In this work, a 4.54 kW near-single-mode narrow-linewidth fiber amplifier (NLFA) seeded by a conventional fiber Bragg grating (FBG)-based oscillator is demonstrated. A TMI model including the heat load and bending-induced eigenmode variation is established. The TMI characteristics of step-index fiber (SIF) and trench-assisted fiber (TAF) are numerically investigated. The simulation results show that the TAF provides a higher LP11 bending loss and a higher TMI threshold than the SIF. In the experiment, a low-numerical aperture (NA) 25/400 μm trench-assisted Yb-doped fiber is employed in the main amplifier to enhance the bending loss of the LP11 mode and increase the TMI threshold. A maximum output power of 4.54 kW is achieved with an optical efficiency of 78.4%. The 3-dB linewidth is 170 pm, and the Raman suppression ratio is 31 dB. The beam quality factors are Mx2 = 1.34 and My2 = 1.41, indicating near-single-mode operation. These results indicate that low-NA trench-assisted YDFs can effectively balance nonlinear effect suppression and TMI threshold enhancement in high-power NLFAs.
Data-driven control provides a flexible way to address complex vehicle dynamics. However, integrated stability control of four-wheel independent-drive autonomous electric vehicles (4WID-AEVs) remains challenging due to the strong coupling among longitudinal, lateral, and roll dynamics. This study develops a Control Lyapunov Function (CLF)-guided reinforcement learning framework for unified four-wheel torque control. The learned policy coordinates the coupled vehicle dynamics, while CLF information provides stability-oriented guidance during learning. The resulting Vehicle Dynamics Learning for Integrated Stability Control (VDL-ISC) policy directly maps vehicle states to four-wheel torque commands in simulation. Its reward design considers velocity tracking, yaw-roll dynamics, tire-load utilization, and actuator smoothness. For physical deployment, the learned policy is integrated into a Safety-Augmented VDL (SA-VDL) architecture. A deterministic quadratic-programming-based safety/allocation layer enforces prescribed actuator and physical constraints on the policy-generated torque references. Co-simulation results show that VDL-ISC improves the Safety Performance Index by more than 25% relative to direct-yaw-moment MPC(DYC-MPC) under the tested extreme maneuvers. Real-vehicle experiments show real-time operation of the complete SA-VDL architecture, with the Safety Performance Index increasing by 13.6% and 16.9% relative to the factory onboard strategy in the U-turn and double-lane-change tests, respectively.
With the rapid development of internet and communication technologies, telecom fraud has become increasingly frequent, posing a serious threat to users’ financial security. Traditional detection methods based on rules or shallow models often suffer from low accuracy and poor generalization when facing complex and evolving fraud patterns. This paper addresses the problem of telecom fraud detection by constructing a relation-oriented detection framework using real call detail records. By incorporating advanced graph neural network models and accounting for class imbalance through cost-sensitive learning, the proposed method significantly improves the identification of fraudulent numbers. Experimental results demonstrate outstanding performance in key metrics such as accuracy, recall, and F1-score, highlighting its practical value and potential for wide application.
Conventional Gaussian Restricted Boltzmann Machines (GRBMs) and Discriminative Restricted Boltzmann Machines (DRBMs) process continuous data using deterministic parameters, making it difficult to capture the parameter uncertainty commonly encountered in practical applications. Existing fuzzy RBM models based on type-1 fuzzy sets enhance robustness to some extent, but they treat the membership functions as precisely defined, so the uncertainty inherent in the membership functions themselves remains unrepresented. To address this, this study proposes Interval Type-2 Gaussian Fuzzy Restricted Boltzmann Machine (IT2GFRBM) and Interval Type-2 Discriminative Fuzzy Restricted Boltzmann Machine (IT2DFRBM), which extend the model parameters to interval type-2 symmetric triangular fuzzy numbers (IT2STFNs). This study extends the crisp possibilistic mean defuzzification method, derives the approximate free energy function along with a sharp, computable bound on its approximation error and an exact error ledger, and presents a training algorithm based on Contrastive Divergence (CD). Experimental results on the MNIST, Fashion-MNIST, CIFAR-10, and BloodMNIST datasets demonstrate that the proposed models outperform GRBM, DRBM, their fuzzy variants, and several baseline methods in terms of reconstruction quality, classification accuracy, and noise robustness, with one MNIST comparison excepted. These results validate the effectiveness of the interval type-2 fuzzy mechanism in capturing the uncertainty inherent in the membership functions.
Unmanned Aerial Vehicles (UAVs) have shown great application potential in the low-altitude economy, intelligent logistics, and emergency communication. In this paper, we first replace traditional ground APs with UAVs as aerial access points in a Cell-Free massive MIMO (CF-mMIMO) system. Subsequently, some UAVs are equipped with Simultaneously Transmitting and Reflecting Reconfigurable Intelligent Surfaces (STAR-RISs) to form a STAR-RIS-UAV-CF-mMIMO architecture, achieving full-domain coverage and efficient resource coordination. Different from existing research, in this paper, we simulate user heterogeneity by using different numbers of antennas, and then consider the differentiated Quality of Service (QoS) requirements of heterogeneous users in the CF-mMIMO system. Based on this, we propose a DDPG-based Resource Allocation for STAR-UAV-RIS-CF-mMIMO System (DRA-SUCS) scheme. This scheme aims to optimize the position coordinates of UAVs, the phase shift matrix and reflection amplitude coefficients of STAR-RISs, as well as the transmit and receive beamforming matrices, so as to maximize the information transmission rate in this system. Simulation results show that this scheme significantly improves the total information transmission rate of the STAR-RIS-UAV-CF-mMIMO communication system through effective resource allocation.