A Refined Alternating Optimization for Sum Rate Maximization in SIM-Aided Multiuser MISO Systems

TL;DR

This paper proposes a refined optimization method for SIM-aided multiuser MISO systems that significantly improves sum rate performance and overcomes the limitations of existing alternating optimization approaches.

Contribution

The paper introduces a novel optimization strategy that prioritizes SIM phase shift design and employs iterative projected gradient methods, leading to substantial performance gains.

Findings

Achieves up to 115.53% higher sum rate than benchmarks.

Optimizing SIM phase shifts first improves performance.

Iterative PG methods outperform single-step PG in optimization.

Abstract

Stacked intelligent metasurfaces (SIMs) have emerged as a disruptive technology for future wireless networks. To investigate their capabilities, we study the sum rate maximization problem in an SIM-based multiuser (MU) multiple-input single-output (MISO) downlink system. A vast majority of pioneer studies, if not all, address this fundamental problem using the prevailing alternating optimization (AO) framework, where the digital beamforming (DB) and SIM phase shifts are optimized alternately. However, many of these approaches suffer from suboptimal performance, quickly leading to performance saturation, when the number of SIM layers increases assuming the \emph{fixed SIM thickness}. In this letter, we demonstrate that significant performance gains can still be achieved, and such saturation does not occur with the proposed method in the considered setting. To this end, we provide practical design guidelines to improve AO-based optimization of digital precoders and SIM phase shifts. Specifically, we show that (i) optimizing the SIM phase shifts first yields significant performance improvements, compared to optimizing the DB first; and (ii) when applying projected gradient (PG) methods, which are gradually becoming more popular to optimize the phase shifts thanks to their scalability, we find that using an iterative PG method achieves better performance than the single PG step, which is commonly used in existing solutions. Based on these customizations, the proposed method achieves a higher achievable sum rate (ASR) of up to $\ensuremath{115.53\%}$, compared to benchmark schemes for the scenarios under consideration.