AquaIntellect: A Semantic Self-learning Framework for Underwater Internet of Things Connectivity

TL;DR

An intelligent sensing strategy for exploring the semantics of the underwater environment by judiciously selecting the data to transmit, thereby minimizing redundancy for utmost relevant data transmission is presented.

Abstract

The emerging paradigm of Non-Conventional Internet of Things (NC IoT), which is focused on the usefulness of information as opposed to the notion of high volume data collection and transmission, will be an important and dominant part of human life in the near future. This paper proposes a novel semantic-based approach for addressing the unique challenges posed by underwater NC IoT. We present an intelligent sensing strategy for exploring the semantics of the underwater environment by judiciously selecting the data to transmit, thereby minimizing redundancy for utmost relevant data transmission. We introduce an evolutionary function for the selection of the semantic-empowered messages relevant to the specific task within a minimum Age of Information (AoI), a freshness metric of the collected information, and for monitoring the underwater environment for performance optimization. A DNN-empowered Bayesian integrated with an adaptive surrogate model optimization will determine the optimal placement strategy of the sensors and the uncertainty level of the underwater landscape. An Adaptive Expected Improvement (AEI) mechanism is introduced to predict the optimal arrival rate for enabling a synchronized data sensing and transmission ecosystem, ensuring efficiency and timeliness. Simulation results show that the proposed solution outperforms conventional approaches.

Figures (4)

• Figure 1: The proposed terrestrial and underwater cohesive NC IoT, introducing smart buoys, called Smart Semantic Gateways (SSGs). A semantics-driven approach interprets and prioritizes the information based on relevance and significance in the presence of uncertainties.
• Figure 2: (left) The optimal number of sensors from DNN-based Bayesian Optimization for a search space of huge uncertainty; (center) The distance between two sensors for an optimal sensor placement with maximum sensing information; (right) Visualization of $E(X)$ values across varying wakeup probabilities for a set of sensors.
• Figure 3: Comparative analysis of delays over time for our proposed approach with random and non-adaptive placement strategies.
• Figure 4: (left) The acquisition function from Bayesian Optimization (BO) for the optimal number of sensor selections with their respective distance $d$; (center) The behavior of the surrogate modeling-based BO of $r$ to determine the optimal values of $k$; (right) The comparative efficiency of the AEI method over the EI approach in achieving optimal arrival rates for underwater IoT subnets.