Networked Intelligent Reflecting Surfaces for Communications, Sensing, and Learning in 6G Wireless Systems

In the past decades, the expeditious development of wireless networks, culminating in the advent of 5G by 2020, has profoundly reshaped our daily lives. Qualcomm anticipated that 5G will contribute up to $13.2 trillion to the global digital economy by 2035. This remarkable success has ignited the exploration of 6G wireless networks in both academia and industry. To shore a plethora of thrilling applications, e.g., the Internet of Everything, big data analytics, and augmented reality, in addition to communications, 6G is expected to expand its capabilities to encompass sensing and learning as services. Meeting these ambitious requirements necessitates the development of evolutionary and transformative technologies for the evolution of versatile 6G networks.  Recent advancements in reconfigurable metamaterials have introduced intelligent reflecting surfaces (IRSs) as a cornerstone of future wireless systems, due to their extraordinary abilities to customize smart radio environments. IRSs in the form of artificial thin films can be seamlessly attached to existing infrastructures, which tremendously reduces the operational expenditure and implementation complexity, facilitating the widespread deployment of networked IRSs. However, deploying networked IRSs is not merely an increase in the quantity of IRSs, but more importantly, a qualitative paradigm shift in modern wireless systems. In particular, networked IRSs offer unique opportunities for distinct application scenarios in future 6G networks, spanning communications, sensing, and learning, while simultaneously presenting new challenges. First, networked IRSs are able to bypass blockages and provide virtual line-of-sight links for wireless communications via beam routing. Nevertheless, the challenge lies in scaling up this technique efficiently. To circumvent this difficulty, we shall leverage the high scalability of graph neural networks to develop efficient network-level beam routing methods. Second, for wireless sensing applications, networked IRSs enable observing targets from various perspectives without installing additional power-intensive access points, leading to networked sensing. However, how to coherently exploit the sensing information from networked IRSs is paramount. In this regard, we will compare the sensing performance with active and passive IRSs, respectively, and provide valuable system insights into networked IRS-enabled sensing systems. Last, in distributed learning systems reliant on wireless networks, e.g., federated learning, networked IRSs create the possibility to improve learning performance by equipping edge users with dedicated IRSs. To fully unleash their potential, our research will focus on designing communication-learning frameworks with networked IRSs, including convergence analysis and optimizing communication-computation resource allocation.