Beam Tracking for SIM-Assisted LEO Satellites: From Orbital Prediction to Wave-Domain Mobility Management

Abstract

Low-Earth-orbit (LEO) constellations are becoming a key component of global broadband, direct-to-device, and machine-type connectivity, but their high mobility turns beam maintenance into a continuous control problem. Stacked intelligent metasurfaces (SIMs) can reduce radio-frequency chains and onboard digital processing by beamforming directly in the wave domain; however, their multilayer electromagnetic coupling makes rapid reconfiguration fundamentally different from phased-array steering. This article offers a system-level perspective on SIM beam tracking for LEO satellite communications. Building on a recently proposed one-shot geometry-aided tracker, we explain how predicted orbital geometry is turned into multilayer phase updates, present a unified predict--correct architecture, and compare four tracking paradigms---CSI-driven reactive, geometry-aided predictive, sensing-assisted, and hybrid physics/learning---in terms of the information, signaling, and computation they require and the errors they can correct; the latter two are discussed as research directions. A case study with a perturbed orbit then quantifies when prediction alone suffices: one-shot tracking updated every second retains about 85--91\% of the rate of per-second full reoptimization with roughly 35--54 times less computation per update; adding a predictive full reoptimization only every $20$~s raises this to about 90--95\%; and one scalar SNR report per user and update limits the loss caused by a $2^\circ$ attitude bias to a few percentage points, whereas open-loop prediction loses up to 21 points. A roadmap toward autonomous SIM payloads concludes the article.