Transforming wireless systems with large AI models : toward intelligent and adaptive connectivity

Abstract

The rapid evolution of wireless communication technologies demands intelligent and adaptive solutions to address the growing complexity of cross-technology interoperability and large-scale connectivity management. Large AI models, with their ability to process vast amounts of data and learn intricate patterns, offer a promising approach to two key challenges: seamless cross-technology interoperability and reliable connectivity management in mobile IoT networks. The primary aim of this research is to harness these capabilities to create frameworks that enhance the efficiency and scalability of wireless systems. This dissertation presents two fundamental contributions: one for the cross-technology communication and the other for the network connectivity optimization. Previous reverse engineering approaches in cross-technology communication include non-reversibility, symbol misalignment, scalability issues, and over-reliance on empirical methods, making them labor-intensive and less effective for modern applications. To address these issues, we propose XiTuXi, which enables seamless cross-technology communication between WiFi and diverse IoT protocols (e.g., ZigBee, Bluetooth, LoRa) via neural machine translation. By employing Transformer models and genetic algorithm-driven forward engineering, XiTuXi automates the translation of bit sequences across heterogeneous protocols, eliminating reliance on manual reverse engineering. In the context of network connectivity optimization, we concentrate on reducing disconnection risks within large-scale mobile IoT networks that utilize eSIM technology. To achieve this, we introduce NeSIM, which employs a lookahead multi-operator access strategy to predict operator switch events in advance, thereby optimizing proactive profile provisioning and minimizing handover latency.