We present a systematic design that ensures end-to-end service level agreements (SLAs) while minimizing the operational expenditure (OPEX) in an Open Radio Access Network (O-RAN). Unlike existing frameworks, which adopted theoretical models that are often misaligned with industrial standards, we devise our scheme based on standard-compliant, particularly 3GPP-specified models. We develop an algorithm that decouples the complex resource management into three coordinated closed-loop control stages: long-term infrastructure planning, medium-term computing resource allocation, and short-term radio resource scheduling. The algorithm has a short execution time, making it suitable for real-world implementations. We corroborate the efficacy of the proposed scheme through extensive numerical simulations and a prototype built upon an O-RAN testbed called INA-Infra, implemented using Nephio. Experimental results demonstrate that, compared to the baselines, our scheme significantly reduces SLA violations and enhances resource efficiency under non-stationary traffic conditions, confirming its engineering feasibility.

Embedding spectrum sensing directly into 5G radio access networks (RANs) enables continuous and fine-grained monitoring of interference under active traffic. However, existing approaches typically rely on spectrogram-based analysis of raw received I/Q signals, making low-power interference difficult to detect when it overlaps with legitimate transmissions. This demo presents an in-RAN spectrum sensing system implemented directly inside a 5G gNodeB using a GPU-accelerated High-PHY architecture. The system isolates noise-and-interference (NaI) from uplink signals and performs interference detection and classification within the gNB. Experiments on an actual AI-RAN testbed show that NaI isolation enables reliable detection and classification of interference at power levels up to 10 dB lower than raw-signal-based approaches, even when interference is fully overlapped with active uplink traffic.

With the rapid development of sixth-generation (6G) wireless networks, ensuring secure and covert communication has become increasingly important. Meanwhile, autoencoder (AE)-based communication systems have emerged as a promising paradigm due to their ability to jointly optimize the transmitter and receiver in an end-to-end manner. This paper aims to propose an AE-based covert communication system that generates transmission signals with low autocorrelation (ACF), making them difficult for eavesdroppers to detect, while maintaining the block error rate (BLER) performance for the intended communication. To address the scalability limitation of conventional AE systems, where input dimensionality grows exponentially with message size, we develop an AE architecture that integrates error-correction coding (ECC). The source message is first encoded into a binary sequence and fed into the AE, while an ECC decoder at the receiver recovers the original message, enabling efficient binary-input processing and improved BLER performance. Besides, to reduce ACF and detection probability without compromising intended communication performance, we propose a novel training mechanism that incorporates a Kullback-Leibler (KL) divergence term into the conventional cross-entropy loss. Simulation results show that the proposed communication system achieves BLER performance comparable to that of conventional AE-based systems over both AWGN and Rayleigh fading channels. Meanwhile, under the ACF-based detection framework, the proposed AE system trained with the proposed loss exhibits a notably lower detection rate than the conventional AE system. In particular, with a false alarm rate of 10-3, the proposed AE-based system can achieve a detection probability of 0.0457, compared to 0.9089 for the conventional AE-based system, under an AE setting where each 4-bit message block is modulated into 10 transmitter symbols by the AE encoder.

Embodied AI requires sub-second inference near the Radio Access Network (RAN), but deployments span heterogeneous tiers (on-device, RAN-edge, cloud) and must not disrupt real-time baseband processing. We report measurements from a 5G Standalone (SA) AI-RAN testbed using a fixed baseline policy for repeatability. The setup includes an on-device tier, a three-node RAN-edge cluster co-hosting a containerized 5G RAN, and a cloud tier. We find that on-device execution remains multi-second and fails to meet sub-second budgets. At the RAN edge, SLA feasibility is primarily determined by model variant choice: quantized models concentrate below 0.5 s, while unquantized and some larger quantized models incur deadline misses due to stalls and queuing. In the cloud tier, meeting a 0.5 s deadline is challenging on the measured WAN path (up to 32.9% of requests complete within 0.5 s), but all evaluated variants meet a 1.0 s deadline (100% within 1.0 s). Under saturated downlink traffic and up to N=20 concurrent inference clients, Multi-Instance GPU (MIG) isolation preserves baseband timing-health proxies, supporting safe co-location under fixed partitioning.

How to minimize data delivery time in time-varying networks is a fundamental problem with applications ranging from data centers to satellite systems. This problem can be solved by computing maximum flow on a Time-Expanded Graph (TEG). However, existing approaches rely on repeated maximum flow computations over different time horizons, resulting in high computational overhead. In this work, we propose OPSS, a One-Pass and Smart-Search solver that leverages the temporal structure of TEGs to efficiently identify the earliest feasible delivery interval. OPSS incrementally expands the time horizon while reusing intermediate results, thereby avoiding redundant recomputation. The proposed smart-search strategy further reduces the search space and improves efficiency. We prove the optimality of OPSS and evaluate it using data delivery problem over real-world satellite constellations. The results show that OPSS achieves up to 38× speedup over state-of-the-art methods, demonstrating its effectiveness and scalability in large-scale time-varying networks.

Modern 5G/6G radio access networks are increasingly programmable through O-RAN, yet their operational complexity has grown with disaggregation, open interfaces, and fine-grained control parameters. While RAN-side analytics and telemetry mechanisms, such as KPI-based monitoring and mobility event reporting, provide visibility into network behavior, operators still face challenges in correlating heterogeneous events and safely translating observations into actionable configuration changes. This paper presents an LLM-based Net Analyzer rApp for the O-RAN Non-RT RIC that enables explainable and safe, human-in-the-loop automation for RAN operations. The proposed rApp adopts an event-informed, batch-triggered reasoning framework in which mobility events are first interpreted, anomalies are confirmed through targeted log inspection, configurations are inspected via tool-gated access, and minimal configuration changes are proposed only after explicit operator approval. The architecture enforces a strict separation between reasoning and actuation, ensuring auditability and operational safety. The system is implemented and demonstrated on a real O-RAN testbed using a reproducible ping-pong handover scenario, illustrating how large language models can function as reasoning co-pilots that transform raw RAN telemetry into structured explanations and controlled remediation workflows, complementing existing analytics-only approaches in the NonRT RIC.

Satellite networks will be a critical part of the 6G infrastructure, offering ubiquitous coverage and resilience to natural disasters on Earth. However, the expected increase in the number of services with high communications and compute demands, such as remote sensing and rural Internet of Things (IoT) data analytics, poses challenges to limited resources in satellite networks. Service Function Chain (SFC), an application-driven network technology, offers a promising solution by flexibly orchestrating virtualized functions into a service chain to support these demands. Our work studies the SFC-constrained maximum flow problem that aims to find the maximum flow possible between a source and a destination node subject to the SFC constraints, where the data must flow through a predefined sequence of service functions. We propose the Flexible Function-Time Expanded Graph (F-2-TEG) to represent the SFC ordering requirements and the time-varying network topology, while uniformly modeling the compute, storage and communication resources. We prove that F-2-TEG models the entire feasible solution space of the SFC-constrained max-flow problem, in particular, incorporating the full flexibility in the allocation of compute resources. This flexibility is not captured by the state-of-the-art (SOTA) work. For smaller networks, the optimal solution can be found effectively using F2-TEG through a linear programming (LP) solver, since F-2-TEG supports effective pruning. For larger networks, we further propose an efficient graph algorithm over F-2-TEG that achieves lower computation complexity via local search. Simulation over the real-world Starlink constellation shows that the enlarged solution space compared to SOTA work improves the maximum flow by 50-142% and the F-2-TEG-based LP scheme is 3.5x faster than other graph-model-based LP schemes. For larger networks, our graph-based algorithm can be more than 130x faster than LP-based schemes, while still improving the maximum flow by 46% compared to SOTA work.

Time-Triggered Communication is a key technology for Industrial Internet of Things (IIoT), 5G/6G ultra-reliable and low-latency communication (URLLC) and Vehicle-to-Everything (V2X).We name the data over the time-triggered communications time-triggered flows. Each flow consists of periodic packets that need to be delivered from a source to a destination node. Each packet needs to reach its destination before its deadline. Different flows can have different cycle lengths. To achieve assured transmission of time-triggered flows, existing efforts constrain the packets of a flow to be cyclically transmitted along the same path based on the flow’s given cycle length. Under such Fixed Cyclic Scheduling (FCS), reservation for flows with different cycle lengths can become incompatible over a shared link, limiting the total number of admissible flows. Considering the cycle lengths of different flows, a hypercycle has length equal to their least common multiple (LCM). It determines the time duration over which the scheduling compatibility of the different flows can be checked. In this work, we propose a more flexible scheduling scheme called the Hypercycle-level Flexible Scheduling (HFS) scheme, where a flow’s resource reservation can change over its different cycles within the same hypercycle. HFS can significantly increase the number of admitted flows by providing more scheduling options while remaining perfectly compatible with existing time-triggered system. We show that, (theoretically) the possible capacity gain provided by HFS over FCS can be unbounded. We formulate the time-triggered joint pathfinding and scheduling problem under HFS as an integer linear programming problem which we prove to be NP-Hard. To solve HFS efficiently, we further propose a least-load-first heuristic (HFS-LLF), solving HFS as a sequence of shortest path problems. Extensive study under real-world settings shows that HFS admits 1.5× the number of flows scheduled by FCS. Moreover, our proposed HFS-LLF can run 104× faster than solving HFS using a generic solver.

Random Access Channel (RACH) jamming poses a critical security threat to 5G and beyond (B5G) networks. This paper presents an analytical model for predicting the impact of Msg1 jamming attacks on RACH performance. We use the OpenAirInterface (OAI) open-source user equipment (UE) to implement a Msg1 jamming attacker. Over-the-air experiments validate the accuracy of the proposed analytical model. The results show that low-power and stealthy Msg1 jamming can effectively block legitimate UE access in 5G/B5G systems.

Inter-satellite data routing is a fundamental issue that must be addressed to achieve efficient global satellite directto- device communication in the large-scale LEO satellite network (LSN). Existing routing algorithms are inadequate for enabling satellite direct-to-device communication in LSN due to protocol incompatibility, excessive overhead, or performance limitations. Facing these challenges, this paper investigates a novel pathinformed distributed routing algorithm based on collaborative deep reinforcement learning (PI-CDRL) to achieve highperformance and scalable distributed inter-satellite routing in dynamic LSN. We first design a virtual geographic node-assisted routing process to help devices and satellites establish mappings under dynamic conditions. Building on this foundation, the sustained path information following packet transmission completion is leveraged to facilitate collaborative training among satellites, effectively mitigating the sub-optimal performance limitations induced by partial observation and enabling satellite nodes to make accurate real-time routing decisions. Simulation results demonstrate that the successful transmission rate and average latency of PI-CDRL outperform those of baseline distributed routing methods under identical network conditions. Furthermore, our proposed method exhibits robustness in performance as the network scale expands and topology changes due to the ability to fully utilize the path information.

The Rapid advancement of the satellite industry offers unprecedented opportunities for enabling Internet of Everything (IoE) applications over satellite networks. A key characteristic of such applications is that computation cannot begin until the entire application data has been fully received at the destination. To meet strict end-to-end delay constraints, minimizing the total application delay is essential. However, this requirement violates the optimal substructure property commonly assumed in traditional shortest path routing problems. Existing routing solutions often overlook these unique computation constraints and rely on substructure-preserving heuristics, resulting in suboptimal delay performance. Moreover, they lack reliability in producing delay-guaranteed routing solutions, which leads to low task completion ratios under stringent application deadlines. To overcome this problem, we propose FlexSatIoE-a routing scheme that allows for flexible buffering data over satellite networks. FlexSatIoE formulates this routing problem as an integer linear programming (ILP) problem, to provide the optimal solution. As the network scales, considering the computational intractability of ILP, FlexSatIoE further modifies the storage time-aggregated graph to comprehensively model the satellite networks' compute, storage and transmission resources. Based on the graph extension, FlexSatIoE designs an efficient routing algorithm, enabling flexible use of buffer resources by using a flow reassignment mechanism. We conduct extensive experiments over the setting of real-world satellite networks. The results show that FlexSatIoE reduces the average delay and increases the number of completed tasks by up to 50% and 40%, respectively, as compared to the existing schemes, demonstrating the superior capability and reliability of FlexSatIoE in ensuring deterministic application delays.

This paper demonstrates the feasibility of transformer-based split inference for real-time video object detection over dynamic 5G AI-RAN networks. We extend throughput-aware adaptive splitting from CNNs to a Swin Transformer backbone and show that practical split execution is achievable for transformer-based vision models without retraining. To address the large intermediate activations inherent to transformers, we introduce an efficient, accuracy-preserving activation compression pipeline that substantially reduces uplink payload. The complete system – including adaptive split selection, transformer inference, and compression – is implemented and validated end-to-end on a real-time detection workload, with distributed UPF (dUPF) integration further reducing user-plane latency and improving runtime stability. Extensive measurements on an NVIDIA Aerial-based AI-RAN testbed jointly account for inference and 5G communication energy, quantifying the latency-energy-privacy trade-offs in realistic deployments.

When applying the game-theoretic approach to find the optimal strategy for industrial cyber-physical systems defenders, most existing work assumes both the system states (e.g., for a power grid system, the system state captures which buses are compromised) and the attacker’s instant actions are observable and can be used to make the decision for the defender’s next move. Also, the reward and expected utilities are calculated based on the most likely system state and attack action. However, there is uncertainty in determining the system state and attack actions as the attack unfolds in the system in practice. This work shows that such an approximation is non-optimal in determining the defense strategy. Instead, we propose a framework that models the uncertainty in the system state and attack action. We derive the defender’s optimal strategy under such uncertainty by calculating the expected utilities across different action spaces and redefining the immediate reward within the deep learning algorithm based on the expected utilities and our estimation of the probabilistic distribution of the system state and attack action, ultimately employing the agent system for learning and generating the optimal defense strategy. The simulation experiments are carried out based on the generic industrial cyber-physical system testbed and the numerical results confirmed that the proposed solution can improve the defender’s expected utilities by 38.8% compared to the state-of-the-art.

In vehicular cooperative perception (CP), numerous resource allocation strategies have been proposed to enhance urban autonomous driving. However, existing studies often overlook the competition between self-perception and cooperative perception, where degrading a ground vehicle's (GV's) self-perception may introduce safety risks and reduce passenger comfort. Moreover, air-ground cooperation-which can improve sensing precision, reduce task execution delay, and enhance CP service availability-has received limited attention. It is worth noting that autonomous aerial vehicles are unavailable for cooperative perception during the recharging process.To address these issues, this paper investigates on-demand scheduling strategy in air-ground cooperative perception. At the millisecond timescale, resource competition is considered in real-time sensing, communication, and computation (SC$<^>{2}$2) resource allocation. At the minute timescale, the idle flying period between consecutive tasks is utilized for autonomous aerial vehicles' recharging through attachment to GVs along the route. Specifically, we first develop a model that captures the mutual influence between autonomous aerial vehicles and GVs on perception performance under resource constraints. Then, a mixed-timescale solution is proposed: at the small timescale, a multi-agent deep reinforcement learning algorithm with gradient-free projection and auxiliary supervision is designed to schedule SC2 resources; at the large timescale, a Hungarian-based algorithm is employed to control autonomous aerial vehicles' recharging. Simulation results show that the proposed approach outperforms benchmark schemes by reducing task execution delay and energy consumption, and enhancing CP service availability, while satisfying sensing precision, GV safety, and passenger comfort requirements.

Conventional satellite beam-hopping (BH) systems rely on phased-array antennas for multiple-input-multiple-output (MIMO) beamforming, requiring numerous high-power radio frequency chains that degrade energy efficiency (EE). To address this, this article employs stacked intelligent metasurfaces (SIM) to enable a nonorthogonal multiple access (NOMA)-based low-Earth-orbit satellite BH network. SIM can facilitate electromagnetic wave-domain beamforming, eliminating energy-intensive high-resolution digital-to-analog converters-based digital beamforming. We formulate a joint optimization problem to maximize system EE by designing wave-based beamforming at SIM and NOMA user power allocation, subject to transmit power constraints and user quality-of-service requirements. The nonlinear variables and fractional formulation induce nonconvexity in this problem, and an alternating quadratic transform-central gradient algorithm is proposed as the solution to this challenge. Simulation results demonstrate that our proposed algorithm can achieve convergence under various SIM configurations, and compared to conventional MIMO beamforming schemes, the proposed SIM-NOMA scheme achieves higher system EE with the minimum number of antennas. Even when the number of antennas in the MIMO system is increased threefold, SIM-NOMA still maintains a significant performance advantage.

Time-Triggered Communication is a key technology for many safety-critical systems, with applications spanning the areas of aerospace and industrial control. Such communication relies on time-triggered flows, with each flow consisting of periodic packets originating from a source and destined for a destination node. Each packet needs to reach its destination before its deadline. Different flows can have different cycle lengths. To achieve assured transmission of time-triggered flows, existing efforts constrain the packets of a flow to be cyclically transmitted along the same path. Under such Fixed Cyclic Scheduling (FCS), reservation for flows with different cycle lengths can become incompatible over a shared link, limiting the total number of admissible flows. Considering the cycle lengths of different flows, a hyper-cycle has length equal to their least common multiple (LCM). It determines the time duration over which the scheduling compatibility of the different flows can be checked. In this work, we propose a more flexible schedule scheme called the Hypercycle-level Flexible Schedule (HFS) scheme, where a flow's resource reservation can change across cycles within a hypercycle. HFS can significantly increase the number of admitted flows by providing more scheduling options while remaining perfectly compatible with existing Time-Triggered Ethernet system. We show that, theoretically the possible capacity gain provided by HFS over FCS can be unbounded. We formulate the joint pathfinding and scheduling problem under HFS as an ILP problem which we prove to be NP-Hard. To solve HFS efficiently, we further propose a least-load-first heuristic (HFS-LLF), solving HFS as a sequence of shortest path problems. Extensive study shows that HFS admits up to 6 times the number of flows achieved by FCS. Moreover, our proposed HFS-LLF can run 104 times faster than solving HFS using a generic solver.

Large-scale low-Earth-orbit (LEO) satellite systems are increasingly valued for their ability to enable rapid and wide-area data exchange, thereby facilitating the collaborative training of artificial intelligence (AI) models across geographically distributed regions. Due to privacy concerns and regulatory constraints, raw data collected at remote clients cannot be centrally aggregated, posing a major obstacle to traditional AI training methods. Federated learning offers a privacy-preserving alternative by training local models on distributed devices and exchanging only model parameters. However, the dynamic topology and limited bandwidth of satellite systems will hinder timely parameter aggregation and distribution, resulting in prolonged training times. To address this challenge, we investigate the problem of scheduling federated learning over satellite networks and identify key bottlenecks that impact the overall duration of each training round. We propose a discrete temporal graph-based on-demand scheduling framework that dynamically allocates communication resources to accelerate federated learning. Simulation results demonstrate that the proposed approach achieves significant performance gains over traditional statistical multiplexing-based model exchange strategies, reducing overall round times by 14.20% to 41.48%. Moreover, the acceleration effect becomes more pronounced for larger models and higher numbers of clients, highlighting the scalability of the proposed approach.

Digital Twin (DT) technology has emerged as an important enabler for enhancing cellular network operations, particularly in complex environments such as transportation hubs. However, existing DT efforts for cellular network often focus on modeling of physical propagation channel and communication protocols. Many of them lack a systematic approach to model how cellular network is affected by system-specific operational states (e.g., flight arrival schedule or various possible faults in an airport). In this work, we propose an operation-aware digital twin framework that incorporates the knowledge of system operations into the optimization of AI-driven radio access networks. This framework integrates a "what-if" operational scenario generator to enumerate diverse operational scenarios. Furthermore, it can sync up with the actual system’s operation states in near real time to provide on-the-fly fine-tuning of AI/ML models. We developed an end-to-end prototype of our proposed framework and validated its benefits through performance evaluation using a set of AI/ML rApps running on RAN Intelligent Controller. Our proposed framework can help advance DT technology for its use in 5G networks and beyond, enabling intelligent and responsive network management in dynamic environments.

Service availability (SA) is essential for maintaining productivity and operational continuity in Industrial Internet of Things (IIoT) systems. However, accurately evaluating availability is challenging due to the complexity of dynamic recovery techniques integrated failure detection, localization, and rerouting restoration. These processes involve state-dependent behaviors and interactions across multiple network layers, making it difficult to pre-enumerate all recovery paths before failures occur. Thus, it is crucial to develop availability models that account for the dynamic coupling between service failure and recovery factors to achieve more accurate evaluations. To address this, we propose a novel two-stage state-transition availability evaluation method (TSAE) to quantify dynamic service availability in IIoT systems. In the first stage, fault conditions of network components and service mapping states are sampled to identify events that lead to service outages. In the second stage, a set of transition rules is defined to model the complex interactions between components and services during dynamic recovery processes. The effectiveness and accuracy of the proposed method are validated using two typical network scenarios. Extensive sensitivity experiments on a large-scale network case, considering factors, such as recovery priority, resource demand-supply ratio, and recovery performance, reveal key availability bottlenecks under various deployment configurations. Furthermore, tradeoffs between global and local recovery strategies under varying failure conditions are discussed, providing insights for network designer to select options that balance service continuity and resource efficiency.

The rapid rise of botnet attacks has disrupted telecom network services for millions of users. Detecting botnet traffic is therefore important for maintaining the availability and quality of these services. Existing methods primarily focus on developing attack detection techniques but ignore the adverse effects of various real-world factors, such as changes in attack behaviors and network conditions, on model performance. In response to this challenge, we propose a generalizable multi-view framework for botnet flow detection, namely GM-BFD. GM-BFD learns flow representations from intra-flow and inter-flow views, where the intra-flow view extracts the complex relationship between flow features, and the inter-flow view extracts flow behaviors between hosts while avoiding host representations. This multi-view modeling provides an expressive representation of flows and avoids overfitting host information during training. To enhance the model's generalizability, GM-BFD designs an un-supervised domain adaptation model with prediction confidence filtering to learn discriminative and transferable representations of benign and malicious flows across scenarios as accurately as possible. Experimental validation using the CTU dataset with 13 scenarios demonstrates the superior performance and generalizability of GM-BFD, with an $F_{1}$-score improvement of approximately 11 % over the best baseline.