Due to network congestion, the uplink communication of local models is slow and unpredictable in cloud-based Federated Learning (FL), which will make it difficult to achieve the goal of Hyper Reliable Low Latency Communication (HRLLC) in the 6G era. To minimize communication latency and also to achieve a larger range of user participation, Hierarchical Federated Learning (HFL) has been proposed in academia. Nevertheless, HFL still faces many challenges, such as time-varying channels, user mobility, and data heterogeneity. To address these difficulties, we design a dynamic cell association scheme for multi-cell over-the-air computation-based HFL (MC-AirCompFL). This dynamic strategy innovatively integrates the channel state information (CSI) driving mechanism with the data distribution distance sensing technique to achieve dual-dimensional cooperative optimization. Firstly, we analyze the convergence behavior of MC-AirCompFL at different global communication rounds. Secondly, to relieve the pressure from imbalanced data and nonideal wireless channels, we minimize the optimality gap and data distributed distance by jointly optimizing the cell association and transmission power at user equipment (UE) and the de-noising factors at base stations(BSs). Finally, numerical results based on the MNIST datasets validate the superiority of the proposed scheme over the traditional cell association strategy in the multi-cell FL.
Cloud computing has transformed data storage and management, offering scalability, flexibility, and cost-efficiency for applications. However, ensuring data security and maintaining performance in distributed cloud environments remain a significant challenge. Sensitive data stored in the cloud is vulnerable to unauthorized access, raising concerns about confidentiality and integrity. One particular vulnerability is caused by file deletion operations in computer systems, which usually only remove index entries, with the content of the files remaining on the disk. This can pose a potential threat to sensitive information in the cloud storage. This paper addresses the challenge by proposing a shredding-based approach for effective file management. The proposed approach proactively shreds files into independently meaningless shreds when storing a file in the cloud, and then distributes these shreds across multiple cloud sites. When the file is deleted, it is much harder to recover the content of the original file. A mathematical model is developed to perform shredding and reconstruction efficiently. It is further enhanced by randomizing the shredding process and adding modifiers to improve security. The method addresses the security problem of files stored in the cloud from a different perspective, compared to traditional encryption-based methods. Performance evaluation demonstrated that the proposed method achieves a balance between security and efficiency. The method is further evaluated using FABRIC infrastructure, and the results show that the shredding overhead is insignificant, and the approach can be used in practical cloud-based systems.
With the widespread adoption of internet of things (IoT) devices, dynamic spectrum sharing emerges as a technology for the efficient utilization of spectrum resources. To achieve this, sensing spectrum usage and maintaining collected information are essential for effective allocation decisions. In blockchain-enabled dynamic spectrum sharing, each IoT device collects and stores sensing data locally to automate spectrum sharing in a decentralized manner. However, IoT devices with limited storage find it challenging to participate because each IoT device must keep its sensing data. In this paper, we adopt a distributed storage technology within the blockchain-enabled dynamic spectrum sharing to store sensing data without depending on the storage capacity of individual devices. Additionally, we propose a sensing data allocation algorithm based on the value of sensing data and availability of participating device storage.
Log anomaly detection plays a pivotal role in ensuring system stability and security, particularly in largescale environments characterized by the generation of log data at exceptionally high volumes and velocities. Conventional approaches often struggle to effectively filter log information and fully leverage temporal dynamics, resulting in challenges such as information loss, semantic drift, and heightened computational overhead. To overcome these limitations, we present QYXLAD, an innovative log anomaly detection framework. QYXLAD enhances log representation accuracy by seamlessly integrating temporal and semantic information. It introduces a MPMM(Multi-level Progressive Masking Mechanism)-based Feature Fusion designed to capture temporal dependencies and semantic features across diverse pattern combinations, thereby significantly improving the sensitivity and precision of anomaly detection. Furthermore, QYXLAD utilizes a Mamba-based classifier for anomaly identification. Comprehensive theoretical analysis and empirical evaluations demonstrate that QYXLAD achieves state-of-the-art performance on multiple public log datasets, surpassing existing methods in key metrics such as precision, recall, and F1-score. These results underscore the framework’s efficacy and superiority in addressing log anomaly detection challenges.
The convergence of wireless communications and transportation technologies has propelled the evolution of Intelligent Transportation Systems (ITS), enhancing connectivity, safety, and efficiency in vehicular networks. As electric vehicle (EV) adoption grows within connected ecosystems, the need for intuitive charging management solutions has surged. This paper presents a task-oriented, bilingual chatbot to optimize EV charging control via an intelligent human-machine interface (HMI). Using Natural Language Processing (NLP) and Artificial Intelligence (AI), the system enables seamless interaction with charging infrastructure through wireless channels, supporting English and Spanish to address linguistic diversity. Integrated with ITS frameworks, it facilitates real-time monitoring and control, aligning with cooperative sensing and user experience goals. Two approaches are assessed: Voiceflow, for fast prototyping, and BERT (Bidirectional Encoder Representations from Transformers)-based models, implemented as a multilingual system and separate monolingual variants. Results show monolingual BERT models outperform the multilingual model in terms of intent classification accuracy, reinforcing reliability in dynamic settings. By linking wireless connectivity with AI-driven HMI, our proposal enhances EV charging scalability and accessibility within ITS, advancing sustainable transportation and efficient vehicular networking.
This paper investigates coded distributed matrix-vector multiplications in the context of eavesdropping, denial-of-service attacks, and computational accuracy. It is demonstrated that the coding problem can be reduced to analog secret sharing. The first analog zigzag-decodable secret-sharing scheme is introduced, and its mutual information security is derived. The encoding and decoding algorithms are shown to be numerically stable, relying solely on shift and addition operations, in contrast to existing methods that require matrix inversions or the solution of linear systems. The proposed scheme achieves comparable security levels while exhibiting a significantly smaller relative computation error than the analog Shamir’s secret-sharing scheme. Additionally, it is shown to outperform the analog Shamir’s scheme in model training accuracy when applied to linear regression and logistic regression problems.
A video streaming user is usually interested in some semantically-important scenes of a video content, e.g. goal scene in a football game. If the user could watch the video segments consisting that scene with high quality, the quality of experience (QoE) of the user would be substantially improved more than ever. Although Adaptive Bitrate (ABR) algorithm considering this aspect can achieve better QoE, design of current approach has been limited only inside the application layer. It is expected that higher QoE could be achieved when the underlying transport layer also handle these semantics information for its congestion control mechanism. In this paper, we propose a new concept for quality of service (QoS) control with differentiated explicit congestion notification (ECN) to improve QoE and reveal that throughput differentiation by ECN considering video semantics can improve QoE of a video streaming user. Evaluation results show that throughput differentiation by congestion control conducted inside the network improves the QoE of the user up to 57.9%.
Ciphertext policy attribute-based encryption (CP-ABE) is one of the most widely used encryption schemes for cloud data sharing due to its ability to protect data confidentiality and access controllability. However, most of the traditional CP-ABE schemes face challenges of distributed deployment difficulties and privacy leakage due to the fact that the attribute key is distributed by a central attribute authority, and the access policy is stored in plaintext. In this paper, we design a decentralized attribute encryption scheme (DABE-PU) with policy update privacy for data sharing. DABE-PU effectively integrates policy privacy and updating while avoiding the single point of failure risk arising from a single attribute authority. Therefore, DABE-PU helps to achieve data sharing that can be implemented in a distributed deployment and supports privacy for policy updating. In addition, DABE-PU ensures that users can complete ciphertext updating with low computation and communication overhead. Finally, we performed a formal security analysis and performance analysis of DABE-PU. The results show that DABE-PU can guarantee the privacy of policy updating as well as efficient ciphertext updating.
The rapid evolution of telecommunications technology underscores the vast potential of Low-Earth Orbit (LEO) satellite networks as a complement to traditional terrestrial systems, effectively addressing their inherent limitations. Nonetheless, LEO networks encounter formidable challenges. The high-velocity movement of satellites frequently leads to service disruptions, necessitating swift and adaptive rerouting mechanisms. Moreover, routing decisions must satisfy multiple constraints to ensure Quality of Service (QoS) and system reliability. These challenges are further intensified by the complexity of large-scale, multi-layer satellite constellations. To overcome these issues, this study formulates a multi-constrained optimization model. The integration of node-splitting techniques and multi-path routing strategies serves to simplify network topology and enhance overall reliability. Two efficient routing algorithms are introduced: the Optimization-based Routing Algorithm (OBRA) and the a-based Heuristic (H-AB). Computational evaluations reveal that OBRA achieves superior QoS optimization, while H-AB demonstrates significant advantages in computational efficiency. Both algorithms are capable of delivering real-time solutions, even within networks comprising up to 10,000 satellites. This work contributes to the advancement of routing methodologies for large-scale, multi-layer LEO satellite networks and supports the development of future 6G communication systems.
Massive Machine-Type Communications are one of the service types supported in 5G. They are characterized by massiveness and sporadic traffic, as well as to be energy efficient. While the scattered nature of the traffic occurrence does not pose a serious burden on efficient network planning, the requirement to serve a massive number of devices and to be energy-efficient certainly does. Moreover, besides the successful transmission/reception, these data need to be processed too. With limited resources on both the Radio Access Network part (responsible for communication) and edge cloud part (responsible for processing), as well as with the competition among a large number of users with this traffic in the cellular network, we need to address the problem of admission control, so that the network can successfully serve all the users that were admitted. To that end, in this paper we model the behavior of the system using a queueing network and perform the analysis that leads to admission policies for mMTC traffic with computation requirements. We do this both for homogeneous and heterogeneous users. Using data from a 5G trace, we validate our analytical results and provide further insights. Results show that the number of admitted users almost completely depends on the traffic pattern and that the entity with the lower capacity determines the number of admitted users.
The integration of large language models (LLMs) with mobile edge computing (MEC) systems presents a novel approach to enhancing vehicle-to-everything (V2X) connected autonomous driving. This study aims to address the prevalent challenges in multi-model sensor data fusion, such as latency, privacy preservation, and the need for dynamic adaptation to evolving environmental conditions, by leveraging real-time data from LiDAR sensors. We propose an LLM-based framework to improve V2X driving assistance systems’ operational efficiency, safety, and reliability, where pictures and image recognition work as integrated data from multiple sensors to train various vehicle and lane detection models. Based on the benefits of federated learning, i.e., distributed at each MEC server and optimising models accordingly, these training models can avoid the data privacy issue in V2X driving assistance implementation. The application of generated test data significantly improves the success rate of the lane detection feature and pedestrian detection, by 95% and 85%, respectively. The experiment results demonstrate that our proposed framework is effective and feasible.
For 5G/B5G communication systems, to cater the explosive demands in communication rates, extending communication band to millimeter wave (mmWave) seems to be an essential solution, and hence massive multiple-input-multiple-output (MIMO) architecture is usually adopted to combat the severe path-loss of mmWave signals. However, the traditional fully digital precoding manner is inapplicable owing to its high hardware cost and power consumption rendered by the individual requirements on radio frequency (RF) chains of each antenna. In this paper, for massive MIMO systems with multiple users, a novel hybrid precoding scheme is proposed based on the concept of equivalent channel. Specifically, for each user in the system, its analog combiner and analog precoder are designed jointly aiming to maximize the achievable rate with its equivalent channel. After the analog precoder/combiner phase-shifters for each user has been obtained, see them as a part of the channel and form a comprehensive channel of the multi-user system, and then the total baseband digital precoding at the BS is implemented with the block diagonalization method to delete the interference among users. Simulation results show that the proposed scheme can achieve higher spectral efficiency with low complexity compared to the existing schemes.
PTZ cameras offer drones rapid focusing and perspective adjustments, significantly simplifying drone operation. However, manual control of both the drone and camera can increase complexity and error rates. Therefore, our work develops a voice-controlled PTZ camera system for drones, integrating YOLOv8 to assist in target detection, enhancing convenience and efficiency. To achieve precise voice-command execution, the system employs advanced AI models. Whisper converts speech to text with high accuracy, while GPT-3.5 Turbo and LangChain extract key commands for camera control. The system adjusts the pan, tilt, and zoom features of the PTZ camera by utilizing the obtained keywords through the Raspberry Pi. By integrating these technologies, the system delivers a seamless and efficient user experience.
For a colocation data center with power oversubscription, its power capacity can be inadequate when the power loads of all the tenants peak simultaneously, due to lack of collaboration. In this paper, we point out that such simultaneous power peaks can be common and regular challenges rather than occasional events. To address the problem, we propose to exploit the virtual energy storage device (vESD) as an incentive mechanism in the pricing of rent rates and energy costs, to guide the power management of their tenants and eliminate the regular simultaneous power peaks. In addition, we design a dynamic allocation and pricing strategy to support the tenants to flexibly rent vESD for their respective cost optimization, and the fairness is guaranteed with Nash Equilibrium. The experiments based on real-world data center traces show that using vESD can improve the utilization of power infrastructures and save the provisioning costs by 32%.
The O-RAN architecture introduces unprecedented flexibility and openness into modern cellular networks, allowing mix-and-match of components from different vendors and the rapid deployment of innovative solutions across the RAN vertical. Despite its openness, some fundamental technical challenges associated with 5G/Next-G still remain in O-RAN. A well known example is joint optimization of Resource Block (RB) allocation, Modulation and Coding Scheme (MCS) selection, and Beamforming (BF) design. In this paper, we present Savitar—an O-RAN scheduler that jointly optimizes these components, with the objective of minimizing spectrum usage while meeting per-UE probabilistic data rate requirements. Following the multi-timescale design principle in O-RAN, we present three components (each at a different time scale) of Savitar that can be seamlessly integrated with O-RAN RICs: (i) hyperparameter tuning in the Non-Real-Time (Non-RT) RIC, (ii) parallel RB Group (RBG) allocation and MCS selection in the Near-RT RIC, and (iii) BF vector design in the RT Open Distributed Unit (O-DU). A unique design in these components is our handling of CSI uncertainty with limited data samples. Experimental results show that Savitar achieves competitive spectrum efficiency performance while meeting our design requirements (i.e., per-UE probabilistic data rate requirement and real-time requirement in O-DU).
In this paper, we design a novel unmanned aerial vehicle (UAV) aided covert cooperative cognitive radio (CR) scheme, where a UAV as the secondary transmitter can send its own covert signal to a secondary receiver while guaranteeing the quality of service for the primary user (PU). To accomplish the covert transmission of the secondary user, the PU’s signal is applied as a friendly interference to interrupt the detection of wardens. We first calculate the minimum detection error probability and Kullback-Leibler divergence under the finite blocklength constraint. Then, the average effective throughput maximization problem under the probabilistic line-of-sight channel is constructed by jointly optimizing the UAV’s transmit power and trajectory. Finally, simulation results demonstrate that the proposed UAV-assisted cooperative CR scheme is effective for covert air-ground transmissions against multiple wardens.
Federated learning is a collaborative training paradigm designed to protect private data and is widely used in the cooperative training of Internet of Things (IoT) devices. However, despite its focus on privacy protection, federated learning remains susceptible to poisoning attacks from malicious clients. These attacks can degrade system performance and potentially lead to data privacy breaches. Moreover, real-world IoT datasets are often heterogeneous, further increasing the difficulty of detecting malicious clients. Existing defense mechanisms often struggle to effectively identify malicious clients while maintaining high model performance. To address this issue, we propose a defense mechanism called Malicious client exclusion aggregation (MaEA). This method utilizes KL divergence to preliminarily filter out anomalous clients, aggregates the remaining (preliminarily filtered) clients to obtain a pre-center model, and then identifies and excludes malicious clients by measuring their deviations from this pre-center model. We executed a series of extensive experiments on the CIFAR-10 dataset to demonstrate the effectiveness of MaEA. The results demonstrate that our approach can efficiently detect and identify malicious clients while correcting model performance.
Efficient recovery in smart grids is critical to minimize outages and maintain stability. This paper addresses the backup power scheduling problem using Deep Reinforcement Learning (DRL). By modeling the recovery process as a Markov Decision Process, we propose a Deep Q-Learning (DQN) solution that dynamically activates backup power units and selects power nodes for repair. Simulation results on a standard test system demonstrate that our method reduces recovery time, improves resource utilization, and enhances grid stability compared to heuristic approaches.
Power grids integrated with advanced networking technologies and edge computing are rapidly evolving to meet modern energy demands. However, the growing reliance on IoT devices for real-time monitoring poses significant challenges in network congestion, load balancing, and efficient data routing. This research presents an edge-enabled IoT framework that dynamically integrates load balancing, resource allocation, and communication optimization in smart grids. The solution provided here incorporates the introduction of edge computing in the smart grids. Firstly, the edge devices are set up in the physical and the cyber layer to process the data, followed by introducing the load balancing and resource management framework in the edge devices to improve the resource management and load balancing of the smart grids. In addition, the proposed framework significantly enhances network efficiency by integrating edge computing to process energy data locally, thereby minimizing latency, reducing network congestion, and optimizing real-time power distribution. Experimental results demonstrate significant improvements, including increased throughput (11.8 Mbps), reduced delay (0.002%), and 15-20% lower energy consumption compared to existing methods. These findings establish a resilient, communication-driven model for future smart grids.
Radio frequency (RF) fingerprinting is a technique used to identify a wireless device based on its specific and unique hardware characteristics. In recent years, deep learning has been utilized for RF fingerprinting due to its superiority in feature extraction and higher classification accuracy. However, one major challenge of deep learning-based RF fingerprinting is that wireless signals are highly sensitive to environmental conditions, causing the device fingerprints captured in one environment to not transfer well to another. Hence, deep learning models are found to perform well in the same condition but lose their ability to classify devices in the new condition. In this paper, we examine three transfer learning techniques to mitigate the domain shift problem in RF fingerprinting and compare them with two well-defined baselines. The three RF fingerprinting datasets under various scenarios are examined to explore how environmental factors impact RF fingerprinting, such as transmitter locations, transmitter distance, and device configurations. We identify the most challenging scenarios and study how environmental factors lead to model deterioration through t-SNE visualization.
