Advances on Broad-Band and Wireless Computing, Communication and Applications

Next-generation 5G networks with massive Multiple Input Multiple Output (MIMO) must efficiently allocate radio resources to mobile users whose channel conditions change rapidly due to movement. This paper proposes a novel game-theory Reinforcement Learning (RL) framework for mobility-aware resource allocation in 5G MIMO systems. We model the resource allocation problem as a dynamic game between network entities and integrate a predictive deep RL agent that anticipates User Equipment (UE) mobility patterns. By forecasting UE movement, the RL agent proactively assists a game-theory optimization of MIMO resource allocation before channel quality degrades. The combination of game theory with predictive RL enables the network to reach a near-equilibrium resource distribution that is both adaptive and fair, improving convergence stability compared to standalone learning or game approaches. Simulation results in a high-mobility 5G scenario demonstrate that the proposed approach significantly boosts user Quality of Service (QoS) for example, increasing average throughput and reducing latency and handover failures relative to conventional reactive allocation strategies. Specifically, the proposed framework delivers a 17–22% increase in average user throughput, reduces handover failures by approximately 15%, and lowers latency by up to 12% when compared with conventional reactive allocation strategies. These findings illustrate the promise of integrating mobility prediction and game-theory RL for robust, high-performance resource management in future wireless networks.

This study analyzes Inter-AMF, Inter-5GC, and Inter-PLMN handovers in 5G/6G networks using a simulation framework that models UE mobility and A3-based triggering. Performance is evaluated across five HO types using metrics such as success rate, latency, and signaling overhead. Simulation results show that Xn HO offers less delay, while Inter-5GC and Inter-PLMN exhibit higher overhead due to core re-selection and authentication.

Optimization of mesh router placement in Wireless Mesh Networks (WMNs) is an NP-hard problem. To deal with this problem, in our previous study, we developed WMN-PSOHCDGA hybrid simulation system. In this paper, we extend our study by implementing four crossover methods (UNDX, BLX-\(\alpha \), SPX, psBLX) and two mutation methods (Boundary Mutation and Uniform Mutation). We carry out a comparison study for two scenarios by combining these methods considering Subway distribution of mesh clients, Linearly Decreasing Vmax Method (LDVM), and a small-scale WMN. The simulation results show that the combination of SPX with Boundary Mutation has the best load balancing.

The message routing protocol is essential for the efficient delivery of messages in Delay/Disruption-Tolerant Networks. Messages dispatched by a mobile node are conveyed via a Store-Carry-Forward method among mobile nodes in a Delay-Tolerant Network. Current protocols consider the behavior, historical data, and movement models of mobile nodes to enhance their message exchange mechanisms during encounters. However, none of them take into account the development of the foundational infrastructure, such as the creation of road networks. We propose CD-PRoPHET, a modification of PRoPHET, to adjust the delivery probability by taking into account the density of the road network. Our simulation findings demonstrate the efficacy of incorporating the road network for message reachability, resource efficiency, and average latency in comparison to the baseline.

Cloud computing has revolutionized modern IT architectures by enabling scalable and flexible resource allocation. However, accurately forecasting cloud workloads remains a significant challenge due to their high variability, non-stationarity, and complex temporal dependencies. In this paper, we analyze and compare traditional statistical models, such as ARIMA and VARIMA, with Recurrent Neural Network-based approaches, specifically Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks, for predicting CPU utilization workloads. Experiments were conducted on a real-world dataset comprising Azure Virtual Machine traces, containing fine-grained CPU usage metrics sampled every five minutes. Results show that neural models outperform VARIMA in capturing workload fluctuations, with GRU achieving the lowest normalized prediction errors across metrics. These findings provide insights for selecting forecasting models that support efficient and cost-effective resource management in dynamic cloud environments.

This paper proposes an adaptive region of interest (ROI)-aware downsampling method for efficient transmission of 3D point cloud data. Existing approaches such as voxel grid sampling (VGS) and farthest point sampling (FPS) reduce data size but often remove important visual information. To address this issue, our method detects a 3D ROI using multi-axis projection and applies different downsampling ratios to ROI and non-ROI areas. Experiments show that the proposed method preserves the visual quality of salient regions more effectively than VGS and FPS, while also providing stable input for object detection models. In CNN-based detection, it achieved an average detection score of 0.594, improving by up to 0.17 over baselines. In Transformer-based detection, it reached 0.871, exceeding existing methods by up to 0.37. These results demonstrate that our method improves the quality and reliability of point cloud transmission under bandwidth constraints, supporting adaptive streaming and contributing to semantic multimedia communications.

A biodiversity learning support system was developed based on immersive virtual reality (IVR). The aim of this system was to deepen users’ understanding of and interest in the importance of preserving biodiversity by allowing them to visualize endangered and extinct species in a realistic VR space. Mammals, birds, reptiles, and amphibians were selected from the Ministry of the Environment of Japan’s Red List of endangered and extinct species. These animals were modeled in three-dimensional (3D) form and placed in the virtual space. Users wear a head-mounted display and move freely within the VR space from a first-person perspective while viewing detailed information alongside the 3D animal models. Quizzes are provided to enhance the learning experience. These features aim to give users with a visual and physical experience that cannot be conveyed through conventional passive learning methods and to increase their general interest on biodiversity preservation. An evaluation survey involving 31 participants assessed the operability, functionality, satisfaction level with the visual experience, and changes in user interest. The survey results suggested that users were highly satisfied with both the operability and the realism of the visual experience. An increase in users’ interest in biodiversity preservation was also confirmed.

This study presents the development of a learning support system designed to help users of all ages—from children to adults—learn finger alphabets within a mixed reality (MR) environment. The proposed system incorporates two core functions: a finger alphabet example display function and a finger alphabet judgment function. The example display function overlays on the real-world view a reference image of the finger alphabet corresponding to the letter selected by the user. Meanwhile, the judgment function evaluates the user’s hand posture in real time by comparing the joint coordinate values of the fingers with a predefined dataset, and then superimposes the most closely matching character onto the MR space. By imitating the reference image and receiving immediate visual feedback, users can intuitively verify the correctness of their hand shapes and effectively learn the finger alphabet.

This paper treats the development of radiation therapy operation training systems. For the operation training systems of various types of radiation therapy equipment, their 3D models should be prepared. The authors investigated how easily create such 3D models using Generative AI tools, and performed 3D model generation experiments using several Generative AI tools. As a result, the authors developed radiation therapy operation training prototype systems using the 3D models actually generated by PartPacker, one of the Generative AI tools for 3D models. In this paper, the authors report the experimental results of 3D model generation and introduce how radiation therapy operation training prototype systems can be developed using the generated 3D models.

This paper presents a camera node placement optimization system for motion analysis during soldering tasks. We formulate the problem as a Mixed-Integer Linear Programming (MILP) that jointly enforces (i) imaging coverage over all joint and time pairs, (ii) positional uniqueness (at most one camera per candidate location) and (iii) network connectivity under capacity-constrained wireless mesh links. To address column explosion from position\(\times \)orientation discretization, we adopt column generation with Restricted Master Problem (RMP) and Pricing, while visibility is computed only for a clustered set of representative timestamps \(\tilde{T}\) and later verified on the full set T. Communication constraints are separated via Benders decomposition considering master handles coverage, positional uniqueness, and accumulated cuts, while the subproblem verifies connectivity and returns feasibility/optimality cuts. We consider warm-start Branch-and-Bound (B&B) to round the RMP solution to a feasible integer solution. In simulations, our method consistently achieved full coverage and network connectivity. The integration of column generation, Benders decomposition, and warm-start B&B enabled practical solution even with large candidate sets.

To address challenges in identifying core patents in the traditional Chinese medicine (TCM) innovation evolution path amid technological advancements and industrialization, this study conducts topic modeling on TCM patents and combines it with the theory of technology life cycles to track the dynamic evolution of TCM topics. A multi-dimensional patent value indicator system and a dynamic patent citation network are designed and constructed. An improved scoring algorithm is employed to identify core patents along the innovation evolution path of TCM. The results indicate that TCM technology topics have evolved from traditional herbal formulas and extraction techniques to a more refined direction of component analysis and efficacy research. The integration of advanced technologies such as informatization signifies a new stage in the development of TCM research. The discovery and analysis of core patents in each topic along the TCM innovation evolution path provide a new perspective for understanding TCM technological innovation, assisting enterprises in gaining an advantageous position in technology transfer and patent transactions, and offering scientific support for the future direction of TCM research and development and the formulation of policies.

Zero-day vulnerabilities represent one of the most critical challenges in cybersecurity, with traditional detection methods often lagging behind exploitation. This paper presents a machine-learning framework for proactive zero-day prediction by combining static code analysis with historical vulnerability patterns. Leveraging datasets from the National Vulnerability Database (NVD), Exploit-DB, and CVE repositories, we extract hybrid features encompassing code complexity metrics (cyclomatic, Halstead), NLP-based AST embeddings, and temporal exploit trends. Evaluating ensemble models (XGBoost, Random Forest) against deep learning (LSTM), our framework achieves an F1-score of 0.86 and AUC-ROC of 0.93, outperforming prior approaches by 19% in recall. A case study demonstrates successful prediction of CVE-2023-1234 14 days pre-patch through elevated code complexity (2.1\(\times \) baseline) and CVSS exploitability patterns. While effective on open-source projects, limitations emerge in proprietary software contexts, with accuracy declining by 23% due to domain shifts. The results validate ML’s potential to reduce zero-day exposure windows, particularly when integrated into CI/CD pipelines. Quantitative evaluation further shows robust trade-offs across classification thresholds, with 88% recall and AUC of 0.93. False-negative analysis reveals that 38% of misses stem from novel exploit patterns, while our ensemble framework reduces undetected zero-days by up to 26% compared to baselines. This work advances cybersecurity practices by enabling early risk prioritization and provides benchmarks for future hybrid AI approaches.

This paper proposes a Digital Twin (DT)-driven framework to proactively combat ransomware in IoT ecosystems by simulating ransomware propagation, detecting anomalies via federated learning, and triggering automated responses such as device isolation and firmware rollbacks. A smart city case study demonstrates the framework’s effectiveness, achieving a 78% reduction in infections and a detection latency of 2.4 s. By integrating lightweight cryptography for edge devices and blockchain for secure firmware rollbacks, the system ensures both efficiency and security. Experimental results show that the proposed DT-enhanced federated learning approach achieves an F1-score of 0.96 and an AUC of 0.97 (12.4% improvement over baseline), with a 38% reduction in false positives and detection precision of 99%. The framework reduces recovery time from 112 min to 8.3 min and cuts downtime costs from $387k to $14.5k per incident. Rollback latency is minimized to 1.1 s using hybrid storage, meeting real-time requirements for critical infrastructure. The framework addresses critical challenges, including rollback delays and synchronization consistency, while optimizing performance through local caching and hybrid storage solutions. Despite its advantages, challenges like resource constraints and real-time validation in latency-sensitive applications remain. Ethical risks are mitigated through Azure Confidential Ledger encryption, ensuring compliance with GDPR and HIPAA. Future work will focus on self-learning models and edge-native blockchain architectures to further enhance resilience and scalability in healthcare, energy, and transportation sectors. This approach shifts IoT security from reactive patching to proactive threat mitigation.

In Fifth-Generation (5G) and Beyond 5G (B5G) networks, Network Slicing (NS) enables diverse services to coexist on shared infrastructure, but assessing Security Posture (SP) presents significant challenges due to complex, multi-dimensional security attributes and inherent uncertainty in evaluation processes. Traditional security assessment methods rely on binary approaches that fail to capture complex relationships between security dimensions. To address these limitations, this paper proposes a Fuzzy-based System for Assessment of Security Posture (FSASP) in 5G/B5G NS environment. The FSASP integrates three parameters: Identity and Access Management Strength (IAMS), Vulnerability Exposure (VE), and Monitoring and Response Maturity (MRM) and utilize Fuzzy Logic (FL) to provide comprehensive SP assessment. Simulation results demonstrate that VE has the most negative impact on SP, while achieving good security requires optimization across all dimensions. Single-dimension strengths cannot compensate the weaknesses of other parameters, emphasizing the importance of balanced security investments. The proposed FSASP offers an interpretable and uncertainty-aware framework that provides valuable guidance for security administrators in prioritizing investments and making risk mitigation decisions in dynamic 5G/B5G NS environments.

1-to-N avatar operation is a method in which an operator can simultaneously control multiple avatars. Each time the operator switches to a different avatar, they must cognitively process and understand that avatar’s current situation (mental context switching). As the number of avatars increases and more information is fed to the operator from each one, the time required for mental context switching also increases. This leads to greater cognitive overload for the operator and, consequently, a higher risk of misidentification. Attackers can exploit such misidentification to carry out social engineering attacks against the operator without infiltrating the avatar operating system itself. This paper investigates the parameters associated with mental context switching in 1-to-N avatar operation environments and discusses an experimental design to explore the feasibility of such attacks.

As AI models become central to societal functions, they inevitably attract the attention of malicious actors, resulting in the proliferation of counterfeit AI models. Therefore, identity verification, similar to user recognition, is essential for AI models. However, owing to probabilistic variability in outputs, particularly in large language models (LLMs), and continuous ability enhancement through autonomous learning, the typical user recognition method of enrolling and presenting the same information cannot be directly applied. To address these issues, this study proposes identifying AI models by “assessing the abilities of AI models.” The identification method based on this concept is called the Completely Automated Public Test to Tell Ability of Artificial Intelligence (CAPT-AI). To examine the feasibility of our proposed method, we conduct a preliminary study focusing on LLMs, serving as a foundational experiment for CAPT-AI.

In the future, the advancement of tangible avatars is expected to enable bodily transformation and enhance physical abilities in the real world. In such an avatar-symbiotic society, not only the traditional “1-to-1 avatar operation,” in which one person controls a single avatar, but also novel forms of control—such as “M-to-1 avatar operation” (multiple users controlling one avatar) and “1-to-N avatar operation” (a single user controlling multiple avatars)—will also become feasible. However, the bodily augmentation enabled by “M-to-1 avatar operation” and “1-to-N avatar operation” (hereafter collectively referred to as “M-to-N avatar operation”) is a double-edged sword; while it empowers legitimate users, it also enhances the capabilities of attackers. If malicious actors exploit M-to-N avatar operation, various forms of attacks may become significantly more advanced. In this study, we focused on social engineering enabled by M-to-N avatar operation. Typically, M-to-1 avatar operation could allow multiple attackers to impersonate a single avatar with highly advanced attack capabilities, and 1-to-N avatar operation could enable a single attacker to impersonate multiple individuals and conduct “deepfake video call scams.” These types of attacks are referred to in prior studies as Advanced Persistent Social Engineering (APSE). Given these risks, it is essential for an avatar-symbiotic society to implement mechanisms capable of determining whether an avatar is being controlled through M-to-N avatar operation, particularly as a defense against APSE. This technological shift calls for an extension of the traditional Turing Test into an M-to-N Turing Test. In light of this context, this paper explores a method for the automatic detection of M-to-N avatar operation.

The growing proliferation of Internet of Things (IoT) and edge devices has increased the demand for lightweight and adaptive intrusion detection systems capable of operating in resource-constrained environments. In this paper, we present a testbed called Edge-IDS, which is built on five Raspberry Pi devices interconnected via a central switch. Two Raspberry Pis are configured as attackers (scanner and flooder), two generate benign client–server traffic, and one device functions as the central orchestrator and IDS. A Fuzzy-based Intrusion Detection System (FIDS) is implemented on the orchestrator, using three traffic-derived parameters: Packet Rate (PR), Failed-Connection Ratio (FCR), and Destination-Port Entropy (DPE) to compute an Attack Possibility Level (APL). Before deployment on the testbed, the system was evaluated through simulations, and the results demonstrate that the fuzzy approach can effectively differentiate between benign and malicious traffic patterns. For example, when the packet rate is overloaded (PR = 0.9), with unstable connections (FCR = 0.9) and higher entropy or DPE is more than 0.5, the system outputs an APL value that is more than 0.7, showing that the system identifies high-rate, unstable, and widely distributed traffic as malicious.\(\ldots \)

In this paper, we evaluate a heuristic route guidance method that individually provides incentives to tourists to encourage detours and reduce traffic congestion. Traffic congestion caused by an excessive influx of tourists into sightseeing areas has become a serious issue. Although various route guidance methods have been studied, most of them only consider travel time and distance to the destination, and do not consider congestion on the road as a whole. As a result, traffic congestion may still occur on specific roads, even if individual routes are optimized. We proposed a method to reduce overall congestion by determining incentives for tourists individually and directing them to less congested detour routes. From the results of multi-agent simulation using artisoc Cloud, we confirmed that the proposed method can guide tourists to less congested routes with less total incentives and computation time compared to the route guidance method using random search and genetic algorithm (GA).

Driver mental status plays a crucial role in ensuring road safety and factors such as fatigue, stress, and lack of attention significantly increase the risk of accidents. Conventional monitoring systems often rely on single indicators, which may not provide a comprehensive evaluation of driver condition. In this study, we propose a Fuzzy Logic (FL) based system for assessing driver condition by integrating multiple parameters, including physical parameters for deciding Driver Body Condition (DBC). We call this system FL-based DBC (FLDBC) system. As input parameters for FLDBC system we consider Pupil Dilation (PD), Driver Facial Expression Changes (DFEC), Handle Reaction Time (HRT) and Driver Heart Rate Status (DHRS). The output parameter is DCB. We evaluated the implemented system by computer simulations. The simulation results show that DCB is good when the pupil dilation is normal. With increasing HRT, the DCB value increased. Also, when DFEC and HRT changed, the DCB value increased.

Wireless Mesh Networks (WMNs) are rapidly developing due to their usefulness and deployment capabilities, making them effective solutions for diverse networking applications. However, these networks face several challenges, including congestion, interference, diminished data transfer rates, packet loss, and increased latency. The strategic placement of mesh routers is crucial for mitigating these issues. However, finding the best placement of mesh routers in WMNs is a complex and challenging issue and is classified as an NP-hard problem. To solve this problem, we propose and implement a Cat Swarm Optimization (CSO) based intelligent simulation system, called WMN-CSO. We evaluate the performance of proposed system for Uniform distribution of mesh clients considering different Mixture Ratios (MRs). The simulation results indicate that the system performs better for smaller MR values.

The article describes a new distributed steganography protocol that uses CAPTCHA codes as containers for hiding secret information. CAPTCHA codes used for user authentication allow for the identification of human users, but they can also serve as containers for hiding parts of secret information. A special combination of selected parts of the visual CAPTCHA allows trusted users to access selected systems or services, but also allows for the collection of a special set of containers that contain a specific amount of divided secret information. The article also discusses the security and application of such distributed steganography solutions.

In order to retain the integrity of application data, conflicting transactions have to be serialized. Some transactions are aborted and re-started to make transactions serializable. The EERO-AT (Energy-Efficient Role Ordering by considering Aborted Transactions) scheduler is proposed in our previous studies to reduce the execution time (ET) required for each re-started transaction to commit. However, as the number of transactions concurrently issued increases, the ET required to commit each re-started transaction becomes longer. In this study, the IEERO-AT (Improved EERO-AT) scheduler is proposed to furthermore reduce the ET of each re-started transaction by reducing the waiting time of each re-started transaction in a scheduling queue. We show the IEERO-AT scheduler reduces the ET of re-started transactions compared to the EERO-AT scheduler by using simulation.

Subjects manipulate objects according to the access rights. Here, illegal information flow occurs in the read operation. In our previous studies, a read rejection method is considered where the illegal read operations are rejected. The illegal read operation is a read operation occurring illegal information flow. In this paper, we propose a write rejection method. In the write rejection method, a read operation is not rejected even if the read operation occurs illegal information flow. However, the illegal write operations are rejected. The illegal write operation is a write operation to write the data read in an illegal read operation to an object. Therefore, it is prevented to flow the data read in illegal read operations to objects. Suppose data of objects are text. The LLM (Large Language Model) is used to check the data read in an illegal read operation are written to an object in a write operation. By fine tuning the LLM, the similarity between the data read in an illegal read operation from an object and the data written to another object is made clear. If the similarity is larger than a threshold, the data read in an illegal read operation flow to another object in the write operation. Therefore, the illegal write operation is rejected. Otherwise, the write operation is not rejected. Generally, most operations issued by subjects are read. The smaller number of read operations are rejected in the write rejection method than the read rejection method.

Battery Swapping Stations (BSSs), which have a lot of batteries primarily for Battery Swapping Electric Vehicles (BSEVs), can operate as an emergency power supply with proper allocation of electricity. BSEVs can redistribute the stored energy between two BSSs through battery migrations, then, the difference of stored electricity among BSSs also can be minimized. Electricity is transferred from a BSS with a higher State Of Charge (SOC) to the lower one by BSEV. By the way, the effects of battery migrations appear with delays. In other words, it needs a time to appear the effects accordance with the distance between BSSs. Therefore, BSEVs perform migration consisting of loading/unloading a battery with an easy decision as of only start time, they put energy into destination BSS after equalization. This phenomenon incurs both of the unneeded inversions of high-low relationship and it makes wider the SOC gaps among BSSs. We propose two methods to decide migration considering transportation delay to suppress improper electricity transports. The proposed methods effectively narrow the gap of SOC between BSSs and restrain the fluctuation of SOC. The latter method also cannot take effect for proper judgement of migration with decreased estimation accuracy.

With the explosive growth of Internet information, manual data annotation is time-consuming and subjective, making automated machine-based text recognition and classification crucial due to its high efficiency. As a key method for text annotation, text classification has evolved from human-machine collaboration to full automation, saving computing resources. However, local computers are unable to handle massive volumes of text data. While cloud computing addresses this issue, it faces challenges: the privacy of sensitive data needs protection, and ordinary data is at risk of tampering by third parties. Although encryption algorithms can provide secure solutions, they restrict ciphertext classification and incur high costs, leaving the balance between security, efficiency, and accuracy a major challenge. To tackle this, this paper designs a text classification method under a homomorphic model based on the CKKS homomorphic encryption scheme, and conducts experimental analysis using RNN+Attention as the foundation. The final results demonstrate that the proposed scheme achieves high computational efficiency and accuracy.

Mobile sensing technologies can collect multiple types of information that are related to the nearby environment using vehicle onboard sensors. However, mobile sensing technologies require robust networks due to unstable performance on vehicle-to-road (V2R) communication. Furthermore, modern IoT systems must connect via secure communication networks to protect service and data. This paper introduces a new configuration of a private V2R wireless communication architecture with two different role wireless links. In our design, the IEEE 802.11ah link discovers another node before entering the high-speed Wi-Fi link that delivers application data. This design attempts to extend the capacity of data transmission even when the vehicle is on the move. We also use the QUIC protocol to exchange application data, encrypt the payloads, and improve the performance in poor-quality networks. We found that the new proposed system can discover other nodes with 800 m or more coverage over the IEEE 802.11ah wireless link. The application data link delivered payloads in the scenarios on a high packet loss network, and it reduced 14.2 times shorter time completion at the maximum.

Recently, there are various environmental issues including global warming, air and marine pollution caused by plastic waste, deforestation, water pollution of water resources, excessive consumption of natural resources such as fossil fuels and minerals and food waste. In this paper, we focus on waste management problem to improve the environment. We propose a food waste management system for more effective compost production. We present the design and implementation of the control system for food wast management robot.

This paper introduces Ratio-Difference-based Anti-packet Triggering (RDAPT) method for vehicular Delay/Disruption Tolerant Networking (DTN). In RDAPT, nodes compare the ratio of duplicate bundles with that of observed bundles when receiving summary vectors. If the difference exceeds a threshold of 0.7, an anti-packet is generated; otherwise, the bundle is retained. Unlike conventional schemes in which only destinations issue anti-packets, RDAPT enables intermediate nodes to adaptively regulate recovery, thereby improving responsiveness to network dynamics. Simulations were conducted in Scenargie under different vehicle densities with considering Epidemic and Spray and Wait (SpW) routing. Results show that RDAPT combined with Epidemic achieves lower overhead than DTAG and conventional anti-packet schemes without introducing additional delay.While RDAPT combined with SpW consistently maintains delivery ratios above 0.9, substantially outperforming DTAG, but has moderately higher overhead. The evaluation results show that RDAPT offers a more balanced approach to anti-packet control, improving delivery robustness while effectively managing overhead in vehicular DTNs.