Meta Reinforcement Learning-Based Spectrum Sharing Between RIS-Assisted Cellular Communications and MIMO Radar
Prajwalita Saikia, Keshav Singh, Omid Taghizadeh, Wan-Jen Huang, Sudip Biswas
Abstract
New wireless networks together with fixed spectrum allocation have resulted in spectrum paucity, which has led to the idea of spectrum sharing between radar and communication systems. In this work, we consider a spectrum-sharing framework between a reconfigurable intelligent surface (RIS)-assisted cellular system and a multiple-input multiple-output (MIMO) radar and investigate its performance. In particular, we formulate an optimization problem to jointly optimize the communication transmit precoder matrix, RIS phase shift matrix, and transmit waveform of radar while maintaining the operational fairness of the proposed system, including the limitation on permissible interference towards the radar system. Thereafter, to address the non-convexity of the problem, we propose a low-complexity meta-reinforcement learning (MRL) algorithm that solves the problem in continuous action spaces by reducing the overall training overhead. Exhaustive simulation results are presented that demonstrate the benefit of using the MRL algorithm for the proposed spectrum-sharing framework along with the utility of the deployment of RIS in terms of controlling the interference from the base station to the radar. It is also shown that the proposed MRL technique outperforms traditional block coordinate descent (BCD)-based solutions, meta-heuristic approaches and other reinforcement learning (RL) algorithms such as twin delayed deep deterministic (TD3) and deep deterministic policy gradient (DDPG).