Internet of Underwater Things — Cognitive Networking

Overview

My doctoral and ongoing research at the James Watt School of Engineering, University of Glasgow focuses on the Internet of Underwater Things (IoUT) — one of the most demanding application areas in wireless engineering due to harsh acoustic channels, low bandwidth, high latency, and severe energy constraints.

Key Engineering Contributions

DEKCS — Dynamic Clustering Protocol

A clustering and routing protocol for underwater wireless sensor networks that selects cluster heads dynamically based on residual energy and distance. Extends network lifetime by more than 70% compared to standard protocols.

Published in IEEE Sensors Journal (2021); cited 109+ times.
The most-downloaded article in its IEEE Xplore category, Q1 2022.
Led to an invited presentation at the IEEE Sensors Conference, Dallas.

RL-SWIPT — Sustainable Underwater Charging

A reinforcement-learning approach for simultaneous wireless information and power transfer (SWIPT) using AUVs. Models the joint throughput / harvested-power problem as an MDP — achieves up to 207% improvement in energy efficiency vs random-trajectory baselines.

Published in IEEE Internet of Things Journal (2024).
First RL-based approach for this application.

Q-Learning Route Selection

Models routing as a reinforcement-learning problem prioritising energy efficiency and packet delivery — yielding orders-of-magnitude energy savings vs direct transmission.

Published in IEEE WCNC.

Current Manuscripts

Hierarchical Federated Anomaly Detection (IEEE IoT Journal, under review)

Preprint: arXiv:2603.24648

A three-tier federated learning framework for IoUT anomaly detection. Feasibility-aware sensor-to-fog association, compressed model updates, and selective cooperative aggregation.

31–33% energy reduction vs always-on inter-fog exchange.
71–95% total energy reduction with compressed uploads.

Transforms AUVs from passive recorders into intelligent edge nodes:

FathomNet-trained YOLOv8 for semantic detection.
64-byte semantic tokens replace raw images (99.9%+ bandwidth savings).
Persistent underwater structures act as visual anchors for drift correction.

ML for IoUT — Survey & Tutorial (IEEE COMST, revised and resubmitted)

Preprint: arXiv:2603.07413

A comprehensive tutorial-survey synthesising ~300 papers (2012–2025) on machine learning across the IoUT protocol stack — from physical layer through application layer — with a technology roadmap to 2035+.

Technologies Used:

Python
PyTorch
Reinforcement Learning
Federated Learning
YOLOv8