Extremely large-scale multiple-input multiple-output (XL-MIMO) significantly enhances wireless communication capability, yet the inherent strong near-field effects incur prohibitive pilot overhead and severely constrain the exploitable sparsity for channel estimation. Conventional far-field estimators become inadequate in this regime, and existing near-field techniques usually suffer from basis mismatch, high computational complexity, and limited adaptability to complex propagation behaviors. To overcome these challenges, this paper proposes a robust near-field channel estimation framework that integrates model-driven inference, data-driven refinement, and adaptive learning. First, we develop a dual-stage optimization network (DONet) based on approximate message passing (AMP). By jointly learning the sensing and sparse transform matrices in a gridless manner and performing hierarchical-global optimization of inference parameters, DONet achieves enhanced representational capability and high-fidelity channel reconstruction. To further suppress unmodeled residuals, a lightweight residual refinement network (RRNet) is designed as a plug-and-play correction module. RRNet leverages learnable linear projections and a residual-enhanced multilayer perceptron to capture nonlinear deviations while preserving the DONet backbone for interpretability and stable convergence. Finally, a transfer learning-aided network adaptation (TLNA) strategy is introduced to improve reliability under dynamic environments via a pretraining-fine-tuning mechanism, enabling fast domain adaptation with minimal retraining cost. Simulation results demonstrate that the proposed framework achieves substantial performance gains over representative baselines, offering an efficient and robust solution for near-field XL-MIMO channel estimation.

This paper explores the security issues of a rate-splitting multiple access (RSMA)-aided integrated sensing and communication (ISAC) system. A dual-function base station communicates with multiple users via rate-splitting technology and simultaneously senses multiple targets, which also act as colluding malicious eavesdroppers. For the case where coarse target locations are known but subject to estimation errors, we construct an equivalent wiretap channel model incorporating channel uncertainty under the collusion scenario. Based on this model, the transmit covariance matrix is optimized for the sensing-only task, representing an ideal sensing-oriented transmitter design. However, in ISAC systems, the practical transmit covariance also needs to account for communication performance. In order to improve the sensing performance while ensuring communication security, we formulate an optimization problem that jointly optimizes transmit beamforming and rate-splitting to minimize covariance mismatch, subject to secrecy rate and transmit power constraints. This non-convex problem is transformed using convex relaxation techniques and solved via a robust block coordinate descent algorithm. We further consider the case where the locations of malicious sensing targets are unknown, and then formulate an optimization problem to minimize communication power while enhancing sensing capability via an omnidirectional beam, subject to rate constraints. We then develop an efficient block-wise algorithm based on convex reformulation. Simulation results confirm the efficacy of the proposed RSMA scheme in addressing system uncertainties, as well as reveal influences of various key parameters.

This paper develops a simultaneously transmitting and reflecting reconfigurable intelligent surface (STAR-RIS)-assisted symbiotic radio (SR) system, in which the STAR-RIS is deployed to transmit extra Internet of Things (IoT) data and simultaneously enhance the downlink transmission. A simple and efficient ON-OFF keying modulation scheme is applied by the STAR-RIS to modulate IoT data, which allows a low-complexity IoT transceiver and avoids the signal ambiguity. This work aims to maximize the weighted sum-rate (WSR) of downlink users, subject to the minimum received energy requirement of IoT transmission. Under the assumption of perfect channel state information (CSI), an efficient penalty dual decomposition (PDD)-based algorithm is proposed to solve the WSR maximization problem. By leveraging the PDD framework, the STAR-RIS’s coefficients are updated with close-form expressions. For the imperfect CSI case, the WSR maximization problem becomes a challenging stochastic optimization task. To address it, the constrained stochastic successive convex approximation framework is employed. Additionally, an efficient projection method is proposed to handle the STAR-RIS’s amplitude and coupled phase-shift constraints. Simulation results reveal the performance trade-off between the downlink transmission and the IoT transmission and validate the superiority of the proposed algorithms over the benchmarks.

Reconfigurable intelligent surface (RIS) is able to create favorable reflecting channels for different users and piggyback additional data in the reflected signals. The former brings benefits to non-orthogonal multiple access (NOMA), while the latter enables a mechanism of symbiotic radio (SR). Inspired by these unique advantages, we consider a general SR-NOMA system model where an RIS is deployed to assist both the NOMA in an uplink multi-channel system and the Internet-of-Things (IoT) data transmission. This general model also allows for different performance objectives from the NOMA users. In particular, the users can be either energy-efficiency oriented or spectrum-efficiency oriented. To strike the performance trade-off between these two types of users, a performance metric called resource efficiency (RE) is leveraged to formulate the optimization problem. We jointly design the time-frequency resource allocation, multi-user power control and RIS phase shifts to maximize the weighted sum-RE of the system, subject to the quality-of-service constraints of the SR-NOMA system. An efficient alternating optimization framework with a series of algorithms, including matching theory, fractional programming method, and inner majorization-minimization method, is developed to solve this highly complex and non-convex problem.