Machine Learning for Underwater Acoustic Communications

Abstract

Energy-efficient and link-reliable underwater acoustic communication (UAC) systems are of vital importance to both marine scientific research and oceanic resource exploration. However, owing to the unique characteristics of marine environments, underwater acoustic (UWA) propagation experiences arguably the harshest wireless channels in nature. As a result, traditional model-based approaches to communication system design and implementation may no longer be effective or reliable for UAC systems. In this article, we resort to machine learning (ML) techniques to empower UAC with intelligence capabilities, which capitalize on the potential of ML in progressively improving system performance through task-oriented learning from data. We first briefly overview the literature of both UAC and ML. Then, we illustrate promising ML-based solutions for UAC by highlighting one specific niche application of adaptive modulation and coding (AMC). Lastly, we discuss other key open issues and research opportunities layer-by-layer, with focus on providing a concise taxonomy of ML algorithms relevant to UAC networks.