Physically consistent modeling and optimization of stacked intelligent metasurfaces for wideband multi-user communication
Year: 2026
Type: article
Abstract: Stacked intelligent metasurfaces (SIMs) have emerged as an alternative to single-layer reconfigurable intelligent surfaces (RISs). A SIM consists of multiple programmable metasurface layers placed in front of the base station (BS) antennas and performs beam- forming in the electromagnetic wave domain. Its wideband operation introduces several challenges. Closely spaced elements and layers are coupled, the resulting response is frequency dependent, and a common phase configuration must serve multiple users and subbands. This thesis develops circuit-theoretic models and optimization methods for SIM-assisted wideband and multi-band multi-user downlink transmission. First, we develop a coupling-aware circuit-theoretic model for a SIM-assisted wide- band multi-user multiple-input multiple-output (MIMO) system. The model accounts for frequency-dependent mutual coupling, inter-layer reflections, and frequency-selective channels across subbands. Based on this model, the BS precoder and SIM phase shifts are jointly optimized, with power allocation across subbands providing carrier aggrega- tion. We also investigate a partially reconfigurable SIM in which only the last layer is updated while the first layer remains fixed, reducing the control overhead. For a fixed total number of elements, the comparison between SIM and single-layer RIS shows that their relative performance depends on the operating conditions. The RIS performs bet- ter in narrowband and low signal-to-noise ratio (SNR) regimes, with fewer users, and when power allocation is performed at the BS. In contrast, the SIM provides higher spectral efficiency in wideband and high-SNR regimes with more users. Moreover, ac- counting for mutual coupling improves the modeled spectral efficiency, while partial reconfiguration can achieve near-optimal performance with fewer control bits. We then consider a multi-band multi-user multiple-input single-output (MU-MISO) downlink with inter-band carrier aggregation, where one SIM configuration serves a low band and a high band with different bandwidths and noise levels. Since conventional phase updates need many iterations, we propose a multi-band deep-unfolding network (MBDU-Net), which unrolls projected-gradient phase updates into a fixed number of stages and learns band-specific step sizes and momentum parameters. For the same number of iterations, MBDU-Net achieves a higher sum rate than projected gradient ascent and basic deep unfolding, and generalizes to numbers of subcarriers not seen during training.
Source: Mspace (University of Manitoba)
Author: Ibrahim, Muhammad
Institution: University of Manitoba
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7223163602