We propose and comprehensively validate a wavelength conversion repeater with a simple digital signal processing (DSP) that compensates for fiber nonlinearity through phase conjugation and optimized chromatic dispersion compensation (CDC). First, we demonstrate in a 960-km single channel transmission that the proposed method improves the Q-factor by 1.3 dB compared to an optical-analog-optical (OAO) repeater. Using a validated simulation model, we further show a potential reach extension of 560 km beyond the OAO case. System performance is maximized by our proposed graphical analytical model. This model requires only basic link parameters and avoids computationally intensive simulations, enabling practical configuration for flexible optical networks involving wavelength conversion. Second, we investigate scalability to a multi-channel transmission system. To overcome the limitation of per-channel processing under wavelength division multiplexing (WDM) transmission, we introduce a subcarrier multiplexing repeater scheme that collectively processes subcarriers. A proof-of-concept experiment confirms effective compensation of both intra- and inter-subcarrier nonlinearities, achieving a Q-factor gain of 0.4 dB. These results highlight the proposed repeater as a practical solution for future optical networks requiring flexible wavelength conversion and enhanced transmission performance.

In Internet of Things (IoT) ecosystems, users and service providers deploy fog nodes to exchange and utilize IoT data for various high-quality services. IoT data often contain private and user-critical information; thus, privacy protection and efficient data distribution must be ensured. Previously, we proposed a ticket-based user data access control mechanism (UDAC-TK) that uses a ticket authentication and authorization infrastructure to mediate IoT data exchange between users and service providers, and evaluated its performance and security. In this study, we assessed its applicability to specific IoT services in terms of processing performance and security. We focused on next-generation smart meters, which are scheduled for gradual deployment across Japan from 2025 onward. We propose a distribution framework for electricity, water, and gas data and a model in which electricity smart meters function as gateways for distributing water and gas data. We analyzed the authentication processing load on the nodes constituting the ticket-based infrastructure. Performance evaluations confirmed municipal-scale feasibility only for low-frequency data collection scenarios, such as daily intervals, and showed that the required number of authentication and authorization nodes can be estimated from the total number of deployed smart meters.

Beyond 5G and 6G networks must fulfill diverse requirements of not only smartphones but also the internet of things (IoTs). The existing network control, which is common to all devices, cannot fulfill the network requirements of each device. However, controlling networks for each device is not feasible due to computational complexity. Therefore, controlling networks by device types, which are groups of connected devices with similar requirements, is necessary. Traffic classification is an approach to categorize connected devices into types according to user behaviors. Packet-level analysis of packet series is effective for network control because it recognizes differences in traffic characteristics with similar flow statistics, such as continuous, burst, and sparse traffic. This paper proposes a traffic classification method that analyzes encrypted packet series using only the size and the inter-arrival time of each packet. The proposed method represents packet series as graphs to learn the relation between the features. The proposed method represents the packet sizes and the packet inter-arrival times as node features and edge weights of graphs. The evaluation results show that the graph representations of the inter-arrival times as the node features are effective to classify traffic for a short duration. We also discuss the relation among the classification accuracies, the graph representations, and the classification models.

Generative AI (Artificial Intelligence), particularly large language models (LLMs), has been receiving widespread attention recently, and numerous approaches are being explored to apply it to daily operations across various industries. In the telecommunications sector, efforts are underway to utilize LLMs for the operation and management of network devices. One of the tasks that network engineers find particularly demanding is creating network configurations for these devices. This task requires highly specialized knowledge and extensive experience, but the number of engineers with these skills is limited, resulting in an increased workload for certain individuals. To mitigate this workload, attempts have been made to apply LLMs for generating network configurations. However, several issues arise when engineers use LLMs to generate network configurations for actual system integration cases. Network engineers are required to extract and provide relevant documents for the LLMs, to intervene in the control flow according to the output during the network configuration process, which consists of multiple steps, and configurations generated from insufficiently written design documents tend to be inaccurate. To address these issues, the authors proposed using ReAct (Reasoning and Acting), multi-agent, and human-in-the-loop architectures when applying LLMs for network configuration generation. We implemented these approaches into a prototype system, which generated network configurations without requiring engineers to extract and provide relevant documents or to intervene in the control flow between steps, and consulted them only when essential information was judged to be missing. This demonstrates the potential for LLMs to generate network configurations for actual system integration cases without increasing the operational workload for engineers.

The development of the Internet of Things (IoT) and artificial intelligence has led to a surge in data traffic, exceeding the processing capabilities and low-latency demands of terrestrial terminal devices. UAV-assisted mobile edge computing provides a flexible aerial communication platform, effectively addressing the limitations of traditional MEC services in specific scenarios and enhancing service quality. However, multi-UAV collaboration scenarios face challenges such as the coupling of frequency and computational resources, as well as partial observable environments. This paper proposes a multi-agent proximal policy optimization (MAPPO) algorithm framework, breaking through the limitations of traditional static resource allocation strategies. Initially, a dynamic regret model is designed to quantify historical resource competition loss, constructing a hybrid reward function that balances immediate and long-term benefits through the trade-off between discount factors. Subsequently, an event-driven collaborative training framework is constructed, reducing multi-device communication overhead based on an event-driven mechanism. The experiment shows that compared to other three benchmark algorithms, the proposed scheme achieves stable convergence of cumulative system rewards in dynamic channel scenarios, optimizing the total reward value, and verifying the coordination capability of dynamic regret mechanisms in multi-target conflict scenarios.

This paper proposes overloaded multiple-input-multiple-input (MIMO) assisted bi-directional communication with resource allocation in 2-hop relay networks that achieves higher throughput by utilizing more relays. Non-linear precoding for overloaded MIMO spatial multiplexing is applied in the networks, because the precoding can increase the number of the relays in spite of that of the terminal antennas. The proposed communication can increase the throughput as more relays are employed. Furthermore, this paper proposes resource allocation for the communication that selects relays based on the optimization with respect to the number of the relays. The throughput of the proposed communication is evaluated by computer simulation. Even when only 2 antennas are placed on the terminals with the quaternary phase shift keying (QPSK), the proposed bi-directional communication achieves 14 bits/Hz at the E_b/N₀ of about 30 dB in the 2-hop relay network based on the physical layer network coding.

In recent years, radar-based remote and non-contact vital sign detection has garnered significant attention. However, multiple subjects at the same distance and range bin overlap caused by walking can induce phase interference, making it one of the most challenging issues in non-contact vital sign detection. This problem can hinder vital sign detection and significantly degrade accuracy. Furthermore, most studies assume that subjects remain stationary as a prerequisite for measurement, which limits the feasibility and applicability of radar-based vital sign monitoring systems in real-world scenarios. Therefore, in this study, we proposed the application of a compact multiple-input multiple-output millimeter-wave frequency-modulated continuous wave radar for multi-person localization and heart rate estimation. A point cloud-based signal processing algorithm utilizing minimum variance distortionless response beamforming was designed to separate vital signs from stationary and moving subjects. The experiment conducted simultaneous heart rate measurements for multiple subjects across five experimental scenarios, achieving an average accuracy of 93.51%. Furthermore, this study evaluated heart rate estimation in real-world-like conditions, such as range bin overlap, demonstrating that the proposed method is capable of accurately measuring heart rates across various scenarios.

Given the increasing demand for localization technologies that leverage wireless communication, a wide range of techniques utilizing Time of Arrival (TOA) measurements have been proposed in recent years. Once TOA data are obtained, the next crucial step is to estimate the position of the target node using appropriate localization algorithms. Although numerous methods have been developed for this purpose, few papers provide a theoretical comparison of these techniques. Therefore, there remains the issue of having to choose among these algorithms without solid theoretical guarantees. In this paper, we provide a theoretical analysis comparing the relationships among Linear Least Squares (LLS), Weighted Linear Least Squares (WLLS), and Taylor Series (TS) which are widely used in TOA-based localization techniques. Specifically, the necessary and sufficient conditions for defining WLLS are provided. In addition, a comparison is conducted on the estimation accuracy and localization availability among the LLS, WLLS, and TS. Consequently, it is shown that TS is superior to WLLS, and that WLLS is superior to LLS, in terms of both estimation accuracy and localization availability.