International Journal of Communication Systems最新文献

Lightweight image encryption for the Internet of Things (IoT) enables secure image data transmission between connected devices while accommodating the limited processing memory and power of IoT systems. A general drawback of lightweight image encryption is that it often sacrifices security strength for efficiency, making it more vulnerable to attacks compared to more robust encryption methods. In this manuscript, the development of a lightweight image encryption algorithm for ensuring confidentiality and privacy in Internet of Things devices (DLWIE-ECP-IoTD) is proposed. Firstly, the input image is gathered from the Caltech-101 dataset. Then, a chaotic pattern generation using a one-dimensional discrete chaos mapping system is applied to generate unpredictable patterns to secure image encryption, followed by a chaotic model of information encryption used for securing the image through unpredictable transformations. Then, a Uniform Physics Informed Neural Network (UPINN) is adapted to produce pseudo-random encryption keys from chaotic parameters, providing an efficient and secure key-generation mechanism. Finally, Fully Dynamic Advanced Encryption Standard (FDAES) is utilized for lightweight encryption of image data. The proposed DLWIE-ECP-IoTD approach is implemented in Python. The proposed DLWIE-ECP-IoTD approach achieves 1.120 s of computational time and 0.03% mean squared error (MSE) with existing methods, such as a lightweight multichaos-based image encryption scheme for IoT networks (LWMC-IES-IoTN), a lightweight image encryption scheme for IoT environments and machine learning-driven robust S-box selection (LWIE-IoTE-ML-DRSBS), and a lightweight image encryption scheme for the safe Internet of Things using a novel chaotic technique (LWIE-NCT-SIoT).

The Social Internet of Things (SIoT) allows smart devices to form social relationships for efficient service discovery and resource sharing. However, trust management is challenged by attacks like ballot stuffing and bad mouthing, which manipulate trust scores through biased recommendations. Existing approaches often evaluate recommenders based on their service provider role, overlooking their behavior as recommenders due to data sparsity. This paper presents a resilient trust management model using reinforcement learning. It combines multiple trust features—direct trust, service reliability, social ties, recommendation benevolence, and referred trust—to compute trust-based rewards. A filtering mechanism is also introduced to detect dishonest recommenders and reduce the impact of biased feedback. Experimental results demonstrate strong resilience against Bad mouthing attack (BMA), ballot stuffing attack (BSA), and Sybil attacks compared to state-of-the-art models, along with faster convergence on both SIoT/IoT network and Epinions datasets. These findings confirm the model's effectiveness in preserving trust and resisting attacks in SIoT environments.

A tunable terahertz (THz) multi-input, multi-output (MIMO) dielectric resonator (DR) antenna (DRA) with the self-diplexing ability is implemented. Two separated annular disc DRs with the top coated graphene material connected with two individual ports. And, these graphene coating can be connected to the separate DC power supplies to find the tunable response through individual port. The narrow-intensified resonance spectrum of annular disc DRs operating with the $��\mathrm{HE}�M_{�40�\delta �}�$ mode can provide the highly sensitive tunable MIMO response. This can also help in offering the tunable self-diplexing capability to antenna. The usage of DRs can provide the antenna with the high gain and efficiency in the THz frequency range. The antenna operation is validated using the circuit theory approach. Moreover, antenna operation is validated to work with the four ports and radiators, which can provide the operation with four port MIMO or self-multiplexing capability. The usage of array of the four radiators enhances the directivity significantly by reducing the minor lobe and confining the radiated power as in pencil beam. The multiport antenna operation is validated by calculating envelop correlation coefficient and diversity gain which remain around 0.06 and 10 dB in the passband, respectively. These prove the proposed antenna suitable for working in the multiport systems.

Big data are difficult to process because of its volume and frequent updates. Big data are used to predict network traffic, which allows for further analysis at the application level. Network traffic prediction is essential for effective network planning and management. Deep learning (DL) has emerged as an effective way of capturing complex spatiotemporal relationships, with graph neural network (GNN) models being especially popular in this area. Nonetheless, conventional GNN techniques have inefficiencies in long-term forecasting in network traffic prediction, resulting in suboptimal predictive performance. To overcome the difficulties in forecasting network traffic, an integrated deep graph neural network (DeepGNN) model is presented in this work. First, create an integrated learning module that takes advantage of spatial correlation. Furthermore, sequence convolutional neural networks (sequence convolutional neural network [CNN]) are used for nonlinear dependencies, whereas attention mechanism incorporation is designed for heterogeneous features. In this study, integrated DeepGNN is evaluated on two network traffic datasets, Milan and Trentino. Three services, SMS, call, and internet, are also included for evaluation services in the first dataset and cumulative services in the second. Integrated DeepGNN is compared with the various existing models considering mean square error (MSE), root mean square error (RMSE), and mean absolute error (MAE). The proposed technique achieves a 4.923 MAE rate, which is lower than other techniques. The performance of the proposed technique is analyzed and compared with some related techniques to describe the superiority of the proposed model.

Traditional Mobile Wireless Sensor Networks (MWSNs) often use mobile sinks to mitigate issues like energy holes. However, mobile sinks introduce new challenges, such as significant delays and buffer overflows due to fixed trajectories and variable moving speeds. These issues are particularly problematic for delay-sensitive applications, as most existing researches either focus on delay-tolerant scenarios or rely on energy-intensive greedy data collection methods. This research proposes an improved deep learning model with STGRN for delay-sensitive, energy-efficient routing and mobile sink prediction in MWSNs. STGRN-HRFO addresses mobility challenges in MWSNs through optimized routing, energy-efficient transmission, adaptive resource allocation, predictive mobility models, dynamic topology adjustments, and machine learning, enhancing stability, reducing energy consumption, and improving performance. Based on the projections from STGRN, an effective cluster-based routing system is implemented. A Hybrid Red-billed Frilled lizard magpie Optimizer (HRFO) is introduced to optimize the selection of cluster heads and improve routing efficiency. Significant gains are made using the STGRN-HRFO framework, which reduces the end-to-end path's hop count to 0.2 s, minimizes network energy usage to 3.6 J, and boosts throughput to 90%. Additionally, the energy consumed per packet is minimized to 2.2 mJ. Comparative analysis demonstrates that the STGRN-HRFO protocol effectively enhances network performance, ensuring low latency, high packet delivery ratios, and efficient energy use, particularly in real-world scenarios with complex optimization needs.

Location-based services (LBS) offer useful information based on a user's location, but they raise privacy risks since sensitive data can be misused by third parties. Traditional peer-to-peer (P2P) systems try to protect privacy but struggle to balance anonymization with service accuracy, and the problem worsens as the number of users and queries increases. These issues are overcome by introducing a novel Xception Convolutional Scaling Wide Residual Network (XCovSWideR-Net) model to improve location privacy in P2P systems through anonymization. The anonymization process involves the user, an anonymization server, and the LBS server. The anonymizer first hides the user's personal and location details and then adds dummy locations to mask the real query. Privacy is further improved using the XCovSWideR-Net model, which combines Xception Convolutional Network (XCovNet) and Scaling Wide Residual Network (SwideRes-Net). The anonymized query is sent to the LBS server, which returns the requested information to the anonymization server, and finally to the user without revealing their actual location. The XCovWideR-Net model achieved maximum location privacy of 0.967, location preservation of 0.974, anonymous entropy of 8.268, and a minimum computation time of 2.480 s for 800 users in Scenario 3. These findings highlight the ability of the proposed method to effectively balance privacy, accuracy, and efficiency, providing a promising solution for secure and scalable LBS applications.

The industrial Internet of Things (IIoT) depends on wireless sensor networks (WSNs) to enable low-power, low-data-rate communication in resource-limited settings. While the IEEE 802.15.4 standard provides the communication foundation, its medium access control (MAC) protocols face challenges including energy consumption, latency, scalability, and adaptability. Traditional MAC protocols cannot keep up with the demands of IIoT networks as the number of connected devices continues to increase. Therefore, edge artificial intelligence (Edge AI) and tiny machine learning (TinyML) represent emerging approaches that show potential for improving the performance of traditional MAC protocols directly on IIoT devices. Edge AI and TinyML allow intelligent decision-making at the edge, which enables efficient data processing and adaptability to the environment without the need for cloud infrastructure, which may reduce latency and energy consumption. This paper systematically examines the emerging paradigm of combining Edge AI and TinyML to improve MAC protocols for WSNs in IIoT networks. We explore advanced machine learning (ML) methods applicable to resource-limited devices, and we investigate how these methods can improve key performance metrics for MAC protocols, including energy efficiency, throughput, and network lifetime. We also discuss the challenges and limitations of applying AI solutions in WSNs, including computational constraints, data scarcity, and model scalability. Finally, we propose potential future research directions to improve the application of AI and ML techniques to develop more efficient, adaptive, and intelligent MAC protocols for future IIoT networks.

The intelligent transportation system (ITS) is enabled by the vehicular ad hoc networks (VANETs), but the security threats, such as node impersonation, node tampering, and eavesdropping, are the greatest challenges and cause security concerns within the system. The large-scale vehicular environment is not effectively handled by the previous static and centralized security approaches, which can greatly increase the latency and data integrity problems within the network. Thus, this research proposes a deep learning–assisted blockchain approach for enabling the decentralized, reliable, and secure communication in the VANET. The main contribution of the research is to perform secure authentication and task offloading to enable secure task offloading within the VANET and to guarantee communication performance with minimum energy consumption and delays. First, the data confidentiality, privacy of the task offloading, authentication, and integrity are achieved by introducing blockchain technology. Second, the node authentication is performed using adaptive and attention-based hybrid serial learning (AAHSL), which is developed with the combination of a deep belief network (DBN) and temporal convolution network (TCN). After authenticating the data within the nodes, the adaptive deep reinforcement learning (ADRL)–based task offloading is proposed for reducing the task completion time within the network. In both models, the parameters are tuned using the pattern improvement parameter–based poor and rich optimization algorithm (PIP-PROA). The experimental results demonstrate that the proposed approach achieves an FNR of about 3.46% during the authentication process, and the reward score achieved by the designed model during the task offloading process is 9.22. Thus, the effectiveness of the suggested model is confirmed by the experimental analysis.

Energy efficiency is one of the main factors that determine the durability and dependability of WSNs. The performance and lifespan of a network are usually limited by the small battery capacity of the sensor nodes. Traditional cluster–based routing protocols enhance energy efficiency by creating clusters and sending data to the base station (BS) through the group-head nodes. Nevertheless, if a node that is between the BS and its cluster head sends data through the cluster head instead of directly to the BS, redundant reverse transmission can happen, which results in unnecessary energy dissipation. Hence, a Path Direction-Sensitive Routing Protocol (PDSRP) is introduced to resolve this problem. It uses signal strength indicators to locate the best transmission paths dynamically. The new protocol is instrumental in equalizing energy consumption, cutting down on reverse communication, and improving the general routing efficiency level. According to simulation results, PDSRP is able to considerably reduce power wastage and increase the longevity of the network; thus, it is a viable solution for IoT-based WSN applications that is scalable and energy-efficient.