This paper proposes a precisely optimized sub-connected architecture to address the high power consumption and feedback-type self-interference issues in active fully-connected reconfigurable intelligent surface(RIS) systems. In this architecture, multiple RIS elements share power amplifiers, and beamforming design is carried out using techniques such as fractional programming and alternating direction method of multipliers. These can reduce the number of power amplifiers and lower power consumption. Further, a self-interference suppression mechanism is incorporated, and the optimization problem is subsequently solved using the sequential unconstrained minimization technique, which improves the system's stability and signal quality. Simulations show a 25.0%-33.3% energy efficiency gain with a 6.7%-7.4% rate reduction compared to the fully-connected architecture. Compared to existing sub-connected architectures, it achieves an 11.1%-14.2% improvement in energy efficiency and significantly reduces the impact of feedback-type self-interference, thereby validating its advantages in both energy efficiency and interference suppression.

Mobile edge computing (MEC) and reconfigurable intelligent surface (RIS) are two highly promising key technologies in future wireless communications, dedicated to enhancing the computational capabilities of communication systems and optimizing channel transmission performance, respectively. To address the problem of minimizing the task delay in a multi-user non-orthogonal multiple access (NOMA)-MEC network , RIS is deployed to improve the communication performance, enabling multiple users to offload tasks simultaneously through the same frequency band. We jointly optimize the RIS phase shift, offloading power, beamforming, and distribution factors, and design an equalizer at the receiving end. Since the problem is non-convex, the semidefinite relaxation (SDR) algorithm is used to optimize the phase shift of RIS, the successive convex approximation (SCA) algorithm is used for the power and allocation factors, the equalizer is designed by the minimum mean square error (MMSE) criterion, and the delay is minimized by the alternating optimization.

Unlike outdoor environments, indoor spaces exhibit complex electromagnetic conditions that complicate the localization of targets using wireless signals. Based on the Fresnel propagation model, this paper studies the propagation law of Wi-Fi signals in a rich multi-path environment. First, to address the phase offset in channel state information caused by hardware, a correction method based on least squares regression analysis is proposed, and environmental interference signals are eliminated by spectrum analysis. Second, the target position is estimated by using multiple groups of signal links, and the estimated result is matched with the reference position to reconstruct the target motion path. Experimental results show that the theoretical basis of this method is consistent with the actual movement patterns of the target, achieving a positioning error of about 11 cm without data training. At this error level, this method can distinguish adjacent targets in lateral or longitudinal proximity, thereby reducing misidentification. Compared with the existing deep learning-based methods, the proposed method reduces the positioning error by 50.4% with less training data, and by 5% with ample training data.

To address the challenges of environmental sensitivity in communication links and limited terminal access in cell-free communication systems, this paper introduces reconfigurable intelligent surface (RIS) and rate splitting multiple access (RSMA) technologies to assist communication. A sum-rate optimization algorithm is proposed for RIS-aided cell-free uplink communication systems with RSMA. The algorithm aims to maximize the system sum-rate by jointly optimizing RIS passive beamforming, access point selection, power allocation of sub-messages, and decoding order. Given the non-convex nature of this discrete nonlinear programming problem, we decompose it into multiple subproblems and address them by alternatively using an optimization approach. First, a user-centric matching algorithm based on channel gain is used to obtain the optimal matching scheme. Second, the successive convex approximation (SCA) algorithm is used to solve the sub-message power distribution problem, and the Riemann conjugate gradient algorithm is utilized to solve the RIS passive beamforming problem. Third, a decoding order algorithm is developed based on channel gain and the sub-message allocation ratio. Simulation results demonstrate that the proposed algorithm achieves superior sum-rate performance compared to traditional algorithms.

Existing distributed anomaly detection models based on federated learning can hardly deal with the balance between anomaly detection performance and data privacy protection. In this regard, a federated learning anomaly detection model is proposed based on adaptive noise and hybrid attention mechanism. First, built on the convolutional neural network, this model integrates spatial and multi-head hybrid attention mechanisms to extract complex features in a multidimensional and deep manner, enabling high-precision anomaly detection. Second, based on both local and centralized differential privacy, this model utilizes the adaptive noise and the privacy budget allocation to further improve the privacy and robustness. Validated experiments are exerted on public datasets NSL-KDD and UNSW-NB15. The results show that compared with the existing mainstream approaches, the proposed model can achieve higher-quality anomaly detection while ensuring user data privacy.

Reconfigurable refractive surface (RRS) is regarded as a prospective technology for optimizing signal paths and enhancing directional propagation in complex environments, while non-orthogonal multiple access (NOMA) serves as an empowering technology to meet the high spectrum efficiency and large capacity requirements of the next-generation cellular communications. This paper investigates the problem of maximizing the sum rate of the downlink NOMA system assisted by RRS deployed in the near field of the base station (BS). The sum rate is maximized by alternately optimizing the power distribution and the phase shift matrix of the RRS. The Lagrange dual decomposition method is used to optimize the power allocation and then a phase shift rotation in turn (PSRT) method is proposed to optimize the phase shifts of RRS. Simulation results show that the sum rate of RRS-assisted NOMA systems is higher than that of RRS-assisted orthogonal multiple access (OMA) systems, NOMA systems and OMA systems.

The covert communication system investigated in this paper employs reconfigurable intelligent surface(RIS) and non-orthogonal multiple access(NOMA) technologies. A full-duplex transceiver is used at the public user to introduce uncertainty from friendly interference, thereby enhancing covertness. A covert performance optimization scheme is proposed, where closed-form expressions for the eavesdropper's detection error probability and the effective covert transmission rate are derived. The scheme intends to maximize the system's covert performance through jointly optimizing the covert information transmission probability and the power range of the friendly interference noise. A step-by-step optimization method is employed to obtain the joint optimal design of the system parameters. Numerical simulation results demonstrate that the proposed scheme can achieve covert information transmission , and the optimal solutions for the optimization variables can serve as a guidance for parameter design in practical scenarios.

Significant advancements have been made in image deblurring through multi-layer networks, but their performance remains limited by challenges in feature extraction and residual connections. To address these issues, this paper proposes a multi-scale feature extraction and fusion network (MSFN) for image deblurring. The core idea of the network is to enhance image feature extraction through multi-scale inputs and outputs. Further, MSFN utilizes its feature adaptive detail enhancement (ADE) modules and cross-scale feature fusion (CSFF) modules to capture multi-scale features at different network depths, thereby optimizing the residual connection process and effectively integrating multi-scale information. Experimental results demonstrate that the proposed algorithm achieves superiority in quantitative analysis and significantly improves subjective visual effects, exhibiting an advanced performance.

The study investigates the bipartite consensus of linear multi-agent systems under sequential scaling attacks based on observers. First, since the attack degree of the sequential scaling attack is unknown, observers are designed to estimate the real states of the agents to reduce the impact on the stability of the error system. Second, the attack signals are injected into all controller-to-actuator channels throughout the network. and a distributed controller based on observer's state-estimates is proposed for the linear multi-agent system, which can effectively avoid the use of information from the system itself. Finally, the Lyapunov function proves that the designed controller can enable the linear multi-agent system subject to sequential scaling attacks to achieve bipartite consensus. And numerical simulations validate the theoretical results.

In response to the scarcity of annotated medical image data and the imitations of existing models in segmenting multi-scale target images, this paper proposes a few-shot medical image segmentation method based on multi-scale feature fusion and contrastive learning. First, a sequential concatenation-based multi-scale skip connection method is introduced to replace traditional skip connections, enabling effective fusion of multi-scale feature maps from the encoder and their transmission to the corresponding decoder. Second, considering the dual-branch structure of the model, a contrastive learning module based on multi-scale features is proposed, and a loss function is designed to enhance the model's discriminative ability at the pixel level. Experiments show that our method achieves cross-domain data segmentation for medical images, mitigates performance degradation due to dataset scarcity, and improves the segmentation accuracy and generalization for different-scale targets , outperforming current mainstream few-shot medical image segmentation methods.