Channel capacity optimization based on Riemannian trust region algorithm in IRS-aided communication system

Abstract

Intelligent Reflecting Surface (IRS) can offer unprecedented channel capacity gains since it can reconfigure the signal propagation environment. We decide to maximize the channel capacity by jointly optimizing the transmit-power-constrained precoding matrix at the base station and the unit-modulus-constrained phase shift vector at the IRS in IRS-assisted multi-user downlink communication. We first convert the resulting non-convex problem into an equivalent problem, then use the alternate optimization algorithm. While fixing the phase shift vector, we can obtain the optimal precoding matrix directly by adopting standard optimization packages. While fixing the precoding matrix, we propose the Riemannian Trust-Region (RTR) algorithm to solve this optimization problem. And the key of the RTR algorithm is the solution of the trust-region sub-problem. We first adopt the accurate solution based on Newton's (ASNT) method to solve this sub-problem, which can obtain the global solution but cannot guarantee that the solution is optimal since the initial iteration point is difficult to choose. Then, we propose the Improved-Polyline (IPL) method, which can avoid the difficulty of the ASNT method and improve convergence speed and calculation efficiency. The numerical results show that the RTR algorithm has more significant performance gains and faster convergence speed compared with the existing approaches.