This paper presents a simulation based evaluation of a blockchain supported multisector platform for sustainable smart city ecosystems. The proposed model integrates sustainable urban mobility with health and cultural services through token based incentives and smart contracts. An agent based simulation with 10,000 heterogeneous agents over a 30-day period was combined with a simulated permissioned blockchain layer. Three scenarios were compared: a baseline without incentives, a linear incentive model rewarding green transport, and a cross sector ecosystem model linking green kilometers with health, cultural, and tourism benefits. The results show that token based incentives can increase the use of sustainable transport modes, reduce CO₂ emissions, and improve value circulation across sectors. The cross sector scenario achieved the strongest ecological, economic, and social effects. The study demonstrates that blockchain can support Triple Bottom Line sustainability by connecting emission reduction, automated value distribution, and citizen oriented services within a unified smart city ecosystem.
To enhance user Quality of Experience (QoE) and avoid rebuffering caused by bandwidth degradation in short video applications, it typically sets up a buffer to preload videos in the recommended queue when bandwidth is sufficient. However, users exhibit unique "early departure behavior" during short video browsing, which significantly increases the likelihood of bandwidth waste. Existing methods struggle to balance between QoE and waste rate. This paper proposes a preload framework for short video applications, which includes bandwidth prediction, buffer threshold, and preload decisions. The preload problem is modeled as a stochastic multi-stage optimization problem, and the revenue is divided into immediate decision revenue and future state value. Immediate decision revenue is estimated by predicting user behavior. Solving subproblems at each stage produces a series of fitted hyperplanes for the state value function. The expected state value is then calculated by solving these affine functions, which ultimately determines the optimal loading sequence and bitrate. Experimental results show that, compared to several common algorithms, the proposed framework significantly reduces bandwidth waste while improving QoE.
This paper proposes a fine-grained performance analysis framework for multi-antenna multi-user networks with interference nulling (IN), where the success probability (SP), the variance of link reliability, and the signal-to-interference ratio (SIR) meta distribution are studied as the metrics for system performance evaluation. With these three metrics, the average performance, the fairness of the individual links, and the link reliability distribution in the network can be respectively investigated. Two IN schemes, namely, fixed IN (FxIN) scheme and flexible IN (FlIN) scheme are studied using the framework. First, the expression of SP for each scheme is derived by using stochastic geometry analysis techniques. Then, the approximate expressions for the first and second moments on the link reliability are obtained, with which the variance of link reliability as well as the approximation of the SIR meta distribution for each scheme is derived. Finally, the performance of the two IN schemes are compared within the framework. It is shown that the FlIN scheme always provide higher fairness among individual links compared to the FxIN scheme. Nevertheless, the superiority of the SP and the SIR meta distribution of the two schemes are dependent on the system parameter design.
Building information modelling (BIM) is now routine in how construction firms design and document projects; the open question is what artificial intelligence (AI) can add on top of that foundation. Little is known about what moves a firm from BIM to AI, particularly in smaller Central European markets and in small and medium-sized enterprises. The paper asks three questions: how widely BIM is used across project stages, whether that use shows in cost, waste, and sustainability figures, and what stands in the way of AI adoption. Evidence comes from two surveys grounded in official records: one covering firms in Slovakia, Croatia, and Slovenia, the other on Slovak firms’ AI experience and obstacles. Responses were analyzed with descriptive statistics, Kruskal-Wallis tests, and Spearman correlations, with the BIM-sustainability link cross-checked against a convergent model. BIM maturity differs clearly between the three countries, from a mean of 1.33 in Slovakia to 2.49 in Croatia and 4.33 in Slovenia on a five-point scale (H = 86.77, p < 0.001), and firms using BIM more report better cost, waste, and sustainability outcomes (ρ up to 0.93 for material cost, 0.92 for emission reduction). AI use remains rare. What distinguishes AI-ready firms is digital maturity (ρ = 0.49 for digital-tools use, 0.37 for digitalization) rather than size (ρ = 0.01, not significant), and the barriers cited most concern people and skills rather than technology. With the regional sample in mind, the findings point to a solid digital and BIM base as AI’s realistic starting point.
With the rapid advancement of sixth-generation wireless communication systems (6G), traditional fixed-infrastructure-based mobile edge computing (MEC) struggles to meet performance demands such as ultra-high reliability, ultra-low latency, and high data rates in scenarios with sparse communication facilities or sudden disasters. The space-air-ground integrated network (SAGIN), leveraging its extensive coverage capabilities and high communication capacity, offers a novel solution featuring global seamless coverage, low latency, and efficient energy utilization. This effectively addresses the limitations of MEC in large-scale, dynamic environments. Specifically, by integrating heterogeneous devices such as satellites, unmanned aerial vehicles (UAVs), and ground stations, SAGIN supports coverage across extremely vast areas, effectively compensating for the shortcomings of traditional terrestrial networks in remote regions and disaster scenarios. However, SAGIN’s high dynamism, device heterogeneity, and distributed deployment introduce complex resource scheduling and task allocation challenges. Against this backdrop, digital twin (DT) technology provides critical support for network optimization and management by delivering precise real-time mapping of physical systems. We designed and implemented a digital twin-based integrated air-ground-space network simulation platform (DT-SAG) to achieve collaborative modeling and intelligent scheduling of satellites, UAVs, and ground stations. To address the collaborative control and resource optimization challenges in SAGIN, we combined the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm with a federated learning (FL) framework. The MADDPG algorithm resolves the instability issues of traditional algorithms in complex multi-agent environments through centralized training and distributed execution. Meanwhile, FL reduces communication overhead, ensures data privacy, and enhances training efficiency by performing local training with global parameter aggregation. By integrating these techniques, the proposed FLMADDPG algorithm enables efficient collaboration among agents in SAGIN while avoiding the data privacy and communication bottlenecks associated with centralized learning. Experimental results demonstrate that the proposed method effectively enhances network coverage efficiency, energy management, and communication performance, showcasing the immense potential of integrating digital twins with federated learning.
Information sharing between patients and medical officers is the core of wireless communication in the medical field. Wireless technology and the Internet of Things (IoT) form the Wireless Body Area Network (WBAN). The WBAN sensors can thoroughly track the patient’s biological factors from head to toe and communicate with the destination, such as doctors, caretakers, family, etc., for remote diagnosis and prediction. Maintaining data integrity is crucial for precise diagnosis, necessitating robust protection of patient’s biological data against malicious threats. Automated anomaly detection of WBAN data is required to ensure data integrity in real-world scenarios efficiently for high-speed diagnosis. The contribution of the work lies in creating a wearable WBAN model, performing anomaly detection using Machine Learning (ML) and automated ML (AutoML) models, and analyzing their performance for real-world data. The smartwatch wearable device is deployed in this work to estimate vital signs like heartbeat, pulse, oxygen intake/exhalation, blood pressure, and oxygen saturation. The dataset generated with these essential signs is trained and tested with the ML classifiers to diagnose whether the generated WBAN data is healthy (normal) or unhealthy (abnormal). Traditional and auto ML are used to evaluate the biological dataset for anomaly detection. The novelty of the work is the real-time data used for anomaly detection using the AutoML Tree-based Pipeline Optimization Tool (TPOT) classifier algorithm, which generates a pipeline for future data classification. The complexity and processing overhead of anomaly detection using TPOT is reduced compared to traditional ML. The various performance metrics, such as precision, accuracy, recall, and F1 score, are compared between conventional and auto ML classifiers. TPOT is executed, resulting in the best XGB classifier pipeline with a test accuracy of 98.91
Short-term radar-based precipitation nowcasting is a critical component of real-time early-warning systems, especially in regions like Northwest Vietnam where the impacts of climate change are increasingly severe. Yet achieving reliable predictions several steps ahead while still preserving the spatial organization of rainfall patterns remains a substantial challenge for most existing approaches. This study introduces SR2-GAN, a Generative Adversarial Network tailored for multi-step radar nowcasting. The generator is built upon a ConvLSTM backbone enhanced with refined CBAM attention modules to better capture the dynamic features of rainfall. In parallel, the discriminator evaluates outputs at the patch level, encouraging the model to produce sharper and more coherent structures. A key aspect of the proposed framework is a composite multi-objective loss that jointly leverages adversarial learning, MSE, and SSIM. This design allows the network to balance pixel-level fidelity with the preservation of morphological characteristics. Experiments conducted on radar data from the Pha Đin station (Vietnam) and the DWD dataset (Germany) show that SR2-GAN significantly reduces MSE by 44.4
With the rising demand for 5G networks, there is a demand for efficient strategies that allocate resources to ensure optimal Quality of Service (QoS), bandwidth management, and congestion control. However, the traditional rule-based approach does not suit the complex and fluctuating 5G environments. This study addresses these challenges by introducing a classification-based machine learning (ML) framework for intelligent 5G resource allocation and benchmarking different classification models such as CNNs, SVMs, Decision Tree, and Ensemble Learning (Bagging Boosting). The training and evaluation were performed using a real-world 5G network dataset of key performance indicators (KPIs), namely latency, signal strength, allocated bandwidth, and the level of network congestion. The data augmentation follows Hybrid Adversarial Sampling (HAS), which preserves the realizable network traffic fluctuations for robust model generalization. Standardized performance metrics like accuracy, precision, recall, F1-score, and sensitivity were applied to assess the classification models. Of all the classifiers, the CNN-based model gained an accuracy of 98.7
Ascon, the winner of the NIST Lightweight Cryptography competition, was standardized in 2025 as a lightweight cryptographic primitive for securing the Internet of Things and other resource-constrained devices. Although Ascon is well suited to embedded deployment, its implementation security against fault attacks remains insufficiently understood. In this paper, we propose differential fault analysis (DFA) methods based on 5-bit fault injections and the differential properties of the Ascon S-box. We introduce the concept of fault trails to characterize the evolution of candidate S-box input sets under 5-bit faults. Based on this framework, two fault models are developed to recover the intermediate state in the Finalization stage and ultimately the 128-bit key. Furthermore, we derive the expected numbers of fault injections required to recover a single S-box input under fault models A and B, and simulation results are shown to be in close agreement with the theoretical values. In addition, we consider several fault-injection settings that better reflect practical attack scenarios, including non-uniform fault distributions, different fault invalidation probabilities, and different fault widths. Under fault model A, 391 5-bit faults or 13.3 register-width faults are required on average to recover the 128-bit key. Under fault model B, the corresponding numbers are 289 and 13.1, respectively, outperforming the existing DFA results on Ascon.
Intrusion Detection Systems (IDS) play a crucial role in safeguarding computer networks against malicious activities. However, current IDS models based on machine learning (ML) and deep learning (DL) encounter challenges in accurately classifying malicious network traffic, especially when confronted with adversarial perturbations and the transferability of adversarial attacks. To address these challenges, numerous studies have proposed and demonstrated the effectiveness of a novel multimodal approach, which leverages multiple sources of information to achieve better accuracy and the ability to identify more unknown attacks. In this study, we evaluate the robustness of multimodal learning-based IDS against transferable adversarial examples (AEs) generated by Generative Adversarial Networks (GANs). Additionally, we integrate adversarial training techniques to enhance the IDS’s capability to identify attack patterns with small perturbations. Our proposed strategy, MAT, achieves near-perfect detection rates across all attack types and the highest overall F1 score (0.7595), substantially outperforming all baselines under second-round adversarial attack conditions. While the attained results may not be as remarkable as desired, implementing the Multimodal approach makes a notable contribution to the field, paving the way for further research and advancements in addressing the challenges encountered by existing IDS models.
Efficient IP address lookup is a growing challenge due to the rapid expansion of routing tables and evolving network demands. Traditional methods like Ternary Content Addressable Memory (TCAM) offer fast lookups (5 ns) but are limited by high power consumption (5.5 W), poor memory efficiency (500 MB), and scalability constraints (512K entries). Trie-based alternatives such as General and Multi-bit Tries improve scalability but suffer from slower speeds (50–80 ns) and high memory overhead (250 MB). This study introduces an adaptive enhancement to the Vector Coupled Trie (VCT), which dynamically reorganizes trie nodes based on real-time lookup patterns and incorporates Bloom Filters to minimize redundant memory access. The proposed model achieves a 70
This paper proposes an Enhanced Deep Reinforcement Learning Approach (EDRLA) for spectrum allocation in Cognitive Radio Networks (CRNs) to address spectrum scarcity and improve spectral efficiency. The approach enables secondary users to exploit underutilised spectrum holes left by primary users while minimising interference. EDRLA integrates Deep Reinforcement Learning (DRL) with an Adaptive War Strategy (AWS) to leverage historical channel data to learn both channel correlations and temporal dynamics. The Q-table in DRL is updated through AWS, allowing accurate identification and allocation of available spectrum. The method is evaluated through performance and comparative analyses with conventional techniques, including Enhanced Threshold Energy Detection (ET-BED), Whale Optimisation Algorithm (WOA), and Grey Wolf Optimisation (GWO). The results demonstrate that EDRLA achieves superior performance, with an average throughput of 2.8 Mbps and an energy efficiency of 4.0, highlighting its effectiveness in improving spectrum utilisation. The proposed approach combines reinforcement learning with an adaptive strategy, offering an efficient solution for dynamic spectrum management in CRNs.
Coverage hole is a significant challenge in the Industrial Internet of Things (IIoT) which leads to system disconnection and the inability to monitor critical areas. Effective device deployment is crucial for addressing this issue. While various heuristic, metaheuristic, and machine learning-based approaches were proposed, few pieces of literature address the challenges of real-world IIoT environments with non-penetrable obstacles. To deal with this gap, we introduce a novel strategy for device deployment in IIoT monitoring environments with potential obstacles, called Artificial Bee Colony-based Device Deployment (ABC-DD). The primary objective of ABC-DD is to deploy IIoT devices in a manner that ensures comprehensive system coverage with a minimum number of devices. By using the strengths of the Artificial Bee Colony (ABC) algorithm, ABC-DD optimizes performance criteria, including device overlap, overlap between devices and obstacles, overlap with areas outside the monitoring environment, and inter-device distances. Our strategy enhances IIoT coverage to an average of 83.68
Intelligent vehicles necessitate offloading compute-intensive tasks to proximal edge servers or cloud infrastructures, where accurate end-to-end (E2E) latency prediction becomes essential for optimal offloading destination selection. However, the inherent complexity and stochasticity of vehicle-to-everything (V2X) scenarios pose significant challenges for precise network latency forecasting. To address this issue, an interval prediction method combining variational mode decomposition (VMD) and a non-parametric estimation-based module is proposed in this paper and applied to computational offloading scenarios. This method mitigates noise in raw data and reduces stochastic fluctuations by decomposing signals into intrinsic mode functions (IMFs) via VMD, which enhances prediction accuracy. Error distributions are subsequently modeled using kernel density estimation (KDE) to generate confidence intervals, which improves prediction reliability. Additionally, a centralized architecture leveraging the network data analytics function (NWDAF) is designed to facilitate computational offloading decisions. Simulation results show the proposed method achieves 1.721
Model extraction attacks enable adversaries to replicate the functionality of black-box models deployed as AI services in edge-cloud mobile and ubiquitous systems using only query access, posing serious security and intellectual property risks. In practice, model providers often disclose the list of supported classes and representative reference images to document capabilities and facilitate integration, inadvertently exposing exploitable vision-language semantic information. However, existing attacks exploit only single-modal cues and mainly rely on short-horizon query feedback to guide subsequent optimization, leaving the available information underutilized. We propose VLSMEA, a Vision-Language Semantics Guided Model Extraction Attack, which constructs prompts with a class name-anchored prefix and a learnable suffix, and optimizes them via long-horizon contrastive prompt learning in a shared vision-language space. A semantic scoring mechanism samples hard positives and negatives from long-horizon query histories to drive an alignment-repulsion objective, improving the efficiency and diversity of synthetic queries under a fixed budget. Experiments on five benchmarks with a 30K-query budget show that VLSMEA outperforms prior baselines in clone accuracy and significantly strengthens downstream adversarial transferability against victim models. These results demonstrate that, when combined with exposed semantic metadata in deployed AI services, the growing capabilities of vision-language models can substantially amplify the practical threat of model extraction against black-box deployed vision models, calling for strengthened defenses to improve privacy, integrity, and trustworthiness in secure AI deployment. Our source code is available here .
The primary objective of this study is to investigate the impact of base station placement on network efficiency through experimental analyses conducted on Random Entity Mobility Models (REMMs), which are commonly employed for mobile data collectors. Two widely adopted random mobility models from the literature Random Walk (RW) and Random Waypoint (RWP) are compared with two novel models -Random Point (RP) and Random Journey (RJ)- proposed by the author. The study is grounded in performance analyses conducted across varying area sizes and fixed node counts. The findings clearly demonstrate that base station location plays a critical role in determining model efficiency, and that identifying the optimal position significantly enhances system performance. This research makes an original contribution by offering a comprehensive evaluation of the most effective base station placements for four distinct mobility models. Among them, the RP model consistently achieves the highest efficiency across all simulation scenarios, underscoring its superiority in performance. Furthermore, the evaluation of base station positions reveals that while the central location yields the highest efficiency, random and right-side placements also emerge as viable alternatives. The study also highlights how the optimal layout evolves with an increased number of base stations and how these arrangements contribute to overall system performance improvements. By systematically analyzing the relationship between base station placement strategies and efficiency in mobile data transport systems, this study offers significant contributions to the field of REMMs. It also establishes a solid foundation for future research, particularly in scenarios involving multiple base stations and the exploration of effects in three-dimensional environments.
In multiprocessor Mixed-Criticality System (MCS), tasks must share resources (e.g., memory, I/O devices). Traditional synchronization mechanisms (e.g., suspension locks, spin locks) suffer from priority inversion and remote blocking issues in multiprocessor environments. Existing protocols (e.g., MSRP, FMLP, MrsP) optimize resource contention through mechanisms like priority ceiling and priority inheritance. However, these protocols primarily target static MCS and struggle to adapt to dynamic criticality mode switching in MCS. Since in MCS with escalating resource contention level such as autonomous driving systems: high resource contention levels may lead to the discarding of all LO-Criticality (low criticality) tasks, reducing system utilization. In this paper, we propose a Switchable Protocol Framework (SWPF), which achieves adaptability to resource contention levels by dynamically switching between optimistic and pessimistic locking protocols. Experimental results demonstrate that SWPF outperforms traditional FMLP with an average performance improvement of 5.5
The rapid proliferation of the Internet of Things (IoT) across diverse domains has intensified cybersecurity concerns, particularly within lightweight communication protocols such as Message Queuing Telemetry Transport (MQTT), whose publish–subscribe architecture makes it a frequent target for cyberattacks. Motivated by the growing limitations of conventional intrusion detection systems (IDSs) in handling dynamic IoT environments, this study develops an efficient model that enhances detection accuracy while minimizing inference time. A hybrid intrusion detection framework that integrates the Adaptive Neuro-Fuzzy Inference System (ANFIS) with the Sailfish Optimization Algorithm (SOA) is proposed to achieve this objective. The proposed system is validated using uni-flow and bi-flow versions of the MQTT-IoT-IDS2020 dataset after dimensionality reduction via Principal Component Analysis (PCA). Experimental results demonstrate that the ANFIS–SOA model outperforms traditional classifiers, including Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Multi-Layer Perceptron (MLP), and Naïve Bayes Classifier (NBC), achieving an accuracy of 99.7
Industrial revolutions have always marked breakthrough periods that fundamentally changed the way of production, work organisation and the overall functioning of industry. From the first mechanised machines in the 18th century to today’s intelligent technologies, industry has undergone enormous changes affecting production, logistics, transport and trade. Each industrial revolution has brought new technologies that have helped businesses become more efficient and better respond to market demands. The research and the submitted manuscript aimed to examine different ways of managing the supply chain through innovative information technologies and to analyse their effectiveness and risks at individual stages of the supply chain. The study also focused on identifying optimal ways of implementing these technologies and their possible impacts on the performance and competitiveness of the company.
