A deep learning method for reflective boundary estimation.

Environment estimation is a challenging task in reverberant settings such as the underwater and indoor acoustic domains. The locations of reflective boundaries, for example, can be estimated using acoustic echoes and leveraged for subsequent, more accurate localization and mapping. Current boundary estimation methods are constrained to high signal-to-noise ratios or are customized to specific environments. Existing methods also often require a correct assignment of echoes to boundaries, which is difficult if spurious echoes are detected. To evade these limitations, a convolutional neural network (NN) method is developed for robust two-dimensional boundary estimation, given known emitter and receiver locations. A Hough transform-inspired algorithm is leveraged to transform echo times of arrival into images, which are amenable to multi-resolution regression by NNs. The same architecture is trained on transform images of different resolutions to obtain diverse NNs, deployed sequentially for increasingly refined boundary estimation. A correct echo labeling solution is not required, and the method is robust to reverberation. The proposed method is tested in simulation and for real data from a water tank, where it outperforms state-of-the-art alternatives. These results are encouraging for the future development of data-driven three-dimensional environment estimation with high practical value in underwater acoustic detection and tracking.

Machine-learning based detection of marine mammal vocalizations in snapping-shrimp dominated ambient noise.

Passive acoustics is an effective method for monitoring marine mammals, facilitating both detection and population estimation. In warm tropical waters, this technique encounters challenges due to the high persistent level of ambient impulsive noise originating from the snapping shrimp present throughout this region. This study presents the development and application of a neural-network based detector for marine-mammal vocalizations in long term acoustic data recorded by us at ten locations in Singapore waters. The detector's performance is observed to be impeded by the high shrimp noise activity. To counteract this, we investigate several techniques to improve detection capabilities in shrimp noise including the use of simple nonlinear denoisers and a machine-learning based denoiser. These are shown to enhance the detection performance significantly. Finally, we discuss some of the vocalizations detected over three years of our acoustic recorder deployments using the robust detectors developed.

Depth-dependence of the underwater noise emission from melting glacier ice.

Submarine-melting of ice at the glacier-ocean interface accounts for a large portion of the ice-loss at tidewater glaciers and produces sound via bubble-release. The sound production is dominant in the sub-surface region near the glacier-ocean interface. This depth-dependence of the sound is studied by melting ice blocks in a glacial bay at various depths up to 20 m and recording their acoustics over a large frequency range. The acoustic energy decreases with depth in line with expectations from the physics of the phenomenon and is fit to an exponentially decaying curve. The estimated variation will be useful for interpreting the sound in marine-terminating glaciers bays in terms of the submarine-melting activity.

An architecture for passive joint localization and structure learning in reverberant environments.

Passive localization and tracking of a mobile emitter, and joint learning of its reverberant three-dimensional (3D) acoustic environment, where critical structural features are unknown, is a key open problem. Unaccounted-for occluders are potentially present, so that the emitter can lose line-of-sight to the receivers, and can only be observed through its reflected raypaths. The locations of reflective boundaries must therefore be jointly estimated with the emitter's position. A multistage global optimization and tracking architecture is developed to solve this problem with a relatively unconstrained model. Each stage of this architecture establishes domain knowledge such as synchronization and initial environment estimation, which are inputs for the following stages of more refined algorithms. This approach is generalizable to different physical scales and modalities and improves on methods that do not exploit the motion of the emitter. In one stage of this architecture, particle swarm optimization is used to simultaneously estimate the environment and the emitter location. In another stage, a Hough transform-inspired boundary localization algorithm is extended to 3D settings, to establish an initial estimate of the environment. The performance of this holistic approach is analyzed and its reliability is demonstrated in a reverberant watertank testbed, which models the shallow-water underwater acoustic setting.

A dolphin-inspired compact sonar for underwater acoustic imaging.

Underwater imaging sonars are widely used for oceanic exploration but are bulky and expensive for some applications. The sonar system of dolphins, which uses sound pulses called clicks to investigate their environment, offers superior shape discrimination capability compared to human-derived imaging sonars of similar size and frequency. In order to gain better understanding of dolphin sonar imaging, we train a dolphin to acoustically interrogate certain objects and match them visually. We record the echoes the dolphin receives and are able to extract object shape information from these recordings. We find that infusing prior information into the processing, specifically the sparsity of the shapes, yields a clearer interpretation of the echoes than conventional signal processing. We subsequently develop a biomimetic sonar system that combines sparsity-aware signal processing with high-frequency broadband click signals similar to that of dolphins, emitted by an array of transmitters. Our findings offer insights and tools towards compact higher resolution sonar imaging technologies.

Vertical directionality and spatial coherence of the sound field in glacial bays in Hornsund Fjord.

Arctic glacial bays are among the loudest natural environments in the ocean, owing to heavy submarine melting, calving, freshwater discharge, and ice-wave interactions. Understanding the coherence and vertical directionality of the ambient sound there can provide insights about the mechanisms behind the ice loss in these regions. It can also provide key information for operating technologies such as sonar, communication, and navigation systems. To study the unexplored sound coherence and vertical directionality in glacial bays, a vertical hydrophone array was deployed, and acoustic measurements were made at four glacier termini in Hornsund Fjord, Spitsbergen, in June and July 2019. The measurements show that the sound generated by melting glacier ice is more dominant in the upper portion of the water column near the glacier terminus. The melt water from the submarine melting and the freshwater discharge from the glacier create a glacially modified water duct near the sea surface. This disrupts the inter-sensor vertical coherence in the channel. However, some coherence across the duct is preserved for sound arising from spatially localized events at low frequencies. Overall, the observations in this study can help improve the understanding of the submarine melting phenomenon in glacial bays.