This paper presents two training-assisted (pilot-aided) implementations of a K-means clustering-based beamformer that operate without prior knowledge of the angle of arrival (AOA). In the training phase, the transmitted symbols (class labels) corresponding to the received vectors are known from a predefined pilot sequence. The proposed methods infer cluster centroids directly from the received vectors corresponding to this known training sequence. In the first approach, the M-dimensional centroid corresponding to each constellation point is estimated by averaging a sufficient number of received vectors known to belong to that cluster. During detection, the received vector is associated with the nearest cluster using the Euclidean distance. The second approach extends this method by estimating not only the centroids but also the covariance of each cluster, enabling the use of Mahalanobis distance for classification. This covariance-aware classifier is consistent with a Gaussian-mixture viewpoint when trained from labeled pilot data. MATLAB simulations confirm that incorporating cluster covariance improves the bit error rate (BER) performance. The results demonstrate the effectiveness of both approaches in providing robust beamforming without AOA estimation, with the second method offering superior detection accuracy.
Human activity recognition (HAR) plays a pivotal role in ambient assisted living, particularly for monitoring the elderly and patients with chronic conditions. However, traditional approaches relying on wearable sensors or video cameras face significant challenges regarding user compliance and privacy intrusion. To mitigate these issues, this paper proposes a device-free sensing (DFS) ( https://github.com/mestrelan/MDA-CSI ) framework utilizing Wi-Fi channel state information (CSI), named DFS-CSI. We introduce a robust Transformer-based architecture designed to capture long-range temporal dependencies in wireless signals. DFS-CSI was validated using a comprehensive dataset from 86 volunteers, ensuring high generalization capabilities across diverse human motion patterns.
Low-power wide-area networks (LPWAN), in particular the Long-Range (LoRa) and Long-Range Wide-Area Network (LoRaWAN), have become key enablers of the Internet of Things (IoT). Their efficiency depends on parameters such as the spreading factor (SF), transmit power, and the signal-to-noise ratio, as well as the features of the protocol and regulatory constraints. Several simulators have been developed; most of them remain partial: some lack complete LoRaWAN protocol compliance, and others omit multichannel operation, mobility, security mechanisms, or higher device classes. This complicates fair evaluation and comparison between studies. To address these shortcomings, we present LoRaFlexSim, a modular and event-driven simulator designed to offer both fidelity and extensibility. It is also designed for the large-scale performance evaluation of Long Range (LoRa)/LoRaWAN networks. LoRaFlexSim reproduces the main packet-level performance trends of existing reference frameworks while extending them with multichannel support, Classes A/B/C, Over-the-Air Activation (OTAA) with session key management, mobility models, Adaptive Data Rate (ADR), per-node energy accounting, and interactive dashboards for scenario design and monitoring. Its architecture allows rapid prototyping of new mechanisms while ensuring consistency with LoRaWAN specifications. By combining protocol completeness with extensibility, LoRaFlexSim provides researchers and practitioners with a practical tool to investigate scalability, energy efficiency, and adaptive mechanisms, while paving the way for future integration of machine learning and reinforcement learning for intelligent resource management.
Split learning (SL) is a promising distributed learning approach for fifth-generation/sixth-generation (B5G/6 G) networks. However, its performance is affected by bandwidth constraints and statistical heterogeneity in the data. Prior works have considered either channel-aware or data-aware device selection in isolation, but few jointly consider both under per-round participation limits. This article extends the “A Network Metric-Aware Split Learning Architecture for B5G/6 G Mobile Networks” (SLArch) framework with explicit bandwidth constraints, partial user equipment (UE) participation, and non-independent and identically distributed (non-IID) data modelled via Dirichlet distributions. We introduce four device selection policies: random, channel-quality-based (ChQual), data-quality-based (DataQual), and a joint channel-and-data-aware policy (ChDataQual) weighted by parameters φ and β . Simulations with a total of 150 UEs and 10 participants per round demonstrate key network and SL performance results. DataQual achieves the highest accuracy across most slices, reaching 94.8
The rapid growth of 5G-enabled Internet of Things (IoT) systems increases the need for intrusion detection systems (IDSs) that can adapt continuously to evolving attack patterns, heterogeneous telemetry sources, and changing operational domains. However, conventional deep learning IDS models are usually trained under static assumptions and may suffer from catastrophic forgetting when exposed sequentially to new tasks, domains, or attack classes. This study aims to evaluate the deployability and robustness of continual-learning (CL) strategies for 5G-IoT intrusion detection under three incremental learning paradigms: task-incremental learning (Task-IL), domain-incremental learning (Domain-IL), and class-incremental learning (Class-IL). Using the multi-source ToN-IoT dataset, lightweight MLP-based architectures are implemented with task-specific heads, shared heads, and expanding heads, and six CL strategies are evaluated using accuracy, macro-F1, forgetting, false-positive rate, false-negative rate, confusion matrices, ablation studies, and computational profiling. The results show that Replay provides the most reliable retention across all scenarios, achieving 76.36
Data transmission at the nano-scale faces constraints that are notably different from those of conventional wireless systems, especially in terms of communication reliability, resource availability, and security. In the Internet of Nano-Things (IoNT) environments, protecting sensitive information while preserving acceptable network performance is still an open challenge, particularly when cryptographic mechanisms introduce additional processing and communication overhead. To address such an issue, this paper proposes an IoNT framework that integrates a DNA-based bio-molecular cryptographic scheme with a machine learning (ML) module aimed at dynamically predicting suitable security configurations under changing network conditions. The proposed approach is evaluated in a healthcare scenario based on a simulated artery environment, considering different routing/MAC combinations and varying numbers of nano-devices. Results show that the integration of ML enables improved robustness under congestion conditions, reducing packet loss by approximately 33–37
Distributed ledger technologies (DLT) can enhance trust and auditability in the Internet of Things (IoT). Among them, IOTA has been specifically designed to support machine-to-machine interactions and IoT data anchoring through scalable DLT architectures. However, their integration with Low-Power Wide-Area Networks (LPWANs) remains limited due to device constraints, strict timing requirements, and the operational costs of on-chain transactions. The transition from the fee-less Stardust to the fee-based IOTA Rebased model introduces explicit transaction costs, questioning the viability of continuous IoT data anchoring. IOTA provides a suitable platform to examine the challenges of integrating distributed ledger technologies with LPWAN-based IoT systems. Its transition to a fee-based execution model raises important questions regarding cost predictability and performance in continuous data anchoring scenarios, particularly under the constraints of resource-limited and latency-sensitive environments. This article investigates the practicality of the execution and payment model introduced by IOTA Rebased for IoT scenarios requiring continuous data notarization. We provide an empirical evaluation of continuous IoT data notarization on the public IOTA Rebased Mainnet and characterize the performance implications on edge-oriented deployments, including resource-constrained and resource-rich devices. We implement a notarization oracle that ingests LoRaWAN uplinks from The Things Network (TTN), canonicalizes payloads, generates SHA-256 commitments, and records them on-chain through reusable notarization objects. The oracle enables continuous anchoring of IoT telemetry while minimizing transaction overhead through object reuse. Two 24-h experimental campaigns compare a notarization oracle on resource-constrained and resource-rich hardware under periodic workloads. Results show consistent steady-state gas consumption for UPDATE operations, indicating that object reuse enables stable on-chain cost behavior in IOTA Rebased regardless of the deployment platform. From a performance perspective, both environments achieve stable execution; however, the resource-constrained edge deployment exhibits higher median and tail latency, alongside tighter memory margins compared to the resource-rich centralized baseline. These findings confirm the feasibility of deploying notarization services on constrained edge infrastructure under the new fee-based model.
We propose BTDM, a decentralized blockchain-assisted authentication and trust-management framework for the Internet of Vehicles. Unlike conventional certificate-based approaches, BTDM eliminates dependence on a centralized Certificate Authority by allowing vehicles and RSUs to generate their own cryptographic key pairs and authenticate through session-bound zk-SNARK proofs. Authentication combines OTP-based freshness verification, privacy-preserving credential validation, and blockchain-assisted authorization checking. Following successful authentication, forward-secret session keys are established using ephemeral ECDHE and HKDF-based key derivation to support lightweight authenticated communication. BTDM further incorporates adaptive trust management with evidence-backed RSU voting, reputation-weighted penalties, trust recovery, and quorum-based revocation. To reduce revocation overhead, Bloom-filter-based certificate revocation lists are anchored on the blockchain, while TESLA-based authentication enables efficient V2V beacon dissemination. Closed-form communication models and NS-2.35 simulations are used to evaluate the proposed framework. Results demonstrate that BTDM reduces communication overhead by up to 9.3 × compared with BCPPA, achieves approximately 70 × faster authentication, and reduces average revocation latency to below one second in dense urban vehicular environments.
The expansion of connected systems and the increasing structural complexity of modern networking infrastructures introduce significant challenges to digital security. Limitations of traditional perimeter-based security models, together with the growing exposure of resource-constrained devices, further complicate the protection of data and services. This paper examines zero trust architecture (ZTA) as a security paradigm that enforces continuous authentication, microsegmentation, and context-aware access control across heterogeneous network environments. The study combines a conceptual analysis of ZTA principles with a detailed examination of its core architectural components. Furthermore, it critically assesses the challenges and opportunities associated with ZTA adoption in next-generation networks. Representative use cases from enterprise and federated networking environments are presented to support the discussion.
Sparse vector transmission, designed for constructing sparse linear underdetermined systems, has garnered significant interest in short-packet ultra-reliable and low-latency communications (URLLC), owing to its advantages such as simple implementation and reliable transmission. A key challenge in optimizing system performance lies in the construction of short sparse vectors. To address this issue, this paper proposes a novel multi-layer merged sparse vector transmission framework, which partitions the index bits into multiple layers and sequentially determines the non-zero entries of each layer within a predefined sparse vector via layer-wise sparse mapping. By enabling cross-layer sharing of position resources, this structure achieves efficient sparse vector compression and higher resource utilization than conventional global selection. Simulation results demonstrate a clear block error rate (BLER) advantage over prior sparse vector transmission schemes, particularly under high coding rates where sparse vectors contain many non-zero entries.
The expected growth in the number of connected devices, together with the spectrum crunch in overcrowded radio bands, motivates the exploration of complementary wireless technologies. Optical Wireless Communication (OWC), including visible light systems, provides a vast unlicensed spectrum but introduces unique networking challenges: strong directionality, line-of-sight requirements, sensitivity to shadowing, and fast link intermittency due to mobility. This paper presents PARC (Proactive and Adaptive Routing Control), a proactive routing protocol designed specifically for light-based ad hoc networks (LANETs). PARC combines proactive topology dissemination with link-quality-aware route selection and adaptive control mechanisms. We evaluate PARC in ns-3 OWC scenarios ranging from static baselines to highly dynamic and high-density settings, including an additional extended stress-test with simultaneous mobility and full traffic load. Overall, PARC sustains significantly higher packet delivery and throughput than the OLSR reference protocol under challenging conditions.
Reconfigurable intelligent surfaces (RIS) offer a cost-effective solution for enhancing both spectral and energy efficiency in next-generation wireless networks by employing numerous low-cost passive elements that reflect incident signals with tunable phase shifts. These capabilities are particularly advantageous for long range (LoRa) systems in internet of things (IoT) applications, where reliable long-distance communication is essential. This paper investigates a RIS-assisted LoRa communication system operating under non-line-of-sight (NLoS) conditions, where a LoRa device transmits data to a gateway via a RIS. Under a narrowband assumption, the RIS is modeled by frequency-flat reflection coefficients across the signal bandwidth. To avoid practical hardware constraints, discrete phase shifts are considered at each RIS element. Two phase-optimization strategies are proposed: an exhaustive search method, suitable for small RIS arrays, and a genetic algorithm-based approach, scalable to large deployments. Both strategies aim to maximize the received signal power at the gateway. Bit error rate (BER) performance is evaluated at the output of a coherent receiver. Simulation results show that the proposed RIS-assisted LoRa system significantly improves BER compared with conventional LoRa over a Rayleigh fading channel, achieving 10^-4 BER. The gap with respect to the ideal additive white Gaussian noise (AWGN) channel is about 10 dB, and increasing the number of RIS elements further enhances performance via improved beamforming and higher SNR, highlighting the potential of RIS-assisted LoRa for reliable and energy-efficient low-power-wide-area-network (LPWAN) operation.
Federated learning (FL) enables machine learning (ML) on sensitive data across multiple participants without sharing their local data, thereby preserving privacy. To ensure the FL process is both fair and effective, participants must be incentivized to contribute high-quality updates. However, evaluating contributions accurately, without compromising data privacy or trusting a central authority, is challenging. Evaluation mechanisms must also resist manipulation by dishonest participants attempting to gain undeserved rewards or sabotage the process. We propose BlockFL-SPACE, a blockchain-based FL framework offering secure, privacy-preserving, and tamper-resistant contribution evaluation. To our knowledge, it is the first framework to support secure contribution scoring, attack attribution, and potential recovery from model poisoning in both centralized and decentralized FL settings. BlockFL-SPACE supports various FL schemes and contribution metrics and derives a public validation set from private participant data using methods proposed in this paper. Our system allows secure contribution evaluation using any method based on local model predictions while maintaining privacy and robustness against tampering.
Low Earth Orbit (LEO) satellite networks have reshaped broadband access by extending connectivity to regions where terrestrial infrastructure is limited or unavailable. In tropical mountainous environments characterized by dense vegetation and abrupt elevation changes, however, the behavior of mobile LEO connectivity remains insufficiently explored. This paper presents an empirical performance assessment of Starlink connectivity along a round-trip route in the sub-equatorial region of Rio de Janeiro, Brazil, encompassing urban, semi-urban, and forested segments. A drive-test campaign was conducted to collect throughput, latency, jitter, packet loss, altitude, antenna orientation, obstruction time, and power consumption data under real mobility conditions. Beyond descriptive and correlation analyses, multivariate regression was employed to investigate partial effects and assess potential confounding between altitude and antenna obstruction. Results show uplink rates around 30 Mb/s, stable latency between 20 and 40 ms, and highly variable downlink throughput ranging from near-zero to peaks above 250 Mb/s. The findings provide practical insights into the performance limits and reliability of LEO connectivity in rugged sub-equatorial scenarios, contributing evidence on the operational challenges of mobile satellite access in complex topographies.
The Domain Name System (DNS), one of the pivotal components of the Internet, translates the domain name portion of a URL into its corresponding IP address, which is necessary to establish connections with websites. Initially, UDP and TCP were used as transport layer protocols to transmit DNS messages. The traditional unencrypted nature of DNS queries and their underlying protocols leaves them susceptible to interception and alteration, posing serious privacy and security risks. Recognizing these vulnerabilities, recent advancements in DNS technology have led to the development of encrypted alternatives DoT, DoH, and DoQ. Evaluating the performance of the encrypted DNS protocols is crucial to know their impact on DNS query resolution, which directly influences user experience. In this paper, we compare the three encrypted DNS protocols based on time-based metrics. To assess the performance of each protocol, a local server that supports all encrypted DNS protocols was created, and also, servers supporting all of the encrypted DNS protocols globally were considered. Nearly 80
This study demonstrates that intelligent reflecting surfaces (IRSs) can complement multiple-input multiple-output (MIMO) systems by reducing the number of vehicle antennas required to achieve the same spectral efficiency (SE) in realistic high-speed downlink communication systems. Building on previous analyses of SE with multiple IRSs in a single-input single-output (SISO) setup, which incorporated hardware impairments (HWIs) in line-of-sight (LOS) and Rician fading channels, this work considers selected Rician factor parameters. Further, it investigates the impact of varying Rician factors with distance, drawn from scenario-based simulations, providing a more realistic approach. The analysis is extended to MIMO configurations, exploring how IRSs can complement MIMO by varying the number of IRSs and MIMO configurations to reduce the number of vehicle antennas while maintaining comparable SE performance.
The increasing volume of network traffic and the growing sophistication of cyber-attacks pose challenges for the scalability and accuracy of Intrusion Detection Systems (IDS). A specific limitation is the difficulty of detecting intrusions and estimating the intensity of attacks in resource constrained environments, such as access networks with IoT devices. This article proposes a modular framework that encodes network flows into fixed size Bloom Filter matrices, enabling scalable and efficient learning through different modeling strategies. The methodology was validated using real traffic collected from a Brazilian broadband network. Four modeling approaches were evaluated under a unified training protocol: a baseline linear predictor, an XGBoost ensemble, a lightweight Simple CNN, and a Full CNN. The linear model exhibited limited performance, achieving an accuracy of 0.9012 and an R ^2 of 0.0103 . The XGBoost ensemble improved results substantially, reaching an accuracy of 0.9605 and an R ^2 of 0.5512 . The Simple CNN achieved near perfect classification with an accuracy of 0.9929 and an R ^2 of 0.6258 . The Full CNN delivered the highest overall performance, attaining an accuracy of 0.9929 and an R ^2 of 0.7824 , with the narrowest confidence intervals across ten independent runs. These findings confirm the effectiveness of Bloom filter based summarization for intrusion detection and demonstrate that the convolutional architectures best exploit the spatial structure of the encoded representation, offering a favorable trade off between accuracy and computational cost for deployment in heterogeneous network environments.
Predictive maintenance systems rely on data sharing across organisations, yet commercially sensitive information requires precise access control to prevent competitive disadvantage. Existing centralised mechanisms require blind trust among participants, creating significant barriers to collaborative machine learning in industrial settings. This paper extends SDDK-AC (Secure Decentralised Data and Knowledge Access Control for Predictive Machinery Maintenance), an access control mechanism that couples Attribute-Based Access Control policies with blockchain and smart contracts, by implementing contextual attributes for geolocation verification and data integrity via hash comparison. The mechanism runs on a Hyperledger Besu permissioned blockchain, integrated with Keycloak and an Access Control Proxy. This paper evaluates 30,000 policy decisions across 30 experimental rounds, each comprising 1000 transactions, using a custom-developed Python evaluation script. The results show that most SDDK-AC functions achieve throughput above 60 transactions per second with an average latency of 14 ms, incurring approximately 16
The rapid deployment of the new generation of mobile networks and terminals has significantly intensified public concern over potential health risks associated with electromagnetic field (EMF) exposure, particularly in the context of millimeter-wave frequency bands. The current measurement system based on point-wise sampling of electric field (E-field) strength in the given volume is inefficient at the frequency band. Therefore, super-resolution construction of the E-field from the low-resolution values is in great need. This study introduces a generative adversarial network (GAN) integrated with a field gradient branch and loss function to achieve super-resolution reconstruction of electric fields (E-fields), for the purpose of evaluating millimeter-wave (mmW) exposure. Utilizing a dataset created based on plane wave integral representation (PWIR) and randomized parameter incidence, the model effectively captures the wave propagation characteristics of diverse antennas. The incorporation of gradient information sharpens the E-field distribution details. When combined with cubic interpolation, this approach is validated for frequencies of 30 GHz and 60 GHz. Results by numerical validation show that this method achieves a maximum mean relative error (MRE) below 8
