Transactions on Emerging Telecommunications Technologies最新文献

The integration of Vehicular Ad Hoc Networks (VANETs) has changed intelligent transportation systems by making it possible for vehicles, roadside units (RSUs), and traffic management infrastructure to talk to each other in real time. This feature makes the roads safer, more convenient for drivers, and more efficient for traffic, but it also makes VANET ecosystems vulnerable to many types of cyberattacks, including Denial-of-Service (DoS) and false data injection, which can be very dangerous for safety and privacy. Conventional security solutions frequently struggle to address the highly dynamic, decentralized, and latency-sensitive characteristics of VANET environments. Intrusion Detection Systems (IDS) powered by Artificial Intelligence (AI) have become promising solutions, but there are still issues with computational overhead, secure model updates, and data privacy in distributed vehicular networks. To address these challenges, we introduce Quantum Lightweight Federated Learning (FL), an innovative hybrid machine learning framework that integrates the exponential computational power of Quantum Computing (QC) with the decentralized, privacy-preserving advantages of FL. The suggested method combines knowledge distillation with the FL process to create a lightweight detection model that works well on vehicle nodes with limited resources. Moreover, QKD Encryption is used to protect model parameters during federated aggregation, making sure that end-to-end privacy is maintained without slowing down processing. Lastly, SHAP, an Explainable AI method, is used to make sense of the choices made by the proposed model. Using the CICDDoS-2019 dataset for experimental validation shows that the proposed model is strong, with an accuracy of 99.36%, a high recall rate of 99.53%, and a precision rate of 99.38% across different attack scenarios.

As Internet of Things (IoT) technologies continue to advance, their adoption in smart cities, healthcare, and smart grids has increased significantly. Wireless sensor networks (WSNs) serve as a key enabling technology for IoT-based data monitoring and transmission. An IoT-integrated WSN (IWSN) involves the deployment of numerous sensor nodes in heterogeneous and challenging environments, necessitating efficient and reliable communication mechanisms. A significant issue that requires urgent attention is the potential for security breaches, such as intrusions within WSN traffic. Ineffective intrusion detection can lead to excessive energy consumption by Sensor Nodes (SNs), potentially causing node failures and resulting in diminished network coverage and overall lifespan. Detecting such attacks has led to considerable computational complexity in the existing research. Considering the limited resources of SNs and their deployment in challenging environments, it is essential to design clustering and routing protocols for WSNs that emphasize energy efficiency and security. This study aims to address these issues by developing a clustering and routing protocol that enhances energy efficiency while ensuring robust security and integrating intrusion detection to boost network longevity and data integrity. Initially, clusters have been formed using the fuzzy clustering means (FCM) algorithm. The crested porcupine optimization (CPO) technique is then used to select the optimal cluster heads (CHs). Following the clustering process, an adaptive secretary bird optimization algorithm (ASBO) is used to select the most efficient data transmission routes between the clusters, thereby, the network's energy efficiency is increased. Finally, to enhance the security of clustered WSNs, an advanced intrusion detection system (IDS) based on a multilevel attention dilated residual neural network (MADR-Net) has been used to detect and mitigate network intrusions. The experimental findings indicate that the proposed method surpasses the existing techniques across various performance metrics. Quality of service (QoS) parameters are measured using a packet delivery ratio (PDR) of 98%, dispersion value of 0.1133, end-to-end delay (E2ED) of 45 ms, and energy consumption of 23 J. The MADR-Net algorithm has outperformed the existing algorithms by achieving 98.5% accuracy on the CICIDS-2017 dataset and 98.8% accuracy on the NSL-KDD 2015 dataset.

Due to the emergence of cloud and its functions, it provides adaptable and dynamic scaled computing power used at a reasonable price. Efficiently assigning tasks with high computational resources to the cloud server is a critical problem that must be addressed to enhance system efficiency and ensure the satisfaction of cloud users. This process involves the consideration of factors such as the availability of resources, task priority, and dependencies among tasks. Despite the fact that there are numerous task-planning algorithms, current methods mostly concentrate on reducing the total time taken to finish while disregarding load balance. With the aid of available assets, cloud computing has proven to be an effective technique for providing services to the customers. Due to the heavy stress on the assets, the network performance eventually suffers. One of the more challenging factors in the cloud is the effective use of the available computing power. This necessitates the creation of a task-scheduling approach that is effective and efficient; thus, it has the potential to significantly impact the online computing system's functionality and performance as a whole. In a dynamic way, the scheduling process becomes critical while changing the environmental structure and managing the virtual computers in an optimal manner. Though several models are implemented for improving scheduling tasks in cloud environments, the problem is still unresolved. Additionally, the manual scheduling does not provide a feasible solution. To combat the above difficulties, a novel task scheduling process is to be carried out. Hence, a new optimal task scheduling model is developed using hybrid optimization algorithms for assigning the tasks to the machines on several assumptions. Here, the task scheduling process is executed via the hybridization of Iterative Concept of Drawer and Mother Optimization (ICDMO). This optimization algorithm allocates the tasks after checking whether the machines are in an idle state or not. During the task scheduling, a few multi-objective constraints like makespan, energy, cost, active servers, throughput, and resource utilization are considered to enhance the performance. The developed model's efficiency is determined with the conventional task scheduling approaches to find the effectiveness of the developed task scheduling model.

In recent years, there has been a continuous growth in the demand for intelligent, decentralized and transparent systems in the field of sports training management. The combination of blockchain technology and multi-agent systems has provided secure and trustworthy technical support for autonomous training management scenarios. This paper proposes a Blockchain-based Autonomous Smart Contract Framework for Sports Training Management (BASM-STM), aiming to solve key problems such as low operational efficiency, lack of trust mechanisms, and inflexible coordination mechanisms that are commonly found in traditional training platforms. This framework adopts a six-layer architecture design, achieving a deep integration of modular smart contracts, multi-agent collaborative decision-making, and dynamic trust evaluation mechanisms. Its core innovation contributions are as follows: (1) Builds a modular smart contract system based on the Solidity language; (2) Designs a dynamic trust update mechanism driven by Bayesian networks; (3) Proposes a hybrid coordination engine of genetic algorithm and reinforcement learning. Experimental results show that compared with traditional centralized solutions and existing decentralized benchmark models, the comprehensive performance of BASM-STM has been significantly improved: Orchestration latency reduced by $$, the session matching accuracy, fault tolerance under 5 node failures, and data tamper detection rate are all improved to varying degrees. The above experimental results verify the technical feasibility and system robustness of BASM-STM in secure and trustworthy, intelligent and efficient sports training management scenarios.

As more individuals use public cloud networks for data storage and handling, providing the security, integrity, and confidentiality of private data becomes essential. Classical encryption solutions are unable to address the growing threats and large-scale data processing required in such environments. Therefore, a Blockchain-Based Framework for Secure Cloud Data Encryption Using Heterogeneous Bi-Directional Recurrent Neural Network (BCF-SCDE-HBDRNN) is proposed in this paper. The input data is gathered from the IDS 2018 Intrusion CSVs Dataset. Initially, the data is encrypted using the Martino Homomorphic Encryption Algorithm (MHEA) for data security. The data is then secured using the Fair Proof-of-Reputation Consensus Algorithm (FPoR) based on blockchain technology for secure data storage. A Multi-Agent Cubature Kalman Optimizer (MACKO) is employed for optimal key management. Finally, Heterogeneous Bi-Directional Recurrent Neural Networks (HBDRNN) are used for threat detection such as normal and attack. Experimental evaluation demonstrates that the proposed framework enhances encryption efficiency, strengthens key management, and provides highly reliable threat detection compared to existing methods. The overall results highlight the framework's effectiveness as a robust and scalable solution for secure cloud data protection.

Ensuring people's safety in public places is a significant challenge for administrations today. The importance of automated crowd-monitoring systems has recently expanded beyond their role in addressing security concerns in densely populated areas. These systems have become increasingly vital for safeguarding human lives by helping to mitigate the spread of lethal infectious viruses, such as H3N2, SARS-CoV-2, Influenza, and COVID-19. Artificial intelligence (AI) has added a new dimension to this effort by addressing novel and real-world human safety challenges through automated crowd-monitoring frameworks. The proposed AI framework for crowd surveillance (AIFCS) employs a deep C2DN network to count people and issue warning signals for images exceeding a specified crowd threshold. Four datasets, including three publicly available ones (Mall, Beijing-BRT, and SmartCity) and one self-constructed dataset (Indiana), were used to evaluate the alarm-based congestion monitoring efficiency. The people-counting results for highly crowded frame detection accuracy on the Mall, Beijing-BRT, SmartCity, and Indiana datasets were 98.21%, 86.23%, 75.0%, and 87.01%, respectively. The proposed AIFCS framework ensures real-time predictions across diverse sequences to prevent overcrowding in public places.

Nowadays, we increasingly encounter with highly complex real-world optimization problems across various domains, including engineering, economics, healthcare, and artificial intelligence. Finding optimal or near-optimal solutions to these problems remains a significant challenge. In the existing body of literature, numerous stochastic-based optimization algorithms have been proposed to address such issues. However, ensuring consistent efficiency, robustness, and convergence across diverse problem landscapes remains an important concern. This paper introduces a novel and effective optimization algorithm called the Recruitment-Based Optimization Algorithm (RBOA), which draws inspiration from institutional recruitment and hiring process. The algorithm simulates the dynamic interactions and decision-making mechanisms involved in the selection of internal and external candidates during the recruitment process. Balancing exploration and exploitation is essential for any optimization approach and is achieved through the modeled behaviors of these two candidate types. External candidates facilitate global exploration, while internal candidates enhance local exploitation, together ensuring a comprehensive search of the solution space. Furthermore, the proposed RBOA has been effectively applied to an intelligent attack detection framework for Vehicular Ad Hoc Networks (VANETs), where it optimizes feature selection and classification parameters to enhance detection accuracy and reduce false alarms. In real-world validation for VANET attack detection, RBOA achieved 97.38% accuracy and a false-positive rate of 0.031, demonstrating its practical effectiveness in securing vehicular communications. To rigorously validate its performance, numerous benchmark functions have been used to test RBOA, encompassing multimodal, unimodal, and fixed-dimensional optimization problems. Comparative analysis with 11 well-established optimization algorithms reveals that RBOA consistently outperforms the compared algorithms.

Real-world, unpredictable conditions for autonomous vehicles continue to be a challenge. This study presents a Python-developed “Redundancy Net” as a multi-model ensemble method in improving the autonomous vehicle systems in perception, decision making, fault tolerance, adaptability, and robustness. Evaluated with the KITTI Vision Benchmark Suite, the framework was developed through the integration of stereoscopic cameras, LiDAR, GPS, and IMU data fusion with the implementation of hardware sensor fusion. For analysis, redundancy net data pre-processing sequence normalization, voxelization, Kalman filtering, and sequence temporal alignment, the multimodal systems. Redundancy net uses the complementary features of CNNs for spatial perception, LSTMs for sequence learning, and transformers for global spatio-temporal integration. In the redundancy and reliability layer, model predictions are dynamically fused with “confidence” based weighting through sensor faulted downgrades ensuring dependable outcomes. A fail-safe decision module with Monte Carlo dropout and entropy-based uncertainty evaluates system low confidence outcomes and activates safety holds. The Online Validation and Self-Adaptation mechanism improves and enhances Self-Adaptation mechanisms in real time by adjusting a model's parameters based on continuous performance evaluations and feedback. The experimental evaluations show that Redundancy Net provides a performance boost of 94.8% accuracy as compared to standalone CNN, LSTM, and Transformer models, which speaks to the framework's contribution to developing safe, flexible, and robust autonomous navigation in complicated driving environments.

This research introduces an AI-powered adaptive spatio-temporal anomaly detection framework (AST-ADF) to address vulnerabilities in vehicular ad hoc networks (VANETs), which are susceptible to spoofing, message tampering, and denial-of-service attacks. The framework integrates three intelligence layers: (i) spatio-temporal feature extraction using graph convolutional networks to process mobility traces, channel states, and packet statistics, (ii) an adaptive deep detection layer that uses a hybrid CNN-BiLSTM model for capturing sequential dependencies and local fluctuations in communication channels, and (iii) a self-learning anomaly refinement module using reinforcement learning for dynamic detection updates. AST-ADF reduces false alarms, adapts to environmental changes, and provides robust detection against zero-day anomalies, outperforming static models. Simulation experiments in real VANET environments show over 96% detection accuracy with only 3% false positives. The framework's minimal computational overhead makes it suitable for vehicular edge devices and roadside infrastructure. AST-ADF demonstrates 97%–95% accuracy, 2.8%–3.5% false positives, 96%–93% precision, 97%–94% recall, and strong zero-day adaptability. Unlike previous AI-based detection frameworks, AST-ADF unifies spatial and temporal correlations, significantly improving robustness against noise and adversarial interference. Additionally, it supports real-time deployment with efficient inference through model pruning and edge optimization techniques, ensuring secure, reliable, and intelligent VANET communication for future transportation systems.

Reliable road-condition monitoring is essential for safe and efficient transportation. This study proposes a deep learning-based dynamic channel allocation (DLDCA) framework for Vehicular Ad Hoc Networks (VANETs) integrated with Fiber Bragg Grating (FBG) sensors to support real-time detection of cracks, potholes, and structural defects. The framework employs a Deep Reinforcement Learning (DRL) model that continuously adapts channel assignments based on traffic density, interference, signal strength, and latency constraints, enabling efficient spectrum usage under varying network conditions. Simulation results demonstrate significant gains over static allocation approaches, including a 90% increase in packet delivery rate, 25 Mbps improvement in throughput, 65% rise in channel utilization, and a 40 packets/s reduction in congestion. The end-to-end delay is consistently maintained below 80 ms. These outcomes confirm that the DLDCA framework enhances communication reliability and supports proactive road maintenance in next-generation intelligent transportation systems.