 m (Fig. 4.6). With few bottom features, and a bottom slope of approximately O.l%, the test site meets a prime criterion for a range- independent environment. Correspondingly, all ORCA acoustic models generated for full field processing were strictly range independent.

4.2.3 Initial Geoacoustic Parameter Estimates

As discussed in Section 3.2.2, the sensitivity of full field processing to environmental mismatch necessitates the use of an effective geoacoustic model. The inconsistency of historical parameter estimates near the RDS-2 site, combined with the scarcity of geoacoustic data below the top of the sediment layer, result in a model with uncertainties too great to provide faithful source localization. As such, these a przorz

parameter estimates are used not to produce a geoacoustic model, but to define the bounds used to refine parameter values. Using geoacoustic inversion, as detailed in the Chapter 6, the initial geoacoustic estimates described below provide search bounds within which the inversion determines individual parameter values used to produce an effective geoacoustic model.

Although significant exploration of the Tiinor Seafloor has been conducted for oil exploration, the bulk of available data regarding seafloor properties has been collected from bottom grabs and shallow cores. Consequently, although estimates of sedimentary properties in the first few metres below the waterlsediment interface are available, they have large uncertainties, and any information regarding bottom properties any deeper than the top few metres is scarce.

3auSEABED is a GIC database of seabed properties based primarily on coring and acoustic data collected from oil exploration and scientific surveys in Australian waters [Jenkins, 19951.

(48)

Chapter 4: RDS-2: Rapidly Deployable Systems 11

Longitude (E)

Figure 4.6 Timor Sea bathymetry in the vicinity of the RDS-2 experiment site. Contours are depth soundings from mean surface level in 10 m increments. All acoustic trials were conducted within the 5 by 8 lun box indicated.

Analysis of seafloor d a t a presented in auSEABED reveals a sedimentary layer with large variation in its composition. With grain size ranging between 0.3 and 7.0

4,4

4Grain size is dehed: 4 = - l o g z ( d ) where d is the individual grain diameter in mm, and is an index property that is commonly used in the estimation of sedimentary acoustic properties such as attenuation and sound velocities [Folk, 19741.

Matched field and matched beam source localization using a bottom moored horizontal array

MATCHED

FIELD

AND

MATCHED

BEAM SOURCE

LOCALIZATION

USING

A

BOTTOM MOORED

HORIZONTAL

ARRAY

Matched Field and Matched Beam Source

Localization Using a Bottom Moored Horizontal

Array

MASTER

SCIENCE

Table of Contents

. . .

. . .

. . .

. . .

. . .

List

of

Tables

. . .

List of Figures

. . .

. . .

. . .

. . .

. . .

. . .

. . .

. . .

. . .

. . .

. . .

R

Z

R

. . .

Chapter

1

Introduction

Chapter

2

Background Acoustic Theory

,

,

~ , h

>>

>> 1). Correspondingly, substitution of (2.20) into (2.22) produces the equa-

(3)

,

2

+

P

$

Chapter

3

Acoustic Data Processing

+

-

m)

.

/"

-

p

-

2

I

F2,

C

J I

l2

+

. . .

+