Machine Learning Enables Uplink Interference Suppression in RIS-Aided Proactive Mobile Network

The proactive mobile network (PMN) is an innovative architecture to support minimal latency communications. It adopts an open-loop communication mode, which cancels all real-time control feedback processes and utilizes virtual cells to allocate radio resources to users. However, this design also results in significant potential mutual interference and deteriorates the communication reliability. To address this issue, we propose introducing multi-reconfigurable intelligent surface (RIS) technology, aiming to enhance the PMN's resilience to interference during the more challenging uplink process. Unlike conventional networks, the PMN faces not only the lack of real-time channel state information (CSI), but also the unknown transmission requirements of users. Therefore, managing RISs to serve PMN is an arduous task. We propose a scheme that trains the long short-term memory network (LSTM) using joint CSI, which is iteratively updated with an asynchronous advantage actor-critic (A3C)-based method over time scales. This enables the acquisition of predicted CSI, thus achieving accurate control of RIS to suppress interference. The simulation results highlight the effectiveness of the proposed scheme, demonstrating an improvement of over 21% in the probability of successful physical connections between users and access points (APs), compared to schemes based on outdated CSI and time-correlated CSI.