2025_programme: A Lightweight AI-powered Framework for Multimodal Underwater Networks

• Day: June 19, Thursday
    Location / Time: D. CHLOE at 18:20-18:40
• Last minutes changes: -
• Session: 21. Underwater Communications and Networking
  Organiser(s): Charalampos Tsimenidis, Paul Mitchell, Konstantinos Pelekanakis
  Chairperson(s): Charalampos Tsimenidis, Paul Mitchell, Konstantinos Pelekanakis
• Lecture: A Lightweight AI-powered Framework for Multimodal Underwater Networks [Invited]
  Paper ID: 2248
  Author(s): Filippo Donegà, Filippo Bragato, Filippo Campagnaro, Michele Zorzi
  Presenter: Filippo Donegà
  Abstract: Underwater acoustic communication enables numerous long-range applications, but the low bitrate and long propagation delay that characterize the underwater acoustic channel make it difficult to support demanding applications such as real-time video transmission, which is essential, e.g., for piloting Work-Class ROVs. Alternative technologies such as optical (the most promising for underwater video transmission), electromagnetic radio frequency, and magneto inductive communications can provide significantly higher rates at short distances. A single best-overall underwater wireless communication technique does not exist, hence combining multiple communication technologies in the so-called underwater multimodal networks [1] is currently the best strategy for optimizing the communication performance. It is especially important to understand how to combine the transmission technologies optimally, and how to select which one to use depending on the channel conditions and the requirements imposed by the intended application. Traditional strategies select the best mode either by observing the received power or probing the channels periodically, while recent solutions focus on resource-demanding machine learning algorithms for channel selection. In particular, Reinforcement Learning (RL) is a practical way to deal with unknown Markov Decision Problems, and multimodal communication can be easily thought of as one of them. In this work we showcase how Artificial Intelligence can be used to optimize multimodal communication by presenting and evaluating through DESERT [2] simulations a lightweight QoS-based RL framework that optimally selects the best transmission medium in multimodal underwater networks. Results highlight how this framework outperforms traditional approaches, achieving better performance in many communication parameters while still being suitable for practical deployments onboard lightweight single-board computers.\n[1] Zhilin, Igor V., et al. "A universal multimode (acoustic, magnetic induction, optical, rf) software defined modem architecture for underwater communication." IEEE TWC 2023.\n[2] Campagnaro, Filippo, et al. "The DESERT underwater framework v2: Improved capabilities and extension tools." IEEE UComms 2016.
    Download the full paper
  This paper is a candidate for the "Prof. John Papadakis award for the best paper presented by a young acoustician(under 40)"
• Corresponding author: Mr Filippo Donegà
  Affiliation: Università degli Studi di Padova
  Country: Italy