Design and implementation of reconfigurable environments towards AI-driven integrated sensing and communication

Abstract

Integrated Sensing and Communication (ISAC) has pivotal as a cornerstone technology for future wireless systems, perfectly positioned to integrate sensing and communication functionalities using the same frequency bands and hardware components. Concurrently, smart reconfigurable environments have become crucial in the evolution of contemporary wireless networks, which employ advanced technologies to manipulate the propagation of signals across spatial and temporal dimensions, significantly enhancing network capabilities. This dissertation explores deploying smart reconfigurable environments to augment communication and sensing performance within wireless systems. The challenges and opportunities addressed in this research focus on two distinct applications for reconfigurable environments enhanced ISAC systems: enhancing system efficiency and improving security. The primary objective of this research is to bridge the gap between theoretical advancements and practical applications by developing innovative algorithms that elevate the performance of ISAC systems, thereby ensuring their effectiveness in real-world settings. In this thesis, we present three fundamental contributions: two aimed at enhancing efficiency and one at improving authentication security. Previous ISAC systems have often been hindered by low energy and communication efficiency. To address these issues, we propose two new methods. The first method, MetaMosaic, constructs Reconfigurable Intelligent Surfaces (RIS) by utilizing RFID tags repurposed as unit cells, significantly boosting signal strength and reducing communication latency. MetaMosaic offers a scalable and cost-effective solution for dynamically optimizing wireless environments, thus improving communication efficiency via a bespoke neural radiance field. Second, we proposed the Radio-Frequency Neural Network (RFNN), which incorporates machine learning directly within wireless sensors. By processing data at light speed close to the source, RFNN drastically reduces both time and energy consumption, minimizing the need for extensive data transmission. It maintains high accuracy across various wireless sensing tasks while adhering to the stringent power limits typical of Artificial Intelligence of Things (AIoT) environments. Regarding security enhancement, we introduce Voltmark, which leverages ambient AC light flickering to enhance security and traceability. This approach employs natural watermarks produced by ambient lighting, detectable through standard smartphone cameras, to authenticate the spatial and temporal contexts of transactions without additional hardware. Seamlessly integrating into existing lighting infrastructures, Voltmark exemplifies the potential of ISAC with optical wireless (ISAC-OW), a pivotal component of 6G. This integration not only converts every lamp into a sensing element but also significantly enhances the security framework, boosting the robustness and reliability of the emerging network architectures in the 6G era. This thesis explores the physical properties of environments, combines hardware and software innovations, and introduces novel algorithms drawn from principles in wireless sensing, signal processing, and machine learning. Through various experiments, these systems are implemented and evaluated, demonstrating their capability to support many real-world applications, including behavior and material sensing, signal enhancement, and security enhancement.