We propose a meshless geoacoustic inversion framework based on differentiable modeling to estimate seabed parameters from spatially distributed sensor array measurements. The method establishes an optimization loop comprising a forward pass strictly adhering to physical constraints and a backward pass that leverages the implicit function theorem to compute gradients efficiently for iteratively updating seabed parameters. By operating in a continuous search space, the approach avoids the quantization errors and dimensionality curse of grid-based methods. Numerical experiments in a Pekeris waveguide and a stratified waveguide demonstrate that the proposed method can accurately recover seabed sound speed, density, and attenuation.

Predicting underwater acoustic transmission loss rapidly and accurately is crucial for real-time ocean acoustic applications. While Fourier Neural Operators (FNO) have emerged as powerful surrogate models due to their global receptive fields, they suffer from spectral bias. The frequency truncation mechanism in FNO filters out high-frequency components, resulting in over-smoothed predictions that fail to capture fine-grained interference patterns. To overcome this limitation, this paper proposes a Spectral-Spatial Residual Learning (S2RL) framework. S2RL decomposes the prediction task into a coarse-to-fine process: a spectral Global Propagator first generates a globally consistent prediction, and a spatial Local Refiner subsequently recovers the high-frequency residuals. Experimental results on a South China Sea dataset show that the proposed method significantly outperforms FNO baselines while maintaining millisecond-level inference speeds.

We demonstrate Bayesian optimization (BO), an efficient global optimization method for acoustic inversion in which an ocean acoustic source is localized while simultaneously estimating propagation environment properties. This problem requires Monte Carlo techniques with thousands of forward model evaluations to solve; however, BO can approximate the optimal solution within dozens or hundreds of evaluations. BO performs a sequential search of the parameter space for the global optimum of an objective function, here defined as the correlation between predicted and measured data. At each step, a Gaussian process (GP) surrogate model is fit to the observed data and a new point to evaluate is chosen using an acquisition function. Conventionally, the GP is fit to the entire parameter space. We adopt trust regions based on observed data to enable fitting of GPs on local subsets of the parameter space, leading to improved optimization results. The method is demonstrated on simulated and experimental data over a 7-dimensional search space encompassing source location and four geoacoustic parameters.

Falls represent a critical health hazard for elderly individuals, often leading to severe injuries and decreased quality of life. While existing fall detection systems predominantly rely on threshold-based algorithms or conventional machine learning approaches like CNNs and RNNs, they frequently exhibit limitations in capturing the complex temporal dependencies and spatial features inherent in fall data. This paper presents a novel multi-head attention transformer architecture specifically designed for wearable in situ fall detection. Our approach leverages the transformer’s self-attention mechanism to effectively model long-range dependencies in inertial measurement unit (IMU) sensor data collected from a shinbone-mounted device. The proposed architecture introduces several key innovations: (1) a positionaware embedding layer that preserves the temporal structure of IMU signals, (2) multi-head self-attention blocks that simultaneously attend to different signal features across varying time scales, and (3) a hybrid feature fusion module that combines global context with local patterns. Experimental results on our human subjects dataset demonstrate the critical importance of addressing dataset imbalance. A baseline model trained on imbalanced data achieved only 48.0% fall sensitivity, rendering it unsafe for practical use. By training our proposed transformer architecture on a large, balanced dataset with a specialized Focal Loss function, we increased fall sensitivity to 90.5% on the test set while maintaining an overall accuracy of 93.1% (validation best epoch: sensitivity 92.5% and specificity 95.8%, see Fig. 3). Furthermore, we present a complete hardware/software co-design, including a custom-designed wearable PCB with optimized power management and a comprehensive strategy for deploying our model on an ultra-low-power Lattice Semiconductor iCE40UP5K FPGA, achieving efficient operation with only 22mAcurrent consumption.We frame this study as technical feasibility and edge deployment validation; direct elderly clinical validation remains future work.

Distributed acoustic sensing (DAS) turns fiber-optic cables into distributed, passive sensors suited for continuous, spatially extended ocean monitoring. Specifically, an optoelectronic interrogator injects laser pulses into the fiber and measures the phase modulation of Rayleigh backscattered light, which is caused by external acoustic wavefields inducing elastic strain over a gauge length. A DAS sensor (channel) is commonly considered a point sensor for signal processing, i.e., introduces no spatial coherence to the measured signal. However, DAS sensors have a non-negligible spatial extent due to the transfer function between the measured optical signal and the external acoustic field. The transfer function between optical and acoustic quantities is factorized into four terms, describing the filtering effect of the acquisition gauge window and the spatial averaging window, and of the acoustic wavenumber and direction of arrival. The resulting sensitivity as a function of frequency and angular direction of individual DAS sensors is related to the sensor’s equivalent spatial aperture. The spatial shape of an individual DAS sensor is derived theoretically, and is quantified in common array signal processing terms, such as equivalent spatial aperture, directivity, and beampattern. The shape of the spatial aperture determines the spatial coherence of DAS measurements in a diffuse acoustic wavefield, as demonstrated on publicly available data. The corresponding spatial coherence predicts the statistical characteristics of the speckle pattern in DAS.

Accurate localization and mapping are challenging in environments with poor visibility or noise interference, such as underwater, at night, or in fog, where visual data are unreliable or limited. We present an approach to enhance 3D drone localization and sound mapping using acoustic Simultaneous Localization and Mapping. This method incorporates Angle of Arrival and Time Difference of Arrival measurements of sound sources, as well as quaternions to represent orientation. We simulate a drone equipped with three receiver stations to map multiple stationary sound sources, along with an inertial measurement unit for drone localization. This paper shows drone localization and source mapping using only acoustic and inertial measurements.

Electronic activity in digital systems unintentionally emits radio frequency (RF) signals called emanations. These emanations compromise data security, which is important for corporate and military establishments. This work focuses on detecting anomalous activity that compromises data security through emanations. An example of such anomalous activity is emanations from damaged peripherals, such as a mouse or keyboard, which can be used to steal digital data. Prior work on emanation detection uses profiling on specific hardware (HW). However, this is not scalable across all types of HW. We propose a HW-agnostic solution for finding anomalous activity using emanations by scanning the signature of harmonics from leakages of clock signals. An algorithm for multi-harmonic pitch estimation is introduced for wireless applications. A preprocessing technique is developed that removes the effect of artifacts. Thorough mathematical derivations demonstrate the algorithm theoretically. In-phase and Quadrature-phase (IQ) data are collected from emanation sources placed in a shielded room from 0.1-1.1 GHz using software-defined radios (SDR). Results are presented for use cases emulating anomalous activity that compromises data security, such as damaged peripherals and unauthorized data copy onto external devices.

Distributed acoustic sensing (DAS) turns a fiber-optic cable into a long, dense array of sensors for passive acoustic monitoring, which is well-suited for ocean acoustic measurements. DAS measures phase differences of the backscattered light from a laser pulse in an optical fiber to provide localized strain estimates. These strain estimates enable distributed sensing of external acoustic vibrations. The transfer function between the measured optical signal and the external acoustic quantities is factorized into four terms, describing the filtering effect of the acquisition gauge window and spatial averaging window and the acoustic wavenumber and direction of arrival. The resulting beampattern, i.e., the sensitivity as a function of frequency and angular direction, of individual DAS sensors is derived, and it is related to the sensor's equivalent spatial aperture. The shape of the spatial aperture determines the spatial coherence of DAS measurements in a diffuse acoustic wavefield, as demonstrated on publicly available data. The corresponding spatial coherence predicts the statistical characteristics of the speckle pattern in DAS.

We address three key challenges in learning continuous kernel representations: computational efficiency, parameter efficiency, and spectral bias. Continuous kernels have shown significant potential, but their practical adoption is often limited by high computational and memory demands. Additionally, these methods are prone to spectral bias, which impedes their ability to capture high-frequency details. To overcome these limitations, we propose a novel approach that leverages sparse learning in the Fourier domain. Our method enables the efficient scaling of continuous kernels, drastically reduces computational and memory requirements, and mitigates spectral bias by exploiting the Gibbs phenomenon.

Environment-aware underwater acoustic detection and communication require precise forecasting of the range-dependent sound speed field (SSF) at any given time. Recently, methods such as Gaussian process regression (GPR) and conditional diffusion models have shown advanced performance in SSF forecasting. However, limitations remain: standard GPR fails to capture the range-depth spatial correlations, and conditional diffusion models struggle with continuous forecasting. To address these issues, we integrate multi-output GPR and conditional diffusion models to enable continuous forecasting of range-dependent SSFs, employing careful designs for diffusion noise, neural architecture, and training strategies. Our experiments, conducted on HYCOM hindcast datasets from the South China Sea, demonstrate that our proposed model outperforms state-of-the-art baselines in forecasting range-dependent SSFs at any given time and the associated underwater transmission losses.

Underwater acoustic target localization (UATL) is challenging but has achieved some success with limitations. For example, matched field processing (MFP) is sensitive to environmental noise and inefficient in processing large-scale data, making real-time with accurate performance difficult. This paper presents a Bayesian optimization-tuned machine learning approach for UATL and conducts comparative studies with MFP and other parameter tuning methods. The environment used is from the seatrial conducted on October 26, 1993, in the shallow sea area north of Elba Island. First, the simulated training data is generated by the KRAKEN propagation code on grids of ranges and depths. Second, MFP and two machine learning methods (k-nearest neighbor, support vector regression) with distinct hyperparameter optimization approaches are employed for localization. The results show that the machine learning approaches achieve higher localization accuracy than MFP, identifying the underwater target located at a 5.6 km range (error < 0.1 km) and 79 m depth (error < 0.5 m), while Bayesian optimization proves more efficient than alternative tuning methods.

This paper focuses on improving 3D sound mapping using acoustic simultaneous localization and mapping (SLAM), angle of arrival (AOA), and time difference of arrival (TDOA). We use quaternion for orientation tracking. Through simulations involving a drone equipped with microphone arrays and an inertial measurement unit (IMU), the paper shows the effective localization of a drone moving in a room and mapping multiple stationary sources in a dynamic environment.

Extensive monitoring of acoustical activities is important for many fields, including biology, security, and ocean and Earth science. Distributed acoustic sensing (DAS) is an evolving technique for continuous, wide-coverage measurements of mechanical vibrations across oceans. DAS illuminates a fiber-optic cable with laser pulses and measures the backscattered wave due to small random variations in the refractive index of the material. Specifically, DAS uses coherent optical interferometry to measure the phase difference of the backscattered wave from adjacent locations along the fiber. External stimuli, such as mechanical strain due to acoustic wavefields impinging on the fiber-optic cable, modulate the backscattered wave. Hence, the differential phase measurements of the optical backscatter are proportional to the underlying physical quantities of the surrounding wavefield. Continuous measurement of the backscattered electromagnetic signal provides a distributed sensing modality for the external acoustic wavefield that extends spatially along the fiber. We provide a comprehensive overview of DAS technology and detail the underlying physics, from electromagnetic to mechanical and eventually acoustic quantities. We explain the effect of DAS acquisition parameters in signal processing and show the potential of DAS for sound source detection on data collected from the Ocean Observatories Initiative, DOI: https://doi.org/10.58046/5J60-FJ89.

Representing variable-length and continuous-indexed signals through a linear combination of basis functions poses a fundamental challenge in science and engineering. Current approaches resort to preprocessing steps, such as interpolation and extrapolation, to handle irregular and off-grid measurements, which compromise the physical nature of signals and degrade the representation performance. To address this challenge, rather than utilizing discrete vectors, we introduce a Bayesian functional representation model that capitalizes on the continuous nature and rich expressiveness of Gaussian processes to facilitate interpretable and effective basis function learning. Moreover, an analytical and efficient algorithm based on the variational inference framework is developed. Experimental results using real-life datasets demonstrate the superior performance of our proposed method.

Extensive monitoring of acoustic activities is important for many fields, including biology, security, oceanography, and Earth science. Distributed acoustic sensing (DAS) is an evolving technique for continuous, wide-coverage measurements of mechanical vibrations, which is suited to ocean applications. DAS illuminates an optical fiber with laser pulses and measures the backscattered wave due to small random variations in the refractive index of the material. External stimuli, such as mechanical strain due to acoustic wavefields impinging on the fiber-optic cable, modulate the backscattered wave. Continuous measurement of the backscattered signal provides a distributed sensing modality of the impinging wavefield. Considering the potential use of existing telecommunication fiber-optic cables deployed across the oceans, DAS has emerged as a promising technology for monitoring the underwater soundscape. This review presents advances in DAS in the last decade and details the underlying physics from electromagnetic to mechanical and eventually acoustic quantities. To guide the use of DAS for ocean applications, the effect of DAS acquisition parameters in signal processing is explained. Finally, DAS is demonstrated on data from the OOI Regional Cabled Array for the detection of sound sources, such as whales, ships, and earthquakes.

Sound field reconstruction involves estimating sound fields from a limited number of spatially distributed observations. This work introduces a differentiable physics approach for sound field reconstruction, where the initial conditions of the wave equation are approximated with a neural network and the differential operator is computed with a differentiable numerical solver. The use of a numerical solver enables a stable network training while enforcing the physics as a strong constraint, in contrast to conventional physics-informed neural networks, which include the physics as a constraint in the loss function. This study introduces an additional sparsity-promoting constraint to achieve meaningful solutions even under severe undersampling conditions. Experiments demonstrate that the proposed approach can reconstruct sound fields under extreme data scarcity, achieving higher accuracy and better convergence compared to physics-informed neural networks.

We develop a closed-form framework for three-dimensional sound source localization that incorporates the motion of acoustic arrays. Previous closed-form methods for TDOA–AOA localization assume that sensor stations are located at fixed positions; here, the formulation is extended to arrays whose positions and orientations vary over time. Quaternion-based transformations are used to represent array orientation and map measurements into the global frame, thereby avoiding the singularities associated with angle-based representations and enabling the accurate handling of arbitrary rotations. The method fuses Time Difference of Arrival (TDOA) and Angle of Arrival (AOA) observations from three arrays through a weighted least squares solution, providing an analytical estimate of source position that remains stable in the presence of measurement noise. Accounting for array motion eliminates the need for iterative search and ensures that localization accuracy is maintained as the observation geometry changes. This framework supports applications where static sensors are impractical, including underwater environments where visual sensing is limited and positioning systems such as GPS are unavailable. By enabling sound source localization with moving arrays, the method provides a foundation for reliable mapping and navigation in conditions where traditional sensing approaches are ineffective.

This paper proposes a frequency hopping binary frequency shift keying underwater acoustic (UWA) communication system, where a denoising diffusion probabilistic model (DDPM) and a convolutional neural network (CNN) are sequentially used for signal reconstruction and signal demodulation, respectively. Unlike the deep transfer learning (DTL)-based system, this system employs a DDPM to process the received Mel-spectrogram, reconstructing the distorted Mel-spectrogram caused by UWA channel effects, and generating a spectrogram that approximates the transmitted signal, which is then demodulated by the CNN. The proposed system outperforms conventional systems and achieves performance comparable to DTL-based systems in simulation and experiment. DTL requires data samples from new scenarios to learn signal characteristics during deployment; in contrast, this method uses the generative capability of DDPM to enable direct deployment in dynamic underwater environments without additional adaptation processes, offering flexibility and suitability for complex and variable UWA propagation channels.

Accurate, high-resolution three-dimensional (3D) sound speed fields (SSFs) are crucial for characterizing ocean sound propagation, enabling environment-aware underwater acoustic detection and communications. However, SSF data gathered by current observation systems often suffer from low resolution and noise, diminishing their effectiveness in subsequent tasks and necessitating advanced methods to restore their intricate details-referred to as super-resolution (SR) methods. Existing SR methods are either model-based or data-driven. Both paradigms, however, have their limitations in restoring the subtle variations within SSFs, particularly in challenging scenarios with sparse and noisy measurements. To tackle these challenges, we propose a hybrid SR algorithm, named Hybrid-DOT, which combines the strengths of both data-driven and model-driven approaches. Specifically, we employ a pre-trained diffusion model as the data-driven prior to exploit the information inside the data distribution, and a low-rank tensor (LRT) modeling as the model-based prior to integrate domain knowledge. The incorporation of the LRT not only enhances SR performance but also saves the sampling steps of the diffusion model. Experimental results using 3D SSF datasets demonstrate that Hybrid-DOT surpasses state-of-the-art methods across various SR factors and noise levels, enabling accurate characterizations of fine-grained acoustic transmission losses.

Functional tensor decomposition can analyze multi-dimensional data with real-valued indices, paving the path for applications in machine learning and signal processing. A limitation of existing approaches is the assumption that the tensor rank-a critical parameter governing model complexity-is known. However, determining the optimal rank is a non-deterministic polynomial-time hard (NP-hard) task and there is a limited understanding regarding the expressive power of functional low-rank tensor models for continuous signals. We propose a rank-revealing functional Bayesian tensor completion (RR-FBTC) method. Modeling the latent functions through carefully designed multioutput Gaussian processes, RR-FBTC handles tensors with real-valued indices while enabling automatic tensor rank determination during the inference process. We establish the universal approximation property of the model for continuous multi-dimensional signals, demonstrating its expressive power in a concise format. To learn this model, we employ the variational inference framework and derive an efficient algorithm with closed-form updates. Experiments on both synthetic and real-world datasets demonstrate the effectiveness and superiority of the RR-FBTC over state-of-the-art approaches. The code is available at https://github.com/OceanSTARLab/RR-FBTC.