Phase stability of convergence zone propagation at mid-frequency.

Mid-frequency transmissions (1.5 to 7.5 kHz) from a towed source to a drifting array one convergence zone away demonstrate smoothly varying phase. Calculations of the strength and diffraction parameters Φ and Λ for a representative background environment predict partially saturated propagation. For a single convergence zone path, the evolving phase rate due to the changing internal wave field corresponds to a narrowband coherent integration time, defined by the length of a discrete Fourier transform that likely captures the signal, from ∼370 s at 10 kHz to ∼1370 s at 1 kHz in the absence of source motion. Simulated propagation with a split-step parabolic equation solver through a sound speed field with thermocline fine structure and a Garrett-Munk internal wave displacement field provides a full wave picture that is consistent with the statistical predictions. Experimental data from the Philippine Sea demonstrates signal phase that is highly correlated with source tow body vertical excursions, measured by a pressure sensor, that couple into horizontal motion. The received signal can be integrated for up to 128 s at 5.5 kHz, with the upper limit thought to be due to non-uniform source motion rather than ocean dynamics.

Inference of source signatures of merchant ships in shallow ocean environmentsa).

An ocean acoustics experiment in 2017 near a shipping lane on the New England continental shelf in about 75 m of water provided an opportunity to evaluate a methodology to extract source signatures of merchant ships in a bottom-limited environment. The data of interest are the received acoustic levels during approximately 20 min time intervals centered at the closest position of approach (CPA) time for each channel on two 16-element vertical line arrays. At the CPA ranges, the received levels exhibit a frequency-dependent peak and null structure, which possesses information about the geophysical properties of the seabed, such as the porosity and sediment thickness, and the characterization of the source, such as an effective source depth. The modeled seabed is represented by two sediment layers, parameterized with the viscous grain shearing (VGS) model, which satisfies causality, over a fixed deep layered structure. Inferred estimates of the implicit source levels require averaging an error function over the full 20 min time intervals. Within the 200-700 Hz band, the Wales-Heitmeyer model captures the inferred frequency dependence of the source levels.

Trans-dimensional inversion for seafloor properties for three mud depocenters on the New England shelf under dynamical oceanographic conditionsa).

This paper presents inversion results for three datasets collected on three spatially separated mud depocenters (hereafter called mud ponds) during the 2022 Seabed Characterization Experiment (SBCEX). The data considered here represent modal time-frequency (TF) dispersion as estimated from a single hydrophone. Inversion is performed using a trans-dimensional (trans-D) Bayesian inference method that jointly estimates water-column and seabed properties along with associated uncertainties. This enables successful estimation of the seafloor properties, consistent with in situ acoustic core measurements, even when the water column is dynamical and mostly unknown. A quantitative analysis is performed to (1) compare results with previous modal TF trans-D studies for one mud pond but under different oceanographic condition, and (2) inter-compare the new SBCEX22 results for the three mud ponds. Overall, the estimated mud geoacoustic properties show no significant temporal variability. Further, no significant spatial variability is found between two of the mud ponds while the estimated geoacoustic properties of the third are different. Two hypotheses, considered to be equally likely, are explored to explain this apparent spatial variability: it may be the result of actual differences in the mud properties, or the mud properties may be similar but the inversion results are driven by difference in data information content.

Deep sediment heterogeneity inferred using very low-frequency features from merchant shipsa).

The very low-frequency noise from merchant ships provides a good broadband sound source to study the deep layers of the seabed. The nested striations that characterize ship time-frequency spectrograms contain unique acoustic features corresponding to where the waveguide invariant ß becomes infinite. In this dataset, these features occur at frequencies between 20 and 80 Hz, where pairs of modal group velocities become equal. The goal of this study is to identify these ß = ∞ frequencies in ship noise spectrograms and use them to perform statistical inference for the deep layer sound speeds and thicknesses in the New England Mudpatch for a larger number of ships and acoustic arrays over a larger geographical region than previously studied. Marginal probability distributions of the data indicate that using singular points for a feature-based inversion yields an estimate of the sound speed and a limiting value for the thickness of the first deep layer. Heterogeneity is examined by correlating spatial variability of the deep layer sound speeds with ship tracks.

Mid-frequency acoustic localization of breaking waves.

During an experiment in deep water off the coast of Southern California, wind speeds ranged from 10 to 15 m/s and wind forcing produced large breaking waves. A mid-frequency vertical planar hydrophone array recorded underwater ambient noise while an airplane equipped with a high-resolution video camera captured images of the sea surface above the array. Beams of ambient noise between 5 and 6 kHz were projected onto the sea surface and synchronized in space and time with the aerial images. Despite the array's limited azimuthal resolution of the surface, due to its modest 1 m horizontal aperture and relatively deep 130 m deployment depth, concentrated areas of high intensity in the acoustic surface projection were observed to match visible breaking events in the aerial images.

Active intensity vortex and stagnation point singularities in a shallow underwater waveguide.

Vector acoustic properties of a narrowband acoustic field are observed as a function of range from a source towed in waters of depth 77 m on the New England Mud Patch. At the source frequency (43 Hz), the waveguide supported three trapped modes, with mode 2 weakly excited owing to the towed source depth. The receiving sensor was positioned 1.45 m above the seafloor with a sampling range aperture of 2500 m. The vector acoustics observations enabled study of vortex regions that encompass two singular points for active acoustic intensity: the vortex point, which is co-located with a dislocation, and stagnation point. Interpretative modeling, based on the normal modes and using a geoacoustic model consistent with those emerging from studies conducted at this location, is in agreement with these measurements. Model-data comparisons were based on the first-order variables of acoustic pressure and velocity along with inverse Hankel transforms, which yield normalized horizontal wavenumber spectra, and second-order variables in the form of horizontal and vertical intensity as well as non-dimensional intensity-based ratios. These measures provide a degree of observational confirmation of some vortex region properties. Both observations and modeling point to a gradual deepening of such regions with increasing range owing to sediment attenuation.

Observations and simulations of caustic formation due to oceanographic fine structure.

An at-sea experiment in deep water was conducted to explore the impact of small-scale sound-speed variability on mid-frequency (1-10 kHz) acoustic propagation. Short-range (1-5 km) acoustic transmissions were sent through the upper ocean (0-200 m) while oceanographic instruments simultaneously measured the ocean environment within 2 km of the single upper turning points of the acoustic transmissions. During these transmissions, acoustic receptions over a 7.875 m vertical line array show closely spaced, sometimes interfering arrivals. Ray and full-wave simulations of the transmissions using nearby sound-speed profiles are compared deterministically to the received acoustic signals. The sensitivity of the acoustic arrivals to the vertical scales of ocean sound speed is tested by comparing the observed and simulated arrival intensity where the sound-speed profile used by the simulation is smoothed to varying scales. Observations and modeling both suggest that vertical fine-scale structures (1-10 m) embedded in the sound-speed profile have strong second derivatives which allow for the formation of acoustic caustics as well as potentially interfering acoustic propagation multipaths.

Moving source ocean acoustic tomography with uncertainty quantification using controlled source-tow observations.

Ocean sound speed and its uncertainty are estimated using travel-time tomography at ranges up to 2 km using a moving source in ∼600 m water depth. The experiment included two 32-element vertical line arrays deployed about 1 km apart and a towed source at ∼10 m depth transmitting a linear frequency modulated waveform. The inversion accounts for uncertainties in the positions and velocities of the source and receivers in addition to the background sound speed state. At these short ranges, the sound speed effects are small and the representational error of the candidate forward models must be carefully evaluated and minimized. This is tested stringently by a separate position parameter inversion and by cross-validating the estimates of sound speed and arrival time, including uncertainties. In addition, simulations are used to explore the effects of adding additional constraints to the inversion and to compare the performance of moving to fixed source tomography. The results suggest that the ray diversity available from the moving source reduces the posterior sound speed uncertainty compared to the fixed source case.

Influence of seabed on very low frequency sound recorded during passage of merchant ships on the New England shelf.

An examination of the received spectrogram levels of about twenty merchant ship recordings on two vertical line arrays deployed on the New England continental shelf during the Seabed Characterization Experiment 2017 has identified an acoustic feature that can be attributed to the group velocities of modes 1 and 2 being equal at a frequency f=F. The observation of such a feature is a result of ßnm(2πF)=∞, where ßnm is the waveguide invariant for modes n and m. For the New England Mudpatch, the average value of F is about 24.5 Hz. An effective seabed model is inferred from a feature inversion method that has a deep sediment layer which lies between 190 m and 290 m beneath the seafloor with sound speeds on the order of 1810 m/s. This effective sediment model appears to be consistent with a previous seismic survey on the New England shelf that identified a deep low speed layer about 250 m beneath the water sediment interface.

Seabed type and source parameters predictions using ship spectrograms in convolutional neural networks.

Broadband spectrograms from surface ships are employed in convolutional neural networks (CNNs) to predict the seabed type, ship speed, and closest point of approach (CPA) range. Three CNN architectures of differing size and depth are trained on different representations of the spectrograms. Multitask learning is employed; the seabed type prediction comes from classification, and the ship speed and CPA range are estimated via regression. Due to the lack of labeled field data, the CNNs are trained on synthetic data generated using measured sound speed profiles, four seabed types, and a random distribution of source parameters. Additional synthetic datasets are used to evaluate the ability of the trained CNNs to interpolate and extrapolate source parameters. The trained models are then applied to a measured data sample from the 2017 Seabed Characterization Experiment (SBCEX 2017). While the largest network provides slightly more accurate predictions on tests with synthetic data, the smallest network generalized better to the measured data sample. With regard to the input data type, complex pressure spectral values gave the most accurate and consistent results for the ship speed and CPA predictions with the smallest network, whereas using absolute values of the pressure provided more accurate results compared to the expected seabed types.

A performance comparison between m-sequences and linear frequency-modulated sweeps for the estimation of travel-time with a moving source.

This manuscript discusses the utility of maximal period linear binary pseudorandom sequences [also referred to as m-sequences or maximum length sequences (MLSs)] and linear frequency-modulated (LFM) sweeps for the purpose of measuring travel-time in ocean-acoustic experiments involving moving sources. Signal design and waveform response to unknown Doppler (waveform dilation or scale factor) are reviewed. For this two-parameter estimation problem, the well-known wide-band ambiguity function indicates, and moving-source observations corroborate, a significant performance benefit from using MLS over LFM waveforms of similar time duration and bandwidth. The comparison is illustrated with a typical experimental setup of a source suspended aft of the R/V Sally Ride to a depth of∼10 m and towed at∼1 m/s speed. Accounting for constant source motion, the root mean square travel-time variability over a 30 min observation interval is 53 µs (MLS) and 141 µs (LFM). For these high signal-to-noise ratio channel impulse response data, LFM arrival-time fluctuations mostly appear random while MLS results exhibit structure believed to be consistent with source (i.e., towed transducer) dynamics. We conclude with a discussion on signal coherence with integration times up to 11 MLS waveform periods corresponding to ∼27 s.

Seabed classification from merchant ship-radiated noise using a physics-based ensemble of deep learning algorithms.

Merchant ship-radiated noise, recorded on a single receiver in the 360-1100 Hz frequency band over 20 min, is employed for seabed classification using an ensemble of deep learning (DL) algorithms. Five different convolutional neural network architectures and one residual neural network are trained on synthetic data generated using 34 seabed types, which span from soft-muddy to hard-sandy environments. The accuracy of all of the networks using fivefold cross-validation was above 97%. Furthermore, the impact of the sound speed and water depth mismatch on the predictions is evaluated using five simulated test cases, where the deeper and more complex architectures proved to be more robust against this variability. In addition, to assess the generalizability performance of the ensemble DL, the networks were tested on data measured on three vertical line arrays in the Seabed Characterization Experiment in 2017, where 94% of the predictions indicated that mud over sand environments inferred in previous geoacoustic inversions for the same area were the most likely sediments. This work presents evidence that the ensemble of DL algorithms has learned how the signature of the sediments is encoded in the ship-radiated noise, providing a unified classification result when tested on data collected at-sea.

Reverberation due to a moving, narrowband source in an ocean waveguide.

In this paper, a model for the bistatic reverberation associated with seafloor scattering of sound from a moving, narrowband source in an ocean waveguide is developed. Studies of the Doppler effect for moving sources in waveguides have typically focused on the forward propagating field where the Doppler shift leads to a splitting or broadening of the received spectrum. In contrast, the contributions to the scattered field come from all directions and as a consequence the spectrum of the received energy is spread across the entire range of Doppler-shifted frequencies possible for the speed of the source. The model developed here uses rays for the incident field, ray-mode analogies for the scattering, and normal modes to propagate the scattered field to the receiver. Results from this model are compared with data collected using a towed source during the Target and Reverberation Experiment 2013. The possible applications of this Doppler reverberation for seafloor characterization are also considered.

Estimating relative channel impulse responses from ships of opportunity in a shallow water environment.

The uncertainty of estimating relative channel impulse responses (CIRs) obtained using the radiated signature from a ship of opportunity is investigated. The ship observations were taken during a 1.4 km (11 min) transect in a shallow water environment during the Noise Correlation 2009 (NC09) experiment. Beamforming on the angle associated with the direct ray-path yields an estimate of the ship signature, subsequently used in a matched filter. Relative CIRs are estimated every 2.5 s independently at three vertical line arrays (VLAs). The relative arrival-time uncertainty is inversely proportional to source bandwidth and CIR signal-to-noise ratio, and reached a minimum standard deviation of 5 µs (equivalent to approximately 1 cm spatial displacement). Time-series of direct-path relative arrival-times are constructed for each VLA element across the 11 min observation interval. The overall structure of these time-series compares favorably with that predicted from an array element localization model. The short-term standard deviations calculated on the direct-path (7 µs) and bottom-reflected-path (17 µs) time-series are in agreement with the predicted arrival-time accuracies. The implications of these observed arrival-time accuracies in the context of estimating sound speed perturbations and bottom-depth are discussed.

Multiple-array passive acoustic source localization in shallow water.

This paper considers concurrent matched-field processing of data from multiple, spatially-separated acoustic arrays with application to towed-source data received on two bottom-moored horizontal line arrays from the SWellEx-96 shallow water experiment. Matched-field processors are derived for multiple arrays and multiple-snapshot data using maximum-likelihood estimates for unknown complex-valued source strengths and unknown error variances. Starting from a coherent processor where phase and amplitude is known between all arrays, likelihood expressions are derived for various assumptions on relative source spectral information (amplitude and phase at different frequencies) between arrays and from snapshot to snapshot. Processing the two arrays with a coherent-array processor (with inter-array amplitude and phase known) or with an incoherent-array processor (no inter-array spectral information) both yield improvements in localization over processing the arrays individually. The best results with this data set were obtained with a processor that exploits relative amplitude information but not relative phase between arrays. The localization performance improvement is retained when the multiple-array processors are applied to short arrays that individually yield poor performance.

Adaptive and compressive matched field processing.

Matched field processing is a generalized beamforming method that matches received array data to a dictionary of replica vectors in order to locate one or more sources. Its solution set is sparse since there are considerably fewer sources than replicas. Using compressive sensing (CS) implemented using basis pursuit, the matched field problem is reformulated as an underdetermined, convex optimization problem. CS estimates the unknown source amplitudes using the replica dictionary to best explain the data, subject to a row-sparsity constraint. This constraint selects the best matching replicas within the dictionary when using multiple observations and/or frequencies. For a single source, theory and simulations show that the performance of CS and the Bartlett processor are equivalent for any number of snapshots. Contrary to most adaptive processors, CS also can accommodate coherent sources. For a single and multiple incoherent sources, simulations indicate that CS offers modest localization performance improvement over the adaptive white noise constraint processor. SWellEx-96 experiment data results show comparable performance for both processors when localizing a weaker source in the presence of a stronger source. Moreover, CS often displays less ambiguity, demonstrating it is robust to data-replica mismatch.

Multi-frequency sparse Bayesian learning for robust matched field processing.

The multi-snapshot, multi-frequency sparse Bayesian learning (SBL) processor is derived and its performance compared to the Bartlett, minimum variance distortionless response, and white noise constraint processors for the matched field processing application. The two-source model and data scenario of interest includes realistic mismatch implemented in the form of array tilt and data snapshots not exactly corresponding to the range-depth grid of the replica vectors. Results demonstrate that SBL behaves similar to an adaptive processor when localizing a weaker source in the presence of a stronger source, is robust to mismatch, and exhibits improved localization performance when compared to the other processors. Unlike the basis or matching pursuit methods, SBL automatically determines sparsity and its solution can be interpreted as an ambiguity surface. Because of its computational efficiency and performance, SBL is practical for applications requiring adaptive and robust processing.

A double-difference method for high-resolution acoustic tracking using a deep-water vertical array.

Ray-tracing is typically used to estimate the depth and range of an acoustic source in refractive deep-water environments by exploiting multipath information on a vertical array. However, mismatched array inclination and uncertain environmental features can produce imprecise trajectories when ray-tracing sequences of individual acoustic events. "Double-difference" methods have previously been developed to determine fine-scale relative locations of earthquakes along a fault [Waldhauser and Ellsworth (2000). Bull. Seismolog. Soc. Am. 90, 1353-1368]. This technique translates differences in travel times between nearby seismic events, recorded at multiple widely separated stations, into precise relative displacements. Here, this method for acoustic multipath measurements on a single vertical array of hydrophones is reformulated. Changes over time in both the elevation angles and the relative arrival times of the multipath are converted into relative changes in source position. This approach is tested on data recorded on a 128-element vertical array deployed in 4 km deep water. The trajectory of a controlled towed acoustic source was accurately reproduced to within a few meters at nearly 50 km range. The positional errors of the double-difference approach for both the towed source and an opportunistically detected sperm whale are an order of magnitude lower than those produced from ray-tracing individual events.

Change-point detection for recursive Bayesian geoacoustic inversions.

In order to carry out geoacoustic inversion in low signal-to-noise ratio (SNR) conditions, extended duration observations coupled with source and/or receiver motion may be necessary. As a result, change in the underlying model parameters due to time or space is anticipated. In this paper, an inversion method is proposed for cases when the model parameters change abruptly or slowly. A model parameter change-point detection method is developed to detect the change in the model parameters using the importance samples and corresponding weights that are already available from the recursive Bayesian inversion. If the model parameters change abruptly, a change-point will be detected and the inversion will restart with the pulse measurement after the change-point. If the model parameters change gradually, the inversion (based on constant model parameters) may proceed until the accumulated model parameter mismatch is significant and triggers the detection of a change-point. These change-point detections form the heuristics for controlling the coherent integration time in recursive Bayesian inversion. The method is demonstrated in simulation with parameters corresponding to the low SNR, 100-900 Hz linear frequency modulation pulses observed in the Shallow Water 2006 experiment [Tan, Gerstoft, Yardim, and Hodgkiss, J. Acoust. Soc. Am. 136, 1187-1198 (2014)].

Spatial filtering in ambient noise interferometry.

Theoretically, the empirical Green's function between a pair of receivers can be extracted from the cross correlation of the received diffuse noise. The diffuse noise condition rarely is met in the ocean and directional sources may bias the Green's function. Here matrix-based spatial filters are used for removing unwanted contributions in the cross correlations. Two methods are used for solving the matrix filter design problem. First a matrix least-square problem is solved with a low-rank approximation of the pseudo-inverse, here, derived for linear and planar arrays. Second, a convex optimization approach is used to solve the design problem reformulated with ad hoc constraints. The spatial filter is applied to real-data cross correlations of elements from a linear array to attenuate the contribution of a discrete interferer. In the case of a planar array and simulated data, a spatial filter enables a passive upgoing/downgoing wavefield separation along with an efficient rejection of horizontally propagating noise. The impact of array size and frequency band on the filtered cross correlations is discussed.