Transmission Design for IRS-Aided MIMO Cognitive Radio Systems With Finite Alphabet Inputs and Imperfect CSI

Abstract

Intelligent reflecting surface (IRS)-aided cognitive radio (CR) is an effective solution for increasing the spectral efficiency of wireless communications. The transmission design for IRS-aided CR systems has received significant attention. However, existing designs are based on the assumption of ideal Gaussian signals, which may not accurately reflect real-world scenarios. This paper aims to address this issue by studying a practical scenario where the inputs are finite alphabet signals, such as quadrature amplitude modulation (QAM). By jointly optimizing the precoding matrix at the secondary-user transmitter (ST) and the phase shifts of the IRS, the goal is to maximize the mutual information between the ST and the secondary-user receiver (SR), while considering the transmit power constraint on the ST and the interference power constraints on the primary-user receivers (PRs). Two imperfect channel scenarios are investigated: one with statistical channel state information (CSI) errors only in the reflected channels, and the other with bounded CSI errors in all channels. Despite the challenges posed by variable coupling and nonconvexity, this paper proposes corresponding algorithms for solving the formulated problems. Specifically, for the former scenario, a stochastic successive convex approximation (SSCA) based algorithm is proposed to maximize the expected mutual information, while for the latter scenario, an algorithm combining the successive convex approximation (SCA) and the semidefinite relaxation (SDR) is developed to maximize the mutual information in the worst case. Simulation results demonstrate the effectiveness of the proposed algorithms.