CAREER: Online Learning-based Underwater Acoustic Communications and Networking

水声通信网络是在科学研究、污染检测、海上勘探和战术监视等广泛应用中实现无人、原位和实时水生监测的使能技术。水下系统的使用寿命从几年到几十年不等，而多尺度水声环境的时空动态对高效可靠的声学数据传输提出了巨大挑战。该项目的目标是为水声通信和网络开发一个基于在线学习的基础和系统框架，其中水声系统1)建模和预测声环境的长期动态，2)主动调整其通信和网络策略以适应环境的动态，从而最大限度地提高系统在能源效率、频谱效率、以及传输可靠性。通过对其环境的明确了解，拟议的框架将允许与其他声学系统（包括海洋动物）和谐共存，以实现生态友好的运作。该项目的研究将通过夏季青少年K-12外展、课程开发、本科生和研究生培训（特别针对女性和代表性不足的少数民族）与教育相结合，并与当地美元湾高中的水下机器人团队合作。这些活动旨在激励和更好地培训农村、女性、少数民族和经济条件不利的学生从事STEM职业。该项目通过在三个相互关联的领域进行创新，解决了基于在线学习的水声通信和网络的基本挑战。首先，将开发新的信号处理和稀疏学习技术来模拟和预测水声环境的大尺度动态和小尺度衰落的统计分布，包括声传输损失、环境声景观和外部（人为和海洋动物）声源的统计特征。其次，基于声环境预测，构建联合传输功率控制、链路调度、节点协同路由和自动驾驶车辆移动控制的优化框架，实现高网络利用率和与其他声系统的和谐共存。第三，声环境探索与开发的权衡将在贝叶斯强化学习框架中解决，这将提供一种有原则的方法来权衡通信和网络策略的直接回报及其揭示环境动态的相关长期利益。利用密歇根理工大学的地理优势和密歇根理工大学大湖研究中心最先进的设施，将在软件定义的网络架构中进行广泛的现场实验，以进行声学测量收集和离线和在线算法评估。本课题所开发的方法和横切技术可应用于智能射频通信网络的设计。

Underwater acoustic communication networks are the enabling technologies for unmanned, in situ, and real-time aquatic monitoring in a wide range of applications, such as scientific studies, pollution detection, offshore exploration, and tactical surveillance. The lifespan of underwater systems varies from a few years to decades, while the spatiotemporal dynamics of underwater acoustic environments at multiple scales pose grand challenges to efficient and reliable acoustic data transmission. The objective of this project is to develop a fundamental and systematic online-learning-based framework for underwater acoustic communications and networking, where the underwater acoustic system 1) models and predicts the long-term dynamics of the acoustic environment, and 2) proactively adapts its communication and networking strategy to the dynamics of the environment, thereby maximizing the long-term system performance in the aspects of energy efficiency, spectrum efficiency, and transmission reliability. Through explicit learning about its environment, the proposed framework will allow harmonious co-existence with other acoustic systems, including marine animals, to achieve eco-friendly operation. This project's research will be integrated with education through summer youth K-12 outreach, curriculum development, undergraduate and graduate student training that will be particularly tailored to females and underrepresented minorities, and collaboration with an underwater robotics team from the local Dollar Bay High School. These activities are designed to motivate and better train rural, female, minority, and economically disadvantaged students to pursue STEM careers.This project tackles fundamental challenges in online-learning-based underwater acoustic communications and networking by innovating across three interrelated domains. First, novel signal processing and sparse learning techniques will be developed to model and predict the large-scale dynamics and the statistical distribution of small-scale fading of underwater acoustic environments, including the acoustic transmission loss, ambient soundscape, and statistical characterization of external (anthropogenic and marine animal) acoustic sources. Second, an optimization framework will be developed, based on the acoustic environment prediction, for joint transmission power control, link scheduling, node-cooperative routing, and autonomous vehicle mobility control to achieve high network utility and harmonious coexistence with other acoustic systems. Third, the acoustic environment exploration-exploitation tradeoff will be tackled in the Bayesian reinforcement learning framework, which will provide a principled approach to weighing the immediate reward of a communication and networking strategy and its associated long-term benefit of revealing the environment's dynamics. Leveraging the geographic advantage of Michigan Tech and the state-of-the-art facilities of Michigan Tech's Great Lakes Research Center, extensive field experiments will be conducted for acoustic measurement collection and for offline and online algorithm evaluation in a software-defined networking architecture. The methodologies and crosscutting techniques developed in this project can be applied to the design of intelligent radio-frequency communication networks.