Impact of data augmentation on supervised learning for a moving mid-frequency source.

Two residual networks are implemented to perform regression for the source localization and environment classification using a moving mid-frequency source, recorded during the Seabed Characterization Experiment in 2017. The first model implements only the classification for inferring the seabed type, and the second model uses regression to estimate the source localization parameters. The training is performed using synthetic data generated by the ORCA normal mode model. The architectures are tested on both the measured field and simulated data with variations in the sound speed profile and seabed mismatch. Additionally, nine data augmentation techniques are implemented to study their effect on the network predictions. The metrics used to quantify the network performance are the root mean square error for regression and accuracy for seabed classification. The models report consistent results for the source localization estimation and accuracy above 65% in the worst-case scenario for the seabed classification. From the data augmentation study, the results show that the more complex transformations, such as time warping, time masking, frequency masking, and a combination of these techniques, yield significant improvement of the results using both the simulated and measured data.

Learning location and seabed type from a moving mid-frequency source.

While source localization and seabed classification are often approached separately, the convolutional neural networks (CNNs) in this paper simultaneously predict seabed type, source depth and speed, and the closest point of approach. Different CNN architectures are applied to mid-frequency tonal levels from a moving source recorded on a 16-channel vertical line array (VLA). After training each CNN on synthetic data, a statistical representation of predictions on test cases is presented. The performance of a single regression-based CNN is compared to a multitask CNN in which regression is used for the source parameters and classification for the seabed type. The impact of water sound speed profile and seabed variations on the predictions is evaluated using simulated test cases. Environmental mismatch between the training and testing data has a negative impact on source depth estimates, while the remaining labels are estimated tolerably well but with a bias towards shorter ranges. Similar results are found for data measured on two VLAs during Seabed Characterization Experiment 2017. This work shows the superiority of multitask learning and the potential for using a CNN to localize an acoustic source and detect the surficial seabed properties from mid-frequency sounds.

Parabolic equation modeling of high frequency acoustic transmission with an evolving sea surface.

The present paper examines the temporal evolution of acoustic fields by modeling forward propagation subject to sea surface dynamics with time scales of less than a second to tens of seconds. A time-evolving rough sea surface model is combined with a rough surface formulation of a parabolic equation model for predicting time-varying acoustic fields. Surface waves are generated from surface wave spectra, and stepped in time using a Runge-Kutta integration technique applied to linear evolution equations. This evolving, range-dependent surface information is combined with other environmental parameters and input to the acoustic model, giving an approximation of the time-varying acoustic field. The wide-angle parabolic equation model manages the rough sea surfaces by molding them into the boundary conditions for calculations of the near-surface acoustic field. This merged acoustic model is validated using concurrently-collected acoustic and environmental information, including surface wave spectra. Data to model comparisons demonstrate that the model is able to approximate the ensemble-averaged acoustic intensity at ranges of about a kilometer for acoustic signals of around 15 kHz. Furthermore, the model is shown to capture variations due to surface fluctuations occurring over time scales of less than a second to tens of seconds.

Observation of sound focusing and defocusing due to propagating nonlinear internal waves.

Fluctuations of the low frequency sound field in the presence of an internal solitary wave packet during the Shallow Water '06 experiment are analyzed. Acoustic, environmental, and on-board ship radar image data were collected simultaneously before, during, and after a strong internal solitary wave packet passed through the acoustic track. Preliminary analysis of the acoustic wave temporal intensity fluctuations agrees with previously observed phenomena and the existing theory of the horizontal refraction mechanism, which causes focusing and defocusing when the acoustic track is nearly parallel to the front of the internal waves [J. Acoust. Soc. Am., 122(2), pp. 747-760 (2007)].

Acoustic field variability induced by time evolving internal wave fields

A space- and time-dependent internal wave model was developed for a shallow water area on the New Jersey continental shelf and combined with a propagation algorithm to perform numerical simulations of acoustic field variability. This data-constrained environmental model links the oceanographic field, dominated by internal waves, to the random sound speed distribution that drives acoustic field fluctuations in this region. Working with a suite of environmental measurements along a 42-km track, a parameter set was developed that characterized the influence of the internal wave field on sound speed perturbations in the water column. The acoustic propagation environment was reconstructed from this set in conjunction with bottom parameters extracted by use of acoustic inversion techniques. The resulting space- and time-varying sound speed field was synthesized from an internal wave field composed of both a spatially diffuse (linear) contribution and a spatially localized (nonlinear) component, the latter consisting of solitary waves propagating with the internal tide. Acoustic simulation results at 224 and 400 Hz were obtained from a solution to an elastic parabolic equation and are presented as examples of propagation through this evolving environment. Modal decomposition of the acoustic field received at a vertical line array was used to clarify the effects of both internal wave contributions to the complex structure of the received signals.

Acoustic normal mode fluctuation statistics in the 1995 SWARM internal wave scattering experiment

In order to understand the fluctuations imposed upon low frequency (50 to 500 Hz) acoustic signals due to coastal internal waves, a large multilaboratory, multidisciplinary experiment was performed in the Mid-Atlantic Bight in the summer of 1995. This experiment featured the most complete set of environmental measurements (especially physical oceanography and geology) made to date in support of a coastal acoustics study. This support enabled the correlation of acoustic fluctuations to clearly observed ocean processes, especially those associated with the internal wave field. More specifically, a 16 element WHOI vertical line array (WVLA) was moored in 70 m of water off the New Jersey coast. Tomography sources of 224 Hz and 400 Hz were moored 32 km directly shoreward of this array, such that an acoustic path was constructed that was anti-parallel to the primary, onshore propagation direction for shelf generated internal wave solitons. These nonlinear internal waves, produced in packets as the tide shifts from ebb to flood, produce strong semidiurnal effects on the acoustic signals at our measurement location. Specifically, the internal waves in the acoustic waveguide cause significant coupling of energy between the propagating acoustic modes, resulting in broadband fluctuations in modal intensity, travel-time, and temporal coherence. The strong correlations between the environmental parameters and the internal wave field include an interesting sensitivity of the spread of an acoustic pulse to solitons near the receiver.

Modeling mode arrivals in the 1995 SWARM experiment acoustic transmissions

As part of the Shallow Water Acoustics in a Random Medium (SWARM) experiment, a 16 element WHOI vertical line array (WVLA) was moored in 70 m of water off the New Jersey coast. A 400-Hz acoustic tomography source was moored some 32-km shoreward of this array, such that an acoustic path was created that was anti-parallel to the primary propagation direction for shelf-generated internal wave solitons. The presence of these soliton internal waves in the acoustic waveguide causes significant coupling of energy between propagating acoustic modes, creating fluctuations in modal intensities and modal peak arrival times, as well as time spreading of the pulses. Two methods by which acoustic propagation and scattering in soliton-filled waveguides can be modeled are presented here in order to understand and explain the scattering observed in the SWARM field data. The first method utilizes the Preisig and Duda [IEEE J. Ocean. Eng. 22, 256-269 (1997)] Sudden Interface Approximation (SIA) to represent the solitons. The second method, which is computationally slower, uses a finely meshed, "propagated" thermistor record to simulate the solitons in the SWARM experiment waveguide. Both numerical methods are found to generate scattering characteristics that are similar to the SWARM field data.