Service Function Chain (SFC) integrates Network Function Virtualization (NFV) for flexible deployment of Virtualized Network Function (VNF) and utilizes Software-Defined Networking (SDN) for establishing dynamic network connections, enabling customized service chains for specific applications. Security risks in NFV/SDN networks directly impact SFC, making them vulnerable to attacks. Traditional security solutions rely on passive defense mechanisms, which are insufficient against the uncertainties brought by vulnerabilities and backdoors in open networks. As an emerging network security technology, Moving Target Defense (MTD) is expected to break through traditional security protection mechanisms such as “enhanced”, “plug-in” and “passive” defense to achieve active defense against both known and unknown threats. Features of NFV/SDN networks, such as dynamic resource scheduling, microservices, and centralized control, also offer critical guidance for implementing MTD strategy. Therefore, in this paper, we propose a novel security enhancement mechanism by equipping SFC and VNFs with MTD to inherently protect SFC and VNFs during the deployment process, using a sub-pool partitioning algorithm to enhance heterogeneity across sub-pools. Additionally, to address Quality of Service (QoS) requirements, we formulate an optimization problem with three criteria: latency, cost, and defense success rate. Following that, we propose a dynamic deployment algorithm based on Deep Reinforcement Learning (DRL). Finally, to validate the effectiveness of the proposed algorithm, we conduct extensive experiment which have demonstrated that our proposed algorithm is good in increasing the security gains of both SFC and VNFs, provided that this improvement comes at the expense of modest cost and latency.

Low-altitude aerial networks play an increasingly critical role in urban logistics, emergency response, and intelligent airspace management. These networks must ensure reliable identity authentication while addressing privacy, efficiency, and security challenges. Traditional authentication mechanisms—such as PKI-based certificates and remote identification—either overexpose identity data or lack support for flexible attribute disclosure. Moreover, existing anonymous credential (AC) schemes, though offering privacy, often incur high computational and communication costs and exist vulnerable to replay attacks. To address these limitations, we propose <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">DVA-ACLF</b> (<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Designated Verifier Asymmetric Anonymous Credential Scheme for Low-Altitude Flight</i>), a lightweight credential system tailored for UAVs. DVA-ACLF leverages an efficient MAC-based construction combined with non-interactive zero-knowledge proofs to enable selective attribute disclosure without relying on pairings. It supports designated verification to prevent credential misuse and replay. Experimental results show that DVA-ACLF reduces credential size by 49% (2720 bits vs. 5369 bits in Idemix) and mitigates the linear growth of credential size with respect to attribute count observed in CL and BBS+ schemes. It requires only four single-base and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$2(k-l)$</tex-math></inline-formula> double-base exponentiations on UAVs, where <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$k$</tex-math></inline-formula> is the number of selected attributes and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$l$</tex-math></inline-formula> is the number of disclosed attributes—both typically small. These features make DVA-ACLF a strong candidate for secure, efficient, and privacy-preserving authentication in low-altitude flight applications.

Graph Neural Networks (GNNs) exploit topological structures—namely, node-to-node connections—to aggregate contextual information, thereby achieving strong performance across diverse domains. Representative models such as Graph Convolutional Networks (GCN) and Graph Attention Networks (GAT) have demonstrated remarkable success in node-level tasks. Nevertheless, two fundamental issues remain: (1) inaccurate topologies directly misguide the neighborhood aggregation process, degrading classification accuracy; (2) the additional structural information introduces redundant or noisy signals, further impairing generalization. To address these issues, we propose a Dual-channel Dual-mode Graph Network (DDGN) that reconstructs a feature graph driven by node-feature similarity. DDGN employs two parallel channels to extract representations from the original topology and the similarity-induced graph, respectively; each channel integrates two complementary feature-extraction schemes and an adaptive switch that dynamically harmonizes their interaction within the channel for richer representation. A lightweight feature-enhancement module and an adaptive weight allocation mechanism are further introduced to synergistically fuse the dual representations. Extensive experiments on multiple public benchmarks show that DDGN consistently outperforms state-of-the-art GNNs in node classification accuracy.

Next generation Radio Access Networks (RANs) introduce programmability, intelligence, and near real-time control through intelligent controllers, enabling enhanced security within the RAN and across broader 5G/6G infrastructures. This paper presents a comprehensive survey highlighting opportunities, challenges, and research gaps for Large Language Model (LLM)-assisted explainable (XAI) Intrusion Detection Systems (IDS) in future RAN environments. Motivated by this, we propose an LLM interpretable anomaly detection system leveraging multivariate time series Key Performance Measures (KPMs), extracted from E2 nodes, within the Near Real-Time RAN Intelligent Controller (Near-RT RIC). A sequence classification model is trained to identify malicious User Equipment (UE) behavior based on these KPMs. To enhance transparency, we apply post-hoc local explainability methods such as LIME and SHAP to interpret individual predictions. Furthermore, LLMs are employed to convert technical explanations into natural-language insights accessible to non-expert users. Experimental results on real 5G network KPMs demonstrate that our framework achieves high detection accuracy (macro F1-score <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$&gt;$</tex-math></inline-formula> 0.96) while delivering actionable and interpretable outputs.

Large language models (LLMs) demonstrate significant potential for enabling intelligent endogenous networks owing to their amazing intelligence level. LLMs are currently deployed on cloud servers as their vast parameter scales introduce a substantial computational burden to pre-training and inference processes. This deployment paradigm faces severe challenges, including insufficient personalization, high inference latency, and privacy concerns, as an increasing number of mobile users enjoy the LLM services. To enhance the personalization of pre-trained LLMs, exploiting massive local datasets distributed across edge devices to fine-tune LLMs is essential. In this paper, we propose a split learning-based cloud-edge-end collaborative training framework (SCCT) to harness the abundant computational resources of cloud and edge servers while exploiting massive distributed datasets on edge devices for LLM fine-tuning. In SCCT, an LLM with a parameter-efficient fine-tuning module is deployed on the cloud server, while a small-scale language model (SLM) is deployed across each edge device participating in SCCT and its associated edge server in the manner of split learning. SLM and LLM collaborate in a serial manner for training and inference. To optimize the collaboration efficiency, we formulate a mixed integer nonlinear programming problem to minimize the training latency of SCCT, considering the resource heterogeneity of edge devices and network dynamics. To solve this problem, a two-timescale optimization algorithm is proposed to determine the optimal split points on the large timescale and the allocation of communication and computational resources of the edge server on the small timescale. Numerical results demonstrate that the proposed two-timescale optimization algorithm assisted SCCT outperforms the state-of-the-art LLMs deployment and training paradigm in terms of training efficiency and inference performance.

The integration of ubiquitous 5G cellular networks with deterministic Ethernet, such as Time-Sensitive Networking (TSN), is essential for future industrial applications, offering high flexibility and strict determinism. A key challenge in this integration is the dynamic mapping of TSN traffic to 5G QoS profiles, especially given the diverse QoS requirements across f lows. While existing methods based on static mapping or approximations can be effective under stable conditions, they fail to adapt to fluctuating network loads and evolving QoS demands, leading to delays and inaccurate profile selection. To overcome these limitations, we propose DQMARS — a Dynamic QoS Mapping Approach with Resource Slicing. In DQMARS, 5G QoS resources are partitioned into n resource slices aligned with TSN traffic types. Each resource slice is associated with multiple 5G QoS profiles and supports flexible selection based on flow-level QoS requirements at admission time. Within each slice, a Bayesian-optimized learning model leveraging feature and attention transformers is employed for dynamic mapping. This model identifies the most appropriate QoS profile for each TSN traffic flow by evaluating multiple QoS attributes, such as bandwidth, packet delay budget, and packet error rate. We evaluate DQMARS across various industrial scenarios, achieving a mapping accuracy exceeding 99% and minimal delay averaging 1.63 × 10<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">−3</sup> ms per traffic flow. Compared to state-of-the-art methods, our approach significantly reduces mapping delay while exhibiting superior adaptability to dynamic network conditions, making it highly suitable for time-critical industrial applications.

The fully-decoupled radio access network (FD-RAN) with multiple base stations (BS) networked cooperation is expected to provide high-quality communication coverage for aerial vehicles while sensing their locations accurately and efficiently. In this paper, we propose a novel multiple BSs cooperative passive integrated sensing and communication (ISAC) framework in FD-RAN for low-altitude aerial vehicles. It utilizes the Doppler shift information from uplink orthogonal frequency division multiplexing (OFDM) signals to achieve efficient localization without incurring additional sensing overhead. Specifically, the carrier frequency offset (CFO) compensation process at each BS captures Doppler shift resulting from the relative motion between the aerial vehicle and the BS. Considering the position and velocity of the aerial vehicle as unknown variables, the localization is achieved by formulating and solving a system of equations describing these Doppler shifts. Furthermore, to enable accurate trajectory tracking, we design a learning-based adaptive double-filter (LADF) algorithm, which leverages temporal correlations in both Doppler shift measurements and vehicle trajectories to improve the time-continuous localization. Extensive simulation results demonstrate the effectiveness of our proposed ISAC framework in low-altitude FD-RAN, which can enhance the coverage performance and efficiently provide accurate localization for aerial vehicles.

This paper investigates a class of time-fractional quaternion-valued diffusive multidirectional associative memory neural networks (MAM-NNs) incorporating double fuzzy operators and distributed delays. By leveraging eigenvalue theory for the Laplace operator within quaternion algebra, we develop a Fourier series-based solution framework for the proposed MAM-NNs. Utilizing properties of the Mittag-Leffler function and the contraction mapping principle, we establish the existence and uniqueness of a weighted pseudo <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$S$</tex-math></inline-formula>-asymptotically <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\omega$</tex-math></inline-formula>-periodic oscillation. Furthermore, we introduce a novel stability criterion for fractional-order differential equations with distributed delays, based on a generalized Halanay-type inequality, and rigorously analyze the global Mittag-Leffler stability of the quaternion-valued fuzzy MAM-NNs. To optimize the convergence rate of Mittag-Leffler stability, we formulate nonlinear optimization problems and propose corresponding algorithms. The effectiveness of our method is validated through numerical simulations, which confirm the theoretical findings, as well as through practical applications in image processing. Notably, our approach conducts all analyses directly in the quaternion domain without relying on Hamilton rule-based decomposition, thereby preserving the intrinsic structure of quaternion-valued systems and enabling more accurate and efficient computations.

Recently, quantum deep reinforcement learning (Q-DRL) has started to gain attention as a potential approach for tackling complex challenges in wireless communication systems. In particular, Q-DRL, integrating quantum operations into deep learning models, can effectively handle dynamic environments and process large-scale optimizations. As future wireless networks continue to evolve, greater emphasis is being placed on context and meaning rather than raw data. New paradigms, such as semantic communications (SemComs) are essential to effectively convey meaning between transmitters and receivers. By linking SemComs with Q-DRL, future wireless networks will be capable of large-scale extractions and decoding of meaning, thereby minimizing reliance on complete context sharing between communicating parties. Together with SemComs, digital twins (DTs) have been considered as key enablers for future wireless networks. As virtual replicas of physical networks, they serve an important role in network operation, optimization, and control. In this regard, Q-DRL will also be highly beneficial for DTs in enhancing critical functions such as data management and security. This study offers fresh outlooks on how to leverage Q-DRL for SemComs in future wireless networks, augmented by the use of DTs.