Low Earth Orbit (LEO) satellite networks are expected to play a central role in future 6G Non-Terrestrial Networks (NTN) envisioned by the ITU IMT-2030 framework. Efficient video delivery over such networks remains challenging because of rapid visibility changes and strong elevation-dependent variations in link capacity. Motivated by the need for adaptive and energy-efficient operation in emerging 6G NTN architectures, this paper presents a preliminary study on elevation-aware video prefetching in LEO networks. A controlled simulation framework emulating stress-test link dynamics is used to evaluate three algorithms: popularity-based (B1), deadline-aware (B2), and a proposed Elevation-Aware Deadline Prefetching (B3) method that adapts prefetch timing to elevation-dependent link quality, along with a no-prefetch baseline (B0). Results show that the proposed elevation-aware approach achieves Quality of Experience (QoE) equivalent to the deadline-aware method while improving link-quality efficiency by 2.9%, reducing low-elevation usage by 3.0 percentage points, and performing 921 fewer operations. These preliminary findings indicate that incorporating elevation information in prefetch scheduling can improve resource utilization and efficiency without compromising user experience, offering early insights for ongoing 6G NTN research focused on adaptive and energy-conscious communication mechanisms.
Multi-Robot Systems (MRS) are becoming increasingly vital across various domains. In many practical applications, inadequate infrastructure necessitates decentralized operations, making inter-robot data sharing a critical challenge. As the scale of the system grows, both in terms of the number of robots and the operational area, achieving effective system-wide (global) data sharing becomes more difficult, which in turn affects the efficiency of task allocation in large-scale MRS. Emerging 5G network technologies provide facilities for localized data sharing, which occurs much faster compared to global ones. In this work, we study multi-robot task allocation under 5G-assisted localized inter-robot communication. Specifically, we provide a comprehensive comparison of three distinct well-known approaches. Our findings reveal that greedy selection-based strategies perform poorly when information availability is limited. Interestingly, task allocation strategies based on regret-maximization heuristics perform better under restricted information conditions than with full system visibility. Furthermore, our results show that a multi-agent reinforcement learning (MARL) strategy outperforms all other approaches, exhibiting minimal performance degradation even with significantly constrained information sharing.
In this paper, a low-cost, 1-bit reconfigurable intelligent surface (RIS) operating at 3.5 GHz is designed to enhance wireless signal coverage in sub-6 GHz frequency ranges. Each RIS unit cell has two square-shaped patches linked by an SMP 1340-040LF PIN diode, which is placed on a dual-layer FR4 epoxy board. The structure provides a phase shift of 180 ° ± 30 °, depending on the diode’s state, while maintaining a reflection loss of below 2.15 dB over a 630 MHz bandwidth. A 20 × 10 element array is created using column-wise diode biasing to support one-dimensional beam steering. Radial stubs are added to the biasing circuit to prevent interference between the DC control signals and the RF signals. The simulated far-field patterns demonstrate that directional beams can be formed at angles of ±15°, ±30°, and ±45°, confirming the array’s effectiveness in adaptable environments. This RIS design could significantly contribute to the development of affordable 5G and future communication technologies.
We propose a novel sparse matrix precoding (SMP) scheme to increase the spectral efficiency (SE) of MIMO systems while leveraging the belief propagation decoding on a sparse bipartite graph at the receiver to obtain near-optimal performance with low computational complexity. Furthermore, conventional MIMO precoding/ equalization has been targeted toward channel compensation. Our proposed SMP scheme is additionally aimed at increasing the SE while lowering the MIMO receiver complexity. Unlike the conventional approach of enhancing SE in bandwidth-constrained MIMO systems by employing higher-order modulation, our proposed method achieves higher SE through sparse matrix precoding applied to a lower-order modulation at the transmitter. This approach offers a key advantage: the resulting MIMO receiver is computationally more efficient. This is because our scheme is designed to enable a message-passing algorithm (on the log-likelihood ratios (LLRs) of the lower-order modulation) on the bipartite graph representing the sparse precoding matrix, rather than directly decoding a higher-order constellation. We provide information-theoretic analysis that shows the conditions under which the SMP-MIMO provides the same channel capacity as the conventional MIMO. Several simulation results showing MIMO channel capacity and bit-error-rate (BER) provide proof of the proposed concept.
Structural Health Monitoring (SHM) ensures the safety and longevity of civil infrastructure. Recent advancements in Artificial Intelligence (AI), particularly deep learning (DL), have transformed SHM by enabling automated, scalable solutions for damage detection and assessment. This paper presents a synthesis of AI-driven SHM methods—from vision-based surface damage detection and vibration analysis to physics-informed models and digital twins. We explore the integration of CNNs, RNNs, GANs, and transformers with non-destructive testing (NDT), UAV-based monitoring, and 3D reconstruction. Emphasis is placed on real-time defect localization, sensor fault mitigation, and data recovery in operational environments. Despite progress, challenges like data scarcity, environmental variability, and generalizability remain. We propose a roadmap using hybrid AI-physics models and autonomous SHM systems to enhance infrastructure resilience.
In this paper, we investigate the combined effect of nodes mobility and imperfect successive interference cancellation (SIC) on the outage performance of a multi-tag ambient backscatter communication (AmBC) system over Rayleigh fading. The AmBC system consists of a mobile ambient RF source, K energy harvesting enabled mobile tags, and a mobile reader. Under the nodes mobility scenario, all the fading links are modeled using a first-order autoregressive process. We also employ a best tag selection scheme, while the reader utilizes both perfect SIC (pSIC) and imperfect SIC (ipSIC) techniques. Under this system setup, we derive the closed-form expressions for the outage probability (OP) under both pSIC and ipSIC scenarios. We also present the asymptotic OP analyses in the high signal-to-noise ratio (SNR) regime to extract key insights into the system's diversity order. In addition, we explore several practical scenarios, including static nodes configuration and large time-varying errors, to characterize their effects on the system performance. Finally, simulation results are provided to validate the accuracy of the derived analytical expressions.
This paper presents an experimental 1-bit Reconfigurable Intelligent Surface (RIS) integrated with a USRP-based Software-Defined Radio (SDR) platform for enhancing wireless communication in indoor and outdoor environments. The RIS, built as a 20×20 PIN diode array on an FR-4 substrate, enables programmable beam redirection through microcontroller-based phase switching. To benchmark its capabilities, a passive metallic plate of identical dimensions is used as a reference. Real-time video transmission and signal analysis reveal that, unlike the static reflection from the metal plate, the RIS provides dynamic wave-front control, improving signal quality, directionality, and throughput. Indoor tests demonstrate higher SINR and data rates, while outdoor measurements confirm stable wideband performance across varying distances. This comparative study not only highlights the superiority of programmable RIS over traditional reflectors but also emphasizes its readiness for practical deployment in future smart wireless systems.
This paper explores the joint impact of two capacity enhancement schemes in optical backbone networks: multi-band expansion from C+L to C+L+S bands and varying 3R regeneration (no, selective, and full). Using a pay-as-you-grow batch upgrade framework that considers deferral benefits, we evaluate their interaction. In the short-haul BT-UK network, S-band upgrade consistently improves throughput and cost efficiency, with the greatest economic gain with no regeneration. In the long-haul USNET network, S-band upgrade reduces throughput because K-least-loaded routing does not consider path distance, yielding low-quality lightpaths with high blocking probability. Thus, C+L bands with full regeneration are more cost-effective.
Hybrid Reconfigurable Intelligent Surface (HRIS)assisted localization is pivotal for several sixth-generation wireless applications. HRIS comprises active and passive reflecting elements (REs) to support multiple users and targets in the near-field (NF) localization framework. In the HRIS system, active REs are connected to the power amplifiers (PAs) and phase shifters, while passive REs are only connected to the phase shifters. Optimizing the number of connections between active REs and PAs can significantly improve localization. This work proposes a novel Q learning-based active RE-PA association method to improve NF localization. In particular, the power allocation between each PA-active RE and the number of connections between PA and active RE are optimized to reduce the system’s power consumption. This reduces the hardware complexity due to the connection of each active RE with a separate PA. The experimental results have confirmed that the proposed algorithm outperforms the existing schemes and effectively compensates for the performance degradation caused by the reduction in PAs.
In this paper, we propose a simultaneous wireless information and power transfer (SWIPT) for a non-orthogonal multiple access (NOMA) and millimeter wave (mmWave) based sixth generation (6G) vehicle-to-vehicle (V2V) cellular communication system. We consider a cooperative full-duplex (FD) V2V communication system that utilizes NOMA and mmWave technology to enhance both spectrum efficiency and channel capacity. In the proposed model, the forward relay node operates in full-duplex (FD) mode, where it simultaneously utilizes the available spectrum for both information transmission and energy harvesting. Specifically, the relay assists in establishing the cellular downlink communication link from the base station (BS) to the V2V end user while concurrently harvesting energy from the BS’s transmitted signal. We derive closed-form analytical expressions for the outage probability and validate the derived results through Monte Carlo simulations, thereby confirming the accuracy and reliability of the proposed analytical framework for the FD cooperative V2V system. Furthermore, we investigate the impact of key system parameters, including residual self-interference (RSI), energy harvesting (EH) efficiency, and signal-to-interference-plus-noise ratio (SINR), on the overall performance of the proposed system.
GPS spoofing attacks pose serious security threats to Autonomous Vehicles (AVs) by compromising their navigation systems, potentially leading to route deviations and critical safety risks. While several Machine Learning (ML)-based detection mechanisms have been proposed, the solutions exhibit multicollinearity among input features, resulting in information redundancy and poor generalizability across diverse Datasets. To address these challenges, this paper proposes SafeRoute, an autoencoder-driven hybrid model for the effective detection of GPS spoofing attacks on AV networks. Initially, SafeRoute employs an autoencoder to perform feature selection by eliminating redundant and correlated features, thereby preserving relevant information and improving the robustness of the model. The refined feature set is then processed using a hybrid model that combines a recurrent neural network with long-short-term memory (RNN-LSTM) to capture temporal dependencies in GPS data and a Random Forest (RF) classifier for accurate detection of GPS spoofing attacks. The proposed mechanism achieves a high accuracy of 99.98% in detecting GPS spoof attacks across multiple Datasets, demonstrating enhanced generalizability. In the future, this framework can be extended to detect more sophisticated attacks targeting AV systems.
In this work, we introduce and analyze a novel performance measure, namely, Reliable Average PHY Secrecy Range (RAPSR) for the uncrewed aerial vehicle (UAV)-assisted, four-node, dual-hop cooperative physical layer (PHY) secure wireless communication system utilizing Reconfigurable Intelligent Surfaces (RISs) as intermediate nodes. We consider the presence of one Eavesdropper (Eav) node. For it, we define RAPSR as the maximum hovering distance of an RIS-equipped UAV that ensures a strictly positive average PHY secrecy capacity in the presence of an Eav. Specifically, we present an exact analysis for RAPSR based on average PHY secrecy capacity and insightful high signal-to-noise ratio (SNR) approximation-based analysis. To validate the accuracy of the approximation, we simulated the system and obtained an insightful plot that explicitly shows the accurate regime. Our exact and approximate analysis and associated plot provide key design insights for secure RIS-assisted UAV deployments in cooperative terrestrial and vehicular communications.
Multi-controller software-defined networks (SDNs) have emerged as a scalable control plane architecture to enhance the availability, reliability, and performance in large-scale networks. However, when a controller fails, it creates substantial operational issues, as the workload from the inactive controllers needs to be efficiently distributed among the functioning ones, which may lead to overload and a drop in performance. To address this problem, we present a failure-aware fractional switch migration (FT-FSM) approach for multi-controller SDNs designed to minimize the maximum utilization ratio of controllers in various failure scenarios. This approach employs a mixed integer linear programming (MILP) formulation to find the optimal fractional reassignment of switch flows, taking into account controller capacity constraints, migration overhead, and fault tolerance needs. The MILP model ensures that in each failure scenario, traffic from failed controllers is rerouted to active ones while adhering to predefined thresholds. For large-scale instances where the MILP may become intractable, a heuristic method can be employed for the same formulation. The numerical results indicate that the proposed approach delivers a lower utilization ratio and decreases the count of overloaded controllers, thus boosting system resilience and efficiency under controller failure conditions, compared to traditional schemes.
The rapid development of smart edge devices, such as smartphones, wearable health trackers, and IoT-enabled sensors, has changed how data is generated, processed, and used in various applications. While this data is necessary for intelligent applications, its stringent privacy requirements introduce new challenges for centralized data sharing. Federated Learning (FL) enables distributed model training across edge devices without sharing original data. In FL, clients train local models on their data using the model parameters shared by the server. However, in smart edge devices, the generated data is heterogeneous, where standalone models often reduce model accuracy due to diverse data distributions in the clients. In this paper, we propose a Hybrid Machine Learning (HML)-based FL framework, HySecFL, that integrates multiple deep learning architectures to enhance model accuracy and reduce training loss, while preserving data security. In HySecFL, each client trains a federated model on hybridized local models. To configure model hybridization, we select the models with the highest performance when trained standalone. These models include Convolutional Neural Networks (CNN), Feedforward Neural Networks (FNN), Recurrent Neural Networks (RNN), and transfer learning-based architectures such as VGG16 and VGG19. To evaluate the performance, we experimented on three standard FL datasets: MNIST, FMNIST, and CIFAR-10. The results show that our framework achieves an overall 3.12% higher accuracy and reduces training loss by 22.56% compared to single models.
Deep learning training workloads are inherently non-stationary and exhibit rapid variability, making static compute reservations in Cloud-native Machine Learning Operations (MLOps) pipelines either wasteful (due to over-provisioning) or risky (due to under-provisioning). We present DeepScale, a forecast-driven, high-frequency, in-place vertical pod autoscaling (VPA) framework that proactively right-sizes training pods without evictions or downtime. DeepScale continuously ingests per-pod CPU telemetry, forecasts near-term demand using time-series models (XGBoost, LSTM, N-BEATS), and applies policy-aware safety margins and buffers to produce robust resource recommendations. By leveraging in-place VPA in recent Kubernetes releases, resources are updated live with sub-minute convergence. Evaluated on three representative training pipelines, i.e., DNN (MNIST), CNN (CIFAR-10), and LSTM (IMDB), DeepScale reduces requested vCPU-hours by 38–91% over static and current in-place VPA baselines, hence lowering the cost proportionally. These results indicate that coupling predictive control with in-place vertical scaling yields a practical, self-optimizing, and cost-aware autoscaling solution for deep-learning training workloads.
Open RAN is a next-generation wireless network technology that promotes flexibility, cost efficiency, and interoperability through disaggregation and open interfaces. Energy efficiency remains a key challenge, especially in scalable deployments with many connected User Equipments (UEs), where dynamic power management at the gNodeB is critical. In this paper PAC-ORAN (Power-saving Adaptive CPU Scheduling for Open RAN) is proposed which is an advanced CPU scheduling algorithm evaluated with a large number of connected UEs to gNodeb. PAC-ORAN includes an advanced method using moving average adaptive threshold selection and dynamically adjusting CPU core states and frequency tuning based on real-time CPU metrics. The evaluation of PAC-ORAN focuses on two key objectives: (1) Sustainability Goals: PAC-ORAN reduces CPU power consumption by 5% to 30%, lowers RAM power usage by 2% to 20%, and brings down CPU temperature by 2 to 3 degree Celsius. (2) Performance Goals: PAC-ORAN maintains bounded throughput even during power-saving operations and balances CPU frequency usage. The key contribution of this paper lies in the proposed PAC-ORAN algorithm tested on large number of connected UEs along with key findings from the PAC-ORAN algorithms impact on CPU and RAM power utilization as well as its impact on CPU temperature and frequency usage.
Ensuring fast and efficient cloud-service restoration after a disaster is critical, yet it is hindered by resource competition and confidentiality concerns among stakeholders, such as network carriers and Datacenter providers (DCPs). To address this, we propose a novel centralized cooperation model led by a neutral entity called Provider Neutral Exchange (PNE). We devise a multi-objective Integer Linear Programming (ILP) optimization model to be executed by PNE to maximize restoration of cloud services requested by the DCPs, based on resource availability of the carriers. The model also tries to minimize restoration time and cost while prioritizing critical connections. Results across various disaster scenarios demonstrate that our approach significantly enhances cloud-service restoration compared to heuristic strategies.
The evolution of Open Radio Access Networks (O-RANs) is reshaping wireless communications by enabling flexibility, virtualization, and multi-vendor interoperability. However, their disaggregated architecture broadens the attack surface, posing new security risks. Effective intrusion detection remains challenging due to emerging threats, limited labeled data, and class imbalance, which renders traditional metrics like accuracy unreliable. To address this, we propose Fense, a framework that combines feature-engineered Tabular Generative Adversarial Networks (TGANs) with Matthews Correlation Coefficient (MCC) for robust performance evaluation and synthetic data generation. Experiments show that Generative AI (Gen AI) enhances detection accuracy, fairness, and resilience, tackling core challenges in O-RAN security.
Asphalt pavements are aging under rising traffic loads and increasingly extreme climates, yet most asset-management workflows remain reactive, relying on periodic surveys or empirical models that do not continuously reflect evolving material states. This paper presents a physics-informed neural network (PINN) framework for structural health monitoring of asphalt pavements subjected to extreme environmental stressors such as temperature, moisture, water-logging, rainfall, and UV exposure. A hybrid dataset is assembled by merging Long-Term Pavement Performance and TxDOT records with synthetically generated extreme-event scenarios and accelerated testing surrogates. The proposed PINN model augments data with mechanistic features and employs physics-based regularization to enforce monotonic and saturation behaviors. Trained and evaluated with scikit-learn and PyTorch, the model predicts remaining service life (RSL) with high fidelity (test R2 =0.92, RMSE = 1.5 years) that outperforms standard ANN and a mechanistic-empirical regression. Case studies show accurate RSL adjustment after a flood and sensitivity to heat and UV aging. The approach enables proactive, climate-aware pavement maintenance decision-making and demonstrates how AI-driven digital twins, when coupled with distributed sensing and edge computing can enable scalable, communication-aware solutions for infrastructure resilience in future 6G IoT networks.
The emerging 6G landscape envisions large-scale distributed Internet of Things (IoT) systems that integrate sensing, communication, and computation across resource-constrained devices. In such environments, achieving energy-efficient and low-latency connectivity is critical to sustain scalable and autonomous operations. The Routing Protocol for Low Power and Lossy Networks (RPL) has become a standard for IoT routing; however, its conventional Objective Functions (OF0 and MRHOF) face limitations in maintaining energy balance and minimizing delay, particularly in dense or dynamic deployments. To address these challenges, this paper proposes two enhanced versions of RPL, termed as Energy-saving RPL (ERPL1 and ERPL2). ERPL1 combines Expected Transmission Count (ETX), energy consumption, and hop count, while ERPL2 incorporates ETX, energy consumption, and latency as key metrics in the parent node selection process. Simulations conducted using the Cooja simulator under both static and mobile network conditions show that the proposed ERPL schemes outperform OF0, MRHOF, and the state-of-the-art ETXRE. Specifically, ERPL1 and ERPL2 demonstrate up to 41% lower average energy consumption, 17% higher Packet Delivery Ratio (PDR), and 23% reduced end-to-end delay across various network topologies. These improvements are achieved through better load balancing and optimized routing paths, making the proposed ERPL protocols highly suitable for energy-constrained and distributed IoT environments. The study highlights the promise of multi-metric, energy-aware routing as a foundation for sustainable and efficient connectivity in future 6G-enabled IoT networks.
