Data-Driven Methods for Structure Learning in Underwater Acoustic Modeling
Abstract
Traditional ocean acoustic modeling and array signal processing are key ingredients in a host of Naval concepts of operations. The last decade has witnessed unprecedented advances in machine learning, data science, and signal processing leading to powerful, data-driven computational paradigms for building models that capture complex structure in high-dimensional data, holding the promise of similar gains in performance and new capabilities in underwater acoustics. Only by coupling ocean physical models into these powerfuldata-driven approaches can these performance gains be realized in a manner that is robust to the variability and uncertainty of theocean environment, as this research will emphasize.This effort pursues advances in these tools for the ocean acoustic environment, taking into account its unique characteristics and the capabilities that are ultimately sought in them. This will involve the development of performance predictions and guarantees, which are key to their use in future applications. Importantly, the effort will include a critical evaluation of the resulting methods, quantifying gains, limitations, robustness, computational requirements, interpretability, versatility, and other metrics, relative to traditional approaches.The proposed effort will include the following components, based on available resources:Task 1: Neural acoustic fields ( NAFs) are promising candidates for efficiently representing acoustic environments in the creation of realistically synthesized views from limited observational data. This task will develop and apply NAFs for underwater acoustic scene modeling, with applications including localization, tracking, object classification, environmental modeling, and other tasks involving the information extraction from acoustic data.Task 2: Diffusion models are promising generative models for acoustic data. These models will be developed for a variety of ocean applications, including self-localization and tracking, de-reverberation, and signal separation. They will further be pursued as a methodology for efficient simulation of underwater acoustic environments, augmenting collected data.Task 3: Contrastive learning is frequently used to significantly improve the accuracy of learning methods when labeled data is limited, making it compelling for underwater applications. These methods will be developed to efficiently exploit widely available unlabeled aggregated acoustic data sets for tasks including classification and localization.Task 4: Physics-informed neural networks (PINNs) represent an important approch to improving the performance and reliability ofneural networks in underwater acoustic applications. Such architectures will be extended to facilitate our development of differentiable acoustic field models to address problems of localization and environment learning.Task 5: Subsea object classification using b oth passive and active optical methods has shown promise for applications in ASW and MCM usingmethods from optical tomography. This task will pursue multi-modal learning methods for jointly exploiting acoustic and optical data in recovering structure in underwater acoustic scenes. Experiemental evaluations will be conducted beneath the SIO pier.Task 6: Key to the proposed effort is the critical couplings between theory, algorithms, implementation, experiment, and hands-on student training in ocean acoustic research. In support, this task comprises an ongoing annual Underwater Acoustic Bootcamp at UCSD/SIO and SBU/MSC for students, at which key resources of the team will be leveraged, including the high-frequency testbed (HiFAAT), the SOARS wavetank at Scripps, Scripps Pier, and the Marine Sciences Center at Stony Brook University.Approved for Public Release.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Nov 21, 2023
- Source ID
- N000142412004
Entities
People
- Andrew C Singer
Organizations
- Office of Naval Research
- Research Foundation for the State University of New York
- United States Navy