In-situ manipulation of wireless link with reinforcement-learning-driven programmable metasurface in indoor environment

It is of great importance to control flexibly wireless links in the modern society, especially with the advent of the Internet of Things (IoT), fifth-generation communication (5G), and beyond. Recently, we have witnessed that programmable metasurface (PM) or reconfigurable intelligent surface (RIS) has become a key enabling technology for manipulating flexibly the wireless link; however, one fundamental but challenging issue is to online design the PM's control sequence in a complicated wireless environment, such as the real-world indoor environment. Here, we propose a reinforcement learning (RL) approach to online control of the PM and thus in-situ improve the quality of the underline wireless link. We designed an inexpensive one-bit PM working at around 2.442 ​GHz and developed associated RL algorithms, and demonstrated experimentally that it is capable of enhancing the quality of commodity wireless link by a factor of about 10 ​dB and beyond in multiple scenarios, even if the wireless transmitter is in the glancing angle of the PM in the real-world indoor environment. Moreover, we also prove that our RL algorithm can be extended to improve the wireless signals of receivers in dual-receiver scenario. We faithfully expect that the presented technique could hold important potentials in future wireless communication, smart homes, and many other fields.