This paper investigates a dual-robust beamforming design for covert satellite communication, where covert transmission is embedded within an overt channel. While prior studies on covert communication over overt channels typically address robustness for covert user under wiretap channel (WC) uncertainty, achieving simultaneous robustness for both overt and covert users remains challenging in satellite-to-ground links with uncertain channel state information (CSI). We propose a dual-robust beamforming framework that jointly designs overt and covert beamformers to maximize the minimum covert rate, subject to the overt user’s rate outage constraints (ROCs) and covertness constraints. The resulting optimization problem is intrinsically nonconvex due to CSI and WC uncertainties, variable coupling, and the absence of closed-form expressions for the ROCs. To solve the problem, we reformulate the problem into a sequence of convex subproblems by utilizing semidefinite relaxation, the S-procedure, large deviation inequality, and fractional programming. We then develop a dual-robust beamforming algorithm (DRBA) that yields near-optimal beamforming vectors. Numerical results demonstrate that the proposed approach achieves superior covert communication performance while ensuring robust service for both overt and covert users, outperforming benchmark schemes.

This paper addresses the critical challenges of fractional delay and Doppler effects in low earth orbit (LEO) satellite orthogonal time-frequency space (OTFS) systems, where conventional channel estimation methods suffer from high complexity and inadequate prior utilization. To overcome these limitations, we propose a fast and robust framework with: 1) A subspace-constrained variational Bayesian inference (SC-VBI) mechanism that reduces complexity by decoupling high-dimensional matrix operations and 2) A statistical Delay-Doppler-prior (SD2P) integration scheme leveraging elevation angle distributions from real-world LEO orbital dynamics to enhance sparse signal recovery (SSR) accuracy. Departing from traditional sparse Bayesian learning (SBL) approaches, our algorithm mitigates the curse of dimensionality through iterative sub-space updates while embedding physical-layer Doppler characteristics into Bayesian priors. Simulations under 3GPP LEO configurations demonstrate that the proposed method achieves a ten-times compression in computational time and 60% normalized mean square error (NMSE) improvement.