Dual connectivity is a key enabler for reliability and service continuity in 5G non-terrestrial networks, particularly in multi-band and multi-orbit satellite scenarios. However, its practical implementation at the user equipment side is challenged by implementation constraints. A primary obstacle is the timing advance mechanism: even with multiple radio interfaces, a software-defined radio-based UE can typically maintain only one valid timing advance value, making simultaneous bidirectional operation toward multiple cells infeasible. In this paper, we present a pragmatic UE architecture that overcomes this limitation to enable dual connectivity-like capabilities. Our design maintains one active link for bidirectional communication (both uplink and downlink), while a second radio interface is dedicated to establishing a monitoring link that maintains downlink-only synchronization and signal quality estimation of a secondary cell. Building on this fully UE-autonomous architecture, we implement a power level-based hard handover mechanism. The UE continuously monitors its active and monitoring links and triggers the handover based on local power level measurements, without any involvement from the base station or 5G core network. This approach provides a practical pathway toward improved link reliability in non-terrestrial systems while reducing network signaling using commercially available software-defined radio platforms.
The Open Networking Foundation (ONF) has been a driving force in the advancement of Software-Defined Networking (SDN), network disaggregation, and cloud native architectures. Its ecosystem fosters programmable high-performance network infrastructures, leveraging SDN control planes, P4-based data plane programmability, white-box switching, and AI-driven network automation. The open source frameworks of ONF, including ONOS, P4Runtime, Stratum, and SEBA, provide scalable, intent-based solutions for broadband access, edge computing, and multi-domain transport networks. With its projects now hosted under the Linux Foundation (LF), the technological innovations of ONF continue to evolve, supporting next-generation carrier-grade, cloud native, and AI-integrated networks. This article explores the history and core technical advances of ONF and their impact on the programmability, scalability, and automation of modern networking architectures.
Malaria is one of the most life-threatening infectious diseases worldwide, necessitating early and accurate diagnosis to ensure effective treatment. Traditional microscopic examination of blood smears is labor-intensive and prone to diagnostic inconsistencies. Additionally, many existing deep learning (DL) approaches employ conventional transfer learning (TL) strategies that fine-tune pretrained convolutional neural networks (CNN) or use them as fixed feature extractors, thereby constraining domain adaptability and increasing reliance on extensive labeled datasets. Therefore, this study introduces a self-supervised DL framework that integrates contrastive representation learning with TL for automated malaria diagnosis. The proposed model employs a dual-objective contrastive fine-tuning strategy that jointly optimizes supervised cross-entropy and self-supervised contrastive objectives. This hybrid training paradigm enforces semantic alignment between positive sample pairs while maximizing inter-class separation, thereby yielding highly discriminative embeddings. Multiple CNN backbones, including EfficientNetB7, Revitalized DenseNet Reloaded, ConvNeXt, and RegNetX, serve as the foundational architectures for implementing this strategy. The contrastive EfficientNetB7 attained the highest accuracy of 96.87%, outperforming all baseline models. Unlike conventional pipelines that rely on handcrafted preprocessing, segmentation, and multi-stage classification, the proposed framework delivers an end-to-end generalizable solution. This work provides a new perspective on the integration of self-supervised learning within TL to enhance diagnostic reliability and representation quality in medical image analysis.
Continuous physiological monitoring facilitates the adaptation of immersive digital therapies, dynamic gaming environments, or virtual reality training simulations to user’s state. The real-time extraction of the vital signs such as breathing, blood pressure, and heart rate requires computationally intensive algorithms and produces a large volume of data. This paper investigates compressed sensing techniques to reduce the transmitted data from sensors to be reconstructed accurately at a centralized server. We propose three strategies to control the sampling ratio accordingly to the quality of the signals. These strategies are assessed and analyzed in a multimodal dataset captured via camera and radar.
This paper presents the implementation of Ant Colony Optimization (ACO) to solve the Travelling Salesman Problem (TSP) on the Chained-Cubic Tree (CCT) interconnection network. The CCT combines hypercube and tree interconnection networks, offering low diameter and high scalability while eliminating the drawbacks of both interconnection networks. Traditional interconnection networks present inherent trade-offs: hypercubes offer low diameter but suffer from increasing node degree as networks grow, while trees provide scalability but exhibit low connectivity. Experimental evaluation using VLSI benchmark datasets containing 3,954 to 10,150 cities demonstrates that solving the TSP using ACO over CCT achieves efficient speedup performance while maintaining superior scalability.
A Global Navigation Satellite System (GNSS) operating in medium Earth orbit (MEO) is widely used for positioning in several applications, including land vehicles. However, GNSS receivers can suffer from both intentional and unintentional jamming, multipath, and degraded performance in urban environments due to the relatively weak satellite signal. GNSS can be aided by measurements from Inertial Navigation Systems (INS) to enhance positioning performance in challenging scenarios; however, the inherent bias drift and scale-factor instabilities of inertial sensors may degrade performance after a short period of time. With stronger signals, higher availability, and greater temporal diversity, this paper explores the benefits of integrating Low Earth Orbit (LEO) satellites with both GNSS and INS to provide a reliable positioning solution during extended GNSS outages or degraded GNSS performance. We propose a centralized, tightly coupled, fusion scheme based on extended Kalman filtering (EKF), that mitigates GNSS challenges and estimates both inertial sensor errors for INS and receiver clock errors for GNSS/LEO receiver, thus enhancing positioning performance. We demonstrate the performance of our new approach on real road-test trajectories collected in the city of Kingston, involving low-cost inertial sensor measurements from a MEMS-based INS integrated with a commercial GNSS receiver, and LEO satellite signals generated by the Safran GSG-8 simulator for the LEO satellite constellation of the Xona Pulsar system. The quasi-real data collected in this research demonstrated the benefits of the proposed GNSS/LEO/INS approach for maintaining reliable, continuous land vehicle positioning in challenging GNSS environments.
The optimal placement of Base Stations (BS) is a critical challenge in the planning of wireless telecommunication networks, directly impacting both CAPital EXpenditure (CAPEX) and service quality. This problem, characterized by its NP-hard complexity, requires balancing conflicting objectives: minimizing deployment costs while maximizing user coverage. This paper presents a metaheuristic approach based on Ant Colony Optimization (ACO) to solve the multi-objective BS positioning problem. We propose a graph-based formulation where the search space is modeled as a fully connected network, allowing artificial ants to construct solutions by selecting a subset of candidate sites. A key contribution of this work is the implementation of a redundancy-aware pheromone update mechanism. Unlike iterative construction approaches which can be driven by locally attractive choices and may lead to clustered stations with overlapping signals, our approach evaluates the global solution to penalize redundant coverage, promoting the maximization of unique users served. Simulation results demonstrate that the proposed ACO algorithm effectively converges to a high-quality solution, achieving coverage for 312 users with an infrastructure cost below 50% of the theoretical maximum, thereby validating its viability for efficient network design.
In recent papers it has been demonstrated, that Time Domain Extended Active Interference Cancellation (TD-EAIC) can provide a significant sidelobe suppression at acceptable self-interference level and feasible computational effort in a non-guarded (NG)-OFDM system. Meanwhile much simulation research has been made to find the best adaption to use it within a cyclic prefix (CP)-OFDM system as well. This paper presents two configurations that were found to be best, concerning either the aspect of coding redundancy or the aspect of signal latency. Both can provide a spectral shaping which reduces out-of-band energy (OOBE) to a negligible level and with self-interference levels below the interference that can be expected from a typical CP-OFDM channel. Compared to previously published, similar EAIC-approaches, they show significant lower self-interference with identical sidelobe suppression. The first configuration uses the differential (D-) version of TD-EAIC that was presented in the last paper, but with necessary adaption to the carrier phase. The second configuration works without any latency to the signal path, making it the stringent choice for transmission systems with a minimal roundtrip requirement. The truncated singular value decomposition (SVD) approach, as presented in the last paper, is again used to make the computational complexity linear and thus practically implementable regardless of FFT size and carrier count. The usage of the proposed methods includes, achieving deep notches in the transmission spectrum of Cognitive Radio systems. Furthermore, steep spectral edges can be accomplished to reduce spectral guard bands between different transmitters and exploit bandwidth efficiently without mutual interference.
Wireless Sensor Networks (WSNs) deployed in harsh environments, such as space missions, face the challenge of maintaining precise time synchronization, especially due to dynamic conditions like rapid temperature fluctuations. This paper introduces a novel, tick-based simulation framework designed to model and evaluate the performance of local clock discipline algorithms. The framework’s methodology is based on a clock error model that incorporates inherent drift, periodic variations, and stochastic noise. An event system simulates dynamic environmental impacts, such as temperature changes derived from actual satellite mission data, allowing for realistic test conditions. In addition, an evolutionary optimization approach is implemented to tune algorithm parameters. To demonstrate its utility, a case study is presented comparing several algorithms using a scenario derived from satellite telemetry. The results validate the simulation framework as a flexible tool for algorithm development and parameter tuning, facilitating the prediction of algorithm performance under extreme conditions.
Modern education faces many challenges today. This includes lack of motivation and engagement among students and trainees, as well as the low efficacy of traditional methods. This paper presents an experiment on immersive education, from the education use case of the AVANZADO-5G-INMERSIVO project, performed by Universidad Politecnica de Valencia and Telefonica Innovacion Digital. The experiment is based around the comparison of a traditional 2D video lesson versus a Virtual Reality application with full-body, free point-of-view volumetric representation of the teacher. The study gathered a total of 62 test subjects which could experience both forms of on-demand content, which consisted of a traditional Valencian hairdressing lesson. Use experience was evaluated via surveys, direct observation and interviews. Results show that, even though the volumetric representation of teachers still doesn’t provide an adequate image quality, users think that this technology has the potential to enhance their learning experience, engagement and promote active learning.
Mobile Positioning Data (MPD) enables large-scale mobility analysis but is prone to anomalies from noise, tower switching, and missing events. This study proposes and validates a unified framework through a detailed single-user case study for trajectory-based anomaly detection that compares supervised and reinforcement learning (RL) models under consistent preprocessing and spatial grouping. The framework integrates multi-phase ground-truth labeling with Google Timeline and heuristic validation, spatial grouping (DBSCAN, ST-DBSCAN, OPTICS, GeoHash, S2, H3), and spatiotemporal feature engineering. Experiments show that ensemble classifiers achieve near-perfect detection (F1 ≥ 0.99), with OPTICS clustering yielding robust results. A Deep Q-Network (DQN-RL) agent also reached competitive achievement (F1 ≥ 0.98) while enabling adaptive sequential decision-making. Overall, results highlight OPTICS combined with LightGBM as the most efficient and accurate configuration, while RL provides an alternative when labels are limited or sequential context is essential.
Officinal plants represent a valuable resource for human society due to their therapeutic, economic, and ecological importance. However, identifying and monitoring these plants in natural environments can be challenging, as they often grow in areas that are difficult to access using traditional field surveys. In recent years, the decreasing cost of unmanned aerial vehicles (UAVs) has enabled new approaches for large-scale environmental monitoring. Previous studies have explored the use of UAV imagery for plant recognition, typically relying on RGB cameras. In this work, we propose an edge to far-edge network framework for officinal plant classification capable of processing and integrating RGB and multispectral aerial imagery acquired by UAV platforms. The collected data are processed using a spectral signature–based classification approach, which exploits per-class multispectral profiles derived from annotated training data to perform automated plant recognition. Training data are acquired through programmed UAV flights, pre-processed on a far-edge aerial layer, and then on an edge computing layer on the ground. These data are then used to extract class-specific spectral signatures, which were then applied to new images for pixel-wise classification. The proposed approach demonstrates the potential of combining RGB and multispectral information to support efficient monitoring of officinal plants.
Low Earth Orbit (LEO) satellite networks have emerged as a promising solution for global high-speed connectivity, yet their performance stability under high-density user demand remains largely unexplored. Specifically, it is unclear how frequently satellite handovers and link variability impact the Quality of Experience (QoE) when a single terminal serves multiple concurrent users in a stationary event. This paper proposes an empirical evaluation of Starlink’s capacity to support a scientific event with up to 90 simultaneous clients as the primary backhaul. We conducted a real-world experiment correlating flow-level traffic profiles, satellite signal quality metrics, and edge router resource usage within an over-provisioned local network infrastructure. Results show that the network successfully sustained aggregated throughput peaks of over 200 Mb/s with latency remaining stable between 30–50 ms, demonstrating that LEO connectivity can reliably support high-density stationary events with minimal infrastructure compromise, outperforming expectations derived from mobility-focused studies.
The demand for high-throughput, low-latency communications has led to impressive advances in wireless technologies. Operating these heterogeneous systems in an overly crowded spectrum mandates intelligent spectrum sharing policies that enable harmonious coexistence. Machine learning (ML) based signal classifiers play a critical role by allowing devices to identify the types and features of contending transmissions without decoding them. For deployment on edge devices, the classifier must be both energy-efficient and capable of real or near-real-time operation to enable fast link adaptation. This paper proposes a two-stage ML approach for protocol-plus-modulation classification in shared spectrum. The framework is evaluated in the unlicensed 5 GHz and 6 GHz UNII bands, where LTE LAA, 5G NR-U, and Wi-Fi 6 coexist. The first stage identifies the communication protocol, while the second stage classifies the modulation scheme, with Wi-Fi 6 and LTE considered as representative use cases. To achieve significant reductions in energy consumption and inference latency, we adopt a Posit number representation for the weights and inputs of the neural networks, and compare its performance against commonly used representations, including floating point (FP) and Brain Float (BF16). Our extensive simulations demonstrate that models quantized with 8-bit Posits can achieve classification accuracy comparable to their default 32-bit FP baselines, leading to lower storage requirements and substantial reductions in energy expenditure and latency. These reductions are demonstrated by executing trained protocol and modulation classification models on a new low-power, compact SIMD processor designed for edge applications that provides hardware support for Posit.
Metaverse-assisted healthcare and immersive telemedicine rely on timely delivery of physiological sensor data to maintain accurate patient monitoring and synchronization with patient-specific Digital Twins (DTs). In such systems, the freshness of transmitted data, measured by the Age of Information (AoI) is crucial for QoS, realistic interactions, and accurate clinical decisions. However, wireless hospital networks operate under limited bandwidth and time-varying channel conditions while supporting multiple sensors with different levels of clinical importance. Conventional transmission scheduling approaches are not well suited for such dynamic environments, particularly when sudden physiological anomalies require immediate prioritization. This paper presents a reinforcement learning based transmission scheduling framework to minimize AoI in Metaverse-assisted healthcare networks. A Q-learning agent dynamically selects sensors for transmission and allocates bandwidth based on network conditions, queue states and clinical priorities. A lightweight statistical anomaly detection module adjusts sensor priority levels when irregular physiological patterns are detected, ensuring timely delivery of critical health data. Simulations experiments using a publicly available physiological signal datasets show significant AoI reduction, demonstrating an effective AI-based strategy for Metaverse-enabled healthcare networks.
This paper presents a real-time localization and navigation framework based on the augmented quaternion unscented Kalman filter (AQUKF) for loosely coupled global positioning system (GPS) and inertial navigation system integration INS). The AQUKF incorporates quaternion-based attitude representation within the nonlinear model and preserves the unit-norm constraint during filtering, enabling accurate and stable pose estimation. The filter is validated experimentally through a real-time autonomous waypoint navigation scenario. Results show that the AQUKF achieves higher localization accuracy and safer obstacle avoidance compared to the extended Kalman filter (EKF), confirming its robustness for outdoor navigation tasks.
Future wireless systems are envisioned to operate at millimeter-wave and sub-Terahertz frequencies to enable ultrahigh data rates. Transitioning to these frequency bands, combined with the use of large antenna apertures, significantly extends the near-field region of transmitters, placing many receivers within this range. Crucially, realizing the full potential of such networks requires acquiring and tracking the real-time location of near-field users. However, this shift renders conventional localization techniques inadequate, as they are primarily designed for far-field operation and fail to account for near-field wavefront characteristics. This paper introduces NearLoc, a novel, few-shot near-field localization framework that enables precise user positioning using a single transmitter array without requiring phase-coherent measurements. We leverage the distinctive propagation characteristics of near-field wavefronts to shape the spatial energy distribution along the depth axis, enabling efficient range estimation using power-only measurements. Particularly, NearLoc leverages non-diffracting Bessel beams that confine power along a controllable propagation depth and iteratively refine the probabilistic belief of the user location through hierarchical spatial division within a Bayesian inference framework, where the transmitted beam’s propagation depth is adapted based on the updated posterior. Extensive experiments in the D-band regime, demonstrate the effectiveness of the proposed scheme, highlighting its potential as a core component for location-assisted communication and services in future wireless networks.
Emerging smart cities and Internet of Things (IoT) ecosystems are creating device-dense environments where many nodes can opportunistically serve as positioning anchors, reducing reliance on dedicated infrastructure. While high anchor density can enhance localization accuracy, indiscriminate use of all available anchors increases computational load, energy consumption, and may degrade performance in dense non-line-of-sight (NLoS) conditions. To address this challenge, this paper proposes two binary integer programming (BIP)–based anchor selection strategies for received signal strength (RSS) positioning. The formulations determine an optimal subset of anchors by jointly accounting for anchor cardinality, geometric quality through Cramér–Rao lower bound (CRLB)–inspired metrics, and signal strength. The first approach applies an RSS threshold to pre-filter anchors and then optimizes geometry and subset size, while the second incorporates RSS constraints directly into the optimization process. The selected anchors are subsequently used within a Gauss–Newton localization framework. MATLAB simulations under NLoS conditions demonstrate that both proposed approaches consistently outperform a baseline solution that uses all anchors. In 2D and 3D scenarios, the methods achieve substantial reductions in root mean square error (RMSE) and improved cumulative distribution function (CDF) performance, while reducing the number of active anchors by approximately fifty percent. These results highlight the importance of intelligent anchor selection for scalable and energy-efficient localization in dense smart-city areas.
With the growing interest in smart electric mobility, autonomous electric vehicles are increasingly adopted in mobility-on-demand (MoD) applications. The operational efficiency of shared autonomous electric vehicle (SAEV) fleets, however, strongly depends on reliable vehicle-to-everything (V2X) communications. In dense urban environments, establishing direct line-of-sight (LoS) links is often challenging due to building-induced shadowing. To address this issue, this paper proposes an IRS-assisted V2X framework in which SAEVs are equipped with intelligent reflecting surface (IRS) modules and operate as communication relays between the base station and target vehicles. A mixed-integer linear programming (MILP) model is developed to jointly optimize vehicle dispatching and relay configurations under signal-to-noise ratio (SNR) constraints, aiming to maximize service rates while minimizing average customer waiting times. A simplified road network is used to evaluate the proposed approach. Numerical results show that cooperative relaying improves fleet availability and service rates, but may introduce latency penalties in small fleets as the proportion of relay vehicles increases. Overall, the results demonstrate the viability and reliability of the proposed IRS-assisted SAEV operations management framework, motivating further investigation into optimal on-vehicle IRS configurations and operating modes.
Video streaming has become a significant part of people’s daily routines, reflecting the pervasive role of media consumption in their lives. With the widespread adoption of high-speed internet and the proliferation of smart devices, the consumption of media content has skyrocketed. As a result of these trends, network operators and service providers are constantly challenged to optimize their infrastructure, enhance bandwidth capacity, and improve Quality of Service (QoS) to meet the growing demands of media traffic. The deployment of new-generation enables operators to manage QoS requirements by softwarizing network functionalities, providing a higher level of adaptability. It is then necessary to facilitate efficient cache management, optimal resource allocation, and enhanced Quality of Experience for end-users, crucial for the modern media landscape.
